University of Southern California Dissertations and Theses

Microsoft Word - $ASQ100489_supp_undefined_9C841A88-B234-11E0-8B96-E32D9E1A67F9.doc

(USC Thesis Other)

Microsoft Word - $ASQ100489_supp_undefined_9C841A88-B234-11E0-8B96-E32D9E1A67F9.doc

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content PROMOTING PRO-ENVIRONMENTAL BEHAVIOR AMONG UNIVERSITY DORMITORY RESIDENTS by Nicole Diana Sintov A Dissertation Presented to the FACULTY OF THE USC GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements of the Degree DOCTOR OF PHILOSOPHY (PSYCHOLOGY) August 2011 Copyright 2011 Nicole Diana Sintov ii Acknowledgments The author would like to acknowledge Margaret Gatz, Daniel Mazmanian, Norman Miller, and Donna Spruijt-Metz, and for their thoughtful input on the design of the project; the members of the Prescott lab for their feedback and support, including the undergraduate research assistants who worked on the project; the staff at Phoenix Energy Technologies for developing and updating the intervention website as needed; the staff at USC Facilities Management for furnishing the electricity data and funding the website; graphic artist Craig Stubing for designing the advertisements; and the many business donors whose generous merchandise donations served as participation incentives. iii Table of Contents Acknowledgments ii List of Tables v List of Figures vii Abstract viii Preface x Chapter One: Effectiveness of a Competition-Based Intervention in Promoting Pro-environmental Behavior in a University Residential Setting Chapter 1 Abstract 1 Chapter 1 Background 3 Chapter 1 Methods 10 Chapter 1 Results 23 Chapter 1 Discussion 42 Chapter 1 References 53 Chapter Two: Explaining Change in Pro-environmental Behavior Based on the Theory of Planned Behavior and Norm Activation Models Chapter 2 Abstract 59 Chapter 2 Background 61 Chapter 2 Methods 69 Chapter 2 Results 80 Chapter 2 Discussion 95 Chapter 2 References 104 Chapter Three: The Influence of Social Desirability and Item Priming Effects on Reports of Pro-environmental Behavior Chapter 3 Abstract 109 Chapter 3 Background 111 Chapter 3 Method 116 Chapter 3 Results 124 Chapter 3 Discussion 130 Chapter 3 References 135 Conclusion 138 Comprehensive Bibliography 145 iv Appendices Appendix 1A: Appeals Posters 155 Appendix 1B: Focus Group Questions 157 Appendix 1C: Exploratory Analyses with Group Identification 158 Appendix 1D: Total Sample Demographic Characteristics 160 Appendix 2A: Measures 162 Appendix 2B: Fitting Alternative TPB Models 170 Appendix 2C: Fitting NAM Alternative Models 172 Appendix 2D: Fitting Alternative Hybrid Models 173 Appendix 2E: Sample Bias Analyses 175 Appendix 2F: Validity Verification 178 Appendix 2G: Latent Difference Score Model 188 Appendix 2H: Testing for Differences on Model Parameters Across Intervention and Control Groups Based on Change Models 189 Appendix 3A: Behavior and Intentions Measures 191 Appendix 3B: Intentions-Behavior Correlations 193 v List of Tables Table 1.1: Frequencies (Percentages) of Survey Respondents Identified Through Various Recruitment Strategies 15 Table 1.2: Participant Demographic Characteristics by Intervention Status 18 Table 1.3. Participant Characteristics on Key Variables by Intervention Status 21 Table 1.4: Mean Daily Electricity Consumption (SD) in kWh among USC Dormitory Buildings in Fall 2008 and 2009 26 Table 1.5: Fixed Effect Parameter Estimates and Standard Errors from Multilevel Regression Models (N=1414) 31 Table 1.6: Sequential R 2 Values Associated with Adding Additional Predictors to Each Model for General Pro-Environmental Behavior (PEB), Electricity Use Subscale Score, and Consumer Behavior Outcomes (n=187) 37 Table 1.7: Standardized Regression Coefficients for Final Models (n=187) 37 Table 2.1: Sample Characteristics for Scales Used in Analyses 74 Table 2.2: Pearson Correlations Among Key Measures at Times 1 and 2 75 Table 2.3: Participant Characteristics by Intervention Status Based on Time 1 Data 77 Table 2.4a: Model Fit Indices and R² Values for Alternative Change Models 89 Table 2.4b: Evaluating Role of ∆Int on ∆PEB 89 Table 2.4c: Model Fit Indices and R² Values for Alternative Change Models 90 Table 2.4d: Evaluating Role of ∆Int on ∆PEB 91 Table 2.5: NAM Change Models 94 Table 3.1: Participant Characteristics (n=254) 120 Table 3.2: Survey Block Sequencing 122 vi Table 3.3: Univariable Regression Results 126 Table 3.4a: Hypothesis 2: BEHAV1 Models, n=254 127 Table 3.4b: Hypothesis 2: BEHAV2 Models, n=254 128 Table 3.5: Model Parameters for Order Effects at the Level of the General Construct vs. Specific Item Content, n=254 130 vii List of Figures Figure 1.1: Study Timeline 14 Figure 1.2. Daily Mean Electricity Consumption Among Target Buildings During the Fall 2009 Intervention and Baseline Periods 24 Figure 1.3. 2008 Daily Electricity Use Averaged by Week for On-Campus Buildings Excluding Building E 27 Figure 1.4. 2009 Daily Electricity Use Averaged by Week for On-Campus Buildings Excluding Building E 27 Figure 2.1. Norm Activation Model 63 Figure 2.2. Theory of Planned Behavior 64 Figure 2.3. Final Theory of Planned Behavior Model with Unstandardized Path Coefficients 83 Figure 2.4. Norm Activation Model Final Model Unstandardized Path Coefficients 84 Figure 2.5. Final Hybrid Model with Unstandardized Path Coefficients 85 Figure 2.6. Final TPB Change Model with Unstandardized Path Coefficients 92 Figure 2.7. Final NAM Change Model with Unstandardized Path Coefficients 94 Figure 4. Combining Features of the Theory of Planned Behavior, Goal Setting Theory, and Transtheoretical Model to Suggest Appropriate Intervention Targets 142 viii Abstract The primary aim of this study was to evaluate the effectiveness of a competition- based intervention in promoting pro-environmental behavior (PEB) among university dormitory residents. The study used a prospective quasi-experimental intervention design with a control group. The intervention, which included appeals and a group-level incentive, was implemented using a building-versus-building competition framework. Additionally, a website was developed with access to different supplementary intervention content assigned by building residence. Participants completed baseline and follow-up self-report surveys regarding energy use behaviors and key constructs derived from the Theory of Planned Behavior and the Norm Activation Model. A total of 298 undergraduate students participated in the first survey, and 225 also took part in the follow-up survey. No significant effects of the intervention were found for self-reported PEB. Changes in electricity use between the baseline and competition phases ranged from a decrease of 7.1% to an increase of 3.1% among the seven intervention buildings. Mixed level models based on electricity data showed that when adjusting for temperature, intervention buildings used significantly more electricity during the intervention phase relative to the control phase during the project year (2009) and the preceding year (2008). A secondary objective of the project was to investigate mechanisms of change in PEB by examining relationships among TPB and NAM constructs and investigating which constructs explain change in PEB. To address this objective, structural equation modeling was applied to the self-report data from both surveys. Results provided preliminary evidence supporting the extension of the TPB to addressing change in PEB. Results also ix indicated that changes in intentions were most important for explaining changes in PEB. A final aim of this project was to evaluate the influences of impression management and item order effects on self-reported PEB. Higher impression management scores predicted higher PEB scores. Weak evidence of item order effects were also observed. These findings suggest that social desirability biases should be considered in designing surveys, conducting analyses, and interpreting study findings. Understanding how interventions work to change PEB and its determinants, and how to measure and examine these processes reliably, remain important avenues for future research. x Preface Human activities have been found to contribute to climate change (Sabine & Tanhua, 2010; Stott et al., 2004). The Intergovernmental Panel on Climate Change estimates that over the next 100 years, the mean global surface temperature will rise by 1.1 – 6.4° Centigrade, depending in part on human energy use (2007). Modifying human behavior is one strategy to mitigate anthropogenic contributions to climate change. In the United States (US) and Europe, the residential sector is a major consumer of energy, making residential settings an important target for intervention efforts that attempt to reduce energy consumption. For instance, energy use among the US residential sector in 2009 accounted for 22% of all electricity consumed during that year (US Energy Information Administration, 2010). Similarly, among the 27 European Union nations, energy use among the residential sector accounted for approximately 29% of all electricity consumed in 2004 (Bertoldi & Atanasiu, 2006). Over the past several decades, efforts to reduce household energy consumption (i.e., electricity, natural gas) and promote other types of pro-environmental behavior have been implemented with varying levels of success (for review see Abrahamse et al. 2005). Residential energy reduction interventions Among interventions that aim to reduce residential energy use, two overarching intervention approaches have been used: behavioral antecedent strategies and behavioral consequence strategies. Antecedent interventions aim to influence a given behavior prior to its performance (Abrahamse et al. 2005). Examples of antecedent interventions include information provision, prompting, and modeling. Consequence strategies are xi implemented after behaviors are performed, with the goal of impacting future engagement in the same behaviors. The primary consequence strategies that have been used in energy reduction interventions have been feedback and rewards. These intervention strategies have received varying levels of empirical support, with some showing promise for reducing energy consumption in residential settings. This literature is reviewed later in Papers 1 and 2. Energy reduction interventions in university settings In addition to residential settings, higher education institutions are a prime target for energy reduction interventions for two main reasons. First, universities consume large quantities of energy. For instance, the University of Southern California (USC) University Park Campus annual resource use includes approximately 155 million kWh of electricity, 4 million therms of natural gas, and 270 million gallons of water (University of Southern California, 2008). This represents a greater annual demand for resources than generated in 2007 by Lompoc, California, a city with a population of over 40,000 (California Department of Energy, 2009). Additionally, university-based interventions can reach large numbers of individuals, and due to the combination of students, faculty, and staff, have the potential to influence behavior beyond residential settings. For instance, at USC, over 30,000 graduate and undergraduate students are enrolled annually, and of these, over 5,000 reside in USC-owned housing. The university also employs over 15,000 faculty and staff. Interventions that result in the adoption of energy reduction behavior among such large communities have some potential to reduce future energy demands, particularly if behavior can be sustained long-term. xii Although some universities have had energy conservation plans in place for decades, outside monitoring of sustainability practices has placed pressure on universities to implement such practices. For example, the Sustainable Endowments Institute’s annual campus evaluations are intended to promote effective sustainability policies among universities by making their sustainability practices public and enabling them to learn from each other’s experiences. Additionally, dozens of dormitory-versus-dormitory sustainability competitions have been conducted at major US universities, but only a few have been empirically evaluated (McClelland & Cook, 1980a; McClelland & Cook, 1980b; Petersen et al., 2007). Without evaluation, it remains unknown exactly which behaviors and underlying determinants of behavior are modified by the interventions, the duration of these effects, and the mechanisms that contribute to change. Paper 1 will discuss the effectiveness of a campus-based intervention in changing various types of pro-environmental behavior, and Paper 2 will examine mechanisms of intraindividual change in pro-environmental behavior. Theoretical models of behavior Related to understanding determinants of behavior, numerous studies have shown that pro-environmental behavior can be described cross-sectionally using a number of theories, including the Theory of Planned Behavior (TPB; Ajzen & Madden, 1986) and Norm Activation Model (NAM; Schwartz, 1994). Refer to pages 68 and 69 in Paper 2 for graphical representations of the NAM and TPB, respectively. Pro-environmental behavior (PEB) is defined here as any behavior that supports the sustainability of natural ecosystems, environmental health, and “contribute[s] towards environmental preservation xiii and/or conservation” (Axelrod & Lehman 1993, p. 153). The above theories have received a fair amount of empirical support in explaining PEB and are discussed in Paper 2. However, studies that have evaluated conceptual models of PEB have generally used cross-sectional designs and have focused on explaining levels of behavior rather than changes in behavior. Therefore, relatively little is known about how such models may be applied to describing processes of behavior change. Related, interventions relevant to reducing energy consumption generally have not related intervention strategies to conceptual models of PEB, nor have they provided for measurement of theory-based determinants of behavior. This has limited current understanding about which behavioral determinants are affected by which intervention strategies, and how these contribute to behavior change. Identifying the factors that contribute to behavior change has the potential to extend the theoretical literature and can help guide intervention work. Methodology The study of theoretical models of PEB relies heavily on the use of self-report methodology, which is subject to a variety of method biases, including social desirability and item priming effects. In particular, a bias toward positive impression management is typically associated with exaggerated reports of prosocial behavior. Related to theories like the TPB that include future intentions as a theoretical construct, available research has found less evidence of impression management bias when questions about future intentions are presented prior to questions about past prosocial behavior, compared to when individuals respond to questions about future intentions after responding to questions about their recent behavior (Beebe et al., 2008; Johnson et al., 2004). This xiv research has been conducted with older adult populations in the area of health promotion. It is important to know whether findings generalize to other populations and topics, such as PEB. Understanding these biases is important not only for the validity of individual study conclusions, but also for the validity of conceptual models that include the constructs intentions and behavior. These models rely on the accumulation of evidence from studies that use common methods and are subject to similar biases. Study goals The present study had three main objectives. The primary aim of this study was to evaluate the effectiveness of a competition-based intervention in promoting PEB among residents of undergraduate residence halls at the University of Southern California in fall 2009. A secondary objective of the project was to investigate mechanisms of change in PEB by examining relationships among TPB and NAM constructs and investigating which constructs explain change in PEB. A final aim was to evaluate the influences of impression management and item order effects on self-reported PEB. The three papers that follow address each of these objectives in turn. 1 Chapter One: Effectiveness of a Competition-Based Intervention in Promoting Pro- environmental Behavior in a University Residential Setting Chapter 1 abstract Background. Many university-based energy reduction efforts have been implemented in recent years, but few have been subjected to empirical evaluation. The aim of the present study was to evaluate the effectiveness of a competition-based intervention in reducing energy consumption in a university residential setting. Methods. The study used a prospective quasi-experimental design with a control group. To engage participants, the project was implemented using a building-versus- building competition framework. A group-level incentive and appeals constituted intervention stimuli. Additionally, a website was developed with access to different supplementary intervention content assigned by building residence. Participants completed baseline and follow-up self-report surveys regarding energy use behaviors and key constructs derived from the Theory of Planned Behavior and the Norm Activation Model. Data from 187 survey respondents, which included 80 intervention and 107 control participants, were used to examine changes in self-reported behavior. Results. Regression analyses based on survey data indicated no statistically significant effect of the intervention on self-reported pro-environmental behavior. Descriptive analyses based on electricity data indicated that changes in electricity use between the baseline and competition phases ranged from a decrease of 7.1% to an increase of 3.1% among the seven intervention buildings. Mixed level models based on 2 electricity data showed that when adjusting for temperature, intervention buildings used significantly more electricity during the intervention phase relative to the control phase during the project year (2009) and the preceding year (2008). Conclusions. The intervention was not effective in promoting pro-environmental behavior. Additionally, only six students registered to use the website, which suggests that the requirement for registration was a barrier to its use. Different methods of intervention content delivery may be better suited to an undergraduate population. That electricity consumption was higher during the intervention phase in the two years studied highlights the need for investigators to examine patterns of energy use prior to implementing interventions for greater likelihood of finding an effect. We discuss issues related to recruitment and provide a set of recommendations for future energy reduction competitions. 3 Chapter 1 Background Energy use among the US residential sector in 2009 accounted for 22% of all energy consumed during that year (US Energy Information Administration, 2010). Similarly, among the 27 European Union nations, energy use among the residential sector accounted for approximately 29% of all electricity consumed in 2004 (Bertoldi & Atanasiu, 2006). Over the past several decades, efforts to reduce household energy consumption and promote other types of pro-environmental behavior have been implemented with varying levels of success (for review see Abrahamse et al. 2005). Pro- environmental behavior (PEB) is defined here as any behavior that supports the sustainability of natural ecosystems, environmental health, and “contribute[s] towards environmental preservation and/or conservation” (Axelrod & Lehman 1993, p. 153). Below we review selected interventions aimed at promoting PEB in residential settings, and then describe the literature specific to college campuses. Two overarching intervention approaches have primarily been used: behavioral antecedent strategies and behavioral consequence strategies. Antecedent interventions aim to influence a given behavior prior to its performance (Abrahamse et al. 2005). The provision of information is a commonly used method in household energy reduction interventions. Information ranges from conceptual, for instance, facts about global warming, to specific, such as recommended behaviors based on audits. Information interventions have been found to be more effective than control conditions in increasing PEB. Based on the results of 8 studies, a meta-analysis by Hines et al. (1987) identified a corrected correlation coefficient of .47 (SD=.29) between information approaches and 4 increased PEB. However, this strategy is generally less effective when compared directly to others (e.g., Seligman & Darley, 1977; Winett et al., 1978), and may be best used as an adjunct to other intervention strategies (Seligman & Darley, 1977). Appeals are another antecedent strategy that involve a request that explicitly or implicitly aims to evoke a given behavior or set of behaviors. Appeals for energy conservation are quite common in the mass media and among university campus energy reduction campaigns, but have received relatively less attention in the empirical literature. Appeals represent a unique approach in that they can be disseminated relatively easily to large numbers of individuals, and even if only small effect sizes are achieved on an individual level, the overall effect across the target population may be of practical significance. One study had begun applying a written appeals and information intervention to residents of a Maryland apartment building during the winter of 1977, three weeks prior to President Carter’s televised energy conservation appeal (Wodarski, 1982). The intervention continued for the following nine weeks. Before the presidential appeal, building gas use increased by 26%, whereas after the presidential appeal, gas use declined by 15.3%. These findings suggest that although the written appeal had limited effectiveness, the presidential appeal may have contributed to observed reductions in gas use. This possibility should be considered with caution given that weather variables were not adjusted for, and other variables associated with the 1970s energy crisis were not assessed. Two additional studies by the same author observed building electricity reductions of 6% and 8% following a combination of appeals, written information, group- level incentives, and group workshops. Due to the combination of intervention 5 components, the independent contribution of appeals in these studies cannot be determined. Overall, these findings highlight the difficulty in evaluating appeals interventions and indicate a need for further investigation of this approach. Another common antecedent intervention strategy is goal-setting, which entails presenting participants with a reference point, for instance to save 5% or 10% electricity relative to use in some prior time period (Abrahamse et al,. 2005). Goal setting is more effective when combined with feedback on goal attainment (Becker, 1978). This may serve to highlight inconsistency between actual behavior and stated goals, thereby evoking cognitive dissonance. For example, in a laboratory study that provided immediate feedback about washing machine electricity use via a display on the machine (McCalley & Midden, 2002), participants who also set a reduction goal saved 20% more water compared to baseline, significantly more than a group that received feedback but did not set a goal (11%). Goals and commitments are more effective when made publicly (Pallak & Cummings, 1976) and when pledges are relatively difficult, though effects have generally been limited in duration (Katzev & Johnson, 1984; Van Houwelingen & Van Raaij, 1989; Winett et al., 1979; Winett et al., 1982). Whereas information and goal setting operate through influencing behavioral antecedents, behavioral consequence strategies are implemented after behaviors are performed, with the goal of impacting future engagement in the same behaviors. Feedback is an empirically supported strategy for reducing home energy use. The Hines et al. (1987) meta-analysis found a corrected correlation coefficient of .28 (SD=0.11) between feedback strategies and increased PEB, supporting this strategy. More 6 specifically, descriptive feedback entails providing individuals with information about their energy consumption after some baseline period. The effectiveness of descriptive feedback is well-documented (Bittle et al., 1979; Kantola et al., 1984; Midden et al., 1983), and tends to increase with increased frequency of provision (McClelland & Cook, 1979-1980; van Houwelingen & Van Raaij 1989). Normative feedback on other’s behavior may influence perceived social norms, and has received mixed support, with some studies finding no differences between treatment groups who receive normative feedback compared to descriptive feedback (Abrahamse et al., 2007; Brandon & Lewis, 1999), and others finding enhanced savings among normative compared to descriptive feedback treatment groups (Midden et al., 1983; Pallak et al., unpublished). These mixed findings may be attributable to the undesirable boomerang effect described by Schultz et al. (2007) and identified in several other intervention studies (e.g., Brandon & Lewis, 1999; Van Houwelingen & Van Raaij, 1989). To prevent this increase in energy use among initially low users of energy following the provision of normative feedback, injunctive feedback that communicates overall approval of a behavior should be administered concurrently with normative feedback (Schultz et al., 2007). The bulk of the literature indicates that rewards also enhance energy savings (e.g., Hayes & Cone, 1977; McClelland & Cook, 1980b; Winett et al., 1978). Significantly greater energy savings have been observed among households that receive rewards versus those that do not (e.g., Winett et al. 1978). However, energy savings associated with rewards have been shown to decline towards the end of interventions or after (McClelland & Cook, 1980b). Also, rewards are typically combined with other 7 intervention components, making it difficult to isolate their effect on behavior. These findings suggest that external rewards should not be relied on to maintain long-term behavior change. Likewise, self-determination theory advocates that interventions promote intrinsic motivation whereby behaviors initially contingent on external stimuli become maintained by intrinsic factors within an individual (e.g., social approval; Deci & Ryan, 1985; Geller, 2002). University-based energy reduction interventions Four studies aimed at reducing energy consumption among college students living in university housing were identified. Over the course of an academic year, McClelland and Cook (1980a) implemented a campus-wide intervention study at a public university in the US targeting residential, office, and mixed-use buildings. Dormitory buildings were randomly assigned to one of two experimental conditions, both of which used feedback and information but distributed the content through different channels. Electricity consumption dropped significantly among all buildings. No differences between experimental groups were observed, which was attributed to student enthusiasm and initiatives that resulted in user participation-based interventions in all buildings. During the academic year following the intervention, average residential housing electricity consumption levels remained at the 15% reduced level that had been achieved during the intervention. In another study at the University of Colorado, six 2-week gas reduction competitions were implemented as part of an intervention among a group of University apartments that housed married students and their children (McClelland & Cook, 1980b). 