University of Southern California Dissertations and Theses

Exposure to alcohol advertising on television and alcohol use among young adolescents

(USC Thesis Other)

Exposure to alcohol advertising on television and alcohol use among young adolescents

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content
EXPOSURE TO ALCOHOL ADVERTISING ON TELEVISION AND ALCOHOL
USE AMONG YOUNG ADOLESCENTS

by

Jerry L. Grenard

A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(PREVENTIVE MEDICINE)

August, 2008

Copyright 2008 Jerry L. Grenard

ii
Dedication
To Brenda.

iii
Acknowledgments
I would like to thank my committee members Drs. Jennifer Unger, Clyde Dent,
Susan L. Ames, Michael Cody, and especially my committee chairman, Alan W. Stacy.
Dr. Stacy provided tireless support and guidance throughout the entire process of
developing the ideas for this project and in reviewing its progress. Special thanks also go
to Dr. Ames for asking tough questions and providing much appreciated encouragement.
Drs. Steve Sussman, Andy Johnson, and Mary Ann Pentz have also provided invaluable
advice and assistance during my time at IPR. In addition to the faculty, I would like to
thank Marny Barovich and James Pike for their support throughout my doctoral program.
Without Marny’s help, I would probably still be trying to figure out how to sign up for
my first class. She guides students through the program with incredible patience and
expertise. Thanks also go to my fellow classmates who made the whole experience great
fun.
This research was supported in part by grants from the National Institute on
Alcohol Abuse and Alcoholism (AA12128) and the National Institute on Drug Abuse
(DA16094).

iv

Table of Contents
Dedication ii
Acknowledgments iii
List of Tables vi
List of Figures vii
Abstract viii
Chapter 1 Introduction 1
Specific Aims 1
Background and Significance 3
Problem Importance 3
Risk Factors for Alcohol Abuse by Adolescents 5
Persuasive Communications 8
Theoretical Models of Alcohol Advertising Effects 17
Empirical Studies on Alcohol Advertising 22
Assessment of Exposure to Alcohol Advertisements 36
Indirect Measurement of Associations in Memory 47
Chapter 2 Measurement Model for Marketing Research Measures Used to
Appraise Alcohol Advertisements on Television. 61
Abstract 61
Introduction 61
Advertising exposure assessments 63
Measurement models, CFA, and SEM 65
Reflective and formative indicators in CFA 67
CFA and testing measurement invariance 69
The current study 72
Methods 73
Participants 73
Procedures 74
Measures 74
Data Analyses 81
Results 85
Demographics 85
Indicator Types 91
Measurement Invariance Testing 94
Discussion 104

v
Chapter 3 Exposure to Alcohol Advertising on Television and Alcohol Use
among Young Adolescents 107
Abstract 107
Introduction 107
Methods 117
Participants 117
Procedures 118
Measures 118
Data analyses 122
Results 125
Measurement model 131
Structural growth models 136
Structural mediation models 142
Discussion 149
Chapter 4 Exposure to Alcohol Advertising and the Development of Alcohol-
Related Associations in Memory 155
Abstract 155
Introduction 155
Methods 162
Participants 162
Procedure 163
Measures of drug-related associations in memory 163
Measures related to televised alcohol advertising exposure 166
Alcohol use measures 169
Data Analysis 170
Results 171
Discussion 187
Chapter 5 Summary and Discussion 192
References 197

vi
List of Tables
Table 2-1 Demographic Information 86
Table 2-2 Alcohol Use and Problem Consequences by Gender 88
Table 2-3 Goodness-of-Fit Statistics for Model Runs on 5 Imputed Data Sets. 95
Table 2-4 Factor Loading Estimates for the Final Constrained Group Model. 99
Table 2-5 Estimates for Factor Means and Variances in the Constrained Model. 101
Table 2-6 Comparison of Parameters for the Split, Cross-Validation Samples. 102
Table 2-7 Correlations among Latent Factors for Girls and Boys. 103
Table 3-1 Demographic Information for Participants in 7
th
Grade. 126
Table 3-2 Bivariate Correlations among Alcohol Use and Ad Exposure Measures. 130
Table 3-3 Standardized Parameter Estimates for the Alcohol Use Growth Model. 139
Table 3-4 Standardized Parameter Estimates for the Mediation Model. 146
Table 4-1 Study Three Demographic Information. 172
Table 4-2 Standardized Estimates for the Parallel Growth Curve Models. 178
Table 4-3 Percentage of Responses to Homograph Cues by Meaning Category. 182

vii
List of Figures
Figure 1-1 Strickland Model (Strickland, 1983). 21
Figure 3-1 Model for Growth of Alcohol Use 115
Figure 3-2 Model for Mediation of Exposure and Problems 116
Figure 3-3 Measurement Model. 133
Figure 3-4 Growth Model by Gender as Predicted by Viewing of Popular Shows. 138
Figure 3-5 Interaction of Exposure to Ads with Liking of Ads. 141
Figure 3-6 Mediation Model for Alcohol-Related Problems. 145
Figure 4-1 Growth of Alcohol-Related Associations for the CBAT. 176
Figure 4-2 Growth of Alcohol-Related Associations for the COBT. 177

viii
Abstract
Three studies provided support for the hypothesis that televised alcohol
advertisements influence underage drinking. Data were collected from 3,890 students in a
prospective study covering the 7
th
through 10
th
grades. Measures of exposure to alcohol
advertising included self-reported observation of alcohol ads on TV and 2 memory
measures (top of mind awareness and cued recall). Opportunity-based measures assessed
exposure to alcohol ads indirectly by first asking about the frequency that participants
watched specific programs (i.e., popular shows and sports programs) and then weighting
the responses by the frequency of alcohol ads broadcast during each of those programs.
One key affective measure assessed how much participants liked alcohol ads compared to
other ads on TV. Additional measures assessed demographic variables, sports
participation, alcohol use, and problems associated with alcohol use. The results from a
confirmatory factor analysis (CFA) in study 1 (chapter 2) demonstrated that each of the
measures related to alcohol ads, except top of mind awareness, had good measurement
properties. The measures loaded well on single latent factors as expected with no cross-
factor loadings and the factor loadings and thresholds were invariant across gender. Study
2 (chapter 3) provided support for a causal relationship between exposure to televised
alcohol advertising and underage drinking. Structural equation modeling of alcohol
consumption showed that exposure to alcohol ads and/or liking of those ads in 7
th
grade
were predictive of latent growth factors for alcohol use after controlling for a range of
covariates. There was a significant total effect for males and a significant mediated effect
for females of exposure to alcohol ads and liking of those ads in 7
th
grade through latent

ix
growth factors for alcohol use on alcohol-related problems in 10
th
grade. Study 3 (chapter
4) results were consistent with the hypothesis that exposure to televised alcohol
commercials contributes to the development of spontaneous alcohol-related associations
in memory. This is a key finding because previous studies have demonstrated that
spontaneous associations such as these are predictive of alcohol use prospectively. Taken
together, these 3 studies support the accumulating evidence that alcohol ads on TV are a
contributing factor in underage drinking.

1
Chapter 1 Introduction
Specific Aims
This project examines the influence of televised alcohol advertisements on
adolescent consumption of alcohol, negative consequences as a result of drinking alcohol,
and development of alcohol-related associations in memory. Alcohol is the most common
psychoactive substance used by adolescents and is often a contributing factor to the
number one cause of morbidity and mortality among this age group, accidental injury or
death especially due to automobile accidents. A few early studies on the effects of
alcohol advertising on youth had mixed results, but more generally, studies have shown
small but significant associations between exposure to alcohol advertising and alcohol
use by adolescents. This project contributes to the literature in three areas. In the first
study (see Chapter 2), detailed psychometric properties of advertising exposure measures
are examined using confirmatory factor analysis (CFA), and these properties inform the
current and future studies on which measures provide the best information on the
constructs of interest. The second study (Chapter 3) examines mediator and moderator
relationships in a prospective study of alcohol use. Affective reactions to beer
commercials on television, for example, are expected to moderate the influence of
exposure to those ads on beer consumption. The frequency of drinking alcohol will
mediate the relationship between exposure to alcohol advertisements and the occurrence
of negative consequences resulting from alcohol consumption. The third study (Chapter
4) involves the development of associations in memory. Dual process theories of
cognition propose that life experiences lead to the development of associations in

2
memory that influence decision-making in future, related contexts with limited conscious
awareness of the original source or influence of those associations. Alcohol
advertisements may contribute to the development of associations in memory between
drinking alcohol and having fun, for example, which later influence an adolescent’s
decision at a preconscious level (and largely bypassing rational decision-making
processes) to drink beer with friends. The results of these studies have important
implications for policy making and the design of prevention programs. Policy making
requires a clear understanding of the relationship of alcohol advertising and the
development of underage drinking to design policies that are effective at protecting our
youth. This study will further that understanding to inform decisions about television
advertisements and to inform the design of prevention programs. More effective
interventions can be designed if the moderators and mediators of alcohol consumption
and problem consequences are understood. The specific aims were as follows:
Study 1 (see Chapter 2)
1. Examine the properties of the alcohol advertising exposure measures using
confirmatory factor analyses.
Study 2 (see Chapter 3)
1. Examine the moderating influence of liking for alcohol advertisements on
exposure to alcohol ads and the prediction of alcohol involvement over time.
2. Test a model where the growth in alcohol use over time mediates the relationship
between exposure to alcohol advertisements and negative consequences related to
alcohol use.

3
3. Determine if the moderating and mediating relationships differ across gender.
Study 3 (see Chapter 4)
1. Determine which life experience factors contribute to the development of alcohol-
related associations in memory over time.
2. Describe the changes that occur in 18 homograph association norms across 3
years among adolescents to obtain some initial indications for when particular
associations do and do not change during this period of development.
Background and Significance
Problem Importance
Prevalence of alcohol use. Alcohol is a major health concern among adolescents
and young adults. In a recent national report, the prevalence of current alcohol use
reported by age increased from 2.9% for persons aged 12 to a peak of 70% for persons
aged 21 (SAMHSA, 2004). There were about 10.9 million (29%) underage adolescents
between 12 and 20 who reported drinking alcohol in the past month prior to the survey.
Of those persons between the ages of 12 and 17 years, 17.7% reported drinking in the
past month, 10.2% were binge drinkers (5 or more drinks on one occasion at least once
during the past month), and 2.6% were heavy drinkers (5 or more drinks on the same
occasion at least 5 different days in the past month) (SAMHSA, 2004). A study by
Lewinsohn, Rohde, and Seeley (1996) interviewed 1507 adolescents aged 14 to 18 years
and found that 16.6% indicated problem use behaviors and an additional 6.2% met DSM-
IV criteria for substance abuse or dependence. Gender and ethnic differences in quantity
of consumption and health consequences had been reported with males more at risk than

4
females and White, American Indian, and Alaska Native adolescents more at risk than
Hispanics, African Americans, and Asians (Windle & Windle, 2006).
Health risks of alcohol use. Several health risks are associated with alcohol use by
adolescents including accidental injuries, homicides, and suicides, which are the three
leading causes of mortality among adolescents (Windle & Windle, 2006). Accidental
injuries cause more than 50% of all deaths in this age group (CDC, 2004). One important
concern is driving under the influence of alcohol. The percentage of persons aged 16 to17
who reported driving under the influence in the past year was 9.7%, and for those 18 to
20 years old, 20.1% reported driving under the influence (SAMHSA, 2004).
Approximately 20% of drivers between the ages of 16 and 20 with blood alcohol content
greater than 0.08 b/dl, were involved in fatal crashes between 1991 and 2001 (NHTSA,
2003). In addition to automobile accidents, alcohol consumption by adolescents has been
associated with depressed mood, risky behaviors, poor academic performance, and
smoking tobacco (Spirito et al., 2001). Other studies have linked extended alcohol use by
adolescents to abnormal brain development including memory and attention problems
(Tapert, Granholm, Leedy, & Brown, 2002), reduced hippocampal volumes (S. A. Brown
& Tapert, 2004), and abnormal brain responses to challenging cognitive tasks (S. A.
Brown & Tapert, 2004). The initiation of alcohol use at an early age increases the risk for
adverse consequences later in life due to its association with heavier and more persistent
consumption (Maggs & Schulenberg, 2006).
Riley (2005) presented recommendations given to the WHO by public health
experts on policies concerning the marketing and promotion of alcohol to adolescents.

5
First, the WHO should assist countries to take regulatory steps necessary to prevent the
exposure of young people to promotional messages about alcohol. Second, the WHO
should assist countries with the development and implementation of programs that
counter the effects of media advertising. Third, the participation of young people is
important to expose the illusions perpetuated by alcohol advertising. Finally, the WHO
should help make sure that negotiations on trade do not inhibit the ability of local
communities to establish local alcohol abuse prevention policies.
Risk Factors for Alcohol Abuse by Adolescents
There have been a large number of studies examining the risk factors for drinking
alcohol. Rather than conduct an extensive review of this research, the current manuscript
will provide a few representative studies across a range of risk factors. The studies
reviewed deal primarily with adolescents or in some cases with college aged participants.
Influence of parents and peers. Parents and peers have been shown to influence
the alcohol-related behavior of adolescents in a number of studies. Higher parental
monitoring of youth and a negative attitude toward alcohol appears to reduce the risk of
drinking by adolescents (Fergusson, Horwood, & Lynskey, 1995; Wood, Read, Mitchell,
& Brand, 2004). Conversely, an increased risk of drinking by adolescents is associated
with a positive attitude toward alcohol and a higher level of drinking by parents
(Feldman, Harvey, Holowaty, & Shortt, 1999; Wood et al., 2004). Adolescents are also at
higher risk of drinking if their friends drink (Feldman et al., 1999), but the influence of
friends may be moderated by the attitudes of the parents toward drinking (Wood et al.,
2004). Wills, Resko, Ainette, & Mendoza (2004) found that parental support was

6
positively related to protective factors, but peer support had a mixed relationship with
alcohol use and protective or risk factors.
Trauma, Stress, and Coping. Alcohol may be consumed to relieve the effects of
trauma and stress. Studies have shown that lifetime traumatic events and other adverse
events are positively associated with alcohol abuse and dependence (Clark, Lesnick, &
Hegedus, 1997; Kilpatrick et al., 2000; Wills, Sandy, Yaeger, Cleary, & Shinar, 2001). In
addition, problem focused coping styles reduce the risk of alcohol abuse while
disengagement coping styles (anger, helplessness, and avoidance) increase the risk of
substance use (Wills et al., 2001).
Alcohol Expectancies. Alcohol expectancies are the anticipated effects of
consuming alcohol, and they may be acquired from both cultural sources and observation
as well as from personal experiences with drinking (Christiansen, Goldman, & Inn,
1982). Examples of expectancies include ‘I am more romantic when I drink’ and ‘alcohol
makes me happy’ (Christiansen, Goldman, & Brown, 1985). The endorsement of the
enhanced social interaction expectancies for alcohol use by adolescents has been shown
to predict alcohol consumption prospectively (Christiansen, Smith, Roehling, &
Goldman, 1989; Killen et al., 1996), and drinking has been shown to mediate the
relationship between positive expectancies and alcohol-related problems (D'Amico et al.,
2002). Coping and emotional enhancement motives have been shown to mediate the
relationship between alcohol expectancies and alcohol consumption (Cooper, Frone,
Russell, & Mudar, 1995).

7
Memory Associations. Persons with strong alcohol-related associations to
ambiguous cues are at higher risk for higher alcohol use (Stacy, 1997; Stacy &
Newcomb, 1998; Weingardt, Stacy, & Leigh, 1996). Automatic or spontaneous
associations or cognitions may be described as habitual patterns of cognitive behavior
that have developed as a result of prior experiences. Associations related to alcohol use
begin to strengthen as a person observes or engages in alcohol drinking episodes (Stacy,
1997). As experiences strengthen associations related to drinking alcohol, cognitions
about drinking become more accessible in memory, and these spontaneous cognitions can
overshadow deliberative reasoning about the pros and cons of drinking behavior.
Spontaneous alcohol-related cognitions may be more closely related to the intuitive
system than to the reasoning system in the dual-process model of judgment and choice
described by Kahneman (2003). Dual-process theories and associations in memory are
described more fully below.
Genetics and Personality. A number of factors with some genetic basis have been
associated with alcohol use. Having parents and family members with substance use
disorders has been identified as a risk factor for adolescents to abuse alcohol whether the
child was raised with those family members or not (Hoffmann & Cerbone, 2002; Miles et
al., 1998). Personality factors that have been associated with alcohol use include poor
behavioral control such as conduct disorder, aggression, and antisocial behaviors (Martin,
Lynch, Pollock, & Clark, 2000; Mustanski, Viken, Kaprio, & Rose, 2003), temperament
(Wills & Stoolmiller, 2002), and excitement seeking (Crawford, Pentz, Chou, Li, &

8
Dwyer, 2003; Mustanski et al., 2003). Attention Deficit Hyperactivity Disorder (ADHD)
is also associated with the risk of alcohol use (e.g., Molina & Pelham, 2001, 2003).
The preceding list highlights some of the known risk factors for the abuse of
alcohol among adolescents. One environmental risk factor that was not discussed
concerns the influence of alcohol advertising broadcast on television. The following
sections address this area in more detail.
Persuasive Communications
Alcohol advertising has been a subject of concern especially for vulnerable
populations such as adolescents. Alcohol companies might or might not intentionally
target adolescents with their advertisements, but the question remains whether alcohol
commercials seen on television are influencing adolescents to initiate early
experimentation with alcohol and whether this early experimentation puts adolescents at
risk for problems related to alcohol use including alcohol-related accidents and alcohol
abuse or dependence later in life. This section will review some of the theories that
describe how alcohol advertisements might affect alcohol use. In addition, the section
will review empirical studies of alcohol advertising including content analyses,
econometric studies, experimental studies, and observational studies.
Theories of persuasion in social psychology. Theories of attitude change and
persuasion developed in social psychology can be applied to the study of the effects of
alcohol advertising exposure on adolescents. This is a large area of study, but a few of the
more influential theories will be reviewed briefly. Some of the first systematic studies of
persuasion were reported in the 1940s and 1950s by Carl Hovland and associates from

9
Yale University (for an overview, see Bettinghaus & Cody, 1994). These researchers
assumed that people must learn and remember messages if persuasive communications
were to be effective. There were four sequential steps proposed underlying the message
learning process. The receiver of a persuasive communication must (a) pay attention to
the message, (b) comprehend the message, (c) yield to the message and accept the
advocated position, and (d) remember the information. Based upon these steps, a wide
range of factors affecting persuasion were studied including the message, source,
receiver, and channel effects (Bettinghaus & Cody, 1994). A shortcoming in the message
learning approach to persuasion was that it did not take into account an active evaluation
and elaboration of the elements in the message by the receiver. Later theories of
persuasion included the concept of self-persuasion.
One of the important self-persuasion theories of attitude formation and change is
the Elaboration Likelihood Model (ELM). Petty and Wegener (1999) provided a
description of the ELM for persuasion. The model depicts two possible routes, central or
peripheral, to the development of judgments that are evaluative (e.g., positive or
negative) or non-evaluative (e.g., likelihood). The central route involves more effortful
cognitive processing of the persuasive message source and arguments in an attempt to
understand the validity of the message. The peripheral route to attitude change, on the
other hand, requires less reflective effort. An important concept in the ELM is the
elaboration continuum that ranges from one extreme where motivation and ability to
process a message are high and result in central processing to the other extreme where the
interest and ability to analyze a message are low leading to peripheral processing. ELM

10
acknowledges that individuals strive to obtain attitudes that are correct at least on a
subjective level. Persons can hold biased attitudes, but they rarely intend to be biased
(Petty & Wegener, 1999).
Attitude assessment and change depends on a large number of variables, however,
that exert their influence in different ways depending upon where an individual falls
along the elaboration continuum (Petty & Wegener, 1999). External variables such as
message source, strength of arguments, and distractions interact with internal variables
such as personal relevance, knowledge, mood, and need for cognition. For example, a
person who is distracted and in a hurry may accept an advertisement’s message simply
because the spokesperson is attractive and is assumed to be intelligent. On the other hand,
a person who is attentive and interested in comparing products might evaluate the
attributes of the attractive spokesperson more carefully and discover that the person has
never actually used the product placing some doubt on the message. Petty and Wegener
argue that although the process of attitude change is complex the ELM provides a
framework for predicting how variables will affect attitude change.
A number of studies have applied the ELM to advertising. These studies tended to
focus on particular hypotheses derived from the theory. For example, Haugtvedt, Petty,
and Cacioppo (1992) examined the influence of the need for cognition (Cacioppo &
Petty, 1982) on the evaluation of products depicted in advertisements. Need for cognition
is a personality-based motivational factor that influences the degree of elaboration on
messages. Those high in a need for cognition enjoy thinking about messages more than
those low in a need for cognition. Haugtvedt, Petty, and Cacioppo (1992) conducted three

11
experiments among undergraduate students and found that those high in a need for
cognition were more influenced by the strength of the arguments in the advertisement
than those low in a need for cognition, and those low in a need for cognition were more
influenced by a peripheral cue (i.e., the attractiveness of the models in the ads) than were
those high in a need for cognition. The authors caution that individuals low in a need for
cognition may be influenced to evaluate messages more carefully if the messages are
more personally relevant or their personal responsibility is increased, and persons high in
a need for cognition are likely to have limited elaboration of messages when the personal
relevance is very low.
In another study of the effects of personality characteristics on persuasion,
Wheeler, Petty, and Bizer (2005) found that argument strength moderates the association
between matching-favorability and attitude change. When a message is matched to a
person’s concept of their typical social behavior and personality (self-schemata), stronger
arguments in the message were more likely to influence attitude change, whereas
matching does not improve persuasion if the arguments are specious. These results were
consistent with the ELM prediction that matching messages to self-schemata will
increase elaboration and improve the persuasive appeal of a message with strong
arguments.
In another example, Schumann, Petty, and Clemons (1990) conducted two studies
on advertising among undergraduate students and found that the ELM successfully
predicted how the effects of variations in advertising content over time would influence
consumer attitudes. The researchers found that minor cosmetic variations in

12
advertisements improved the influence of advertisements on attitudes among those
participants with a lower motivation to process the message (i.e., lower product
relevance). Substantive variations in the arguments presented in advertisements, on the
other hand, were more influential among those participants with a higher motivation to
process the message.
A few models of persuasion have taken a slightly different approach to some of
the concepts developed in the ELM. The Cognitive Resources Model (CRM: Mazzocco
& Brock, 2006) of persuasion considers limitations in attentional resources in a manner
similar to elaboration likelihood model except that the CRM draws upon cognitive
research on the central executive in the Baddeley and Hitch (1974) model of working
memory. The use of the working memory model allows more detailed predictions about
the effects of visual imagery in advertising, for example. The limited resources attributed
to the central executive and the two slave systems, the articulatory loop and the
visuospatial sketchpad, help determine how particular messages may be processed and
how the processing may influence attitude change.
Chen and Chaiken (1999) describe a dual-process theory of attitude change or
judgment formation that is called the Heuristic-Systematic Model (HSM). This model has
a few similarities to the ELM model described above. The central and peripheral modes
of processing in the ELM model are analogous to the systematic and heuristic modes of
information processing, respectively, in the HSM model. Systematic processing is
effortful and requires ability and information, whereas heuristic processing uses
judgmental rules or heuristics stored in memory and requires minimal cognitive effort.

13
The HSM model proposes that a sufficiency principle determines when one or both
systems are used. This principle states that a person will apply the least amount of
cognitive effort necessary to reach a desired level of confidence in their judgment, and
that level of confidence is determined by their level of motivation. Example motivations
include a desire for accuracy of the judgment, defense of a personal belief, or impression
management in a social situation. Multiple motives may be active at any one time, and a
person may be conscious or unconscious of motives (and/or processing modes) that are
influencing a particular judgment. According to Chen and Chaiken, the HSM is different
from the ELM in that the ELM assumes that central processing predominates in situations
of high motivation and ability and peripheral systems function at low levels of motivation
and ability. The HSM, on the other hand, acknowledges that systematic and heuristic
processing can co-occur at any level of motivation and ability.
Austin and colleagues (Austin & Meili, 1994; Austin, Roberts, & Nass, 1990)
developed the Message Interpretation Process (MIP) model of decision-making about
media messages among children and adolescents. The model was influenced by theories
of social cognition (Bandura, 1986), persuasion (Petty & Cacioppo, 1986), decision-
making (Elias, Branden-Muller, & Sayette, 1991), and expectancies (Goldman, Brown, &
Christiansen, 1987). The model predicts a cascading effect of media and social influences
operating in both logical and affective processes. It is proposed that children are capable
of logically comparing images from the media to perception of their environment to
obtain a sense of the realism of the commercials, and following that analysis, children
determine if the images are similar to their experiences at home or with friends.

14
Desirability of the portrayals on the ads is an affective reaction to commercials that
combines with similarity to produce identification by the child with the persons in the
ads. Identification leads to expectancies about alcohol outcomes, and these expectancies
guide behavioral intentions to use alcohol. Austin and her colleagues have found some
support for various aspects of the model (e.g., Austin, Chen, & Grube, 2006), and more
details on their studies are provided below in the section on observational studies.
Theories of persuasion in media communications. Kenney and Scott (2003)
review the status of research in persuasive imagery using methods of visual rhetoric. The
authors first provide justification for the inclusion of visual elements in the study of
rhetoric, which in some traditions has only been applied to verbal language. Arguments
by Aristotle and by Kenneth Burke are presented to demonstrate the importance of
imagery in persuasion. It is argued that even something as mundane as the design of a
spoon carries rhetorical impact. Highly crafted spoons of silver and gold are, of course,
designed to convey a sense of elegance and grandeur in place settings. Similar arguments
are made for the architectural design of banks that convey a sense of stability and
security. Images found in advertisements naturally fall under the domain of visual
rhetoric. The authors indicate that there have been many studies that have analyzed the
rhetorical content of advertisements, and these studies typically have applied one of three
general methods of rhetorical analysis. The classical approach uses the five cannons of
rhetoric including invention, arrangement, style, delivery, and memory. These cannons
were originally concerned with the quality of a speaker but they have been adapted to
study images. Three additional elements may include the logic of the argument, the

15
emotion created in the audience, and the believability of the source. A second rhetorical
method of analysis is called Burkean rhetoric, and it largely is concerned with the motive
behind symbolic communications. Humans are primarily story-telling beings and these
stories have motives behind them. For example, commercial messages may convey brief
slice-of-life stories that provide information that viewers can use to improve their lives.
In this approach to rhetoric, myth is seen to help convey the story being communicated.
The final rhetorical method discussed is called critical analysis. In this case, critical
analysis means that there is a political perspective to the analysis (e.g., Marxist), and the
analysis is undertaken to expose aspects of the power structure within a society. The
authors of this chapter, Kenney and Scott (2003), say that few of these studies in rhetoric
have undertaken the task of evaluating consumer responses to the image or object of
study. Exceptions include McQuarrie and Mick (1999) and Phillips (2003) who study
consumer responses to visual persuasion in advertising.
Mere exposure to visual images has been shown to influence affective response to
the images (Zajonc, 1968). Nordhielm (2003) proposes a level-of-processing model for
the effects of repeated exposure to advertisements. At a higher level of processing, a two
factor model appears to accurately predict that familiarity and an associated positive
affect toward a product occur with repeated exposure to the same commercial, but after a
higher number of repetitions, tedium appears to result in the occurrence of negative
affect. This mere exposure effect is often referred to as the inverted U shaped response to
repeated exposure to an image or commercial. At a low level of processing, however, it
appears that increasing perceptual fluency with repeated exposure leads to a

16
misattribution of the fluency effect (e.g., to a liking of the advertisement) that increases
monotonically. The author provides some empirical evidence for the proposed
moderation of the level of processing on repetition effects (also see Nordhielm, 2002).
An important concept in advertising is brand image, which can influence
consumer behavior. Boush and Jones (2006) provide a brief overview of the
development, content, and the strategic functions of brand images. The authors define
brand image as the overlapping, accumulated associations, both objective and subjective,
with a brand name held across individuals including consumers, retailers, and
competitors. The content of a brand image may include product associations (e.g.,
attributes and benefits), source associations concerning the product
manufacturer/supplier, and consumer associations with a particular lifestyle or self-
expression. The authors describe potential cognitive processes responsible for brand
imaging functions including recall and recognition in memory processes, attitude
formation and evaluative processing, and decision-making processes by consumers and
competitors Two possible models for the way images are represented in memory include
network models of spreading activation (A. M. Collins & Loftus, 1975) and constructive
models of memory frameworks or schemas. Boush and Jones (2006) describe a number
of strategic functions for a brand image in the market place: (a) positive brand images
facilitate sales through new product introductions, product extensions, or through product
alliances, (b) brand image influences the perception of product quality, and (c) brand
image can be leveraged to influence intermediaries like retailers who must decide which
products to stock and how much self-space to allot to each product. The authors propose

17
a strategic model of brand image which includes brand image structure/content and
information processes as mediators between the sources of brand image (e.g.,
advertisements, direct experience) and the strategic functions of brand image. For the
current dissertation proposal, it may be useful to think of alcohol advertising as
developing a brand image for alcohol among young adolescents.
Theoretical Models of Alcohol Advertising Effects
A number of theories have influenced research on the effects of alcohol
advertising on the population as a whole and on adolescents in particular. Research in this
area has been concerned with two important aspects of advertising and alcohol use: (a)
the causal relationship between advertising and consumption and (b) the causal
relationship between advertising, consumption, and alcohol-related problems. The results
of studies in this area could have an important impact on policy related to alcohol
advertising and, as a result, on public health and the alcohol industry. Theory is an
important tool for the design and interpretation of this research. The following discussion
provides some background information for a few of the more influential theories such as
the single distribution theory, social learning theory, and dual-process theories of
persuasion.
The single distribution theory of alcohol consumption as outlined by Schmidt and
Popham (1978) was based upon observations of empirical data collected across countries
with different levels of alcohol consumption on a per capita basis. The theory included
three general propositions: (1) an increase in the mean level of alcohol consumption in a
population will be closely associated with an increase in the number of heavy users of

18
alcohol, (2) a rise in the number of heavy users will result in an increase in alcohol-
related morbidity and mortality, and (3) prevention programs should consider
implementing measures that influence overall consumption in order to control the
prevalence of alcohol-related problems. If alcohol advertising is shown to cause an
increase in consumption, then it would follow from the single distribution theory that
banning alcohol commercials should reduce alcohol-related morbidity and mortality. This
single distribution theory of consumptions is contrasted with a bimodal theory of
consumption. The bimodal theory postulates that the consumptive pattern of alcoholics is
fundamentally different than normal drinkers and there is no close association between an
overall consumption level and the number of heavy drinkers (Fisher, 1993). One
limitation of the single distribution theory is that it does not address the unique issues
surrounding vulnerable populations such as underage adolescents. The theory does
highlight, however, the importance of distinguishing use and abuse of alcohol.
A second influential theory applied to studies of alcohol advertising is Social
Learning Theory (Bandura, 1977). This theory is based in part on studies by Bandura and
his associates on aggression and observational learning. For example, Bandura (1965)
found that preschool children would model the aggressive behaviors of adult male models
on film if the models received either positive reinforcement or no feedback for their
behavior, but the children would not model the behavior if the models received
punishment after their aggressive behavior. Bandura, Ross, and Ross (1963) found that
observational learning and performance of aggressive behaviors occurred at nearly twice
the frequency when preschool children observed models depicting aggression live, on

19
film, or as cartoon characters compared to children that were not exposed to aggressive
behaviors by models. These and other studies suggested that children were learning new
behaviors vicariously and the modeled behavior did not have to be rewarded to influence
the behavior of the children. These results suggested that behavior was not strictly shaped
by environmental stimuli and consequences, but in addition, there appeared to be
cognitive mediators of learning and performance.
Social learning theory as describe by Bandura (1977) attempts to explain human
behavior in terms of interactions among cognitive, behavioral, and environmental
influences. People are influenced by their environment and they influence their
environment. Key elements for human psychological functioning include vicarious,
symbolic, and self-regulatory processes. The thoughts, emotions, and behaviors of
individuals are strongly influenced by observation of others resulting in vicarious
experiences and learning. Cognitive manipulation of symbols to represent events allows
individuals to think about alternative behaviors in anticipation of acting in a future
situation. Self-regulatory processes such as self-generated inducements and consequences
occur in concert with the influence of environmental stimuli so that an individual may
guide personal behavior.
Observational learning, which is a central aspect of Social Learning theory, has
implications for the study of alcohol advertising. While Bandura acknowledged the
influence of more mechanistic types of learning such as classical and operant
conditioning, he suggested that these means of “trial and error” learning were too
inefficient and time consuming to account for all of the learning accomplished by humans

20
(Bandura, 1977). According to social learning theory, the process of learning new
behaviors by observation of models is mediated by attention, retention or memory, motor
reproduction, and motivation. The mediating processes determine in part whether an
observed behavior is learned and performed. Social and economic factors might also
influence the performance of observed behaviors. An adolescent may attend to and retain
the positive messages in an advertisement about drinking, but social pressures from
family members might limit motivation or a lack of money might limit access to alcohol
and prevent the adolescent from drinking.
Another important aspect of social learning theory is the concept that individuals
develop anticipated reinforcements for a particular behavior through observational
learning (Bandura, 1977). These anticipated outcomes, sometimes called outcome
expectancies, develop as a result of observing a close association or correlation between
an observed behavior and an outcome. For example, models in commercials appear to be
drinking alcohol and having fun. This learned association may cause an adolescent to
expect to have fun when drinking. As noted in the earlier section on risk factors, alcohol
expectancies have been association with drinking and alcohol-related problems (e.g.,
D'Amico et al., 2002; Killen et al., 1996). Social learning theory also predicts that
innovations diffuse through populations via social contacts and observational modeling
(Bandura, 1977). This suggests that alcohol expectations may be learned or strengthened
if the adolescent observes parents or peers drinking and having fun. It is important,
therefore, to determine the relative contributions of advertising versus more proximal

21
social influences in the development of anticipated outcomes among adolescent for
drinking alcohol.
One of the first researchers to adapt the social learning theory to the study of
alcohol advertising was Donald E. Strickland (Strickland, 1982, 1983). He was interested
in exploring the link between alcohol advertising and alcohol consumption and problems.
Strickland (1983) hypothesized that more exposure to alcohol advertising would result in
an increase in consumption of alcohol, and increased alcohol consumption would lead to
more alcohol-related problems (see Figure 1-1). Although the model in the figure does
not show any variables inherently tied to social learning, Strickland contended that the
effects of ad exposure result in vicarious learning of normative expectations and
identification with models (not shown), which lead to more consumption. Studies that
have tested this model and other models of the effects of advertising on alcohol
consumption will be discussed in the following section.
.

Figure 1-1 Strickland Model (Strickland, 1983).
Advertising
Exposure
Controls
Age, Gender,
Ethnicity,
TV viewing
Consumption Alcohol
Abuse

22

Empirical Studies on Alcohol Advertising
Review articles. Smart (1988) provided a brief review of some of the early
research on the effects of alcohol advertising. The types of studies reviewed included
those on advertising bans, exposure to advertising, econometric studies, and experimental
studies. The author discussed the theory behind the studies and highlighted problems
associated with the designs of the various studies. Tentative conclusions by the author
suggested that there was little evidence for a relationship between advertising and alcohol
consumption and the effect was small for the few studies showing an association.
Fisher (1993) provides an excellent review of studies related to advertising and
alcohol consumption. Of particular interest to the author was the relationship between
advertising and alcohol abuse. If there was a clear relationship, the author felt that some
regulation of alcohol advertising would be indicated. In general, however, the studies
reviewed were mixed on this issue, and the author was unable to provide a clear
indication that advertising influences alcohol abuse. The author concluded the following:
(1) there was no experimental evidence showing an association between alcohol
advertisements and an increase in consumption, (2) survey data in observational designs
showed a consistent yet small effect (1-4%) of advertising on consumption that is usually
not statistically significant, (3) econometric studies have shown a relationship between
brand advertising expenditures and an increase in brand sales, but there was no apparent
increase in overall alcohol consumption with increased advertising, and (4) bans and
regulatory restrictions did not appear to affect overall consumption of alcohol. The
author provided a useful approach to reviewing the issues. He examined the following

23
areas: (a) theoretical basis for media influences, (b) content analysis of media and
advertising, (b) experimental and observational studies, and (d) macroeconomic studies.
Fisher (1999) updated his earlier review (Fisher, 1993) and came to a similar conclusion
that alcohol advertising appears to have no more than a small effect on consumption (see
also Atkin, 1995; J. D. Brown & McDonald, 1995).
In a review of econometric studies, Saffer (2002) concluded that more recent
cross-sectional econometric studies have shown a small but positive relationship between
alcohol advertising expenditures and underage alcohol consumption. Earlier studies that
reported null findings were subject to methodological problems. Those studies used
yearly, national aggregate expenditure data, and Saffer suggested that this high level of
aggregation lacked the variance needed to observe effects and the aggregate expenditure
levels were in the upper, flat portion of the advertising response function (i.e., the area of
diminishing returns) making it difficult to observe expenditure effects. Taking cross-
sections of marketing regions that vary in expenditures per capita has been more
successful at demonstrating the relationship between advertising and consumption. In
addition, studies on advertising bans have shown a relationship although the results are
mixed. It is difficult to reduce advertising through regulation, according to the author,
because advertisers shift marketing dollars to media outlets that are not regulated, and it
would be difficult to impose a complete ban on all advertising in all media. Saffer also
reviewed counter-alcohol use advertising campaigns that have been effective, and
concluded that use of counter-advertisements would be more effective than regulation of
alcohol advertising in reducing underage drinking.

24
Content analysis. Grube (1993) reviewed literature on portrayals of alcohol use on
TV shows and content analysis of alcohol ads. The frequency of ads in prime time
comedies and dramas is roughly 0.25 per hour whereas alcohol ads on sports
programming ranges from 1.48 per hour for college sports to 5.68 for major professional
sports. Content analysis shows that ads link drinking with desirable attributes such as
sociability, elegance, and physical attractiveness, and with positive outcomes such as
success, relaxation, romance, and adventure.
Econometric studies. Gius (1995) conducted an econometric study of brand sales
for distilled spirits. The researchers modeled sales for 16 brands over 14 years (1976 -
1989) in the United States. The results indicated that the per capita consumption of a
brand was significantly and positively related to advertising expenditures for that brand,
but the advertising expenditures for rival brands was not a significant predictor. Control
variables in the model included brand price, rival brand price, median income, time trend
of consumption, and lagged consumption. The author concluded that the evidence was
consistent with the theory that advertising has a significant effect on demand primarily
through brand switching and not by raising the level of consumption across all brands.
An econometric study on motor vehicle fatalities showed that the price of alcohol
and, to a lesser degree, alcohol advertising were positively related to fatalities (Saffer,
1997). This relationship was true for all ages and for 18 to 20 year olds when
demographic variables were included in the model. The author concluded that limiting
advertising (e.g., via media bans) or raising the price of alcohol (e.g., through higher
taxes) would save lives.

25
In a recent econometric study, Saffer and Dave (2006) found that alcohol
advertising had a modest but significant effect on the prevalence of alcohol use (past year
and past month) and on binging (past month) among adolescents in two large, nationwide
data sets. Self-report survey data on a range of items including alcohol use were obtained
from the Monitoring the Future project (N=63,000) and from the National Longitudinal
Survey of Youth (N=10,000). Aggregate alcohol advertising expenditures and price data
from 75 metropolitan areas were appended to the survey data for the adolescents either by
region or by state depending upon the residence information provided by the data set. The
results for multiple regression on past year use of alcohol, past month use, and binging in
the past month were similar across the two data sets. Alcohol advertising had a small but
positive effect on alcohol use, and the price index for alcohol had a negative effect on
alcohol use. Blacks participated less than whites in alcohol use. Females were more
sensitive to alcohol advertising and price than males. One important finding indicated
that controlling for individual variables such as family relationships and working status
increases the observed alcohol advertising effects. The authors simulated the effects of a
policy change on alcohol use and reported that a 28% reduction in alcohol advertising
across all media would result in a drop in alcohol 30-day prevalence from 25% to
between 21% and 24%. The prevalence of binging would drop from 12% to between 8%
and 11%. The authors suggest that a complete ban would reduce underage drinking even
further.
Based in part on information published in the literature and in national surveys,
Hollingworth et al. (2006) estimated the savings in lives that would occur if alcohol

26
advertising were banned. Deaths related to alcohol consumption including injuries,
homicides, alcohol poisoning, cirrhosis, and hemorrhagic stroke were predicted over a 60
year period for a cohort of 4 million persons aged 20 in the year 2000. Hollingworth et al.
predicted that a nationwide ban on advertising would save 7,610 lives, which was 16.4%
of the total alcohol-related deaths for the cohort. The study found that an additional 1,490
(2.7%) lives would be saved if the tax on alcohol sales was raised by the equivalent of
$1.00 per six pack of beer.
Observational studies. Strickland (1983) found that there was a small association
between exposure to alcohol advertising and alcohol consumption in a cross-sectional
study among 772 current users of alcohol in the 7
th
, 9
th
, and 11
th
grades, but there was
neither a direct or indirect association (through consumption) between advertising
exposure and problems related to alcohol abuse. In contrast, there was a much stronger
association between social influences (e.g. associating with peers that approve of using
alcohol) and both alcohol consumption and alcohol abuse. Strickland concluded that
advertising has a relatively small effect if any affect on alcohol abuse, and a ban on
alcohol advertising was not warranted based upon the his study.
Atkin, Hocking, & Block (1984) found that exposure to alcohol advertisements on
TV and in magazines was positively related to liquor, beer, and to a lesser extend to wine
consumption in a cross-sectional study of 665 students in 7
th
through 12
th
grade.
Intentions to drink in the future also were associated with exposure to alcohol
advertisements.

27
A cross-sectional study by Aitken et al. (1988) of 433 youth between the ages of
10 and 17 showed that drinkers were better able to correctly identify alcohol commercials
and appreciated alcohol advertising more than non-drinkers even after controlling for
variables known to be associated with underage drinking (i.e. age, gender, social class,
number of friends that drink, and parents’ attitude toward underage drinking).
Adlaf and Kohn (1989) used structural equation modeling to analyze cross-
sectional data collected by Strickland (1983) from 772 adolescents. Results showed that
the influence of exposure to alcohol advertisements on alcohol intoxication and abuse
was mediated by alcohol consumption. Advertising exposure had a significant direct
effect on alcohol use, but associating with peers that used alcohol had a stronger effect. In
addition, increased viewing of TV was directly associated with more exposure to alcohol
ads, but TV viewing also had an inverse effect on alcohol consumption. The authors
suggest that a ban on alcohol ads would have a minimal effect on reducing underage
drinking.
Austin and Meili (1994) examined how 154 at-risk youth 10 to 11 years old
interpreted TV ads in relation to their home life and perceived social norms. Alcohol ads
on TV had an influence on the development of alcohol outcome expectancies and
intentions to drink, but the results also suggest that the youth are actively using both
logical and emotional approaches to evaluate and compare the messages in the ads to
other information available to them. Parents and family appear to have a strong influence
on how the messages in alcohol ads are interpreted.