8 Participants received printed information and were informed that the winning building of each 2-week contest would receive a cash reward to use at the residents’ discretion. Savings over the 12-week intervention period averaged 6.6%, which was a significant reduction compared to baseline use. The largest decrease (10%) occurred during the 1 st 2- week contest. No significant changes in gas use were observed in the subsequent five contests. Post-intervention survey results indicated that knowledge of contest rules was minimal: only 28% of survey respondents (10% of the entire sample) were correct about how many times their building won a prize, which highlights the need to track individual- level intervention exposure in group-based programs. Petersen et al. (2007) evaluated the effects of a competition-based intervention implemented among undergraduate dormitory residents at Oberlin University in Ohio. The intervention was framed as a dormitory-versus-dormitory energy reduction contest, and although no tangible rewards were mentioned, implicit rewards associated with winning the contest may have influenced outcomes. Dormitories were randomized to receive either continuous feedback that was displayed in building lobbies and could be accessed via website, or limited web-based feedback that was administered once in the middle of the intervention and again at its conclusion. On-campus advertising provided information about the importance of energy conservation to all participants, but did not provide specific reduction strategies. During the 2-week intervention, the continuous feedback group reduced electricity use by 55% compared to 31% among the other group. Website data indicated that 46% of computers registered to on-campus residents accessed the study website at least once during the competition, and that residents receiving 9 continuous feedback accessed it more than did those in the other group. Follow-up surveys indicated that respondents employed a variety of low-cost and no-cost strategies to conserve during the competition and that they planned to continue using these strategies. These findings lend support to feedback, and the relative advantage of continuous feedback, within a competition framework for reducing energy consumption among college dormitory residents. Limitations to this study include that the competition was of limited duration (i.e., two weeks), analyses of electricity data did not adjust for weather effects, and the only follow-up data collected were intentions to continue conserving energy rather than actual electricity use data. An intervention aimed at reducing student energy use and promoting related knowledge and attitudes was implemented at Tufts University (Marcell et al., 2004). One dormitory that housed mostly upper-classmen was assigned to each condition: (1) a community-based social marketing intervention; (2) a more common education-based program. Both conditions began with an in-person educational session on climate change along with brochures encouraging energy conservation. For the following eight weeks, participants in the community-based social marketing condition also received small incentives (e.g., candy), prompts (e.g., computer monitor stickers), and regular emails, which solicited energy reduction commitments and provided tailored information on energy reduction strategies. This project did not include a competition. A total of nine residents participated in the education-based program, and 15 participated in the community-based social marketing program. No significant change was observed in electricity use for either building. Pre-test and post-test surveys assessed energy use 10 behaviors and attitudes and knowledge related to climate change, but the low participation rate precluded statistical testing. As was noted by the authors of the study, upperclassmen may be less amenable to change, and in fact dormitories that housed upperclassmen exhibited lower levels of change in energy use (i.e., 2%) in the Petersen et al (2007) study relative to buildings that housed lowerclassmen. Study Aims Intervention programs combining several intervention strategies have not systematically examined the unique contributions of each. Understanding the independent contributions of each strategy as well as how they work together is important for informing future interventions. As well, many university-based energy reduction efforts have not been evaluated. The primary aim of the present study was to evaluate the effectiveness of a competition-based intervention in reducing electricity use among residents of undergraduate residence halls at the University of Southern California (USC) in fall 2009. A secondary objective of the project was to assess changes in self-reported PEB across the study time period. A final goal was to investigate the independent effects of several intervention components on changes in building energy use and self-reported PEB. Chapter 1 Method Using a prospective quasi-experimental design, the current study randomly assigned groups of participants to different levels of exposure to intervention content based on their dormitory of residence, including an assessment-only control group not 11 exposed to any part of the intervention. Participants completed baseline and follow-up self-report surveys regarding energy use behaviors and key constructs derived from the Theory of Planned Behavior (Ajzen & Madden, 1986) and the Norm Activation Model (Schwartz, 1994). See Figure 1.1 for the study timeline. Procedures Survey 1. Beginning in September 2009, survey 1 was publicly accessible online for approximately eight weeks. It required approximately 25 minutes to complete. The survey assessed inclusion criteria and ended automatically if participants returned responses that indicated ineligibility (see Participants section below for eligibility criteria). Recruitment continued through the end of October 2009. Emails, poster advertising inside of target buildings and on the USC campus, the Department of Psychology research participant pool, and contact with target residence hall staff were used as recruitment strategies. Experimenters also set up tables to distribute information in a high-traffic area on campus. All participants were entered into raffle drawings to receive a variety of prizes, ranging in value from $10-$300. Participants who were enrolled in the Psychology Department research participant pool also received credits for taking the surveys that could be applied to courses. Intervention. Seven target buildings were selected for inclusion in the intervention, on the basis of being similar in the types of appliances that were under the control of residents (e.g., air conditioning units) and in other characteristics such as layout and population (i.e., primarily underclassmen). The intervention began in late September 2009 and lasted for eight weeks. Intervention stimuli common to all target 12 buildings included appeals to reduce energy use and a group-level incentive. In addition, to increase engagement, the intervention was advertised as a dormitory-versus-dormitory competition called the “Ecolympics Energy Reduction Challenge”. Buildings earned points based on building-level survey participation rates and reductions in building-level electricity use during the fall 2009 semester. The building that reduced its electricity use by the largest percentage won a pizza party (group-level incentive). Appeals were made via emails and posters, both of which contained a URL to the study website and information about the competition and incentive. Posters also contained graphics and messages that encouraged energy conservation (e.g., “Which is the greenest dorm on campus?”). New posters were placed in target buildings weekly. All target buildings were treated equally with respect to appeals, recruitment, and incentives. Likewise, appeals and recruitment efforts were absent among assessment-only control buildings. In addition to the shared intervention stimuli, supplementary intervention content was provided via a study website. The target buildings were randomized to different combinations of website intervention components. As part of the registration process required to access the website, users reported their building of residence. Based on this information, the website granted access to the assigned condition. Residents of the partial control building (Building C) served as a partial control group; although they could not access any supplementary content on the website, they were still included in the competition and exposed to the same incentive and appeals as residents of other target buildings. Residents of a second building were granted access to informational modules only (Building E). Three buildings shared one electricity meter and were treated as a 13 single group (Building D); residents of these buildings were assigned access to informational modules and real-time electricity feedback of all participating buildings. Residents of the sixth building had access to informational modules and were asked to set an individual and building-level electricity reduction goal upon registration (Building A). Residents of the seventh building could access the information, feedback, and goal- setting modules (Building B). See Materials section below for additional information on supplementary intervention content. Based on existing literature, this content was expected to have substantial effects on behavior. Intervention recruitment methods were similar to those used for Survey 1. Beginning in late September 2009, participants in the first survey who indicated that they resided in a target building were emailed to encourage their participation in the competition. Emails were resent on several occasions. Participants who completed either survey but were not residents of one of the seven target buildings constituted the assessment-only control group. These participants were not included in the competition, had no access to the supplementary website modules, and were not informed by study staff of the intervention. They completed study measures for course credit or raffle prizes. Survey 2. Survey 2 was available 24 hours per day for three weeks beginning the day after the intervention ended. It was administered online and required approximately 25 minutes to complete. Recruitment methods were the same as for survey 1, with particular focus on emailing Survey 1 participants for follow-up. 14 Focus Groups. In March 2010, focus groups were conducted to obtain qualitative data on participant perceptions of the project. All survey participants who reported residing in target buildings were emailed and invited to participate. Two focus groups lasting approximately 90 minutes each were held among the subset of participants who responded to this solicitation. Students who were not part of the subject pool received $20 for participating, and students in the subject pool received credits that could be applied towards their courses. Each focus group was co-led by the Principal Investigator and an undergraduate research assistant. Figure 1.1. Study timeline. Study Groups August September October November December Intervention: Sep 30 – Nov 25 All targets: Competition, appeals, incentive Residents of the seven target buildings Partial Control (Building C) Electricity Baseline Period Aug 17 - Sep 29 No website modules available Intervention End: Nov 25 Info (Building A) Info module available Info, Goal Setting (Building B) Set reduction goal during initial log-in, Survey 2: Nov 26- Dec 17 then info & goal modules available Info, Feedback (Building D) Info & feedback modules available Info, Goal Setting, Feedback (Building E) Set reduction goal during initial log-in, then info & goal modules available Survey 1: Sep 9 -Nov 1 Assessment Only Control a Survey 1: Sep 9 -Nov 1 Survey 2: Nov 26- Dec 17 a Not included in intervention 15 Participants Surveys. To be eligible to participate in the surveys, participants had to be at least 18 years of age, a current USC undergraduate student, and a resident of USC-owned or USC-managed housing. To be eligible to participate in the intervention, participants also had to be residents in one of the seven selected target buildings. A total of 302 students participated in the first survey and 225 followed up in the second, yielding a follow-up rate of 76%. Methods by which participants were recruited to participate in each survey are shown in table 1.1. Table 1.1. Frequencies (percentages) of survey respondents identified through various recruitment strategies. Friend in dorm RA BG Friend outside dorm Study team Email FB Saw poster Alt Res Subject pool Total N Survey 1 7 (2.4) 54 (18.8) 11 (3.8) 4 (1.4) 27 (9.4) 62 (21.6) 11 (3.8) 34 (11.9) 1 (0.35) 137 (47.7) 287 (100) Survey 2 4 (1.9) 28 (13.0) 12 (5.6) 6 (2.8) 12 (5.6) 36 (16.7) 11 (5.1) 24 (11.2) 0 (0) 128 (59.5) 215 (100) Table note. Data from all survey participants with data on recruitment source are reported here to demonstrate overall recruitment efforts. Participants could report more than one recruitment source. RA= resident advisor. BG=Building Government. FB=www.Facebook.com. Alt Res=Alternativeresearch.org. Subject Pool = USC Psychology Subject Pool. Among survey participants, 11 individuals provided responses on either survey that suggested random or careless responding (see Validity Items below). A total of 6 respondents completed only time 1 demographic questions, and 7 completed time 1 demographics and half of one additional scale. Responses from these participants were excluded. Among the remaining 278 survey 1 respondents, 216 participated in survey 2, and 198 completed measures on the key outcomes examined in this study. Missing scale 16 items were imputed for individuals who were missing less than 10% of items on a given scale. Each participant who was missing more than 10% of items on a given scale was given a missing value for that scale (for each scale, 2-9 of participants were missing more than 10% but less than 100% of items; across the sample, 14 participants were missing between 10 and 100% of items on any scale). With these imputed data, 187 participants had complete scale-level data at both time points and were retained in subsequent analyses. Approximately 75% of the sample was female, and just under half of participants were university freshmen. Parental education level was high in this sample. For 68% and 70% of participants, maternal and paternal education levels, respectively, were 4-year college degrees or higher. See table 1.2 for descriptive characteristics of participants. Intervention. Of the 187 survey participants retained for analyses, 80 reported residing in one of the seven target buildings and were treated as intervention participants. The 107 survey participants who reported residence in other buildings (i.e., not selected for inclusion in the competition) were treated as a control group. To assess the overall effectiveness of the competition, incentive, and appeals on promoting PEB, the intervention and control groups were compared on building-level electricity use and self- reported PEB. Only six individuals registered to access the supplementary website modules, which was the only way such content could be accessed. We were therefore unable to conduct planned analyses on the relative effectiveness of different web intervention modules. Data regarding those who may have participated informally in the competition (i.e., taking actions to save energy) but did not participate in the surveys nor 17 supplementary intervention website modules could not be gathered on an individual-level basis due to the nature of such participation. This is common to group-level interventions (e.g., Marcell et al., 2004; McClelland & Cook, 1980a; McClelland & Cook, 1980b; McMakin et al., 2002). Buildings. Electricity data were gathered for the seven target buildings included in the intervention (i.e., Buildings A-E). Three of these buildings shared an electricity meter and were treated as a single group (i.e., Building D). Electricity data were also gathered for three non-target on-campus dormitories (i.e., Buildings F-H) that were not included in the competition. Electricity data from these buildings were used clarify the effects of the intervention on changes in building-level electricity use. Heat among all on-campus dormitory buildings was under resident control. However, heat was provided via a central boiler system and therefore not reflected in overall building electricity use among any of the buildings. Cooling among intervention buildings was provided via a central chiller system and therefore not reflected in electricity use among intervention buildings. However, apartments in non-target Buildings F and G had occupant-controlled air conditioning units. Use of these units was included in electricity consumption for those buildings. 18 Table 1.2. Participant demographic characteristics by intervention status. Characteristic All (n=187) Intervention (n=80) Control (n=107) Mean age in years (SD) 19.1 (1.3) 18.6 (1.1) 19.5 (1.2) Sex (% female) 75.7 72.5 78.1 Number of semesters in residence hall 1 2 3 4 or more (%) 74.2 2.7 16.1 7.0 81.0 2.5 10.1 6.3 69.2 2.8 20.6 7.4 Class status Freshman Sophomore Junior Senior (%) 45.5 32.1 15.5 6.9 72.5 10.0 11.3 6.3 25.2 48.6 18.7 7.4 Urbanicity of place of origin Rural Suburban Urban (%) 4.3 79.1 16.6 3.8 81.3 15.0 4.7 77.6 17.8 Focus Groups. A total of seven students participated in one of two follow-up focus groups. The mean age of these participants was 19.0 years, and six were female. A total of six were freshmen, and one was a sophomore. The participants represented three residence halls: Buildings A, B, and E. The freshmen had been residing in their residence halls for two semesters, and the sophomore for four. Approximately 30 students expressed interest in participating in a focus group, but additional groups were not held due to budget constraints and conflicting participant schedules. Materials Surveys. The two surveys were identical except that the second included a set of follow-up questions to obtain feedback on the intervention. In addition to the variables 19 below, the surveys assessed a number of other constructs from the TPB and NAM. For brevity, the measurement and results related to these items will not be discussed here. Demographic data. Demographic variables included age, gender, year of university enrollment, parental educational attainment, and urbanicity of region of origin (urban, suburban, or rural). Behavior. Behavior was assessed using three self-report scales. Participant characteristics for these scales are displayed in table 1.3. Frequency of engagement in a variety of PEB for the two weeks preceding the survey was assessed using a 17-item modified version of the Schultz Proenvironmental Behavior Scale (Schultz et al., 2005). Responses were provided on a 5-point Likert scale with response options ranging from “never” to “very often”. Items were scored 1 to 5, with a possible scale score range of 17 to 85. Cronbach’s alpha was calculated to assess internal consistency and was .70 for survey 1 and .75 for survey 2. Because the focus of the appeals was to reduce electricity use, a total of four items from the Schultz PEB scale that pertained to electricity use (i.e., turning off lights, thermostat use, water temperature settings for laundry, turning off computer when not in use) were examined separately as the Electricity Use Subscale. Internal consistency for the Electricity Use Subscale at time 1 was .24 and at time 2 was .34. The Consumer Behavior Subscale from Stern’s Indices of Pro-Environmental Behavior was used to assess consumer behavior (e.g., purchasing organic produce; Stern et al. 1999). Responses to the four items were provided on a 5-point Likert scale. Items 20 were scored 1 to 5, with a possible scale score range of 4-20. Internal consistency alphas for time 1 and 2 were .71 and .65, respectively. Group Identification. Group identification was assessed using the 8-item Arrow- Carini Group Identification Scale, Version 2 (Henry et al., 1999). Respondents provided answers to questions about affiliation with other residents of their dormitory building on a 7-point Likert scale ranging from strongly disagree to strongly agree. Items were scored 1 to 7, with a possible scale score range of 8-56. Internal consistency was .95 for both time 1 and time 2. Validity items. One true/false question was included in each survey to detect random or careless responding. It stated “George W. Bush is the current president of the United States”. As the survey were administered in 2009, the correct answer was false. The Impression Management scale from the Balanced Inventory of Desirable Responding Impression Management Scale (BIDR-IM) was included in both surveys to assess response bias associated with impression management (Paulhus, 2002). The scale included 12 items that assessed engagement in various behaviors (e.g., “When I was young, I sometimes stole things”; “I have done things that I don’t tell other people about”). Answers indicated the extent to which respondents believed the statements to be true for them, and were provided on a 7-point Likert scale ranging from “definitely false” to “definitely true”. Items were scored 1 to 7, with a possible scale score range of 12-84. Internal consistency was .74 for time 1 and .75 for time 2. A set of follow-up validity analyses investigated the influence of BIDR-IM scores on results. 21 Survey 2 follow-up items. Survey 2 was identical to Survey 1 except for the addition of a set of follow-up questions designed to assess student perceptions of the project. The follow-up items were presented only to individuals who indicated they were residents of target buildings. Items addressed whether participant was aware of the competition, with whom participants discussed the competition, and to what extent they believed these discussions influenced their behavior during the competition. Participants were also asked if they planned to sustain any behavior changes made during the competition. Table 1.3. Participant characteristics on key variables by intervention status. Scale All (n=187) Intervention (n=80) Control (n=107) Mean (SD) PEB Time 1 Time 2 Change (T2-T1) a 48.6 (8.1) 50.5 (8.5) 1.8 (5.7) 49.9 (8.2) 52.0 (8.6) 2.1 (5.8) 47.7 (7.9) 49.4 (8.4) 1.7 (5.8) Electricity Use Subscale Time 1 Time 2 Change (T2-T1) a 12.8 (2.8) 13.7 (2.8) 0.8 (2.3) 13.3 (2.8) 14.3 (2.8) 0.9 (2.4) 12.5 (2.8) 13.2 (2.7) 0.8 (2.1) Consumer Behavior Time 1 Time 2 Change (T2-T1) a 11.6 (2.6) 11.3 (2.5) -0.2 (2.1) 11.7 (2.4) 11.2 (2.3) -0.5 (2.1) 11.5 (2.8) 11.4 (2.7) -0.1 (2.1) a Time 2 – time 1 mean difference. Intervention. Intervention materials included appeals and an intervention website. Appeals. Appeals to reduce energy use were made via poster and email advertising. Example posters are provided in Appendix 1A. 22 Website. To administer the supplementary intervention content, an intervention website with several content modules was developed by website development firm Phoenix Energy Technologies. This website included: an informational module on environmental problems and strategies to reduce energy consumption; a goal-setting component; and a module that provided real-time electricity feedback for included buildings (www.EcolympicsUSC.com). If assigned to receive the goal-setting module, participants were asked to set both individual-level and building-level goals. The feedback module was an interactive graph that displayed real-time electricity consumption for all target buildings, and provided for time-frame and building-vs- building comparisons. Building data. Data on electricity consumption, building occupancy, and weather were gathered to assess the impact of the intervention on building-level electricity use. Electricity use. Data on daily electricity consumption, measured in kilowatt-hours (kWh), were obtained for each on-campus dormitory building for the date range 8/1/09- 1/31/10. Daily consumption data for all on-campus dormitories were also obtained for the fall 2008 time period (date range 8/1/08-1/31/09). All electricity data were prepared by USC Facilities Management and provided to the study team. Anomalous events that influenced building electricity use were discussed with USC Facilities Management Director of Energy Services. Three electricity use observations were deemed invalid on the basis of being more than 50% above the mean compared to the adjacent two weeks. Means for the adjacent two weeks were substituted in for these observations. 23 Building occupancy. University residential staff provided data on the number of residents in each building to study staff for target buildings for fall 2009 (see table 1.4). Occupancies for other buildings and time periods could not be obtained. Weather variables. Sky cover and daily mean, maximum, and minimum ambient temperatures were collected for the USC/downtown Los Angeles location for the time periods 8/1/08-1/31/09 and 8/1/09-1/31/10. Data were gathered from National Oceanic and Atmospheric Administration (NOAA) public datasets (NOAA, 2010). Focus group questions. Focus group questions elicited information on participant perceptions of recruitment strategies, advertising, and incentives for to the intervention and surveys. We also inquired about group cohesion among dormitory residents. See Appendix 1B for the list of focus group questions. Chapter 1 Results The SAS statistical package version 9.1 was used to analyze all electricity and self-report data (SAS Institute Inc., 2002). Electricity Consumption Descriptive analyses. Table 1.4 displays descriptive details on 2008 and 2009 electricity use among all on-campus dormitories. Mean daily electricity consumption during the baseline and intervention phases was compared to assess change in consumption. The baseline period was 8/17/09-9/29/09. Although we had electricity data beginning 8/1/09, dormitory residents began moving in on 8/17/09, so we elected to begin the start of the baseline period on that date. The intervention period was 9/30/09- 11/25/09. As has been done in prior studies (e.g., Becker, 1978; Bittle et al., 1979; 24 Petersen et al., 2007), descriptive changes in electricity use across study phases were calculated by subtracting mean daily use during the intervention phase from mean daily use during the baseline phase, and dividing the resulting figure by baseline mean daily use. Overall electricity use among the target buildings combined (Buildings A-E) declined negligibly (i.e., 0.8%) during the 8-week intervention compared to baseline use. Among individual buildings, changes in use ranged from an increase of 3.1% to a decrease of 7.1%. See Figure 1.2 for a visual representation. Figure 1.2. Daily mean electricity consumption among target buildings during the fall 2009 intervention and baseline periods. 25 Daily electricity consumption data was obtained for all on-campus dormitories for fall 2008, as well as non-target buildings in 2009, in order to examine electricity use patterns across time and different populations of dormitory residents. Figures 1.3 and 1.4 show mean daily electricity use averaged by week for on-campus dormitories during fall 2008 and fall 2009, respectively. The same calculations used for describing changes in electricity use among target buildings in 2009 were also used for 2008 data. The same calendar dates included in the baseline and intervention phases in 2009 were also used for 2008 calculations even though no intervention was conducted in 2008 (i.e., baseline = 8/17/08-9/29/09, intervention = 9/30/08-11/25/08). In 2008, the target buildings combined reduced their electricity use by 24.1% during the intervention relative to baseline period. However, discussions with USC Facilities Management revealed that consumption data for one of the buildings, Building E, included the electricity use associated with a chiller system used to cool water for part of the study period (8/28/2008-10/3/2008), which was not included for any other time periods or any other buildings. Because this obscured interpretations of the data, Building E data were excluded from all subsequent electricity analyses. The remaining target dormitories increased their overall use by 3.7% in 2008. Excluding Building E from 2009 calculations, the remaining target dormitories decreased their use by 2.7%. Non-target buildings (i.e., Buildings F-H) reduced 2008 use by 1.6%, and 2009 use by 18.4%. 26 Table 1.4. Mean daily electricity consumption (SD) in kWh among USC dormitory buildings in fall 2008 and 2009. 2008 2009 Baseline Phase Intervention Phase Percent change a Baseline Phase Intervention Phase Percent change a Building A Occupancy 3255.8 (277.4) n/a 3428.5 (87.9) n/a +5.3% 3094.4 (228.8) 450 3191.1 (80.9) 450 +3.1% Building B Occupancy 749.5 (164.4) n/a 790.1 (214.2) n/a +5.3% 1200.9 (175.8) 295 1124.7 (76.9) 297 -6.3% Building C Occupancy 1489.0 (225.0) n/a 1678.3 (88.1) n/a +12.7% 1272.7 (1074.3) 220 1236.0 (1449.2) 250 -2.8% Building D Occupancy 3007.1 (309.7) n/a 2918.9 (213.3) n/a -2.9% 3071.2 (378.4) 494 2854.0 (187.1) 545 -7.1% Building E Occupancy 11739.4 (4425.9) n/a 6548.9 (1853.2) n/a -44.2% 5830.6 (581.4) 400 5954.9 (150.5) 401 +2.1% Building F Occupancy 6078.5 (934.8) n/a 5571.4 (1000.5) n/a -8.4% 6872.1 (1650.2) n/a 4949.1 (623.9) n/a -28.0% Building G Occupancy 2670.7 (564.0) n/a 2705.2 (332.3) n/a +1.3% 2929.0 (789.2) n/a 2353.3 (190.9) n/a -19.7% Building H Occupancy 6728.2 (1002.2) n/a 6953.1 (421.6) n/a +3.3% 6854.4 (846.0) n/a 6284.5 (304.2) n/a -8.3% Targets combined 20240.8 15364.7 -24.1% 14469.8 14360.7 -0.8% Targets combined excluding Building E 8501.4 8815.8 +3.7% 8639.2 8405.8 -2.7% Nontargets combined 15477.4 15229.7 -1.6% 16655.5 13586.9 -18.4% All buildings 35718.2 30594.4 -14.3% 31125.3 27947.6 -10.2% Average daily temp 74.2 (2.5) 70.9 (5.9) 76.7 (4.9) 67.1 (4.9) Mean maximum daily temp 83.1 (3.8) 82.8 (9.1) 86.6 (7.7) 76.6 (7.1) a Change observed between baseline and intervention phases = (baseline use-intervention use)/baseline use. Table note. For both years, baseline phase = August 17-September 29, intervention phase = September 30-November 25. To compare consumption across years, the 2008 dates were selected to correspond to the 2009 study phases. No intervention was conducted in 2008. 27 Figure 1.3. 2008 daily electricity use averaged by week for on-campus buildings excluding Building E. Figure note. Intervention buildings=Buildings A-D. Control buildings=Buildings F-H. Figure 1.4. 2009 daily electricity use averaged by week for on-campus buildings excluding Building E. Figure note. Intervention buildings=Buildings A-D. Control buildings=Buildings F-H. Electricity use in KwH Electricity use in KwH 28 Regression analyses. In addition to examining patterns in electricity use descriptively, changes in electricity use were evaluated using multi-level regression models. First, a set of preliminary univariable regression models assessed the influence of each weather variable on aggregate electricity use among all on-campus buildings (excluding Building E) in 2008 (8/17/08-11/25/08) and 2009 (8/17/08-11/25/09). Weather variables included maximum daily temperature, minimum daily temperature, average daily temperature, and average daily sky cover. Aggregate daily electricity consumption was regressed separately on each weather variable. Average daily sky cover accounted for less than 3% of explained variance in 2009 and less than 1% in 2008 and therefore was not retained in subsequent analyses. Maximum daily temperature accounted for the largest percentage of explained variance in consumption relative to average and minimum daily temperatures. In addition, on 85% of the days studied, average temperatures were at least 65° F, a standard building temperature set-point (NOAA, 2010). Therefore, electricity use associated with cooling was relatively more important for the time period studied. Along with the fact that maximum daily temperature accounted for more variance in electricity use outcomes in the preliminary regression models, we elected to retain maximum daily temperature for use in the multi- level models. When the multi-level models described below were repeated including minimum temperature to adjust for electricity use associated with heating (e.g., space heaters), the same pattern of results was yielded. These results are therefore not reported. To account for repeated measurements in time among each building, multi-level regression modeling was used to fit mixed effects models examining the effects of 29 maximum temperature, year, phase, building, and target building status on electricity consumption. Again, Building E data were excluded from analyses, so 1404 daily electricity use values from the remaining seven on-campus dormitory meters were included in these analyses. The SAS PROC MIXED procedure was used to test the models. Restricted Maximum Likelihood estimation was used to estimate model parameters. Building was entered as a random effect. Fixed effects were maximum daily temperature, year (2008 = 0, 2009 = 1), phase (intervention phase = 1, baseline phase = 0), and target building status (control = 0, target = 1). A variable was created to code for buildings in which residential spaces had personal air conditioning units (1 = yes; 0 = no). However, initial analyses indicated that this variable contributed less than 1% to the explained variance in electricity use, so it was not retained in subsequent analyses. A stepwise model building procedure was used to model mean daily electricity use based on the set of predictors. On the first step, building was entered as a random effect and target building status was entered as a fixed effect. On the second step, maximum daily temperature was entered as a fixed effect. On the third step, year and phase were added as fixed effects. On the fourth step, interactions among fixed effects were added. Several models with different interactions were evaluated. Results are displayed in table 1.5. The intraclass correlation for the building effect was quite high (0.92), indicating that within- building electricity use observations shared strong resemblance. A significant between- subjects main effect of target status was observed such that overall, target buildings used significantly less electricity than control buildings. This effect accounted for 43.45% of the between building variance in electricity use. Model IV explained the most within- 30 building variance in electricity use. Higher electricity use was predicted by higher maximum temperatures and baseline phase. A significant phase by target status interaction indicated that target buildings used significantly more electricity during intervention relative to baseline phases. In Model VI, a significant three-way interaction between phase, target status, and year indicated that target buildings used relatively more electricity during the 2009 intervention phase than other time periods. 