28
Grube and Wallack (1994) conducted a cross-sectional study among 468 children
between the ages of 10 and 14 years old (M=11.9). Alcohol advertising measures
included TV viewing at various times during the week and types of programs (e.g.,
sports), awareness of beer advertising (recognition of still photographs), and knowledge
of beer brands (listing brands and matching brands to slogans). A structural equation
model based upon information processing concepts was used to test the association
between alcohol advertising and intentions to drink as an adult. The results showed that
children who had a better awareness of alcohol commercials also had positive alcohol
beliefs and intended to drink more often as an adult (see also Grube, 1995).
Unger, Johnson & Rohrbach (1995) surveyed 386 students in the 8
th
grade in
southern California in a cross-sectional study of the recognition and liking of cigarette
and alcohol advertisements on television. The study found that cigarette advertisements
appealed to susceptible nonsmokers just as much as to smokers, and both of these groups
liked cigarette ads more than non-susceptible nonsmokers. For alcohol advertisements,
non-susceptible nondrinkers liked alcohol ads the least, susceptible nondrinkers second,
and drinkers liked ads the most. The results suggested that advertisements might
influence susceptible adolescents to start drinking, not just influence drinkers to be loyal
to or change brands. This seemed to be especially true for cigarette ads and susceptible
nonsmokers.
Bloom, Hogan, and Blazing (1997) administered a questionnaire via mail to
adolescents (mean age = 15.3) in a blue collar neighborhood in a southern state. The
researchers attempted to obtain a sample of participants with similar demographic

29
backgrounds to control certain confounders in the relationship between tobacco and
alcohol promotion at sporting events and use of those products. Although this was a
cross-sectional study with limited ability to make causal inferences, there was an
association between consumption or the intention to consume tobacco and watching of
NASCAR racing, watching of college basketball and pro football on television. Beer
usage was associated with watching college basketball on television and college football
in-person. These associations were observed after adjusting for age, gender, hours
working each week, and friends and parents usage. Friends usage was the strongest
predictor for both tobacco and alcohol outcome variables. The authors concluded that
there appeared to be some association between promotions at sporting events and on
television and the consumption or intention to consume tobacco and alcohol.
Wyllie, Zhang, & Casswell (1998b) conducted a cross-sectional study among 500
young persons aged 10 to 17 years in New Zealand. A structural equation model showed
significant paths leading from (a) liking of alcohol ads on TV, (b) frequency of current
drinking, (c) peer behavior, and (d) age to the self-reported level of drinking expected at
age 20. In addition, there were significant paths from (a) liking of ads, (b) peer behavior,
and (c) parental approval of alcohol to the current level of drinking. There was not a
significant reciprocal path from current drinking to liking of alcohol ads. The authors
concluded that the results were consistent with their hypothesis that the liking of alcohol
commercials was associated with both an increased level of current drinking and a higher
level of drinking expected in the future.

30
Austin, Chen, & Grube (2006) tested the Message Interpretation Process (MIP)
model of how individuals internalize persuasive messages using processing strategies
based upon both logic and emotion. Refer to the section above on persuasion for a
description of the MIP model. Using a structural equation model in a cross-sectional
study among 651 youths 9 to 16 years old, the researchers found that the influence of
exposure to alcohol advertising on alcohol use was mediated by (a) skepticism about ads,
(b) desirability and identification with persons depicted in the ads, (c) positive and
negative expectancies, and (d) liking of beer brands and beer-themed items. Measures of
TV exposure included total TV viewing, sports viewing, and primetime sitcom/drama
viewing. There were no measures of advertising recall. These results showed that
exposure to alcohol advertisements influenced the decisions of individuals to drink, and
these decisions were based in part on logical and in part on affect-based processes.
Interpretation processes such as skepticism, desirability, and identification mediated the
effects of TV viewing on alcohol expectancies, brand and merchandise liking, and
alcohol use. The main point made by the authors appeared to be that it was important to
consider mediating variables when examining the effects of exposure to alcohol
commercials on under-aged drinking. Exposure might have weak direct effects but
important indirect effects.
Expectancy studies. As discussed briefly in the section on risk factors, alcohol
expectancies are associated with the consumption of alcohol among young persons.
Several researchers have suggested that alcohol expectancies may mediate the influence
of alcohol advertisements on underage drinking, but studies in this area have produced

31
mixed results. Lipsitz, Brake, Vincent, & Winters (1993) found no association between
exposure to alcohol commercials and alcohol expectancies in two laboratory studies, one
with 5th and one with 6th grader students. Groups of participants were shown 1 of 3
videos 25 minutes in length: (a) 5 alcohol and 35 filler commercials, (b) 5 soda and 35
filler commercials, or (c) 5 alcohol, 2 anti-drinking, and 33 filler commercials.
Participants then completed the AEQ-A and several memory tasks to determine recall of
the alcohol commercials. Contrary to these early results, Dunn and Yniguez (1999) found
in a cross-sectional study among 551 students in the 4
th
and 5
th
grades that participants in
an experimental group exposed to alcohol commercials were more likely to associate
positive and arousing effects with drinking than were those in the control group exposed
to soft drink commercials. Fleming, Thorson, and Atkin (2004) found in a cross-sectional
study that advertising exposure influenced attitudes and perceptions of alcohol, and these
perceptions in turn influenced alcohol expectancies and intentions to drink as adults
among 15 to 20 year old youth in a nationwide telephone survey. Among 21 to 29 year
old young adults surveyed, however, attitudes influenced by advertising did not predict
alcohol expectancies although expectancies did predict alcohol consumption. In a
longitudinal study, however, Martino et al. (2006) found that among 1410 students that
alcohol advertising exposure in 8
th
grade was positively associated with perceptions of
alcohol’s positivity and potency in 9
th
grade using bivariate analyses but not in
multivariate analyses where peer and adult influences were included in the models. This
last study suggested that proximal social influences were better predictors of alcohol
expectancies than exposure to alcohol commercials.

32
Longitudinal Studies. Connolly, Casswell, Zhang and Silva (1994) conducted a
prospective study on the influence of alcohol in the mass media in New Zealand on
drinking by adolescents. The participants (n=667) were interviewed at the ages of 13, 15,
and 18 years old. Measures included the following: (a) commercial advertisements
recalled from television, radio, magazines, newspapers, and films, (b) moderation
messages recalled, (c) alcohol-use portrayals recalled from entertainment programs, (d)
hours of TV watched per week, (e) perceived approval of alcohol use by peers, and (f)
demographic variables. Multiple regression results showed that out of 6 alcohol-related
variables measured at ages 13 and 15 among males only the number of advertisements
recalled at age 15 was predictive of beer consumption at age 18. For females, portrayals
of alcohol recalled at age 13 was negatively associated the maximum amount of beer
consumed at age 18, and the number of ads recalled at age 13 was positively associated
with the frequency of drinking beer at age 18. Among the control variables, occupation
for males was a significant predictor of beer consumption, and hours of watching TV,
occupation, and living situation for females were significant predictors. Regression
results for the consumption of wine and spirits showed that no alcohol-related media
variables measured at ages 13 or 15 were significant predictors of alcohol consumption
by males or females at age 18. The authors concluded that beer advertising observed by
adolescent males had a significant influence on their drinking behavior at age 18. The
influence of advertising was not as clear for females.
Casswell and Zhang (1998) reported on another longitudinal study that examined
alcohol advertising in New Zealand. The researchers administered questionnaires and

33
interviewed participants at the ages of 18 and 21 years old. Of the 921 surveyed, only the
participants who indicated that they drank beer at age 18 were included in the study
(n=630). Measures included the following: (a) liking for advertising, (b) brand allegiance
across the two ages, (c) consumption of beer, (d) self-reported alcohol-related aggression,
and (e) demographic variables. The data fit a hypothesized, structural equation model
very well. There were significant paths from liking advertisements and brand allegiance
at age 18 to beer consumption at ages 18 and 21, and there were significant paths from
consumption at these ages to aggression at age 21. Gender was also significant in the
model such that men liked advertisements more and were more likely to drink larger
quantities of beer in both age groups. The authors concluded that 18 year-olds who
expressed a liking for beer advertisements and brand allegiance were more likely to drink
larger quantities of beer at age 21 even after adjusting for beer consumption at age 18. In
addition, the 21 year-olds who consumed more beer were more likely to experience
alcohol-related aggression.
Ellickson, Collins, Hambarsoomians, and McCaffrey (2005) reported on a 3-year
prospective study of the effects of alcohol advertising measured among 3111 students in
the 7
th
and in the 8
th
grade on drinking behaviors among those students in the 9
th
grade.
Exposure to advertising at beer concession stands, but not to TV or magazine ads, was
predictive of 9
th
grade drinking for those who were non-drinkers in 7
th
grade after
adjusting for control variables. Exposure to advertising in magazines and at beer
concessions stands, but not TV or in-store displays, were predictive of drinking in 9
th

grade for those who were drinkers in 7
th
grade after adjustment for control variables. The

34
authors concluded that some sources of alcohol advertising may influence adolescent
drinking depending upon prior experience with alcohol by the adolescent.
Stacy, Zogg, Unger, and Dent (2004) examined the predictive effects of a range of
alcohol advertising exposure measures in a prospective study among young adolescents.
Students in public schools were surveyed in the 7
th
grade (n=2998) and then again one
year later (n=2250). This data is a subset of the data that is the focus of the current
research. Exposure measures collected by Stacy et al. included the following: (a) watched
TV shows index, (b) watched TV sports index, (c) self-reported frequency of observing
alcohol ads, (d) cued-recall memory test, and (e) draw-an-event memory test. Outcome
measures included current use of beer or wine/liquor (past 30-days), binge drinking (3 or
more drinks per occasion), and prior alcohol use (past 6 months). The alcohol use
variables were dichotomized due to a strong skew toward zero, and the affects of alcohol
advertising exposure measures on the three outcomes were modeled using logistic
regression. Odds ratios were presented for the models before and after adjusting for prior
alcohol use, frequency of general TV viewing, participation in team sports, perception of
alcohol use by friends, perceived approval by peers of alcohol use, intentions to use
alcohol, perceptions of alcohol use by adults, gender, ethnicity, and school attended.
Results showed that one opportunity-based measure, the watched shows exposure index,
was predictive of each of the 3 alcohol use measures even after adjusting for the potential
confounders. Another opportunity-based measure, the watched TV sports index,
predicted beer use but not wine/liquor use or binging. One memory-based measure, the
self-reported frequency measure, predicted beer use but not wine/liquor use or binging.

35
The other memory-based measures, cued recall and draw-an-event, were not predictive of
alcohol use. There were no significant interactions of the exposure measures with gender,
ethnicity, or prior alcohol use. The authors concluded that exposure to alcohol
commercials was associated with an increased risk of alcohol use and beer consumption
in particular.
Zogg (2004) reported on a prospective study of 1097 students attending 7
th

through 9
th
grades at public schools in Southern California. Note that this data is a subset
of the data that is the subject of the current dissertation research. The study reported by
Zogg included assessments of student exposure to alcohol advertising, consumption of
alcohol, alcohol-related problems, and a number of control variables. Exposure to
advertising included self-reports on frequency of seeing alcohol advertisements on TV,
frequency of watching sports events on TV, and a checklist of frequently watched TV
shows. The data were fit to a modified version of the Strickland (1983) model. In partial
support of the model, results showed that alcohol consumption in 8
th
grade mediated the
relationship between self-reported exposure to ads in 7
th
grade and alcohol-related
problems reported in 9
th
grade. The effects of ad exposure were small, which was
consistent with the findings of Strickland. Contrary to the model, however, Zogg reported
that peer influences on problems were not mediated by consumption. The study modified
the Strickland model by adding an indirect measure of spontaneous alcohol-related
associations as a proxy for implicit cognitions toward alcohol. Seventh grade scores for
this measure of associations significantly predicted alcohol-related problems in 9
th
grade.
Overall, this study provided some additional support for earlier findings that exposure to

36
alcohol has small but significant effects on alcohol consumption and problems associated
with drinking.
In another longitudinal study, Snyder et al. (2006) used both self-reported
exposure to alcohol advertising and industry advertising expenditures in 24 US markets
to show that drinking among adolescents and young adults (aged 15 to 26 years) is
influenced by alcohol advertising. The researchers controlled for age, gender, ethnicity,
school attendance status, and alcohol sales per capita. Although there was a considerable
amount of attrition (68%), the researchers were able to use all data from each participant
across the four years of the study in hierarchical linear models with repeated measures.
The results indicated that a person who saw 1 more advertisement per month was
predicted to consume 1% more drinks per month. For each additional dollar spent per
capita on alcohol advertising, persons in that market were predicted to consume 3% more
drinks per month on average. The results were similar for the sub-sample of youth under
the age of 21. The strengths of this research is the large number of participants recruited
across different markets in the USA , the objective measure of advertising expenditures
coupled with self-reports of advertising exposure, and the longitudinal data collection.
In summary, both cross-sectional and prospective studies have shown a small but
consistent influence of alcohol advertising on adolescent drinking. A key element of the
studies is the measurement of exposure to alcohol advertising.
Assessment of Exposure to Alcohol Advertisements
This section discusses some of the general issues surrounding the question of how
to measure exposure to advertising and some of the more specific studies related to the

37
validity of certain measures. Advertising may have both short-term effects on sales and
longer-term effects on brand identity and loyalty. It is difficult to measure longer-term
effects related to brand development, but it seems worthwhile to briefly review the issues.
The development of brand associations in memory over time is related to one of the aims
of the current study, which is to understand the development of associations (especially
alcohol associations) in memory over time. The current study is also concerned, however,
with reliability and validity of shorter-term measures of advertising that may predict
increased consumption of products.
Wright-Isak, Faber, and Horner (1997) distinguish between measurement of the
shorter-term effects of advertising and the longer-term effectiveness of advertising.
Short-term effects studied in academia often focus on specific elements in commercials
and how viewers are influenced by these elements. Short-term measures that are used by
marketing managers to justify their budgets often include sales, brand perceptions, and
advertising attitudes and awareness. These short-term measures may range in timing from
a few minutes to a year after an advertising campaign. The authors argue, however, that
advertising effectiveness should be measured over longer periods of 10 years or more. In
the longer term, advertising should begin to associate a brand with a particular image
(e.g., Maytag is a reliable brand of appliances), and this association should create positive
beliefs about the brand that will affect purchasing decisions. The measure of the
effectiveness of advertising should be a measure of the strength of this association for the
brand of interest in comparison to the association strength of its competitors. Although
there is a strong desire to justify advertising budgets in the short-term, the authors

38
recommend that both academia and industry begin to focus more on the longer-term,
prospective effects of advertising in the development of brand images.
Cook and Kover (1997) discuss some other issues surrounding the measurement
of advertising effectiveness. Academic researchers tend to focus on general theory,
affective responses, attitude change, and decision-making. On the other hand, increasing
pressure to show results for advertising expenditures has caused advertising agency
researchers and marketing managers to focus on sales and limit research on attitude
change. The authors argue that this divergence in research on advertising effectiveness
has created divergent meanings for common terms such as effectiveness, and the
differences in language and reward structures between academicians and practitioners
needs to be addressed, according to the authors, to improve communication and
collaboration.
Shapiro, Heckler, and MacInnis (1997) presented results that showed some
support for the hypothesis that advertisements do not need to be attended to consciously
to produce positive ad evaluations. The researchers recruited 48 undergraduates to
participate in a task that required them to read a computer-based magazine. As they were
reading the magazine, the participants were required to manipulate a cursor in such a way
that the cursor would not hit a word as the article scrolled down the screen. Target and
filler graphic ads were displayed to the side of the main text column for the experimental
groups but not for the control groups. One half of the experimental participants were told
to attend to the ads (attentive group) and the other half received no instructions
concerning the ads (pre-attentive group). Attention to the ads in the attentive group was

39
documented by an increase in the number of errors in controlling the cursor that occurred
for that group relative to the pre-attentive and control groups when ads were displayed. A
degraded picture paradigm was used to measure differential priming for the ads among
the groups, but there were no significant differences among the groups. However, the
authors questioned the validity of the assessment. Measures of ad evaluation and brand
attitude showed a significantly higher evaluation for the pre-attentive relative to the
control group. In a post hoc analysis of intervening variables, the researchers found an
interaction between left hemisphere activation and group in the prediction of ad and
brand evaluations. Those pre-attentive and attentive participants with a high level of
activation were more likely to give a positive evaluation of the ad than the control group.
According to the authors, this was consistent with the concept of matching activation,
which predicts that processing resources in one hemisphere will increase when the
processing resources of the opposite hemisphere are increased. In other words, increased
use of the left hemisphere to read the magazine apparently increased the resources
available in the right hemisphere to process the ad images.
Haugtvedt and Priester (1997) used the Elaboration Likelihood Model (ELM) of
persuasion and its associated research to argue for assessment of attitude strength as a
measure of advertising effectiveness. The authors presented evidence that an attitude can
be developed by either central or peripheral routes of processing, and the route of
processing determines the strength or persistence of the attitude over time. The authors
noted that repetition of an advertisement especially with minor variations will lead to
central route processing due to the development of multiple associations in memory. The

40
strength of an attitude may be measured as it decays over time, or it may be measured as
a resistance to counter-arguments. According to the authors, it was the strength of the
argument that was a better predictor of consumer behavior, which makes it an important
measure of advertising effectiveness.
Stewart and Furse (1985) used cued recall, message comprehension, and
persuasion as performance measures for the relative effectiveness of executional features
used in advertisements made for television. The study used aggregate scores from more
than 300 participants for each one of 1059 commercials across 356 brands, 115 product
categories, and 62 advertising firms. The advertisements were coded for 155 mechanical
features that loaded on 24 factors (e.g., setting, cast, emotional tone, product benefits, and
brand sign-off). The cued-recall measure occurred 72 hours after viewing the
commercial. The brand and product were used as cues, and the participant was asked to
recall some unique aspect of the commercial. For key message comprehension,
participants were asked to recall a specific key message. Persuasion was measured as the
change in brand choice from before to after exposure to the commercial. Multiple
regression results showed that executional factors explained a significant but modest
amount of variance in cued-recall (12%), message comprehension (6%), and persuasion
(4%). Factors related to gaining attention or facilitating encoding of memories (e.g.,
humor and brand differentiating message) were most predictive of cued-recall. Brand
differentiating message was the strongest predictor of message comprehension and
persuasion. The performance measures used in this study were typical of copy-testing

41
measures used by advertising research firms, but there was no link to sales as a measure
of consumer behavior.
In another article examining difficulties associated with measurement, Stewart
(1989) reviews studies in advertising research and suggests that the use of absolute
measures of advertising effectiveness can produce erroneous conclusions. He argues that
it is important to consider the characteristics of the measure, the shape of the advertising
response function, and where a particular brand is located on that function. Measures
based upon memory of advertisements (e.g., recognition, cued-recall) can be influenced
by environmental factors such as prior exposure and pre-existing product knowledge and
by related cognitive factors such as proactive and retroactive interference. Attitude or
persuasion is measured as a change in belief-based or global evaluations of the product
from before to after exposure to the advertisement and the strength of the new attitude
depends upon whether the change occurred via central or peripheral route processing
according to ELM. Action measures capture brand choice, store visits, and other
consumer behaviors, which have complex relationships to sales. Relative as opposed to
absolute measures are more likely to capture the effectiveness of advertising according to
Stewart. Absolute measures assume a replacement model of learning where old
information is forgotten or replaced by new information. In an accumulation model,
however, the old information is retained and the probability of retrieving the new
information increases with exposure or practice. Stewart cites several empirical studies of
advertising that support the accumulation model for the advertising response function. In
this case, relative measures of effectiveness are needed to capture the strength of

42
consumer response to a target product compared to competitors’ products. One other
critical factor in determining the effectiveness of advertising is to determine where a
particular product is on the response curve. If an established product is in the area of
diminishing returns on the curve, using absolute measures of effectiveness for additional
advertising might be misleading because there is little room for the response to change. It
might be possible, however, to assess advertising effectiveness by ceasing advertising to
examine the decay function. Stewart concludes that it is important to consider multiple
factors when trying to establish the effectiveness of advertising.
In a review of research on advertising effects, Gibson (1983) presented evidence
that recall measures were not nearly as useful as attitude measures in determining the
effectiveness of commercials. Although the overall test-retest reliability of recall
measures was reasonably high (range .67 to .87), Gibson presented studies that showed
when product categories were taken into account the weighted average test-retest values
were as low as .29. In addition, a number of studies showed that the variation in the recall
scores due to the advertisement was no larger than the variation due to extraneous factors
such as demographics, exposure context, and measurement variables including
interviewer. In addition, Gibson found little support in the literature for an association
between measures of recall and persuasion where persuasion was measured as a positive
evaluation of the brand. The correlations between recall and persuasion ranged from
virtually none to .32, and there were considerable differences between the items that were
recalled from an advertisement and the positive evaluations of the brand. The reported
correlations between recall and sales were as low as .03, but the correlation between

43
persuasion and sales was much better in the area of .42. Gibson concluded that measures
of persuasion are better than recall for determining the effectiveness of advertising.
In contrast to Gibson (1983), Dubow (1994) concluded that a recall measure can
be a useful tool when combined with other tools in assessing the effectiveness of
advertising. The poor reliability and validity reported by Gibson (1983) where largely
based upon an out-of-date recall methodology and inappropriate assumptions about the
form of the relationship between recall and sales. Gibson reported that recall test-retest
reliability was dependent as much on consumer characteristics, exposure context, and test
methods as upon the actual recall of the consumer, which suggested poor reliability.
Dubow reports that most of the data reported by Gibson was collected using the now
defunct Burke DAR (day after recall) method, which did not control for the confounding
variables as is common in newer methods such as the Gallup’s and Robinson’s In-View
system and the ASI cable-based system. Newer recall methods appear to exhibit better
levels of reliability according Dubow. The results for 105 studies from two advertising
research services showed no significant difference between test and retest scores. A poor
correlation between recall scores and sales or persuasion was sited as evidence for poor
reliability by Gibson, but Dubow contends that the previously assumed linear relationship
is actually curvilinear. It appears that above a certain threshold, an increase in recall
scores does not result in an increase in sales. In one example study, the variance
explained increases from 28% for a linear model to 36% when the curvilinear
relationship is modeled. Dubow concludes that although recall is not by itself a sufficient

44
measure it appears to be a reliable and valid measure of advertising effectiveness that
may be useful especially when used with other measures.
Haley and Baldinger (1991) reported on the validity of copy research measures of
television commercials. The study that was sponsored by the Advertising Research
Foundation collected 5 pairs of commercials for which the sales histories were known
and for which there was a significant difference in performance between the commercials
within the pairs. Products in the ads were packaged goods in the food and health-and-
beauty-aids categories. Copy testing normally used to predict the success of ads were
administered after the sales histories were known in order to reduce the overall cost of the
study. Three off-air and 3 on-air methods were tested across the 5 pairs for a total of 30
cells. There were between 400 and 500 persons surveyed within each cell for a total of
between 12,000 and 15,000 participants in the study. Measures administered included the
following: (a) persuasion indicated by brand choice and purchase interest, (b) salience or
top-of-mind awareness, (c) recall with various types of cues, (d) communication of key
points by the ad, (e) liking of the advertisement, (f) diagnostics such as ‘I learned a lot’
and ‘Tells me a lot about how the product works,’ and (g) negative impact due poor
product performance or lack of useful information. Results showed that copy testing can
be used to predict sales. Each type of measure tested had at least one indicator that would
be useful. The best performing measures included liking of the ad (300% better than
chance prediction), tells me how the product works (234%), brand recall from a category
cue (234%), recall of the main point in the ad (200%), overall brand rating (184%), top-
of-mind awareness (167%), and whether the ad is boring (234%). The use of multiple

45
measures improved the percentage of correct classifications from 73.3% to 93.3% when 6
of the best measures were used simultaneously. Other tentative findings indicated that
off-air methods were as good as or slightly better than on-air methods, post test only was
better than pre/post test methods, single versus multiple exposure methods were similar,
scoring with the mean was generally similar to using the top-box scores (e.g., sum of the
top two positive response categories). The tentative methodological findings might be
attributable to the design and procedures used in the current study. Haley and Baldinger
concluded that copy testing did work and that many of the measures currently used in the
advertising research industry were useful and valid in predicting sales.
Recognition tasks have been used in research on dual processes in cognition.
Seamon et al. (1995) used forced recognition and affective liking tasks to demonstrate the
dissociation of implicit and explicit memories in the mere exposure paradigm.
Participants viewed a series of stimuli for brief periods of time during the study phase
and then they were asked in the test phase to select one of two stimuli in a forced choice
recognition task (explicit memory) or in an affective preference task (implicit memory).
Implicit memory occurred as predicted by the mere exposure hypothesis for stimuli
shown for shorter durations at study whereas explicit memory occurred for stimuli shown
for longer durations.
In a related study of exposure time (or attention) to advertisements, the
effectiveness of advertisements was dependent upon a number of contextual factors that
should be considered in measurement schemes. Thorson and Zhao (1997) showed that a
measure of the amount of time that participants viewed a television screen (EOS: eyes on

46
screen) was a useful measure of attention to commercials. In a naturalistic study among
200 participants ranging in age from 18 to 54 years, EOS significantly predicted
recognition, recall, and attitude toward commercials imbedded in a situation comedy and
a mystery program. The study also demonstrated that creative quality of the commercial
(i.e., relevance, originality, and impact) influenced attention. In addition, the study
identified certain barriers to attention including the location of the break in the program,
location within the break, and location of information within the commercial. The authors
conclude that the effectiveness of a commercial is not just a function of the advertisement
itself. This is, of course, consistent with the ELM theory of persuasion.
Krishnan and Shapiro (1996) reported that the use of both direct and indirect
measures can be useful in determining the effectiveness of commercials. Direct measures
ask participants to consciously search their memories for information related to
advertising (e.g., recognition and recall measures). Indirect measures do not instruct
participants to search their memories. In the study reported by the authors, the stem
completion task required the respondents to complete a word stem with the first word that
came to mind, and the choice intention task required respondents to select one of two
brand names presented as the one they would be most likely to choose for purchase. The
stem completion and choice intention responses were found to be primed by the
advertising study phase of the experiments. The difference between the procedures is
intended to distinguish between explicit and implicit memories, respectively (Schacter,
1987). The studies reported by Krishnan and Shapiro demonstrated dissociations between
the measurement types that might have implications for decisions by brand managers.

47
Brand names that are words frequently found in usage were better remembered implicitly
when compared to low frequency words. This is important for impulse buying (e.g.,
chewing gum) where consumers do not expend effort to consider alternative brands. On
the other hand, products that require conscious consideration (e.g., cars) should have low
frequency words as brand names because it is easier for the consumer to recall these
words. The use of both types of measures can be useful when testing commercials
according to the authors. Indirect tests are more likely to correctly assess the effects of
advertising than direct tests if the level of processing is low for a particular product (e.g.,
soap) where peripheral cues are the primary influence on consumers. Direct tests are
better for high-involvement products.
In summary, studies have shown an association between copy testing measures
such as cued recall, recognition, and liking of advertising and purchase intention or sales
of products (e.g., Haley & Baldinger, 1991) despite the complex nature of the
marketplace and persuasion. Several theories and studies suggest that the influence of
advertisements may operate at a preconscious level as well as at a conscious, deliberative
level (e.g., Krishnan & Shapiro, 1996; Petty & Wegener, 1999).
Indirect Measurement of Associations in Memory
A number of studies have demonstrated the important influence of spontaneous
associative memory processes on judgment and behavior (Damasio, 2003, pp. 149;
Kahneman, 2003; Schacter, 1987). The current research, however, will address a different
aspect these associations. As outlined briefly in the Specific Aims, the current study will
examine the development of these associations over time as a function of life experiences

48
including exposure to alcohol advertisements. A number of theories predict the
development of automatic or implicit associations in memory, but few studies have
attempted to demonstrate the prospective development of these associations. One
difficulty in attempting a study of this type is the measurement of automatic associations,
which are likely to be unavailable to self-reflection (Greenwald, McGhee, & Schwartz,
1998; Schacter, 1987). Word association measures can be constructed as indirect
measures and have been used successfully to study implicit or automatic associations in
basic research in cognitive science (McEvoy, Marschark, & Nelson, 1999; Nelson,
McEvoy, & Dennis, 2000) and in applied studies of health behavior (Ames et al., 2007;
Stacy, 1997; Szalay, Inn, & Doherty, 1996). The current study will use word association
measures to study the development of automatic memory processes.
A brief history of word association. The study of the association of ideas has a
long history. Aristotle developed the concepts of contiguity, similarity, and sequence that
were later refined by the British empiricists (Davison, Vogel, & Coffman, 1997).
Associationist and connectionist models developed in the last century follow some of the
concepts originally proposed by Aristotle. Word association is used to study the
association of ideas for both associative and connectionist models. Francis Galton is
credited with publishing one of the first studies that used word association. Galton was
primarily interested in the study of evolution and heredity, but he also studied the
association of thoughts and ideas (Boring, 1950; Crovitz, 1970). Galton recorded his own
associations to words in an introspective experiment that was published in 1879 (Crovitz,
1970; Thorne & Henley, 2001). Carl Jung was influenced by Galton’s work and

49
developed a technique to study psychological complexes using word associations
(Boring, 1950; Jung, 1910, 1918/1969). Galton’s work also influenced Wundt and his
students including James Catell who published an extensive word association experiment
on German and British subjects in 1889 (Boring, 1950; Cattell & Bryant, 1889). Freud
might have been influenced in part by Galton when Freud developed his method of free
association for psychoanalysis in 1892 (Boring, 1950; Thorne & Henley, 2001). Free
association as used by Freud invites the patient to talk freely about personal experiences
or dreams, but Freud’s definition of free association does not necessarily include word
association (Freud, 1925/1995; Thorne & Henley, 2001).
In contrast to psychoanalysis, the term free association refers to one of several
types of word association tasks used in studies of cognition and verbal behavior. In free
word association, the participant is asked to respond to a cue word or phrase with the first
word that first comes to the participant’s mind (Woodworth, 1921). A continuous free
word association task requests that the participant write a number of responses to the
same cue. A third type of word association task is called controlled word association, and
for this task, the participant is asked to respond to a cue with a word or words from a
particular category such as naming animals that are mammals (Woodworth, 1921).
Free word association norms are established by collecting free word associations
from a number of participants (e.g., 100 to 150 or more participants). The frequency of
each unique response to a cue word is calculated for all of the participants, and the
relative frequency is the strength of the association between the cue and the response. For
example, the stimulus word ‘bud’ from the Nelson, McEvoy, and Schreiber (2003) norms

50
has the following normative response frequencies: beer=0.39, flower=0.15, roses=0.11,
light=0.08, etc. This means that 39% of the participants responded with the word ‘beer’
to the stimulus word ‘bud’ in a free word association task. A number of free word
association norms have been published (e.g., Hirsh & Tree, 2001; Jenkins, 1970; Nelson,
McEvoy, & Schreiber, 2003; Nelson, McEvoy, Walling, & Wheeler, 1980; Palermo &
Jenkins, 1964; Rosenzweig & Miller, 1966)
Both behaviorist and cognitive psychologists have used word association. Skinner
and others studied verbal behavior using word association norms (e.g., S. W. Cook &
Skinner, 1939; Skinner, 1937; Thorndike, 1932). Experimental psychologists later
expanded upon the study of word associations to propose cognitive models based upon
the associations of ideas (e.g., Cramer, 1968; Deese, 1959a, 1965; Pollio, 1966). For
example, Deese (1965) proposed that the interconnectivity of ideas in free word
associations was important in the understanding of associative meaning. In addition,
education and learning was the subject of research from a cognitive perspective in the
1950s and 1960s (e.g., Cramer, 1968; Entwisle, 1966; Postman & Keppel, 1970).
Contemporary research using word associations has drawn upon the work of these
earlier experimenters. Roediger and his colleagues studied illusory memory (Roediger &
McDermott, 1995; Watson, Balota, & Roediger, 2003) based upon the work of Deese
(1959a) who showed that memory intrusions in free recall were predicted by associative
frequencies in word association norms. In addition, contemporary theories concerning
associations (e.g., Nelson, McEvoy, Janczura, & Xu, 1993; Roediger & McDermott,
1995) were influenced by Underwood’s concept of implicit associative response (IAR:

51
Underwood, 1965; Underwood & Reichardt, 1975). Underwood used word association
norms to demonstrate that strong associates were more likely than weak associates of
words in a list to be falsely identified as duplicates in list of words (Underwood, 1965).
This work influenced Nelson, McEvoy, and their associates to develop a predictive model
for cued recall and recognition (Nelson, McKinney, Gee, & Janczura, 1998). The
predictive model utilized the associative strengths found in a large database of word
association norms that they collected during the 1990s (Nelson, McEvoy, & Schreiber,
2003). In addition, Szalay, Deese and their associates were influenced by earlier
researchers such as Noble (1952) to study psychological meaning across cultures using
continuous word association tasks (e.g., Szalay & Brent, 1965; Szalay & Deese, 1978).
As can be seen in this brief historical overview, the use of word association in the
study of memory has continued since Galton’s original work in 1879. In the more recent
studies of cognition (e.g., the false memory paradigm), word association has been used to
investigate dual-process theories of memory and decision-making.
Word association and implicit cognitive processes. A number of empirical
findings show that word association can measure relatively spontaneous, automatic, or
implicit cognitive processes (for review, see Stacy, Ames, & Grenard, 2006). First,
amnesic patients who are severely limited in conscious or explicit recall are successful at
recall when indirect priming (implicit memory associations) techniques are employed
(e.g., Levy, Stark, & Squire, 2004). Second, participants self-report that during word
association tasks they do not use intentional retrieval (explicit memory) processes
(Mulligan, 1998: experiment 2). Third, dissociations between explicit memory tasks (e.g.,

52
cued recall) and word association have been successfully demonstrated (e.g., Goshen-
Gottstein & Kempinsky, 2001). Finally, studies have shown that conceptual priming, not
just perceptual priming, occurs in word association. In one study for example, word
association responses in a priming task depended upon the semantic context of the target
words presented during the study phase (Zeelenberg, Pecher, Shiffrin, & Raaijmakers,
2003), and in a second study, a verb generation task (controlled word association) was
used by Segar, Rabin, Desmond and Gabrieli (1999) to demonstrate that implicit priming
occurred at a conceptual level across languages. These studies show that word association
tasks measure an associative processing system directly implicated in implicit cognitive
processes in comparison to more deliberative, rational, or explicit processes.
Reports on the Reliability of Word Association Tasks. Researchers have estimated
the reliability of word association tasks using various indices. Szlay and Deese (1978)
estimated the test-retest stability of a continuous word association task. Over a period of
one week, the authors found that 61% of the responses given by a group of participants as
first associates in the baseline task were repeated as an associate in some position in the
one-week follow up task (48% of the second associates were repeated, 40% of the third
associates, etc.). Rozin, Kurzer, & Cohen (2002) also examined the test-retest stability of
continuous word association to food related cues (3 responses per cue) and found that
74% of the respondents repeated at least one baseline response in a two-month follow-up
task. Hoz, Champagne, and Klopfer (1992) reported that the test-retest reliability for a
continuous word association measure of student’s knowledge of biology was poor over a
two-month period (r=0.18). These authors compared group level responses in 5x5

53
matrices that included the five stimulus words crossed with the five top associates for
each cue word. However, Preece (1978) used a different technique to show that word
association was a reliable measure over a three-year period (ages 12 to 14) of students’
knowledge of physics (r=0.82). He compared the number of responses at the individual
level at each time period that was related to the stimulus through fundamental equations
in physics. Although comparing word association norms is not a test of reliability, it is
interesting to note that Stacy, Leigh, & Weingardt (1993) reported that word association
norms compared favorably across different samples of participants. The authors reported
that association norms collected on homographs from participants (N=1003) in the
Northwest correlated well (r=0.87 to 0.96) with norms collected from participants in
other parts of the country (Nelson et al., 1980; Wollen, 1980). Except for these few tests
of reliability, little has been published on the reliability of word association tasks.
Development of word associations in memory. A number of theories suggest how
associations develop in memory including association among experiences or concepts.
Three approaches in cognitive science to modeling associations include spreading
activation (J. R. Anderson, 1983), multiple trace theory (Hintzman, 1984, 1986), and the
Hopfield network (Hopfield & Tank, 1986; Masson, 1995).
Anderson (1983) provides a description of a version of the Adaptive Control of
Thought (ACT) model of spreading activation in memory (see also J. R. Anderson et al.,
2004). Memory is composed of cognitive units that consist of a unit node and a set of
elements. The number of elements in a cognitive unit is limited to roughly 5 each, and an
example of a cognitive unit might be a paired associate or a single sentence. A cognitive

54
unit is created when a transient copy of the unit is placed in working memory based upon
an external or internal event. ACT assumes a constant probability over many
manipulations that the unit in working memory will be transferred into a long-term
memory trace. This probability does not vary with intention, motivation, or duration of
residence in working memory. This means that each occurrence of the unit in working
memory increases the likelihood that the unit is encoded as a trace in long-term memory.
The strength of the trace, once it is established, is increased with each additional
occurrence of the unit in working memory. Established traces are not lost, but the
strength of a trace will decay over time according to an inverse power function. Retrieval
of a trace from long-term memory depends upon a spreading activation of cognitive units
in working memory through other units in long-term memory. This process makes
information available to working memory through associative relatedness, and the ACT
model predicts retrieval times and probabilities based upon associative strengths between
nodes and non-linear processes (i.e., feedback to cognitive nodes affects retrieval). Key
elements of the ACT model predict that multiple experiences increase the strength of
memory traces and the likelihood of recalling a trace depends upon the associative
strength of that trace with other traces in memory.
D. L. Hintzman described a multiple trace model of memory (Hintzman, 1984,
1986). In this model, an experience is made up of primitive units (e.g., features of a name
and a face), and each experience is encoded as a collection of these primitive units in a
single memory trace. Concepts and abstractions are conceived of as collections of similar
traces in memory, which are retrieved in a way that integrates the individual traces.

55
During the retrieval process, a memory probe in the active, primary memory store
interacts with all traces simultaneously in the dormant, secondary memory. The probe
contains features of primitive units that either match or fail to match features in the
secondary memory traces. The activation level of an echo returned by a memory probe
from the secondary memory depends both upon how well the probe matches each
memory trace and upon the number of traces with matching features. The summing of
matching features over a number of specific memory traces provides the experience of
generalizations or abstractions. The key features of this model related to acquiring and
retrieving memories are that each experience creates a new trace in memory and that the
strength of memory activation depends in part on the number traces with similar features
(i.e., more experiences lead to stronger memory activations).
Hopfield and Tank (1986) described a model for optimization computations that
was based upon a distributed network of neuron-like elements with symmetrical
connections. According to the authors, optimization computations can be applied to
problems of perception, behavioral choice, and motor control. For example, optimization
is needed to match a word printed on a page to the semantic meaning in memory that it
most resembles. The association is determined by the most stable state reached in the
activation of the network. Learning is accomplished through repeated activations of the
network. Masson (1995) described a distributed memory model with features similar to a
Hopfield network (Hopfield & Tank, 1986), which Masson used to model semantic
priming. The similarity in the patterns of activation of the prime and target result in the
priming effect. Two concepts (e.g., beer, fun) are “associated” in memory to the degree

56
that they share similar patterns of activation in memory. The pattern of activation across
the network of processing units represents knowledge about a concept, and this pattern of
activation is influenced by past experiences with the concept.
These models suggest that associations among concepts as represented by
associations among words in memory should develop over time as experiences
accumulate. Word associations would then be expected to change across historical time
as the meanings or popularity of words change in the culture and across the development
of children as they age. There is evidence that historical changes do occur. Jenkins and
Palermo (1965) examined 7 word association norms collected between 1910 and 1960
and found that associations change gradually over time. For example, the percentage of
identical responses with those obtained in 1910 were 81%, 79%, 74%, 66%, 58%, and
64% in 1925, 1927, 1933, 1942, 1952 and 1960, respectively. The strongest associates in
1910 were the most stable across the time periods. Seventy-one of the most common
responses to 100 cue words were the same in 1927 and in 1960. The popular, more
frequent responses increased while superordinate (i.e., more general class name)
responses decreased.
There is also evidence for a change in word associations among children as they
develop. Palermo and Jenkins (1964) reported word association norms for 200 words
collected in a cross-sectional study of the development of associations among students in
grades 4, 5, 6, 7, 8, 10, 12, and college. An examination of the response frequencies
showed that for some cue words there is little change in the responses across those ages,
but responses for other cue words do change across the grade levels. For example, the

57
response ‘drink’ to the cue word ‘whiskey’ remains somewhat steady across the age
groups at roughly 60% whereas the response ‘wine’ declined gradually from 10% among
4
th
and 5
th
graders to 1% among college students. Palermo (1971) observed
developmental changes in word association norms collected in a cross-sectional study of
1
st
through 4
th
grade students. The authors reported that the trends observed in these
younger participants matched trends observed previously in data collected from 4
th

through college level participants (Palermo & Jenkins, 1964). The frequency of the most
popular responses increased from the 1
st
grade through college as did contrast and
paradigmatic (e.g., noun response to a noun cue) responding. Female participants
appeared to respond more often with popular responses than males. Superordinate
responses increased until 6
th
grade and then declined through the college level. The
authors concluded that there are developmental changes occurring as children age and
that these changes in word association norms should be taken into account when
conducting studies influenced by memory associations such as paired associate learning,
associative generalizations, clustering in recall, and verbal discrimination. Some of these
changes in language usage may be associated with cognitive development (Palermo &
Molfese, 1972).
In a cross-sectional analysis of age differences in associative memory, Coronges,
Valente, and Stacy (2007) found a difference between 7
th
grade and college age students
using social network analysis techniques on free-word association responses. The authors
compared network measures across the groups for associations to 16 ambiguous cue
words (e.g., count, draft, and field). College students provided more unique responses

58
than the younger students, and according to the authors, this suggests that the older
students had more associations due to the accumulation of more life experiences. The
network for the college students was more centralized, however, indicating that the
responses that were given more frequently by college students played a stronger role as
‘hubs’ in associative memory processes and provided a more efficient organization of
concepts in memory. The authors conclude that these differences and others observed
between the groups of students provide evidence that the structure of associations in
memory develop with age and life experiences, and that individual differences in this
development are likely.
It is clear that cognitive scientists and those interested in verbal behavior have
been interested in how associations develop in memory including word associations.
Most of this research has been conducted on word association norms among the general
population or among sub-groups by age, gender, or ethnicity. Little research has been
conducted, however, concerning individual differences in associations in memory. One
exception is the use of word association in the study of spontaneous processes in
substance use and abuse.
Word association and research on substance use. Free word associations have
been used to study drug use behaviors. The continuous association task was used by
Szalay and associates to measure the psychological meanings surrounding drug use
(Szalay, Strohl, & Doherty, 1999). For example, these authors compared word
associations by students on a campus (n=354 students) where the level of alcohol
consumption was high to word associations by students at number of other campuses

59
(n=21,991 students) where the alcohol consumption was more moderate. The students at
the high use campus had higher vulnerability scores (number of word association
responses that align with frequent alcohol users) than the control campuses, and in fact,
even students that did not drink alcohol showed higher vulnerability scores at the high
use campus. These results appeared to show that the social environment on campus can
affect the vulnerability of all students on that campus to alcohol consumption (see also,
Szalay et al., 1996).
Stacy, Leigh, and Weingardt (1994) compiled a list of the most common positive
outcomes for alcohol use generated by two groups of college students. The authors asked
another group of college students to write down the first word that came to mind when
reading each outcome. The authors found that the number of alcohol related responses to
the positive outcomes was significantly associated with the self-reported frequency of
alcohol use after adjusting for gender, age, and ethnicity. The results were consistent
with theory that suggests positive drinking outcomes are based on culturally derived
associations and then strengthened as an individual gains more experience with drinking
alcohol (Stacy et al., 1994). Stacy (1995) presented evidence that memory associations
could mediate between cues to drug use and drug use behavior. In a cross-sectional study
with college students, Stacy (1995) showed that free associations to ambiguous cues
related to alcohol were significantly predicted by frequency of alcohol use after
controlling for gender, acculturation, and alcohol use by friends or parents.
A number of other studies have shown that memory associations as measured
with word association tasks are predictive of drug use and related outcome measures.