31 Table 1.5. Fixed effect parameter estimates and standard errors from multilevel regression models (N=1414). Model I Model II Model III Model IV Model IV Reduced Model V Model VI Model VI Reduced % Additional Variance Explained a Between Within 43.45 0.0 43.45 11.88 43.45 12.50 43.45 17.43 -0.03 12.65 43.45 13.54 43.45 14.59 -0.02 12.39 Change in Log Likelihood b -21 -182 -232 -325 -205 -260 -277 -201 Variable Parameter estimate (standard error) p value Maximum Daily Temperature 26.33 (1.91) p<.0001 24.00 (2.02) p<.0001 24.00 (1.95) p<.0001 28.02 (1.95) p<.0001 21.20 (2.11) p<.0001 26.02 (2.02) p<.0001 28.19 (2.00) p<.0001 Year c -37.58 (31.12) p<.23 -37.58 (30.23) p<.22 112.72 (47.10) p<.02 -146.07 (35.75) p<.0001 Phase d -107.24 (32.83) p<.01 -428.93 (47.29) p<.0001 16.23 (43.77) P<.72 -196.36 (35.73) p<.0001 Target Status e -2895.33 (1222.6) p<.02 -2895.33 (1222.6) p<.02 -2895.33 (1222.6) p<.02 -3213.05 (1223.07) p<.009 -2895.33 (1222.58) p<.02 -2993.66 (1222.70) p<.02 Year x Phase -275.85 (65.20) p<.0001 Phase x Target status e 562.97 (61.12) p<.0001 153.71 (42.36) p<.001 Year x Phase x Target status e 348.47 (58.62) p<.0001 142.95 (47.76) P<.01 Table note. Building E was excluded from these analyses due to anomalies in 2008 consumption. a Compared to baseline model with no predictors. Within= within-building variance component. Between=between-building variance component. b Compared to baseline model where -2 Log Likelihood = 22237. c 0=2008, 1=2009. d 0=baseline, 1=intervention. e 0=control, 1= target. Self-Reported Behavior To evaluate the overall effectiveness of the competition, appeals, and incentive on promoting behavior change, survey outcomes were compared among intervention and 32 assessment-only control participants. Because only six participants accessed the intervention content on the study website, we were unable to conduct planned analyses to evaluate the effects of the different website module components on PEB. Three behavior outcomes were assessed to examine whether changes from time 1 to time 2 applied only to a specific domain of behavior or were more generalized. The three outcomes were: total PEB score, electricity use subscale score, and consumer behavior score. All outcomes were coded as continuous. Baseline group equivalence. We tested for baseline differences in demographic characteristics and scores on the three outcome scales between intervention (n=80) versus control participants (n=107). Two multiple logistic regression models were conducted with different conceptual blocks of variables predicting intervention status. In the first block, demographic factors including sex, urbanicity of place of origin (coded 1 for urban and 0 for suburban or rural), maternal education level, and paternal education level were entered. Variables related to age and residence status were used in the second model, including age, number of semesters in residence at residence hall, and year of university enrollment. Younger individuals (Odds ratio [OR] = 0.43, 95% confidence interval [CI] = 0.25-0.75), and those who had lived in their current residence halls longer (OR = 1.44, CI = 1.02-2.02) were significantly more likely to reside in target buildings. Age, length of residence, and year of enrollment were highly intercorrelated (residence length/enrollment year, r = -.62, p<.0001; residence length/age, r = .47, p<.0001; age/enrollment year r = -.64, p<.0001). Year of enrollment coded for the conceptually most relevant effect of length of time as a student in the university environment and 33 significantly predicted intervention status in a univariable model (OR = 2.05, CI = 1.42, 2.96). Enrollment year was therefore retained as a covariate in subsequent analyses. In addition, t-tests were conducted to evaluate baseline differences in the three outcome scales among intervention participants vs. controls. Only baseline electricity use scores differed significantly across groups. Time 1 electricity use score was a mean of 0.88 points higher among intervention relative to control participants, approximately one third of a standard deviation (t = 2.12, p<.04). Time 1 PEB was a mean of 2.2 points higher among intervention relative to controls, approximately one fourth of a standard deviation (t = 1.87, p<.07). Time 1 consumer behavior score was a mean of 0.21 points higher among intervention relative to control participants, less than one tenth of a standard deviation (t = 0.54, p<.60). Attrition bias. Attrition bias was investigated using the same procedures as were used to assess baseline differences among intervention relative to control participants. Two multiple logistic regression models were conducted with conceptual blocks of variables predicting follow-up status (coded yes = 1, no = 0, to indicate whether or not participant had complete data for the present study at time 2). The demographic block included sex, urbanicity of place of origin, maternal education level, and paternal education level. The second block included age, number of semesters in residence at residence hall, location of residence (on or off campus), and year of university enrollment. Those who enrolled in the university relatively later (OR = 0.54, CI = 0.35- 0.86, p<.01), lived on campus (OR = 0.60, CI = 0.39 – 0.92), and reported urban place of 34 origin (OR = 0.47, 0.26-0.86) were significantly less likely to participate in the follow-up survey. T-tests were conducted to evaluate baseline differences in the three outcome scales among those who had complete data at time 2 relative to those who did not. Time 1 PEB score was a mean of 2.5 points higher among those who did not have complete time 2 data, approximately one fourth of a standard deviation (t = 2.30, p<.03). Similarly, time 1 electricity use score was a mean of 0.8 points higher among those who did not follow up, approximately one fourth of a standard deviation (t = 2.22, p<.03). Time 1 consumer behavior score was a mean of 0.4 points higher among those who did not have complete data at time 2, approximately one seventh of a standard deviation, but this difference was not significant (t=1.02, p<.31). See Appendix 1D for additional descriptive characteristics among individuals who participated in survey 2 relative to those who did not. Main analyses. For each outcome, the same procedures were followed. We first used paired t-tests to evaluate the significance of differences between time 1 and time 2 scores. Next, using three multiple regression models, one for each outcome, we modeled survey 2 behavior scores based on a set of predictors. A stepwise model building procedure was used. On the first step, the survey 1 score corresponding to the outcome was entered as a predictor. On the second step, enrollment year was added as a covariate. On the third step, intervention status (1=intervention; 0= control) was added. On the fourth and final step, an interaction term (survey 1 score x intervention status) was added. For all three outcomes, table 1.6 shows the change in R² values associated with adding 35 additional predictors to each model, and table 1.7 displays standardized model parameters from final models. General PEB. The t-test was significant (t 186 = 4.44, p<.0001) and indicated a mean increase of 1.8 from baseline to follow-up, approximately one fifth of a standard deviation. The final regression model was also significant (F = 63.43, p<.0001) and accounted for 58.23% of the variance in time 2 PEB. Time 1 PEB significantly predicted time 2 PEB (t = 11.49, p<.0001). After accounting for time 1 PEB score, none of the other predictors were significant in the final model. Electricity use subscale. The t-test was significant (t 186 =5.09, p<.0001) and indicated a mean increase of 0.8 points from baseline to follow-up, approximately one quarter of a standard deviation. The final regression model was significant (F=39.24, p<.0001), and accounted for 46.3% of the variance in time 2 electricity use. Time 1 electricity use score significantly predicted time 2 electricity use score (t=9.20, p<.0001); none of the other predictors were significant in the final model. On step 3, intervention participants tended to report higher scores at time 2 relative to controls, but this contributed less than 1% to the explained variance in the outcome after accounting for enrollment year and time one electricity use score (b=0.09, p<.13). Consumer behavior. The t-test was not significant (t 197 =-1.53, p<.13), indicating no significant change in consumer behavior across time. The final regression model was significant (F=33.11, p<.0001) and accounted for 40.7% of the variance in time 2 consumer behavior. Higher time 1 consumer behavior scores predicted higher time 2 scores (t=9.89, p<.0001). None of the other predictors were significant in the final model. 36 Impression management bias. To investigate the validity of the above results, the multiple regression analyses for each outcome were repeated to adjust for impression management scores. All final models were run using continuous time 2 impression management score as a covariate among the subset of 186 participants who had data on this variable. In addition, final models were repeated excluding the 46 participants (11 males and 35 females) whose bias score surpassed the sex-specific cutoff at either time. Consistent with guidelines from the scale developer, bias score was defined as the number of items with a score of 6 or 7 (Paulhus, 2002). Sex-specific cutoffs were defined as mean sex-specific bias score plus one standard deviation. Bias scores among the present sample were similar to those identified in the sample upon which the norms for the instrument were based (Paulhus, 2002). Results from both sets of analyses were consistent with findings from analyses that did not adjust for impression management. This suggests that the findings were not strongly influenced by an impression management bias, so results that did not adjust for impression management are reported. 37 Table 1.6. Sequential R 2 values associated with adding additional predictors to each model for general Pro- environmental Behavior (PEB), electricity use subscale score, and consumer behavior outcomes (n=187). PEB Electricity use Consumer behavior Model I Time 1 score 57.8 45.6 44.5 Model II Year of enrollment 0.3 0.0 0.1 Model III Intervention status (yes=1) 0.1 0.7 0.4 Model IV Intervention status x Time 1 score interaction 0.0 0.0 0.3 Total 58.2 46.3 45.3 Table note. To calculate sequential R², variables were added in a stepwise manner using the order listed. Table 1.7. Standardized regression coefficients for final models (n=187). PEB Electricity use Consumer behavior Time 1 score 0.75 P<.0001 0.67 P<.0001 0.71 P<.0001 Year of enrollment 0.05 P<.37 -0.01 P<.92 -0.01 P<.94 Intervention status (yes=1) 0.03 P<.93 0.15 P<.57 0.20 P<.46 Intervention status x Time 1 score interaction 0.01 P<.97 -0.07 P<.81 -0.27 P<.31 Mixed-effects models. To account for the effect of correlated observations that could potentially result from participants living in the same buildings (i.e., within- building group cohesion or structural building effects), multi-level mixed modeling analyses were conducted. SAS PROC MIXED was used to account for individuals nested within residence hall. Restricted Maximum Likelihood estimation procedures were used 38 to estimate the model parameters. For each outcome, residence hall was entered as a random effect; intervention status, enrollment year, time 1 scores, and an interaction between intervention status and time 1 scores were entered as fixed effects. Baseline models run with no fixed effects showed that residence hall accounted for less than .002% of the variance in each outcome; the same results were observed when adding the fixed effects to the models. SAS PROC MIXED uses an additional degree of freedom for each additional cluster (i.e., residence hall) estimated. In study, survey participants resided in 27 residence halls, but only six of these had 10 or more survey respondents. To maximize the possibility of detecting potential effects resulting from correlated observations, another set of analyses were conducted using data from participants residing in the six buildings, which included 116 individuals. Findings were similar to those using the larger sample: residence hall accounted for less than .02% of the variance in any of the three outcomes. Because residence hall was not a major source of variation in outcomes, we concluded that the stepwise regression models reported above were the more appropriate analytic approach. Exploratory analyses. The competition was advertised as a group event with a group-level incentive (building pizza party), and therefore it was possible that level of group identification influenced outcomes. To examine this possibility, additional exploratory analyses examined the effects of self-reported group identification on outcomes in two ways. First, group identification was added as a predictor to the stepwise regression models. Second, group identification was examined as a mediator of the effect of intervention status on each outcome. 39 In the first set of analyses, a series of regression analyses considered the effect of group identification on each behavior outcome. Results are detailed in Appendix 1C. In a set of preliminary univariable models, each of the three behavior outcomes (i.e., overall PEB, electricity use subscale, consumer behavior) was separately regressed on time 1 and time 2 group identification scores. Time 1 and time 2 group identification scores accounted for 0.01-1.5% and 0.9-3.5% of the variance across the outcomes, respectively. Next, the same model building procedures used to test the three behavior outcomes above were used again, with group identification entered on step 3, intervention status entered on step 4, and the intervention status x time 1 score interaction term entered on step 5. This was done separately for time 1 and time 2 group identification scores. In the final models, time 1 group identification had a significant positive effect on time 2 PEB and consumer behavior scores; time 2 group identification had a significant positive effect on time 2 electricity use and PEB scores. Relative to the final models excluding group identification (table 1.7), those that included group identification measured at either time also reduced the effect size of target status in the final models, particularly for the electricity use and consumer behavior outcomes. This suggests that group identification may mediate the effects of intervention status on time 2 scores for these outcomes. Other model parameters did not vary appreciably between the models excluding and including group identification scores. As these were exploratory analyses, results are not elaborated upon further. In the second set of analyses, we examined time 2 group identification score as a potential mediator of intervention status on time 2 behavior using the procedures outlined 40 by Baron and Kenny (1986). The meditational hypothesis was not supported for any of the three behavior outcomes. As these were exploratory analyses and hypotheses were not supported, results are not shown. Qualitative results Qualitative data were provided from two follow-up focus groups as well as survey 2 follow-up items. Focus Groups. Some of the most salient topics addressed in focus groups were related to project awareness, recruitment, group identity, and incentives. Project awareness and recruitment. Focus group participants indicated that they lacked sufficient information about how the competition worked. They also reported that the study’s lack of physical presence on campus contributed to a lack of awareness in this regard. Participants mentioned that they knew the general goal of the project was to engage in “green” behavior, but many were unaware of the existence of the website. Mixed responses were provided regarding resident advisors as a recruitment source. Some participants stated that resident advisors (RAs) were the most powerful source of information in dormitories, followed by building governments. Participants reported that RAs could encourage participation in an event or group through physical advertising or word of mouth, and that residents would have been more likely to participate in events if told by RAs versus another source. Others commented that they only attended activities of interest and that the influence of RAs decreased with time after move-in. Participants reported that the study’s Facebook page was underdeveloped and recommended more active use of Facebook by study staff, including more events and sending more Facebook 41 messages. Participants also suggested the following recruitment strategies: presentations about the project in classes, events (e.g., movie night), advertise project through student government and other large student organizations, always have a physical presence, on- campus tables, large and artistic posters, in-person surveys with immediate incentives. Incentives and group identity. All participants agreed that the raffle prizes were motivating incentives for completing the surveys. However, participants reported that the pizza party was not a good incentive for winning the competition in part because the party was for the entire building. Instead, they recommended free t-shirts, and overall believed that individual-level incentives would be more successful. Participants stated that they did not share a strong sense of group identity with their entire buildings, which included up to 450 residents. Instead, they reported feeling closer with individual suites, floors, and sides (of buildings). Participants also stated that they felt closer with residents in their buildings at the beginning of the school year, and as the year went on, people began to drift apart and/or friendships and cliques solidified independent of building. One participant remarked: “I maybe know only about 50 people in the entire building, but I’m not even sure if my friends in other buildings know everyone. That’s why I think a pizza party for the entire dorm is not as good of an incentive…. The 50 people that I know are just people that I met and hang out with. Even if I did the survey, I don’t think I would have told them about it so that we could get the pizza party”. 42 Survey 2 follow-up items. A total of 79 intervention participants completed follow-up items at the end of Survey 2. Among these respondents, 97% (n=77) reported they were aware of the competition. Among the 44 respondents who reported that they discussed the project with friends inside their dorm, approximately 66% (n=29) reported that these discussions had little to no impact on their behavior during the competition, and 32% (n=14) reported that the discussions had somewhat of an effect on their behavior. Among the 26% (n=20) of individuals who indicated they discussed the project with friends who lived outside their dorm, 75% (n=15) reported that these discussions had little to no effect on their behavior, whereas 35% (n=7) reported somewhat of an effect, and one participant reported that her behavior was affected “a lot”. A total of 71 (90%) respondents indicated that they planned to sustain behavior changes after the competition. Chapter 1 Discussion An 8-week intervention was applied to a group of university dormitory residents in fall 2009. The intervention was framed as a competition and included an appeal to reduce dormitory energy consumption as well as a group-level reward. Intervention effectiveness Summary of dormitory electricity use results. During the 8-week intervention, changes in electricity use from the baseline to intervention phase ranged from a decrease of 7.1% among one building to an increase of 3.1% among another without adjusting for temperature effects. Multi-level mixed models adjusting for average temperature, year, and study phase indicated significantly higher electricity use among target buildings 43 during the intervention phase relative to the baseline phase in both years studied (i.e., 2008 and 2009). Higher maximum temperatures were also significantly associated with greater electricity use. Summary of self-reported behavior results: within-person change. Based on self- report data, PEB, including behaviors specifically related to electricity use, increased significantly from baseline to post-intervention across the sample. There was no overall change in consumer behavior across time. Regression results indicated that time 1 behavior was the most important predictor of time 2 behavior for all three of the self- reported outcomes, accounting for a majority of the explained variance in each. This demonstrates that baseline levels of several types of PEB strongly predict subsequent levels, and suggests challenges in increasing PEB among individuals with low baseline levels. Summary of self-reported behavior results: between-group change. No significant effects of the intervention were identified for the general PEB, electricity use, or consumer behavior outcomes in regression models. Synthesis. Taken together, the data do not support a competition, group-level incentive, and appeals intervention in promoting changes in electricity use behaviors among a university undergraduate population. These null results contrast with findings of prior university-based energy reduction interventions. For instance, as part of a gas reduction intervention, information and group rewards were applied to a group of university apartment residents using a competition framework (McClelland & Cook, 1980b). Gas use declined significantly by 6.6% over the 12-week intervention. Whereas 44 that study applied group monetary rewards as an incentive, the group reward in the present study was a building pizza party, which focus groups indicated was not incentivizing. This points to the need for investigators to identify appropriate incentives. University-based energy reduction interventions that have applied different intervention stimuli (e.g., feedback) have been more successful in promoting building energy reductions of 15% (McClelland & Cook, 1980a) and 32% (Petersen et al. 2007). This suggests that stimuli other than group-level incentives and appeals may be needed to achieve considerable reductions in electricity use among university residential populations. It should also be noted that the Petersen et al. study (2007) did not adjust for the effects of weather, so it is possible that the observed changes were influenced by weather variables as well as intervention content. Larger effect sizes would have been anticipated in the present study had intervention participants accessed the supplementary intervention website content. Unexpectedly, only six students registered to use the website, which suggests that the requirement of registration was a barrier to use, or that different methods of intervention content delivery may be better suited to undergraduate populations. This precluded planned individual-level analyses of the effects of the different intervention components on behavior change. Our findings also contrast with those of other studies that have supported the effectiveness of group-level rewards combined with appeals in other multi-tenant settings. For instance, Wodarski (studies 2 and 3; 1982) found preliminary evidence supporting the effectiveness of appeals and group-level incentives in reducing apartment building electricity use up to 8% relative to baseline periods. These findings may differ 45 from those of the present study due to the different populations studied (elderly and low- income apartment residents). Alternatively, findings may differ because the Wodarski interventions included information booklets and a group workshop, making it difficult to determine the independent contributions of the appeals and incentives. Future work should evaluate the independent contributions of intervention components. Findings of the multi-level models based on electricity data suggest that temporal variables explain electricity consumption even when adjusting for temperature. We therefore recommend that future intervention planners examine temporal patterns in electricity consumption to identify the optimal time to implement an intervention that is likely to have an observable effect. Whereas building electricity use indicated significantly higher use among target buildings during the intervention phase, self-report data indicated a modest, non- significant improvement in electricity use behaviors among intervention participants. These findings may conflict because building energy use is too rough a measure to capture any such individual-level changes, and intervention participants constituted only 8% of the dormitory resident population across intervention buildings. With small effect sizes among a minority of the dormitory population, these changes may be unlikely to be detected at the building level. Changes may be more likely to be observed if a larger proportion of residents in a building participate, as has been demonstrated in the Petersen et al. study (2007), or if electricity use is assessed at a more granular level. Participation and project awareness 46 That few students registered to use the intervention website in our study adds to the mixed success of prior published studies in recruiting college dormitory residents to participate in energy reduction events. Based on website visitation data in the Petersen et al. study (2007), 46% of on-campus dormitory residents accessed the study website at least once, indicating a fairly high level of recruitment. Furthermore, those in the real- time feedback group made an average of 4.8 visits to the website compared to 2.5 visits among low resolution feedback group, suggesting enhanced interest in high resolution feedback. A total of 418 students (23% of dormitory residents) completed a follow-up survey that was administered after the intervention. In the present study, approximately 10% of eligible residents in the target buildings participated in the first survey, which is less than the Petersen et al. study (2007) but higher than other voluntary university-based intervention projects. For instance, an electricity reduction education program among residents of two Tufts University undergraduate dormitory buildings recruited a total of 24 participants (Marcell et al., 2004). In the current study, the surveys may have attracted higher levels of participation due to the raffle prize incentives; focus group participants indicated that they participated in the surveys because of the incentives. Dormitory residents may not have registered to use website for a variety of reasons. First, focus groups indicated a lack of knowledge about competition rules and the website. Although many strategies were used to recruit participants and disseminate information about the project, limited funding and time to plan recruitment activities, both of which have been identified as barriers to recruitment in randomized clinical trials (Mitchell & Abernethy, 2005), may have hindered recruitment efforts in the present 47 study. An initial delay in funding for the study website left a relatively small (i.e., one month) window of time open to forge partnerships with residential hall staff. According to focus group data, residential staff is important in encouraging residents to participate in on-campus programs. Additionally, just over half of intervention participants indicated that they had discussed the competition with friends. This may have contributed to a perception that peers were not actively participating in the project. Such a perception may have functioned as counterproductive normative feedback, discouraging others from participating. The findings of the present study were consistent with another campus-based energy reduction program that suffered recruitment problems (Marcell et al., 2004) and did not observe any sizeable changes in building electricity use. Institutional rules regarding advertising were cited by the authors of that study as barriers to recruitment and engagement. This indicates the need for investigators to develop collaborative relationships with institutional organizations in order to effectively recruit participants. In addition, our project shared similarities with the McClelland and Cook (1980b) study, which provided group-level rewards and printed information in a competition context. In that study, only 10% of that sample had accurate knowledge about competition rules; likewise, focus groups in our study indicated a lack of awareness about competition rules. Overall, these findings highlight the difficulties in disseminating accurate information about and recruiting participants for campus-based energy reduction interventions. Recommendations 48 We used recruitment methods that were expected to be effective based on prior work, but we were not successful in recruiting participants to register for the website and access the supplementary intervention content. Based on our experiences, we developed a set of recommendations for investigators who attempt to implement interventions in university residential settings: Maintain an appropriate physical presence. On campuses where face-to-face contact is the norm for campus events and activities, interventions should strive to maintain a physical presence. Although the present study used a kitchen-sink approach to recruitment, a physical presence on campus was lacking. This contributed to confusion among students who were accustomed to in-person meetings and events. Insufficient physical presence may hinder recruitment by reducing perceived trust or credibility, both of which are related to participation rates in household energy conservation programs (Becker & Seligman, 1978; Costanzo et al., 1986; McGuire, 1985). Prioritize recruitment and obtain stakeholder buy-in. As noted in prior work, poor recruitment can lead to high costs, reduced statistical power, and consequently reduced scientific value of studies (Mapstone et al., 2007). Therefore, recruitment should be considered an important component of a study, with sufficient time and funding devoted to it. In the present study, difficulty with outreach combined with limited funding delayed the launch of the intervention and slowed the forging of partnerships with residential hall staff and campus groups, which according to focus group data may have affected participation. 49 Recruitment source has been found to influence participation rates in interventions that aim to encourage PEB (Craig & McCann, 1978; Miller & Ford, 1985). Related, research on perceived trust and credibility (Becker & Seligman, 1978; Costanzo et al., 1986; McGuire, 1985) suggests that recruitment efforts should operate at least in part through familiar and trusted communications channels. For the present study, leaders of student groups and residence hall staff, who were presumably more familiar to the target audience than study staff, were contacted and asked to disseminate messages about the project to potential participants. However, it is possible that the study failed to obtain sufficient buy-in from these parties, and that messages about the study were not forwarded on to potential participants. It is also possible that these parties did not accurately understand the project, or did not communicate it in a positive light to potential participants. Future work should provide for careful assessment of recruitment protocol fidelity, as well as the effects of levels of perceived trust and credibility of recruitment source on participation. Additionally, prior to beginning recruitment, it may be useful to hold in-person information sessions with leaders of groups through whom the project will be advertised in order to obtain buy-in. Depending on study design, it may also be useful to hold initial meetings with groups of target participants. Identify appropriate content delivery formats. In addition to having an interactive web portal, the Petersen et al. study (2007) placed marquees in residence hall lobbies that displayed feedback on electricity use, increasing the likelihood that students would be exposed to intervention content on a day-to-day basis. In the present study, supplementary intervention content could only be accessed online, and the information 50 about how to access the website, or existence thereof, seems not to have been communicated clearly. College students may be more likely to subsequently access web- based feedback applications if feedback is first presented in highly visible common areas. Select optimal project dates. Our data indicated that electricity use was likely to be higher during the intervention period relative to the baseline period. Investigators should examine temporal patterns in energy use to identify the optimal time to implement an intervention that will have observable results. Identify appropriate incentives. Individual-level incentives seemed to be effective in recruiting survey participants in our study, but the group-level incentive for the competition did not appear to function as hoped. One reason for this may be attributable to relatively low levels of group cohesion among dormitory building residents. Focus group participants indicated that dormitory residents typically are not familiar with many other residents in their building, and therefore have little motivation to participate in a project that would benefit the entire building. Group cohesion among a competing entity should be assessed and may need to be relatively high for a successful group-based intervention project. Additional research on this topic is needed. Also, when conducting an intervention that makes assumptions about underlying group processes, segmenting participants on the basis of meaningful units may be a better strategy. Alternatively, individual-based incentives may be more effective in evoking behavior change. Foster positive social norms about the project. The present study’s recruitment source data indicate that less than 5% of survey participants heard about the surveys through friends inside or outside of their dormitory buildings. Along with focus group 51 data indicating student confusion about competition rules, this suggests that there may have been a perception among residents that friends were not participating in the project. Intervention planners should make use of social norms by emphasizing positive aspects of participating (injunctive feedback) and indicating that others are participating (normative feedback). Limitations Several limitations of this study should be noted. First, our study was conducted at a private university in the southern California region with distinct weather patterns. The sample may have been biased such that students interested in environmental issues may have been more likely to participate. Therefore, the results may not generalize to other populations or settings. Next, few students registered to use the intervention website, which precluded examination of several key hypotheses. Finally, because buildings to be included in the intervention were pre-selected based on structural characteristics, the study did not have true random assignment to condition. Strengths A key strength of this study was the combination of an energy reduction intervention with the baseline and follow-up assessments of theoretical determinants of behavior. The inclusion of qualitative follow-up data was also a strength that aided in interpreting findings. Similarly, combining electricity data with self-report data provided for a more thorough evaluation of intervention effectiveness. The original design of the study provided for testing the effectiveness of individual intervention components. Although these issues could not be examined due to unanticipated barriers to 52 participation, we provided a useful set of recommendations for future researchers interested in energy reduction interventions. Recommendations may also be useful in non-university settings, such as corporate office parks. 