60
Word association measures have been found to predict alcohol and marijuana use (Ames
& Stacy, 1998; Stacy, Ames, Sussman, & Dent, 1996; Stacy, Leigh, & Weingardt, 1997),
driving under the influence mediated by marijuana use (Ames, Zogg, & Stacy, 2002),
alcohol related problems mediated by alcohol use (Stacy & Newcomb, 1998). In addition,
Stacy (1997) found that memory association predicted alcohol and marijuana use
prospectively after adjusting for prior drug use. These studies and others suggest that
word association tasks have some criterion validity in the prediction of substance use.
In a rare study of the predictors of the development of word associations, Stacy,
Leigh, and Weingardt (1997) found that prior behavior predicted responses on a word
association test. In a cross-sectional study among 1,003 undergraduates, the self-reported
number of days using alcohol, studying chemistry or physics, or using computers in the
prior 30 days significantly predicted whether the participants responded to ambiguous
word cues with a word related to one of the target categories. The authors concluded that
these results were consistent in general with a number of associative theories of memory
in social and cognitive science (see sections above). Further investigation of those factors
which influence the development of associations on an individual level seems warranted.
The current dissertation research investigated the influence of televised alcohol ads on the
development of alcohol-related associations in memory.

61
Chapter 2 Measurement Model for Marketing Research Measures Used to Appraise
Alcohol Advertisements on Television.
Abstract
The influence of televised alcohol advertisements on adolescents is of practical interest,
but few studies have examined the properties of measures designed to assess this
influence. The current study examines the properties of some common measures of media
exposure and provides an example of how to use confirmatory factor analysis (CFA) in
this endeavor. Measures of advertising exposure including cued recall, self-reported
exposure, liking of alcohol ads, and propensity to watch TV were assessed among 2,986
adolescent students. These measures loaded well on single latent factors as expected, and
the factor loadings and thresholds were invariant across gender. Further, the study
provides a useful example of the steps to follow when developing a measurement model
that can be used to test hypothesized relationships among latent factors in a structural
model.

Introduction
The alcoholic beverage industry spent an estimated 1 billion dollars for televised
advertising (CAMY, 2006) on sales of 142 billion (USDA, 2007) in 2005. Clearly, the
industry spends a significant amount of money on advertising, but there is still much to
learn about how that advertising is able to influence the purchase decisions of consumers
(e.g., Tellis, 2005). An understanding of this influence is important not only for the
advertisers, but it is also important from a public health standpoint. The alcoholic

62
beverage industry may not intentionally target underage youth, but there is growing
evidence that exposure to alcohol ads is one of the factors influencing underage drinking
(e.g., Ellickson et al., 2005; Snyder et al., 2006). Measurement of exposure and memory
for alcohol ads is a critical aspect of these studies, but few articles report more than
minimal measurement properties of these assessments or provide comprehensive
information on how to evaluate media-related measures before incorporating them into
predictive or explanatory models.
One popular method of modeling psychological processes is structural equation
modeling (SEM). This statistical methodology is capable of modeling associations among
latent constructs and accounting for measurement error. Although SEM is commonly
used now, there are few examples in the literature that describe comprehensively how to
prepare to test hypotheses using SEM in advertising and other media research. The
current study will utilize data collected from a research project on alcohol advertising to
demonstrate all of the steps necessary to prepare a measurement model using
confirmatory factor analyses (CFA) prior to fitting a structural model to the data. The
following text will briefly describe measures commonly used in alcohol advertising
research and some of the properties of those measures. Following that, some details will
be provided on the steps necessary to prepare a measurement model useful for media
research.
Although application of a minimal form of CFA has become standard practice for
some researchers, there are a number of critical issues that are rarely if ever addressed in
media assessment and other applied areas. These include, for example, decisions about

63
formative versus reflective indicators (see definition below), level of across-group
invariance, cross-validation, and appropriate analysis when some data are missing. This
article addresses these frequently ignored issues while providing a thorough illustration
of CFA analysis for common measures of advertising exposure. This work also
summarizes results that are important for future research on the measurement of exposure
to alcohol advertisements among adolescents.
Advertising exposure assessments
A range of measures have been used in observational studies on the influence of
televised alcohol advertising on sales and on underage drinking. Exposure to ads has been
measured variously as the self-reported time spent watching TV (Adlaf & Kohn, 1989;
Austin et al., 2006), self-reports on the time of day and type of programs viewed (Grube
& Wallack, 1994), self-reports on viewing specific programs for which the number of ads
per program was known (Ellickson et al., 2005; Strickland, 1983), self-reported
frequency of viewing alcohol ads (Stacy, Zogg et al., 2004), and regional data on
expenditures by marketers to broadcast alcohol ads on TV (Saffer & Dave, 2006).
Memory for alcohol ads has been measured using recall of ads (Connolly et al., 1994),
cued recall (Stacy, Zogg et al., 2004), recognition (Grube & Wallack, 1994; Unger et al.,
1995), and non-verbal, top of mind awareness (Stacy, Pearce, Zogg, Unger, & Dent,
2004). Reactions to ads have included measures of alcohol expectancies (Austin et al.,
2006; Fleming, Thorson, & Atkin, 2004), liking of ads (Casswell & Zhang, 1998; Unger
et al., 1995), brand allegiance (Casswell & Zhang, 1998), and skepticism or identification

64
with actors (Austin et al., 2006). Despite the use of these assessments in studies of
advertising, few researchers have provided information on the properties of the measures.
Two studies have provided some useful information on measures of advertising
effectiveness. Haley and Baldinger (1991) reported on the predictive validity of copy
research measures typically used by marketing departments to estimate the potential sales
impact of television commercials. This study collected 5 pairs of commercials for which
the sales histories were known and for which there was a significant difference in
performance between the commercials within the pairs. Products in the ads were
packaged goods in the food and health-and-beauty-aids categories. Copy tests, normally
used to predict the success of ads, were administered after the sales histories were
collected. This reduced the cost of the study in comparison to a standard prospective
study that would require a larger sample of advertisements to ensure finding some
variance in the outcomes (i.e., sales volumes). Three off-air and 3 on-air methods were
tested across the 5 pairs for a total of 30 cells. Between 400 and 500 persons were
surveyed within each cell for a total of between 12,000 and 15,000 participants in the
study. Results showed that copy testing can be used to predict sales. The best performing
measures included liking of the ad (300% better than chance prediction), tells me how the
product works (234%), brand recall from a category cue (234%), recall of the main point
in the ad (200%), overall brand rating (184%), top-of-mind awareness (167%), and
whether the ad is boring (negative 234%). Haley and Baldinger concluded that copy
testing worked and that many of the measures currently used in the advertising research
industry were useful and valid in predicting sales.

65
In a second major study of adverting, Lodish et al., (1995) conducted a meta-
analysis of advertising experiments conducted in the field. Contrary to the findings by
Haley and Baldinger (1991) described above, the meta-analysis found that recall and
recognition were poor measures of the sales effectiveness of ads. The measures that did
predict sales included the following: (a) changes in the brand/copy and strategy, (b)
introducing new brands or line extensions, and (c) the relative frequency of running ads
at the beginning or end of a marketing campaign. Although these findings are helpful to
brand managers, they have limited use in understanding how ads influence the behavior
of individuals or understanding the reliability of advertising measures. In addition,
neither of these two studies provides useful examples of how to examine the properties of
measures using the more versatile techniques available in CFA and SEM.
Measurement models, CFA, and SEM
Most psychometric researchers follow a two stage approach when applying SEM
to the study of psychological processes (J. C. Anderson & Gerbing, 1988; T. A. Brown,
2006). The first stage examines the properties of the proposed measures and associated
latent constructs in the study, and the second step tests the hypothesized relationships
among the latent constructs. In the first step, CFA techniques are used to determine
whether individual indicators load on their respective latent constructs as expected. In
addition, the various indicators and latent constructs are modeled simultaneously, which
provides the opportunity to examine the simple structure of the measurement model (T.
A. Brown, 2006). Tests for simple structure determine whether indicators load adequately
on their respective factors and whether there are any undesirable cross-loadings of

66
indicators on more than one latent construct. There are no causal relations among the
latent constructs tested in this first step, but the correlation among constructs is estimated
freely providing the opportunity to examine whether expected patterns of association are
observed. Measurement models that fit the data well should provide some assurance
regarding the internal consistency and validity of the measures although more rigorous
methods are available for this purpose when developing a new measure (T. A. Brown,
2006; Campbell & Fiske, 1959). In summary, the development of a measurement model
allows the researcher the opportunity to examine the properties of latent constructs and
their indicators prior to examining the causal relations among the constructs. This first
step is especially helpful when relatively complicated causal models are proposed for
which it might be difficult mathematically to fit the model to a particular data set.
The second stage in SEM procedures is to prepare a structural model by assigning
regression pathways among the latent variables according to hypothesized causal
relationships. This structural model is then fit to the data and the parameter estimates
with associated standard errors are used to infer whether the proposed relationships are
supported by the data. The focus of the current study is on measurement modeling using
CFA methods and does not discuss the details involved in this second step of SEM. The
reader is referred to other sources for examples of ways to test structural models. (e.g.,
Bollen, 1989; Duncan, Duncan, & Strycker, 2006; Marcoulides & Schumacker, 2001;
Raykov & Marcoulides, 2006).

67
Reflective and formative indicators in CFA
Prior to analyzing measures using CFA, it is important to determine how existing
indicators in a data set relate to latent constructs of interest. A latent construct or factor is
a theoretical or hypothesized variable that cannot be directly observed (e.g., depression)
but can be inferred by assessing certain behaviors, and these measures serve as indicators
of the latent construct (Raykov & Marcoulides, 2006). Indicators can be observed
behaviors or responses to questions. Bollen and Lennox (1991) compare reflective
(effect) and formative (causal) indicators with their respective latent variables
(constructs). Observed values for reflective indicators are assumed to be due to the level
of a latent variable, and a person’s response to a question is a reflection of that latent
variable. It is expected that an increase in the latent variable will result in an increase in
all of the observed reflective indicators, which should be correlated with each other. It is
also expected that removing one reflective indicator will not change the value of the
latent variable. That is, the effect indicators should be interchangeable assuming similar
reliability and measurement error. Given three or more indicators, the model for a
reflective latent variable is fully identified in CFA, which means that the parameters for
the indicators and latent variable can be estimated statistically.
For formative indicators, however, the latent variable is formed or ‘caused’ by the
indicators and the indicators need not be correlated with each other. An example of a
formative latent variable is stress measured by the life events checklist. The items on the
life events scale, such as ‘divorce’ or ‘loss of a job’, are assumed to be the cause of stress,
which is a latent variable. Stress as a latent variable does not cause someone to check the

68
item ‘loss of a job’ on the life events scale. All of the indicators on the checklist are not
expected to increase because one does, and each indicator is unique in that by removing
one indicator (e.g., death of a spouse) from the checklist, the nature of the formative
latent variable can change. In addition, the model for a causal latent variable is under-
identified in CFA no matter how many formative indicators are included in the model. In
order to estimate the parameters for a formative latent variable, the latent variable must
predict two reflective indicators (or two reflective latent factors) in addition to the
formative indicators that are predicting the latent variable (Bollen & Lennox, 1991).
Recently, several authors have addressed the importance of considering the nature of
indicators in SEM. Errors can occur in parameter estimates and goodness of fit indices if
formative indicators are modeled as reflective indicators (Jarvis, MacKenzie, &
Podsakoff, 2003; MacKenzie, Podsakoff, & Jarvis, 2005).
Howell, Breivik, and Wilcox (2007) provide a convincing argument that
formative indicators should not be used to define a causal latent construct. First, the
nominal meaning of a formative measure may not be the same as the empirical meaning
depending upon the structural relationships it has in a model. This may occur because a
formative measure is under-identified and the parameters cannot be estimated without
reference to some other variables. This conditional meaning can lead to interpretational
confounding where the nominal or intended meaning of the latent construct is changed
empirically by the variables added to the model for identification. This type of
confounding could be part of the reason for inconsistent findings across some studies of
advertising. Second, formative items are not required to be correlated or have the same

69
nomological net and therefore, do not necessarily have the same antecedents or
consequences. Finally, these two conditions make it questionable whether a single point
variable can accurately represent the items. Howell, Breivik, and Wilcox recommend
avoiding the use of formative measures in theory testing for these reasons. If formative
measures already exist in a data set, the authors recommend including each formative
item as a single indicator in the model. In certain cases, however, it might be possible to
form an index of the formative items where the index will not be used for theory testing
in a structural model (Arnett, Laverie, & Meiers, 2003; Diamantopoulos & Winklhofer,
2001). One special case for which an index of formative indicators may be suitable is
when the indicators have the same units and the sum of the indicators functions like
adding up individual distances to create a measure of total distance (Howell et al., 2007).
CFA and testing measurement invariance
CFA methods can be used to evaluate measures in the total sample of participants
from a population, or the methods can be used to study the properties of measures across
two or more groups within the sample. A study, such as the current one, might
hypothesize that the impact of televised alcohol advertising will differ by gender. In order
to test that hypothesis, it is important to determine whether the individual indicators and
measures in the self-report surveys carry the same meaning and weight among girls and
boys in the study. In principle, measurement invariance between the groups would
indicate that girls and boys with the same level of a latent construct would respond, with
equal probability, in the same way to all indicators on the measure of that construct (e.g.,
Borsboom, 2006; Gregorich, 2006; Teresi, 2006b; Vandenberg & Lance, 2000). Meredith

70
and Teresi (2006) describe degrees of invariance as follows: (a) pattern invariance that
requires equivalent factor loadings across groups, (b) strong factorial invariance has the
additional requirement that specific factor means represented as invariant intercepts
(thresholds) be constant among the groups, and (c) strict factorial invariance adds a third
requirement that residual variances are equal. The level of invariance necessary depends
upon the intended use of the measure (Borsboom, 2006; Gregorich, 2006). If the sum or
score calculated from the observed responses is used to determine employment,
admission to a program, or how resources are assigned, then strict factorial invariance
will help ensure fairness and equity. Strong factorial invariance is required in research
where differences in specific factor means or intercepts on items measuring a particular
factor may have important implications in understanding differences among groups.
Pattern invariance, however, may be sufficient when relationships among the latent
constructs for each group are the primary interest and differences in the magnitudes of
those relationships across groups is not the focus of the research. At a minimum, pattern
invariance is required to demonstrate that the measure has a similar meaning for each
group (Gregorich, 2006). In the current study, strict factorial invariance is desirable in
order to reliably compare the regression weights and factorial means in the hypothesized
structural model across gender.
There are a range of methods available to test for measurement invariance. Teresi
(2006a) provides an excellent review of the various methods that can be used to evaluate
invariance or differential item functioning. In particular, the factor analytic method for
multiple indicators and multiple causes (MIMIC) has several advantages in that it can

71
simultaneously model the item response and underlying trait or construct across groups,
model multiple traits, adjust for the impact of invariance, include covariates, analyze
dichotomous and polytomous responses, and is less affected by lack of purification (i.e.,
removal of non-invariant indicators). The disadvantages, however, include no inclusion
of a guessing parameter as in the Item Response Theory (IRT) methods, requirement of
large sample sizes to assess non-uniform invariance, and no direct participant estimate is
available. The authors provide similar evaluations of other methods including non-
parametric methods such as the Mantel-Haenszel approach and parametric methods such
as the IRT log-likelihood ratio test. Although each method has its own strengths and
weaknesses, the current study uses factor analytic methods that have been shown in
simulation studies to provide effective assessment of invariance (Meade &
Lautenschlager, 2004a, 2004b; Stark, Chernyshenko, & Drasgow, 2006). For this study,
the factorial method had the advantage of being able to assess multiple latent factors
simultaneously across gender groups. In addition, the large sample size in the current
study provided confidence that the factorial method would detect non-invariant items. A
number of authors have provided examples of using factorial methods for determining
measurement invariance (Byrne & Stewart, 2006; Gonzalez-Roma, Tomas, Ferreres, &
Hernandez, 2005; Gregorich, 2006; Meade & Lautenschlager, 2004b; Pentz & Chou,
1994; Steenkamp & Baumgartner, 1998)
In practice, it may be difficult to obtain complete measurement invariance where
all of the items in a measure are equivalent in factor loadings and intercepts across groups
(Byrne, Shavelson, & Muthen, 1989; Steenkamp & Baumgartner, 1998). It often becomes

72
necessary to evaluate the importance of partial measurement invariance on the study
hypotheses and interpretation of modeling results. This strict invariance is not necessary
in situations, such as the current study, where basic research is used to examine the
relative influence of alcohol advertising on girls versus boys. Studies have shown that
partial invariance, which occurs when some but not all of the items are invariant, does not
preclude the use of a measure in a structural model (Byrne et al., 1989; Millsap & Kwok,
2004; Steenkamp & Baumgartner, 1998). Byrne, Shavelson, & Muthen (1989) argued
that as long as one item (in addition to the indicator item for which the factor loading is
fixed at 1.00 to establish the factor measurement scale) was invariant across groups, the
assessment of invariance could continue, and the authors demonstrated the procedures for
assessing partial measurement invariance. In practice, only the intercepts and factor
loadings for invariant items are constrained across groups, and these constrained items
are the ones that will contribute to mean differences across groups in the model (Byrne et
al., 1989; Gregorich, 2006). In summary, structural model evaluation may proceed under
conditions of partial invariance where group level factor means (but not the raw score
means) are to be compared (Byrne et al., 1989; Gregorich, 2006; Steenkamp &
Baumgartner, 1998).
The current study
The central aim of the current study is to provide researchers with an example
showing each of the steps involved in the preparation of a measurement model using
CFA methods as the first stage in SEM. Previous articles have presented important details
for the various steps, but this study attempts to provide a comprehensive example.

73
Included in the example is an evaluation of measurement invariance across gender, which
can provide practical information for evaluation of invariance across other groups such as
treatment condition or ethnicity.
The development of a measurement model depends to some degree on the
hypotheses for the study that will ultimately be tested in a structural model. The specific
hypotheses for this study are shown below:
H1. Exposure to alcohol commercials, the liking of those commercials (affective
reaction), and the interaction of exposure and liking will predict the use of alcohol
and alcohol-related problems reported by the sample of young adolescents after
adjusting for age, gender, ethnicity, close peer alcohol use, close adult alcohol use,
general TV watching, and participation in sports. The level of exposure to alcohol ads
will be more predictive of alcohol use among adolescents who like alcohol
advertisements than among students who do not like alcohol ads.
H2. The model is hypothesized to differ by gender, but not by ethnicity. Boys will be
more strongly influenced by alcohol advertising than girls.
Methods
Participants
The data were collected as part of a study on alcohol advertising (Stacy, Zogg et
al., 2004; Zogg, 2004). Schools were randomly selected from a list of all schools in Los
Angeles County to be recruited for the study, and a total of 23 public middle schools
agreed to participate in the study across 11 school districts. All students in the 7th grade
at each school at the time of enrollment were invited to participate in the study (see

74
results section for demographic characteristics of the participants). The students signed
student assent forms prior to completing the surveys, and their parents or guardians
signed parental consent forms or gave verbal consent over the telephone. The university
institutional review board approved the study procedures. Of the 4,186 students invited to
participate, 3,890 (93%) provided the required assent/consent forms, and of those
consenting, 2,986 (77%) were surveyed in the 7
th
grade.
Procedures
Students were asked to complete paper-and-pencil questionnaires in their school
during regularly scheduled school hours. Students completed one of two forms of the
survey depending upon random assignment of each form at the school level. The two
forms included the same questions except for two alcohol advertising memory measures
that were administered between forms. One form of the survey contained a cued recall
measure for alcohol advertising on TV and the second form contained a top of mind
awareness measure (see descriptions below).
Measures
Propensity to watch TV. The general frequency of viewing television can be
associated with exposure to alcohol advertising and alcohol consumption (Grube, 1995;
Robinson, Chen, & Killen, 1998), but the influence of this variable on consumption can
be mixed (Adlaf & Kohn, 1989). The first 3 of the 7 total items in the scale assessed
weekday viewing: “On a typical weekday, how many hours a day do you watch TV…”
(a) “before school,” (b) “after school before dinner,” and (c) “from dinner until bedtime.”
The next 4 items included the following: “On a typical weekend, how many hours a day

75
do you watch TV…” (d) “Saturday morning until noon,” (e) “Saturday noon until
bedtime,” (f) “Sunday morning until noon,” and (g) “Sunday noon until bedtime.” The
response options for the 7 items ranged from 1 (I do not watch TV) to 5 (5 hours or
more).
Memory for alcohol ads: Cued recall. Participants viewed pictures captured from
2 example and 15 test commercials (Unger et al., 1995). Among the 15 test commercials,
9 were alcohol advertisements that were being aired at the time of the survey, 3 were
advertisements for non-alcoholic drinks, 1 was for an electronics retail outlet, 1 was a
public service announcement, and 1 was an old beer commercial that was out-of-date but
might elicit guessing. Participants responded to an open-ended item that asked them to
identify the product being advertised. The responses were coded by independent judges
as being related to the commercial (1) or not (0). The agreement of the judges was very
good (kappa = .88).
Top of mind awareness for alcohol ads: Draw-an-event memory test. Participants
attempted to freely recall and draw an event from 3 TV commercials as an assessment of
top of mind awareness for advertisements (Stacy, Pearce et al., 2004; Stacy, Zogg et al.,
2004). For the first event, the participants were instructed to recall any TV commercial
and form a picture of a scene from the ad in their minds. The participants were then asked
to draw a sketch of the picture they saw in their minds and were allowed 3 minutes to do
so. The second and third draw-an-event tasks asked the participants to think of alcohol
commercials and draw a sketch of the picture in their minds. After completing all of the
drawings, the participants were asked to label the key features of their drawing. Each

76
drawing was coded by two independent judges as to whether the drawing was clearly
based upon 1 of 50 currently broadcast alcohol commercial (score=1) or not (score=0).
This was intended to distinguish between a general memory for alcohol commercials and
a memory for a specific commercial. A third judge helped obtain a consensus when the
first two judges were in disagreement. The agreement of the two judges was poor
(weighted kappa = .35). Although this measure appeared to have low reliability, it was
included in the CFA analyses for illustrative purposes.
Self-reported observation of alcohol advertising. Participants were asked 4
questions about how pervasive they thought that alcohol commercials were on TV. The
first 2 questions were adapted from Schooler, Feighery, and Flora (1996), “When you
watch TV, how often do you see commercials for alcohol drinks, like beer, wine, or
liquor?” The response options ranges from 1 (a lot) to 5 (I never watch TV). The second
question asked, “In the past week, how many TV commercials have you seen for alcohol
drinks, like beer, wine, or liquor (circle ONE answer).” The response options ranged
from 0 (none) to 6 (6 or more commercials). The third and fourth questions asked
specifically about what the participants did in the past 6 months, “Saw a beer commercial
on TV?” and “Saw wine or liquor advertised on TV?” Response options for these last two
items ranged from 1 (every day) to 7 (never).
Exposure to alcohol advertising on sports programs. Alcohol advertising occurs
more frequently on televised sporting events than other programming (Madden & Grube,
1994). The measure asked how many times in the past month participants had watched
the following sports programs: professional baseball, college and professional basketball,

77
professional soccer and hockey, and Sports Center on ESPN. Response options ranged
from 1 (never) to 6 (every day). The responses were weighted with the average monthly
frequencies for alcohol advertising on each show across a 10-month period prior to data
collection (Strickland, 1983). The alcohol ad frequencies were obtained from Nielson
Media Research.
Exposure to alcohol advertising on popular shows. Participants marked how often
they watched each one of 20 popular television shows during the past month on a 6-point
scale ranging from 1 (never) to 6 (every day). The shows were selected based upon the
size of the teen audience and the number of alcohol commercials aired on the programs in
the 6 months prior to administering the surveys. Examples of these popular shows include
Seinfeld, Friends, and Soul Train. Sports shows were excluded from this measure, but
they were included in a separate measure as described above. The frequency of watching
the shows was weighted by the mean frequency of alcohol advertising on each show
(Strickland, 1983) over a 10-month period prior to each wave of surveys as reported by
Nielsen Media Research.
Liking of alcohol advertisements. Liking of advertisements has been shown to be
a valid method of copy testing by predicting sales of a number of consumer products
(Haley & Baldinger, 1991). In addition, alcohol consumption among adolescents and
young adults has been predicted by desirability and identification with characters in
alcohol advertisements (Austin et al., 2006) and with liking of advertisements (Casswell
& Zhang, 1998; Wyllie, Zhang, & Casswell, 1998a). Three items measured how well
participants in the present study liked the alcohol advertisements they observed on TV

78
(Unger, Schuster, Zogg, Dent, & Stacy, 2003). The first two questions asked “When you
see alcohol commercials on TV….” (a) “Do you think they are funny?” and (b) “Do you
think they are sexy?” Response options ranged from 1 (yes, always) to 5 (never saw any).
The third question asked “Of all the commercials you see on TV, how much do you like
the TV commercials for alcohol?” The response options ranged from 1 (I like the alcohol
commercials the most) to 5 (I have never seen an alcohol commercial on TV). The 3
items were reverse coded such that a higher score indicated greater liking.
Alcohol use in the past 30 days. The participants answered questions about their
use of alcohol (Kann, 2001) in the 30 days prior to the survey. Participants responded to
the following 5 items pertaining to how many days they had…. (1) “had at least one drink
of beer,” (2) “had at least one drink of wine or liquor,” (3) “had 3 or more drinks of beer
in a row,” (4) “had 3 or more drinks of wine or liquor in a row,” and (5) “drank enough to
get drunk.” The eight response options for the 5 items asking about the 30 days prior to
the survey ranged from: 1 (0 days) to 8 (all 30 days).
Problems due to alcohol use. Per the scale developed by Winters, Stinchfield,
and Henly (1993), participants were asked to respond to 8 items following this question:
“How many times has each event happened EVER while you were drinking alcohol or
because of your drinking alcohol? If you NEVER drank alcohol in your life, check never
for each item below.” The 8 items were as follows: (a) “Not able to do your homework
or study for a test because of your drinking” (b) “Got into fights, acted bad or did mean
things because of your drinking” (c) “Went to school or work high or drunk because of
your drinking” (d) “Caused shame or embarrassment to someone because of your

79
drinking” (e) “Neglected your responsibilities because of your drinking” (f) “Passed out
or fainted suddenly because of your drinking” (g) “Had a fight, argument or bad feeling
with a friend because of your drinking” (h) “Was told by a friend or neighbor to stop or
cut down drinking.” The five response options for each item ranged from 1 (Never) to 5
(More than 10 times).
Exposure to drinking by friends and adults. Associations with peers and adults
who drink can have an important influence on underage drinking (e.g., Feldman et al.,
1999; Wood et al., 2004). In order to control for these associations, participates
responded to two sets of questions. The first set asked about friend’s drinking in the past
30 days (1 item), the past 6 months (2 items), and during lifetime (1 item). The second set
asked about drinking by an adult the participant knew well (3 items).
Propensity to play sports. Adolescents and young adults who participant in sports
are more likely to drink alcohol (Aaron, Dearwater, Anderson, Olsen, & et al., 1995;
Leichliter, Meilman, Presley, & Cashin, 1998; Wechsler, Davenport, Dowdall,
Grossman, & et al., 1997). In addition, Slater et al. (1996) found a correlation between
participation in sports and a preference for beer advertisements with sports-related
content. Prior research has shown that a measure of viewing sports on television loaded
on a sports activity factor, not on a factor for exposure to alcohol advertising as expected
(Stacy, Pearce et al., 2004). Therefore, sports participation may be a confounder in the
relationship between exposure to alcohol advertising and alcohol consumption. The
current measure included 5 items (Thorlindsson, Vilhjalmsson, & Valgeirsson, 1990) that
asked participants how often in the past 6 months they had played football, baseball,

80
basketball, hockey or soccer, and other sports. The response options ranged from 1
(never) to 7 (every day).
Ethnicity. The students were asked to select the ethnic background of their father
and their mother. The response options included the following items: 1 (White/Non-
Latino), 2 (Latino/ Hispanic), 3 (Asian), 4 (Black/African-American), 5 (Native Hawaiian
or other Pacific Islander), 6 (American Indian or Alaska Native, Please Specify
Tribe:_____), and 7 (Don’t know). The father’s race was assigned to the student either as
a single race or as mixed race if the student selected more than one category. In addition,
the student was assigned to be of mixed race if the category selected for the mother was
different from that selected for the father or if she was of mixed race.
Language acculturation. The students indicated to what degree they use English
at home and with friends. This measure of language use is a proxy for acculturation. This
scale was adapted from a measure of acculturation previously validated for Latinos and
Whites (Marin, Sabogal, Marin, Otero-Sabogal, & Perez-Stable, 1987) and modified for
any group whose first or second language is English (Stacy, 1995). The three survey
items asked students which language they were most likely to use at home, with friends,
or while watching TV. The five response options ranged from: 1 (only English) to 5 (only
another language, not English). The items were reverse coded, and the mean was taken
as an index score.
Socioeconomic Status. SES was assessed by measuring the participant’s living
arrangement, parents’ occupations, and parents’ education. One item asked with whom
the participant lived, and the response options were 1 (both my parents most or all the

81
time), 2 (with my mother most or all of the time), 3 (with my father most or all of the
time), 4 (sometimes with my mother and sometimes with my father), and 5 (other persons
most or all the time). The occupation of each parent was asked in two open-ended items.
The open-ended items for occupation were independently coded by two judges into the
following categories: (0) unemployed, student, househusband/wife, (1) unskilled worker,
(2) machine operator, cook, waitress/waiter, (3) electrician, plumber, tailor, (4) clerk,
salesperson, flight attendant, mechanic, truck driver, military enlisted, (5) small business
owner, manager, (6) teacher, engineer, nurse, pilot, military officer, and (7) doctor,
lawyer, large business owner. The final two items asked the highest grade completed by
each parent, and the response options ranged from 1 (not completed elementary school) to
6 (completed graduate school). Each of the 3 SES proxy measures including the living
arrangement, the mean category for the occupations of the parents, and the mean
education level of the parents were used to describe the characteristics of the participants.
Data Analyses
Descriptive statistics were determined for the demographic variables and several
other variables that differed by gender including those related to the use of alcohol. Chi-
squared and t-tests were used to examine differences by gender among the variables.
Missing data. Missing data is often a concern in social science research, and the
current study is no exception. The current data had both planned missing data across two
survey forms and unplanned missing due to inadequate time to complete surveys and
participants skipping miscellaneous items. In a review of issues surrounding missing
data, Schafer and Graham (2002) recommended multiple imputation or direct maximum

82
likelihood over a range of other options that may result in biased statistical estimates (for
a similar recommendation, see L. M. Collins, 2006). Listwise deletion, for example, is a
common method for removing observations with missing data, but the approach can
result in biased estimates of means, variances, and covariances unless the data are
missing completely at random (MCAR). MCAR only occurs if the cause of the missing
data does not depend upon the observed or the missing values. According to Schafer and
Graham, data collected in the social sciences are rarely MCAR.
The current study used multiple imputations to replace missing values. The
procedure is an extension of single imputation using regression parameters generated
from non-missing data (Schafer & Olsen, 1998). A number of sets (m>1) of plausible
values are generated for the missing values. The multiple sets of values are drawn from
an assumed prior distribution for the missing values (e.g., multivariate normal). By
creating multiple data sets, it is possible to make statistical adjustments to the model
parameter estimates for the uncertainty associated with missing data. Typically, only 3 to
5 datasets are required to provide unbiased parameter estimates and efficient standard
error estimates (Rubin, 1987; Schafer & Olsen, 1998). After creating the multiple data
sets, the multiple imputation method uses standard statistical methods in analysis of the
theoretical model. The overall parameter estimates are the mean of the multiple
imputation estimates, and the total variance is comprised of the within-imputation and
between-imputation variances (Rubin, 1987; Schafer & Olsen, 1998).
The application of the multiple imputation method to the current data involved
two software packages. Multiple data sets were generated using the Markov Chain Monte

83
Carlo method within the SAS MI procedure (SAS, 2005). The multiple data sets were
then analyzed using Mplus (DATA TYPE = IMPUTATION), which is capable of
analyzing each dataset using CFA models and then reducing the multiple sets of
estimates to a single set of parameters and standard errors (Muthen & Muthen, 1998-
2007).
Formative versus reflective items. The decision rules described by Jarvis,
MacKenzie, and Podsakoff (2003) were applied to the measures used in the current study
as follows: (a) does the direction of causality flow from the latent construct to the
indicators, (b) are the indicators interchangeable due to similar content, (c) are the
indicators expected to covary with each other, and (d) should the indicators have similar
antecedents and consequences? The indicators were reflective if the responses to the four
decision rules were positive. Only factors with reflective indicators were included in the
CFA measurement model (see following text).
Measurement model. Consistent with the recommendations of Anderson &
Gerbing (1988), a CFA measurement model was fit for the theoretical latent factors of
interest. A robust, weighted least squares estimator, WLSMV, was used in Mplus to
account for non-normal distributions in the categorical indicators (Muthen & Muthen,
1998-2007; Wirth & Edwards, 2007) The fit of the measurement model was assessed
using the estimated values for the chi-squared probability, CFI, TLI, and RMSEA. Factor
loadings were examined for the expected sign, significance, and adequate size (T. A.
Brown, 2006). When the overall model fit was poor, areas of strain (i.e., undesirable
parameter estimates such as low factor loadings for certain indicators) were examined

84
using a predetermined set of modification indices and expected parameter change values
(T. A. Brown, 2006; Muthen & Muthen, 1998-2007). Any changes suggested by these
statistical parameters were evaluated in terms of relevant theory before implementing any
changes in the original model. The measurement model was fit to data for boys and girls
separately first and then simultaneously to test for measurement invariance as described
below.
Measurement invariance testing. The performance of the measurement model was
examined across gender. A number of procedures have been proposed for testing
measurement invariance using CFA methods (for review, see Vandenberg & Lance,
2000). The current study implemented the following steps (Gregorich, 2006): (a) assessed
whether the same number of factors were present in each gender to establish dimensional
invariance, (b) showed that the same indicators load on the same factors in each gender to
demonstrate configural invariance, (c) tested whether factor loadings were the same
across gender to demonstrate metric (pattern) invariance, and (d) tested whether item
intercepts were constant across gender to show strong factorial invariance. Partial
invariance testing procedures were adapted from various sources (Byrne et al., 1989;
Steenkamp & Baumgartner, 1998; Vandenberg & Lance, 2000). When the model fit at
any of the steps listed above was inadequate, a set of predetermined modification indices
and expected change parameters were examined to identify areas of strain. Constraints
were relaxed only for those parameters for which a priori conceptual support could be
justified, and once a parameter was relaxed in one step, it remained unconstrained in
subsequent steps.

85
Typically, measurement invariance is examined using chi-squared difference
tests (Byrne et al., 1989). Despite the fact that chi-square tests are sensitive to sample
sizes (Marsh, Hau, & Grayson, 2005), it has been shown using simulation studies that
chi-squared difference tests can maintain nominal Type I error rates of .05 or .01 across
models of varying complexity and invariance (French & Finch, 2006). Additional fit
indices were considered per recommendations in the literature. Chen (2007) provided
guidelines for changes in fit indices based upon simulation studies when testing
measurement invariance. For example, when sample sizes are adequate (N > 300) and
sample sizes are equal across groups, non-invariance is indicated where the CFI changes
by 0.010 or more and RMSEA changes by 0.015 or more for tests that constrain factor
loadings, intercepts, or residual variances (but see, French & Finch, 2006). Chen (2007)
cautions, however, that testing measurement invariance is a complex task, and
researchers should use good judgment in addition to the recommended guidelines.
Results
Demographics
Demographic characteristics shown in Table 2-1 indicated that the students in 7
th

grade were 12.51 (SD=0.54) years old and included the following ethnicities: 13.37%
non-Hispanic Whites, 47.87% Latino, 17.02% Asian, 3.08% African American, 0.77%
Native Hawaiian or Pacific Islander, 0.95% American Native, 4.32% mixed, and 12.62%
didn’t know. The Los Angeles county population racial demographics reported for the
2000 census were 51.24% White, 10.29% African American, 0.30% Native Hawaiian or
Pacific Islander, 12.57% Asian, 0.85% American Native, and 24.75% some other race

86
(U.S. Census Bureau, 2000). The percentage of mixed race persons was reported to be
5.19%. The report on the 2000 census also indicated that 46.88% of LA county residents
were Hispanic or Latino, which was an ethnicity that crossed racial categories according
to the Census Bureau guidelines for reporting race and ethnicity (U.S. Census Bureau,
2000).
Table 2-1 Demographic Information
Item Total Girls Boys
Gender: N (%) 3890 (100) 1905 (50.14) 1894 (49.86)

Age: M (SD) 12.51 (0.54) 12.51 (0.54) 12.51 (0.53)
Ethnicity: N (%)

White / non Latino 520 (13.37) 261 (13.78) 259 (13.60)
Latino / Hispanic 1862 (47.87) 937 (49.47) 923 (48.45)
Asian 662 (17.02) 324 (17.11) 338 (17.74)
Black / African American 120 (3.08) 56 (2.96) 64 (3.36)
Native Hawaiian or Pacific Islander 30 (0.77) 15 (0.79) 15 (0.79)
American Indian or American Native 37 (0.95) 17 (0.90) 20 (1.05)
Don’t know 491 (12.62) 196 (10.35) 206 (10.81)
Mixed 168 (4.32) 88 (4.65) 80 (4.20)

Language acculturation: M (SD) 4.22 (0.76) 4.14 (0.79) 4.28 (0.72)

Living arrangement: N (%)
Both parents 1200 (71.13) 632 (71.82) 567 (70.43)
Mother only 327 (19.38) 179 (20.34) 147 (18.26)
Father only 55 (3.26) 29 (3.30) 26 (3.23)
Alternate 54 (3.20) 18 (2.05) 36 (4.47)
Other 51 (3.02) 22 (2.50) 29 (3.60)
Education: M (SD)
1

Father 3.52 (1.63) 3.50 (1.62) 3.55 (1.64)
Mother 3.49(1.57) 3.45 (1.56) 3.55 (1.57)
Occupation of Father: N (%)
2

Unemployed, student,
househusband/wife
50 (3.62) 24 (3.28) 26 (3.99)
Unskilled worker 53 (3.84) 27 (3.69) 26 (3.99)
Machine operator, cook,
waitress/waiter
148 (10.71) 86 (11.76) 62 (9.52)
Electrician, plumber, tailor 552 (39.94) 283 (38.71) 269 (41.32)

87
Table 2-1, Continued.
Clerk, salesperson, flight attendant
mechanic, truck driver, military
enlisted
188 (13.60) 104 (14.23) 84 (12.90)
Small business owner, manager 217 (15.70) 117 (16.01) 100 (15.36)
Teacher, engineer, nurse, pilot,
military officer
140 (10.13) 67 (9.17) 73 (11.21)
Doctor, lawyer, large business owner 34 (2.46) 23 (3.15) 11 (1.69)

Occupation of Mother: N (%)
2

Unemployed, student,
househusband/wife
353 (24.77) 191 (25.16) 162 (24.32)
Unskilled worker 149 (10.46) 83 (10.94) 66 (9.91)
Machine operator, cook,
waitress/waiter
117 (8.21) 67 (8.83) 50 (7.51)
Electrician, plumber, tailor 126 (8.84) 63 (8.30) 63 (9.46)
Clerk, salesperson, flight attendant
mechanic, truck driver, military
enlisted
304 (21.33) 164 (21.61) 140 (21.02)
Small business owner, manager 162 (11.37) 87 (11.46) 75 (11.26)
Teacher, engineer, nurse, pilot,
military officer
191 (13.40) 95 (12.52) 96 (14.41)
Doctor, lawyer, large business owner 23 (1.61) 9 (1.19) 14 (2.10)

Notes:
1
Education by parent gender t(3175)=-0.49, p=.62;
2
Occupation by parent gender
was significant ( χ
2
(49)=186.81, p<.001), but parent occupation by student gender was
non-significant (both p>.05).

Gender and ethnicity differences were examined for the students surveyed in 7
th

grade (see Table 2-1). Age and ethnicity was similar across gender in the present study
(p=.86 and p=.86, respectively). The mean language acculturation was 4.22 (SD=0.76)
and this fell between the response categories 5 (only English) and 4 (English more than
another language). Girls were somewhat lower in language acculturation than boys
(M=4.14 and 4.28, respectively; p<.001). Most students lived with both parents (71.13%)
or with their mother only (19.38%) whereas smaller proportions lived with their father
only (3.26%), alternated between their mother and father (3.20%), or lived with someone

88
other than their parent (3.02%). These living arrangements were significantly different by
gender (p=.03) with more girls living with both parents or with their mother only than
boys. The education level of parents reported by the students was similar (p=.62) for
mothers (M=3.49; SD=1.57) and fathers (M=3.52; SD=1.63) and the reported education
level was independent of the gender of the student (both p>.05). The mean education
level for each parent fell between the response codes for 3 (completed high school &
received diploma) and 4 (some college or job training - 1 to 3 years). Differences for
occupations of the parents by gender were significant ( χ
2
(49)=186.81, p<.001). More
mothers appeared to stay at home (24.77%) than fathers (3.62%). Differences for parent
occupation by student gender were non-significant (both χ
2
(7)<14.07, p>.05).
Alcohol use by student gender was significant for past 30 day use of beer, lifetime
binging with beer, and past 30 days binging with beer (all χ2(7)>14.07, p<.05). Males
appeared to report higher levels of use than females in these categories (see Table 2-2).
All other comparisons of alcohol use by student gender were non-significant (all p>.05).
Males reported more negative consequences due to alcohol use (t(2648)=-2.15, p<.05).