53 Chapter 1 References Abrahamse, W., Steg, L., Vlek, C., & T. Rothengatter. 2005. A review of intervention studies aimed at household energy conservation. Journal of Environmental Psychology, 25, 273-291. Ajzen, I. 2009. Behavioral interventions based on the Theory of Planned Behavior. Retrieved on May 30, 2009, from http://people.umass.edu/aizen/pdf/tpb.intervention.pdf. Ajzen, I., & Madden, T.J. 1986. Prediction of goal-directed behaviour: attitudes, intentions, and perceived control. Journal of experimental social psychology, 22, 453- 474. Aronson, E. 1969. The theory of cognitive dissonance: A current perspective. In Zanna, M.P, & Olson, J.M., (Eds.). Advances in Experimental Social Psychology, Vol. 4, pp. 1- 34. New York: Academic Press. Axelrod, L.J., & Lehman, D.R. 1993. Responding to environmental concerns: What factors guide individual action?. Journal of Environmental Psychology 13, 149-159. Baron, R. M., & Kenny, D.A. 1988. The moderator-mediator variable distinction in social psychological research: conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51, 1173-1182. Bamberg, S. 2003. How does environmental concern influence specific environmentally related behaviors? A new answer to an old question. Journal of Environmental Psychology, 23, 21-32. Bamberg, S., & G. Moser. 2007. Twenty years after Hines, Hungerford, and Tomera: A new meta-analysis of psycho-social determinants of pro-environmental behavior. Journal of Environmental Psychology, 27, 14-25. Baron, R.M., & Kenny, D.A. 1986. The moderator-mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51, 1173-1182. Becker, L.J., & Seligman, C. 1978. Reducing air conditioning waste by signaling it is cool outside. Personality and Social Psychology Bulletin, 4, 412-415. Bertoldi, P., & Atanaiu, B. 2006. Electricity consumption and efficiency trends in the enlarged European union: Status report 2006. Institute for environment and sustainability, 2007. http://re.jrc.ec.europa.eu/energyefficiency/pdf/EnEff%20Report%202006.pdf 54 Bittle, R.G., Valesano, R.M., & Thaler, G. 1979. The effects of daily cost feedback on residential electricity consumption. Behavior Modification, 3, 187-202. Brandon, G., & Lewis, A. 1999. Reducing household energy consumption: A qualitative and quantitative field study. Journal of Environmental Psychology, 19, 75-85. Breckon, D.J., Harvey, J.R., & R.B. Lancaster. 1989. Community health education: Settings, roles, and skills. 2 nd ed. Rockville, MD: Aspen Publishing. California Department of Energy. 2009. Energy consumption data management system. http://www.ecdms.energy.ca.gov/Elecbyutil.asp#results. Cheung, S.F., Chan, D.K.S., & Z.S.Y. Wong. 1999. Reexamining the theory of planned behavior in understanding wastepaper recycling. Environment and Behavior, 31, 587- 612. Craig, S., & McCann, J.M. 1978. Assessing communication effects on energy conservation. Journal of Consumer Research, 5, 82-88. Costanzo, M., Archer, D., Aronson, M., & Pettigrew, T. 1986. Energy conservation behavior: The difficult path from information to action. American Psychologist, 41, 521- 528. Deci, E.L., & Ryan, R.M. 1985. Intrinsic motivation and self-determination in human behavior. New York: Platinum Press. Donaldson, S.I., Graham, J.W., & W.B. Hanse. 1994. Testing the generalizability of intervening mechanism theories: Understanding the effects of adolescent drug use prevention interventions. Journal of Behavioral Medicine, 17, 195-216. Festinger, L. 1957. A theory of cognitive dissonance. Stanford: University Press. Fried, C. B. & E. Aronson. 1995. Hypocrisy, misattribution, and dissonance reduction. Personality and Social Psychology Bulletin, 21, 925-934. Geller, E. S. (2002). The challenge of increasing proenvironmental behavior. In R. B. Betchel, & A. Churchman (Eds.), Handbook of environmental psychology (pp. 525- 540). New York: John Wiley & Sons. Guagnano, G.A., Stern, P.C., & T. Dietz. 1995. Influences on attitude-behavior relationships: A natural experiment with curbside recycling. Environment and Behavior, 27, 699-718. 55 Hayes, S.C., & Cone, J.D. 1977. Reducing residential electrical energy use: Payments, information, and feedback. Journal of Applied Behavior Analysis, 10, 425-435. Henry, K.B., Arrow, H., & Carini, B. 1999. A tripartite model of group identification: Theory and measurement. Small Group Research, 30, 558-581. Hines, J.M., Hungerford, H.R., & A.N. Tomera. 1987. Analysis and synthesis of research on responsible environmental behavior: A meta-analysis. Journal of Environmental Education, 18, 1-8. Hirst, E., & S. Grady. 1982-1983. Evaluation of a Wisconsin utility home energy audit program. Journal of Environmental Systems, 12, 303-320. Hopper, J., & J. Nielsen. 1991. Recycling as altruistic behavior: Normative and behavioral strategies to expand participation in a community recycling program. Environment and Behavior, 23, 195-220. Kahn, E.B., Ramsey, L.T., Brownson, R.C., Heath, G.W., Howze, E.H., Powell, K.E., Stone, E.J., Rajab, M.W., Corso, P., & the Task Force on Community Preventative Services. The effectiveness of interventions to increase physical activity. American Journal of Preventive Medicine, 22, 73-107 Kaiser, F.G., 1998. A general measure of ecological behavior. Journal of Applied Social Psychology, 28, 395-422. Kaiser, F.G., Wolfing, S., & U. Fuhrer. 1999. Environmental attitude and ecological behaviour.. Journal of environmental psychology, 19, 1-19. Kantola, S.J., Syme, G.J., & Campbell, N.A. 1984. Cognitive dissonance and energy conservation. Journal of Applied Psychology, 69, 416-421. Katzev, R., & Johnson, T.R. 1984) Comparing the effects of monetary incentives and foot-in-the-door strategies in promoting residential electricity conservation. Journal of Social Psychology, 14, 12-27. Mapstone, J., Elbourne, D., & I. Roberts. 2007. Strategies to improve recruitment to research studies. Cochran Database of Systematic Reviews, 18. Marcell, K., Agyeman, J., & A. Rappaport. 2004. Cooling the campus: Experiences from a pilot study to reduce electricity use at Tufts university, USA, using social marketing methods. International Journal of Sustainability in Higher Education, 5, 169-189. 56 McCalley, M.T., & Midden, C.J.H. 2002. Energy conservation through product- integrated feedback: The roles of goal-setting and social orientation. Journal of Economic Psychology, 23, 589-603. McClelland, L., & Cook, S.W. 1979-1980. Energy conservation effects of continuous in- home feedback in all-electric homes. Journal of Environmental Systems, 9, 169-173. McClelland, L, & S.W. Cook. 1980a. Energy conservation in university buildings: Evaluation Review, 4. 119-133. McClelland, L, & S.W. Cook. 1980b. Promoting energy conservation in master-metered apartments through group financial incentives. Journal of Applied Psychology, 10, 20-31. McGuire, W.J. 1985. Attitudes and attitude change. In G. Lindzey & E. Aronson Eds., Handbook of social psychology pp. 223-246. New York: Random House. Midden, J.F., Weenig, M.H., & Zieverink, H.J.A. 1983. Using feedback, reinforcement, and information to reduce energy consumption in households: A field experiment. Journal of Economic Psychology, 3, 65-86. Miller, R.D., & Ford, J.M. 1985. Shared savings in the residential market: A public/private partnership for energy conservation. Baltimore, MD: Urban consortium for technology initiatives, Energy task Force. Mitchell, G. K., & A. P. Abernethy. 2005. A comparison of methodologies from two longitudinal community-based randomized controlled trails of similar interventions in palliative care: What worked and what did not?, Journal of Palliative Medicine, 8, 1226- 1237. National Oceanic and Atmospheric Administration. 2009. National weather service observed weather. Retrieved on March 9, 2010, from http://www.weather.gov/climate/index.php?wfo=lox National Oceanic and Atmospheric Administration. 2010. U.S. Department of Commerce National Oceanic and Atmospheric Administration National Environmental Satellite, data, and information service. Retrieved on January 20, 2011, from http://www.ncdc.noaa.gov/oa/documentlibrary/hcs/cdd.200801-200912.pdf Pallak, M.S., Cook, D.A., & Sullivan, J.J. (unpublished). The effects of comparison feedback on electricity usage. University of Iowa. Pallak, M.S., & Cummings, N. (1976). Commitment and voluntary energy conservation. Personality and Social Psychology Bulletin, 2, 27-31. 57 Paulhus, D.L. 2002. Socially desirable responding: The evolution of a construct. In H. Braun, D.N. Jackson, & D.E. Wiley (Eds.), The role of constructs in psychological and educational measurement (pp. 67-88). Hillsdale, NJ: Erlbaum. Petersen, J.E., Shunturov, V., Janda, K., Platt, G., and K. Weinberger. 2007. Dormitory residents reduce electricity consumption when exposed to real-time visual feedback and incentives. International Journal of Sustainability in Higher Education, 8, 16-33. Pitts, R.E., and J.L. Wittenbach. 1981. Tax credits as a means of influencing consumer behavior. Journal of Consumer Research, 8, 335-338. SAS Institute Inc. SAS 9.2. Cary, NC: SAS Institute, 2002. Schwartz, S.H. 1994. Are there universal aspects in the structure and contents of human values? Journal of Social Issues, 50, 19-45. Schultz, P.W., Gouveia, V.V., Cameron, L.D., Tankha, G., Schmuck, P., & M. Franek. 2005. Values and their relationship to environmental concern and conservation behavior. Journal of Cross-cultural Psychology, 36, 457-475. Schultz, P.W., Nolan, J.M., Cialdini, R.B., Goldstein, N.J., & Griskevicius, V. 2007. The constructive, destructive, and reconstructive power of social norms. Psychological Science, 18, 429-434. Seligman, C., & J.M. Darley. 1977. Feedback as a means of decreasing residential energy consumption. Journal of Applied Psychology, 62, 363-368. Staats, H., Harland, P., & Wilke, H.A.M. 1996. Communicating the greenhouse effect ot he public: Evaluation of mass media campaign from a social dilemma perspective. Journal of environmental management, 45, 189-203. Stern, P.C., Dietz, T., Abel, T., Guagnano, G.A., & L. Kalof. 1999. A value-belief-norm theory of support for social movements: the case of environmentalism. Research in Human Ecology, 6, 81-97. Stern, P.C., Dietz, T., Kalof, L., & G.A. Guagnano. 1995. Values, beliefs, and proenvironmental action: Attitude formation toward emergent attitude change. Journal of Applied Social Psychology, 25, 1611-1636. Stutzman, T.M., & S.B. Green. 1982. Factors affecting energy consumption: Two field tests of the Fishbein-Ajzen model. Journal of Social Psychology, 117, 183-201. US Energy Information Administration 2010. Retrieved on June 17, 2010 from http://www.eia.doe.gov/aer/consump.html 58 Van Houwelingen, J.H., & F.W. Van Raaij. 1989. The effect of goal setting and daily electronic feedback on in-home energy use. Journal of Consumer Research, 16, 98-105. Van Liere, K., & R. Dunlap. 1978. Moral norms and environmental behavior: An application of Schwartz’s norm-activation model to yard burning. Journal of Applied Social Psychology, 8, 174-188. Winett, R.A., Hatcher, J.W., Foret, T.R., Leckliter, I.N., Love, S.Q., Riley, A.W., & Fishback, J.F. (1982). The effects of videotape modeling and daily feedback on residential electricity conservation, home temperature and humiity, perceived comfort, and clothing worn: Winter and Summer. Journal of Applied Behavior Analysis, 15, 381- 402. Winnett, R.A., Kagel, J.H., Battalio, R.C., & R.C. Winkler. 1978. Effects of monetary rebates, feedback, and information on residential electricity conservation. Journal of Applied Psychology, 63, 73-80. Winett, R.A., Neale, M.S., & Grier, H.C. 1979. Effects of self-monitoring and feedback on residential electricity consumption. Journal of Applied Behavior Analysis, 12, 173- 184. Wodarski, J.S. 1982. National and state appeals for energy conservation: A behavioral analysis of effects. Behavioral Engineering, 7, 119-130. 59 Chapter Two: Explaining Change in Pro-environmental Behavior Based on the Theory of Planned Behavior and Norm Activation Models Chapter 2 Abstract Background. Numerous studies have shown that pro-environmental behavior (PEB) can be explained cross-sectionally using a number of theories, including the Theory of Planned Behavior (TPB) and Norm Activation Model (NAM). However, relatively little is known about how these models perform in explaining behavior change. Understanding factors that influence behavior change is a stronger method for identifying targets for intervention efforts relative to an approach that identifies predictors of static level of PEB. Methods. The aim of the present study was to examine factors that predict change in PEB based on the TPB and NAM. Participants were 206 undergraduates who completed baseline and follow-up self-report surveys that assessed behavior and other key constructs from the TPB and NAM. These included attitudes, expected outcomes, perceived control, personal and social norms, ascribed responsibility, awareness of consequences, and behavioral intentions. Structural equation modeling was used to evaluate the relative fits of the models and contributions of model constructs to explaining variation in behavior and behavior change. First, a series of cross-sectional models representing the TPB, NAM, and a hybrid model combining features of both theories were fitted to the data. Next, based on the TPB, a stepwise model testing procedure was used whereby latent change scores were substituted in place of cross- 60 sectional model constructs. This provided for evaluation of how changes in PEB determinants predicted changes in PEB. Results. Results provided preliminary evidence for the extension of the TPB to addressing change in PEB. Results also indicated that changes in intentions were most important for explaining change in PEB. Other variables contributed a fractional amount of explained variance in PEB change. Conclusions. These results suggest that the TPB can be extended to address behavior change. Additionally, our findings indicate that PEB intervention design should consider how intervention strategies can directly or indirectly promote changes in intentions. Finally, this study points to a need for reliable and valid self-report instruments to assess theoretical constructs relevant to PEB. 61 Chapter 2 Background Pro-environmental behavior (PEB) is defined here as any behavior that supports the sustainability of natural ecosystems, environmental health, and “contribute[s] towards environmental preservation and/or conservation” (Axelrod & Lehman, 1993, p. 153). Under this umbrella term are behaviors such as recycling, driving and engaging in other modes of transportation, eating, consumerism, electricity, gas, and water consumption and conservation, donation of time and money to environmental organizations, and political support for environmental causes. A number of theoretical models have been proposed to explain PEB, and although empirical support has been found for pieces of models, tests of the full models are sparse. Understanding relationships among predictors of PEB is important for guiding intervention strategies that aim to promote PEB. The evidence for selected conceptual models is reviewed below. Explaining behavior with theoretical models Research supports the application of several theoretical models to PEB, including the Norm Activation Model (NAM; Schwartz 1994) and the Theory of Planned Behavior (TPB; Ajzen & Madden, 1986). The Norm Activation Model (NAM), depicted in figure 2.1, posits that PEB is a type of altruistic behavior predicted directly by personal norms and indirectly by social norms. The relationship between personal norms and behavior is proposed to be moderated by awareness of consequences (AC) and ascribed responsibility (AR), such that personal norms are activated only when a person is both aware of the consequences of performing or not performing a particular behavior, and ascribes responsibility for these consequences to the self. When these conditions are met, 62 NAM predicts that a person will act in accordance with personal norms. Several studies support the application of pieces of the NAM to PEB (Guagnano et al. 1995; Hopper & Nielsen, 1991; Stern et al. 1995; Van Liere & Dunlap 1978). For instance, Van Liere and Dunlap (1978) found a significant interaction between AC and AR in predicting yard- burning behavior, which contributes to air pollution. Among 307 US residents surveyed by telephone, those reporting both high AR and high AC were less likely to burn yard waste than others. Schultz and colleagues (2005) replicated and examined the NAM using self-transcendence values to approximate personal norms. Based on survey data from 988 college students in six countries, regression analyses yielded main effects of self-transcendence values (β=.18), AC (β=.18) and AR (β=.15) on PEB, as well as a significant three-way interaction between self-transcendence, AC, and AR. Partial correlations between self-transcendence and PEB were .24 among the high AC-high AR group, and .07 among the low AC-low AR group, lending support to the NAM. In another study based on a community sample of 420 US adults, 17.6% of environmentally-relevant consumer behavior was explained by the NAM, although AR was not included in this model (Stern et al., 1999). Adding attitudes and values to the model increased the explained variance in consumer behavior to 22.7%. These findings suggest that modifying existing theoretical models may increase their explanatory power. 63 Figure 2.1. Norm Activation Model. The Theory of Planned Behavior (TPB), depicted in figure 2.2, postulates that behavior is directly determined by intention to perform the behavior, and intention is predicted by three factors: attitudes, social norms, and perceived control (Ajzen & Madden, 1986). Although full tests of the model are few, there is considerable empirical support for the application of pieces of TPB to PEB (Cheung et al. 1999; Kaiser et al. 1999; Stern et al. 1995; Stutzman & Green 1982). Additionally, Bamberg (2003) fit a TPB model to data from a study of 380 university students. The model explained 60% of the variance in PEB and supported the TPB: PEB was predicted proximally by a behavioral intention factor, which was predicted by environmental attitudes, social norms, and perceived control. Ascribed Responsibility Awareness of Consequences Behavior Social Norms Personal Norms 64 Figure 2.2. Theory of Planned Behavior. Promoting behavior change through interventions A variety of interventions have been implemented to successfully increase PEB. Two overarching intervention approaches have been used: behavioral antecedent strategies and behavioral consequence strategies. Below we briefly review select strategies based on these approaches and hypothesize about how intervention strategies might act on theorized PEB determinants to result in behavior change. For exhaustive reviews of PEB intervention strategies, see Abrahamse et al. (2005) and Geller (1992). Antecedent interventions aim to influence a given behavior prior to its performance. For instance, information provision ranges from conceptual, for instance, providing facts about global warming, to specific, such as tailored advice based on home audits. Related to the NAM, information may increase awareness of problems and knowledge about possible solutions. Related to the TPB, Ajzen (2009) suggests that information can influence attitudes and perceived control by modifying existing beliefs Behavioral Intention Expected Outcome Attitude Perceived Control Behavior Social Norms Personal Norms 65 related to those constructs. Information is a very common intervention strategy, and a meta-analysis based on the results of 8 studies found a corrected correlation coefficient of .47 (SD=.29) between information approaches and increased PEB (Hines et al., 1987). However, it is unclear whether information was combined with other approaches in these interventions. Other studies have found that relative to other intervention strategies, information alone produces smaller effects and is often used as a control condition (Seligman & Darley, 1977; Winett et al., 1978). In fact, a preponderance of evidence has shown that information alone is not sufficient to produce substantial changes in PEB (for reviews see Stern, 1992; Syme et al., 2000; Winett & Ester, 1983). Information may be best used as an adjunct to other approaches Another common antecedent intervention strategy is goal-setting, which entails presenting participants with a reference point, for instance to save 5% or 10% electricity relative to use in some prior time period (Abrahamse et al. 2005). Goal setting is most often combined with information or feedback strategies that highlight conflicts between actual behavior and stated goals. This has the potential to evoke cognitive dissonance. For example, goal setting combined with feedback has been found to be more effective than feedback alone (McCalley & Midden, 2002) or goal setting alone (Becker, 1978). Goal setting does not neatly map on to the TPB and NAM, highlighting the limitations of these theories in explaining behavior change. In contrast to behavioral antecedent approaches, behavioral consequence strategies are implemented after behaviors are performed, with the goal of impacting future engagement in the same behaviors. Overall, feedback is an effective intervention 66 strategy. The Hines (1987) meta-analysis found a corrected correlation coefficient of .28 (SD=0.11) between feedback strategies and increased PEB, supporting this strategy. With respect to energy use in particular, several types of feedback have been tested. Descriptive feedback entails providing individuals with information about their behavior after some baseline period. This enables people to associate their behaviors with outcomes and therefore has the potential to influence awareness of consequences, ascribed responsibility, and perceived control. Normative feedback provides individuals with information about behavior of neighbors or other relevant social groups, which may influence perceived social norms. Descriptive feedback is generally an effective strategy for reducing household energy use (Abrahamse et al. 2005), with effectiveness increasing with increased frequency of provision (Petersen et al., 2007; van Houwelingen & Van Raaij 1989). Normative feedback has received mixed support, with some studies finding no differences between treatment groups who receive normative feedback compared to descriptive feedback (Abrahamse et al., 2007; Brandon & Lewis, 1999), and others finding enhanced savings among normative compared to descriptive feedback treatment groups (Midden et al., 1983; Pallak et al., unpublished). These mixed findings may be attributable to the undesirable boomerang effect described by Schultz et al. (2007) and identified in several other intervention studies as an increase in energy use among initially low users of energy following the provision of normative feedback (e.g., Brandon & Lewis, 1999; Van Houwelingen & Van Raaij, 1989). Administering injunctive feedback, which communicates overall approval or disapproval, along with normative feedback, has been shown to prevent this boomerang effect. 67 Rewards represent another consequence strategy that is well-documented for promoting PEB (Abrahamse et al., 2005) ranging from reducing electricity use and littering to increasing recycling (for review see Geller, 1992). However, positive changes associated with rewards have been shown to decline towards the end of interventions (McClelland & Cook 1980b) or after rewards are withdrawn (Pitts & Wittenbach 1981). Rewards do not neatly fit into either of the above theories, and may operate as an extrinsic reinforcer. This highlights the limitations of these theories to explain some mechanisms of change in PEB. Explaining behavior change Together, the theoretical and intervention literatures tell a story about the many factors associated with PEB and the variety of strategies that can be used to promote PEB. Understanding level of PEB and its correlates is an essential foundation, but to advance this field, information is needed about factors that influence behavior change. To the authors’ knowledge, no studies have directly examined correlates or models of behavior change related to PEB. In other fields, such as obesity research, studies of predictors of behavior change are also sparse, whereas a great deal of work has addressed correlates and theoretical models of level of behavior (Baranowski et al., 1999; Sutton, 2000). Two main issues have hindered the ability of the existing literature to address causal processes related to behavior change. First, theoretical studies of PEB have generally used cross-sectional designs, and consequently, have focused on explaining cross-sectional associations among variables related to PEB. This approach has limited 68 the information yielded about mechanisms of behavior change. Second, the intervention literature largely has not provided for the assessment of theoretical determinants of PEB, limiting current understanding about which behavioral determinants are affected by intervention strategies, and how these contribute to behavior change. As an example, interventions that have aimed to modify social norms are well-supported in evoking change in PEB in the desired direction (Cialdini et al., 1990; Goldstein et al., 2008). The construct of social norms is a determinant of behavior under both the TPB and NAM, so it seems logically sound to conclude that behavior change in these interventions resulted from modification of social norms. However, there is little direct empirical support for this conclusion, in part because social norms have typically not been assessed in intervention work. Additionally, even if the assumption is correct that social norms change as a result of an intervention, several questions remain: do changes in one model construct, such as intentions, account for all observed changes in behavior, or do additional variables explain behavior change as well? To the present authors’ knowledge, no study has directly evaluated these theoretical mechanisms of change. Study goals The present study aimed to describe a theoretical set of relationships that explains change in PEB. Given prior studies based on the TPB that have found intentions to be the key predictor of PEB and account for up to 88% of PEB variance (Kaiser et al., 2007), another objective was to examine whether changes in intentions alone best account for changes in PEB, or if changes in more distal model constructs contribute to explaining change in PEB as well. 69 Chapter 2 Method Participants completed baseline and follow-up self-report surveys regarding PEB and related constructs based on the Theory of Planned Behavior (Ajzen & Madden, 1986) and the Norm Activation Model (Schwartz, 1994). The surveys were part of a larger study involving a competition-based intervention intended to reduce energy use. Results from the intervention portion of the study are reported in another paper (see Paper 1). Procedures Beginning in early September 2009, the baseline survey was accessible online to participants 24 hours per day for eight weeks. It required approximately 25 minutes to complete. The intervention began in late September and continued through the end of November, 2009, at which point the follow-up survey became accessible. The follow-up survey was also administered online and was available 24 hours per day for three weeks. It was identical in content to the baseline survey but included an additional set of follow- up items regarding perceptions of the intervention. The intervention was advertised as a dormitory-versus-dormitory competition with a pizza party as the reward for the building that reduced its electricity usage by the largest percentage compared to its own baseline. A total of seven on-campus dormitories were selected to take part in the 8-week competition. All participating dormitories were treated equally with respect to intervention content: posters were hung weekly encouraging residents to participate in the competition, and individuals who completed the baseline survey and indicated they resided in one of target buildings were emailed and encouraged to participate in the competition. As well, residential staff among all 70 participating buildings were contacted and asked to distribute information about the competition to residents of their buildings. Recruitment strategies for the surveys and intervention included emails, poster advertising, the social networking website www.Facebook.com, the psychology subject pool, and contact with leaders of student groups. Survey participants were entered into raffle drawings to receive a variety of prizes, ranging in value from $10-$300. Additionally, participants who were enrolled in participating Psychology Department courses received credit for survey participation that could be applied towards courses. Measures The surveys assessed demographic variables constructs included as part of the NAM and TPB. Table 2.1 displays descriptive statistics on the following scales based on unstandardized participant scores, and table 2.2 shows sample demographic characteristics. Demographic data. Demographic variables included age, gender, year of university enrollment, and parental educational attainment. The TPB asserts that behavior is likely to be explained best when measurement of all model constructs are compatible (Ajzen, 2009). This suggests that studies may only support the TPB when measures are consistent with respect to context, time, and specificity of behavior. Empirical support for this concept has been found; a model based on measures that were not compatible explained only 3.2% of the variance in behavior relative to 50.2% when the measures were compatible (Kaiser et al., 2007). Several measures were created by the first author based on guidelines and examples from Ajzen 71 (2009). All items were formulated to be exactly compatible with the behaviors in the behavior scale. Appendix 2A contains the measures that were used. PEB. Frequency of engagement in a variety of PEB for the two weeks preceding the survey was assessed using a 17-item version of the Schultz Proenvironmental Behavior Scale (Schultz et al., 2005). Five items were added by the first author to the original 12-item scale to include behaviors particularly relevant to college students (e.g., turning off a computer when not in use). Responses were provided on a 5-point Likert scale with response options ranging from “never” to “very often”. Items were scored 1 to 5, for a possible scale score range of 17 – 85. Behavioral intentions. Behavioral intentions were assessed using a 17-item modified scale developed by the first author, with items reflecting the same behaviors assessed by the modified Schultz PEB scale. The stem question was developed by Bamberg (2003). Respondents rated their likelihood of engaging in each behavior in the next two weeks on a 5-point Likert scale ranging from “likely” to “unlikely”. Items were scored 1 to 5, for a possible scale score range of 17 – 85. Personal and social norms. Based on stem questions used previously by Bamberg (2003) and Garling et al. (2003), two scales were designed by the first author to assess perceived social norms and personal norms, respectively. The items on both scales corresponded to the 17 behavior items assessed by the modified Schultz PEB scale. The Social Norms Scale stem question stated “Most people who are important to me would support my decision to _____”. Responses were provided on a 4-point Likert scale ranging from “very unlikely” to “very likely”. Items for were scored 1 to 4, for a possible 72 scale score range of 17 – 68. The Personal Norms Scale stem question stated: “I feel a moral obligation to ____”. Responses were provided on a 7-point Likert scale ranging from “strongly disagree” to “strongly agree”. Items for were scored 1 to 7, for a possible scale score range of 17 – 119. Ascribed responsibility. Based on recommendations by Ajzen, a 17-item Ascribed Responsibility Scale was constructed by the 1st author to assess the extent to which participants judged their contribution to climate change would be modified by engaging in each of the behaviors assessed by the Schultz PEB scale. The stem question asked “How much would your participating in each of the following activities help reduce your personal contribution to future global warming? (If you already do something listed below, then rate how much you believe this reduces your personal contribution to global warming currently.)”