Table 2-2 Alcohol Use and Problem Consequences by Gender
Item Total Females Males
At Least One Drink of Beer
In Lifetime N(%)

0 days 1595 (56.94) 842 (59.21) 753 (54.60)
1 day 532 (18.99) 260 (18.28) 272 (19.72)
2 days 242 (8.64) 123 (8.65) 119 (8.63)
3 to 9 days 216 (7.71) 101 (7.10) 115 (8.34)
10 to 19 days 86 (3.07) 39 (2.74) 47 (3.41)
20 to 39 days 50 (1.79) 24 (1.69) 26 (1.89)
40 to 99 days 30 (1.07) 15 (1.05) 15 (1.09)
100 or more days 50 (1.79) 18 (1.27) 32 (2.32)

89
Table 2-2, Continued
At Least One Drink of Beer
In Past 30 Days N(%)
1

0 days 2414 (83.18) 1243 (84.44) 1171 (81.89)
1 day 281 (9.68) 140 (9.51) 141 (9.86)
2 days 90 (3.10) 40 (2.72) 50 (3.50)
3 to 5 days 55 (1.90) 20 (1.36) 35 (2.45)
6 to 9 days 27 (0.93) 16 (1.09) 11 (0.77)
10 to 19 days 9 (0.31) 6 (0.41) 3 (0.21)
20 to 29 days 6 (0.21) 3 (0.20) 3 (0.21)
All 30 days 20 (0.69) 4 (0.27) 16 (1.12)
At Least One Drink of Wine
or Liquor In Lifetime N(%)

0 days 1799 (64.67) 934 (66.15) 865 (63.14)
1 day 455 (16.36) 215 (15.23) 240 (17.52)
2 days 210 (7.55) 113 (8.00) 97 (7.08)
3 to 9 days 153 (5.50) 78 (5.52) 75 (5.47)
10 to 19 days 69 (2.48) 33 (2.34) 36 (2.63)
20 to 39 days 40 (1.44) 17 (1.20) 23 (1.68)
40 to 99 days 23 (0.83) 0 (0.64) 14 (1.02)
100 or more days 33 (1.19) 13 (0.92) 20 (1.46)
At Least One Drink of Wine
or Liquor Past 30 Days N(%)

0 days 2422 (83.81) 1246 (85.05) 1176 (82.53)
1 day 272 (9.41) 124 (8.46) 148 (10.39)
2 days 105 (3.63) 54 (3.69) 51 (3.58)
3 to 5 days 34 (1.18) 17 (1.16) 17 (1.19)
6 to 9 days 23 (0.80) 14 (0.96) 9 (0.63)
10 to 19 days 10 (0.35) 5 (0.34) 5 (0.35)
20 to 29 days 6 (0.21) 2 (0.14) 4 (0.28)
All 30 days 18 (0.62) 3 (0.20) 15 (1.05)
3 or More Drinks of Beer in a
Row In Lifetime N(%)
1

0 days 2432 (88.12) 1258 (89.92) 1174 (86.26)
1 day 134 (4.86) 61 (4.36) 73 (5.36)
2 days 70 (2.54) 33 (2.36) 37 (2.74)
3 to 9 days 45 (1.63) 13 (0.93) 32 (2.35)
10 to 19 days 26 (0.94) 13 (0.93) 13 (0.96)
20 to 39 days 25 (0.91) 14 (1.00) 11 (0.81)
40 to 99 days 8 (0.29) 2 (0.14) 6 (0.44)
100 or more days 20 (0.72) 5 (0.36) 15 (1.10)

90
Table 2-2, Continued.
3 or More Drinks of Beer in a
Row In Past 30 Days N(%)
1

0 days 2688 (92.91) 1383 (94.40) 1305 (91.39)
1 day 105 (3.63) 47 (3.21) 58 (4.06)
2 days 34 (1.18) 14 (0.96) 20 (1.40)
3 to 5 days 25 (0.86) 9 (0.61) 16 (1.12)
6 to 9 days 11 (0.38) 5 (0.34) 6 (0.42)
10 to 19 days 7 (0.24) 3 (0.20) 4 (0.28)
20 to 29 days 6 (0.21) 2 (0.14) 4 (0.28)
All 30 days 17 (0.59) 2 (0.14) 15 (1.05)
3 or More Drinks of Wine or
Liquor In Lifetime N(%)

0 days 2448 (89.15) 1263 (90.67) 1185 (87.58)
1 day 135 (4.92) 55 (3.95) 80 (5.91)
2 days 58 (2.11) 31 (2.23) 27 (2.00)
3 to 9 days 43 (1.57) 20 (1.44) 23 (1.70)
10 to 19 days 20 (0.73) 9 (0.65) 11 (0.81)
20 to 39 days 17 (0.62) 7 (0.50) 10 (0.74)
40 to 99 days 6 (0.22) 2 (0.14) 4 (0.30)
100 or more days 19 (0.69) 6 (0.43) 13 (0.96)
3 or More Drinks of Wine or
Liquor In Past 30 Days N(%)

0 days 2707 (93.73) 1384 (94.60) 1323 (92.84)
1 day 92 (3.19) 43 (2.94) 49 (3.44)
2 days 30 (1.04) 16 (1.09) 14 (0.98)
3 to 5 days 18 (0.62) 10 (0.68) 8 (0.56)
6 to 9 days 13 (0.45) 4 (0.27) 9 (0.63)
10 to 19 days 7 (0.24) 2 (0.14) 5 (0.35)
20 to 29 days 6 (0.21) 2 (0.14) 4 (0.28)
All 30 days 15 (0.52) 2 (0.14) 13 (0.91)

Consequences of alcohol use:
M (SD)
2
0.09 (0.41) 0.08 (0.38) 0.11 (0.44)

Notes:
1
Alcohol use by student gender was significant for past 30 day use of beer,
lifetime binging with beer, and past 30 days binging with beer (all χ
2
(7)>14.07, p<.05),
but all other comparisons of alcohol use by student gender were non-significant (all
p>.05);
2
Consequences of alcohol use differed by gender (t(2648)=-2.15, p<.05);
P=proportion; SD=standard deviation; P = proportion; M = mean; SD = standard
deviation; N=number; %=percentage.

91
Indicator Types
Reflective indicators. Reflective indicators are predicted by a latent psychological
construct, covary, are interchangeable, and have similar antecedents and consequences
(Jarvis et al., 2003). Indicators (excluding covariates) that fit this description included
those for a propensity to watch TV, memory for alcohol ads (cued recall or top of mind
awareness), self-reported exposure to alcohol ads, liking for alcohol ads, alcohol use in
the past 30 days, propensity to have problems when drinking alcohol, and propensity to
play sports. Each of these latent constructs was expected to predict how a participant
would answer each associated indicator (survey question). For example, a propensity to
watch TV was expected to cause a participant to respond positively to watching TV at
various times during the week. The higher the propensity to watch TV, the more likely a
participant is to watch TV at each time period. Common antecedents such as poor
parental supervision would be likely to influence all of the indicators, and common
consequences such as more or less exposure to alcohol ads were also likely. It seemed
clear that memories for alcohol ads as measured with cued recall and top of mind
awareness would predict the responses on those indicators. It also seemed clear that self-
reports of exposure to alcohol ads and liking for ads required some meta-cognition about
exposure and liking that would then predict responses to those respective indicators. A
less obvious case was that of a propensity to play sports, but it was assumed that those
participants with greater athletic ability and interest in sports as antecedents were more
likely to play various sports. A common consequence for being active in sports is a
greater likelihood of drinking alcohol (e.g., Aaron et al., 1995).

92
Assignment of alcohol use and problems related to drinking alcohol as reflective
constructs deserve some detailed discussion. In the case of alcohol use in the past 30
days, the indicators asked about types of alcoholic drinks (beer and wine or liquor) and
quantity consumed (3 drinks of beer, 3 drinks of wine or liquor, and drink enough to get
drunk). A participant may have a favorite alcoholic drink such as a beer or wine cooler
when beginning to experiment with drinking, but as that person becomes more involved
with alcohol, an increase in the consumption of other types of alcohol is likely and binge
drinking also becomes more likely. Common antecedents across the indicators include
risk factors such as associating with peers who drink, and common consequences include
increased risk of accidents and abuse (S. A. Brown & Tapert, 2004).
A propensity to have problems associated with drinking alcohol seemed likely to
predict problems identified in the reflective indicators. Some persons have more
difficulty controlling their drinking behavior probably due to common antecedents such
as deficits in executive functioning, conduct disorder (Giancola & Mezzich, 2003; Myers,
Brown, & Mott, 1995) and poor affective control (A. Johnson et al., in press). Common
consequences include problems at school, at home, and with social relationships. The
indicators were assumed to covary and problems in all areas were likely to increase as a
person became more involved with alcohol and was less able to control his/her behavior.
Formative indicators. Formative indicators predict a psychological construct and
do not necessarily covary with each other (Jarvis et al., 2003). In accord with the
recommendations of Howell, Breivik, and Wilcox l (2007), multiple formative indicators
were not used to predict a single latent variable. For example, SES indicators such as

93
parental education and occupation would be entered in a structural model as individual
indicators (covariates). In a structural model, the formative latent variable is not
necessary, and the formative indicators can be used as covariates to predict a reflective
latent variable as in MIMIC (multiple indicator, multiple causes) models (T. A. Brown,
2006, p 304). Formative indicators including demographic variables, however, were used
only to describe the participant characteristics and not entered in the current measurement
model.
A unique type of formative construct is simply a summation of the formative
indicators (Howell et al., 2007, p 215). In the current study for example, the participants
indicated the frequency that they watched individual shows on a list of TV programs.
These indicators were weighted by the monthly mean number of alcohol ads shown on
each program. The sum or index of these indicators was simply an indication of the
number of alcohol ads to which a participant was exposed on average in a month. After
weighting the items, each indicator had the same units: alcohol ads per unit of time. One
index was created for popular shows and one was created for sports-related shows.
Similar summation indices were formed for exposure to drinking by friends or
adults. The two indices were sums of the number of times the participant observed
alcohol use by a friend or by an adult they knew well. The idea here is based upon Social
Learning Theory (Bandura, 1977) such that an increased exposure to drinking by models
(friends or adults) will increase the likelihood that the participant will be willing to drink
alcohol. It could also be argued that involvement with alcohol as described above could
cause a participant to seek out friends who drink alcohol. For the current age group,

94
however, it is assumed that learning about alcohol from models whether they are friends,
close adults, or actors on alcohol commercials is an influential source of exposure to
drinking alcohol. These exposure indices could be used as covariates that predict
involvement with alcohol in a future structural model designed to test hypothesized
relationships among the latent factors, but they were not included in the current
measurement model.
Measurement Invariance Testing
Prior to starting the tests of measurement invariance, the participants were split
into two groups using a random number generator in SAS. The first group of participants
numbered 1504 (744 girls and 760 boys) and the second group numbered 1480 (768 girls
and 712 boys). The first group was used to establish measurement invariance and the
details for this procedure are provided below. The second group of participants was used
to cross-validate the results observed in the first group and those comparisons are also
provided below.
Models were tested independently for each gender to examine the goodness-of-fit
and areas of strain (i.e., parameter estimates such as small factor loadings that appear to
be making the goodness of fit worse). Each of the models had reasonably good fit
statistics as can be seen in Table 2-3. The first indicator for each factor was used to set
the factor scale as is the default in Mplus (Muthen & Muthen, 1998-2007). The original 8
factor models for girls and boys had CFI and TLI values above 0.95 and RMSEA values
were less than 0.050. There were similar areas of strain in each of the models. First, the
latent factor based upon top of mind awareness of alcohol ads exhibited poorly fitting

95
indicators. The factor loadings for 2 of 3 indicators were less than 0.2 in both the girls
and boys models. For this reason and because the inter-rater reliability for the measure
was low (k=.35), the factor was excluded from the measurement model. A second
common area of strain across the two models occurred in the latent factor for propensity
to watch TV. The indicator that asked about viewing TV on weekday mornings before
school had a loading that was less than 0.4 whereas the other indicator factor loadings
were 0.59 or greater. This indicator with a low factor loading was excluded from the
model due to the difference in loadings on the factor plus the fact that the poor indicator
had a lower probability and more skewed distribution of endorsement across the
categories than the other items. The adjusted models with 7 factors each had good fit
statistics: CFI and TLI > 0.97 and RMSEA < 0.041 (see Table 2-3).
Table 2-3 Goodness-of-Fit Statistics for Model Runs on 5 Imputed Data Sets.
Sample/Model Chi-sq Percentiles CFI
2
TLI
2
RMSEA
2
Girls only
8 factor model
Expected Observed
137.066 402.500
0.965
(0.001)
0.976
(0.001)
0.042
(0.001)
Girls only
Adjusted 7 factor model
1

Expected Observed
114.400 329.060
0.970
(0.001)
0.980
(0.001)
0.041
(0.001)
Boys only
8 factor model
Expected Observed
123.957 314.599
0.972
(0.001)
0.978
(0.001)
0.037
(0.001)
Boys only
Adjusted 7 factor model
1

Expected Observed
104.034 262.017
0.977
( 0.002)
0.982
(0.001)
0.035
(0.002)
Simultaneous 2 group
model
Unconstrained factor
loadings & thresholds
Expected Observed
238.738 616.591
0.972
(0.001)
0.980
(0.000
0.039
( 0.001)
Simultaneous 2 group
model
Constrained factor
loadings & thresholds
Expected Observed
254.124 659.831
0.969
(0.001)

0.980
(0.000)
0.040
(0.001)

96
Table 2-3, Continued.

Cross-validation sample

Simultaneous 2 group
model
Unconstrained factor
loadings & thresholds
Expected Observed
232.416 629.054
0.968
(0.002)
0.978
(0.001)
0.041
(0.001)
Simultaneous 2 group
model
Constrained factor
loadings & thresholds
Expected Observed
249.594 707.199
0.963
(0.002)
0.976
( 0.001)
0.043
( 0.001)
Notes:
1
One factor (Memory: top of mind awareness) and 1 other indicator were
removed due to poor fit.
2
Mean values across the 5 data sets with standard deviations in
parentheses.

After modifying the gender models as noted above, the simple structure of the two
gender models was examined through consideration of a set of modification indices pre-
determined to be important to the current evaluation of simple structure. Specifically,
modification indices for factor cross-loadings were examined among the alcohol
advertising exposure and alcohol involvement factors. The Mplus program does not
calculate modification indices when the type of analysis includes multiple imputed data
sets. It is possible, however, to run each of the 5 imputed data sets individually and
compare the modification indices across the runs. A minimum chi-square difference was
set at a p-value of 0.001 ( χ
2
(1) > 10.0) to correct for multiple comparisons. For girls, no
modification indices met this requirement for 3 or more of the data sets. In fact, 2 of the 5
data sets did not produce any modification indices at the required significance level. For
boys, only one factor cross-loading met the chi-squared significance level in more than 3
data sets (i.e., 4 of 5 data sets in this case). The modification indices suggested that the
indicator for watching TV after dinner on weekdays should load on the latent factor for
alcohol use in the past 30 days. The expected parameter change was low, however, at less

97
than 0.2 across the 4 estimates. The cross-loading was not tested in the model due to the
low expected parameter change and the lack of any theory or past studies that would
justify this particular cross-loading. It was determined, therefore, that the adjusted 7
factor models for girls and boys had the same number of factors and the indicators loaded
on the same factor in the girls and boys models demonstrating dimensional and
configural invariance across gender (Gregorich, 2006).
The next step in the process of testing measurement invariance was to fit a model
for data from both sexes simultaneously. A baseline model was fit with unconstrained
factor loadings and thresholds (intercepts for categorical indicators) for each group. The
factor means were constrained to zero, however, and the scale factors were fixed at 1.0
for both genders to ensure identification of the model (Muthen & Muthen, 1998-2007).
The goodness-of-fit indices for this simultaneous model were good: CFI = 0.972; TLI =
0.980; and RMSEA = 0.039 (see Table 2-3), and there were no signs of strain or cross-
loadings for either gender in the modification indices for the factor loadings or
thresholds. The factor loadings were all significant (p>.001). Out of the 40 indicators for
girls, 32 were greater than 0.60, and only 1 was less than 0.40 at 0.371 for a cued recall
memory item. For boys, 32 were greater than 0.6 and none were less than 0.4. Based
upon these indices, it was determined that the simultaneous, baseline model fit the data
well.
The baseline, unconstrained model was compared to a model with both factor
loadings and thresholds constrained to be equal across groups. The factor means were
fixed at zero for girls and freely estimated for boys, and the scale factors were fixed to 1

98
for girls and freely estimated for boys (Muthen & Muthen, 1998-2007). It was not
possible to calculate the difference in chi-square between the two models using the
imputation type of analysis in Mplus. The fit of the constrained model was very good
(CFI=0.969, TFI=0.980, and RSMEA=0.040; see Table 2-3). The differences in the
goodness-of-fit indices across the unconstrained and constrained models suggested that
there was no loss of fit by constraining the factor loadings and thresholds according to
critical values recommended by Chen (2007). The changes in three indices were as
follows: ΔCFI=0.003, ΔTLI=0.000, and ΔRMSEA=0.001. In addition, there were no
areas of strain or cross-loadings suggested for the thresholds or factor loadings by the set
of pre-determined modification indices obtained by running the constrained model on
each of the 5 imputed data sets. Out of the 40 indicators for girls, 30 standardized factor
loadings were greater than 0.60, and only 1 was less than 0.40 at 0.389, which was a
loading for a cued recall memory item. For boys, 32 factor loadings were greater than 0.6
and none were less than 0.4. These results can be seen in Table 2-4 that lists the
standardized loading estimates and residual variances.
The factor loadings for indicators on all latent variables for females were
compared to those for males by examining the correlation between the two sets of factor
loadings. A high correlation coefficient would be expected for latent factors that are
invariant across gender. The results were as follows: Pearson’s correlation r=0.92,
minimum difference (girls – boys loading) = -.318, maximum difference = 0.097, and a
median difference of -0.001. A comparison of residual variances for indicators across
gender produced these results: correlation r=.55, minimum difference = 0.072, maximum

99
difference 0.846, and median difference = -0.511. In summary, the factor loadings were
very similar across gender as expected, but the residual variances tended to be higher for
boys particularly for the cued recall memory indicators and for the alcohol-related
problem indicators.
Table 2-4 Factor Loading Estimates for the Final Constrained Group Model.
Factor/Indicator Constrained
1
Girls Boys
Loading S.E.
3
Std.
4
Resid.
5
Std.
4
Resid.
5
Estimate Est. Var. Est. Var.
Self-Reported Exposure to Ads
See alcohol ads how often
2
1.000 0.000 0.660 0.565 0.685 0.569
Alcohol ads in past week 1.079 0.045 0.712 0.493 0.741 0.480
Beer ads past 6 months 1.364 0.055 0.900 0.190 0.883 0.264
Wine/liquor ads past 6 months 1.257 0.050 0.829 0.312 0.773 0.537

Memory for Ads: Cued Recall
Miller Lite - women at bar
2
1.000 0.000 0.679 0.537 0.619 1.508
Miller Lite - men on phone 0.950 0.124 0.641 0.588 0.628 1.300
Bud Light – coach yelling 0.701 0.144 0.471 0.776 0.496 1.385
Miller Lite – man/woman at bar 1.017 0.147 0.686 0.529 0.661 1.314
Miller Draft - bowling shoes 0.833 0.134 0.561 0.684 0.526 1.700
Corona - palm tree/lighthouse 0.751 0.128 0.506 0.743 0.519 1.446
Miller Draft - people at tables 0.670 0.110 0.453 0.794 0.505 1.256
Bud Ice - penguin 0.692 0.127 0.466 0.782 0.432 2.050
Budweiser - lizard 0.576 0.136 0.389 0.846 0.707 0.313
Alcohol Use Past 30 Days
Number days drank beer
2
1.000 0.000 0.891 0.206 0.904 0.185
Days drank wine/liquor 0.994 0.022 0.885 0.216 0.897 0.199
Drank 3 or more beers 1.078 0.021 0.960 0.079 0.963 0.077
Drank 3 or more wine/liquor 1.065 0.019 0.949 0.099 0.945 0.113
Days got drunk 1.068 0.020 0.952 0.094 0.943 0.119

Problems Due to Alcohol
Not able to do homework
2
1.000 0.000 0.936 0.123 0.898 0.991
Got into fights 1.029 0.034 0.963 0.072 0.898 1.046
Went to school drunk 0.983 0.039 0.920 0.153 0.958 0.374
Embarrassed someone 0.945 0.047 0.885 0.217 0.946 0.439
Neglected responsibilities 0.974 0.036 0.912 0.169 0.912 0.778
Passed out 0.936 0.051 0.876 0.231 0.934 0.523
Argued with a friend 1.000 0.035 0.936 0.124 0.933 0.616
Told to stop drinking 0.938 0.045 0.878 0.229 0.892 0.921

100
Table 2-4, Continued.
Propensity to Watch TV
After school before dinner
2
1.000 0.000 0.591 0.650 0.671 0.532
After dinner 1.063 0.046 0.628 0.605 0.694 0.531
Saturday morning 1.119 0.050 0.662 0.562 0.679 0.640
Saturday afternoon/evening 1.262 0.052 0.746 0.444 0.702 0.717
Sunday morning 1.154 0.051 0.682 0.535 0.683 0.667
Sunday afternoon/evening 1.256 0.054 0.743 0.448 0.730 0.603

Like Alcohol Ads
Alcohol ads are funny
2
1.000 0.000 0.620 0.616 0.632 0.757
Alcohol ads are sexy 1.213 0.088 0.752 0.435 0.704 0.758
Like alcohol ads the most 1.044 0.076 0.647 0.581 0.737 0.463

Propensity to Play Sports
Baskeball
2
1.000 0.000 0.628 0.605 0.591 0.738
Football 1.394 0.084 0.875 0.234 0.875 0.235
Hockey or soccer 0.885 0.058 0.556 0.691 0.537 0.764
Baseball 0.940 0.054 0.590 0.651 0.601 0.619
Other sports 0.817 0.056 0.513 0.736 0.416 1.266
Notes:
1
Factor loadings are constrained to be equal across gender.
2
S.E. = standard error
of the factor loading estimate (constrained across gender).
3
The loading for the first
indicator was set to 1.0 to fix the scale of the factor.
4
Std. Res. = standardized residual.
5

Resid. Var. = residual variance.

The factor means and variances for the constrained model are shown in Table 2-5.
The means for girls were fixed at zero in order to identify those means for boys, which
were significantly different from girls. Three factor means were significantly higher for
boys than girls including cued recall memory for alcohol ads (M=0.21; p<.05), liking of
alcohol ads (M=0.81; p<.001), and propensity to play sports (M=0.58; p<.001). Factor
variances were standardized to one (see Table 2-5), but all of the factor variances where
significant (all p<.01) except for the latent alcohol-related problems among boys (p>.05).

101
Table 2-5 Estimates for Factor Means and Variances in the Constrained Model.
Factor/Parameter Girls Boys
Param.
Est.
1
S.E
2
. Std.
Est.
3
Param.
Est.
1
S.E.
2
Std.
Est.
3
Means
Self-reported alcohol ad
exposure 0.000 0.000 0.000 0.040 0.040 0.056
Memory for ads: Cued recall 0.000 0.000 0.000 0.200 0.079 0.207
Alcohol use past 30 days 0.000 0.000 0.000 0.088 0.126 0.096
Alcohol-related problems 0.000 0.000 0.000 -1.692 1.363 -0.828
General TV viewing 0.000 0.000 0.000 0.063 0.036 0.095
Like alcohol ads 0.000 0.000 0.000 0.575 0.054 0.809
Play sports 0.000 0.000 0.000 0.365 0.042 0.580

Variances
Self-reported alcohol ad
exposure 0.435 0.031 1.000 0.503 0.053 1.000
Memory for ads: Cued recall 0.463 0.095 1.000 0.944 0.294 1.000
Alcohol use past 30 days 0.794 0.030 1.000 0.833 0.175 1.000
Alcohol-related problems 0.877 0.049 1.000 4.114 3.452 1.000
General TV viewing 0.350 0.027 1.000 0.437 0.042 1.000
Like alcohol ads 0.384 0.041 1.000 0.505 0.072 1.000
Play sports 0.395 0.036 1.000 0.396 0.053 1.000
Notes:
1
Param. Est.= parameter estimate.
2
S.E.=standard error.
3
Std. Est.=standardized
parameter estimate.

Cross-validation of the results involved completing the same series of models as
described above with a second group of participants randomly split from the first (see
description of participant groups above). The independent gender models were similar to
those observed in the first groups, and the same exclusions were implemented by
removing the top of mind awareness memory factor and one weekday morning indicator
from the propensity to watch TV measure. The results for the goodness-of-fit indices for
the unconstrained and constrained models are shown in Table 2-3. The fit was good for
the final constrained model for this group as indicated by the following values:
CFI=0.963, TLI=0.976, and RMSEA=0.043. The differences between the unconstrained

102
and constrained models were larger that those observed in the first group but still
relatively small in overall magnitude (F. F. Chen, 2007). The differences were as follows:
ΔCFI=0.005, ΔTLI=0.002, and ΔRMSEA=0.002. The differences between the fit indices
for the final constrained models in the first and second cross-validation groups were also
small in magnitude: ΔCFI=.006, ΔTLI=.004, and ΔRMSEA=.003. The factor loadings
and residual variances were also compared across groups. The difference for the cross
validation samples in the mean factor loadings was 0.003 for girls and -0.011 for boys,
and the difference in residual variances was -0.005 for girls and -0.086 for boys (see
Table 2-6). The factor loadings correlated very well for girls (r=0.987) and boys
(r=0.976). The residual variances correlated very well for girls (r=0.985), and there was a
strong correlation for boys (r=0.822) although the magnitude was not as large as for girls.
Taken together, these cross-validation results provide additional confidence that there is
measurement invariance across the samples.
Table 2-6 Comparison of Parameters for the Split, Cross-Validation Samples.
Measurement Girls Boys

Std.
Load.
1
Resid.
Var.
2
Std.
Load.
1
Resid.
Var.
2
Mean parameter value for both samples 0.727 0.444 0.742 0.702
Minimum difference in values across samples -0.061 -0.085 -0.108 -0.868
Maximum difference in values across samples 0.059 0.079 0.055 0.958
Median difference in values across samples 0.005 -0.008 -0.009 0.015
Mean difference in values across samples 0.003 -0.005 -0.011 0.086
Standard deviation for mean difference 0.031 0.042 0.036 0.299
Difference as a percentage of parameter mean 0.47% -1.20% -1.48% 12.28%
Notes:
1
Std. Load.=standardized loading.
2
Resid. Var.=residual variance.

The next measurement concern was related to construct validity. While it is not
possible to rigorously examine construct validity without using multiple traits and
multiple methods (e.g., Campbell & Fiske, 1959), it is possible to examine correlations

103
among latent factors for indications that the factors are related to each other as expected
based upon applicable theories or prior research. This type of evaluation is often
described as examining the nomological net of a measure. Table 2-7 provides the factor
correlation estimates with the values for girls above the diagonal and the values for boys
below the diagonal. The largest values for both sexes are the correlations between alcohol
use and alcohol-related problems (r=.69 girls and r=.67 boys), which is to be expected.
The second largest correlations were between alcohol use and liking of alcohol ads (r=.56
girls and r=.52 boys). As might be expected from these high correlation values, liking of
alcohol ads and alcohol-related problems were also correlated (r=.47 girls and r=.37
boys). It is important to note that despite these high correlation values between latent
factors, the indicators did not exhibit any cross-loadings among these factors.

Table 2-7 Correlations among Latent Factors for Girls and Boys.
Latent Factors 1 2 3 4 5 6 7
1. Memory for Ads: Cued Recall 1 .329 .257 .242 ns ns ns
2. Self-Reported Ad Exposure .338 1 .191 .195 .174 .134 .209
3. Propensity to Watch TV .164 .323 1 .193 ns ns ns
4. Like Alcohol Ads .323 .314 .251 1 .565 .471 .124
5. Alcohol Use (past 30 days) ns .198 .154 .515 1 .687 .243
6. Alcohol-Related Problems ns .289 ns .366 .673 1 .274
7. Propensity to Play Sports ns .192 ns .217 .200 .211 1
Factor means for boys
(girls fixed at zero )
.207 ns ns .809 ns ns .580
1
Correlations above the diagonal are for girls and those below the diagonal are for boys.
ns = non-significant (p>.05).

Measures of exposure to alcohol advertisements were correlated as expected.
Cued recall memory and self-reported exposure were correlated (r=.33 girls and r=.34
boys), and each of these measures was correlated with liking of alcohol ads although the
relationship was stronger for boys than girls (see Table 2-7). The exposure measures had

104
low or non-significant correlations with alcohol use and alcohol-related problems. While
lower values were expected, the non-significant values between cued recall for ads and
alcohol use and problems were unanticipated because it is hypothesized that exposure to
ads does have an influence on alcohol use. The full extent of these relationships will be
explored further in the structural modeling phase of the analyses.
Discussion
The current study provides an example of how to examine the properties of
measures and construct a measurement model in preparation for testing hypothesized
relationships among latent constructs in a structural model. The example steps included
imputation of missing data, delineation of reflective versus formative measures, exclusion
of indicators and measures with poor psychometric properties, and tests of measurement
invariance across gender. Results of these analyses in the current study showed that
alcohol advertising measures (e.g., cued recall for ads and self-reported exposure to ads)
and alcohol use measures (e.g., past 30 day use and problems associated with alcohol
use) have good factor loadings and a simple structure without any cross-loadings of
indicators on the latent factors. In addition, the analyses showed that the factor loadings
and thresholds are invariant across gender, which will allow comparison of structural
model parameters across gender. Although it was anticipated that partial invariance might
occur, none was indicated. Finally, the results for the measurement model were
successfully cross-validated with the second half of the sample of participants, which was
randomly split at the beginning of the analyses.

105
Measures were examined for the type of relationship between the indicators and
the latent variables, and most of the principle measures of memory for exposure to
alcohol advertising were determined to be reflective indicators such that a latent
psychological construct predicts how participants will respond to certain questions on a
survey (Bollen & Lennox, 1991). One measure of exposure, however, was determined to
more likely be formative. Exposure to alcohol advertising on popular shows measures the
number of ads observed on each TV program watched in a month. The sum of the ads
observed over a period of time was a measure of exposure to alcohol ads but did not
necessarily reflect some latent psychological construct that predicted how participants
chose programs to watch.
The CFA measurement model simultaneously fit the data to all of the latent
factors across gender, and the factors were allowed to freely correlate within each gender
(T. A. Brown, 2006). This was useful in confirming that the latent factors were strongly
associated or poorly associated with other factors as expected (i.e., evaluation of the
nomological net). For example and as anticipated, the largest correlation among both girls
and boys was between alcohol use in the past 30 days and alcohol-related problems, and
the correlation between the propensity to watch TV was not significantly correlated with
past 30-day alcohol use or alcohol-related problems. In addition, differences were
observed by gender providing some support for the hypotheses that alcohol ads would
have a different influence on boys and girls (Casswell & Zhang, 1998; Connolly et al.,
1994). These differences were observed among latent constructs such as cued-recall

106
memory for ads, self-reported exposure to ads, liking of alcohol ads, and propensity to
watch TV.
Several limitations apply to the current study. First, the participants were recruited
from schools in Los Angeles County and might not be representative of students located
in other regions of the country. Second, African American students were under-
represented in the study sample. Finally, differences across ethnicities were not
anticipated (Stacy, Zogg et al., 2004) and measurement invariance across ethnicities was
not tested due to the complexity of fitting a CFA measurement model to multiple groups
by gender and ethnicity. It is possible, however, that some differences in the current
measures might exist across ethnicity. This possibility will be controlled to some degree
by including ethnicity as a covariate in the structural model.
In summary, the current study provides a useful example of how to prepare a
measurement model. The techniques apply to the study of measures that are used to
assess psychological constructs across a range of disciplines including marketing
research, health behavior, and other basic and applied social sciences.

107

Chapter 3 Exposure to Alcohol Advertising on Television and Alcohol Use among
Young Adolescents
Abstract
This study provides important support for a causal relationship between exposures
to televised alcohol advertising and underage drinking. In the current study, a total of
3,890 students were surveyed across 4 years from the 7
th
through the 10
th
grades. Results
for structural equation modeling of alcohol consumption showed that exposure to alcohol
ads and/or liking of those ads in 7
th
grade were predictive of the latent growth factors for
alcohol use (past 30 days and past 6 months) after controlling for a range of covariates. In
addition, there was a significant total effect for males and a significant mediated effect
for females of exposure to alcohol ads and liking of those ads in 7
th
grade through latent
growth factors for alcohol use on alcohol-related problems in 10
th
grade.
Introduction
Alcohol use among adolescents and young adults is a major health concern in the
USA. In a recent national report, there were 10.9 million (29%) adolescents between the
ages of 12 and 20 who reported drinking alcohol in the past month prior to the survey
(SAMHSA, 2004). Of those persons between the ages of 12 and 17 years, 17.7% reported
drinking in the past month, 10.2% were binge drinkers (5 or more drinks on one occasion
at least once during the past month), and 2.6% were heavy drinkers (5 or more drinks on
the same occasion at least 5 different days in the past month). A study by Lewinsohn,
Rohde, and Seeley (1996) found that 16.6% of adolescents aged 14 to 18 years reported

108
problem behaviors related to alcohol use and an additional 6.2% met DSM-IV criteria for
substance abuse or dependence.
Several health risks are associated with alcohol use by adolescents including
accidental injuries, homicides, and suicides, which are the three leading causes of
mortality among adolescents (Windle & Windle, 2006). Accidental injuries cause more
than 50% of all deaths in this age group (CDC, 2004), and one important concern is
driving under the influence of alcohol. Approximately 20% of drivers between the ages
of 16 and 20 involved in fatal crashes between 1991 and 2001 had a blood alcohol
content greater than 0.08 b/dl (NHTSA, 2003). Alcohol consumption by adolescents has
been associated with depressed mood, risky behaviors, poor academic performance, and
smoking tobacco (Spirito et al., 2001). Other studies have linked extended alcohol use by
adolescents to abnormal brain development including memory and attention problems
(Tapert et al., 2002). The initiation of alcohol use at an early age increases the risk for
adverse consequences later in life due to its association with heavier and more persistent
consumption (Maggs & Schulenberg, 2006).
A range of risk factors has been identified for underage drinking. Parental
monitoring and parental drinking influence adolescents (Fergusson et al., 1995; Wood et
al., 2004) as well as drinking by friends (Feldman et al., 1999). Alcohol is used
sometimes to relieve stress from traumatic events (Clark et al., 1997; Kilpatrick et al.,
2000; Wills et al., 2001) and sometimes as part of a disengagement coping style (Wills et
al., 2001). Expectancies about the anticipated effects of drinking alcohol and, in
particular, the endorsement of the enhanced social interaction expectancies for alcohol

109
use by adolescents has been shown to predict alcohol consumption prospectively
(Christiansen et al., 1989; Killen et al., 1996). Persons with strong, spontaneous alcohol-
related associations to ambiguous cues are at higher risk for higher alcohol use (Stacy,
1997; Stacy & Newcomb, 1998; Weingardt et al., 1996) as predicted by dual-process
theories of judgment and decision-making (Stacy, Ames, & Knowlton, 2004; Tiffany,
1990). Having parents and family members with substance use disorders has been
identified as a risk factor for adolescents to abuse alcohol whether the child was raised
with those family members or not (Hoffmann & Cerbone, 2002; Miles et al., 1998).
In addition to the risk factors noted above, considerable attention has been given
to the influence of alcohol advertising on underage drinking using both cross-sectional
and longitudinal research. Cross-sectional studies have consistently shown a small but
significant association between exposure to alcohol ads and alcohol use (e.g., Adlaf &
Kohn, 1989; Atkin et al., 1984; Austin et al., 2006; Wyllie et al., 1998b). More
importantly, prospective studies have shown similar findings providing support for a
causal relationship between exposure to ads and alcohol use.
Two longitudinal studies in New Zealand demonstrated that exposure to alcohol
advertising at an earlier age was predictive of later alcohol use. The participants (n=667)
in a study by Connolly, Casswell, Zhang and Silva (1994) were interviewed at the ages of
13, 15, and 18 years old. Multiple regression results showed that the number of
advertisements recalled from television, radio, magazines, newspapers, and films at age
15 was predictive of beer consumption at age 18 for males, and the number of ads
recalled at age 13 was positively associated with the frequency of drinking beer at age 18

110
for females. The results for other measures of ad exposure such as alcohol-use portrayals
in the media were not predictive of later alcohol use. The authors concluded that beer
advertising observed by adolescent males had a significant influence on their drinking
behavior at age 18, but the influence of advertising was not as clear for females.
In the second study reported from New Zealand, Casswell and Zhang (1998)
administered questionnaires and interviewed participants at the ages of 18 and 21 years
old. There were significant paths in the structural equation model from liking
advertisements and brand allegiance at age 18 to beer consumption at ages 18 and 21, and
there were significant paths from consumption at these ages to aggression at age 21.
Gender was also significant in the model such that men liked advertisements more and
were more likely to drink larger quantities of beer in both age groups. The authors
concluded that 18 year-olds who expressed a liking for beer advertisements and brand
allegiance were more likely to drink larger quantities of beer at age 21 even after
adjusting for beer consumption at age 18. In addition, the 21 year-olds who consumed
more beer were more likely to experience alcohol-related aggression.
Ellickson, Collins, Hambarsoomians, and McCaffrey (2005) reported on a 3-year
prospective study of the effects of alcohol ad exposure on 3,111 middle school students
located in South Dakota, USA. Exposure to advertising at beer concession stands, but not
to TV or magazine ads, was predictive of 9
th
grade drinking for those who were non-
drinkers in 7
th
grade after adjusting for control variables. Exposure to advertising in
magazines and at beer concessions stands, but not TV or in-store displays, were
predictive of drinking in 9
th
grade for those who were drinkers in 7
th
grade after

111
adjustment for covariates. The authors concluded that some sources of alcohol
advertising may influence adolescent drinking depending upon prior experience with
alcohol by the adolescent.
In another longitudinal study, Snyder et al. (2006) used both self-reported
exposure to alcohol advertising and industry advertising expenditures across 24 US
markets to show that drinking among adolescents and young adults (aged 15 to 26 years)
is influenced by alcohol advertising. The researchers adjusted for age, gender, ethnicity,
school attendance status, and alcohol sales per capita. The results indicated that
consumption was positively associated with the number of advertisements observed each
month and with alcohol company expenditures per capita on advertising. The results were
similar for the sub-sample of youth under the age of 21. The strengths of this research,
which showed a significant association between alcohol ad exposure and consumption
was the large number of participants recruited across different markets in the USA, the
objective measure of advertising expenditures coupled with self-reports of advertising
exposure, and the longitudinal data collection.
The current study reports on data collected prospectively across four years from
7
th
through 10
th
grade students in Southern California. Two prior studies have reported on
the effects of advertising observed in the first two and first three years of this data
collection. Stacy, Zogg, Unger, and Dent (2004) examined the early predictive effects of
a range of alcohol advertising exposure measures in a prospective study among young
adolescents. Exposure measures collected by Stacy et al. included the following: (a)
watched TV shows index, (b) watched TV sports index, (c) self-reported frequency, (d)

112
cued-recall memory test, and (e) draw-an-event memory test. Outcome measures
included current use of beer or wine/liquor (past 30-days), binge drinking (3 or more
drinks per occasion), and prior alcohol use (past 6 months). Results showed that one
opportunity-based measure, the watched shows exposure index, was predictive of alcohol
use measures even after adjusting for potential confounders, and a second opportunity-
based measure, the watched TV sports index, predicted beer use but not wine/liquor use
or binging. A memory-based measure, the self-reported frequency of seeing alcohol ads,
predicted beer use but not wine/liquor use or binging, but two other memory-based
measures, cued recall and draw-an-event, were not predictive of alcohol use. In the
second study reporting on the Southern California students, Zogg (2004) fit the first 3
years of data to a modified version of the Strickland (1983) model. In partial support of
the model, results of multiple regression showed that alcohol consumption in 8
th
grade
significantly mediated the relationship between self-reported exposure to ads in 7
th
grade
and alcohol-related problems reported in 9
th
grade. The effects of ad exposure were small,
which was consistent with the findings of Strickland.
Taken together, these cross-sectional and prospective studies have shown small
but consistent effects of exposure to advertising on the consumption of alcohol by
adolescents. These effects have been observed across a number of measures for exposure
to advertising. An opportunity measure often used is one that asks participants to self-
report how frequently they watch certain TV programs, and that reported frequency is
weighted by the number of alcohol ads broadcast on those programs. This watched shows
index of advertising exposure has been used successfully in models that significantly

113
predicted alcohol use (Adlaf & Kohn, 1989; Austin et al., 2006; Bloom et al., 1997;
Stacy, Zogg et al., 2004). A second self-report measure asks participants to reflect on
how many alcohol advertisements they have observed in the recent past, and this measure
has also been a significant predictor of alcohol use (Snyder et al., 2006; Zogg, 2004).
Cued recall of alcohol ads is a memory-based measure that has had some mixed results.
The measure has successfully predicted intentions to use alcohol (Grube & Wallack,
1994), but has both significant (Connolly et al., 1994) and non-significant results (Stacy,
Zogg et al., 2004) in the prediction of alcohol use. One particularly strong measure of
exposure in the literature is self-reported liking of ads although this is more of an
affective reaction to ads than a strict measure of exposure. This measure has been used
successfully, however, to predict susceptibility to use alcohol (Unger et al., 1995) and
alcohol use (Casswell & Zhang, 1998; Wyllie et al., 1998a). In addition, liking of
commercials was the best measure of several alternative measures used to predict sales
of various consumer products (Haley & Baldinger, 1991). While there is no single ‘gold
standard’ measure for exposure to advertising, there are several measures that have been
applied successfully to the study of alcohol advertising.
The current study examined the effects of alcohol ad exposure on consumption
across 4 years of data collection. Liking of alcohol advertisements was hypothesized to
be a moderator of the relationship between alcohol advertising exposure and alcohol use.
Liking of alcohol advertisements was expected to vary among the participants for a
number of reasons including social influences. For example, peers may share and
encourage liking of the ads, while some parents may train their children to be skeptical of

114
claims made in commercials. It was anticipated that adolescents who like alcohol
advertisements will be more likely to elaborate on the content of the ads (e.g., image
themselves in the scene), and as a result, they will be more likely to be persuaded to try
the product (cf., Petty & Wegener, 1999). Adolescents who disliked or were skeptical of
alcohol ads were less likely to be persuaded by the ads. The specific hypotheses for this
study were as follows:
H1. Exposure to alcohol ads, the liking of those ads (affective reaction), and a synergistic
interaction of exposure and liking measured at time 1 will predict the initial level of
alcohol use and the growth of involvement with alcohol from time 1 through time 3
after adjusting for potential confounders (see Figure 3-1). When liking of ads is low,
exposure to alcohol ads will have little effect on initial alcohol use or the growth of
alcohol use over time. When liking is higher, then those participants with more
exposure will have a higher initial use and a faster rate of growth in alcohol use (i.e.,
a significant increase in alcohol use over the three time periods).
H2. Advertising exposure will interact with liking of commercials at time 1 in the
prediction of problems associated with alcohol consumption in time 4, and this
relationship will be mediated by the growth in alcohol use from time 1 through time 3
after adjusting for potential confounders (see Figure 3-2). This moderated-mediation
hypothesis suggests that those students who like alcohol advertisements will be more
strongly influenced by exposure to those ads to begin drinking and drink more than
those students who do not like alcohol ads, and those who drink more will experience
more negative consequences from drinking. The intention here is to provide support

115
for a causal relation for the effects of alcohol advertising on alcohol use and negative
consequences due to alcohol use.
H3. Based upon previous studies, the models are hypothesized to differ by gender. Boys
will be more strongly influenced by alcohol advertising than girls.

Figure 3-1 Model for Growth of Alcohol Use
Notes: I = growth curve intercepts; S = linear slope.
Wave 1
Alcohol
Use
Wave 3
Alcohol
Use
Wave 2
Alcohol
Use
I S
Wave 1 covariates:
Age, Gender, Ethnicity
Peer alcohol use
Adult alcohol use
Sports participation
General TV watching
Wave 1 predictors:
Ad Exposure
Ad Liking
Exposure*Liking

116

Figure 3-2 Model for Mediation of Exposure and Problems
Notes: I = growth curve intercepts; S = linear slope; O = latent outcome variable; ab=indirect effect mediation pathway; c’=direct
pathway.

Wave 1
Alcohol
Wave 3
Alcohol
Wave 2
Alcohol
I S
Wave 1 covariates:
Age, Gender, Ethnicity
Peer alcohol use
Adult alcohol use
Sports participation
General TV watching
Wave 1 (latent variables
or composite scores):
Ad Exposure
Ad Liking
Exposure*Liking
O
Wave 4:
Problems 1
Problems 2
.

.
a
b
c’

117
Methods
Participants
The current data were collected as part of a prospective study on the influence of
alcohol advertising on underage drinking (Stacy, Zogg et al., 2004; Zogg, 2004).
Participants recruited from public schools were surveyed during regular school hours
from the 7
th
through the 10
th
grades. Of the 4,186 students recruited to participate in the
study, 3,890 (93% of consented) students completed the survey in at least one wave:
2,986 (77%) were surveyed in 7
th
grade, 2849 (73%) in the 8
th
grade, 2093 (54%) in the
9
th
grade, and 1,609 (41%) in the 10
th
grade. Dropout in the 9
th
and 10
th
grades was
primarily due to failure of entire schools to remain in the study after initial agreements to
participate. Five of nineteen high schools refused to allow surveys to be administered in
class to those students who had been surveyed previously in middle schools.
A total of 23 public middle schools, randomly selected from all middle schools in
Los Angeles County, agreed to participate in the study, and these schools represented 11
school administrative districts across the county. The characteristics of the participants
differ somewhat from those for Los Angeles County in part due to the refusal of some
middle schools to participate in the study. All 7
th
grade students in each school at the time
of the study were invited to participate. Parents of the students either signed a consent
form or gave verbal consent via telephone, and students signed assent forms prior to
completing the surveys. The surveys and all procedures were approved by the university
Institutional Review Board.

118
Procedures
Students completed paper-and-pencil questionnaires during regular classroom
hours at their school. Two forms of the survey were administered between subjects based
upon random assignment at the school level. One form included a cued recall measure
and the second form included a top of mind awareness measure of exposure to alcohol
advertising on TV (see measures section below). All other questions were the same for
the two forms.
Measures
Memory for alcohol ads: Cued recall. Surveys included still pictures captured
from TV advertisements including 2 example and 15 test ads (Unger et al., 1995). Among
the 15 test ads, 9 were alcohol ads that were being shown on TV at the time of the survey,
3 were ads for soft drinks, 1 was for an electronic store, 1 was an anti-alcohol PSA, and 1
was an old alcohol that ad might elicit guessing. An open-ended item asked participants
to write down what product was being advertised. Independent judges coded the
responses as being related to the commercial (1) or not (0), and there was good inter-rater
agreement (kappa = .88) and good internal consistency among the items (coefficient
alpha = .74).
Self-reported observation of alcohol advertising. Participants were asked about
how often they saw alcohol commercials on TV. The first 2 questions were adapted from
Schooler, Feighery, and Flora (1996), “When you watch TV, how often do you see
commercials for alcohol drinks, like beer, wine, or liquor?” The options ranged from 1 (a
lot) to 5 (I never watch TV). The 2
nd
question asked, “In the past week, how many TV

119
commercials have you seen for alcohol drinks, like beer, wine, or liquor (circle ONE
answer).” The options ranged from 0 (none) to 6 (6 or more commercials). The 3
rd
and 4
th

questions asked about participant activities in the past 6 months, “Saw a beer commercial
on TV?” and “Saw wine or liquor advertised on TV?” Options for these last two
questions ranged from 1 (every day) to 7 (never). The 4 items showed good internal
consistency (coefficient alpha = .72).
Exposure to alcohol advertising on popular shows. Participants indicated how
frequently they watched 20 each TV shows during the past month on a 6-point scale
ranging from 1 (never) to 6 (every day). Selection of the shows listed on the surveys
involved consideration of the popularity of the shows among teens and the frequency of
alcohol ads on them. The shows included Seinfeld, Friends, and Soul Train among others.
Sports programs were included in a separate measure as described below. The self-
reported frequency of watching each show was multiplied by the average frequency of
alcohol advertising for each show (Strickland, 1983) as reported by Nielsen Media
Research over a 10-month period prior to administering the surveys. The weighted items
were summed to yield an index score for the number of alcohol ads each participant was
exposed to during a typical day of watching popular show (coefficient alpha = .79).
Exposure to alcohol advertising on sports programs. A separate measure of
exposure to alcohol ads was needed because televised sporting events air many more
alcohol ads than other programming (Madden & Grube, 1994). Survey questions asked
participants how many times in the past month they watched professional baseball,
college and professional basketball, professional soccer and hockey, and Sports Center on

120
ESPN. Response options ranged from 1 (never) to 6 (every day). These responses were
weighted in a manner similar to the popular shows described above. The mean of the
items provided an index for the number of alcohol ads observed per hour of watching TV
(coefficient alpha = .80).
Liking of alcohol advertisements. Studies on copy testing by advertisers have
shown that liking of advertisements is predictive of sales for consumer products (Haley &
Baldinger, 1991). In addition, drinking among adolescents and young adults is associated
with desirability and identification with characters in alcohol ads (Austin et al., 2006) and
with liking of alcohol ads (Casswell & Zhang, 1998; Wyllie et al., 1998a). The current
surveys included 3 items assessing how much participants liked alcohol ads on TV
(Unger et al., 2003). The first two items asked “When you see alcohol commercials on
TV….” (a) “Do you think they are funny?” and (b) “Do you think they are sexy?”
Options ranged from 1 (yes, always) to 5 (never saw any). The 3rd item asked “Of all the
commercials you see on TV, how much do you like the TV commercials for alcohol?”
The options ranged from 1 (I like the alcohol commercials the most) to 5 (I have never
seen an alcohol commercial on TV). Reverse coding yielded items where a higher score
indicated greater liking (coefficient alpha = .78).
Current alcohol use. Current use of alcohol was assessed for the past 30 days
(Kann, 2001) and past 6 months. Five items assessed use of alcohol in the 30 days prior
to the survey by asking how many days participants had…. (1) “had at least one drink of
beer,” (2) “had at least one drink of wine or liquor,” (3) “had 3 or more drinks of beer in a
row,” (4) “had 3 or more drinks of wine or liquor in a row,” and (5) “drank enough to get

121
drunk.” The 8 options ranged from: 1 (0 days) to 8 (all 30 days). A second group of
items asked how often in the past 6 months participants had… (1) “drank beer,” (2)
“drank wine or wine coolers,” (3) “drank liquor,” or (4) “got drunk.” Options ranged
from 1 (every day) to 8 (never). The 6 month questions were reverse coded and then an
index was formed from all 9 items (coefficient alpha = .91).
Problems due to alcohol use. Participants responded to 8 items that were based
upon the measure developed by Winters, Stinchfield, and Henly (1993): “How many
times has each event happened EVER while you were drinking alcohol or because of
your drinking alcohol? If you NEVER drank alcohol in your life, check never for each
item below.” The 8 items were as follows: (a) “Not able to do your homework or study
for a test because of your drinking” (b) “Got into fights, acted bad or did mean things
because of your drinking” (c) “Went to school or work high or drunk because of your
drinking” (d) “Caused shame or embarrassment to someone because of your drinking” (e)
“Neglected your responsibilities because of your drinking” (f) “Passed out or fainted
suddenly because of your drinking” (g) “Had a fight, argument or bad feeling with a
friend because of your drinking” (h) “Was told by a friend or neighbor to stop or cut
down drinking.” The 5 options for each item ranged from 1 (Never) to 5 (More than 10
times). An index score was calculated as the mean of the 8 items (coefficient alpha = .93).
Potential confounding variables. The following assessments were selected
primarily because they have shown some association with alcohol ads and/or alcohol use
in prior research, and it was important to control for these potentially confounding
variables. One such measure is the amount of time that is spent watching television,

122
which has mixed results in its association with alcohol use (Grube, 1995; Robinson et al.,
1998). This was assessed with 7 items asking about the amount of time spent watching
TV on weekdays and weekends. Associations with peers and adults who drink can have
an important influence on underage drinking (e.g., Feldman et al., 1999; Wood et al.,
2004). In order to control for these associations, participates responded to two sets of
questions, one set about drinking by friends and a second set asking about drinking by
adults the participant knew well. Adolescents and young adults who participate in sports
are more likely to drink alcohol (Aaron et al., 1995; Leichliter et al., 1998; Wechsler et
al., 1997), so participants were asked about how often in the past 6 months they had
played football, baseball, basketball, hockey or soccer, and other sports (Thorlindsson et
al., 1990). Demographic variables assessed included age, gender, ethnicity, language
acculturation (Marin et al., 1987; Stacy, 1995) , and parents’ occupation and education.
Data analyses
Descriptive statistics were calculated for demographic and alcohol use variables
among all participants and by gender. Chi-squared statistics provided information on
differences in these variables by gender. Bivariate correlations showed preliminary
information about the associations between pairs of the variables.
Some adjustments were required for missing data that occurred due to skipping
items by participants or due to attrition. A number of studies have shown that ad hoc
procedures for handling missing data such as list wise deletion or mean substitution often
result in biased parameter and/or standard error estimates whereas procedures that use
maximum likelihood methods or multiple imputation yield more accurate estimates while

123
adjusting for the uncertainty associated with the missing data (e.g., Schafer & Graham,
2002). The current analyses utilized full information maximum likelihood estimation (R.
J. A. Little & Rubin, 2002) as implemented in Mplus statistical software (Muthen &
Muthen, 1998-2007).
Construction of the structural equation models used to test the hypotheses
involved two steps (J. C. Anderson & Gerbing, 1988). First, a confirmatory factor
analyses of the measurement models established the following: (a) simple structure of the
model such that each item loaded well on its hypothesized latent factor and not on any
other factors, (b) measurement invariance for factor loadings and indicator intercepts or
thresholds across gender and time of assessment (for review, see Vandenberg & Lance,
2000), and (c) parcels for indicator items where appropriate (T. D. Little, Cunningham,
Shahar, & Widaman, 2002). The second step involved sequential fitting of the model as
follows: (a) latent growth curve models across times 1 - 3, (b) the addition of the time 1
predictor and covariates (centered on their means), and (c) for hypothesis H2, the final
addition of the time 4 outcome variable, alcohol-related problems. Fit of the models were
assessed with a range of goodness-of-fit statistics (for review, see Marsh et al., 2005)
including the Chi-squared test, Comparative Fit Index (CFI), Tucker-Lewis Index (TLI),
Root Mean Squared Error of Approximation (RMSEA), and the Standardized Root Mean
Square Residual (SRMR).
The moderation and mediation hypotheses were tested as a part of the model
evaluation. Moderation of the exposure to alcohol ads by liking of the ads was tested by
creating an interaction, manifest variable (product of the two variables at time 1) after

124
centering each of the continuous variables. An interaction was indicated by the
significance of the regression coefficient of the latent growth variables or the latent
outcome variable on the interaction variable.
As stated in hypothesis H2, predictors at time 1 were expected to influence the
outcome at time 4 through the growth of alcohol use as a mediator. Both the latent
intercept and slope for the growth of alcohol use were expected to act as mediators. Three
approaches were used to evaluate the significance of the mediated effects. First,
mediation was tested by assessing the joint significance of the coefficient for regression
of the mediator on a predictor and the coefficient for the regression of the outcome on the
mediator adjusted for the predictor. The joint significance test for mediation has been
shown in simulation studies to have low Type I error rates and good power to detect
mediation (MacKinnon, Lockwood, Hoffman, West, & Sheets, 2002), but it does not
provide a measure of the mediation effect size. Second, mediation effect sizes (i.e.,
specific and total indirect effects) were assessed using the multivariate delta method
(Bollen, 1989) as implemented in Mplus (Muthen & Muthen, 1998-2007). This method
estimates significance for the product of two regression coefficients, the coefficient for
the mediator regressed on the predictor and the coefficient for the outcome regressed on
the mediator adjusted for the predictor. The delta method has been shown to produce
accurate standard errors in simulation studies (MacKinnon et al., 2002). Finally, the
confidence intervals for the indirect effects were calculated using the distribution of the
product method (PRODCLIN software: MacKinnon, Fritz, Williams, & Lockwood, in
press) to adjust for the non-normal distribution of the product of two normally distributed

125
variables that is the product of the two regression coefficients in the mediation model
(MacKinnon, 2008; MacKinnon, Lockwood, & Williams, 2004). The distribution of the
product method has more power to detect significant mediation effect sizes than the delta
method by adjusting the critical values used to estimate the confidence interval for the
non-normal distribution of the product term.
Results
Demographic characteristics for time 1 of the study as shown in Table 3-1
indicated that the students in 7
th
grade were 12.51 (SD=0.54) years old and included the
following ethnicities: 13.37% non-Hispanic Whites, 47.87% Latino, 17.02% Asian,
3.08% African American, 0.77% Native Hawaiian or Pacific Islander, 0.95% American
Native, 4.32% mixed, and 12.62% didn’t know. Age and ethnicity was similar across
gender in the present study (p=.86 and p=.86, respectively). The mean language
acculturation was 4.22 (SD=0.76) and this fell between the response categories for
language spoken: 5 (only English) and 4 (English more than another language). Females
reported somewhat lower language acculturation than males (M=4.14 and 4.28,
respectively; p<.001). The education level of parents reported by the students was similar
(p=.62) for mothers (M=3.49; SD=1.57) and fathers (M=3.52; SD=1.63) and the reported
education level was independent of the gender of the student (both p>.05). The mean
education level for each parent fell between the response codes for 3 (completed high
school & received diploma) and 4 (some college or job training - 1 to 3 years).
Occupation of the parents by gender was significantly different ( χ
2
(49)=186.81, p<.001).
It appeared that more mothers (24.77%) stayed at home than fathers (3.62%). Parent

126
occupation by student gender was non-significant (both χ
2
(7)<14.07, p>.05). Alcohol use
by student gender was significant for past 30 day use of beer, lifetime binging with beer,
and past 30 days binging with beer (all χ2(7)>14.07, p<.05) with males reporting higher
levels of use than females in these categories (see Table 3-1). All other comparisons of
alcohol use by student gender were non-significant (all p>.05). Males reported more
negative consequences due to alcohol use (t(2648)=-2.15, p<.05).
Table 3-1 Demographic Information for Participants in 7
th
Grade.
Item Total Females Males
Gender: N (%) 3890 (100) 1905 (50.14) 1894 (49.86)
Age: M (SD) 12.51 (0.54) 12.51 (0.54) 12.51 (0.53)
Ethnicity: N (%)

White / non Latino 520 (13.37) 261 (13.78) 259 (13.60)
Latino / Hispanic 1862 (47.87) 937 (49.47) 923 (48.45)
Asian 662 (17.02) 324 (17.11) 338 (17.74)
Black / African American 120 (3.08) 56 (2.96) 64 (3.36)
Native Hawaiian or Pacific Islander 30 (0.77) 15 (0.79) 15 (0.79)
American Indian or American Native 37 (0.95) 17 (0.90) 20 (1.05)
Don’t know 491 (12.62) 196 (10.35) 206 (10.81)
Mixed 168 (4.32) 88 (4.65) 80 (4.20)
Language acculturation: M (SD) 4.22 (0.76) 4.14 (0.79) 4.28 (0.72)
Education: M (SD)
1

Father 3.52 (1.63) 3.50 (1.62) 3.55 (1.64)
Mother 3.49(1.57) 3.45 (1.56) 3.55 (1.57)

Occupation of Father: N (%)
2

Unemployed, student,
househusband/wife
50 (3.62) 24 (3.28) 26 (3.99)
Unskilled worker 53 (3.84) 27 (3.69) 26 (3.99)
Machine operator, cook, waitress/waiter 148 (10.71) 86 (11.76) 62 (9.52)
Electrician, plumber, tailor 552 (39.94) 283 (38.71) 269 (41.32)
Clerk, salesperson, flight attendant
mechanic, truck driver, military enlisted
188 (13.60) 104 (14.23) 84 (12.90)
Small business owner, manager 217 (15.70) 117 (16.01) 100 (15.36)
Teacher, engineer, nurse, pilot, military
officer
140 (10.13) 67 (9.17) 73 (11.21)
Doctor, lawyer, large business owner 34 (2.46) 23 (3.15) 11 (1.69)

127
Table 3-1, Continued.
Occupation of Mother: N (%)
2

Unemployed, student,
househusband/wife
353 (24.77) 191 (25.16) 162 (24.32)
Unskilled worker 149 (10.46) 83 (10.94) 66 (9.91)
Machine operator, cook, waitress/waiter 117 (8.21) 67 (8.83) 50 (7.51)
Electrician, plumber, tailor 126 (8.84) 63 (8.30) 63 (9.46)
Clerk, salesperson, flight attendant
mechanic, truck driver, military enlisted
304 (21.33) 164 (21.61) 140 (21.02)
Small business owner, manager 162 (11.37) 87 (11.46) 75 (11.26)
Teacher, engineer, nurse, pilot, military
officer
191 (13.40) 95 (12.52) 96 (14.41)
Doctor, lawyer, large business owner 23 (1.61) 9 (1.19) 14 (2.10)

At Least One Drink of Beer
In Lifetime N(%)

0 days 1595 (56.94) 842 (59.21) 753 (54.60)
1 day 532 (18.99) 260 (18.28) 272 (19.72)
2 days 242 (8.64) 123 (8.65) 119 (8.63)
3 to 9 days 216 (7.71) 101 (7.10) 115 (8.34)
10 to 19 days 86 (3.07) 39 (2.74) 47 (3.41)
20 to 39 days 50 (1.79) 24 (1.69) 26 (1.89)
40 to 99 days 30 (1.07) 15 (1.05) 15 (1.09)
100 or more days 50 (1.79) 18 (1.27) 32 (2.32)
At Least One Drink of Beer
In Past 30 Days N(%)
3

0 days 2414 (83.18) 1243 (84.44) 1171 (81.89)
1 day 281 (9.68) 140 (9.51) 141 (9.86)
2 days 90 (3.10) 40 (2.72) 50 (3.50)
3 to 5 days 55 (1.90) 20 (1.36) 35 (2.45)
6 to 9 days 27 (0.93) 16 (1.09) 11 (0.77)
10 to 19 days 9 (0.31) 6 (0.41) 3 (0.21)
20 to 29 days 6 (0.21) 3 (0.20) 3 (0.21)
All 30 days 20 (0.69) 4 (0.27) 16 (1.12)
At Least One Drink of Wine or Liquor
In Lifetime N(%)

0 days 1799 (64.67) 934 (66.15) 865 (63.14)
1 day 455 (16.36) 215 (15.23) 240 (17.52)
2 days 210 (7.55) 113 (8.00) 97 (7.08)
3 to 9 days 153 (5.50) 78 (5.52) 75 (5.47)
10 to 19 days 69 (2.48) 33 (2.34) 36 (2.63)
20 to 39 days 40 (1.44) 17 (1.20) 23 (1.68)
40 to 99 days 23 (0.83) 0 (0.64) 14 (1.02)

128
Table 3-1, Continued.
100 or more days 33 (1.19) 13 (0.92) 20 (1.46)
At Least One Drink of Wine or Liquor
In Past 30 Days N(%)

0 days 2422 (83.81) 1246 (85.05) 1176 (82.53)
1 day 272 (9.41) 124 (8.46) 148 (10.39)
2 days 105 (3.63) 54 (3.69) 51 (3.58)
3 to 5 days 34 (1.18) 17 (1.16) 17 (1.19)
6 to 9 days 23 (0.80) 14 (0.96) 9 (0.63)
10 to 19 days 10 (0.35) 5 (0.34) 5 (0.35)
20 to 29 days 6 (0.21) 2 (0.14) 4 (0.28)
All 30 days 18 (0.62) 3 (0.20) 15 (1.05)
3 or More Drinks of Beer in a Row
In Lifetime N(%)
3

0 days 2432 (88.12) 1258 (89.92) 1174 (86.26)
1 day 134 (4.86) 61 (4.36) 73 (5.36)
2 days 70 (2.54) 33 (2.36) 37 (2.74)
3 to 9 days 45 (1.63) 13 (0.93) 32 (2.35)
10 to 19 days 26 (0.94) 13 (0.93) 13 (0.96)
20 to 39 days 25 (0.91) 14 (1.00) 11 (0.81)
40 to 99 days 8 (0.29) 2 (0.14) 6 (0.44)
100 or more days 20 (0.72) 5 (0.36) 15 (1.10)
3 or More Drinks of Beer in a Row
In Past 30 Days N(%)
3

0 days 2688 (92.91) 1383 (94.40) 1305 (91.39)
1 day 105 (3.63) 47 (3.21) 58 (4.06)
2 days 34 (1.18) 14 (0.96) 20 (1.40)
3 to 5 days 25 (0.86) 9 (0.61) 16 (1.12)
6 to 9 days 11 (0.38) 5 (0.34) 6 (0.42)
10 to 19 days 7 (0.24) 3 (0.20) 4 (0.28)
20 to 29 days 6 (0.21) 2 (0.14) 4 (0.28)
All 30 days 17 (0.59) 2 (0.14) 15 (1.05)
3 or More Drinks of Wine or Liquor
In Lifetime N(%)

0 days 2448 (89.15) 1263 (90.67) 1185 (87.58)
1 day 135 (4.92) 55 (3.95) 80 (5.91)
2 days 58 (2.11) 31 (2.23) 27 (2.00)
3 to 9 days 43 (1.57) 20 (1.44) 23 (1.70)
10 to 19 days 20 (0.73) 9 (0.65) 11 (0.81)
20 to 39 days 17 (0.62) 7 (0.50) 10 (0.74)
40 to 99 days 6 (0.22) 2 (0.14) 4 (0.30)

129
Table 3-1, Continued.
100 or more days 19 (0.69) 6 (0.43) 13 (0.96)
3 or More Drinks of Wine or Liquor
In Past 30 Days N(%)

0 days 2707 (93.73) 1384 (94.60) 1323 (92.84)
1 day 92 (3.19) 43 (2.94) 49 (3.44)
2 days 30 (1.04) 16 (1.09) 14 (0.98)
3 to 5 days 18 (0.62) 10 (0.68) 8 (0.56)
6 to 9 days 13 (0.45) 4 (0.27) 9 (0.63)
10 to 19 days 7 (0.24) 2 (0.14) 5 (0.35)
20 to 29 days 6 (0.21) 2 (0.14) 4 (0.28)
All 30 days 15 (0.52) 2 (0.14) 13 (0.91)

Consequences of alcohol use: M (SD)
4
0.09 (0.41) 0.08 (0.38) 0.11 (0.44)

Notes:
1
Education by parent gender t(3175)=-0.49, p=.62;
2
Occupation by parent gender
was significant ( χ
2
(49)=186.81, p<.001), but parent occupation by student gender was
non-significant (both p>.05);
3
Alcohol use by student gender was significant for past 30
day use of beer, lifetime binging with beer, and past 30 days binging with beer (all
χ
2
(7)>14.07, p<.05), but all other comparisons of alcohol use by student gender were
non-significant (all p>.05);
4
Consequences of alcohol use differed by gender (t(2648)=-
2.15, p<.05); P=proportion; SD=standard deviation; N=number; %=percentage.

Measures of alcohol use and consequences of use were weakly correlated with
measures of exposure to alcohol advertising in bivariate analyses (see Table 3-2). This
was true for self-reported frequency of seeing alcohol ads (all .08 ≤r ≤.13, all p<.001),
watching popular shows (all .05 ≤r ≤.10, all p<.05), watching sport shows (all .04 ≤r ≤.09,
for 8 of 9 p<.05), and general TV watching (all .07 ≤r ≤.13, all p<.001). Product cued
recall had small but significant correlations with lifetime beer use (all r=.06, p<.05) but
not to other alcohol use measures (all p>.05). The strongest correlations were between
liking of alcohol ads and alcohol use measures (all .15 ≤r ≤.33, all p<.001).

130
Table 3-2 Bivariate Correlations among Alcohol Use and Ad Exposure Measures.
Variable Lifetime
Beer
Lifetime
Wine/Liquor
Lifetime
Beer
Binges
Lifetime
Wine/Liquor
Binges
Past
Month
Beer
Past Month
Wine/Liquor
Past
Month
Beer
Binges
Past Month
Wine/Liquor
Binges
Alcohol-
Related
Problems
Self-
reported
frequency
.14*** .13*** .09*** .09*** .12*** .10*** .08*** .08*** .08***
Product
cued recall
.06* .03 .01 .00 -.01 -.02 -.02 -.03 -.01
Watched
popular
shows
.10*** .09*** .06** .05* .10*** .08*** .09*** .09*** .09***
Watched
sports
shows
.05* .04* .06** .04 .05** .06** .09*** .08*** .08***
Liking of
alcohol ads
.33*** .30*** .30*** .29*** .29*** .27*** .28*** .24*** .15***
General
TV
watching
.13*** .13*** .09*** .07*** .09*** .10*** .08*** .07*** .08***
Notes: Wave 1 data; * p<.05; ** p<.01; *** p<.001

131
Measurement model
The measurement model examined the factor loading, simple structure, and
measurement invariance of the latent variables proposed for the models. Indicators loaded
well on their hypothesized latent variables in separate models for females and males.
Loadings of indicators on the alcohol use factors across times 1, 2, and 3 were all above
.800 for males and females with the exception of one item for females that was .755.
Indicators for the self-reported exposure to advertising and liking of alcohol ads all
loaded on their respective factors at greater than .630 for males and females, and
indicators for the time 4 consequences of alcohol use were all greater than .700 for male
and females. Examination of a priori hypothesized modification indices for cross-
loadings among the alcohol use, consequences, self-reported ad exposure, and liking of
ads (adjusted for multiple comparisons: critical χ
2
(1)>10.0) indicated 3 cross-loadings for
females and 2 for males. In an adjusted model, however, the fit of the indicators for these
cross-loadings was poor (all loadings<.200 and p>.10) providing support for a simple
structure among the factors, and therefore, the cross-loadings were excluded from the
measurement models. The measurement models for males and females each fit the data
well: CFI>.97, TLI>.98, and RSMEA<.044. The chi-square statistics were significant
(p<.001), but this is likely due to the large sample size.
The next step in evaluation of the measurement model was to test invariance
across gender and across time. The fit indices for the model with unconstrained factor
loadings and thresholds across gender (Fit: χ
2
(258)=1497.94, p<.001; CFI=.98; TLI=.99;
RMSEA=.050) had only a trivial differences (F. F. Chen, 2007) from the fit of the model

132
where loadings and thresholds were constrained (Fit: χ
2
(275)=1490.18, p<.001;
CFI=.98; TLI=.99; RMSEA=.048). This observed invariance of loadings and thresholds
was adequate to compare structural models across gender (e.g., Gregorich, 2006). When
evaluating invariance across time for the alcohol use factors, the fit for the model with
constrained factor loadings and thresholds across time (Fit: χ
2
(23)= 1218.751, p<.001;
CFI=.98; TLI=.99; RMSEA=.127) was similar to the fit for the unconstrained model (Fit:
χ
2
(11)= 1238.462, p<.001; CFI=.98; TLI=.97; RMSEA=.185) providing evidence for
invariance across time. A measurement model for males and females combined is
provided in Figure 3-3.
The preceding analyses demonstrated that items loaded well on their respective
factors for each gender and did not exhibit any cross-loadings across factors providing
evidence for unidimensionality of the factors. This evidence plus the fact that the goal of
the current research is to test relationships among latent factors (and not evaluation of
individual scale indicators) meets standards that warrant the use of parcels in the
structural model (Bandalos & Finney, 2001; T. D. Little et al., 2002). The 8 indicators in
the time 4 measure of alcohol-related problems were parceled by summing 2 indicators
each yielding a total of 4 parcels. The use of parcels in SEM, when warranted, provides
for more stable model estimation especially when the distribution of responses is non-
normal (Bandalos & Finney, 2001; T. D. Little et al., 2002) or skewed toward zero as is
the case for alcohol-related problems.

133

Figure 3-3 Measurement Model.
Notes: Factor loadings and thresholds were invariant across gender; Larger ovals
represent latent factors and rectangles designate measured variables; Numbers to the left
of the measured variables and within small circles are residual, or unique, variances of
indicators. Standardized estimates are provided; All factor loadings are significant based
on unstandardized estimates at p<.01; Fit indices: χ
2
(238)=1620.68, p<001; CFI=.98;
TLI=.99; RMSEA=.039.
Time 1
Alcohol Use
Drank beer: past 30 days
Drank wine/liquor: 30 days
Binged beers: past 30 days
Binged wine/liquor: 30 days
Got drunk: past 30 days
Drank beer: past 6 months
Drank wine: past 6 months
Got drunk: past 6 months
Drank liquor: past 6 months
.24
.27
.16
.23
.19
.11
.12
.10
.26
.87
.86
.95
.94
.94
.90
.88
.85
.92
Time 2
Alcohol Use
Drank beer: past 30 days
Drank wine/liquor: 30 days
Binged beers: past 30 days
Binged wine/liquor: 30 days
Got drunk: past 30 days
Drank beer: past 6 months
Drank wine: past 6 months
Got drunk: past 6 months
Drank liquor: past 6 months
.22
.20
.12
.21
.15
.19
.07
.09
.22
.89
.88
.95
.96
.93
.92
.89
.85
.94

134

Figure 3-3, Continued.

Time 3
Alcohol Use
Drank beer: past 30 days
Drank wine/liquor: 30 days
Binged beers: past 30 days
Binged wine/liquor: 30 days
Got drunk: past 30 days
Drank beer: past 6 months
Drank wine: past 6 months
Got drunk: past 6 months
Drank liquor: past 6 months
.15
.18
.08
.17
.13
.11
.09
.06
.16
.92
.92
.97
.96
.94
.93
.91
.90
.96
Time 4
Alcohol Use
Problems
Not able to study
Got into fights
Went to school drunk
Caused shame to someone
Neglected responsibilities
Passed out or fainted
Had an argument with a friend
Was told to stop drinking
.43
.42
.21
.40
.36
.42
.32
.30
.76
.84
.82
.76
.80
.77
.89
.76

135

Figure 3-3, Continued.
Time 1
Cued Recall
of Alcohol
Ads
Miller Lite: 2 women at bar
Miller Lite: Men on cell
Bud Light: Coach yelling
Miller Lite: Couple at bar
Miler Draft: Bowling shoes
Corona Extra: Palm tree/house
Miller Draft: At bar cheering
Budweiser: Frogs/lizards
Bud Ice: Penguin
.48
.48
.70
.66
.67
.56
.43
.69
.45
.72
.74
55
.76
.66
.57
.59
.72
.55
Time 1
Self-Report
Alcohol Ad
Exposure
How often see alcohol ads
How many ads: past week
How many beer ads: 6 months
How many wine/liquor: 6 mo.
Are alcohol ads funny
Are alcohol ads sexy
How much like alcohol ads
.54
.41
.47
.53
.33
.15
.46
.68
.74
.92
.82
.69
.73
.77
Time 1
Liking of
Alcohol Ads

136
Structural growth models
The latent growth factors for alcohol use over times 1 through 3 were regressed
on each of the alcohol ad exposure measures in 4 separate models. The hypothesized
moderator, liking of alcohol ads, was included in each of those 4 models (see Figure 3-1).
In addition, the growth factors were simultaneously regressed on covariates measured at
time 1 including age, observing peers drink, observing adults drink, playing sports,
general TV watching, language acculturation, and socio-economic status measured by the
occupations and education of each participant’s parents. Standard errors were adjusted for
clustering by school (Muthen & Muthen, 1998-2007). All structural growth models
differed by gender, so only the results for multi-group models by gender are presented
here.
The model with exposure to alcohol ads on popular shows fit the data well (Fit
indices: χ
2
(26)=62.44, p<001; CFI=.97; TLI=.92; RMSEA=.027, SRMR=.021) and
yielded significant interactions between exposure to alcohol ads and liking of those ads in
the prediction of the intercept for the growth curve. As shown in Figure 3-4 and Table 3-
3, the interaction term was significant for males (b=.112, p=.006) and for females
(b=.093, p=.037). Figure 3-5 depicts this interaction illustrating that the level of exposure
to ads was more predictive of a higher level of alcohol use in 7
th
grade for those students
who reported a greater liking of alcohol ads. The size of the interaction term for males
(95%CI: 0.033 - 0.192) was not significantly different from that for females (95%CI:
0.005 - 0.180). There was no interaction in the prediction of the slope for the latent

137
growth for alcohol use, but exposure to ads predicted the slope for females (b=.160,
p=.005) and liking of ads predicted the slope for males (b=.283, p=.003).
The intercepts for the latent growth factors were both significant for males and
females (both p<.001), but there was no significant difference for the latent growth
intercept factor between males (95%CI: 0.437 - 0.556) and females (95%CI: 0.422 -
0.573) or for the latent growth slope factor for males (95%CI: 0.360 – 1.127) and females
(95%CI: 0.333 - 0.570). The residual variances for the latent growth intercept and slope
were each significant for females (both p>.05), and the residual intercept variance was
significant for males (p<.05). The residual variance for the latent growth slope factor for
males was small and insignificant, however, and this variance had to be fixed at zero for
the parameters to be estimable.
Several covariates in the latent growth model that were significant in predicting
the intercept for females included observing friends drink (b=.472, p<.001) and
observing close adults drink (b=.137, p=.001), and for males, the significant covariates
included observing friends drink (b=.605, p<.001), observing close adults drink (b=.174,
p=.001), and language acculturation (b=-.100, p=.03). Covariates that significantly
predicted the slope for females included playing sports (b=-.131, p=.04), and for males,

138

Figure 3-4 Growth Model by Gender as Predicted by Viewing of Popular Shows.
Notes: Alcohol use = past 30 days + past 6 months; I = growth curve intercepts; S = growth curve slopes; Standardized parameter
estimates: male/female (p<.05), ns=non-significant; Adjusted for age, observing peers drink, observing adults drink, playing sports,
general TV watching, acculturation, parents’ jobs, parents’ education, and clustering by school; Fit indices: χ
2
(26)=62.44, p<001;
CFI=.97; TLI=.92; RMSEA=.027, SRMR=.021.
Wave 1
Alcohol
Wave 3
Alcohol
Wave 2
Alcohol
I S
Like Ads
Shows x Like
Popular Shows
.112/.093
.518/.499
.201/.438
.653/.632 .834/.814
.319/.692
.304/.337
ns/.950 .340/.535
.851/.398
.283/ns
ns/.160
.189/.261
.709/.510

139
Table 3-3 Standardized Parameter Estimates for the Alcohol Use Growth Model.
Girls Boys
Parameter
Estimate
SE Parameter
Estimate
SE
Intercept on
T1 alcohol use .814*** .061 .834*** .040
T2 alcohol use .632*** .059 .653*** .047
T3 alcohol use .499*** .058 .518*** .030
Slope on
T1 alcohol use .000 .000 .000 .000
T2 alcohol use .438*** .043 .201*** .036
T3 alcohol use .692*** .070 .319*** .061
Intercept on T1 predictors
Popular shows -.031 .030 -.024 .031
Liking of ads .261*** .048 .189*** .030
Shows x Liking .093* .044 .112** .041
Age .042 .033 .034 .030
Peer drinking .472*** .059 .605*** .042
Playing sports .013 .040 -.007 .024
Adult drinking .137** .040 .174** .052
General TV viewing .003 .035 .015 .041
Language acculturation .048 .042 -.100* .045
Parents’ jobs -.002 .044 .093 .048
Parents’ education -.042 .040 .002 .029
Slope on T1 predictors
Popular shows .160** .057 .184 .120
Liking of ads -.023 .075 .283* .095
Shows x Liking -.076 .064 -.223 .149
Age .024 .041 -.136 .113
Peer drinking .020 .076 -.820*** .176
Playing sports -.131* .065 -.005 .134
Adult drinking -.023 .062 -.191 .195
General TV viewing -.016 .064 -.099 .116
Language acculturation .027 .069 .374* .165
Parents’ jobs .109 .084 -.254 .205
Parents’ education -.079 .065 .070 .159
Slope correlation with Intercept -.237 .132 na na
Latent Variable Intercepts
Growth intercept .497*** .038 .469*** .030
Growth slope .451*** .061 .744*** .196

140
Table 3-3, Continued.
Residual variances
T1 acohol use .337** .099 .304*** .066
T2 alcohol use .510*** .047 .709*** .053
T3 alcohol use .398*** .081 .851*** .028
Intercept .535*** .095 .340*** .064
Slope .950*** .033 .000 na
Notes: T1= time 1, T2=time 2, T3=time3; * p<.05; **p<.01; ***p<.001;
na=not available, slope variance fixed at 0.

the significant covariates included observing friends drink (b=-.820, p<.001) and
acculturation (b=.374, p=.02). The change in sign for the parameters for males between
the intercept and slope suggest that those higher in alcohol use at time 1 had smaller
growth slopes in use, but it should be noted that, when freed to evaluate this hypothesis,
the correlations between the intercept and slope factors were non-significant for both
males and females (both p>.05).
Growth models for the other 3 exposure measures including cued recall of alcohol
ads, self-reported exposure to alcohol ads, and exposure to alcohol ads on sports shows
had similar patterns of fit as the model for exposure to ads on popular shows (i.e.,
CFI>.95 and RMSEA<.030). The model for cued recall had a significant interaction of
exposure and liking in the prediction of the intercept (p<.01) for males but not for
females. No significant interactions between exposure and liking were observed for the
self-reported exposure or the sports shows models. Liking of alcohol ads consistently
predicted the intercept for both males and females across these 3 models (all p<.001).
None of the 3 exposure measures predicted the intercepts or slopes for females

141

Part a. Males
0.000
0.500
1.000
1.500
2.000
2.500
00.5 11.5 22.5 33.5 4
Exposure to Ads on Popular Shows
Intercept for Growth Curves
Liking at mean + 1 std Liking at mean Liking at mean - 1 std

Part b. Females
0.000
0.200
0.400
0.600
0.800
1.000
1.200
1.400
0 0.5 1 1.5 2 2.5 3 3.5 4
Exposure to Ads on Popular Shows
Intercept for Growth Curves

Figure 3-5 Interaction of Exposure to Ads with Liking of Ads.
Notes: Males in part ‘a’ and females in part ‘b’; Liking of ads plotted at the mean, the
mean plus 1 standard deviation, and the mean minus 1 standard deviation.

142
(all p>.05), but cued recall predicted the intercept and self-reported exposure predicted
both the intercept and slope for males in those 2 respective models (all p<.05). Among
the covariates for females, observing friends and close adults drinking were significant
predictors of the intercepts, but not the slopes, across the three models (all p<.01). For
males, observing friends drink was predictive of both intercepts and slopes in the three
models (all p<.001), and similar to the popular shows model described previously, the
coefficient for friends drinking was positive when predicting the intercept and negative
when predicting the slope. A similar pattern was observed for language acculturation
among males except that the coefficient was negative when predicting the intercept and
positive when predicting the slope (all p<.05). Observing close adults drink by males was
predictive of the intercepts across the three models (all p<.01), but this covariate was not
predictive of the slopes (all p>.05).
Structural mediation models
The hypothesized mediation model (see Figure 3-2) fit the data well for exposure
to alcohol ads on popular shows (fit indices: χ
2
(130)=182.66, p=.002; CFI=.98; TLI=.97;
RMSEA=.015, SRMR=.026), and important mediating relationships were observed in the
model (see Figure 3-6 and Table 3-4). Both the intercept and slope for the growth in
alcohol use across time 1 through time 3 were significant mediators for females between
exposure to alcohol ads at time 1 and alcohol-related problems at time 4. The joint
significance test for mediation by the slope among females was met by the coefficient for
the regression of the slope for alcohol use on exposure to ads on popular shows (a=.190,
p=.001) and the coefficient for the regression of alcohol-related problems on the slope for

143
alcohol use (b=.478, p<.0001). The mediation effect size or indirect effect was also
significant in agreement with the joint significance test (delta method indirect effect:
ab=.091, p=.02; PRODCLIN 95%CI: 0.032 – 0.166). The joint significance test for
mediation by the intercept among females was met by the coefficient for the regression of
the intercept for alcohol use on liking of ads (a=.267, p<.001) and the coefficient for the
regression of alcohol-related problems on the intercept for alcohol use (b=.393, p<.02).
The mediation effect size or indirect effect was also significant in agreement with the
joint significance test (delta method indirect effect: ab=.105, p=.03 PRODCLIN 95%CI:
0.018 – 0.207). The joint significance test for mediation by the intercept among females
was also met by the coefficient for the regression of the intercept for alcohol use on the
interaction term for popular shows and liking of ads (a=.091, p=.042) and the coefficient
for the regression of alcohol-related problems on the intercept for alcohol use (b=.393,
p=.02). The mediation effect size or indirect effect was not significant, however (delta
method indirect effect: ab=.036, p=.16; PRODCLIN 95%CI: did not converge). Among
males, the joint significance test for mediation was not met for any of the hypothesized
pathways, but there was a significant total effect of the interaction term for popular shows
and liking of ads at time 1 on problems at time 4 (see Table 3-4), which included the
direct effect on time 4 problems and indirect effects through the intercept and slope (delta
method total effect: b=.164, p=.02). These effects among females and males were
significant even after adjustment for time 1 problems, age, friends drinking, adults
drinking, playing sports, general TV watching, acculturation, parents’ jobs, parents’
education, and clustering by school.

144
The intercepts shown in Table 3-4 for the latent growth factors were both
significant for males and females (both p<.001), but there was no significant difference
for the intercept factors between males (95%CI: 0.433 - 0.558) and females (95%CI:
0.463 - 0.590) or for the latent growth slope factor for males (95%CI: 0.237 – 0.646) and
females (95%CI: 0.380 - 0.610). The intercept for the time 4 alcohol-related problems
was fixed at zero for females by default, and the intercept for this factor among males
was non-significant (b=.232, p=.21).The residual variances for the latent growth
intercept and slope and for the time 4 problems factor were each significant for males and
females (all p<.01).
The covariates, alcohol-related problems at time 1 and friends and close adult
drinking at time 1, were significant predictors of the intercept for females (all p<.05) as
shown in Table 3-4. The same covariates for males plus language acculturation and
parent jobs were significant predictors of the intercept (all p<.05). For males, drinking by
friends and language acculturation were significant predictors of the slope (both p<.05),
and once again, the sign of the coefficients for these predictors changed between the
intercept and the slope suggesting that those higher in alcohol use at time 1 might have
had lower growth rates than those lower in use at time 1. None of the time 1 variables
were significant predictors of alcohol-related problems at time 4 for males or females (all
p<.05).