. Responses were given on a 7-point Likert scale ranging from “would not help at all” to “would help a lot”. Items for were scored 1 to 7, for a possible scale score range of 17 – 119. Awareness of consequences. The 46-item Climate Change Quiz developed by Sundblad et al., 2007 was abridged to increase its relevance to the target population in the current study. The abridged scale included 13 items that assessed factual knowledge about global warming causes and consequences. Items were answered in true/false format and are provided in Appendix 2A. Correct answers were scored 1 and incorrect answers were scored 0, for a possible scale score range of 0-13. Attitudes. General attitudes towards the environment were assessed with the widely used New Ecological Paradigm (Dunlap et al., 2000). The scale had 15 items that 73 were rated on a 5-point Likert scale ranging from “strongly disagree” to “strongly agree”. Items for were scored 1 to 5, for a possible scale score range of 15 – 75. Expected outcome. Based on guidelines outlined by Ajzen (2009), a 17-item Expected Outcome Scale was developed by the first author to assess expected outcomes that would result from engaging in the behaviors assessed by the modified Schultz PEB scale. Responses were provided on a 7-point Likert scale with response options ranging from “very bad” to “very good”. Items for were scored 1 to 7, for a possible scale score range of 17 – 119. Perceived control. Based on a stem question developed by Zhao et al. (in preparation) and the guidelines of Ajzen (2009), a set of items was developed by the first author to create a Perceived Control Scale. This scale measured perceived ability to engage in each behavior included in the modified Schultz PEB scale. The stem question asked about the extent to which participants believe they could engage in the 17 behaviors if they wanted to. Responses were provided on a 5-point Likert scale ranging from “completely unsure” to “completely sure”. Items for were scored 1 to 5, for a possible scale score range of 17 – 85. Validity items. One true/false question was included in each survey to detect random responding. It stated “George W. Bush is the current president of the United States”. As the survey was administered in Fall 2009, the correct answer to this item was false. 74 Table 2.1. Sample characteristics for scales used in analyses. Theory Model constructs Scale Mean (SD) Internal consistency (α) Mean Time 2- Time 1 differenc e (SD) Time 1 Time 2 Theory of Planned Behavior (TPB) Intentions Intentions Scale 55.20 (8.81) N=206 .77 55.14 (9.2) N=203 .79 -0.05 (6.09) t=-0.12 Attitude New Ecological Paradigm 53.67 (7.5) N=195 .82 52.91 (8.00) N=194 .86 -0.37 (5.48) t=-0.92 Expected Outcome Expected Outcome Scale 101.04 (12.63) N=204 .89 99.18 (13.86) N=204 .91 -1.73 (13.16) t=-1.86 Perceived Control Perceived Control Scale 71.15 (10.0) N=194 .91 70.10 (11.73) N=194 .92 -1.15 (11.85) t=-1.32 Norm Activation Model (NAM) Ascribed Responsibility Ascribed Responsibility Scale 93.13 (18.90) N=198 .96 92.11 (18.37) N=197 .96 -0.94 (17.17) t=-0.75 Awareness of Consequences Sundblad Climate Change Quiz 7.62 (1.87) N=206 .57 7.63 (1.94) N=206 .48 0.01 (2.62) t=0.03 TPB and NAM PEB Modified Schultz PEB Scale 48.84 (8.05) N=206 .70 50.53 (8.47) N=206 .75 1.70 (5.68) t=4.3** Social Norms Social Norms Scale 55.15 (9.13) N=195 .95 54.89 (9.55) N=196 .95 -0.08 (10.58) t=-0.11 Personal Norms Personal Norms Scale 86.43 (17.43) N=197 .95 83.58 (20.58) N=198 .96 -2.95 (14.44) t=-2.82* **p<.0001. *p<.01. 75 Table 2.2. Pearson correlations among key measures at times 1 and 2. PEB Int PNor SNor Att Ex Out Per Cont AR AC PEB 1.0 .87 .59 .27 .26 .30 .27 .40 -.03 Int .84 1.0 .70 .38 .32 .40 .34 .54 .00 PNor .46 .55 1.0 .49 .33 .45 .30 .61 .03 SNor .25 .31 .37 1.0 .25 .50 .33 .41 .07 Att .26 .28 .39 .35 1.0 .34 .44 .45 .15 ExOut .33 .42 .54 .38 .31 1.0 .47 .55 .17 PerCont .41 .41 .46 .39 .22 .49 1.0 .46 .16 AR .32 .40 .61 .40 .35 .55 .35 1.0 .03 AC .08 .11 .06 .23 .28 .05 .07 .17 1.0 Table note. Bottom (shaded) triangle represents correlations among measures at time 1, and top triangle represents correlations at time 2. Int=intentions. PNor=Personal Norms. SNor=Social Norms. Att=Attitude. ExOut=Expected Outcome. PerCont=Perceived Control. AR=Ascribed Responsibility. AC= Awareness of Consequences. Participants To be eligible to participate in the surveys, participants had to be at least 18 years of age, a current USC undergraduate student, and a resident of USC-owned or USC- managed housing. To be eligible to participate in the intervention, participants also had to be residents in one of the seven selected target buildings. A total of 302 students participated in the first survey and a subset of 225 followed up in the second survey. Among these, 11 individuals provided responses on either survey that suggested random or careless responding (see Validity Items above). A total of 6 respondents completed only time 1 demographic questions, and 7 completed only time 1 demographics and half of the PEB scale. Responses from these 24 participants were excluded. Among the remaining 278 respondents, 216 participated at time 2. Missing scale items were imputed using regression techniques for individuals who were missing fewer than 10% of items on a given scale. Each participant who was missing more than 10% of items on a given scale was given a missing value for that scale (for each scale, 1-7 participants were 76 missing more than 10% but less than 100% of items; across the sample, 14 participants were missing between 10 and 100% of items on any scale). With these imputed data, 206 participants had data on the PEB scale at time 2, and of these, 179 had complete data on all key measures at both time points (i.e., measures listed in table 2.1). Mplus Version 5.21 was used for analyses. Mplus uses all data available to estimate a model using full information maximum likelihood. We therefore elected to use the sample of 206 individuals who had data on the PEB scale at times 1 and 2. Possible bias introduced as a result of defining the sample in this way was investigated and is reported below in the Results section. A total of 94 survey participants who reported residence in one of the seven dormitory buildings included in the competition were considered intervention participants in analyses. A total of 112 survey participants who did not reside in participating buildings were considered control participants. Based on survey 1 data, approximately 77% of participants were female, and just less than half were university freshmen. Parental education level was high among the sample. For 68% and 71% of participants, maternal and paternal education levels, respectively, were 4-year college degrees or graduate degrees. A preliminary multiple logistic regression analysis was conducted to tested for differences among intervention and control participants. Intervention status (1=intervention, 0=control) was regressed on age, sex, enrollment year, maternal education level, and paternal education level. Results indicated that younger participants (Odds ratio [OR] =0.41, 95% confidence interval [CI] =0.24, 0.73) were more likely to be 77 intervention versus control participants. See table 2.3 for additional sample characteristics. Table 2.3. Participant characteristics by intervention status based on time 1 data. Characteristic All (n=206) Intervention (n=94) Control (n=112) Mean age in years (SD) 19.02 (1.22) 18.50 (0.99) 19.46 (1.22) Sex (% female) 77 79 74 Class status Freshman Sophomore Junior Senior 48 31 15 6 73 11 11 5 26 48 19 7 Analyses Using structural equation modeling (SEM), a series of models were fit to the data using maximum likelihood (ML) estimation. First, a set of models explaining cross- sectional level of PEB was tested. Next, latent difference scores were added and modifications to each model were made in a stepwise fashion to test the direct effects of changes distal model constructs on explaining change in PEB. Testing for bias among sample. Potential bias among the sample was investigated by examining scores on the Intentions and PEB scales. These variables were selected because they were of greatest importance in the study. In addition, differences in correlations among the key time 1 variables were examined among alternative sample sizes. Validity verification. Psychometric properties of the survey measures were assessed. Internal consistencies of each scale at both measurement occasions were 78 estimated using Cronbach’s alpha. Based on time 1 data, the factor structures of the scales were tested using confirmatory factor analyses. Cross-sectional models of PEB. Using time 1 survey data, the following theoretical models were tested to evaluate which one explained the most variance in time 1 PEB: (1) TPB (figure 2.2); (2) NAM (figure 2.1); (3) Hybrid model that combined features of the TPB and NAM (figure 2.5). For all models, scores on the modified Schultz PEB scale (PEB) represented the behavior outcome. For the TPB models, observed scores on the following scales were included to represent the model constructs: Intentions Scale (Int), Perceived Control Scale (PerCont), Social Norms Scale (SNor), Personal Norms (PNor), Expected Outcomes Scale (ExOut), and New Ecological Paradigm (to assess attitudes; Att). For NAM models, observed scores on the following scales were included: SNor, PNor, Sundblad Climate Change quiz (to assess awareness off consequences; AC), Ascribed Responsibility Scale (AR). The Hybrid Model included all of those constructs used in the TPB and NAM models. To adjust for wide variation in scale score ranges, all variables were transformed to z-scores for analysis. Each theoretical model was fit to the data separately. Figure 2.2 shows graphically the theoretically assumed model of structural relationships among TPB constructs that was fit to the data. In addition to testing this initial model, modifications were made in a stepwise fashion to determine the set of relationships among TPB variables that best explained PEB. For instance, holding all other model parameters constant, a direct regression path from PNor to PEB was added to the model and the change in PEB R² and model fit indices evaluated against the baseline model. The final TPB model was selected 79 using several criteria, including amount of explained variance in PEB and Int, comparative fit index (CFI) value, and root mean square error of approximation (RMSEA). Parsimony was also considered. The same procedures were repeated for the NAM and Hybrid models. Figures 2.1 and 2.5 show graphically the baseline NAM and Hybrid models that were tested, respectively. Change model. Among the final TPB, NAM, and hybrid models, the model that explained the most variance in PEB was selected as a framework to test hypotheses about change. Latent difference scores were created for each model construct based on time 1 and time 2 scores on each variable. In contrast to directly calculating a difference score, latent difference scores provide for the separation of measurement error from systematic change. Different combinations of paths among the latent difference scores were tested in explaining ∆PEB. For instance, a direct path from ∆SNor was tested in predicting ∆PEB. Next, a direct path from ∆PNor to ∆PEB was tested. This was done for each predictor. Multiple groups analyses. To test whether there were differences in model parameters across intervention and control groups, multiple groups analyses were conducted as part of evaluating cross-sectional models as well as the change model. Previous analyses (i.e., Paper 1) did not observe a significant effect of intervention status on the outcome examined in this study, so we did not expect to observe any substantial differences across groups for the present study. Group differences were therefore evaluated only for select models. Analyses followed a multiple groups SEM approach in which an initial constrained model was specified with regression paths set equal across the two groups. A subsequent model freed regression paths across groups. Model fit 80 indices were compared to determine whether forcing equal parameters across groups produced a significant decrement in fit. The SAS version 9.1 software package was used for descriptive analyses and to prepare data for use in structural models (SAS Institute Inc., 2002). Mplus Version 5.21 was used to specify and test all structural models (Muthen & Muthen, 2006). Chapter 2 Results Testing for bias among sample Attrition analyses examined scores on the PEB and Int scales among several subgroups which could have been defined as alternative sample sizes for the present study: the 278 participants with data on the PEB scale at time 1, the subset of 206 participants who also had PEB data at time 2, and the 179 participants with complete data on all key variables. The correlations between intentions and PEB were .86, .84, and .83 across the three groups, respectively. Mean scores differed by less than one eighth of a standard deviation across the groups. In addition, correlations among all key time 1 variables were computed within each of the three possible samples. Corresponding pairs of correlations were then compared across the samples and differences of greater than .05 were identified. Of 108 comparisons, a total of 3 pairs of correlations differed by more than .05. The difference was .06 among these three pairs. Taken together, these analyses suggest that results based on the 206 individuals included in the present study would be unlikely to differ substantially if the sample were expanded to include all respondents with time 1 PEB 81 scores, or if the sample was limited to those with only complete data. See Appendix 2E for details. Validity verification The internal consistencies of the measures were generally acceptable at both time points. Cronbach’s alpha was .70 or greater for all scales except for the Sundblad Climate Change Quiz, which had somewhat lower values of .57 and .48 at times 1 and 2, respectively. See table 2.1 for details. Data from survey 1 were used to test a 1-factor structure for each scale using confirmatory factor analyses. A highly constrained model was run first equating item loadings and residuals. Next, item loadings and residuals were freed. Overall, the data had considerable misfit to the 1-factor models, which suggests that the scales represented a heterogeneous set of constructs. See Appendix 2F for details. These findings are consistent with the fact that the PEB scale captured a variety of pro-environmental behaviors, and the scales developed by the authors were designed to reflect these diverse behaviors. As we were limited to the measures available, we proceeded with the scales as originally constructed. Limitations of the measures are addressed in the Discussion section of this paper. Cross-sectional models Figures 2.3, 2.4, and 2.5 show the estimated standardized path coefficients for the final TPB, NAM, and Hybrid models, respectively. Table 2.2 shows the matrix of correlations among the measures used in the models. 82 For each theoretical model, a baseline model was specified based on the theoretically assumed set of relationships among model constructs (figures 2.1, 2.2, and 2.5). Modifications to each model were then made in a stepwise fashion. For the TPB, alternative models tested direct paths from the distal model constructs (ExOut, SNor, PerCont, Att, and PNor) to PEB to evaluate whether additional paths increased the explained variance in PEB or changed the model fit. For instance, holding other model parameters constant, a direct path from PNor to PEB was added. Next, the PNor PEB path was removed and a SNor PEB path was added. Direct paths from each distal construct were tested in this manner. As well, the effect of adding direct paths from all distal constructs to PEB was evaluated, with and without Int in the model (Appendix 2B, Models IX-XII). This provided for assessment of the role of Int in the TPB model. Compared to the baseline TPB model, modifications to the TPB model did not increase the amount of explained variance in PEB or Int, and did not improve the statistical fit of the model. See Appendix 2B for R² values and model fit indices associated with fitting alternative TPB models. The baseline TPB model was therefore retained as the final TPB model, which provided moderate misfit to the data (χ² (12) = 91.42, CFI=.82, RMSEA= .18). In the final model, Int had a strong direct effect on PEB and explained 70% of variance in PEB. Intentions were determined by PNor and PerCont, which together explained 29% of the variance in Intentions. Attitudes explained only .64% of the variance in Intentions and the Attitudes-Intention path was not significant. PNor had the strongest effect on Int, followed by PerCont. SNor had a significant effect on PNor, as did ExOut on Att. 83 Figure 2.3. Final Theory of Planned Behavior model with unstandardized path coefficients. Relative to the baseline NAM model, modifications to the NAM model did not increase the amount of explained variance in PEB nor improve the statistical fit. The baseline model was therefore retained as the final NAM model, which misfit to the data to a relatively minor degree (χ² (2) = 10.23, CFI=.90, RMSEA= .14). PNor had a direct effect on PEB and explained 21% of the variance in PEB. SNor had a significant effect on PNor and explained 14% of the PNor variance. The effect of the three-way interaction between PNor, AR, and AC on PEB was not significant. Adding main effects of AR and AC also did not improve the fit of the model or contribute to explained variance in PEB. See Appendix 2C for results of fitting alternative NAM models. Attitude Social Norms PEB Perceived Control Personal Norms Intentions Expected Outcome .32** .39** .19* .07 .40** .88** .35** .37** . 4 5 * * *p<.001,** p<.0001 ?² (12) = 91.42, CFI=.83, RMSEA=.18 84 PEB Personal Norms Social Norms Ascribed Responsibility Awareness of Consequences .38** .44** -.01 **p<.0001 ?² (2) =10.43, CFI=.90, RMSEA=.14 Figure 2.4. Norm Activation Model final model unstandardized path coefficients. Evaluating modifications to the Hybrid model included testing for 3-way interactions between AC, AR, and either Int or PNor. Relative to the baseline Hybrid model, modifications to the Hybrid model did not increase the amount of explained variance in PEB nor improve the statistical fit appreciably. Based on these findings as well as parsimony, the baseline Hybrid model was retained as the final model. The final Hybrid Model misfit the data considerably (χ² (17) = 107.4 CFI=.81, RMSEA= .16). See Appendix 2D for results of fitting alternative Hybrid models. 85 Attitude Social Norms PEB Perceived Control Personal Norms Intentions Expected Outcome .32** .39** .19* .001 .40** .88** .35** .37** . 4 5 * * *p<.001,** p<.0001 ?² (17) = 107.44, CFI=.81, RMSEA=.16 Ascribed Responsibility Awareness of Consequences -.02 .07 Figure 2.5. Final hybrid model with unstandardized path coefficients. Testing for intervention vs. control participant differences in cross-sectional model relations Potential differences in the relationships among model constructs between intervention and control participants were tested using a SEM multiple groups approach. For each theoretical model (i.e., TPB, NAM, Hybrid), select cross-sectional alternative models (in Appendices 2B, 2C, and 2D) were used as a basis to evaluate potential group differences. For each theoretical model, the same procedures were followed. First, using the baseline model, four preliminary multiple groups models tested whether greater misfit was produced by a model equating means and/or variances across the groups relative to a model in the means and/or variances were freed. Regression paths were set equal in all 86 four preliminary models. Across the TPB, NAM, and Hybrid models, no substantial differences in model fit were observed among these preliminary models. We therefore elected to proceed using models in which means and variances were free to vary. Differences between a partially constrained and free model were then tested. The partially constrained model forced equal regression paths across the intervention and control groups, but permitted means, correlations, and variances to vary. In a subsequent model, regression paths were freed across groups. Results are shown in Appendices 2B, 2C, and 2D. Results generally indicated no substantial improvement in model fit between the constrained and free models, suggesting that the intervention and control participants did not differ with respect to relationships among variables. These findings support the evaluation of the models based on the entire sample (n=206). Change model The final TPB model accounted for more variance in PEB relative to the NAM final model (70% vs. 21%, respectively). Compared to the final TPB model, the final Hybrid model did not contribute additional explained variance in PEB or Int. Therefore, the final TPB model was retained as a framework for testing the change model. Table 2.1 shows mean score changes from time 1 to time 2 among the key measures. PEB scores increased significantly by a mean of 1.7 points, which was approximately one fifth of a standard deviation. PNor scores decreased significantly by a mean of 3.0 points, representing approximately one sixth of a standard deviation. Mean scores on the other scales did not differ significantly from time 1 to time 2. 87 To construct the change model, a cross-sectional model of the TPB was first specified. For each TPB construct, a latent difference score was created using time 1 and time 2 observed scores (see Appendix 2G for a diagram). A number of baseline TPB models were tested to evaluate whether to retain regression paths among the time 1 model constructs when evaluating the change model, and to test different constraints on the correlations between observed variables and latent change scores. Two of these baseline models were selected and used in subsequent testing of alternative change models: one included the theoretical regression paths among time 1 model constructs (table 2.4c and 2.4d, baseline model), and one did not (tables 2.4a and 2.4b, baseline model). The goals of the following analyses were two-fold: 1) identify changes in model constructs that explain ∆PEB; 2) determine the relative importance of ∆Int in explaining ∆PEB in the context of other ∆scores. Similar to the procedures used in testing the cross-sectional models, a step-wise model building approach was used to evaluate the effects of changes in determinants of PEB on changes in PEB. Direct paths were estimated from the latent change score associated with each model construct to the latent difference score for PEB. See tables 2.4a-2.4d for results of fitting alternative change models. First (tables 2.4a and 2.4c), individual regression paths were estimated from each ∆score to ∆PEB to determine which model constructs were more important in explaining ∆PEB. With all theoretically consistent regression paths specified among ∆scores (table 2.4a), ∆Int accounted for 33% of variance in ∆PEB. Adding direct paths from ∆Att, ∆PNor, and ∆PCont to ∆PEB did not improve model fit nor contribute to explaining ∆PEB. 88 Next (tables 2.4b and 2.4d), the role of ∆Int in explaining ∆PEB was tested. In a global test (table 2.4b, Model II), ∆PEB was regressed on all ∆scores. Then, paths from other ∆scores to and from ∆Int were modified. Results indicated that the other ∆scores were not useful in explaining ∆PEB. For instance, table 2.4b, Model III retained ∆Int but did not specify a path from ∆Int → ∆PEB. In this model, the other predictors explained just 2% of the variance in ∆PEB. In Model IV (table 2.4b), ∆Int was removed and the remaining predictors explained 3% of ∆PEB variance. Results based on the other baseline change model were similar. In table 2.4d, Model II, ∆Int alone accounted for 44% of the variance in ∆PEB. When direct paths from the other ∆ scores were added to ∆PEB, explained variance in ∆PEB increased by just 1%. Dropping ∆Int from the model, the other ∆scores combined accounted for 4% or less of the variance in ∆PEB. These findings suggest that changes in more distal model constructs do not influence ∆PEB, and of the tested model constructs, ∆Int is most important in accounting for ∆PEB. The final change model was selected based on amount of explained variance in PEB, CFI, RMSEA, and parsimony. The final change model is represented in figure 2.6 and model fit parameters are in table 2.4d, Model IV. In the final change model, ∆Int had a significant direct effect on ∆PEB and explained 44% of the variance in ∆PEB. ∆PerCont and ∆PNor had strong effects on ∆Int, whereas the effect of ∆Att on ∆Int was quite small. Together, ∆PNor, ∆PerCont, and ∆Att explained 11% of the variance in ∆Int. ∆PNor had a significant effect on ∆SNor and explained 4% of the variance in ∆SNor. ∆ExOut explained just 0.1% of the variance in ∆Att. 89 Table 2.4a. Model fit indices and R² values for alternative change models. Baseline Model Model I Baseline add direct ∆PNor →∆PEB path Model II baseline add direct ∆Att→ ∆PEB path Model III baseline add direct ∆PerCont → ∆PEB path Model IV baseline add direct ∆SNor →∆PEB path Model V baseline add direct ∆ExOut →∆PEB path χ² (df) 126.4 (53) 126.7 (53) 126.7 (53) 124.1 (53) 125.6 (53) 126.4 (53) R² ∆PEB ∆Int ∆PNor ∆Att .33 .07 .01 .00 32 .08 .01 .00 .32 .08 .01 .00 .33 .07 .01 .00 .33 .07 .01 .00 .33 .07 .01 .00 CFI .94 .94 .94 .94 .94 .94 RMSEA .08 .08 .08 .08 .08 .08 Table note. Baseline model consisted of theoretically consistent regression paths specified among latent change scores. Regression paths among observed scores were not specified. Correlations among latent and observed variables were set to 0. Table 2.4b. Evaluating role of ∆Int on ∆PEB. Baseline Model Model I Baseline plus direct paths from ∆Att, ∆PNor, ∆PerCont→ ∆PEB Model II Baseline plus direct paths from all ∆scores→ ∆PEB Model III Baseline but remove ∆Int →∆PEB Model IV Baseline but remove ∆Int from model, add ∆Att, ∆PNor, ∆PerCont → ∆PEB paths Χ² (df) 126.4 (53) 126.3 (51) 125.1 (50) 132.3 (51) 91.4 (37) R² ∆PEB ∆Int ∆PNor ∆Att .33 .07 .01 .00 .33 .08 .01 .00 .33 .08 .01 .00 .02 .07 .01 .00 .03 -- .01 .00 CFI .94 .94 .94 .93 .93 RMSEA .08 .09 .09 .08 .08 Table note. Baseline model consisted of theoretically consistent regression paths specified among latent change scores. Regression paths among observed scores were not specified. Correlations among latent and observed variables were set to 0. 90 Table 2.4c. Model fit indices and R² values for alternative change models. Baseline Model Model I Baseline add ∆Int→ ∆PEB Model II Baseline add ∆PNor → ∆ PEB Model III Baseline add ∆SNor → ∆ PEB Model IV Baseline add ∆PerCont →∆ PEB Model V Baseline add ∆Att→ ∆ PEB Model VI Baseline add ∆ExOut →∆PEB χ² (df) 205.7 (42) 198.0 (50) 337.8 (50) 341.1 (50) 340.8 (50) 337.6 (50) 339.7 (50) R² ∆PEB PEB Int PNor Att -- .67 .29 .15 .09 .44 .71 .31 .15 .09 .02 .69 .31 .15 .09 .00 .69 .31 .15 .09 .001 .69 .31 .15 .09 .02 .69 .31 .15 .09 .01 .69 .31 .15 .09 CFI .90 .91 .82 .82 .82 .82 .82 RMSEA .14 .12 .17 .17 .17 .17 .17 Table note. Baseline model consisted of theoretically consistent regression paths specified among observed time 1 variables. Latent change scores were specified but regression paths among them were not. Correlations among latent and observed variables were set to 0. 91 Table 2.4d. Evaluating role of ∆Int on ∆PEB. Baseline Model I Baseline add ∆Int→ ∆PEB Model II Baseline add direct paths from all ∆ scores→ ∆PEB Model III Model II keep direct paths from all ∆scores → ∆PEB except ∆Int Model IV Full change model (TPB model based on ∆scores specified) χ² (df) 205.7 (42) 198.0 (50) 192.7 (45) 333.1 (46) 253.3 (66) R² PEB ∆PEB Int ∆Int PNor ∆PNor Att ∆Att .67 -- .29 -- .15 -- .09 -- .71 .44 .31 -- .15 -- .09 -- .70 .45 .31 -- .15 -- .09 -- .69 .04 .31 -- .15 -- .09 -- .73 .44 .36 .11 .16 .04 .09 .001 CFI .90 .91 .91 .82 .89 RMSEA .14 .12 .13 .17 .12 Table note. Shading represents final change model. Baseline model consisted of theoretically consistent regression paths specified among observed time 1 variables. Latent change scores were specified but regression paths among them were not. Correlations among latent and observed variables were set to 0. 92 Figure 2.6. Final TPB change model with unstandardized path coefficients. Figure note. For clarity, the latent difference score models (Appendix 2G) used to construct the latent difference scores based on time 1 and time 2 observed scores are not shown, nor are paths among time 1 model constructs. Additional Analyses: Change model based on NAM In a cross-sectional TPB model that eliminated intentions (Appendix 2B, table 2, Model XI), the other TPB variables accounted for 21% of the explained variance in TPB in the absence of Int, which was the same as the final NAM model. Because Int and PEB were highly correlated in this sample, we chose to evaluate change models based on the NAM final model as well. To construct the change model, a cross-sectional model of the NAM was first specified. For each NAM construct, a latent difference score was created using time 1 and time 2 observed scores (see Appendix 2G for a diagram). A baseline NAM model was Attitude Social Norms PEB Perceived Control Personal Norms Intentions Expected Outcome .03 .13** .09* .11 .25*** .70*** .13* .08 . 2 7 * * * . *p<.05, **p<.001,** p<.0001 ?² (66)= 253.32, CFI=.89, RMSEA=.12 ? Expected Outcome ? Attitude ? Intentions ? PEB ? Social Norms ? Perceived Control ? Personal Norms .39 .28 93 specified which included theoretical paths among the time 1 variables, and creation of latent change scores with no paths among them. The goal of this analysis was to identify the changes in model constructs that explain ∆PEB. As in the prior analyses, a step-wise model building approach was used. See table 2.5 for results of fitting alternative NAM change models. Individual regression paths were estimated from each ∆score to ∆PEB to determine which model constructs were more important in explaining ∆PEB. ∆PNor explained 5% of variance in ∆PEB. Adding direct paths from ∆SNor and ∆Interact to ∆PEB did not improve model fit nor contribute to explaining ∆PEB. Sequentially testing independent contributions of ∆scores to ∆PEB in the absence of a ∆PNor→∆PEB path showed that the other ∆SNor and ∆Interact explained 0% and .1% of the variance in ∆PEB, respectively. The final change model was selected based on amount of explained variance in ∆PEB, CFI, RMSEA, and parsimony. The final change model is represented in figure 2.7 and model fit parameters are in table 2.5, Model V. In the final change model, ∆PNor had a significant direct effect on ∆PEB and explained 5% of the variance in ∆PEB. ∆PNor had a significant effect on ∆SNor and explained 4% of the variance in ∆SNor. The interaction term did not explain ∆PEB. 94 Social Norms PEB Personal Norms .13* .21* *p<.01 ?² (15)= 74.36, CFI=.89, RMSEA=.14 ? PEB ? Social Norms ? Personal Norms AR x AC x PNor ? AR x AC x PNor .001 .44 Table 2.5. NAM change models. Baseline Model Model I Baseline add ∆PNor →∆PEB path Model II ∆SNor → ∆PEB Model III ∆Interact→ ∆PEB Model IV ∆SNor, ∆PNor, ∆Interact → ∆PEB Model V NAM model specified among change scores Model VI Model V add ∆SNor→ ∆PEB path χ² (df) 60.6 (10) 73.9 (14) 84.5 (14) 84.4 (14) 73.8 (12) 74.4 (15) 74.3 (14) R² PEB ∆PEB PNor ∆PNor .24 -- .14 -- .28 .05 .14 -- .25 .00 .14 -- .25 .001 .14 -- .28 .05 .14 -- .28 .05 .16 .04 .28 .05 .16 .04 CFI .91 .89 .87 .87 .89 .89 .89 RMSEA .16 .14 .16 .16 .16 .14 .15 Table note. ∆Interact = ∆(PNor x AR x AC). Baseline model consisted of specifying a cross-sectional NAM model, including specification of latent difference scores but no paths among them. Figure 2.7. Final NAM change model with unstandardized path coefficients. Testing for intervention vs. control participant differences in change model relationships Based on the TPB final change model and the NAM final change model, potential differences in the relationships among change model constructs between intervention and 95 control participants were tested using a SEM multiple groups approach. For each model, a series of multiple groups tests were conducted constraining different parameters to be equal across intervention and control participants. First, an initial highly constrained model forced equal regression paths, latent variable means, and latent variable variances across the two groups, but permitted correlations and observed variable means and variances to vary. In the second model, regression paths were freed across groups. In the third model, regression paths were again constrained to be equal across groups, but latent variable means and variances were freed. In the fourth model, all parameters were freed across groups. Results are shown in Appendix 2H and generally indicated no substantial improvement in model fit between the constrained and free models for the TPB, suggesting that the intervention and control participants did not differ with respect to relationships among variables. Modest improvements in fit were observed for the NAM change models that permitted latent variable means and variances to vary across groups, but no improvements were observed by freeing regression parameters. These findings support the evaluation of the models based on the entire sample (n=206). Chapter 2 Discussion The present study evaluated the Theory of Planned Behavior and Norm Activation Model based on cross-sectional and longitudinal data collected among a sample of college students. Describing cross-sectional PEB For each theoretical model, modifications to structural model paths were tested to evaluate whether adding any direct paths to PEB helped to explain variance in PEB. For 96 both models, modifications did not improve model fits nor increase the explained variance in PEB. Additionally, a hybrid model combining features of the TPB and NAM did not contribute additional explained variance in PEB beyond the TPB. These results support to the structure of the models and are consistent with prior work that has empirically supported the models (Bamberg, 2003; Kaiser et al., 2003; Kaiser et al., 2007; Schultz et al., 2005; Stern et al., 1999). In the cross-sectional TPB model in the present study, intentions explained 71% of the variance in PEB. This finding is in line with prior work that has found intentions to explain 51-52% of general ecological behavior among a community sample of 823 Swiss adults (Kaiser & Gutscher, 2003), 60% of variance in requesting a brochure about green energy among college students (Bamberg, 2003), and 49-88% of several types of PEB among a community sample of 1,394 German adults (Kaiser et al., 2007). Within the context of the TPB, these results stress the importance of building motivation for PEB for actual engagement in PEB. Additionally, 29% of intentions was explained by attitudes, norms, and perceived control. This figure is considerably lower than in prior work, which has found the same set of variables to account for 64% (Bamberg, 2003), 81% (Kaiser & Gutscher, 2003), and 91% (Kaiser et al., 2007) of the variance in PEB intentions. This may be attributable to different measures used or different study samples. In the NAM cross-sectional model, 21% of the variance in PEB was explained. This is consistent with a prior community-based study that found the NAM to explain approximately 18% of variance in PEB (Stern et al., 1999). However, that the personal norms x awareness of consequences x ascribed responsibility interaction effect in the 97 present study did not predict PEB diverges from the hypothesized model and prior empirical support of it (Schultz et al., 2005). This may be due to differences among the measures used between studies, or perhaps the relatively smaller sample size in the present study. Because this interaction effect has received relatively less attention in the literature than other effects included as part of the NAM, further investigation of it is warranted. Describing change in PEB Results based on time 1 and time 2 data provided preliminary support for the extension of the TPB to addressing change in PEB. Effect sizes among the change model were generally smaller than those in the cross-sectional model, consistent with the fact that variation in time 1 scale scores was generally larger than variation in change. For instance, in the final TPB change model, 44% of the variance in change in PEB and 11% of the variance in change in intentions was explained by the model. In contrast, 70% of the variance in PEB and 29% of variance in intentions was explained in the cross- sectional models. These findings suggest that describing PEB cross-sectionally differs from describing change in PEB. A change model based on the NAM was also evaluated. The model explained 5% of the variance in change in PEB, nearly all of which was accounted for by change in personal norms. This suggests that a change model based on TPB is superior in addressing PEB change relative to one based on NAM constructs. 98 Role of intentions A goal of the present study was to evaluate the relative contributions of changes in intentions relative to other variables in explaining change in PEB. Change in intentions accounted for 44% of the variance in change in PEB and was the most important variable in explaining PEB change by far. The other variables combined only accounted for no more than 4% of the variance in PEB change in the absence of change in intentions. This indicates that changes in distal model constructs, such as social norms, have no appreciable direct effects on changes in PEB. Other variables may mediate the effects of changes in distal model constructs, which has implications for intervention planning and outcome assessment. For instance, if the effect of changing social norms on change in PEB is mediated through change in intentions, then intentions should be assessed to better understand processes of change. That change in intentions was found to be the key contributor to PEB change has other implications for interventions. Our findings suggest that to directly effect change in PEB, interventions should target behavioral intentions. However, intentions may not be directly modifiable by existing intervention strategies. For example, how would home energy feedback influence intentions to reduce air conditioning use? To the authors’ knowledge, this question has not been empirically evaluated, and it is difficult to hypothesize about how feedback might directly change intentions of reducing use of air conditioning, or even monitoring of use. Intentions may be modified indirectly by intervention approaches that change other variables that predict intentions. For example, 99 descriptive feedback may enhance perceived control by showing residents the effects of their behavior on energy use, and changing perceived control may increase intentions. On the other hand, changing intentions may not be the optimal starting point for everyone. The Transtheoretical Model describes a series of stages of readiness for behavior change that may be helpful to consider and incorporate into intervention design (Prochaska et al., 1992). Depending on stage of change, some individuals may need to build up additional motivation prior to being ready to change intentions. Alternatively, others may already have well-developed intentions, in which case efforts should focus on translating those intentions into action. Measurement Conclusions drawn from theoretical studies of PEB reflect the quality of the measures used. In contrast to other areas of psychological research, there is a relative scarcity of well-established, commonly used measures available to test psychological theories applied to PEB. This weakness of the field may reflect the fact that PEB is often conceptualized as a multidimensional construct and as such cannot be straightforwardly addressed by a single instrument. Developing measures tailored to specific types of PEB (e.g., energy use, transit use, recycling) may be useful. Defining and measuring theoretical constructs can be quite a challenge. Therefore, rigorous psychometric evaluation of instruments should be conducted in order to improve self-report tools and the reliability of results of PEB studies. 100 Alternative approaches to assessing behavior change In the present study, participants responded to survey questions regarding their current level of each theoretical construct. Work in other fields suggests that assessing constructs as they pertain to change may be appropriate (Baranowski et al., 2003; Brug et al., 2005). For instance, a study of nutrition among a sample of Dutch adults indicated that participants had positive attitudes, norms, and perceived control regarding eating a low fat diet, but a follow-up study found much less positive attitudes, norms, and perceived control related to reducing fat intake (Conner & Norman, 1996). This suggests that when study goals relate to describing change, different information might gained by making relatively minor changes to item phrasing in order to assess constructs as they pertain to change rather than level (e.g., I intend to reduce my use of air conditioning at home vs. I intend to use less air conditioning at home). Limitations The findings of the present study should be considered in light of several limitations. First, the sample comprised university undergraduate students and may have been biased such that students interested in environmental issues were more likely to participate. Therefore, the results may not generalize to other populations or settings. Second, among all final cross-sectional and change models, CFIs were greater than .80, and RMSEAs were less than .20. Although these fit indices did not reach traditionally accepted levels of adequacy (i.e., CFI>.90, RMSEA<.10), the misfit was not so substantial as to reject the models. The misfit observed may reflect weaknesses in the measures used, which is elaborated upon in the limitations section below. 101 Third, many of the measures used were developed by the author and have not been used among other samples. The measures of intentions, expected outcome, social and personal norms, perceived control, and ascribed responsibility were developed to correspond exactly to the behavior items in the PEB outcome scale. Although internal consistencies of the scales were generally acceptable, the data misfit a 1-factor scale model among many of them (see Appendix 2F). This may be an indication that the scales assessed a heterogeneous set of constructs (e.g., recycling, water use, littering). One avenue for future work is to evaluate whether theoretical models perform differently across distinct types of PEB. Kaiser et al. (2007) tested TPB models for six specific PEB (e.g., recycling glass, involvement in environmental organizations, purchasing organic food), as well as several general models of TPB that included measures of various PEBs. In the six behavior-specific models, the TPB was empirically supported, but relationships among the variables differed across the specific behaviors. This suggests variation in the relative importance of TPB model constructs across distinct types of PEB. Related, another limitation was that whereas the measures developed by the first author corresponded with respect to specific PEBs, the measure of attitudes used (i.e., New Ecological Paradigm) was more general. Similarly, the Sundblad Global Warming Quiz used to assess awareness of consequences did not correspond to the PEB items and responses were scored on a correct/incorrect basis rather than a subjective basis. This measure in particular had relatively low correlations with other measures, and may have contributed to model misfit. 102 A final limitation of the current study is that the intentions-behavior measure covariance was quite high. A meta-analysis of 422 hypotheses examining relationships between intentions and a wide variety of behaviors found a correlation range of .40 - .82 between behavior and intentions (Sheeran, 2002). As well, in a meta-analysis of 15 studies that focused specifically on determinants of PEB, correlations between behavior and intentions ranged from .42-.61 (Bamberg & Moser, 2007). In the present study, behavior-intentions correlations were higher than these ranges (.84-.87), suggesting that the measures used to assess these constructs may have captured what was shared among the constructs rather than two distinct constructs. For this reason, additional analyses used the NAM final model as a framework to test change hypotheses, which did not include intentions as a model construct. As well, in testing the cross-sectional hybrid models, direct paths from the three-way interaction to both the intentions and behavior constructs were tested in order to evaluate relationships that excluded intentions. Strengths To the authors’ knowledge, this is the first study that has evaluated the performance of the TPB in explaining behavior change. This extends existing research that has supported cross-sectional TPB models in explaining PEB. Additionally, the present study yielded information that can be used designing interventions to promote PEB. Future directions To better understand which determinants of behavior change should be targeted in intervention efforts, as well as what intervention tools can effectively modify such 103 determinants, rigorous longitudinal intervention work paired with repeated assessments of behavior and behavioral determinants is needed (Baranowski et al., 1999; Jeffrey, 2004; Sutton, 2000). For example, introducing an intervention that aims to change perceived social norms regarding electricity use via feedback, and assesses electricity use behavior and related perceived social norms across multiple occasions would allow for the examination of the relationship between change in social norms and change in electricity use behaviors. This experimental framework represents a strong approach to elucidating the causal processes related to behavior change related to electricity use, and how such changes are influenced by intervention strategies. 104 Chapter 2 References Abrahamse, W., Steg, L., Vlek, C., & Rothengatter, T. 2005. A review of intervention studies aimed at household energy conservation. Journal of Environmental Psychology, 25, 273-291. Ajzen, I. 2009. Behavioral interventions based on the Theory of Planned Behavior. Retrieved on April 24, 2009, from http://people.umass.edu/aizen/pdf/tpb.intervention.pdf Ajzen, I., & Madden, T.J. 1986. Prediction of goal-directed behaviour: attitudes, intentions, and perceived control. Journal of experimental social psychology, 22, 453- 474. Bamberg, S. 2003. How does environmental concern influence specific environmentally related behaviors? A new answer to an old question. Journal of environmental psychology, 23, 21-32. Bamberg, S., & Moser, G. 2007. Twenty years after Hines, Hungerford, and Tomera: A new meta-analysis of psycho-social determinants of pro-environmental behavior. Journal of environmental psychology, 27, 14-25. Becker, L.J. 1978. Joint effect of feedback and goal setting on performance: A field study of residential energy conservation. Journal of Applied Psychology, 63, 428-433. Baranowski, T., Cullen K.W., Baranowski, J. 1999. Psychosocial correlates of dietary intake: advancing dietary intervention. Annual Review of Nutrition, 19, 17-40. Baranowski, T., Cullen, K.W., Nicklas, T., Thompson, D., Baranowski, J. 2003. Are current health behavioral change models helpful in guiding prevention of weight gain efforts? Obesity Research, 11, 235-245. Brandon, G., & Lewis, A. 1999. Reducing household energy consumption: A qualitative and quantitative field study. Journal of Environmental Psychology, 19, 75-85. Brug, J., Oenema, A., & Ferreira, I. 2005. Theory, evidence, and Intervention Mapping to improve behavior nutrition and physical activity interventions. International Journal of Behavioral Nutrition and Physical Activity, 2:2. Cheung, S.F., Chan, D.K.S., & Wong, Z.S.Y. 1999. Reexamining the theory of planned behavior in understanding wastepaper recycling. Environment and Behavior, 31, 587- 612. 105 Cialdini, R. B., Reno, R.R., & Kallgren, C.A. 1990. A focus theory of normative conduct: Recycling the concept of norms to reduce littering in public places. Journal of Personality and Social Psychology, 58, 1015-1026. Clark, C., Kotchen, M.J., & Moore, M.M. 2003. Internal and external influences on pro- environmental behavior: Participation in a green electricity program. Journal of environmental psychology, 23, 237-246. Conner, M., & Norman, P. 1996. Predicting health behavior: Philadelphia, Open University Press. Dunlap, R.E., Van Liere, K., Mertig, A., & Jones, R.E. 2000. Measuring endorsement of the New Ecological Paradigm: A revised NEP scale. Journal of social issues, 56, 425- 442. Garling, T., Fujii, S., Garling, A., & Jakobsson, C. 2003. Moderating effects of social value orientation on determinants of proenvironmental behavior. Journal of Environmental Psychology, 23, 1-9. Geller, E.S. 1992. Solving environmental problems: A behavior change perspective. In S. Staub & P. Green (Eds.), Psychology and social reasonability: Facing global challenges. New York: New York University Press. Goldstein, N.J., Cialdini, R.B., & Griskevicius, V. 2008. A room with a viewpoint: using social norms to motivate environmental conservation in hotels. Journal of Consumer Research, 35, 472-482. Guagnano, G.A., Stern, P.C., & Dietz, T. 1995. Influences on attitude-behavior relationships: A natural experiment with curbside recycling. Environment and Behavior, 27, 699-718. Hines, J.M., Hungerford, H.R., & Tomera, A.N. 1987. Analysis and synthesis of research on responsible environmental behavior: A meta-analysis. Journal of Environmental Education, 18, 1-8. Hopper, J., & Nielsen, J. 1991. Recycling as altruistic behavior: Normative and behavioral strategies to expand participation in a community recycling program. Environment and Behavior, 23, 195-220. Jeffrey, R.W. 2004. How can health behavior theory be made more useful for intervention research? International Journal on Behavioral Nutrition and Physical Activity, 23, 1-10. 106 Kaiser, F.G., & Gutscher, H. 2003. The proposition of a general version of the theory of planned behavior: Predicting ecological behavior. Journal of Applied Social Psychology, 33, 586-603. Kaiser, F.G., Wolfing, S., & Fuhrer, U. 1999. Environmental attitude and ecological behaviour. Journal of environmental psychology, 19, 1-19. Kaiser, F. G., Schultz, W.P., & Scheuthle, H. The theory of planned behavior without compatibility? Beyond method bias and past trivial associations. Journal of Applied Social Psychology, 37, 1522-1544. Latham, G.P., & Locke, E.A. 1991. Self-regulation through goal setting. Organization Behavior and Human Decision Processes, 50, 212-247. McCalley, M.T., & Midden, C.J.H. 2002. Energy conservation through product- integrated feedback: The roles of goal-setting and social orientation. Journal of Economic Psychology, 23, 589-603. McClelland, L, & Cook, S.W. 1980b. Promoting energy conservation in master-metered apartments through group financial incentives. Journal of Applied Social Psychology, 10, 20-31. Midden, J.F., Weenig, M.H., & Zieverink, H.J.A. 1983. Using feedback, reinforcement, and information to reduce energy consumption in households: A field experiment. Journal of Economic Psychology, 3, 65-86. Muthen, L.K., Muthen, B., 2006. M-Plus: The Comprehensive Modeling Program for Applied Researchers Version 5.21 [Computer software]. Muthen and Muthen, Los Angeles, CA. Pallak, M.S., Cook, D.A., & Sullivan, J.J. unpublished. The effects of comparison feedback on electricity usage. University of Iowa. Paulhus, D.L. 2002. Socially desirable responding: The evolution of a construct. In H. Braun, D.N. Jackson, & D.E. Wiley (Eds.), The role of constructs in psychological and educational measurement. Hillsdale, NJ: Erlbaum. Petersen, J.E., Shunturov, V., Janda, K., Platt, G., & Weinberger, K. 2007. Dormitory residents reduce electricity consumption when exposed to real-time visual feedback and incentives. International Journal of Sustainability in Higher Education, 8, 16-33. Pitts, S.M., & Wittenbach, J.L. 1981. Tax credits as a means of influencing consumer behavior. Journal of Consumer Research, 8, 335-338. 107 Prochaska, J.O., DiClemente, C.C., & Norcross, J.C. 1992. In search of how people change: Applications to addictive behaviors. American Psychologist, 47, 1102-1114. SAS Institute Inc. SAS 9.2. Cary, NC: SAS Institute, 2002. Schultz, P.W., Gouveia, V.V., Cameron, L.D., Tankha, G., Schmuck, P., & Franek, M. 2005. Values and their relationship to environmental concern and conservation behavior. Journal of cross-cultural psychology, 36, 457-475. Schwartz, S.H. 1994. Are there universal aspects in the structure and contents of human values? Journal of social issues, 50, 19-45. Seligman, C., & Darley, J.M. 1977. Feedback as a means of decreasing residential energy consumption. Journal of Applied Psychology, 62, 363-368. Sheeran, P. 2002. Intention-behavior relations: A conceptual and empirical review. In Stroebe, W., & Hewstone, M. (Eds.), European Review of Social Psychology (V. 12). New York :John Wiley and Sons. Stern, P. 1992. What psychology knows about energy conservation. American Psychologist, 47, 1224-1232. Stern, P.C., Dietz, T., Abel, T., Guagnano, G.A., & Kalof, L. 1999. A value-beleif-norm theory of support for social movements: The case of environmentalism. Human Ecology Review, 6, 81-97. Stern, P.C., Dietz, T., Kalof, L., & Guagnano, G.A. 1995. Values, beliefs, and proenvironmental action: Attitude formation toward emergent attitude change. Journal of Applied Social Psychology, 25, 1611-1636. Stutzman, T.M., & Green, S.B. 1982. Factors affecting energy consumption: Two field tests of the Fishbein-Ajzen model. Journal of social psychology, 117, 183-201. Sundblad, E-L., Biel, A., & Garling, T. 2007. Cognitive and affective risk judgments related to climate change. Journal of Environmental Psychology, 27, 97-106. Sutton, S., 2000. Interpreting cross-sectional data on stages of change. Psychological Health, 15, 163-171. Syme, G.J., Nancarrow, B.E., & Seligman, C. (2000). The evaluation of information campaigns to promote voluntary household water conservation. Evaluation Review, 24, 539-578. 108 Van Houwelingen, J. H., & Van Raaij, W.F. 1989. The effect of goal-setting and daily electronic feedback on in-home energy use. Journal of Consumer Research, 16, 98-105. Van Liere, K., & Dunlap, R. 1978. Moral norms and environmental behavior: An application of Schwartz’s norm-activation model to yard burning. Journal of Applied Social Psychology, 8, 174-188. Winett, R. A., and Ester, P. 1983. Behavioral science and energy conservation. Conceptualizations, strategies, outcomes, energy policy applications. Journal of Economic Psychology 3: 203-229. Winett, R.A., Kagel, J.H., Battalio, R.C., & Winkler, R.C. 1978. Effects of monetary rebates, feedback, and information on residential electricity conservation. Journal of Applied Psychology, 63, 73-80. 109 Chapter Three: The Influence of Social Desirability and Item Priming Effects on Reports of Pro-environmental Behavior Chapter 3 Abstract Background. Several key theories used to understand pro-environmental behavior (PEB) include both future behavioral intentions and past behavior as model constructs. The study of these constructs relies heavily on the use of self-report methodology, which is subject to a variety of method biases, including social desirability and item priming effects. In particular, impression management (IM) is a subtype of social desirability that is typically associated with exaggerated reports of prosocial behavior and qualities. Available research has found less evidence of IM bias when questions about future intentions are presented prior to questions about past prosocial behavior (Intentions- Behavior ordering), compared to when individuals respond to questions about recent behavior prior to future intentions (Behavior-Intentions ordering; Beebe et al., 2008; Johnson et al., 2005). Methods. The present study investigated IM bias and item order effects among 254 university students who responded to an online survey. The survey was part of a larger study (see Paper 1) that provided for the assessment of PEB and related intentions. The current study focused on data from five blocks of items: two behavior blocks, two intentions blocks, and one expected outcomes block. These blocks were administered to participants in random order to provide for the investigation of item order effects. The survey also assessed IM. 110 Results. Higher IM scores predicted higher scores on both behavior subscales. Regression analyses yielded weak evidence of item order effects, whereby individuals receiving the Intention-Behavior ordering had significantly higher scores on one behavior subscale than individuals with the Behavior-Intention ordering. This effect was observed for only one of the two PEB blocks, and was mediated by other effects in a multiple regression model. Order effects did not moderate the relationship between IM and behavior. Conclusions. The positive association between IM and PEB is consistent with the existing literature, indicating that individuals tend to exaggerate self-reports of prosocial behavior. The finding that responding to intentions items first predicted higher behavior scores contrasts with prior work which has found order effects in the opposite direction. We recommend that IM bias be considered in designing surveys, conducting analyses, and interpreting study findings. 111 Chapter 3 Background Pro-environmental behavior (PEB) is defined here as any behavior that supports the sustainability of natural ecosystems, environmental health, and “contribute[s] towards environmental preservation and/or conservation” (Axelrod & Lehman, 1993, p. 153). Under this umbrella term are behaviors such as recycling, electricity, gas, and water consumption, and donation of time and money to environmental organizations. According to several theoretical models applied to PEB, including the Theory of Planned Behavior (Ajzen & Madden, 1986), the Model of Interpersonal Behavior (Triandis, 1980) and Social Learning Theory (Bandura, 1997), intentions have been found to be the strongest predictors of behavior (Bamberg, 2003; Chapman, 2001; Cheung et al. 1999; Kaiser et al. 1999; Kaiser et al., 2007; Stern et al. 1995; Stutzman & Green 1982). Evaluations of these conceptual models have generally been cross-sectional and have used self-report surveys to assess past behavior and future behavioral intentions on the same measurement occasion. It is widely recognized that self-report measures can be subject to a variety of method biases, including those stemming from socially desirable response patterns and item context effects (for reviews see Podsakoff et al., 2003 and Podsakoff & Organ, 1986). These issues have not been widely examined in the context of PEB. For many psychological variables involved in the study of PEB, including intentions and past behavior, self-report remains the most practical method of assessment. It is therefore important to investigate and refine self-report instruments. An exhaustive review of 112 method biases is beyond the scope of this paper. Here we summarize two biases with implications for the study of PEB. Socially desirable responding. Socially desirable responding (SDR) is a widely recognized issue in self-report research. Several scales have been developed to detect socially desirable response patterns (e.g., Crowne & Marlowe, 1964; Paulhus, 2002), and a recent review identified more than 2,500 published studies across disciplines that have used scales of SDR as a measure of response style or to validate self-report instruments (Uziel, 2010). SDR refers to the tendency of the respondent to provide responses that paint her or him in a favorable light (Crowne & Marlowe, 1964). Conceptually, this includes under-reporting socially undesirable behaviors (e.g., alcohol consumption; Davis et al., 2009) and over-reporting socially desirable behaviors and qualities (e.g., emotional stability; Borkenau et al., 2009). Whereas the negative associations between social desirability and socially stigmatized behavior are well established, relatively less empirical work has examined relationships among social desirability and prosocial behavior, making this a target for investigation. SDR has been empirically divided into two factors, the first of which is self- deception, reflecting the unconscious tendency to regard and describe oneself positively (Paulhus, 1984; Paulhus, 2002). Self-deception questions relate to psychologically uncomfortable but presumably universal experiences that are difficult for respondents to disconfirm (e.g.,“I always know why I like things”), and that self-deceivers tend to deny, because they genuinely believe these not to be true for them (Paulhus, 2002). The other factor is impression management (IM), which reflects respondents’ conscious efforts to 113 deceive others. IM scales include questions about discrete behaviors (e.g., “I always obey laws, even if I’m unlikely to get caught”), so respondents’ true engagement in such acts is presumably more consciously accessible and easier to disconfirm (Paulhus, 2002; Uziel, 2010; Zerbe & Paulhus, 1987). IM is a key focus of the present study. The nature of IM has been debated, primarily because some studies have conceptualized IM as a stable personality trait that reflects a need for acceptance or approval from others, and not necessarily a threat to the validity of research results (Crowne & Marlowe, 1964; Uziel, 2010). Under such circumstances, IM may represent content variance and controlling it may remove the predictive power of a measure (Zerbe & Paulhus, 1987). Determining when to treat IM as a method bias is therefore important, and it has been suggested that IM be considered contamination only when the construct it represents is unrelated conceptually to the constructs of interest (Zerbe & Paulhus, 1987). Item priming effects. Item context effects are another type of method bias, and refer to the properties attributed to an item solely due to its proximal relation to other items in an instrument (Wainer & Keily, 1987). Item priming effects are a particular type of item context effect, and are observed when responding to one particular set of items influences responses to another set (Salancik, 1984; Salancik & Pfeffer, 1977). For instance, if in an individual is asked about her attitudes towards energy conservation, and is subsequently asked whether she intends to engage in behaviors to conserve energy in the future, she is likely to respond to the latter query using information made salient by the initial responses regarding energy conservation. This can artificially elevate the covariation among constructs. 