145

Figure 3-6 Mediation Model for Alcohol-Related Problems.
Notes: Alcohol use = past 30 days + past 6 months; I = growth curve intercepts; S = growth curve slopes; Standardized parameter
estimates: male/female (p<.05); Adjusted for wave 1 problems, age, drinking peers, drinking adults, playing sports, general TV
watching, acculturation, parents’ jobs, parents’ education, and clustering by school; Fit indices: χ
2
(130)=182.66, p=.002; CFI=.98;
TLI=.97; RMSEA=.015, SRMR=.026.
Wave 1
Alcohol
Wave 3
Alcohol
Wave 2
Alcohol
I S
Like Ads
Shows x Like
Popular Shows
.171/.267
.506/.466
.349/.404
.643/.590 .821/.759
.549/.640
.326/.424
.849/.661
.686/.921 .253/.386
Wave 4
Problems
P1
P2
P3
P4
.648/.404
.458/.503
.720/.707
.721/.692
.736/.705
.780/.734
ns/.393
.ns/.478
ns/.190
.631/.513
.093/.091
.129/.ns
.482/.500
.481/.522
.392/.462

146
Table 3-4 Standardized Parameter Estimates for the Mediation Model.
Girls Boys
Parameter
Estimate
SE Parameter
Estimate
SE
Intercept on
T1 alcohol use .759*** .046 .821*** .038
T2 alcohol use .590*** .060 .643*** .047
T3 alcohol use .466*** .056 .506*** .030
Slope on
T1 alcohol use .000 .000 .000 .000
T2 alcohol use .404*** .036 .349*** .057
T3 alcohol use .640*** .056 .549*** .101
T4 alcohol-related problems on
T4 problems 1 .707*** .029 .720*** .035
T4 problems 2 .692*** .039 .721*** .056
T4 problems 3 .705*** .038 .736*** .048
T4 problems 4 .734*** .050 .780*** .037
Intercept on T1 predictors
Popular shows -.052 .034 -.027 .031
Liking of ads .267*** .047 .171*** .028
Shows x Liking .091* .042 .093* .046
T1 problems .297* .123 .264** .084
Age .030 .031 .040 .030
Peer drinking .426*** .060 .539*** .052
Playing sports .006 .043 -.009 .024
Adult drinking .155*** .036 .138** .053
General TV viewing .012 .034 .012 .037
Language acculturation .050 .042 -.098* .040
Parents’ jobs .000 .041 .112* .046
Parents’ education -.041 .045 -.002 .030
Slope on T1 predictors
Popular shows .190** .058 .113 .063
Liking of ads -.021 .078 .129* .060
Shows x Liking -.083 .068 -.112 .081
T1 problems -.125 .135 .076 .156
Age .031 .039 -.075 .068
Peer drinking .057 .075 -.483*** .128
Playing sports -.137 .073 -.015 .074
Adult drinking -.029 .067 -.103 .119

147
Table 3-4, Continued.
General TV viewing -.021 .064 -.059 .062
Language acculturation .029 .073 .227* .097
Parents’ jobs .130 .075 -.135 .109
Parents’ education -.085 .064 .009 .090
T4 alcohol-related problems on
Intercept .393* .166 .177 .303
Slope .478*** .106 .179 .214
Popular shows -.054 .065 -.007 .058
Liking of ads -.102 .064 -.095 .062
Shows x Liking .040 .072 .167 .094
T1 problems .050 .070 .014 .090
Age .036 .049 -.004 .034
Peer drinking -.022 .085 .234 .214
Playing sports .050 .059 .027 .044
Adult drinking -.027 .041 .021 .074
General TV viewing .022 .062 -.021 .063
Language acculturation .013 .063 -.048 .086
Parents’ jobs -.003 .103 .061 .092
Parents’ education .006 .064 -.018 .100
Intercepts for latent factors
Problems with alcohol at T4 .000 .000 .232 .184
Growth curve intercept .526*** .032 .496*** .032
Growth curve slope .495*** .059 .441*** .104
Residual variances
T4 problems 1 .500*** .041 .482*** .050
T4 problems 2 .522*** .054 .481*** .080
T4 problems 3 .502*** .054 .458*** .071
T4 problems 4 .462*** .074 .392*** .058
T1 acohol use ..424*** .070 .326*** .063
T2 alcohol use .513*** .049 .631*** .046
T3 alcohol use .404*** .075 .648*** .067
Intercept .386*** .096 ..253** .077
Slope .921*** .055 .686*** .087
T4 alcohol-related problems .661*** .061 .849*** .054
Effects from Shows to Problems
Total .017 .051 .008 .046
Total indirect .070 .042 .015 .028
Indirect Shows – I – Problems -.021 .018 -.005 .009
Indirect Shows – S – Problems .091* .040 .02 .028

148
Table 3-4, Continued
Direct Shows – Problems -.054 .065 -.007 .058
Effects from Liking to Problems
Total -.007 .063 -.042 .041
Total indirect .095 .057 .053 .058
Indirect Liking – I – Problems .105* .048 .030 .052
Indirect Liking – S – Problems -.010 .038 .023 .030
Direct Liking – Problems -.102 .064 -.095 .062
Effects from Interaction SxL to
Problems

Total .036 .066 .164* .069
Total indirect -.004 .040 -.004 .045
Indirect from SxL – I – Problems .036 .026 .016 .032
Indirect from SxL – S – Problems -.039 .031 -.020 .029
Direct from SxL – Problems .040 .072 .167 .094
Notes: T1= time 1, T2=time 2, T3=time3; * p<.05; **p<.01; ***p<.001; na=not
available, slope variance fixed at 0; I=intercept factor for growth curve, S=slope factor
for growth curve; SxL=interaction term for popular shows and liking of alcohol ads.

Mediation models for the other 3 exposure measures fit the data very well (i.e.,
CFI>.97 and RMSEA<.020). In all 3 models for females, the intercept for the growth of
alcohol use mediated the influence of liking of alcohol ads at time 1 on alcohol-related
problems at time 4 (joint significance and delta method: all p<.05). No other indirect
effects were significant for females or males. There was 1 significant interaction for
males that was between cued recall of alcohol ads and liking of alcohol ads in the
prediction of the growth intercept (p<.05). The sign for this interaction term was
unexpectedly negative, which suggested that for higher levels of cued recall for alcohol
ads, those with more liking of alcohol ads would use less alcohol at time 1 than those
who liked ads less. There was one significant interaction for females that was between
self-reported exposure to alcohol ads and liking of ads in the prediction of alcohol-related
problems at time 4 (p<.05). As expected, the interaction term was positive resulting in
curves similar to those shown in Figure 3-5 such that for higher levels of self-reported

149
exposure to alcohol ads, those reporting more liking of ads would report more alcohol-
related problems at time 4. The only other significant effect of a time 1 covariate on time
4 problems was in the cued recall mediation model where liking of alcohol ads by males
significantly and negatively predicted alcohol-related problems (p<.05). This negative
parameter was in contrast to a significant positive prediction of the intercept by liking of
ads in the same model. In these three mediation models for females, both the intercept
and slope for the growth of alcohol use were positive predictors of the level of alcohol-
related problems at time 4 (all p<.05) whereas this was not the case for males.
The pattern of relationships among the control covariates and the growth curve
factors were similar in these three models as those observed above in the mediation
model for exposure to alcohol ads on popular shows. In particular, alcohol-related
problems at time 1, friends drinking, and adults drinking were significant predictors of
the intercept but not the slope for females (all p<.05). In the case of males, problems,
friends drinking, adults drinking, acculturation, and parents’ jobs predicted the intercepts
in each mediation model, and friends drinking and acculturation predicted the slopes (all
p<.05). As observed in the growth models, the direction of the prediction for friends
drinking and acculturation changed between the intercept and slope factors for males.
Discussion
The current study provides evidence supporting the hypothesis that exposure to
alcohol advertising and affective reactions to those advertisements on television influence
underage drinking and the development of alcohol-related problems. Measures of
exposure to alcohol ads includes self-reported frequency of viewing alcohol ads, memory

150
for alcohol ads cued by still pictures taken from TV ads, and indirect measures of
exposure for which students report the frequency of watching various popular and sports
shows on TV. The frequency of watching shows is weighted by the number of alcohol
ads on those shows as reported by the Nielsen Corporation. The popular shows portion of
this indirect measure provides the best evidence for the influence of alcohol ads on
underage drinking. The growth of alcohol use from the 7
th
through the 9
th
grades is
predicted by the frequency of watching popular shows and self-reports on the liking of
alcohol ads. In partial support of hypothesis 1, there is a significant interaction between
exposure to ads and liking of ads in the prediction of the intercept (but not the slope) for a
growth curve modeled across these grade levels for both male and female students. The
interaction shows that the level of exposure to ads is more predictive of a higher level of
alcohol use in 7
th
grade for those students who report a greater liking of alcohol ads. In
addition to this interaction observed at time 1, the frequency of watching popular shows
at time 1 predicts the slope for the growth of alcohol use for females, and the liking of
alcohol ads at time 1 predicts the slope for males.
In support of hypothesis 2, a mediation model shows that the influence of alcohol
ads on the occurrence of alcohol-related problems at time 4 is mediated by the growth of
alcohol use. There was a significant indirect effect of exposure to ads on popular shows
in time 1 on problems in time 4 through the growth of alcohol use among females and a
significant total effect from the shows and liking interaction term in time 1 to problems in
time 4 among males. These relationships are significant even after adjusting for a range
of other covariates measured at time 1 that are known to be associated with alcohol use

151
including age, alcohol-related problems, observing friends drink, observing close adults
drink, playing sports, general TV watching, acculturation, and socio-economic status.
The other three measures of exposure to alcohol advertising show similar findings
although these measures are somewhat less predictive of the growth in alcohol use and
alcohol-related problems.
Differences in the models by gender did not support hypothesis 3. Although males
and females regression coefficients differed somewhat across models, alcohol advertising
appears to influence both genders to drink more and the increased drinking leads to more
alcohol-related problems. Males and females were nearly equal in the influence of
exposure and liking of ads on the initial level (intercept) of alcohol use. In addition, there
was mediation for both genders, which was evidenced by a significant indirect effect of
exposure to ads through the growth of alcohol use on alcohol-related problems in time 4
among females and a significant total effect of the interaction between exposure and
liking on problems in time 4 including the indirect effect through the growth of alcohol
use among males.
The growth and mediation models in the current study are particularly well suited
to inferences about the causal influence of exposure to alcohol ads on increased use of
alcohol and alcohol-related problems among adolescents. Measures of exposure at time 1
are used to predict the increasing use of alcohol over time and the development of
alcohol-related problems at time 4. These relationships satisfy one necessary condition
for causal inferences, which is a temporal ordering of predictors and outcomes. In
addition, the models for this study control for a range of potentially confounding

152
variables including strong predictors such as prior alcohol-related problems and peer
influences. Findings in the current study are consistent with previous research, which also
strengthens inferences. The indirect measure of exposure to alcohol ads on popular shows
is predictive of alcohol use (Strickland, 1983; Unger et al., 2003) and measures for liking
of alcohol ads are predictive of alcohol use (Casswell & Zhang, 1998; Wyllie et al.,
1998b). Finally, the findings here are consistent with well established theories on
vicarious learning (Social Learning Theory: Bandura, 1977) and persuasive messages in
the media (e.g., Elaboration Likelihood Model: Petty & Wegener, 1999). This
combination of factors provides reasonably good support for the influence of exposure to
alcohol advertisements on alcohol use and alcohol-related problems among adolescents.
The results observed for covariates in the current study are similar to those
reported in previous studies on risk factors for alcohol use among adolescents. Covariates
found to be significant in the current study that are also reported in the literature include
drinking by friends (Feldman et al., 1999), drinking by well known adults (Wood et al.,
2004), and acculturation (Fosados et al., 2007) An interesting effect is observed for males
where the sign of the regression coefficient for drinking by friends changes between
predicting the intercept and slope for the latent growth of alcohol use. Drinking by
friends is a positive predictor of the intercept and a negative predictor of the slope. The
result suggests that those who had a higher number of friends who drank at time 1 also
drank more themselves at time 1 (intercept), and that these same persons had smaller
growth rates (slopes) in alcohol use over time. In other words, those who start out at a
higher level of drinking may not increase their level of drinking over time as much as

153
those who start out at a lower level of drinking. If this is the case, however, a negative
correlation is expected between the intercept and slope for males, but this association did
not reach the level of significance. These conflicting results suggest that further study is
required before drawing inferences from these covariate results.
The current study has a few limitations that warrant discussion. First, all
observational studies including this one are limited in their ability to control for 3
rd

variable effects. Although adjustments for known potential confounders help control for
these potential effects, there are other variables that might also have some effects (e.g.,
depression or anxiety), which cannot be controlled except in a randomized study using a
control versus experimental group design. Second, the current results are technically only
generalizable to public school students in the Los Angeles area. Third, the current study
relies on self-reports without any biochemical assays to validate the measures, but prior
research has shown that self-reports can be reliable and valid (e.g., Stacy, Widaman,
Hays, & DiMatteo, 1985) if certain safeguards are undertaken, such as assurances of
confidentiality, as they were in this study. Fourth, alcohol use measures among young
adolescents are often skewed toward zero and this is true in the current sample. Seventh
graders were actually recruited because of their low levels of alcohol use in order to
examine the early development of alcohol use, but unfortunately, these skewed measures
may have contributed, in part, to some of the null findings in this study. Finally, not all
results converge in the current study across multiple measures of exposure to advertising,
but there is little literature available that indicates which exposure measures are optimal.
However, the use of the indirect measure of exposure on popular shows and liking of ads

154
are used successfully across a range of studies, and in particular, liking of ads, although
not strictly a measure of exposure, is used across product categories to predict the success
of individual ads or ad campaigns (e.g., Haley & Baldinger, 1991).
The influence of televised alcohol advertisements on underage drinking has
important implications for prevention programming. First, children can be taught about
the design of persuasive messages in the media early to help them avoid undue influence
by the media on their behaviors (e.g., Austin & Johnson, 1997). The interaction of liking
of alcohol ads and exposure to ads in the current study in the prediction of alcohol use
suggests that a media literacy approach might have some important protective effects on
alcohol use among young adolescents. Second, policy regarding televised alcohol ads
should be designed to limit exposure to these ads by young adolescents. In regards to
advertising policy, it is important to have a comprehensive policy that not only limits
televised ads but it is also likely that web, print, and display ads may need some
regulation to prevent redistribution of advertising funds to these media outlets. Although
there are other influences on underage drinking, such as observation of friends and adults
drinking as noted in the current study, prevention strategies should address the influence
of alcohol ads as part of an overall strategy to prevent early initiation of alcohol use and
the development of problems related to consumption.

155
Chapter 4 Exposure to Alcohol Advertising and the Development of Alcohol-Related
Associations in Memory
Abstract
The results for this study were consistent with the hypothesis that exposure to
televised alcohol commercials contributes to the development of spontaneous alcohol-
related associations in memory. Paper-and-pencil questionnaires that included word
association tasks were administered to 3,890 students in a prospective study from the 7
th

through the 9
th
grades. Results for parallel growth modeling of alcohol-related
associations and alcohol use showed that cued recall of alcohol ads, self-reported
exposure to alcohol ads, and liking of alcohol ads predicted the number of alcohol-related
associations in the 7
th
grade, and liking of alcohol ads predicted the growth of
associations over time. The initial level of alcohol use did not predict the growth of
associations, but the growth of alcohol use was significantly, and very strongly,
correlated with the growth of alcohol-related associations over the 3 time periods.
Introduction
One important risk factor for the use of substances such as alcohol by adolescents
is the development of substance-related associations in memory. Measures of these
associations are predictive of alcohol (Kelly, Masterman, & Marlatt, 2005; Stacy &
Newcomb, 1998) marijuana (Ames & Stacy, 1998; Stacy et al., 1996; Stacy et al., 1997),
and tobacco (Grenard et al., in press; Kelly, Masterman, & Marlatt, 2006) use even after
controlling for other risk factors such as sensation seeking characteristics, peer
influences, and substance use expectancies. In addition, Stacy (1997) found that memory

156
associations predicted alcohol and marijuana use prospectively after adjusting for prior
drug use (see also, Kelly, Haynes, & Marlatt, 2008; Thush & Wiers, 2007). Despite the
importance of associations in memory as a risk factor for substance use, there are few
reports on what influences the development of these associations in memory. The current
study examines the influence of televised alcohol advertising on the development of
alcohol-related associations in memory.
A number of studies have demonstrated the more general influence of
spontaneous associative memory processes on judgment and behavior in dual-process
theories of cognition (e.g., Damasio, 2003; Kahneman, 2003; Schacter, 1987). These
spontaneous or implicit processes have been described as the ‘default’ process in
judgment and choice whereas deliberative or explicit processes require more cognitive
effort in making judgments (Bargh & Morsella, 2008; Kahneman, 2003). The current
research, however, does not address the influence of these processes on decision-making.
Instead, the current study examines the development of associations in memory over time
as a function of life experiences with alcohol including exposure to alcohol
advertisements. A number of theories predict the development of automatic or implicit
associations in memory (see text below), but few studies have attempted to demonstrate
the prospective development of these associations. One difficulty in attempting a study of
this type is the measurement of automatic associations, which are likely to be unavailable
to self-reflection (Greenwald et al., 1998; Nisbett & Wilson, 1977; Schacter, 1987). Word
association measures can be constructed as indirect measures and have been used
successfully to study implicit or automatic associations in basic research in cognitive

157
science (McEvoy et al., 1999; Nelson et al., 2000) and in applied studies of health
behavior (Ames et al., 2007; Stacy, 1997; Szalay et al., 1996). The current study used
word association measures to study the development of automatic or implicit associations
in memory.
A number of empirical findings show that word association can measure
relatively spontaneous, automatic, or implicit cognitive processes (for review, see Stacy
et al., 2006). First, amnesic patients who are severely limited in conscious or explicit
memory exhibit indirect priming effects indicative of implicit memory associations (e.g.,
Levy et al., 2004). Second, participants self-report that during word association tasks they
do not use intentional retrieval (explicit memory) processes (Mulligan, 1998: experiment
2). Third, dissociations between explicit memory tasks (e.g., cued recall) and word
association have been successfully demonstrated (e.g., Goshen-Gottstein & Kempinsky,
2001). In addition, studies have shown that conceptual priming, not just perceptual
priming, occurs in word association. In one study for example, word association
responses in a priming task depended upon the semantic context of the target words
presented during the study phase (Zeelenberg et al., 2003), and in a second study, a verb
generation task (controlled word association) was used by Segar, Rabin, Desmond and
Gabrieli (1999) to demonstrate that implicit priming occurred at a conceptual level across
languages. These studies show that word association tasks measure an associative
processing system directly implicated in implicit cognitive processes in comparison to
more deliberative, rational, or explicit processes.

158
Word association norms have been collected from a wide range of populations
(e.g., de La Haye, 2003; Nelson, McEvoy, & Schreiber, 2003; Palermo & Jenkins, 1964;
Postman & Keppel, 1970), and these norms have been used in cognitive science to study
memory processes. For example, Nelson, McEvoy, and associates used their association
norms to model how the activation of word associates in memory facilitated cued recall
and recognition during new associates learning (e.g., Nelson & McEvoy, 2005; Nelson,
McEvoy, & Pointer, 2003). Roediger and his colleagues studied illusory memory
(Roediger & McDermott, 1995; Watson, Balota, & Roediger III, 2003) based upon the
work of Deese (1959b) who showed that memory intrusions in free recall were predicted
by associative frequencies in word association norms. Spence and Owens (1990) found
that the strength of association between two nouns in word association norms was
inversely related to the distance between nouns in English texts suggesting that the use of
the first noun activates associates in the author’s memory and increases the likelihood for
use of the second noun. In an applied study of memory associations and consumer
behavior, Berger and Fitzsimons (2008) showed that environmental stimuli, which are
associated with a consumer product in memory, can increase fluency and positive
evaluations of that product thereby increasing the likelihood of purchase. These studies
provide evidence that word associations in memory have important implications for
understanding cognitive processes and behavior.
Various approaches are used in cognitive science to model the development of
associations in memory. Anderson (1983) provided a description of a version of the
Adaptive Control of Thought (ACT) model of spreading activation in memory (see also J.

159
R. Anderson et al., 2004). Key elements of the ACT model predict that multiple
experiences increase the strength of a memory trace, and the likelihood of recalling a
trace depends upon the associative strength of that trace with other traces in memory.
Hintzman (1984, 1986) proposed a multiple trace model of memory. In this model, each
experience creates a new trace in memory, and the strength of memory activation
depends in part on the number traces with similar features (i.e., more experiences lead to
stronger memory activations). Hopfield and Tank (1986) described a model for
optimization computations that was based upon a distributed network of neuron-like
elements with symmetrical connections. The association between a word printed on a
page, for example, and its semantic meaning in memory is determined by the most stable
state reached in the activation of the network. Learning is accomplished through repeated
activations of the network.
Each of these models suggests that associations among concepts as represented by
associations among words in memory should develop over time as experiences
accumulate. Word associations would then be expected to change across historical time
as the meanings or popularity of words change in the culture and across the development
of children as they age. There is evidence that historical changes do occur. Jenkins and
Palermo (1965) examined 7 word association norms collected between 1910 and 1960
and found that associations change gradually over time. For example, the percentage of
identical primary responses with those obtained in 1910 were 81%, 79%, 74%, 66%,
58%, and 61% in 1925, 1927, 1933, 1942, 1952 and 1960, respectively. The strongest

160
associates in 1910 were the most stable across the time periods. Seventy-one of the most
common responses to 100 cue words were the same in 1927 and in 1960.
There is also evidence for a change in word associations among children as they
develop. Palermo and Jenkins (1964) reported word association norms for 200 words
collected in a cross-sectional study of the development of associations among students in
grades 4, 5, 6, 7, 8, 10, 12, and college. An examination of the response frequencies
showed that for some cue words there is little change in the responses across those ages,
but responses for other cue words do change across the grade levels. For example, the
response ‘drink’ to the cue word ‘whiskey’ remains somewhat steady across the age
groups at roughly 60% whereas the response ‘wine’ declined gradually from 10% among
4
th
and 5
th
graders to 1% among college students. Palermo (1971) observed
developmental changes in word association norms collected in a cross-sectional study of
1
st
through 4
th
grade students. The authors reported that the trends observed in these
younger participants matched trends observed previously in data collected from 4
th

through college level participants (Palermo & Jenkins, 1964). The frequency of the most
popular responses increased from the 1
st
grade through college as did contrast and
paradigmatic (e.g., noun response to a noun cue) responding.
In a cross-sectional analysis of age differences in associative memory, Coronges,
Stacy, and Valente (2007) found a difference between 7
th
grade and college age students
using social network analysis techniques on free-word association responses. The authors
compared network measures across the groups for associations to 16 ambiguous cue
words (e.g., count, draft, and field). College students provided more unique responses

161
than the younger students, and according to the authors, this suggests that the older
students had more associations due to the accumulation of more life experiences. The
network for the college students was more centralized, however, indicating that the
responses that were given more frequently by college students played a stronger role as
‘hubs’ in associative memory processes and provided a more efficient organization of
concepts in memory. The authors conclude that these differences and others observed
between the groups of students provide evidence that the structure of associations in
memory develop with age and life experiences, and that individual differences in this
development are likely.
In a rare study of predictors for the development of word associations, Stacy,
Leigh, and Weingardt (1997) found that prior behavior predicted responses on a word
association test. In a cross-sectional study among 1,003 undergraduates, the self-reported
number of days using alcohol, studying chemistry or physics, or using computers in the
prior 30 days significantly predicted whether the participants responded to ambiguous
word cues with words related to one of the target categories. The authors concluded that
these results were consistent in general with a number of associative theories of memory
in social and cognitive science, and with the view that individual differences in
experiences shape associations in memory.
Few studies have examined individual and group differences in the development
of associations in memory prospectively despite the evidence that these associations have
an important influence on consumer behavior. This study tested the following
hypotheses:

162
H1. Exposure to alcohol advertising, liking of those ads, and the growth of alcohol
involvement will influence the development of alcohol-related associations in
memory over time after adjusting for time 1 covariates including age, gender,
acculturation, observed peer and adult alcohol consumption, participation in sports,
and socio-economic status.
H2. The relative frequency of associations in memory for two or more (dictionary)
meanings will change for some but not all of the 18 homographs examined in this
study from time 1 through time 3. The proportion of responses coded in the alcohol-
related categories for the homographs bud, draft, hammered, pitcher, shot, and tap are
expected to change over time due to increasing life experiences related to alcohol For
example, adolescents are likely to transition from watching children’s TV programs
to more mature programs that may be sponsored by alcohol producers, and
adolescents are likely to be exposed to peers using alcohol. Other homographs for
which meaning categories are not expected to change include count, ground, lead,
neon, pupil, rash, and shed. It is anticipated that for these latter words the meaning
categories among adolescents are well established at a younger age (cf., Palermo &
Jenkins, 1964).
Methods
Participants
The self-report survey data were collected as part of a prospective study of
televised alcohol advertising (Stacy, Zogg et al., 2004; Zogg, 2004). Participants
recruited from public schools in Southern California completed surveys during regular

163
school hours once per year from the 7
th
through the 10
th
grades. A total of 4,186 students
agreed to participate in the study and 3,890 (93% of consented) students completed
surveys in at least one wave. Of the total surveyed 2,986 (77%) completed surveys in 7
th

grade, 2,849 (73%) in the 8
th
grade, and 2,093 (54%) in the 9
th
grade.
Schools were selected randomly from a list of all public middle schools in Los
Angeles County to be recruited for the study, and 23 schools across 11 school districts
agreed to participate in the study. A few schools that were approached did not agree to
participate, which resulted in some difference between the study sample and the
population characteristics of Los Angeles County. All students in the 7th grade at each
school were invited to participate in the study. The students signed assent forms, and their
parents or guardians signed parental consent forms or gave verbal consent over the
telephone. The procedures were approved by the university institutional review board.
Procedure
Students completed paper-and-pencil questionnaires in class during regular school
hours. Two forms of the survey were randomly assigned to participants at the school
level. The forms were the same except that one form of the survey included the cued
recall measure for alcohol advertising on TV while the second form included a top of
mind awareness measure, which is not included in the current study due to poor
psychometric properties.
Measures of drug-related associations in memory
Word association tasks. Word association tasks have been predictive of substance
use (Stacy, 1995, 1997). These assessments used indirect instructions such that no

164
references were made about alcohol advertising or alcohol use in the instructions or task
even though the responses were later coded for alcohol-related references. In the cue-
behavior association task (CBAT), each participant was asked to respond to 18
ambiguous cue words (homographs) with a single word that first came to mind. The cue
words were presented in random sequence to help control for order effects. Students were
instructed as follows: “Write next to each word the first word it makes you think of. For
example, if the word is ‘doctor,’ you might write ‘nurse.’” Target cue words for this
measure included five ambiguous words that were selected to potentially elicit alcohol
related responses (bud, tap, pitcher, draft, hammered, and shot), and the remaining eleven
words were ambiguous filler words (season, shed, count, neon, ground, rash, ram,
program, lead, screen, and pupil) that were not expected to elicit alcohol related
responses. The responses given by the participants to each cue word were entered
verbatim into a computer based coding system (Ames et al., 2005). Two judges
independently coded each response: 1 (related to alcohol beverages, consumption, or
associated activities) or 0 (not related to alcohol). If there was a discrepancy between the
two judges, a third judge obtained a consensus coding. The inter-rater agreement was
good. For example, the kappa values in time 3 for the cues tap, draft, hammered, and shot
were greater than .80, and the kappa for the cue pitcher was .60. The test score for the
cue-behavior task was the sum of the coded responses for the six target words (0=no
alcohol related responses and 6=alcohol related responses to all six words).
The responses to the homographs in the CBAT were coded a second time by a
second group of judges to assess the relative change in meanings for the words across

165
time. Two judges were asked to independently assign each response to 1 of 2 or more
dictionary-type meanings for each of the 18 homographs. A third judge obtained a
consensus when the first two judges did not agree on a meaning. The inter-rater kappa
values ranged from .45 to .71, which was fair to good agreement beyond chance (Fleiss,
Levin, & Paik, 2004, p. 604). The proportion of responses attributed to each meaning was
calculated for each homograph.
The cue-outcome-behavior task (COBT) used short phrases instead of single
ambiguous cue words, and the participants were asked to respond with the first behavior
or action that came to mind. This task is a form of verb generation, which has been found
to assess an implicit conceptual form of memory (Seger et al., 1999) as has free word
association. The cue phases were self-generated by similar populations of participants in
previous research (Stacy, 1995, 1997; Zogg, Ma, Dent, & Stacy, 2004) and included 11
target phrases that did not mention alcohol (having fun, being more social, forgetting
problems or worries, having a good time, feeling more relaxed, feeling good, feeling
buzzed, being cool, feeling happy, pleasant time, and tasting something good). In
addition, two filler phrases were included (getting good grades and being nice). The
participants were instructed as follows: “For each result below, write the first behavior or
action it makes you think of.” The responses to the 13 outcome cues were entered
verbatim into a computer based coding system, and judges coded these responses in the
same manner described above for the cue-behavior task. The inter-rater agreement was
good. For example, all kappa values in time 3 for the 11 target cues were greater than .70.

166
The codes for the 11 target words were summed to provide a score for alcohol (0 = no
alcohol-related associations and 11 = all alcohol-related associations).
Measures related to televised alcohol advertising exposure
Cued recall. Participants viewed pictures captured from 2 example and 15 test
commercials (Unger et al., 1995). Among the 15 test commercials, 9 were alcohol
advertisements that were being aired at the time of the survey, 3 were advertisements for
non-alcoholic drinks, 1 was for an electronics retail outlet, 1 was a public service
announcement, and 1 was an old beer commercial that was out-of-date but might elicit
guessing. Participants were asked to write the name of the product or thing advertised.
Independent judges coded the responses as being related to the commercial or not (1/0).
A cued recall index was calculated as the number of alcohol commercials where the
brand was correctly identified (range: 0-9; coefficient alpha = 0.58). In addition, a false
positive recall index was scored as the number of non-alcoholic product commercials that
were incorrectly identified as being related to alcohol (range: 0-4).
Exposure to alcohol advertising on popular shows. Participants marked how often
they watched each one of 20 popular television shows during the past month on a 6-point
scale ranging from 1 (never) to 6 (every day). The shows were selected based upon the
size of the teen audience and the number of alcohol commercials aired on the programs in
the 6 months prior to administering the surveys. Examples of these popular shows include
Seinfeld, Friends, and Soul Train. The frequency of watching the shows was weighted by
the mean frequency of alcohol advertising on each show (Strickland, 1983) as reported by

167
Nielsen Media Research over a 10-month period prior to each wave of surveys. The mean
of this weighted frequency comprised the participant’s score (coefficient alpha = .79).
Exposure to alcohol advertising on sports programs. Alcohol advertising occurs
more frequently on televised sporting events (Madden & Grube, 1994) than other
programming, so a separate instrument was created for this type of exposure. Participants
were asked how many times in the past month they watched the following broadcast
events: professional baseball, college and professional basketball, professional soccer and
hockey, and Sports Center on ESPN. Response options ranged from 1 (never) to 6 (every
day). These responses were weighted with the average monthly frequencies for alcohol
advertising on each show across a 10-month period prior to each wave of data collection
(Strickland, 1983). The frequencies were obtained from Nielson Media Research. An
index score was calculated as the mean of the 7 items (coefficient alpha = .80).
Self-reported observation of alcohol advertising. Participants were asked 4
questions about how pervasive they thought that alcohol commercials were on TV. The
first 2 questions were adapted from Schooler, Feighery, and Flora (1996), “When you
watch TV, how often do you see commercials for alcohol drinks, like beer, wine, or
liquor?” The response options ranged from 1 (a lot) to 5 (I never watch TV). The second
question asked, “In the past week, how many TV commercials have you seen for alcohol
drinks, like beer, wine, or liquor (circle ONE answer).” The response options ranged
from 0 (none) to 6 (6 or more commercials). The third and fourth questions asked
specifically about what the participants did in the past 6 months, “Saw a beer commercial
on TV?” and “Saw wine or liquor advertised on TV?” Response options for these last two

168
items ranged from 1 (every day) to 7 (never). The mean of the 4 items comprised the
index score (coefficient alpha = .78).
Liking of alcohol advertisements. Liking of advertisements has been shown to be
a valid method of ad copy testing by predicting sales for a number of consumer products
(Haley & Baldinger, 1991). In addition, alcohol consumption among adolescents and
young adults has been predicted by desirability and identification with characters in
alcohol advertisements (Austin et al., 2006) and with liking of advertisements (Casswell
& Zhang, 1998; Wyllie et al., 1998b). Three items measured how well participants in the
present study liked the alcohol advertisements they observed on TV (Unger et al., 2003).
The first two questions asked “When you see alcohol commercials on TV….” (a) “Do
you think they are funny?” and (b) “Do you think they are sexy?” Response options
ranged from 1 (yes, always) to 5 (never saw any). The third question asked “Of all the
commercials you see on TV, how much do you like the TV commercials for alcohol?”
The response options ranged from 1 (I like the alcohol commercials the most) to 5 (I have
never seen an alcohol commercial on TV). The 3 items were reverse coded such that a
higher score indicated greater liking, and the mean of the items comprised each
participant’s score (coefficient alpha = .72).
General television viewing frequency. The amount of time spent viewing
television has been associated with alcohol consumption and exposure to alcohol
advertising (Grube, 1995; Robinson et al., 1998). The first 3 of the 7 total items in the
scale assessed weekday viewing: “On a typical weekday, how many hours a day do you
watch TV…” (a) “before school,” (b) “after school before dinner,” and (c) “from dinner

169
until bedtime.” The next 4 items included the following: “On a typical weekend, how
many hours a day do you watch TV…” (d) “Saturday morning until noon,” (e) “Saturday
noon until bedtime,” (f) “Sunday morning until noon,” and (g) “Sunday noon until
bedtime.” The response options for the 7 items ranged from 1 (I do not watch TV) to 5 (5
hours or more). The mean of the 7 items formed the scale score (coefficient alpha = .79).
Alcohol use measures
Current alcohol use. Alcohol use was assessed for the past 30 days (Kann, 2001)
and past 6 months. Past 30 day use included 5 items asking how many days participants
had…. (1) “had at least one drink of beer,” (2) “had at least one drink of wine or liquor,”
(3) “had 3 or more drinks of beer in a row,” (4) “had 3 or more drinks of wine or liquor in
a row,” and (5) “drank enough to get drunk.” The 8 options ranged from: 1 (0 days) to 8
(all 30 days). Past 6 months use included 4 items that asked how often in the past 6
months participants had… (1) “drank beer,” (2) “drank wine or wine coolers,” (3) “drank
liquor,” or (4) “got drunk.” Options ranged from 1 (every day) to 8 (never). The 6 month
questions were reverse coded and then an index was formed from the mean of all 9 items
(coefficient alpha = .91).
Covariates and potential confounding variables. It was important to control for
variables that that have shown some relationship or have the potential to be related to
alcohol use or the development of alcohol-related associations in memory. Observing
peers and adults drink alcohol can influence underage drinking (e.g., Feldman et al.,
1999; Wood et al., 2004) as well as the development of associations in memory.
Therefore, each participant was asked to respond to questions about drinking by friends

170
and drinking by adults the participant knew well. Participation in sports increases the
likelihood of underage drinking (Aaron et al., 1995; Leichliter et al., 1998; Wechsler et
al., 1997), so participants were asked about how often in the past 6 months they had
played football, baseball, basketball, hockey or soccer, and other sports (Thorlindsson et
al., 1990). Demographic variables that were assessed included age, gender, ethnicity,
language acculturation (Marin et al., 1987; Stacy, 1995), and parents’ occupation and
education.
Data Analysis
Descriptive statistics were calculated for demographic and alcohol use variables
among all participants. A two step process was then used to construct the structural
equation models used to test the hypotheses (J. C. Anderson & Gerbing, 1988). First,
confirmatory factor analyses (CFA) of the measurement models established the simple
structure of the measures in the model such that each item loaded well on its
hypothesized latent factor and not on any other factors. The second step involved
sequential fitting of the structural models as follows: (a) latent growth curve models
across times 1 to 3 for the development of alcohol-related associations, (b) parallel
growth curves for the development of associations and the growth of alcohol use over
times 1 to 3, and (c) the addition of the time 1 alcohol advertising exposure variables and
control covariates (all continuous variables were centered on their means). One model
was fit to the CBAT data, and a second model was fit to the COBT data. Adjustments for
missing data in the current analyses utilized full information maximum likelihood
estimation (R. J. A. Little & Rubin, 2002) as implemented in Mplus statistical software

171
(Muthen & Muthen, 1998-2007). This method helps avoid bias in the estimation of
parameters and/or standard errors that can occur when ad hoc methods such as list wise
deletion or mean substitution are used for handling missing data (e.g., Schafer & Graham,
2002). Adjustments were made to the standard error estimates for clustering by school
(Murray & Short, 1995; Muthen & Muthen, 1998-2007). Fit of each model was assessed
with a range of goodness-of-fit statistics (for review, see Marsh et al., 2005) including the
Chi-squared test, Comparative Fit Index (CFI), Tucker-Lewis Index (TLI), Root Mean
Squared Error of Approximation (RMSEA), and the Standardized Root Mean Square
Residual (SRMR).
The relative changes in meanings for the homographs across time were assessed
using omnibus chi-square tests for proportions of responses in time 1 versus time 3.
Planned post hoc analyses were conducted on the alcohol-related categories for the 6
alcohol-related homographs (bud, draft, hammered, pitcher, shot, and tap) by creating a
dichotomous variable for each alcohol-related category for each homograph to indicate if
a participant’s response was related to the alcohol category (1) or not (0). A linear growth
curve model was fit to the data for the dichotomous variable across the 3 time periods for
each homograph. A significant, positive slope for the growth curve would indicate that
the relative frequency of responses assigned to the alcohol-related category was
increasing over the 3 time periods.
Results
Analysis of demographic data for year 1 indicated that the students in 7
th
grade
were 12.51 (SD=0.54) years old and included the following ethnicities: 13.37% non-

172
Hispanic Whites, 47.87% Latino, 17.02% Asian, 3.08% African American, 0.77%
Native Hawaiian or Pacific Islander, 0.95% American Native, 4.32% mixed, and 12.62%
didn’t know (see Table 4-1). The mean language acculturation was 4.22 (SD=0.76) and
this fell between the response categories for language spoken: 5 (only English) and 4
(English more than another language). The education level of parents reported by the
students was similar (p=.62) for mothers (M=3.49; SD=1.57) and fathers (M=3.52;
SD=1.63). The mean education level for each parent fell between the response codes for
3 (completed high school & received diploma) and 4 (some college or job training - 1 to
3 years). The coded occupation class that was based upon an open-ended response by the
student differed for mothers and fathers ( χ
2
(49)=186.81, p<.001). It appeared that more
mothers (24.77%) stayed at home than fathers (3.62%).Of those reporting some lifetime
use of alcohol, 45% reported drinking beer and 35% reported drinking wine or liquor, and
among those who reported alcohol use in the past month, 17% drank beer and 16% drank
wine or liquor (see Table 4-1).
Table 4-1 Study Three Demographic Information.
Item Total
Male Gender: N (%) 1894 (49.86)
Age: M (SD) 12.51 (0.54)

Ethnicity: N (%)

White / non Latino 520 (13.37)
Latino / Hispanic 1862 (47.87)
Asian 662 (17.02)
Black / African American 120 (3.08)
Native Hawaiian or Pacific Islander 30 (0.77)
American Indian or American Native 37 (0.95)
Don’t know 491 (12.62)
Mixed 168 (4.32)
Language acculturation: M (SD) 4.22 (0.76)

173
Table 4-1, Continued.
Education: M (SD)
1

Father 3.52 (1.63)
Mother 3.49(1.57)

Occupation of Father: N (%)
2

Unemployed, student, househusband/wife 50 (3.62)
Unskilled worker 53 (3.84)
Machine operator, cook, waitress/waiter 148 (10.71)
Electrician, plumber, tailor 552 (39.94)
Clerk, salesperson, flight attendant
mechanic, truck driver, military enlisted
188 (13.60)
Small business owner, manager 217 (15.70)
Teacher, engineer, nurse, pilot, military
officer
140 (10.13)
Doctor, lawyer, large business owner 34 (2.46)

Occupation of Mother: N (%)
2

Unemployed, student, househusband/wife 353 (24.77)
Unskilled worker 149 (10.46)
Machine operator, cook, waitress/waiter 117 (8.21)
Electrician, plumber, tailor 126 (8.84)
Clerk, salesperson, flight attendant
mechanic, truck driver, military enlisted
304 (21.33)
Small business owner, manager 162 (11.37)
Teacher, engineer, nurse, pilot, military
officer
191 (13.40)
Doctor, lawyer, large business owner 23 (1.61)

Prevalence of Alcohol Use
Lifetime beer use: P (SD) .43 (.50)
Lifetime wine/liquor use: P (SD) .35 (.48)
Past month beer use: P (SD) .17 (.37)
Past month wine/liquor use: P (SD)

.16 (.37)
Notes:
1
Education by parent gender t(3175)=-0.49, p=.62; ;
2
Occupation by parent
gender was significant ( χ
2
(49)=186.81, p<.001); P=proportion; SD=standard deviation.

The structural model results shown in Figure 4-1 and Table 4-2 for the parallel
growth of alcohol-related associations in memory measured using the CBAT and the
growth of alcohol use fit the data well (Fit: χ
2
(39)=69.98, p=.002, CFI=.99, TLI=.96,

174
RMSEA=.014, SRMR=.014). Liking of alcohol ads at time 1 (b=.158, p<.001) and self-
reported exposure to alcohol ads (b=.058, p<.05) were significant predictors of the
intercept for the growth of associations. Covariates that were positive predictors of the
intercept for associations included peer drinking and acculturation, and those that
positively predicted the slope included acculturation and parents’ occupation (all p<.05).
Contrary to hypothesis H1, none of the predictors related to advertising or the initial level
of alcohol use (i.e., the intercept for the growth in use of alcohol) significantly predicted
the growth of alcohol-related associations. The intercepts for the two growth curves were
not significantly correlated (p>.05). However, the two slopes were significantly, and
strongly, correlated (r=.46, p<.001). Measures that significantly predicted the initial level
of alcohol use included cued recall (b=-.098, p<.01), self-reported observation of alcohol
ads (b=-.052, p<.05), liking of alcohol ads (b=.281, p<.001), and the covariates peer
drinking (b=.573, p<.001) and adult drinking (b=.178, p<.001). In addition, the intercept
for the growth of alcohol-related association was a significant predictor of the slope for
the growth of alcohol use (b=-.168, p<.01).
The structural model shown in Figure 4-2 for the parallel growth of alcohol-
related associations in memory measured using the COBT and the growth of alcohol use
fit the data well (Fit: χ
2
(39)=61.85, p=.011, CFI=.99, TLI=.96, RMSEA=.012,
SRMR=.012). Significant time 1 predictors of the intercept for growth of alcohol-related
associations included cued recall of alcohol ads (b=.140, p<.01) and liking of alcohol ads
(b=.124, p<.01), and in partial support of hypothesis H1, liking of alcohol ads was a
significant predictor of the slope for the growth of associations (b=.183, p<.05).

175
Covariates that were positive predictors of the intercept included peer drinking,
acculturation, and parents’ occupation, and one covariate, playing sports, was a negative
predictor of the intercept (all p<.05). Acculturation was the only covariate that was a
significant, positive predictor of the slope for the growth of alcohol-related associations
(p<.001). The two intercepts (r=.36, p<.01) and the two slopes (r=.80, p<.001) for the
growth of associations and the growth of alcohol use were significantly correlated. There
was an especially strong correlation between the two slopes. Contrary to hypothesis H1,
however, the intercept for the growth of alcohol use did not significantly predict the slope
for the growth of alcohol-related associations. Measures that significantly predicted the
initial level of alcohol use included cued recall (b=-.093, p<.01), self-reported exposure
to alcohol ads (b=-.053, p<.05), liking of alcohol ads (b=.280, p<.001), and the covariates
peer drinking (b=.576, p<.001) and adult drinking (b=.179, p<.001).
The meaning categories for the homographs are shown in Table 4-3 with the
associated proportion of participant responses coded as being related to each category
across the three time periods. The measures of inter-rater agreement (percent agreement
and kappa) are shown in column 3. A chi-square, omnibus test for change in the response
frequencies between time 1 and time 3 indicated that the proportions of responses for one
or more categories changed significantly for each homograph (all p<.05). Post hoc tests
were planned on meaning categories related to alcohol to test hypothesis H2. The target
homograph cue words with alcohol-related categories included bud, draft, hammered,
pitcher, shot, and tap.