114 The strength of IM and item priming biases may vary based on the structure of a self-report instrument. For instance, a study by Beebe et al (2008) varied question order to evaluate method bias in a survey about cancer screening administered to 752 adults aged 50 and older. Individuals who were administered the Past Behavior-Future Intentions question ordering were 1.8 times more likely to report they had been screened (71%) than those who were administered Future Intentions-Past Behavior ordering (58%). Another study found that relative to Past Behavior-Future Intentions ordering, Future Intentions-Past Behavior ordering decreased the level of endorsement and increased the accuracy (as confirmed by medical records) of self-reports of past health promotion screening (i.e., papanicolaou screening, mammograms, and gynecological examinations) among a community sample of 580 women aged 50 and older (Johnson et al., 2005). In regression analyses including covariates such as interview mode, however, this effect was not significant (OR=1.47, 95% CI=0.98-2.2). The authors speculated that an initial inquiry about future intentions relieved respondents in the Future Intentions – Past Behavior group of social pressure to report engagement in past socially desirable behavior, but this was not evaluated statistically. Rationale for present study The results obtained by Beebe et al. (2008) and Johnson et al. (2005) were based on samples of adults aged 50 and older and pertain to health screening. It is important to know whether these findings generalize to other populations and topics, such as PEB, the study of which relies heavily on the use of self-report, and often assesses intentions and behavior at the same occasion. Understanding these biases as they relate to PEB is 115 important not only for the validity of individual study conclusions, but also for the validity of conceptual models that include intentions and behavior constructs. Such models rely on the accumulation of evidence across studies that use common methods and are therefore subject to similar biases. In addition, to the authors’ knowledge, the existence of an interaction between item priming effects and IM has not been previously studied. For instance, if IM effects varied as a function of item order, this would have implications for survey design and the validity of self-report instruments. Study aims and hypotheses The present study aims to evaluate the effects of IM and item priming related to scales that assess future PEB intentions and past PEB. Several hypotheses are evaluated. H1. A main effect of IM will be observed such that IM will positively predict self- reported PEB and other prosocial behavior. H2. A main effect of block order is predicted such that those administered Intentions-Behavior question ordering will have higher behavior scores relative to those administered Behavior-Intentions ordering. In contrast to findings of prior studies, we predict this outcome because consistent with cognitive dissonance theory (Festinger, 1957), if a respondent has already indicated an intention to engage in a behavior, and is then queried about her or his recent participation in the behavior, to avoid dissonance, s/he may over-report past behavior to be consistent with the reported intention. H3. The inference made by Beebe et al. (2008) and Johnson et al. (2005) that Intentions-Behavior ordering leads to more accurate (i.e., lower) behavior scores due to reduced social pressure will be examined indirectly as permitted by our data. Specifically, 116 we will investigate block order as a potential moderator of IM. IM is treated as a proxy for social pressure here. If the reduced social pressure hypothesis is supported, then the positive relationship between IM and behavior should be stronger among those administered Behavior-Intentions ordering relative to those administered Intentions- Behavior ordering. However, findings in the opposite direction represent another possibility: Intentions-Behavior ordering may actually amplify the effect of IM on behavior scores. We reason that consistent with cognitive dissonance, Intentions- Behavior ordering can introduce additional pressure, social or intrapersonal, to exaggerate reported behavior, and thereby amplify the effect of IM. If this hypothesis is supported, block order will moderate the effect of IM on behavior such that the positive effect of IM on behavior score will be greater among those administered the Intentions- Behavior ordering relative to those administered the Behavior-Intentions ordering. Chapter 3 Method Participants completed a self-report survey regarding PEB and key constructs derived from the Theory of Planned Behavior and the Norm Activation Model. The survey was part of a larger study involving a subsequent intervention intended to reduce energy use. Data from the intervention portion of the study are reported in Paper 1. Participants To be eligible to participate in the study, participants had to be at least 18 years of age, a current USC undergraduate student, and a resident of USC-owned or USC- managed housing. A total of 302 students participated in the initial survey. A total of 291 responses were available after eliminating 11 cases that suggested random or careless 117 responding (see Materials section below for more information). Responses from 31 individuals who did not provide data on the impression management scale were excluded from analyses as this was a key variable in our study. Among the remaining 260 respondents, missing scale items were imputed among respondents with 10% or fewer items missing on a scale. Those with more than 10% of items missing on a scale were given a missing score for that scale. A final sample size of 254 participants with complete data on all key variables was retained in subsequent analyses. Approximately 77% of the sample was female, and just over half of participants were university freshmen. Parental education level was high in this sample. For 66% and 71% of participants, maternal and paternal education levels, respectively, were 4-year college degrees or higher. See table 3.1 for additional characteristics of participants. Materials The survey provided for the assessment of the following variables. Mean scores and internal consistencies of key variables are reported in table 3.1. Demographic data. Demographic variables included age, gender, year of university enrollment, and urbanicity of region of origin (urban, suburban, or rural). To approximate parental socioeconomic status, parental educational level was assessed. PEB. Frequency of engagement in a variety of PEB for the two weeks preceding the survey was assessed using a 17-item version of the Schultz Proenvironmental Behavior Scale (Schultz et al., 2005). Five items were added by the first author to the original 12-item scale to include behaviors particularly relevant to college students (e.g., turning off a computer when not in use). Responses were provided on a 5-point Likert 118 scale with response options ranging from “never” to “very often”. Items were scored 1 to 5, with a possible scale score range of 17-85. The scale is provided in Appendix 3A. Behavioral intentions. Behavioral intentions were assessed using a 17-item scale developed by the first author, with items reflecting the same behaviors assessed by the modified Schultz PEB scale. The stem question was developed by Bamberg (2003). Respondents rated their likelihood of engaging in each behavior in the next two weeks on a 5-point Likert scale ranging from “likely” to “unlikely”. Items were scored 1 to 5, with a possible scale score range of 17-85. The scale is provided in Appendix 3A. Outcome expectancies. Based on guidelines provided by Ajzen (2009), a 17-item scale was developed by the first author to assess expected outcomes associated with engaging in the behaviors included as part of the modified Schultz behavior scale. Responses were provided on a 7-point Likert scale with response options ranging from “very bad” to “very good”. Items were scored 1-7, with a possible scale score range of 17-119. Socially desirable responding. The Impression Management scale from the Balanced Inventory of Desirable Responding Impression Management Scale (BIDR-IM) was included to assess level of impression management (Paulhus, 2002). The scale included 12 items that asked about engagement in various behaviors (e.g., “When I was young, I sometimes stole things”; “I have done things that I don’t tell other people about”). Answers indicated the extent to which respondents believed the statements to be true for them, and were provided on a 7-point Likert scale ranging from “definitely false” to “definitely true”. Items were scored 1-7, with a possible scale score range of 12-84. 119 Random Responding. One true/false question was included in each survey to detect random responding. It stated “George W. Bush is the current president of the United States”. As the survey was administered in 2009, the correct answer was false. Additional behavior scales. Two behavior scales were included to examine IM biases across additional behavior constructs. Kaiser’s Scale of General Prosocial Behavior (Kaiser, 1998) included six dichotomous (1 = yes; 0= no) items regarding engagement in general pro-social behavior, such as visiting a sick friend in the hospital. The mean score was 10.49 (SD=0.78, range = 7-12). The Consumer Behavior Subscale from Stern’s Pro-Environmental Behavior Scale assessed frequency of environmentally relevant purchasing (e.g., purchasing organic produce; Stern et al., 1999). Four items were answered on a 5-point Likert scale ranging from “never” to “always”. Items were scored 1 to 5, with a possible scale score range of 4-20. The mean score was 11.63 (SD = 2.79). 120 Table 3.1. Participant characteristics (n=254). Characteristic Mean age in years (SD) 19.0 (1.2) Sex (% female) 77.2 Number of semesters in residency at current residence hall 1 2 3 4 or more (%) 76.8 2.8 14.2 6.3 Class status Freshman Sophomore Junior Senior (%) 52.2 28.5 12.3 7.1 Urbanicity of place of origin Rural Suburban Urban (%) 3.2 77.2 19.7 Key variables Mean (SD) Range α a PEB BEHAV1 BEHAV2 49.4 (8.4) 24.4 (5.2) 25.0 (4.3) 24-79 11-45 11-35 .72 .60 .54 Intentions INTENT1 INTENT2 55.7 (9.1) 28.2 (5.3) 27.6 (4.7) 25-83 12-44 12-40 .80 .72 .63 EXPECT 100.7 (13.2) 41-119 .89 BIDR-IM 45.4 (10.5) 19-74 .71 a Cronbach’s alpha Procedures Survey and recruitment. Beginning in September 2009, the survey was accessible online to participants 24 hours per day for approximately eight weeks. It required approximately 25 minutes to complete. Recruitment strategies for the survey included emails, poster advertising, the social networking website www.Facebook.com, the 121 psychology subject pool, and contact with leaders of student groups, such as student government. All participants were entered into raffle drawings for their participation to receive a variety of prizes, ranging in value from $10-$300. Additionally, participants enrolled in participating Psychology Department courses received credit for participating in the survey that could be applied towards courses. Survey item sequencing. The first section of survey items assessed demographic variables. Next was a section that contained five blocks of items, which were the focus of the present study: two behavior blocks, two intentions blocks, and one outcome expectancies block. To provide for the examination of specific item content effects, the behavior scale was divided into two separate subscales of items (i.e., BEHAV1 and BEHAV2), as was the intentions scale (i.e., INTENT1 and INTENT2). The BEHAV1 and INTENT1 blocks contained corresponding item content, as did the BEHAV2 and INTENT2 blocks. The expected outcomes scale (EXPECT) was administered in its entirety. The scale items can be seen in Appendix 3A. Each participant was assigned to a particular ordering of these five blocks based on the third letter of his or her email address. For instance, participants who indicated the letter a, b, or, c were administered the blocks in the following order: INTENT1, BEHAV1, EXPECT, INTENT2, BEHAV2. Participants with letters d, e, or f were administered the blocks in the following sequence: BEHAV1, INTENT1, EXPECT, BEHAV2, INTENT2. Participants were administered each block one at a time, and once they had completed the items within a block they could not return to previous blocks. See Table 3.2 for details on block sequencing. Third letter of email address was selected as 122 the randomization variable because email address was collected as the contact information and would be constant across study waves (we wanted to keep order block order administration constant and no other identifying information was obtained). There was no basis to expect that the administration sequence would be related to the variables of interest (behavior and intentions). This was confirmed by preliminary analysis of variance analyses (0.52<F<1.18, .31<p<.82). Table 3.2. Survey block sequencing. Block order administration Third letter of email address N (%) Sex (% male) Mean age (SD) 1 2 3 4 5 a, b, c 33 (13) 12 18.8 (1.1) INTENT1 BEHAV1 EXPECT INTENT2 BEHAV2 d, e, f 34 (13) 18 19.2 (1.2) BEHAV1 INTENT1 EXPECT BEHAV2 INTENT2 g, h, i 32 (13) 16 19.1 (1.0) INTENT1 BEHAV1 INTENT2 BEHAV2 EXPECT j, k l 25 (10) 16 19.1 (1.6) BEHAV1 INTENT1 BEHAV2 INTENT2 EXPECT m, n, o 51 (20) 35 19.0 (1.3) INTENT1 BEHAV2 EXPECT INTENT2 BEHAV1 p, q, r 27 (11) 35 18.9 (1.1) BEHAV1 INTENT2 EXPECT BEHAV2 INTENT1 s, t, u 37 (15) 23 19.0 (1.2) INTENT1 BEHAV2 INTENT2 BEHAV1 EXPECT v, w, x, y, z 15 (6) 57 18.9 (1.3) BEHAV1 INTENT2 BEHAV2 INTENT1 EXPECT Analyses BEHAV1 and BEHAV2 were treated as separate outcomes and a set of dummy- coded predictor variables was created for each. For each behavior block, a dichotomous variable called Block Order was created to represent block ordering for the corresponding 123 intention blocks (1 = Intentions-Behavior, 0 = Behavior-Intentions). To adjust for possible intrusive effects of the presentation of new material on subsequent responses, a dichotomous (yes = 1; no = 0) variable called Intervening Blocks was created to represent whether any blocks, such as the outcome expectancies block, had appeared between corresponding behavior and intentions blocks for each outcome. Finally, to adjust for possible effects of fatigue, two variables (one for each outcome) called Preceding Blocks were created to code for the number of blocks that preceded each behavior block among the series of the five key blocks focused on in this study. The SAS statistical package version 9.1 was used for analyses (SAS Institute Inc., 2002). Preliminary analyses examined the associations of age, sex, urbanicity of place of origin, parental education levels, length of residence in residence hall, and year of university enrollment with the Block Order variables and the outcomes. Using regression modeling, we evaluated our hypotheses regarding the influences of IM and item order effects on self-reported PEB. Scores on the BEHAV1 and BEHAV2 blocks were evaluated separately as the outcomes. First, each outcome was regressed on four variables in separate univariable regression models: BIDR-IM, Block Order, Preceding Blocks, and Intervening Blocks. The regression of the outcomes on BIDR-IM scores and Block Order provided for a test of Hypothesis 1 and partial test of Hypothesis 2, respectively. Correlations between BIDR-IM scores and two additional behavior scales were also examined to inform Hypothesis 1. To evaluate Hypothesis 2 while adjusting for other effects, and to test Hypothesis 3, a stepwise regression model building procedure was used to model each outcome 124 based on a set of predictors. On the first step, BIDR-IM score was entered as a predictor (as was a quadratic BIDR-IM function for BEHAV2). On the second step, age was entered. Preceding Blocks was also entered on the second step as a covariate for BEHAV1. On the third step, Block Order was added. On the fourth and final step, an interaction term (BIDR-IM x Block Order) was added. Hypothesis 3 was also tested using general linear modeling. Each outcome was regressed on Block Order, which was treated as a class variable, BIDR-IM scores, and an interaction of the two terms. Planned contrasts tested whether the regression coefficient of the interaction term differed significantly across the levels of Block Order. The contrasts were not significant (p BEHAV1 < .17 and p BEHAV2 <.60) and therefore these analyses are not reported upon here. Chapter 3 Results Preliminary analyses examined the associations of age, sex, urbanicity of place of origin, parental education levels, length of residence in residence hall, and year of university enrollment with the Block Order variables and the outcomes. None of these demographic variables were associated with Block Order. Younger age and later year of enrollment at the university were significantly associated with higher BEHAV1 (β age =- 0.15, p<.02, t=12.47; β enroll =0.13, t=2.09, p<.04) and higher BEHAV2 scores (β age =-0.22, t=-3.58, p<.0004; β enroll =0.12, t=1.99, p<.05). Age and enrollment year were highly correlated (r=-.66, p<.0001), and the effect of age was stronger than enrollment year in multiple regression models that regressed each outcome outcomes on both variables simultaneously. Therefore, age was retained as a covariate for both outcomes in subsequent analyses. 125 Hypothesis 1: BIDR-IM scores will positively predict behavior scores. In the univariable models, higher BIDR-IM scores significantly predicted higher scores on both the BEHAV1 and BEHAV2 outcomes. See table 3.3 for parameter estimates. Preliminary analyses suggested a possible nonlinear relationship between BIDR- IM scores and the outcomes. To address this, the BIDR-IM scale score was squared and centered to determine whether a nonlinear function better described the data. The quadratic BIDR-IM term contributed an additional 1.1% to the explained variance in BEHAV2 after accounting for BIDR-IM, and was retained in subsequent models for this outcome. See tables 3.3 and 3.4b for details. However, for BEHAV1, after accounting for the linear BIDR-IM term, the quadratic term contributed only an additional 0.02% in explained variance, and therefore was not retained in subsequent BEHAV1 models. IM and additional behavior scales. Pearson correlations between IM scores and scores on a scale of general prosocial behavior (r=-.03, p<.67) and consumer behavior (r=.11, p<.10) were not significant. This indicates that among the behavior outcomes included in the survey, IM was related only to the general PEB outcome and not to other measures of prosocial behavior. Hypothesis 2: Intentions-Behavior ordering will predict higher behavior scores. Results from the univariable regression models are displayed in table 3.3. Block order. A main effect of Block Order was found for the BEHAV1 subscale. Intentions-Behavior ordering predicted significantly higher scores on the behavior scale 126 relative to Behavior-Intentions ordering. No Block Order effect was observed for BEHAV2. Intervening blocks. No significant effect of Intervening Blocks was observed for either the BEHAV1 or BEHAV2 outcome. Preceding blocks. A greater number of blocks preceding the BEHAV1 block significantly predicted higher BEHAV1 scores, suggesting that BEHAV1 scores were influenced by amount of preceding material. Preceding Blocks was retained as a covariate for further BEHAV1 analyses to adjust for this effect. Number of blocks preceding BEHAV2 was not significantly associated with BEHAV2 scores. Table 3.3. Univariable regression results. Predictor Outcomes BEHAV1 BEHAV2 BIDR-IM 0.24 P<.0001 0.23 P<.0001 Block order a 0.14 P<.03 0.01 P<.85 Intervening blocks 0.08 P<.20 -0.04 P<.49 Preceding blocks 0.15 P<.02 -0.04 P<.51 Table note. Standardized regression coefficients are reported. BIDR-IM=Balanced Inventory of Desirable Responding Impression Management score. a Block order coding: 1=intentions block administered before corresponding behavior block, 0=behavior block before corresponding intentions block. Hypothesis 3: interaction of IM and block order. In the final model accounting for age and Preceding Blocks, the effect of Block Order on BEHAV1 was reduced (relative to the univarible model) and no longer significant. Block Order did not have a 127 significant effect on BEHAV2 scores. There was no evidence to support an interaction between Block Order and IM for either outcome. These results fail to support the hypothesis that the influence of IM varies as a function of order effects. In the final models, younger age was significantly associated with higher scores on both outcomes. Additionally, higher BIDR-IM scores predicted higher BEHAV2 scores. See tables 3.4a and 3.4b for model parameters. Table 3.4a. Hypothesis 2: BEHAV1 models, n=254. Model I Model II Model III Model IV F 15.52 P<.0001 8.58 P<.0001 6.41 P<.0001 5.67 P<.0001 BIDR-IM 0.24 P<.0001 0.22 P<.0004 0.22 P<.0004 0.09 P<.34 Preceding blocks 0.12 P<.05 0.12 P<.24 0.12 P<.24 Age -0.15 P<.02 -0.15 P<.02 -0.15 P<.02 Block order 0.001 P=.99 -0.42 P<.14 BIDR-IM x Block Order 0.47 P<.12 Model R² .058 .093 .093 .103 Table note. Standardized beta weights are reported. 128 Table 3.4b. Hypothesis 2: BEHAV2 models, n=254. Model I Model II Model III Model IV Model V F 9.50 P<.0001 11.00 P<.0001 8.22 P<.0001 7.17 P<.0001 5.60 P<.0001 BIDR-IM -0.50 P<.25 -0.57 P<.18 -0.58 P<.18 0.24 P<.003 -0.90 P<.11 BIDR-IM² 0.75 P<.09 0.82 P<.06 0.82 P<.06 1.15 P<.04 Age -0.22 P<.0004 -0.22 P<.0004 -0.21 P<.0006 -0.22 P<.0004 Block order 0.004 P<.96 0.03 P<.92 -1.57 P<.38 BIDR-IM x Block Order -0.04 P<.90 1.71 P<.38 BIDR-IM²* Block Order -0.51 P<.36 Model R² .070 .117 .117 .103 .120 Table note. Standardized beta weights are reported. General construct bias vs. specific item bias Given that the univariable models indicated block order effects for the BEHAV1 and not the BEHAV2 outcome, additional analyses were conducted to examine possible effects of specific item content. Item order effects, or in this study, block order effects, may function on the level of a general construct, whereby administering questions about PEB contributes to biased responding on subsequent PEB questions that contain different but related content (e.g., INTENT1 followed by BEHAV2). Alternatively, block order effects may be limited to the content of specific items, such that bias is only observed for items with corresponding content (e.g., INTENT1 followed by BEHAV1). To address this question, we examined data from responses to the first two blocks that participants completed, which included an intentions block and a behavior block for each individual. The outcome was a behavior variable that represented the score of whichever behavior 129 block appeared in those first two blocks (Behavior = BEHAV1 or BEHAV2). A new dichotomous variable called Match was created to represent whether those first two blocks had matching item content (yes = 1; e.g., BEHAV1, INTENT1) or different item content (no = 0; e.g., BEHAV1, INTENT2). The new behavior outcome was then regressed on Match, Block Order, covariates (i.e., BIDR-IM and age), and interaction between match and block order. A significant main effect of Block Order was observed in Model II, though this should be interpreted with caution as the overall model test was not significant. The effect was consistent with the findings of hypothesis 2 and indicated that Intentions-Behavior ordering predicted higher behavior scores. In the final model, Block Order was not significant, though the effect size was roughly the same as in Model II. This lends some support to order bias at the level of the general construct of PEB. In the final model, IM significantly predicted higher scores, as did younger age. The interaction between matching content and block order was not significant, failing to provide evidence for bias at the item content level. See table 3.5 for model parameters. 130 Table 3.5. Model parameters for order effects at the level of the general construct vs. specific item content, n=254. Model I Model II Model III Model IV F 1.56 P<.22 2.86 p<.06 5.44 P<.0003 4.43 P<.0007 Match -0.08 P<.22 -0.06 P<0.36 -0.06 P<.36 -0.01 P<.94 Block order 0.13 P<.05 0.09 P<.14 0.14 P<.14 Age -0.14 P<.03 -0.14 P<.03 BIDR-IM 0.20 P<.002 0.20 P<.002 Match x Block Order -0.07 P<.52 Model R² 0.006 0.022 0.080 0.082 Table note. Standardized beta weights are reported. Chapter 3 Discussion This study evaluated the influence of IM and order effects on self-reported PEB. To our knowledge, this is the first study that has explored these biases among scales that assess PEB. Our hypotheses received mixed support. Supporting our first hypothesis, IM positively predicted scores on both behavior outcomes across several of the models tested. This finding adds to the relatively limited base of evidence that has found positive associations between IM scores and self-reported prosocial behavior (Borkenau et al., 2009; Fernandes & Randall, 1992). Interestingly, however, the present study did not find significant associations between IM and self-reported prosocial behavior or consumer behavior. Significant associations between these constructs and IM would be expected if participants had a general response tendency to present themselves in a positive light, or 131 if IM represented content variance in the context of our study. The lack of such associations suggests that participants may have been aware of the goals of the larger study and adopted a response style that aligned with these goals. These results point to IM as unwanted error variance rather than meaningful trait variance in the context of our study. We therefore recommend the use of IM scales to control for contamination in research on PEB when participants are aware of study goals. Our second hypothesis regarding item order effects received limited support. Univariable regression analyses found a main effect of block order for one behavior outcome such that those administered the Intentions-Behavior ordering had significantly higher behavior scores relative to those administered the Behavior-Intentions ordering. A similar, though nonsignificant, effect of block order was observed in exploratory models that regressed behavior score on block order irrespective of corresponding item content. The direction of the observed effect contrasts with prior work that has found that Intentions-Behavior ordering yields more accurate (i.e., lower) reports of prosocial behaviors (Beebe et al., 2008; Johnson et al., 2005). These conflicting results may be attributable to the fact that the previous work has been conducted with older populations, whereas the present study sample consisted of college students. Alternatively, order effects may behave differently for scales, which were used in the present study, versus individual items, used in prior studies. Although the present study found evidence of order effects in a univariable model, this effect did not contribute to the explained variance in the outcome when other predictors were added to the model. Additionally, order effects were not observed for the 132 second behavior outcome evaluated in this study. Taken together, our findings suggest that order effects did not have a strong, consistent influence on self-reported PEB among the scales used in the present study. Similarly, Johnson et al. (2005) observed a significant item order effect in a bivariate model of concordance, but the effect was no longer significant after adjusting for other predictors (e.g., interview mode, barriers to screening) in a logistic regression model. As well, the order effect was observed only for papanicolaou screening, not for mammograms or complete physical or gynecologic exams. Beebe et al. (2008) observed order effects for mail surveys but not for telephone surveys based on the same items. In sum, order effects may depend on type of behavior being assessed and/or may vary by survey mode. The lack of consistent order effects in the context of the present study suggests that this bias may not pose a considerable threat to the results of self-report research related to PEB. Our findings did not indicate that order effects moderate the influence of IM on behavior scores. This study therefore failed to support the social pressure hypothesis posed by Beebe et al. (2008) and Johnson et al. (2005) that Intentions-Behavior ordering reduces the effect of IM on behavior. Neither did it support our hypothesis regarding Intentions-Behavior ordering amplifying the effect of IM on behavior. This suggests that IM may operate independently of item order. If Intentions-Behavior ordering results in higher behavior scores by inducing cognitive dissonance, which conceptually relates to self-deception, then item order effects may interact with self-deception. Future work should examine this to better understand the processes underlying order effects. 133 The present study also investigated the effect of amount of content preceding behavior block on behavior score. In univariable analyses for the BEHAV1 outcome, higher behavior scores were predicted by greater numbers of blocks preceding the behavior block for one of the outcomes. In other words, the later in the survey that behavior block appeared, the higher the behavior score. This contrasts with prior work that has observed higher scores among initial survey items regardless of item content (Ferber, 1952). This finding should be interpreted with caution for several reasons. First, the effect of preceding content was observed for only one of the outcomes. Next, in the final model testing Hypothesis 2, the effect was not significant, but the effect size was roughly similar to the univariable model (β = 0.12 vs. 0.15). Although this item context effect was not a focus of the present study, it aids in the interpretation of our findings on order effects. For instance, as participants completed additional items about PEB, a prosocial set of behaviors, it is possible that they became increasingly aware of the study goals and their tendency to provide socially desirable responses also increased. Alternatively, our findings may be related to cognitive dissonance: as participants completed additional items related to PEB, they became more motivated to report behaviors that were consistent with the attitudes, intentions, or behaviors they had just reported. Changes in level of IM as participants moved through survey were not measured, but this represents an avenue for future research. Limitations Several limitations of this study should be noted. First, the study participants were undergraduates at a private university, so the results may not generalize to other 134 populations. The sample may also have been more cooperative than others. To evaluate this possibility, BIDR-IM cutoff scores, defined as the number of items with a score of 6 or 7, among the present sample were compared to those among the sample of 338 University of British Columbia undergraduates upon which the instrument norms were based (Paulhus, 2002). Mean cutoff scores among the norming sample were 2.9 (SD = 2.8) for males and 3.2 (SD= 2.8) for females. Among the present sample, mean cutoff scores were 2.8 (1.9) for males and 3.1 (2.3) for females, suggesting that the present sample closely resembled the norming sample with respect to level of IM. Samples not based on undergraduates may have different IM levels. An additional limitation of the current study is that respondents completed surveys online in our study. Item order effects and/or IM may vary by survey mode, an issue that should be addressed by future research. Finally, although we recommended the use of IM scales to control for method bias in PEB research, we acknowledge that the use of IM scales does not address all method biases. Strengths This study makes an important contribution to the literature on self-reported PEB. To the authors’ knowledge, this is the first study to evaluate the effects of IM and order effects on self-reported PEB and related intentions. Also, in contrast to prior work from other fields that has examined order effects for single items, this study evaluated order effects for multiple-item scales, which is a stronger test. Additionally, the present study examined continuous rather than binary outcomes, which also advances prior work on order effects. 