176

Figure 4-1 Growth of Alcohol-Related Associations for the CBAT.
Notes: CBAT = Cue Behavior Association Task; I = growth curve intercepts; S = linear slope; CBAT=cue behavior association task
coded for alcohol-related associations; Standardized parameter estimates (p<.05); Adjusted for age, gender, peer drinking, adult
drinking, playing sports, and clustering by school; Fit: χ
2
(39)=69.98, p=.002, CFI=.99, TLI=.96, RMSEA=.014, SRMR=.014.
Wave 1
CBAT.
Wave 3
CBAT
Wave 2
CBAT
Ic
Sc
Wave 1
Alcohol
Use
Wave 3
Alcohol
Use
Wave 2
Alcohol
Use
Ia
Sa
Popular shows
Cued recall
Self-report exposure
Like ads
.158
-.168
.281
-.098
.137
.458
-.052
Time 1 Measures
Related to Alcohol
Advertising
Sports shows
General TV
False Recall
.058

177

Figure 4-2 Growth of Alcohol-Related Associations for the COBT.
Notes: COBT = Cue Outcome Behavior Task; I = growth curve intercepts; S = linear slope; COBT=cue outcome behavior task coded
for alcohol-related associations; Standardized parameter estimates (p<.05); Adjusted for age, gender, peer drinking, adult drinking,
playing sports, and clustering by school; Fit: χ
2
(39)=61.85, p=.011, CFI=.99, TLI=.96, RMSEA=.012, SRMR=.012.
Wave 1
COBT.
Wave 3
COBT
Wave 2
COBT
Ic
Sc
Wave 1
Alcohol
Use
Wave 3
Alcohol
Use
Wave 2
Alcohol
Use
Ia
Sa
Popular shows
Cued recall
Self-report exposure
Like ads
.124
.280
-.093
.126
.802
-.053
Time 1 Measures
Related to Alcohol
Advertising
.183
.363
Sports shows
General TV
False recall
.140

178
Table 4-2 Standardized Estimates for the Parallel Growth Curve Models.
Item Cue Behavior
Association Task
1

Cue Outcome
Association Task
2

Parameter
Estimate
Standard
Error
Parameter
Estimate
Standard
Error
Association Intercept on
T1 associations .671*** .059 .568*** .049
T2 associations .448*** .048 .326*** .026
T3 associations .312*** .037 .232*** .019
Association Slope on
T1 associations .000 .000 .000 .000
T2 associations .447*** .036 .365*** .034
T3 associations .622*** .050 .520*** .044
Alcohol Use Intercept on
T1 alcohol use .789*** .035 .785*** .036
T2 alcohol use .610*** .036 .603*** .036
T3 alcohol use .484*** .024 .480*** .024
Alcohol Use Slope on
T1 alcohol use .000 .000 .000 .000
T2 alcohol use .365*** .041 .378*** .035
T3 alcohol use .579*** .067 .601*** .060
Associations Slope on
Alcohol Use Intercept
.112 .109 -.164 .203
Alcohol Use Slope on
Associations Intercept
-.168** .062 .017 .077
Associations Intercept on
T1 Predictors

Popular shows .002 .033 .004 .027
Sports shows .044 .040 .013 .036
Cued recall .073 .050 .140** .045
False recall -.058 .053 -.064 .051
Self-report .058 .028 .026 .034
Liking of ads .158*** .020 .124** .043
General TV viewing .001 .027 .005 .048
Gender (male=1, female=0) .021 .039 -.038 .033
Age -.008 .028 -.030 .027
Peer drinking .114 .063 .206** .073
Playing sports -.055* .026 -.097* .040
Adult drinking .005 .048 .005 .043
Language acculturation .108*** .033 .148*** .035
Parents’ jobs -.003 .029 .130** .042
Parents’ education .075 .056 -.053 .038

179
Table 4-2, Continued.
Associations Slope on
T1 Predictors

Popular shows -.011 .028 .056 .045
Sports shows -.033 .041 -.018 .067
Cued recall .093 .067 -.045 .062
False recall .105 .072 .132 .082
Self-report .045 .036 .065 .037
Liking of ads .042 .037 .183* .081
General TV viewing .026 .041 -.035 .041
Gender (male=1, female=0) .051 .030 -.024 .039
Age .054 .031 .049 .038
Peer drinking -.048 .069 .187 .121
Playing sports .029 .039 -.009 .054
Adult drinking -.068 .043 .039 .062
Language acculturation .171*** .041 .197*** .055
Parents’ jobs .173** .054 -.003 .044
Parents’ education -.062 .047 -.130 .094
Alcohol Use Intercept on
T1 Predictors

Popular shows -.016 .029 -.016 .030
Sports shows .026 .030 .026 .030
Cued recall -.098** .035 -.093** .035
False recall -.001 .033 -.005 .032
Self-report -.052* .021 -.053* .021
Liking of ads .281*** .032 .280*** .032
General TV viewing .030 .027 .030 .027
Gender (male=1, female=0) -.004 .031 -.006 .031
Age .048 .026 .049 .026
Peer drinking .573*** .037 .576*** .040
Playing sports .001 .025 -.001 .026
Adult drinking .178*** .043 .179*** .044
Language acculturation -.036 .039 -.038 .039
Parents’ jobs .043 .047 .041 .047
Parents’ education -.019 .024 -.020 .024
Alcohol Use Slope on
T1 Predictors

Popular shows .137*** .039 .126** .038
Sports shows -.056 .041 -.058 .037
Cued recall .096 .052 .078 .055
False recall -.059 .070 -.041 .071
Self-report .086 .045 .073 .045
Liking of ads .074 .050 .040 .043
General TV viewing -.071 .046 -.059 .040

180
Table 4-2, Continued.
Gender (male=1, female=0) .048 .048 .044 .047
Age -.031 .042 -.034 .042
Peer drinking -.243*** .062 -.265*** .070
Playing sports -.085 .063 -.079 .061
Adult drinking -.081 .079 -.080 .075
Language acculturation .157* .071 .125 .069
Parents’ jobs .001 .067 -.004 .067
Parents’ education -.034 .052 -.036 .051
Associations Intercept Correlation
with Alcohol Use Intercept
.050 .053 .363** .116
Associations Slope Correlation
with Alcohol Use Slope
.458*** .095 .802*** .161
Latent Factor Intercepts
Associations intercept .298*** .051 .556*** .048
Associations slope .378*** .057 .855*** .138
Alcohol use intercept .520*** .030 .525*** .031
Alcohol use slope .458*** .066 .389*** .069
Residual Variances
T1 associations .550*** .079 .677*** .056
T2 associations .574*** .036 .749*** .028
T3 associations .491*** .057 .663*** .053
T1 alcohol use .377*** .055 .384*** .056
T2 alcohol use .590*** .040 .588*** .040
T3 alcohol use .550*** .054 .527*** .053
Associations intercept .895*** .027 .852*** .043
Associations slope .879*** .046 .866*** .059
Alcohol use intercept .367*** .065 .365*** .070
Alcohol use slope .840*** .049 .871*** .044
R-Square Latent Variables
Associations intercept .105*** .027 .148** .043
Associations slope .121** .046 .134* .059
Alcohol use intercept .633*** .065 .635*** .070
Alcohol use slope .170** .049 .129** .044
Notes:
1
CBAT model fit: χ
2
(39)=69.98, p=.002, CFI=.99, TLI=.96, RMSEA=.014,
SRMR=.014;
2
COBT model fit: χ
2
(39)=61.85, p=.011, CFI=.99, TLI=.96, RMSEA=.012,
SRMR=.012; T1 = time 1 (7
th
grade), T2 = time 2 (8
th
grade), T3 = time 3 (9
th
grade);
*(p<.05), **(p<.01), ***(p<.001).

The response codes for these alcohol-related categories were dichotomized as
being in the alcohol category (1) or in one of the other categories (0), and then fit to a

181
linear growth curve using Mplus (Muthen & Muthen, 1998-2007). Those
homograph/categories that fit the dichotomized data well (e.g., χ
2
(1)<2.52, p>.05;
RMSEA<.050; SRMR<.050) and had significant slope means included draft/beer
(S=.210; 95%CI: .103 - .318), hammered/drunk (S=.448; 95%CI: .270 - .627), shot/liquor
(S=.564; 95%CI: .127 – 1.001), and tap/spigot (S=.288; 95%CI: .104 - .471). The
significant slope means for these growth curves provide support for the first part of
hypothesis H2. The slope mean for the homograph/category of pitcher/container was not
significant (S=.526; 95%CI: -1.574 – 2.626) probably due to a small growth in this
category (17.88%, 20.03%, and 21.81% for times 1, 2, and 3 respectively). The
homograph/category of bud/beer did not fit the data well (Fit: χ
2
(1)=20.57, p<.001,
CFI=.89, TLI=.68, RMSEA=.074, SRMR=.029) probably due to a non-linear growth rate
(20.87%, 29.30%, and 28.25% for times 1, 2, and 3 respectively).
As noted above, the omnibus chi-squared test was significant for all of the
homographs contradicting the second part of hypothesis H2. The homographs count,
ground, lead, neon, pupil, rash, and shed were not expected to change because it was
believed that the relative proportions for the meaning categories would have been well
established at a younger age. Small albeit significant changes were noted for ground,
rash, and shed, but the other homographs exhibited larger changes in proportions. For
example, the percentage of responses referring to the category of ‘metal or pencil’ for the
homograph lead increased from 38.46% in 7
th
grade to 56.30% in 9
th
grade, and the

182
Table 4-3 Percentage of Responses to Homograph Cues by Meaning Category.
Item Cue Inter-Rater
Agreement
(%)
kappa
Meaning Categories for
Coding the Responses.
Grade 7
(% in each
category)
Grade 8
(% in each
category)
Grade 9
(% in each
category)
Grades 7-9
Chi-Sq/df
p-value

Cues related to alcohol use
1. bud 69.68
0.62
1. flower or leaf bud
2. marijuana
3. friend
4. beer
5. other
16.47
11.94
42.12
20.87
8.59
N=2386
14.66
17.66
33.03
29.30
5.43
2522
15.79
28.25
23.32
28.25
4.38
1805
340.38
16
<.0001
2. draft 68.65
0.59
1. select
2. sport-related selection
3. beer
4. preliminary writing
5. pull
6. cold breeze
7. other
3.77
12.61
3.99
30.73
2.60
25.36
20.94
N=2307
7.67
10.74
5.90
31.37
2.38
27.72
14.23
2439
7.89
12.83
7.95
24.26
2.61
28.85
15.61
1723
231.62
36
<.0001
3. hammered 64.59
0.45
1. struck or used to strike
2. drunk or high
3. other
79.93
6.81
13.25
N=2433
79.90
11.86
8.25
2522
72.97
18.19
8.84
1776
39.16
4
<.0001

183
Table 4-3, Continued.
4. pitcher 79.24
0.68
1. container
2. baseball
3. other
17.88
73.72
8.40
N=2702
20.03
74.03
5.94
27.46
21.81
72.47
5.72
1907
89.96
4
<.0001
5. shot 68.93
0.58
1. discharge
2. sports-related discharge
3. medical or drug related
4. liquor
5. attempts
6. ruined
7. other
57.85
6.60
23.48
1.10
0.11
0.34
10.51
N=2636
66.69
5.51
19.36
2.66
0.11
0.19
5.48
2702
65.23
6.46
15.98
6.14
0.16
0.00
6.03
1858
680.00
30
<.0001
6. tap 72.11
0.53
1. strike lightly
2. dance
3. spigot
4. borrow
5. other
51.49
17.64
11.63
0.78
18.46
N=2443
47.65
21.62
14.16
0.59
15.98
2535
38.32
28.31
18.01
0.39
14.97
1777
66.46
16
<.0001

184
Table 4-3, Continued.
Cues not related to alcohol
1. field 72.58
0.60
1. land or open space
2. sports-related open
space
3. area of vision
4. area of force
5. other
51.74
41.21
0.12
0.08
6.86
N=2582
44.87
50.19
0.15
0.07
4.72
2692
36.82
57.05
0.11
0.33
5.69
1844
59.01
16
<.0001
2. program 69.19
0.57
1. outline of events
2. television
3. sports-related program
4. computer code
5. other
20.32
57.78
1.41
6.51
13.98
N=2475
23.54
56.86
1.66
9.32
8.62
2587
24.90
53.77
1.47
10.10
9.76
1763
56.42
16
<.0001
3. ram 76.47
0.62
1. male sheep
2. strike
3. sports-related
4. other
45.23
21.98
9.90
22.89
N=2193
48.21
20.01
10.48
21.30
2319
48.15
14.05
18.15
19.64
1680
112.60
9
<.0001
4. screen 76.58
0.64
1. partition or screen
2. sports-related
3. project on a screen
4. other
13.64
1.80
74.91
9.64
N=2551
12.27
2.02
79.09
6.63
2625
9.69
3.59
81.97
4.74
1836
23.59
9
.005

185
Table 4-3, Continued.
5. season 80.77
0.71
1. particular time
2. sports-related season
3. flavor food
4. other
81.43
8.99
1.96
7.62
N=2547
80.98
11.83
1.79
5.40
2629
78.82
13.61
2.12
5.44
1837
63.54
9
<.0001
5. count 70.41
0.48
1. number
2. nobleman
3. depend
4. other
89.57
2.24
0.41
7.78
N=2454
91.90
3.65
0.42
4.03
2605
87.13
8.20
0.39
4.27
1780
47.22
9
<.0001
6. ground 72.22
0.51
1. earth
2. grind
3. foundation
4. other
90.62
1.41
0.38
7.59
N=2622
93.20
1.60
0.22
4.98
2691
90.92
2.16
0.49
6.43
1850
18.49
9
.03
7. lead 68.21
0.57
1. guide the way
2. sports-related guide
3. front
4. metal or pencil
5. (blank by mistake)
6. other
35.49
5.77
9.00
38.46
0.00
11.28
N=2322
27.92
4.42
6.29
53.98
0.00
7.39
2464
24.35
5.35
5.93
56.30
0.00
8.06
1737
70.50
16
<.0001

186
Table 4-3, continued.
8. neon 75.86
0.59
1. gas
2. sign
3. bright color
4. other
0.82
13.82
57.49
27.88
N=2077
3.28
22.15
52.29
22.28
2316
1.45
28.98
48.76
20.82
1729
210.40
9
<.0001
9. pupil 80.30
0.70
1. student
2. eye
3. other
44.21
42.99
12.79
N=2298
43.16
47.67
8.72
2465
37.71
55.33
6.96
1811
163.58
4
<.0001
10. rash 75.94
0.55
1. skin irritation
2. reckless
3. other
80.74
4.01
15.25
N=2419
84.78
2.97
12.25
2490
85.27
3.30
11.42
1786
51.44
4
<.0001
11. shed 75.60
0.61
1. throw off
2. shelter
3. other
34.20
42.92
22.88
N=2155
35.92
47.21
16.87
2294
33.13
47.61
19.26
1630
54.84
4
<.0001

187
percentage for the category of ‘sign’ for the homograph neon increased from 13.82% to
28.98%. It appears that it may be difficult to predict how the proportion of responses
across meaning categories might change for some homographs.
The meaning category proportions for the current study (9
th
grade students) were
compared to word association norms collected from undergraduate psychology students
by Nelson, McEvoy, Walling, and Wheeler (1980). The Pearson correlation coefficient
was r=.59 (p<.01) across dominant and non-dominate proportions for 12 overlapping
homographs (count, draft, field, ground, pitcher, pupil, ram, rash, screen, shed, shot, and
tap). This correlation coefficient was significant but smaller than those reported by
Nelson et al. between their norms and other norms in the literature (range .79 ≤r ≤.91). It
appears that the data collected in the current study differed to some extent from the
Nelson norms possibly due to age differences among participants and/or historical trends
in the norms.
Discussion
This was the first study to examine the influence of televised alcohol advertising
on the development of spontaneous alcohol-related associations in memory among young
adolescents. The influence of alcohol ads on implicit memory has important implications
given that in prior research from several sources spontaneous alcohol-related associations
are predictive of alcohol use by adolescents. In partial support of hypothesis H1, the
current study showed that certain measures related to alcohol advertising predicted these
associations in memory. Self-reported exposure to alcohol ads on TV and liking of those
ads predicted the number of spontaneous associations provided in response to the

188
alcohol-related homograph cue words (CBAT). Liking of alcohol ads and cued recall for
alcohol ads on TV predicted the initial number of spontaneous associations provided in
response to phrases depicting alcohol use outcomes (COBT). An important prospective
finding in this study was that the liking of alcohol ads significantly predicted the growth
(slope) of associations (COBT) over the three time periods.
There was a significant and large correlation between the growth of alcohol-
related associations in memory and the growth in use of alcohol. Some relationship was
expected here since personal experiences with alcohol are expected to increase the
number of (and strength of) associations related to alcohol in memory, and previous
prospective research has shown the effects of associations on later alcohol use. This
strong correlation in the parallel growth model is tantamount to “commitment variation,”
one of Mill’s several necessary but not sufficient criteria for a causal process. The finding
is also consistent with a reciprocal model, in which alcohol use experiences and
associations in memory fuel each other over time. It is important to underscore that this is
a dynamic process involving concomitant changes over time in both alcohol use and
associations in memory. The initial (static) level of alcohol use did not predict the growth
of associations prospectively in either model. However, some covariates were positive
predictors of the growth of associations, including language acculturation and an SES
proxy, parents’ occupations. Covariates that might have been expected to influence the
growth of associations, but did not, included gender and observation of friends and adults
drinking alcohol.

189
Concerning alcohol use, an important finding was that exposure to alcohol
advertising on popular shows at time 1 predicted the growth (slope) of alcohol use, which
is consistent with the view that exposure to alcohol ads influences alcohol use. In
addition and as expected, observation of friends drinking was a strong predictor of the
growth of alcohol use. Liking of alcohol ads predicted the initial level (intercept) of
alcohol use as expected. It was not anticipated, however, that self-reported observation of
alcohol ads and cued recall of alcohol ads would be negative predictors of the intercept
for alcohol use. These latter results suggest that those who used more alcohol in the 7
th

grade reported seeing fewer alcohol ads and had a poorer memory for images from
specific alcohol ads. The reason for this is unclear, but it does appear to contradict the
notion that those who are drinking more tend to be more aware of alcohol ads. It is also
unclear why the initial level of alcohol-related associations was a negative predictor of
the slope for the growth of alcohol use in the CBAT model. It is possible, however, that
the negative relationship might be an indication that those high in initial level of alcohol
use and alcohol-related associations had lower slopes for the growth of alcohol use (i.e.,
because their use was already high at time 1).
In support of hypothesis H2, the relative frequency increased over time for
alcohol-related responses compared to other categories of responses to homograph cue
words with an alcohol-related meaning. The more general increase in alcohol-related
categories in the current study are important to illustrate that as experiences with alcohol
accumulate, spontaneous alcohol-related associations begin to ‘edge out’ other
associations that may be more benign where health behavior is concerned. As noted

190
previously, stronger associations in memory have been shown to increase the likelihood
of consuming alcohol (e.g., Kelly et al., 2005; Stacy & Newcomb, 1998). In addition, this
type of change in word association responses may be indicative of a conceptual change in
how alcohol use is viewed by the participants (cf., Hovardas & Korfiatis, 2006)
Limitations in the current study include those associated with the measures and
the population sampled. First, we cannot be absolutely certain whether some participants
filtered their responses in the word association task, but similar procedures have been
used successfully in cognitive science (e.g., Nelson et al., 2000), education (Hovardas &
Korfiatis, 2006), and health behavior studies (Stacy, 1995, 1997) to measure spontaneous
associations. Second, the alcohol-related association measures were skewed toward zero
among the young adolescents in the current sample, which might have contributed to
some of the null findings in the current study. The association measures were developed
using high school and college participants, and the resulting measures might have been
somewhat less than optimal for the middle school students in the current study. Third,
generalization of the results is limited to students with similar characteristics as the
sample of urban students surveyed in the current study. Finally, not all of the results
converge across assessments of alcohol advertising. We are unaware of any consensus on
which measures are the best indicators of exposure to alcohol advertising, but liking of
ads has been used successfully to predict the success of individual ads (i.e., predict sales)
as well as ad campaigns across a range of consumer products (e.g., Haley & Baldinger,
1991).

191
The current study suggests that advertising, and the liking of alcohol ads in
particular, has some influence on the development of alcohol-related associations, and
these associations could influence underage drinking. It also appears that there may be a
strong reciprocal influence of alcohol-related associations on alcohol use and alcohol use
on the development of associations. Future research should examine in more depth the
life experiences that contribute to the development of alcohol-related associations and in
particular the development of those early associations that contribute to experimentation
with alcohol at a young age.

192
Chapter 5 Summary and Discussion
Consistent with previous research in this area, empirical evidence was presented
showing that televised alcohol advertisements have a small but significant influence on
underage drinking and the problems associated with alcohol use by adolescents. First, the
psychometric properties of alcohol advertising exposure measures were examined
successfully using confirmatory factor analysis (CFA) procedures. Measures of alcohol
advertising were shown to have good simple structure with individual items loading on a
single factor as expected, and these factors were correlated with each other in expected
patterns. Second, latent growth curve modeling demonstrated that exposure to alcohol ads
and liking of those ads in 7
th
grade influenced the growth of alcohol use over time among
the students. Self-reported liking of alcohol ads moderated the influence of exposure to
alcohol ads on alcohol use. Among those who liked alcohol ads more, exposure to the ads
was a stronger predictor of increased alcohol use than among those students who liked
alcohol ads less. In addition, an increase in the use of alcohol over time influenced the
number of problems reported in the 10
th
grade. The mediated or indirect effect of alcohol
advertising on alcohol-related problems through alcohol use was clearly demonstrated for
females and there was a total effect observed for males. Finally, measures related to
alcohol advertising exposure were shown to be predictive of alcohol-related associations
in memory. In particular, liking of alcohol ads was a positive predictor of the growth of
alcohol-related associations. This is of particular interest because a number of past
studies have shown that alcohol-related associations are predictive of alcohol use by older

193
adolescents. The present finding of strong parallel growth in pro-alcohol memory
associations and alcohol consumption is also consistent with previous research.
These results have relevance for the design of interventions and health-related
policy. Concerning health policy, it is difficult to limit exposure by adolescents to alcohol
advertising. If advertising is limited on TV, then more advertising dollars may be applied
to print ads, event-related promotions, and point of sale displays (Austin & Hust, 2005;
Garfield, Chung, & Rathouz, 2003; Saffer, 2002). Some limits on alcohol advertising on
shows with adolescent audiences might have some small but important protective effects
despite the possible switch to other advertising outlets (Jernigan, Ostroff, & Ross, 2005;
Riley, 2005; Roche, 2005). In particular, it might be especially important to limit beer
and hard alcohol advertisements on cable TV shows where there may be an increasing
number of these ads.
Few studies have evaluated media literacy training as an intervention to
counteract the influence of advertising on substance use by adolescents (Austin, 2006; J.
D. Brown, 2006), but those reported in the literature have been successful. For example,
Austin and Johnson (1997) tested the effects of a media literacy training program among
225 third graders. The program was designed to teach skepticism about advertising based
upon a model where children interpret media messages using both logical and emotional
processes. Results showed that the media literacy training reduced positive alcohol-
outcome expectancies, and an improved understanding of the persuasive intent of
advertisements was associated with less perceived desirability to be like the characters
portrayed in the ads. Another study found that an intervention designed to educate 6th

194
graders about the persuasion tactics of advertisers and to promote critical analysis of the
alcohol ads was successful at creating more critical attitudes toward alcohol ads and
reduced intentions to drink in the future (Goldberg, Niedermeier, Bechtel, & Gorn, 2006).
Similar encouraging results were obtained for media literacy interventions targeting
cigarette ads (e.g., Austin, Pinkleton, Hust, & Cohen, 2005; Primack, Gold, Land, &
Fine, 2006). These literacy interventions attempt, in part, to create skepticism toward the
intent of advertisers and the images portrayed in alcohol ads, and this approach seems
congruent with the findings of the current study such that adolescents who disliked
alcohol ads appeared to use less alcohol.
The current media literacy interventions, however, might still rely too much on
rational or deliberative approaches to decision-making by adolescents when there is some
evidence that much of decision-making is more automatic or implicit in nature. New
methods need to be employed to provide adolescents with the cognitive tools they need to
counteract the development and influence of these associations in memory. Intervention
techniques need to be developed and implemented that create associations between risky
alcohol-use situations (e.g., party on Friday night with alcohol present) and healthy
behaviors (e.g., choose a non-alcoholic drink or go to another social event that does not
have alcohol). These healthy options will be more likely to come to mind in critical
decision-making situations if they are associated with those situations in advance.
Otherwise, young adolescents may be inclined to imitate the behaviors of actors in
alcohol ads or their peers, which could be the default decision in new situations that are
likely to overload deliberative processes.

195
Future research should address several outstanding questions regarding the
influence of alcohol advertising on underage drinking. One issue that deserves attention
is the determination of what is measured by the assessment of liking of alcohol ads (M. J.
Chen, Grube, Bersamin, Waiters, & Keefe, 2005; Unger et al., 2003). Do participants like
all ads or do they generalize from one recent ad or ad campaign? Do different advertising
styles appeal to different characteristics of individuals (as would be expected), and is
there a particular ad style that promotes alcohol use by those young adolescents most
vulnerable to abuse of alcohol? Is the measure truly a measure of the affective response
to alcohol ads or is it some relatively unconscious attempt to conform to perceived
opinions of friends or family? Do participants have a general affinity for and approval of
alcohol use, which they then attribute to a liking of alcohol ads? CFA results indicate that
the items in this measure do not significantly cross-load on other factors such as alcohol
use or exposure to alcohol ads, but the liking measure might tap into some sort of
approach/avoidance processes. The liking measure was such a robust indicator of alcohol
use and the development of alcohol-related associations in memory that the measure
deserves further research to better delineate the mechanisms underlying its influence on
alcohol use.
Another area of interest for future research is the influence of peers on underage
drinking. Underage drinking appears to be influenced by alcohol advertising and by peer
drinking (e.g., Bray, Adams, Getz, & McQueen, 2003; Guilamo-Ramos, Turrisi, Jaccard,
Wood, & Gonzalez, 2004). It seems natural to hypothesize than that the influence of
friends on alcohol use will partially mediate the influence of alcohol ads on alcohol use.

196
Alcohol advertising is likely to influence an entire cohort of adolescents (albeit, some
persons more than other), and this influence will include both a direct impact on alcohol
use and an indirect impact through friends (and adults). The overall influence of alcohol
ads could be significantly larger than those observed in the current study when both the
direct and indirect influences are considered.
One final area to consider in future research is the development of alcohol-related
associations in memory. Little research has been conducted on how life experiences
influence the individual development of these associations despite the importance of the
associations in learning and behavior (Stacy et al., 1997). For example, younger children
generally have few alcohol-related associations, but as they develop into teenagers, they
develop more associations in memory related to alcohol (see Chapter 4 of this study).
How do these changes in associations occur? Are they due to cultural influences? If so
how are these influences transmitted to children, and how can they be disrupted in an
intervention? Whatever the mechanisms for transmitting cultural associations to children,
it is clear that these associations contribute to alcohol use and risky behavior. It is also
becoming clear from research in neuroscience, cognitive science, and social psychology
that much of this risky behavior is the result of decision-making that is influenced by
automatic or implicit associations rather than by more deliberative or rational approaches
(e.g., Bargh & Williams, 2006; Bechara & Damasio, 2002; Grenard et al., in press; C. A.
Johnson et al., 2008; Kahneman, 2003). It is critical, therefore, to understand how these
associations develop and how to direct these associations toward more adaptive and
healthy behaviors.

197
References
Aaron, D. J., Dearwater, S. R., Anderson, R., Olsen, T., & et al. (1995). Physical activity
and the initiation of high-risk health behaviors in adolescents. Medicine & Science
in Sports & Exercise, 27(12), 1639-1645.
Adlaf, E. M., & Kohn, P. M. (1989). Alcohol advertising, consumption and abuse: A
covariance-structural modeling look at Strickland's data. British Journal of
Addiction, 84(7), 749-757.
Aitken, P. P., Eadie, D. R., Leathar, D. S., McNeill, R. E., & et al. (1988). Television
advertisements for alcoholic drinks do reinforce under-age drinking. British
Journal of Addiction, 83(12), 1399-1419.
Ames, S. L., Gallaher, P. E., Sun, P., Pearce, S., Zogg, J. B., Houska, B. R., et al. (2005).
A Web-based program for coding open-ended response protocols. Behavior
Research Methods, 37(3), 470-479.
Ames, S. L., Grenard, J. L., Thush, C., Sussman, S., Wiers, R. W., & Stacy, A. W.
(2007). Comparison of indirect assessments of association as predictors of
marijuana use among at-risk adolescents. Experimental and Clinical
Psychopharmacology, 15(2), 204-218.
Ames, S. L., & Stacy, A. W. (1998). Implicit cognition in the prediction of substance use
among drug offenders. Psychology of Addictive Behaviors, 12(4), 272-281.
Ames, S. L., Zogg, J. B., & Stacy, A. W. (2002). Implicit cognition, sensation seeking,
marijuana use and driving behavior among drug offenders. Personality &
Individual Differences, 33(7), 1055-1072.
Anderson, J. C., & Gerbing, D. W. (1988). Structural equation modeling in practice: A
review and recommended two-step approach. Psychological Bulletin, 103(3),
411-423.
Anderson, J. R. (1983). A spreading activation theory of memory. Journal of Verbal
Learning & Verbal Behavior, 22(3), 261-295.
Anderson, J. R., Bothell, D., Byrne, M. D., Douglass, S., Lebiere, C., & Qin, Y. (2004).
An Integrated Theory of the Mind. Psychological Review, 111(4), 1036-1060.
Arnett, D. B., Laverie, D. A., & Meiers, A. (2003). Developing parsimonious retailer
equity indexes using partial least squares analysis: A method and applications.
Journal of Retailing, 79(3), 161.

198
Atkin, C. K. (1995). Survey and experimental research on effects of alcohol advertising.
In S. E. Martin (Ed.), The effects of the mass media on the use and abuse of
alcohol (Research Monograph No. 28). Bethesda, MD: National Institutes of
Health.
Atkin, C. K., Hocking, J., & Block, M. (1984). Teenage drinking: Does advertising make
a difference? Journal of Communication, 34(2), 157-167.
Austin, E. W. (2006). Why advertisers and researchers should focus on media literacy to
respond to the effects of alcohol advertising on youth. International Journal of
Advertising, 25(4), 541-544.
Austin, E. W., Chen, M.-J., & Grube, J. W. (2006). How does alcohol advertising
influence underage drinking? The role of desirability, identification and
skepticism. Journal of Adolescent Health, 38(4), 376-384.
Austin, E. W., & Hust, S. J. T. (2005). Targeting Adolescents? The Content and
Frequency of Alcoholic and Nonalcoholic Beverage Ads in Magazine and Video
Formats November 1999-April 2000. Journal of Health Communication, 10(8),
769-785.
Austin, E. W., & Johnson, K. K. (1997). Effects of general and alcohol-specific media
literacy training on children's decision making about alcohol. Journal of Health
Communication, 2(1), 17-42.
Austin, E. W., & Meili, H. K. (1994). Effects of interpretations of televised alcohol
portrayals on children's alcohol beliefs. Journal of Broadcasting & Electronic
Media, 38(4), 417-435.
Austin, E. W., Pinkleton, B. E., Hust, S. J. T., & Cohen, M. (2005). Evaluation of an
American Legacy Foundation/Washington State Department of Health Media
Literacy Pilot Study. Health Communication, 18(1), 75-95.
Austin, E. W., Roberts, D. F., & Nass, C. I. (1990). Influences of family communication
on children's television-interpretation processes. Communication Research, 17(4),
545.
Baddeley, A. D., & Hitch, G. (1974). Working memory. In G. A. Bower (Ed.), Recent
advances in learning and motivation (Vol. 8, pp. 47-90). New York: Academic
Press.
Bandalos, D. L., & Finney, S. J. (2001). Item parceling issues in structural equation
modeling. In G. A. Marcoulides & R. E. Schumacker (Eds.), New developments
and techniques in structural equation modeling. (pp. 269-296). Mahwah, NJ, US:
Lawrence Erlbaum Associates Publishers.

199
Bandura, A. (1965). Influence of models' reinforcement contingencies on the acquisition
of imitative responses. Journal of Personality and Social Psychology, 1(6), 589-
595.
Bandura, A. (1977). Social Learning Theory. Upper Saddle River, NJ: Prentice Hall.
Bandura, A. (1986). Social foundations of thought and action: A social cognitive theory.
Englewood Cliffs, NJ: Prentice Hall.
Bandura, A., Ross, D., & Ross, S. A. (1963). Imitation of film-mediated aggressive
models. Journal of Abnormal & Social Psychology, 66(1), 3-11.
Bargh, J. A., & Morsella, E. (2008). The unconscious mind. Perspectives on
Psychological Science. Special Issue: From philosophical thinking to
psychological empiricism, 3(1), 73-79.
Bargh, J. A., & Williams, E. L. (2006). The Automaticity of Social Life. Current
Directions in Psychological Science, 15(1), 1-4.
Bechara, A., & Damasio, H. (2002). Decision-making and addiction (part I): Impaired
activation of somatic states in substance dependent individuals when pondering
decisions with negative future consequences. Neuropsychologia, 40(10), 1675-
1689.
Berger, J., & Fitzsimons, G. i. (2008). Dogs on the street, Pumas on your feet: How cues
in the environment influence product evaluation and choice. Journal of Marketing
Research, 45(1), 1-14.
Bettinghaus, E. P., & Cody, M. J. (1994). Persuasive Communication (5th ed.). Fort
Worth, TX: Harcourt Brace.
Bloom, P. M., Hogan, J. E., & Blazing, J. (1997). Sports promotion and teen smoking and
drinking: An exploratory study. American Journal of Health Behavior, 21(2),
100-109.
Bollen, K. A. (1989). Structural equations with latent variables. Wiley series in
probability and mathematical statistics. Applied probability and statistics section.
Oxford, England: John Wiley & Sons, 514.
Bollen, K. A., & Lennox, R. (1991). Conventional Wisdom on Measurement - a
Structural Equation Perspective. [Article]. Psychological Bulletin, 110(2), 305-
314.
Boring, E. G. (1950). A history of experimental psychology (2nd ed.). New York:
Appleton-Century-Crofts.

200
Borsboom, D. (2006). When Does Measurement Invariance Matter? Medical Care.
Special Issue: Measurement in a multi-ethnic society, 44(11, Suppl 3), S176-
S181.
Boush, D. M., & Jones, S. M. (2006). A Strategy-Based Framework for Extending Brand
Image Research. In L. R. Kahle & C.-H. Kim (Eds.), Creating images and the
psychology of marketing communication. Advertising and Consumer Psychology.
(pp. 3-29). Mahwah, NJ, US: Lawrence Erlbaum Associates Publishers.
Bray, J. H., Adams, G. J., Getz, J. G., & McQueen, A. (2003). Individuation, peers, and
adolescent alcohol use: a latent growth analysis. Journal of Consulting & Clinical
Psychology, 71(3), 553-564.
Brown, J. D. (2006). Media literacy has potential to improve adolescents' health. Journal
of Adolescent Health, 39(4), 459-460.
Brown, J. D., & McDonald, T. (1995). Protrayals and effects of alcohol in television
entertainment programming. In S. E. Martin (Ed.), The effects of the mass media
on the use and abuse of alcohol (Research Monograph No. 28). Bethesda, MD:
National Institutes of Health.
Brown, S. A., & Tapert, S. F. (2004). Adolescence and the trajectory of alcohol use:
Basic to clinical studies. Annals of the New York Academy of Sciences, 1021, 234-
244.
Brown, T. A. (2006). Confirmatory factor analysis for applied research. New York:
Guilford Press.
Byrne, B. M., Shavelson, R. J., & Muthen, B. (1989). Testing for the equivalence of
factor covariance and mean structures: The issue of partial measurement
invariance. Psychological Bulletin, 105(3), 456-466.
Byrne, B. M., & Stewart, S. M. (2006). The MACS approach to testing for multigroup
invariance of a second-order structure: A walk through the process. Structural
Equation Modeling, 13(3), 287-321.
Cacioppo, J. T., & Petty, R. E. (1982). The need for cognition. Journal of Personality and
Social Psychology, 42(1), 116-131.
Campbell, D. T., & Fiske, D. W. (1959). Convergent and discriminant validation by the
multitrait-multimethod matrix. Psychological Bulletin, 56, 81-105.
CAMY. (2006). Still growing after all these years: Youth exposure to alcohol advertising
on television, 2001-2005. Washington, D.C.: Center on Alcohol Marketing and
Youth at Georgetown University.

201
Casswell, S., & Zhang, J.-F. (1998). Impact of liking for advertising and brand allegiance
on drinking and alcohol-related aggression: A longitudinal study. Addiction,
93(8), 1209-1217.
Cattell, J. M., & Bryant, S. (1889). Mental association investigated by experiment. Mind,
14, 230-250.
CDC. (2004). Surveillance Summaries (No. MMWR 2004:53(No. SS-2).). Atlanta, GA:
Centers for Disease Control and Prevention.
Chen, F. F. (2007). Sensitivity of goodness of fit indexes to lack of measurement
invariance. Structural Equation Modeling, 14(3), 464-504.
Chen, M. J., Grube, J. W., Bersamin, M., Waiters, E., & Keefe, D. B. (2005). Alcohol
Advertising: What Makes It Attractive to Youth? Journal of Health
Communication, 10(6), 553-565.
Chen, S., & Chaiken, S. (1999). The heuristic-systematic model in its broader context. In
S. Chaiken & Y. Trope (Eds.), Dual-process theories in social psychology. (pp.
73-96). New York: Guilford Press.
Christiansen, B. A., Goldman, M. S., & Brown, S. A. (1985). The differential
development of adolescent alcohol expectancies may predict adult alcoholism.
Addictive Behaviors, 10(3), 299-306.
Christiansen, B. A., Goldman, M. S., & Inn, A. (1982). Development of alcohol-related
expectancies in adolescents: Separating pharmacological from social-learning
influences. Journal of Consulting & Clinical Psychology, 50(3), 336-344.
Christiansen, B. A., Smith, G. T., Roehling, P. V., & Goldman, M. S. (1989). Using
alcohol expectancies to predict adolescent drinking behavior after one year.
Journal of Consulting & Clinical Psychology, 57(1), 93-99.
Clark, D. B., Lesnick, L., & Hegedus, A. M. (1997). Traumas and other adverse life
events in adolescents with alcohol abuse and dependence. Journal of the
American Academy of Child & Adolescent Psychiatry, 36(12), 1744-1751.
Collins, A. M., & Loftus, E. F. (1975). A spreading-activation theory of semantic
processing. Psychological Review, 82(6), 407-428.
Collins, L. M. (2006). Analysis of longitudinal data: The integration of theoretical model,
temporal design, and statistical model. Annual Review of Psychology, 57, 505-
528.

202
Connolly, G. M., Casswell, S., Zhang, J.-F., & Silva, P. A. (1994). Alcohol in the mass
media and drinking by adolescents: A longitudinal study. Addiction, 89(10),
1255-1263.
Cook, S. W., & Skinner, B. F. (1939). Some factors influencing the distribution of
associated words. Psychological Record, 3, 178-184.
Cook, W. A., & Kover, A. J. (1997). Research and the meaning of advertising
effectiveness: Mutual misunderstandings. In W. D. Wells (Ed.), Measuring
advertising effectiveness. Advertising and consumer psychology. (pp. 13-20).
Mahwah, NJ, US: Lawrence Erlbaum Associates Publishers.
Cooper, M. L., Frone, M. R., Russell, M., & Mudar, P. (1995). Drinking to regulate
positive and negative emotions: a motivational model of alcohol use. Journal of
Personality & Social Psychology, 69(5), 990-1005.
Coronges, K. A., Stacy, A. W., & Valente, T. W. (2007). Structural comparison of
cognitive associative networks in two populations. Journal of Applied Social
Psychology, 37(9), 2097-2129.
Cramer, P. (1968). Word association. Saint Louis: Academic Press.
Crawford, A. M., Pentz, M. A., Chou, C.-P., Li, C., & Dwyer, J. H. (2003). Parallel
developmental trajectories of sensation seeking and regular substance use in
adolescents. Psychology of Addictive Behaviors, 17(3), 179-192.
Crovitz, H. F. (1970). Galton's walk: Methods for the analysis of thinking, intelligence,
and creativity. Oxford, England: Harper & Row.
D'Amico, E. J., Barnett, N. P., Monti, P. M., Colby, S. M., Spirito, A., & Rohsenow, D. J.
(2002). Does alcohol use mediate the association between alcohol evaluations and
alcohol-related problems in adolescents? Psychology of Addictive Behaviors,
16(2), 157-160.
Damasio, A. R. (2003). Looking for Spinoza: Joy, sorrow, and the feeling brain. Orlando,
FL: Harcourt.
Davison, G. C., Vogel, R. S., & Coffman, S. G. (1997). Think-aloud approaches to
cognitive assessment and the articulated thoughts in simulated situations
paradigm. Journal of Consulting & Clinical Psychology, 65(6), 950-958.
de La Haye, F. (2003). Word association norms for 9-, 10-, and 11-year-old children
(CE2, CM1, CM2) and adults. Annee Psychologique, 103(1), 109-130.
Deese, J. (1959a). Influence of inter-item associative strength upon immediate free recall.
Psychological Reports, 5, 305-312.

203
Deese, J. (1959b). On the prediction of occurrence of particular verbal intrusions in
immediate recall. Journal of Experimental Psychology, 58, 17-22.
Deese, J. (1965). The structure of associations in language and thought. Baltimore: Johns
Hopkins Press.
Diamantopoulos, A., & Winklhofer, H., M. . (2001). Index construction with formative
indicators: An alternative to scale development. Journal of Marketing Research,
38(2), 269.
Dubow, J. S. (1994). Point of view: Recall revisited: Recall redux. Journal of Advertising
Research, 34(3), 92.
Duncan, T. E., Duncan, S. C., & Strycker, L. A. (2006). An introduction to latent
variable growth curve modeling: Concepts, issues, and applications (2nd ed.).
Mahwah, NJ: Lawrence Erlbaum Associates.
Dunn, M. E., & Yniguez, R. M. (1999). Experimental Demonstration of the Influence of
Alcohol Advertising on the Activation of Alcohol Expectancies in Memory
Among Fourth- and Fifth-Grade Children. Experimental and Clinical
Psychopharmacology, 7(4), 473-483.
Elias, M. J., Branden-Muller, L. R., & Sayette, M. A. (1991). Teaching the foundations of
social decision making and problem solving in the elementary school. In J. Baron
& R. V. Brown (Eds.), Teaching decision making to adolescents. (pp. 161-184).
Hillsdale, NJ, England: Lawrence Erlbaum Associates, Inc.
Ellickson, P. L., Collins, R. L., Hambarsoomians, K., & McCaffrey, D. F. (2005). Does
alcohol advertising promote adolescent drinking? Results from a longtitudinal
assessment. Addiction, 100(8), 235-246.
Entwisle, D. R. (1966). Word associations of young children. Oxford, England: Johns
Hopkins Press, Oxford, England.
Feldman, L. A., Harvey, B., Holowaty, P., & Shortt, L. (1999). Alcohol use beliefs and
behaviors among high school students. Journal of Adolescent Health, 24(1), 48-
58.
Fergusson, D. M., Horwood, L. J., & Lynskey, M. T. (1995). The prevalence and risk
factors associated with abusive or hazardous alcohol consumption in 16-year-olds.
Addiction, 90(7), 935-946.
Fisher, J. C. (1993). Advertising, alcohol consumption, and abuse: A worldwide survey.
Westport, CT Greenwood Press.

204
Fisher, J. C. (1999). Media influence. In R. T. Ammerman, P. J. Ott & R. E. Tarter
(Eds.), Prevention and societal impact of drug and alcohol abuse. (pp. 235-260).
Mahwah, NJ: Lawrence Erlbaum Associates.
Fleiss, J. L. A., Levin, B. A., & Paik, M. C. A. (2004). Statistical Methods for Rates and
Proportions (3rd ed.). Hoboken: John Wiley & Sons.
Fleming, K., Thorson, E., & Atkin, C. K. (2004). Alcohol advertising exposure and
perceptions: Links with alcohol expectancies and intentions to drink or drinking
in underaged youth and young adults. Journal of Health Communication, 9(1), 3-
29.
Fosados, R., McClain, A., Ritt-Olson, A., Sussman, S., Soto, D., Baezconde-Garbanati,
L., et al. (2007). The influence of acculturation on drug and alcohol use in a
sample of adolescents. Addictive Behaviors, 32(12), 2990-3004.
French, B. F., & Finch, W. H. (2006). Confirmatory factor analytic procedures for the
determination of measurement invariance. Structural Equation Modeling-a
Multidisciplinary Journal, 13(3), 378-402.
Freud, S. (1925/1995). An autobiographical study. In P. Gay (Ed.), The Freud Reader
(pp. 3-41). New York: W. W. Norton & Co.
Garfield, C. F., Chung, P. J., & Rathouz, P. J. (2003). Alcohol advertising in magazines
and adolescent readership. JAMA: Journal of the American Medical Association,
289(18), 2424-2429.
Giancola, P. R., & Mezzich, A. C. (2003). Executive functioning, temperament, and drug
use involvement in adolescent females with a substance use disorder. Journal of
Child Psychology & Psychiatry & Allied Disciplines, 44(6), 857-866.
Gibson, L. D. (1983). "Not recall.". Journal of Advertising Research, 23(1), 39-46.
Gius, M. P. (1995). Using panel data to determine the effects of advertising on brand-
level distilled spirits sales. Journal of Studies on Alcohol, 57(1), 73-76.
Goldberg, M. E., Niedermeier, K. E., Bechtel, L. J., & Gorn, G. J. (2006). Heightening
Adolescent Vigilance Toward Alcohol Advertising to Forestall Alcohol Use.
Journal of Public Policy & Marketing, 25(2), 147-159.
Goldman, M. S., Brown, S. A., & Christiansen, B. A. (1987). Expectancy theory:
Thinking about drinking. In H. T. Blane & K. E. Leonard (Eds.), Psychological
theories of drinking and alcoholism. New York: Guilford Press.

205
Gonzalez-Roma, V., Tomas, I., Ferreres, D., & Hernandez, A. (2005). Do Items That
Measure Self-Perceived Physical Appearance Function Differentially Across
Gender Groups? An Application of the MACS Model. Structural Equation
Modeling, 12(1), 148-162.
Goshen-Gottstein, Y., & Kempinsky, H. (2001). Probing memory with conceptual cues at
multiple retention intervals: A comparison of forgetting rates on implicit and
explicit tests. Psychonomic Bulletin & Review, 8(1), 139-146.
Greenwald, A. G., McGhee, D. E., & Schwartz, J. L. K. (1998). Measuring individual
differences in implicit cognition: The implicit association test. Journal of
Personality & Social Psychology, 74(6), 1464-1480.
Gregorich, S. E. (2006). Do Self-Report Instruments Allow Meaningful Comparisons
Across Diverse Population Groups? Testing Measurement Invariance Using the
Confirmatory Factor Analysis Framework. Medical Care. Special Issue:
Measurement in a multi-ethnic society, 44(11, Suppl 3), S78-S94.
Grenard, J. L., Ames, S. L., Wiers, R. W., Thush, C., Sussman, S., & Stacy, A. W. (in
press). Working memory moderates the predictive effects of drug-related
associations on substance use. Psychology of Addictive Behaviors.
Grube, J. W. (1993). Alcohol portrayals and alcohol advertising on television: Content
and effects on children and adolescents. Alcohol Health & Research World, 17(1),
54-60.
Grube, J. W. (1995). Television alcohol portrayals, alcohol advertising, and alcohol
expectancies among children and adolescents. In S. E. Martin (Ed.), The effects of
the mass media on the use and abuse of alcohol (Research Monograph No. 28).
Bethesda, MD: National Institutes of Health.
Grube, J. W., & Wallack, L. (1994). Television beer advertising and drinking knowledge,
beliefs, and intentions among schoolchildren. American Journal of Public Health,
84(2), 254-259.
Guilamo-Ramos, V., Turrisi, R., Jaccard, J., Wood, E., & Gonzalez, B. (2004).
Progressing from light experimentation to heavy episodic drinking in early and
middle adolescence. Journal of Studies on Alcohol, 65(4), 494-500.
Haley, R. I., & Baldinger, A. L. (1991). The ARF Copy Research Validity Project.
Journal of Advertising Research, 31(2), 11-32.
Haugtvedt, C. P., Petty, R. E., & Cacioppo, J. T. (1992). Need for cognition and
advertising: Understanding the role of personality variables in consumer behavior.
Journal of Consumer Psychology, 1(3), 239-260.

206
Haugtvedt, C. P., & Priester, J. R. (1997). Conceptual and methodological issues in
advertising effectiveness: An attitude strength perspective. In W. D. Wells (Ed.),
Measuring advertising effectiveness. Advertising and consumer psychology. (pp.
79-93). Mahwah, NJ, US: Lawrence Erlbaum Associates Publishers.
Hintzman, D. L. (1984). MINERVA 2: A simulation model of human memory. Behavior
Research Methods, Instruments & Computers, 16(2), 96-101.
Hintzman, D. L. (1986). "Schema abstraction" in a multiple-trace memory model.
Psychological Review, 93(4), 411-428.
Hirsh, K. W., & Tree, J. J. (2001). Word association norms for two cohorts of British
adults. Journal of Neurolinguistics, 14(1), 1-44.
Hoffmann, J. P., & Cerbone, F. G. (2002). Parental substance use disorder and the risk of
adolescent drug abuse: an event history analysis. Drug & Alcohol Dependence,
66(3), 255-264.
Hollingworth, W., Ebel, B. E., McCarty, C. A., Garrison, M. M., Christakis, D. A., &
Rivara, F. P. (2006). Prevention of Deaths from Harmful Drinking in the United
States: The Potential Effects of Tax Increases and Advertising Bans on Young
Drinkers. Journal of Studies on Alcohol, 67(2), 300-308.
Hopfield, J. J., & Tank, D. W. (1986). Computing with Neural Circuits - a Model.
[Article]. Science, 233(4764), 625-633.
Hovardas, T., & Korfiatis, K. J. (2006). Word associations as a tool for assessing
conceptual change in science education. Learning and Instruction, 16(5), 416-
432.
Howell, R. D., Breivik, E., & Wilcox, J. B. (2007). Reconsidering formative
measurement. Psychological Methods, 12(2), 205-218.
Hoz, R., Champagne, A. B., & Klopfer, L. E. (1992). Multitechnique-multimethod matrix
methodology for construct validation and its application for the construct of
cognitive structure. Perceptual and Motor Skills, 74, 3-34.
Jarvis, C. B., MacKenzie, S. B., & Podsakoff, P. M. (2003). A critical review of construct
indicators and measurement model misspecification in marketing and consumer
research. [Review]. Journal of Consumer Research, 30(2), 199-218.
Jenkins, J. J. (1970). The 1952 Minnesota Word Association Norms. In L. J. Postman &
G. Keppel (Eds.), Norms of Word Association (pp. 466). New York: Academic
Press.

207
Jenkins, J. J., & Palermo, D. S. (1965). Further data on changes in word-association
norms. Journal of Personality and Social Psychology, 1(4), 303-309.
Jernigan, D. H., Ostroff, J., & Ross, C. (2005). Alcohol Advertising and Youth: A
Measured Approach. Journal of Public Health Policy, 26(3), 312.
Johnson, A., Xiao, L., Palmer, P., Sun, P., Wang, Q., Wei, Y., et al. (in press). Affective
decision-making deficits, linked to a dysfunctional ventromedial prefrontal cortex,
revealed in 10th grade Chinese adolescent binge drinkers. Neuropsychologia.
Johnson, C. A., Xiao, L., Palmer, P., Sun, P., Wang, Q., Wei, Y., et al. (2008). Affective
decision-making deficits, linked to a dysfunctional ventromedial prefrontal cortex,
revealed in 10th grade Chinese adolescent binge drinkers. Neuropsychologia,
46(2), 714-726.
Jung, C. G. (1910). The association method. American Journal of Psychology, 21(2),
219-269.
Jung, C. G. (1918/1969). Studies in word associations: Experiments in the diagnosis of
psychopathological conditions carried out at the psychiatric clinic of the
University of Zurich (M. D. Eder, Trans.). New York: Russell & Russell.
Kahneman, D. (2003). A perspective on judgment and choice: Mapping bounded
rationality. American Psychologist, 58(9), 697-720.
Kann, L. (2001). The Youth Risk Behavior Surveillance System: measuring health-risk
behaviors. American Journal of Health Behavior, 25(3), 272-277.
Kelly, A. B., Haynes, M. A., & Marlatt, G. A. (2008). The impact of adolescent tobacco-
related associative memory on smoking trajectory: An application of negative
binomial regression to highly skewed longitudinal data. Addictive Behaviors, 33,
640-650.
Kelly, A. B., Masterman, P., & Marlatt, G. A. (2006). Adolescent tobacco-related
associative memory: A cross-sectional and contextual analysis. Nicotine &
Tobacco Research, 8(1), 49-55.
Kelly, A. B., Masterman, P. W., & Marlatt, G. A. (2005). Alcohol-related associative
strength and drinking behaviours: Concurrent and prospective relationships. Drug
and Alcohol Review, 24(6), 489-498.
Kenney, K., & Scott, L. M. (2003). A review of the visual rhetoric literature. In L. M.
Scott & R. Batra (Eds.), Persuasive imagery: A consumer response perspective.
Advertising and consumer psychology. (pp. 17-56). Mahwah, NJ, US: Lawrence
Erlbaum Associates Publishers.

208
Killen, J. D., Hayward, C., Wilson, D. M., Haydel, K. F., Robinson, T. N., Taylor, C. B.,
et al. (1996). Predicting onset of drinking in a community sample of adolescents:
the role of expectancy and temperament. Addictive Behaviors, 21(4), 473-480.
Kilpatrick, D. G., Acierno, R., Saunders, B., Resnick, H. S., Best, C. L., & Schnurr, P. P.
(2000). Risk factors for adolescent substance abuse and dependence: Data from a
national sample. Journal of Consulting & Clinical Psychology, 68(1), 19-30.
Krishnan, H. S., & Shapiro, S. (1996). Comparing implicit and explicit memory for brand
names from advertisements. Journal of Experimental Psychology: Applied, 2(2),
147-163.
Leichliter, J. S., Meilman, P. W., Presley, C. A., & Cashin, J. R. (1998). Alcohol use and
related consequences among students with varying levels of involvement in
college athletics. Journal of American College Health, 46(6), 257-262.
Levy, D. A., Stark, C. E. L., & Squire, L. R. (2004). Intact conceptual priming in the
absence of declarative memory. Psychological Science, 15(10), 680-686.
Lewinsohn, P. M., Rohde, P., & Seeley, J. R. (1996). Alcohol consumption in high
school adolescents: frequency of use and dimensional structure of associated
problems. Addiction, 91(3), 375-390.
Lipsitz, A., Brake, G., Vincent, E. J., & Winters, M. (1993). Another round for the
brewers: Television ads and children's alcohol expectancies. Journal of Applied
Social Psychology, 23(6), 439-450.
Little, R. J. A., & Rubin, D. B. (2002). Statistical analysis with missing data (2nd ed.).
Hoboken, N.J.: Wiley.
Little, T. D., Cunningham, W. A., Shahar, G., & Widaman, K. F. (2002). To parcel or not
to parcel: Exploring the question, weighing the merits. Structural Equation
Modeling, 9(2), 151-173.
Lodish, L. M., Abraham, M., Kalmenson, S., Livelsberger, J., & et al. (1995). How T.V.
Advertising Works: A Meta-Analysis of 389 Real World Split Cable T.V.
Advertising Experiments. JMR, Journal of Marketing Research, 32(2), 125.
MacKenzie, S. B., Podsakoff, P. M., & Jarvis, C. B. (2005). The problem of measurement
model mis specification in behavioral and organizational research and some
recommended solutions. [Article]. Journal of Applied Psychology, 90(4), 710-
730.
MacKinnon, D. P. (2008). Introduction to statistical mediation analysis. New York:
Laurence Erlbaum.

209
MacKinnon, D. P., Fritz, M. S., Williams, J., & Lockwood, C. M. (in press). Distribution
of the product confidence limits for the indirect effect program PRODCLIN.
Behavioral Research Methods.
MacKinnon, D. P., Lockwood, C. M., Hoffman, J. M., West, S. G., & Sheets, V. (2002).
A comparison of methods to test mediation and other intervening variable effects.
Psychological Methods, 7(1), 83-104.
MacKinnon, D. P., Lockwood, C. M., & Williams, J. (2004). Confidence Limits for the
Indirect Effect: Distribution of the Product and Resampling Methods.
Multivariate Behavioral Research, 39(1), 99-128.
Madden, P. A., & Grube, J. W. (1994). The frequency and nature of alcohol and tobacco
advertising in televised sports, 1990 through 1992. American Journal of Public
Health, 84(2), 297-299.
Maggs, J. L., & Schulenberg, J. E. (2006). Initiation and Course of Alcohol Consumption
among Adolescents and Young Adults. In M. Galanter (Ed.), Alcohol problems in
adolescents and young adults: Epidemiology, neurobiology, prevention, and
treatment. (pp. 29-47). New York, NY, US: Springer Science + Business Media.
Marcoulides, G. A., & Schumacker, R. E. (2001). New developments and techniques in
structural equation modeling. Mahwah, NJ, US: Lawrence Erlbaum Associates
Publishers, 333.
Marin, G., Sabogal, F., Marin, B. V., Otero-Sabogal, R., & Perez-Stable, E. J. (1987).
Development of a short acculturation scale for Hispanics. Hispanic Journal of
Behavioral Sciences: Special Acculturation Research, 9(2), 183-205.
Marsh, H. W., Hau, K.-T., & Grayson, D. (2005). Goodness of Fit in Structural Equation
Models. In A. Maydeu-Olivares & J. J. McArdle (Eds.), Contemporary
psychometrics: A festschrift for Roderick P. McDonald. Multivariate applications
book series. (pp. 275-340). Mahwah, NJ, US: Lawrence Erlbaum Associates
Publishers.
Martin, C. S., Lynch, K. G., Pollock, N. K., & Clark, D. B. (2000). Gender differences
and similarities in the personality correlates of adolescent alcohol problems.
Psychology of Addictive Behaviors, 14(2), 121-133.
Martino, S. C., Collins, R. L., Ellickson, P. L., Schell, T. L., & McCaffrey, D. (2006).
Socio-environmental influences on adolescents' alcohol outcome expectancies: A
prospective analysis. Addiction, 101(7), 971-983.
Masson, M. E. J. (1995). A distributed memory model of semantic priming. Journal of
Experimental Psychology: Learning, Memory, & Cognition, 21(1), 3-23.

210
Mazzocco, P. J., & Brock, T. C. (2006). Understanding the Role of Mental Imagery in
Persuasion: A Cognitive Resources Model Analysis. In L. R. Kahle & C.-H. Kim
(Eds.), Creating images and the psychology of marketing communication.
Advertising and Consumer Psychology. (pp. 65-78). Mahwah, NJ, US: Lawrence
Erlbaum Associates Publishers.
McEvoy, C. L., Marschark, M., & Nelson, D. L. (1999). Comparing the mental lexicons
of deaf and hearing individuals. Journal of Educational Psychology, 91, 312-320.
McQuarrie, E. F., & Mick, D. G. (1999). Visual rhetoric in advertising: Text-interpretive,
experimental, and reader-response analyses. Journal of Consumer Research,
26(1), 37-54.
Meade, A. W., & Lautenschlager, G. J. (2004a). A Comparison of Item Response Theory
and Confirmatory Factor Analytic Methodologies for Establishing Measurement
Equivalence/lnvariance. Organizational Research Methods, 7(4), 361-388.
Meade, A. W., & Lautenschlager, G. J. (2004b). A Monte-Carlo study of confirmatory
factor analytic tests of measurement equivalence/invariance. Structural Equation
Modeling, 11(1), 60-72.
Meredith, W., & Teresi, J. A. (2006). An essay on measurement and factorial invariance.
Medical Care, 44(11 Suppl 3), S69-77.
Miles, D. R., Stallings, M. C., Young, S. E., Hewitt, J. K., Crowley, T. J., & Fulker, D.
W. (1998). A family history and direct interview study of the familial aggregation
of substance abuse: the adolescent substance abuse study. Drug & Alcohol
Dependence, 49(2), 105-114.
Millsap, R. E., & Kwok, O.-M. (2004). Evaluating the Impact of Partial Factorial
Invariance on Selection in Two Populations. Psychological Methods, 9(1), 93-
115.
Molina, B. S. G., & Pelham, W. E. (2001). Substance use, substance abuse, and LD
among adolescents with a childhood history of ADHD. Journal of Learning
Disabilities, 34(4), 333-342.
Molina, B. S. G., & Pelham, W. E. (2003). Childhood predictors of adolescent substance
use in a longitudinal study of children with ADHD. Journal of Abnormal
Psychology, 112(3), 497-507.
Mulligan, N. W. (1998). The role of attention during encoding in implicit and explicit
memory. Journal of Experimental Psychology: Learning, Memory, & Cognition,
24(1), 27-47.

211
Murray, D. M., & Short, B. (1995). Intraclass correlation among measures related to
alcohol use by young adults: Estimates, correlates and applications in intervention
studies. Journal of Studies on Alcohol, 56(6), 681-694.
Mustanski, B. S., Viken, R. J., Kaprio, J., & Rose, R. J. (2003). Genetic influences on the
association between personality risk factors and alcohol use and abuse. Journal of
Abnormal Psychology, 112(2), 282-289.
Muthen, L. K., & Muthen, B. O. (1998-2007). Mplus user's guide (4th ed.). Los Angeles,
CA: Muthen & Muthen.
Myers, M. G., Brown, S. A., & Mott, M. A. (1995). Preadolescent conduct disorder
behaviors predict relapse and progression of addiction for adolescent alcohol and
drug abusers. Alcoholism: Clinical & Experimental Research, 19(6), 1528-1536.
Nelson, D. L., & McEvoy, C. L. (2005). Implicitly Activated Memories: The Missing
Links of Remembering. In C. Izawa & N. Ohta (Eds.), Human learning and
memory: Advances in theory and application: The 4th Tsukuba International
Conference on Memory. (pp. 177-198): Lawrence Erlbaum Associates,
Publishers, Mahwah, NJ, US.
Nelson, D. L., McEvoy, C. L., & Dennis, S. (2000). What is free association and what
does it measure? Memory & Cognition, 28, 887-899.
Nelson, D. L., McEvoy, C. L., Janczura, G. A., & Xu, J. (1993). Implicit memory and
inhibition. Journal of Memory & Language, 32(5), 667-691.
Nelson, D. L., McEvoy, C. L., & Pointer, L. (2003). Spreading activation or spooky
action at a distance? Journal of Experimental Psychology: Learning, memory, and
cognition, 29, 42-52.
Nelson, D. L., McEvoy, C. L., & Schreiber, T. A. (2003). The University of South
Florida word association, rhyme, and word fragment norms., from
http://cyber.acomp.usf.edu/FreeAssociation/Intro.html
Nelson, D. L., McEvoy, C. L., Walling, J. R., & Wheeler, J. W. (1980). The University of
South Florida homograph norms. Behavior Research Methods & Instrumentation,
12(1), 16-37.
Nelson, D. L., McKinney, V. M., Gee, N. R., & Janczura, G. A. (1998). Interpreting the
influence of implicitly activated memories on recall and recognition.
Psychological Review, 105, 299-324.
NHTSA. (2003). Traffic safety facts 2001 (No. DOT HS 809 476). Washington, D.C.:
National Highway Traffic Safety Administration.

212
Nisbett, R. E., & Wilson, T. D. (1977). Telling more than we can know: Verbal reports
on mental processes. Psychological Review, 84(3), 231-259.
Noble, C. E. (1952). An analysis of meaning. Psychological Review, 49, 421-430.
Nordhielm, C. L. (2002). The influence of level of processing on advertising repetition
effects. Journal of Consumer Research, 29(3), 371.
Nordhielm, C. L. (2003). A levels-of-processing model of advertising repetition effects.
In L. M. Scott & R. Batra (Eds.), Persuasive imagery: A consumer response
perspective. Advertising and consumer psychology. (pp. 91-104). Mahwah, NJ,
US: Lawrence Erlbaum Associates Publishers.
Palermo, D. S. (1971). Characteristics of word association responses obtained from
children in grades one through four. Developmental Psychology, 5(1), 118-123.
Palermo, D. S., & Jenkins, J. J. (1964). Word association norms: Grade school through
college. Oxford, England: U. Minnesota Press, Oxford, England.
Palermo, D. S., & Molfese, D. L. (1972). Language acquisition from age five onward.
Psychological Bulletin, 78(6), 409-428.
Pentz, M. A., & Chou, C.-P. (1994). Measurement invariance in longitudinal clinical
research assuming change from development and intervention. Journal of
Consulting and Clinical Psychology, 62(3), 450-462.
Petty, R. E., & Cacioppo, J. T. (1986). Communication and persuasion: Central and
peripheral routes to attitude change. New York: Springer-Verlag.
Petty, R. E., & Wegener, D. T. (1999). The elaboration likelihood model: Current status
and controversies. In S. Chaiken & Y. Trope (Eds.), Dual-process theories in
social psychology. (pp. 37-72). New York: Guilford Press.
Phillips, B. J. (2003). Understanding visual metaphor in advertising. In L. M. Scott & R.
Batra (Eds.), Persuasive imagery: A consumer response perspective. Advertising
and consumer psychology. (pp. 297-310). Mahwah, NJ, US: Lawrence Erlbaum
Associates Publishers.
Pollio, H. R. (1966). The structural basis of word association behavior. Paris: Mouton.
Postman, L. J., & Keppel, G. (1970). Norms of word association. New York: Academic
Press.
Preece, P. F. W. (1978). Three-year stability of certain word-association indices.
Psychological Reports, 42, 25-26.

213
Primack, B. A., Gold, M. A., Land, S. R., & Fine, M. J. (2006). Association of Cigarette
Smoking and Media Literacy about Smoking among Adolescents. Journal of
Adolescent Health, 39(4), 465-472.
Raykov, T., & Marcoulides, G. A. (2006). A first course in structural equation modeling
(2nd ed.). Mahwah, NJ, US: Lawrence Erlbaum Associates Publishers, 238.
Riley, L. (2005). Drinking it in: Findings of the Valencia meeting on marketing and
promotion of alcohol to young people. In M. Grant & J. O'Connor (Eds.),
Corporate social responsibility and alcohol: The need and potential for
partnership. ICAP series on alcohol in society. (pp. 75-79). New York:
Routledge.
Robinson, T. N., Chen, H. L., & Killen, J. D. (1998). Television and music video
exposure and risk of adolescent alcohol use. Pediatrics, 102(5).
Roche, A. (2005). Establishing Good Practice in Responsible Drinks Promotion:
Illustrations of Good and Bad Practice From a Public Health Perspective. In M.
Grant & J. O'Connor (Eds.), Corporate social responsibility and alcohol: The
need and potential for partnership. ICAP series on alcohol in society. (pp. 115-
132). New York, NY, US: Routledge.
Roediger, H. L., & McDermott, K. B. (1995). Creating false memories: Remembering
words not presented in lists. Journal of Experimental Psychology: Learning,
Memory, & Cognition, 21(4), 803-814.
Rosenzweig, M. R., & Miller, K. M. (1966). Comparisons of word association responses
obtained in the United States, Australia, and England. Journal of Verbal Learning
& Verbal Behavior, 5(1), 35-41.
Rozin, P., Kurzer, N., & Cohen, A. B. (2002). Free associations to "food": The effects of
gender, generation and culture. Journal of Research in Personality, 36(5), 419-
441.
Rubin, D. B. (1987). Multiple imputation for nonresponse in surveys. New York: Wiley.
Saffer, H. (1997). Alcohol advertising and motor vehicle fatalites. Review of Economics
and Statistics, 79(3), 431-442.
Saffer, H. (2002). Alcohol advertising and youth. Journal of Studies on Alcohol, Suppl14,
173-181.
Saffer, H., & Dave, D. (2006). Alcohol advertising and alcohol consumption by
adolescents. Health Economics, 15(6), 617.

214
SAMHSA. (2004). Results from the 2003 national survey on drug use and health:
National Findings (No. DSDUH Series H-25, DHHS Publication No. SMA 04-
3964). Rockville, MD: Substance Abuse and Mental Health Services
Administration.
SAS. (2005). SAS statistical software (Version 9.1). Cary, NC: SAS Institute.
Schacter, D. L. (1987). Implicit memory: History and current status. Journal of
Experimental Psychology: Learning, Memory, & Cognition, 13(3), 501-518.
Schafer, J. L., & Graham, J. W. (2002). Missing data: Our view of the state of the art.
Psychological Methods, 7(2), 147-177.
Schafer, J. L., & Olsen, M. K. (1998). Multiple imputation for multivariate missing-data
problems: A data analyst's perspective. Multivariate Behavioral Research. Special
Issue: Innovative methods for prevention research, 33(4), 545-571.
Schmidt, W., & Popham, R. E. (1978). The single distribution theory of alcohol
consumption. A rejoinder to the critique of Parker and Harman. Journal of Studies
on Alcohol, 39(3), 400-419.
Schooler, C., Feighery, E., & Flora, J. A. (1996). Seventh graders' self-reported exposure
to cigarette marketing and its relationship to their smoking behavior. American
Journal of Public Health, 86(9), 1216-1221.
Schumann, D. W., Petty, R. E., & Clemons, D. S. (1990). Predicting the effectiveness of
different strategies of advertising variation: A test of the repetition-variation
hypotheses. Journal of Consumer Research, 17(2), 192-202.
Seamon, J. G., Williams, P. C., Crowley, M. J., Kim, I. J., & et al. (1995). The mere
exposure effect is based on implicit memory: Effects of stimulus type, encoding
conditions, and number of exposures on recognition and affect judgments.
Journal of Experimental Psychology: Learning, Memory, and Cognition, 21(3),
711-721.
Seger, C. A., Rabin, L. A., Desmond, J. E., & Gabrieli, J. D. (1999). Verb generation
priming involves conceptual implicit memory. Brain & Cognition, 41(2), 150-
177.
Shapiro, S., Heckler, S. E., & MacInnis, D. J. (1997). Measuring and assessing the impact
of preattentive processing on ad and brand attitudes. In W. D. Wells (Ed.),
Measuring advertising effectiveness. Advertising and consumer psychology. (pp.
27-44). Mahwah, NJ, US: Lawrence Erlbaum Associates Publishers.
Skinner, B. F. (1937). The distribution of associated words. Psychological Record, 1, 71-
76.

215
Slater, M. D., Rouner, D., Murphy, K., Beauvais, F., & et al. (1996). Male adolescents'
reactions to TV beer advertisements: The effects of sports content and
programming context. Journal of Studies on Alcohol, 57(4), 425-433.
Smart, R. G. (1988). Does alcohol advertising affect overall consumption? A review of
empirical studies. Journal of Studies on Alcohol, 49(4), 314-323.
Snyder, L. B., Milici, F. F., Slater, M., Sun, H., & Strizhakova, Y. (2006). Effects of
alcohol advertising exposure on drinking among youth. Archives of Pediatrics &
Adolescent Medicine, 160(1), 18-24.
Spence, D. P., & Owens, K. C. (1990). Lexical co-occurrence and association strength.
Journal of Psycholinguistic Research, 19(5), 317-330.
Spirito, A., Barnett, N. P., Lewander, W., Colby, S. M., Rohsenow, D. J., Eaton, C. A., et
al. (2001). Risks associated with alcohol-positive status among adolescents in the
emergency department: a matched case-control study. Journal of Pediatrics,
139(5), 694-699.
Stacy, A. W. (1995). Memory association and ambiguous cues in models of alcohol and
marijuana use. Experimental & Clinical Psychopharmacology, 3(2), 183-194.
Stacy, A. W. (1997). Memory activation and expectancy as prospective predictors of
alcohol and marijuana use. Journal of Abnormal Psychology, 106(1), 61-73.
Stacy, A. W., Ames, S. L., & Grenard, J. L. (2006). Word association tests of associative
memory and implicit processes: Theoretical and assessment issues. In A. W.
Stacy & R. W. Wiers (Eds.), Handbook on implicit cognition and addiction (pp.
75-90). Thousand Oaks, CA: Sage.
Stacy, A. W., Ames, S. L., & Knowlton, B. J. (2004). Neurologically plausible
distinctions in cognition relevant to drug use etiology and prevention. Substance
Use & Misuse, 39(10-12), 1571-1623.
Stacy, A. W., Ames, S. L., Sussman, S., & Dent, C. W. (1996). Implicit cognition in
adolescent drug use. Psychology of Addictive Behaviors, 10(3), 190-203.
Stacy, A. W., Leigh, B. C., & Weingardt, K. R. (1993). An Individual-difference
perspective applied to normative associative stregnth. Paper presented at the
Annual meeting of the Psychonomic Society, Washington D.C.
Stacy, A. W., Leigh, B. C., & Weingardt, K. R. (1994). Memory accessibility and
association of alcohol use and its positive outcomes. Experimental and Clinical
Psychopharmacology, 2, 269-282.

216
Stacy, A. W., Leigh, B. C., & Weingardt, K. R. (1997). An individual-difference
perspective applied to word association. Personality & Social Psychology
Bulletin, 23(3), 229-237.
Stacy, A. W., & Newcomb, M. D. (1998). Memory association and personality as
predictors of alcohol use: Mediation and moderator effects. Experimental &
Clinical Psychopharmacology, 6(3), 280-291.
Stacy, A. W., Pearce, S. G., Zogg, J. B., Unger, J., & Dent, C. W. (2004). A nonverbal
test of naturalistic memory for alcohol commercials. Psychology & Marketing,
21(4), 295-322.
Stacy, A. W., Widaman, K. F., Hays, R. D., & DiMatteo, M. R. (1985). Validity of self-
reports of alcohol and other drug use: A multitrait-multimethod assessment.
Journal of Personality & Social Psychology, 49(1), 219-232.
Stacy, A. W., Zogg, J. B., Unger, J. B., & Dent, C. W. (2004). Exposure to Televised
Alcohol Ads and Subsequent Adolescent Alcohol Use. American Journal of
Health Behavior, 28(6), 498-509.
Stark, S., Chernyshenko, O. S., & Drasgow, F. (2006). Detecting Differential Item
Functioning With Confirmatory Factor Analysis and Item Response Theory:
Toward a Unified Strategy. Journal of Applied Psychology, 91(6), 1292-1306.
Steenkamp, J., & Baumgartner, H. (1998). Assessing measurement invariance in cross-
national consumer research. [Article]. Journal of Consumer Research, 25(1), 78-
90.
Stewart, D. W. (1989). Measures, methods, and models in advertising research. Journal
of Advertising Research, 29(3), 54-60.
Stewart, D. W., & Furse, D. H. (1985). The effects of television advertising execution on
recall, comprehension, and persuasion. Psychology & Marketing, 2(3), 135.
Strickland, D. E. (1982). Alcohol Advertising: Orientations and Influence. International
Journal of Advertising, 1(4), 307.
Strickland, D. E. (1983). Advertising exposure, alcohol consumption and misuse of
alcohol. In M. Grant, M. Plant & A. Williams (Eds.), Economics and alcohol:
Consumption and controls. New York: Gardner Press.
Szalay, L. B., & Brent, J. E. (1965). Cultural meanings and values: A method of
empirical assessment. Washington, DC: American University.
Szalay, L. B., & Deese, J. (1978). Subjective meaning and culture: An assessment
through word associations. Hillsdale, NJ: Lawrence Erlbaum Associates.

217
Szalay, L. B., Inn, A., & Doherty, K. T. (1996). Social influences: Effects of the social
environment on the use of alcohol and other drugs. Substance Use & Misuse,
31(3), 343-373.
Szalay, L. B., Strohl, J. B., & Doherty, K. T. (1999). Psychoenvironmental forces in
substance abuse prevention. Dordrecht, Netherlands: Kluwer Academic
Publishers.
Tapert, S. F., Granholm, E., Leedy, N. G., & Brown, S. A. (2002). Substance use and
withdrawal: Neuropsychological functioning over 8 years in youth. Journal of the
International Neuropsychological Society, 8(7), 873-883.
Tellis, G. J. (2005). Advertising's role in capitalist markets: What do we know and where
do we go from here? [Article]. Journal of Advertising Research, 45(2), 162-170.
Teresi, J. A. (2006a). Different approaches to differential item functioning in health
applications. Advantages, disadvantages and some neglected topics. Medical
Care, 44(11 Suppl 3), S152-170.
Teresi, J. A. (2006b). Overview of quantitative measurement methods. Equivalence,
invariance, and differential item functioning in health applications. Medical Care,
44(11 Suppl 3), S39-49.
Thorlindsson, T., Vilhjalmsson, R., & Valgeirsson, G. (1990). Sport participation and
perceived health status: A study of adolescents. Social Science & Medicine, 31(5),
551-556.
Thorndike, E. L. (1932). The significance of responses in the free association test.
Journal of Applied Psychology, 16, 247-253.
Thorne, B. M., & Henley, T. B. (2001). Connections in the history and systems of
psychology (2nd ed.). Boston, MA: Houghton Mifflin Co.
Thorson, E., & Zhao, X. (1997). Television viewing behavior as an indicator of
commercial effectiveness. In W. D. Wells (Ed.), Measuring advertising
effectiveness. Advertising and consumer psychology. (pp. 221-237). Mahwah, NJ,
US: Lawrence Erlbaum Associates Publishers.
Thush, C., & Wiers, R. W. (2007). Explicit and implicit alcohol-related cognitions and
the prediction of future drinking in adolescents. Addictive Behaviors, 32(7), 1367-
1383.
Tiffany, S. T. (1990). A cognitive model of drug urges and drug-use behavior: Role of
automatic and nonautomatic processes. Psychological Review, 97(2), 147-168.

218
U.S. Census Bureau. (2000). Geographic comparison table: California - counties.
Retrieved May 28, 2007, from http://factfinder.census.gov
Underwood, B. J. (1965). False recognition produced by implicit verbal responses.
Journal of Experimental Psychology, 70(1), 122-129.
Underwood, B. J., & Reichardt, C. S. (1975). Implicit associational responses produced
by words in pairs of unrelated words. Memory & Cognition, 3(4), 405-408.
Unger, J. B., Johnson, C. A., & Rohrbach, L. A. (1995). Recognition and liking of
tobacco and alcohol advertisements among adolescents: Relationships with
susceptibility to substance use. Preventive Medicine, 24(5), 461-466.
Unger, J. B., Schuster, D., Zogg, J. B., Dent, C. W., & Stacy, A. W. (2003). Alcohol
advertising exposure and adolescent alcohol use: A comparison of exposure
measures. Addiction Research & Theory, 11(3), 177-193.
USDA. (2007). Food CPI, Prices and Expenditures: Alcoholic Beverages. Retrieved.
from
http://www.ers.usda.gov/briefing/CPIFoodAndExpenditures/Data/table4.htm.
Vandenberg, R. J., & Lance, C. E. (2000). A review and synthesis of the measurement
invariance literature: Suggestions, practices, and recommendations for
organizational research. Organizational Research Methods, 3(1), 4-69.
Watson, J. M., Balota, D. A., & Roediger, H. L. (2003). Creating false memories with
hybrid lists of semantic and phonological associates: Over-additive false
memories produced by converging associative networks. Journal of Memory &
Language, 49(1), 95-118.
Watson, J. M., Balota, D. A., & Roediger III, H. L. (2003). Creating false memories with
hybrid lists of semantic and phonological associates: Over-additive false
memories produced by converging associative networks. Journal of Memory &
Language, 49(1), 95-118.
Wechsler, H., Davenport, A. E., Dowdall, G. W., Grossman, S. J., & et al. (1997). Binge
drinking, tobacco, and illicit drug use and involvement in college athletics: A
survey of students at 140 American colleges. Journal of American College
Health, 45(5), 195-200.
Weingardt, K. R., Stacy, A. W., & Leigh, B. C. (1996). Automatic activation of alcohol
concepts in response to positive outcomes of alcohol use. Alcoholism: Clinical &
Experimental Research, 20(1), 25-30.

219
Wheeler, S. C., Petty, R. E., & Bizer, G. Y. (2005). Self-Schema Matching and Attitude
Change: Situational and Dispositional Determinants of Message Elaboration.
Journal of Consumer Research, 31(4), 787-797.
Wills, T. A., Resko, J. A., Ainette, M. G., & Mendoza, D. (2004). Role of Parent Support
and Peer Support in Adolescent Substance Use: A Test of Mediated Effects.
Psychology of Addictive Behaviors, 18(2), 122-134.
Wills, T. A., Sandy, J. M., Yaeger, A. M., Cleary, S. D., & Shinar, O. (2001). Coping
dimensions, life stress, and adolescent substance use: A latent growth analysis.
Journal of Abnormal Psychology, 110(2), 309-323.
Wills, T. A., & Stoolmiller, M. (2002). The role of self-control in early escalation of
substance use: A time-varying analysis. Journal of Consulting & Clinical
Psychology, 70(4), 986-997.
Windle, M., & Windle, R. C. (2006). Alcohol Consumption and Its Consequences among
Adolescents and Young Adults. In M. Galanter (Ed.), Alcohol problems in
adolescents and young adults: Epidemiology, neurobiology, prevention, and
treatment. (pp. 67-83). New York, NY, US: Springer Science + Business Media.
Winters, K. C., Stinchfield, R. D., & Henly, G. A. (1993). Further validation of new
scales measuring adolescent alcohol and other drug abuse. Journal of Studies on
Alcohol, 54(5), 534-541.
Wirth, R. J., & Edwards, M. C. (2007). Item Factor Analysis: Current Approaches and
Future Directions. Psychological Methods, 12(1), 58-79.
Wollen, K. A. (1980). Frequency of occurrence and concreteness ratings of homograph
meanings. Behavior Research Methods & Instrumentation, 12(1), 8-15.
Wood, M. D., Read, J. P., Mitchell, R. E., & Brand, N. H. (2004). Do Parents Still
Matter? Parent and Peer Influences on Alcohol Involvement Among Recent High
School Graduates. Psychology of Addictive Behaviors, 18(1), 19-30.
Woodworth, R. S. (1921). Psychology: A study of mental life. New York, NY, US: Henry
Holt and Co, Inc.
Wright-Isak, C., Faber, R. J., & Horner, L. (1997). Comprehensive measurement of
advertising effectiveness: Notes from the marketplace. In W. D. Wells (Ed.),
Measuring advertising effectiveness. Advertising and consumer psychology. (pp.
3-12). Mahwah, NJ, US: Lawrence Erlbaum Associates Publishers.
Wyllie, A., Zhang, J. F., & Casswell, S. (1998a). Positive responses to televised beer
advertisements associated with drinking and problems reported by 18 to 29-year-
olds. Addiction, 93(5), 749-760.

220
Wyllie, A., Zhang, J. F., & Casswell, S. (1998b). Responses to televised alcohol
advertisements associated with drinking behaviour of 10-17-year-olds. Addiction,
93(3), 361-371.
Zajonc, R. B. (1968). Attitudinal Effects of Mere Exposure. Journal of Personality and
Social Psychology, 9(2, Pt.2), 1-27.
Zeelenberg, R., Pecher, D., Shiffrin, R. M., & Raaijmakers, J. G. W. (2003). Semantic
context effects and priming in word association. Psychonomic Bulletin & Review,
10(3), 653-660.
Zogg, J. B. (2004). Adolescent exposure to alcohol advertising: A prospective extension
of Strickland's Model. Unpublished Doctoral Dissertation, University of Southern
California, Los Angeles.
Zogg, J. B., Ma, H., Dent, C. W., & Stacy, A. W. (2004). Self-generated alcohol
outcomes in 8th and 10th graders: Exposure to vicarious sources of alcohol
information. Addictive Behaviors, 29, 3-16.

Abstract (if available)

Abstract Three studies provided support for the hypothesis that televised alcohol advertisements influence underage drinking. Data were collected from 3,890 students in a prospective study covering the 7th through 10th grades. Measures of exposure to alcohol advertising included self-reported observation of alcohol ads on TV and 2 memory measures (top of mind awareness and cued recall). Opportunity-based measures assessed exposure to alcohol ads indirectly by first asking about the frequency that participants watched specific programs (i.e., popular shows and sports programs) and then weighting the responses by the frequency of alcohol ads broadcast during each of those programs. One key affective measure assessed how much participants liked alcohol ads compared to other ads on TV. Additional measures assessed demographic variables, sports participation, alcohol use, and problems associated with alcohol use. The results from a confirmatory factor analysis (CFA) in study 1 (chapter 2) demonstrated that each of the measures related to alcohol ads, except top of mind awareness, had good measurement properties. The measures loaded well on single latent factors as expected with no crossfactor loadings and the factor loadings and thresholds were invariant across gender. Study 2 (chapter 3) provided support for a causal relationship between exposure to televised alcohol advertising and underage drinking. Structural equation modeling of alcohol consumption showed that exposure to alcohol ads and/or liking of those ads in 7th grade were predictive of latent growth factors for alcohol use after controlling for a range of covariates. There was a significant total effect for males and a significant mediated effect for females of exposure to alcohol ads and liking of those ads in 7th grade through latent growth factors for alcohol use on alcohol-related problems in 10th grade.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

A network analysis of online and offline social influence processes in relation to adolescent smoking and alcohol use

PDF

The role of depression symptoms on social information processing and tobacco use among adolescents

PDF

Sociocultural stress, coping and substance use among Hispanic/Latino adolescents

PDF

Intrapersonal and environmental factors associated with Chinese youth alcohol use experimentation and binge drinking behaviors

PDF

Adolescent social networks, smoking, and loneliness

PDF

Anxiety symptoms and nicotine use among adolescents and young adults

PDF

Psychosocial and behavioral ractors associated with emotional eating in adolescents

PDF

Genetic variants and smoking progression in Chinese adolescents

PDF

The effects of mindfulness on adolescent cigarette smoking: Measurement, mechanisms, and theory

PDF

Cultural risk and protective factors for tobacco use behaviors and depressive symptoms among American Indian adolescents in California

PDF

Examining the longitudinal relationships between community violence exposure and aggressive behavior among a sample of maltreated and non-maltreated adolescents

PDF

Factors and correlates of sexual behaviors among young adults from continuation high schools

PDF

Energy drink consumption, substance use and attention-deficit/hyperactivity disorder among adolescents

PDF

Social network influences on depressive symptoms among Chinese adolescents

PDF

Affective decision-making predictive of Chinese adolescents drinking behaviors: a longitudinal study

PDF

Social self-control and adolescent substance use

PDF

Problematic alcohol use in Hispanic emerging adults: the role of perceived discrimination, cultural identity, and salient cultural values

PDF

A sociocultural and developmental approach to intimate partner violence among a sample of Hispanic emerging adults

PDF

Effects of near-roadway air pollution exposure on obesity, obesity-related behavior, and neurobehavioral deficits during peripuberty

PDF

Risk factors associated with smoking initiation among Chinese adolescents: a matched case-control study

Asset Metadata

Creator Grenard, Jerry L. (author)

Core Title Exposure to alcohol advertising on television and alcohol use among young adolescents

School Keck School of Medicine

Degree Doctor of Philosophy

Degree Program Preventive Medicine (Health Behavior)

Publication Date 07/14/2008

Defense Date 05/30/2008

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag Adolescents,advertising,Alcohol,implicit cognition,OAI-PMH Harvest

Language English

Advisor Stacy, Alan W. (committee chair), Ames, Susan L. (committee member), Cody, Michael J. (committee member), Dent, Clyde (committee member), Unger, Jennifer B. (committee member)

Creator Email grenard@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m1344

Unique identifier UC1438515

Identifier etd-Grenard-20080714 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-190930 (legacy record id),usctheses-m1344 (legacy record id)

Legacy Identifier etd-Grenard-20080714.pdf

Dmrecord 190930

Document Type Dissertation

Rights Grenard, Jerry L.

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Repository Name Libraries, University of Southern California

Repository Location Los Angeles, California

Repository Email cisadmin@lib.usc.edu