135 Chapter 3 References Ajzen, I. 2009. Behavioral interventions based on the Theory of Planned Behavior. Retrieved on April 24, 2009, from http://people.umass.edu/aizen/pdf/tpb.intervention.pdf Ajzen, I., & Madden, T.J. 1986. Prediction of goal-directed behaviour: attitudes, intentions, and perceived control. Journal of experimental social psychology, 22, 453- 474. Axelrod, L.J., and Lehman, D.R. 1993. Responding to environmental concerns: What factors guide individual action?. Journal of Environmental Psychology 13,149-159. Bamberg, S. 2003. How does environmental concern influence specific environmentally related behaviors? A new answer to an old question. Journal of environmental psychology, 23, 21-32. Bandura, A. 1997. Self-efficacy: The exercise of control. New York: W.H. Freeman. Beebe, T.J., Jenkins, S.M., Anderson, K. J., Davern, M.E, & Rockwood, T.H. 2008. The effects of survey mode and asking about future intentions on self-reports of colorectal cancer screening. Cancer Epidemiology, Biomarkers, and Prevention, 17, 785-290. Borkenau, P. Zaltauskas, K., & Leising, D. 2009. More may be better but there may be too much: Optimal trait level and self-enhancement bias. Journal of Personality, 77, 825- 858. Chapman, K.J. 2001. Measuring intent: There’s nothing “mere” about mere measurement effects. Psychology and Marketing, 18, 811-841. Cheung, S.F., Chan, D.K.S., & Wong, Z.S.Y. 1999. Reexamining the theory of planned behavior in understanding wastepaper recycling. Environment and Behavior, 31, 587- 612. Crowne, D., & Marlowe, D. 1964. The approval motive: Studies in evaluative dependence. New York: Wiley. Davis, CG., Thake, J., & Vilhena, N. 2010. Social desirability biases in self-reported alcohol consumption and harms. Addictive Behaviors, 35, 302-311. Ferber, R. 1952. Order bias in a mail survey. Journal of Marketing, 17, 171-178. Fernandes, M.F., & Randall, D.M. 1992. The nature of social desirability response effects in ethics research. Business Ethics Quarterly, 2, 183-205. 136 Festinger, L. 1957. A theory of cognitive dissonance. Stanford: University Press. Johnson, T.P., O’Rourke, DP., Burris, J.E., & Warnecke, R.B. 2005. An investigation of the effects of social desirability on the validity of self-report of cancer screening behaviors. Medical Care, 43, 565-573. Kaiser, F.G. 1998. A general measure of ecological behavior. Journal of Applied Social Psychology, 28, 395-422. Kaiser, F.G., Wolfing, S., & Fuhrer, U. 1999. Environmental attitude and ecological behaviour. Journal of environmental psychology, 19, 1-19. Kaiser, F. G., Schultz, W.P., & Scheuthle, H. 2007. The theory of planned behavior without compatibility? Beyond method bias and past trivial associations. Journal of Applied Social Psychology, 37, 1522-1544. McCrae, R.R., & Costa, P.T. 1983. Social desirability scales: More substance than style. Journal of Consulting and Clinical Psychology, 51, 882-888. Paulhus, D. L. 1982. Individual differences, self-presentation, and cognitive dissonance: Their concurrent operation in forced compliance. Journal of Personality and Social Psychology, 43, 838-852. Paulhus, D.L. 1984. Two-component models of socially desirable responding. Journal of Personality a Social Psychology, 46, 598-609. Paulhus, D.L. 2002. Socially desirable responding: The evolution of a construct. In H. Braun, D.N. Jackson, & D.E. Wiley Eds., The role of constructs in psychological and educational measurement pp. 67-88. Hillsdale, NJ: Erlbaum. Podsakoff, P.M., MacKenzie, S.B., Lee, J-Y., & Podsakoff, N.P. 2003. Common method biases in behavioral research: A critical review of the literature and recommended remedies. Journal of Applied Psychology, 88, 879-903. Podsakoff, P.M., & Organ, D.W. 1986. Self-reports in organization research: Problems and prospects. Journal of Management, 12, 69-82. SAS Institute Inc. SAS 9.2. Cary, NC: SAS Institute, 2002. Salancik, G.R. 1984. On priming, consistency, and order effects in job attitude assessment: With a note on current research. Journal of Management, 10, 250-254. Salancik, G.R., & Pfeffer, J. 1977. An examination of the need-satisfaction models of job attitudes. Administrative Science Quarterly, 22, 427-456. 137 Schultz, P.W., Gouveia, V.V., Cameron, L.D., Tankha, G., Schmuck, P., & M. Franek. 2005. Values and their relationship to environmental concern and conservation behavior. Journal of Cross-cultural Psychology, 36, 457-475. Stern, P.C., Dietz, T., Kalof, L., & Guagnano, G.A. (1995). Values, beliefs, and proenvironmental action: Attitude formation toward emergent attitude change. Journal of Applied Social Psychology, 25, 1611-1636. Stern, P.C., Dietz, T., Abel, T., Guagnano, G.A., & Kalof, L. 1999. A value-belief-norm theory of support for social movements: the case of environmentalism. Research in Human Ecology, 6, 81-97. Stutzman, T.M., & Green, S.B. 1982. Factors affecting energy consumption: Two field tests of the Fishbein-Ajzen model. Journal of social psychology, 117, 183-201. Triandis, H.C. 1980. Values, attitudes, and interpersonal behavior. In H. Howe ed., Nebraska symposium on motivation, pp. 195-259. Lincoln, University of Nebraska Press. Uziel, L. 2010. Rethinking social desirability scales: From impression management to interpresonally oriented self-control. Perspectives on Psychological Science, 5, 243-262. Wainer, H., & Kiely, G. L. 1987. Item clusters and computerized adaptive testing: A case for testlets. Journal of Educational Measurement, 24, 185-201. Zerbe, W.J., & Paulhus, D.L. 1987. Socially desirable responding in organizational behavior: A reconception. Academy of Management Journal, 12, 250-264. 138 Conclusions Intervention effectiveness Many university-based energy reduction efforts have been implemented in recent years, but few have been subjected to empirical evaluation. One of the aims of the present study was to evaluate the effectiveness of a competition-based intervention in reducing energy consumption and promoting other PEB among students living in a university residential setting. The study used a prospective quasi-experimental design with a control group. To engage participants, the project was implemented using a building-versus- building competition framework with a group-level reward. A website was developed with buildings assigned access to different intervention content. Survey participants, including assessment-only controls, completed baseline and follow-up self-report surveys regarding energy use behaviors and key constructs derived from the Theory of Planned Behavior (TPB) and the Norm Activation Model (NAM). A total of 302 students participated in the first survey and 225 followed up in the second. Regression analyses indicated no statistically significant effect of the competition on overall self-reported PEB. However, at time 2, electricity use outcome were higher among those who resided in target buildings compared to residents of control buildings, though the group difference did not reach traditional levels of significance. Only six students registered to use the web portal, precluding further planned between-groups analyses. Among the seven target buildings included in the competition, daily average electricity consumption during 8-week competition did not differ substantially relative to baseline levels. Overall, these findings suggest that an intervention that includes appeals, 139 a group-level incentive, and competition is not sufficient to induce substantial changes in PEB. Adjusting for average temperature, the intervention buildings used significantly more electricity during the intervention phase in 2009 and 2008 than during the control periods of both years. This highlights the need for investigators to examine patterns of energy use prior to implementing interventions for greater likelihood of finding an effect. Based on our findings, we made several additional recommendations for future university-based energy reduction interventions that are of practical and theoretical importance. Mechanisms of individual-level behavior change Another aim of the present study was to examine factors that predict change in PEB based on the TPB and NAM. Numerous studies have shown that pro-environmental behavior can be explained cross-sectionally by the TPB and NAM, but relatively little is known about how these models might perform in explaining behavior change. Results provided preliminary support for the extension of the TPB to addressing change in PEB. Additionally, we found that change in intentions was most important for explaining change in PEB. Other variables contributed a fractional amount of explained variance in PEB change. This indicates that PEB intervention design should consider how intervention strategies can directly or indirectly promote intention changes. The intervention implemented as part of the present study included appeals and a group-level incentive within a competition framework; it is difficult to understand from a theoretical perspective how these stimuli might influence intentions. Multiple groups analyses indicated that relationships among TPB and NAM model constructs did not vary across 140 the intervention and control groups. Follow-up analyses indicated that intervention status was not associated with intentions scores at time 2, suggesting that the intervention was not successful in altering intentions for PEB. The supplementary website modules that participants did not access may have been more effective in promoting PEB intentions. Alternative models of behavior change: motivation and action The TPB and NAM explain elements required for motivation to engage in a behavior, but they do less to address the processes that follow motivation and translate behavioral intentions into action. A number of theories address the gap between motivation and action. For example, Goal Setting Theory describes cognitive, affective, and behavioral factors that lead from goal motivation to goal achievement (Locke & Latham, 2002). This model posits that goal achievement, or behavior change, follows from goal setting, implementation intentions, action planning, and learning the skills required to accomplish a goal. Intervention work has demonstrated the effectiveness of setting goals in promoting energy conservation (Becker, 1978; McCalley & Midden, 2002). Another theory that elaborates upon the link between intentions and behavior is the Transtheroetical Model. This theory asserts that an individual progresses through a series of stages of readiness for behavior change when deciding to adopt a given healthy behavior (Prochaska et al., 1992). The model distinguishes between the earlier phases of motivation (i.e., the precontemplation, contemplation, and preparation stages) and the later stages of action (i.e., action and maintenance phases). The findings of the present study point to intentions as a key target for intervention work, but according to the 141 Transtheoretical Model, intentions may not always be a feasible place to begin intervening in behavior. For instance, individuals in the precontemplation stage are typically not considering changing their behavior and therefore attempting to intervene directly with intentions is unlikely to be effective for individuals at this stage. Conversely, among individuals who are already motivated to change, efforts may be better spent in promoting action/goal achievement rather than building motivation for change. For instance, a self-monitoring intervention aimed at increasing dental flossing led to higher levels of action control (defined as self-monitoring and evaluating ongoing behavior) at follow-up, but did not change intention formation (Schuz et al., 2007). Additionally, higher levels of action control predicted increased flossing behavior among individuals who already had relatively high levels of motivation to floss, but not among those with lower motivation levels. The intervention therefore had no motivational effect, but did have a volitional effect among motivated individuals, suggesting that matching intervention strategies to stages of readiness for change can be effective in intervention planning. Combining features of the Theory of Planned Behavior, Goal Setting Theory, and Transtheoretical Model as displayed in figure 4 below may aid in selecting appropriate intervention targets. To utilize this framework, a baseline assessment would have to be conducted to provide for an assessment of readiness for change. Next, intervention strategies that target different theoretical constructs corresponding to stage of change should be chosen. For example, if an individual is in the precontemplation stage, interventions that promote positive social norms might be selected. If an individual is in 142 the preparation stage, s/he may already have sufficient motivation, and a strategy that promotes action, such as goal setting, may be appropriate. If an individual is in the action stage, strategies that affirm engagement in the target behavior and promote behavior maintenance, perhaps self-monitoring, might be appropriate. Figure 4. Combining features of the Theory of Planned Behavior, Goal Setting Theory, and Transtheoretical Model to suggest appropriate intervention targets. PRECONTEMPLATION & CONTEMPLATION N Norms Perceived Control Attitudes PREPARATION ACTION MAINTENANCE New Behavior Sustained Behavior Change Implementation Intentions Goal setting & planning Skill acquisition Behavioral Intentions Note that this is a suggested framework to aid in intervention design that takes into account stages of readiness for change. In addition to the Theory of Planned Behavior and Goal Setting Theory, consideration of other conceptual models of behavior and behavior change may be useful. A personalized approach based on stage of change may not be useful in all situations and would certainly be difficult to implement among a broad population with wide variation in readiness for change. This may be one reason why interventions that have applied several strategies simultaneously have been successful. For instance, by combining goal setting with normative and descriptive feedback, intervention strategies appropriate for several stages of change are applied, and 143 can be successfully applied to a broader population. Future PEB intervention work should investigate whether intervention strategies vary in effectiveness based on stage of change. Measuring PEB and related constructs The present study also investigated issues related to the measurement of PEB and PEB intentions. Specifically, impression management bias and item order effects were evaluated among a subset of respondents to the first survey. Consistent with prior work, this study found that higher IM scores predicted higher scores on two PEB subscales. However, there was no association between IM and other types of prosocial behavior, suggesting that the effect of IM was limited to certain behaviors that were more closely aligned with study goals. Weak evidence of item order effects was also found, whereby individuals receiving the Intention-Behavior question ordering had significantly higher scores on the behavior scale than individuals with the Behavior-Intention ordering. This finding contrasts with prior work that has found order effects in the opposite direction. It is possible that Intention-Behavior order effects may result from cognitive dissonance generated by first stating intentions. Such a process may be more closely related to self- deception rather than IM, which is a form of other-deception. We recommend that IM and self-deception bias be considered in designing surveys, conducting analyses, and interpreting study findings. Conclusion Thoughtful actions need to be taken to retard the trajectory of increasing energy needs. Since the 1970s, interventions aimed at reducing residential energy consumption represent have been conducted with varying levels of success, demonstrating that 144 thoughtfully implemented efforts can be effective in attenuating energy demand. Additionally, several conceptual models of behavior have been applied to understand energy use and other types of environmentally relevant behavior. Understanding how interventions work to change behavior and its determinants, and how to measure and examine these processes reliably, remain important avenues for future research. 145 Comprehensive Bibliography Abrahamse, W., Steg, L., Vlek, C., & T. Rothengatter. 2005. A review of intervention studies aimed at household energy conservation. Journal of Environmental Psychology, 25, 273-291. Ajzen, I. 2009. Behavioral interventions based on the Theory of Planned Behavior. Retrieved on May 30, 2009, from http://people.umass.edu/aizen/pdf/tpb.intervention.pdf. Ajzen, I., & Madden, T.J. 1986. Prediction of goal-directed behaviour: attitudes, intentions, and perceived control. Journal of experimental social psychology, 22, 453- 474. Aronson, E. 1969. The theory of cognitive dissonance: A current perspective. In Zanna, M.P, & Olson, J.M., (Eds.). Advances in Experimental Social Psychology, Vol. 4, pp. 1- 34. New York: Academic Press. Axelrod, L.J., & Lehman, D.R. 1993. Responding to environmental concerns: What factors guide individual action?. Journal of Environmental Psychology 13, 149-159. Bamberg, S. 2003. How does environmental concern influence specific environmentally related behaviors? A new answer to an old question. Journal of Environmental Psychology, 23, 21-32. Bamberg, S., & G. Moser. 2007. Twenty years after Hines, Hungerford, and Tomera: A new meta-analysis of psycho-social determinants of pro-environmental behavior. Journal of Environmental Psychology, 27, 14-25. Bandura, A. 1997. Self-efficacy: The exercise of control. New York: W.H. Freeman. Baranowski, T., Cullen K.W., & Baranowski, J. 1999. Psychosocial correlates of dietary intake: advancing dietary intervention. Annual Review of Nutrition, 19, 17-40. Baranowski, T., Cullen, K.W., Nicklas, T., Thompson, D., Baranowski, J. 2003. Are current health behavioral change models helpful in guiding prevention of weight gain efforts? Obesity Research, 11, 235-245. Baron, R.M., & Kenny, D.A. 1986. The moderator-mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51, 1173-1182. Becker, L.J. 1978. Joint effect of feedback and goal setting on performance: A field study of residential energy conservation. Journal of Applied Psychology, 63, 428-433. 146 Becker, L.J., & Seligman, C. 1978. Reducing air conditioning waste by signaling it is cool outside. Personality and Social Psychology Bulletin, 4, 412-415. Beebe, T.J., Jenkins, S.M., Anderson, K. J., Davern, M.E, & Rockwood, T.H. 2008. The effects of survey mode and asking about future intentions on self-reports of colorectal cancer screening. Cancer Epidemiology, Biomarkers, and Prevention, 17, 785-290. Bertoldi, P., & Atanaiu, B. 2006. Electricity consumption and efficiency trends in the enlarged European union: Status report 2006. Institute for environment and sustainability, 2007. http://re.jrc.ec.europa.eu/energyefficiency/pdf/EnEff%20Report%202006.pdf Bittle, R.G., Valesano, R.M., & Thaler, G. 1979. The effects of daily cost feedback on residential electricity consumption. Behavior Modification, 3, 187-202. Borkenau, P. Zaltauskas, K., & Leising, D. 2009. More may be better but there may be too much: Optimal trait level and self-enhancement bias. Journal of Personality, 77, 825- 858. Brandon, G., & Lewis, A. 1999. Reducing household energy consumption: A qualitative and quantitative field study. Journal of Environmental Psychology, 19, 75-85. Breckon, D.J., Harvey, J.R., & R.B. Lancaster. 1989. Community health education: Settings, roles, and skills. 2 nd ed. Rockville, MD: Aspen Publishing. Brug, J., Oenema, A., & Ferreira, I. 2005. Theory, evidence, and Intervention Mapping to improve behavior nutrition and physical activity interventions. International Journal of Behavioral Nutrition and Physical Activity, 2:2. California Department of Energy. 2009. Energy consumption data management system. http://www.ecdms.energy.ca.gov/Elecbyutil.asp#results. Chapman, K.J. 2001. Measuring intent: There’s nothing “mere” about mere measurement effects. Psychology and Marketing, 18, 811-841. Cheung, S.F., Chan, D.K.S., & Z.S.Y. Wong. 1999. Reexamining the theory of planned behavior in understanding wastepaper recycling. Environment and Behavior, 31, 587- 612. Cialdini, R. B., Reno, R.R., & Kallgren, C.A. 1990. A focus theory of normative conduct: Recycling the concept of norms to reduce littering in public places. Journal of Personality and Social Psychology, 58, 1015-1026. 147 Clark, C., Kotchen, M.J., & Moore, M.M. 2003. Internal and external influences on pro- environmental behavior: Participation in a green electricity program. Journal of environmental psychology, 23, 237-246. Conner, M., & Norman, P. 1996. Predicting health behavior: Philadelphia, Open University Press. Costanzo, M., Archer, D., Aronson, M., & Pettigrew, T. 1986. Energy conservation behavior: The difficult path from information to action. American Psychologist, 41, 521- 528. Craig, S., & McCann, J.M. 1978. Assessing communication effects on energy conservation. Journal of Consumer Research, 5, 82-88. Crowne, D., & Marlowe, D. 1964. The approval motive: Studies in evaluative dependence. New York: Wiley. Davis, CG., Thake, J., & Vilhena, N. 2010. Social desirability biases in self-reported alcohol consumption and harms. Addictive Behaviors, 35, 302-311. Deci, E.L., & Ryan, R.M. 1985. Intrinsic motivation and self-determination in human behavior. New York: Platinum Press. Donaldson, S.I., Graham, J.W., & W.B. Hanse. 1994. Testing the generalizability of intervening mechanism theories: Understanding the effects of adolescent drug use prevention interventions. Journal of Behavioral Medicine, 17, 195-216. Dunlap, R.E., Van Liere, K., Mertig, A., & Jones, R.E. 2000. Measuring endorsement of the New Ecological Paradigm: A revised NEP scale. Journal of social issues, 56, 425- 442. Ferber, R. 1952. Order bias in a mail survey. Journal of Marketing, 17, 171-178. Fernandes, M.F., & Randall, D.M. 1992. The nature of social desirability response effects in ethics research. Business Ethics Quarterly, 2, 183-205. Festinger, L. 1957. A theory of cognitive dissonance. Stanford: University Press. Fried, C. B. & E. Aronson. 1995. Hypocrisy, misattribution, and dissonance reduction. Personality and Social Psychology Bulletin, 21, 925-934. Garling, T., Fujii, S., Garling, A., & Jakobsson, C. 2003. Moderating effects of social value orientation on determinants of proenvironmental behavior. Journal of Environmental Psychology, 23, 1-9. 148 Geller, E.S. 1992. Solving environmental problems: A behavior change perspective. In S. Staub & P. Green (Eds.), Psychology and social reasonability: Facing global challenges. New York: New York University Press. Geller, E. S. 2002. The challenge of increasing proenvironmental behavior. In R. B. Betchel, & A. Churchman (Eds.), Handbook of environmental psychology (pp. 525- 540). New York: John Wiley & Sons. Guagnano, G.A., Stern, P.C., & T. Dietz. 1995. Influences on attitude-behavior relationships: A natural experiment with curbside recycling. Environment and Behavior, 27, 699-718. Goldstein, N.J., Cialdini, R.B., & Griskevicius, V. 2008. A room with a viewpoint: using social norms to motivate environmental conservation in hotels. Journal of Consumer Research, 35, 472-482. Hayes, S.C., & Cone, J.D. 1977. Reducing residential electrical energy use: Payments, information, and feedback. Journal of Applied Behavior Analysis, 10, 425-435. Henry, K.B., Arrow, H., & Carini, B. 1999. A tripartite model of group identification: Theory and measurement. Small Group Research, 30, 558-581. Hines, J.M., Hungerford, H.R., & A.N. Tomera. 1987. Analysis and synthesis of research on responsible environmental behavior: A meta-analysis. Journal of Environmental Education, 18, 1-8. Hirst, E., & S. Grady. 1982-1983. Evaluation of a Wisconsin utility home energy audit program. Journal of Environmental Systems, 12, 303-320. Hopper, J., & J. Nielsen. 1991. Recycling as altruistic behavior: Normative and behavioral strategies to expand participation in a community recycling program. Environment and Behavior, 23, 195-220. Intergovernmental Panel on Climate Change (2007). Climate Change 2007: Synthesis Report. Retrieved on February 12, 2008, from http://www.ipcc.ch/pdf/assessment-report/ar4/syr/ar4_syr.pdf Jeffrey, R.W. 2004. How can health behavior theory be made more useful for intervention research? International Journal on Behavioral Nutrition and Physical Activity, 23, 1-10. Johnson, T.P., O’Rourke, DP., Burris, J.E., & Warnecke, R.B. 2005. An investigation of the effects of social desirability on the validity of self-report of cancer screening behaviors. Medical Care, 43, 565-573. 149 Kahn, E.B., Ramsey, L.T., Brownson, R.C., Heath, G.W., Howze, E.H., Powell, K.E., Stone, E.J., Rajab, M.W., Corso, P., & the Task Force on Community Preventative Services. The effectiveness of interventions to increase physical activity. American Journal of Preventive Medicine, 22, 73-107 Kaiser, F.G., 1998. A general measure of ecological behavior. Journal of Applied Social Psychology, 28, 395-422. Kaiser, F.G., & Gutscher, H. 2003. The proposition of a general version of the theory of planned behavior: Predicting ecological behavior. Journal of Applied Social Psychology, 33, 586-603. Kaiser, F. G., Schultz, W.P., & Scheuthle, H. The theory of planned behavior without compatibility? Beyond method bias and past trivial associations. Journal of Applied Social Psychology, 37, 1522-1544. Kaiser, F.G., Wolfing, S., & Fuhrer, U. 1999. Environmental attitude and ecological behaviour. Journal of environmental psychology, 19, 1-19. Kantola, S.J., Syme, G.J., & Campbell, N.A. 1984. Cognitive dissonance and energy conservation. Journal of Applied Psychology, 69, 416-421. Katzev, R., & Johnson, T.R. 1984) Comparing the effects of monetary incentives and foot-in-the-door strategies in promoting residential electricity conservation. Journal of Social Psychology, 14, 12-27. Latham, G.P., & Locke, E.A. 1991. Self-regulation through goal setting. Organization Behavior and Human Decision Processes, 50, 212-247. Locke, E.A., & Latham, G. P. 2002. Building a practically useful theory of goal setting and task motivation. American Psychologist, 57, 705-717. Mapstone, J., Elbourne, D., & I. Roberts. 2007. Strategies to improve recruitment to research studies. Cochran Database of Systematic Reviews, 18. Marcell, K., Agyeman, J., & A. Rappaport. 2004. Cooling the campus: Experiences from a pilot study to reduce electricity use at Tufts university, USA, using social marketing methods. International Journal of Sustainability in Higher Education, 5, 169-189. McCalley, M.T., & Midden, C.J.H. 2002. Energy conservation through product- integrated feedback: The roles of goal-setting and social orientation. Journal of Economic Psychology, 23, 589-603. 150 McClelland, L., & Cook, S.W. 1979-1980. Energy conservation effects of continuous in- home feedback in all-electric homes. Journal of Environmental Systems, 9, 169-173. McClelland, L, & S.W. Cook. 1980a. Energy conservation in university buildings: Evaluation Review, 4. 119-133. McClelland, L, & S.W. Cook. 1980b. Promoting energy conservation in master-metered apartments through group financial incentives. Journal of Applied Psychology, 10, 20-31. McCrae, R.R., & Costa, P.T. 1983. Social desirability scales: More substance than style. Journal of Consulting and Clinical Psychology, 51, 882-888. McGuire, W.J. 1985. Attitudes and attitude change. In G. Lindzey & E. Aronson Eds., Handbook of social psychology pp. 223-246. New York: Random House. Midden, J.F., Weenig, M.H., & Zieverink, H.J.A. 1983. Using feedback, reinforcement, and information to reduce energy consumption in households: A field experiment. Journal of Economic Psychology, 3, 65-86. Miller, R.D., & Ford, J.M. 1985. Shared savings in the residential market: A public/private partnership for energy conservation. Baltimore, MD: Urban consortium for technology initiatives, Energy task Force. Mitchell, G. K., & A. P. Abernethy. 2005. A comparison of methodologies from two longitudinal community-based randomized controlled trails of similar interventions in palliative care: What worked and what did not?, Journal of Palliative Medicine, 8, 1226- 1237. Muthen, L.K., Muthen, B., 2006. M-Plus: The Comprehensive Modeling Program for Applied Researchers Version 5.21 [Computer software]. Muthen and Muthen, Los Angeles, CA. National Oceanic and Atmospheric Administration. 2009. National weather service observed weather. Retrieved on March 9, 2010, from http://www.weather.gov/climate/index.php?wfo=lox National Oceanic and Atmospheric Administration. 2010. U.S. Department of Commerce National Oceanic and Atmospheric Administration National Environmental Satellite, data, and information service. Retrieved on January 20, 2011, from http://www.ncdc.noaa.gov/oa/documentlibrary/hcs/cdd.200801-200912.pdf Pallak, M.S., Cook, D.A., & Sullivan, J.J. unpublished. The effects of comparison feedback on electricity usage. University of Iowa. 151 Pallak, M.S., & Cummings, N. (1976). Commitment and voluntary energy conservation. Personality and Social Psychology Bulletin, 2, 27-31. Paulhus, D. L. 1982. Individual differences, self-presentation, and cognitive dissonance: Their concurrent operation in forced compliance. Journal of Personality and Social Psychology, 43, 838-852. Paulhus, D.L. 1984. Two-component models of socially desirable responding. Journal of Personality a Social Psychology, 46, 598-609. Paulhus, D.L. 2002. Socially desirable responding: The evolution of a construct. In H. Braun, D.N. Jackson, & D.E. Wiley (Eds.), The role of constructs in psychological and educational measurement (pp. 67-88). Hillsdale, NJ: Erlbaum. Petersen, J.E., Shunturov, V., Janda, K., Platt, G., and K. Weinberger. 2007. Dormitory residents reduce electricity consumption when exposed to real-time visual feedback and incentives. International Journal of Sustainability in Higher Education, 8, 16-33. Pitts, R.E., and J.L. Wittenbach. 1981. Tax credits as a means of influencing consumer behavior. Journal of Consumer Research, 8, 335-338. Podsakoff, P.M., MacKenzie, S.B., Lee, J-Y., & Podsakoff, N.P. 2003. Common method biases in behavioral research: A critical review of the literature and recommended remedies. Journal of Applied Psychology, 88, 879-903. Podsakoff, P.M., & Organ, D.W. 1986. Self-reports in organization research: Problems and prospects. Journal of Management, 12, 69-82. Prochaska, J.O., DiClemente, C.C., & Norcross, J.C. 1992. In search of how people change: Applications to addictive behaviors. American Psychologist, 47, 1102-1114. Sabine, C.L., & Tanhua, T. 2010. Estimation of anthropogenic CO2 inventories in the ocean. Annual Review of Marine Science, 2, 175-198. Salancik, G.R. 1984. On priming, consistency, and order effects in job attitude assessment: With a note on current research. Journal of Management, 10, 250-254. Salancik, G.R., & Pfeffer, J. 1977. An examination of the need-satisfaction models of job attitudes. Administrative Science Quarterly, 22, 427-456. SAS Institute Inc. SAS 9.2. Cary, NC: SAS Institute, 2002. Schwartz, S.H. 1994. Are there universal aspects in the structure and contents of human values? Journal of Social Issues, 50, 19-45. 152 Schultz, P.W., Gouveia, V.V., Cameron, L.D., Tankha, G., Schmuck, P., & M. Franek. 2005. Values and their relationship to environmental concern and conservation behavior. Journal of Cross-cultural Psychology, 36, 457-475. Schultz, P.W., Nolan, J.M., Cialdini, R.B., Goldstein, N.J., & Griskevicius, V. 2007. The constructive, destructive, and reconstructive power of social norms. Psychological Science, 18, 429-434. Schuz, B., Sniehotta, F.F., & Schwarzer, R. 2007. Stage-specific effects of an action control intervention on dental flossing. Health Education Research, 22, 332-341. Schwartz, S.H. 1994. Are there universal aspects in the structure and contents of human values? Journal of social issues, 50, 19-45. Seligman, C., & J.M. Darley. 1977. Feedback as a means of decreasing residential energy consumption. Journal of Applied Psychology, 62, 363-368. Sheeran, P. 2002. Intention-behavior relations: A conceptual and empirical review. In Stroebe, W., & Hewstone, M. (Eds.), European Review of Social Psychology (V. 12). New York :John Wiley and Sons. Staats, H., Harland, P., & Wilke, H.A.M. 1996. Communicating the greenhouse effect ot he public: Evaluation of mass media campaign from a social dilemma perspective. Journal of environmental management, 45, 189-203. Stern, P. 1992. What psychology knows about energy conservation. American Psychologist, 47, 1224-1232. Stern, P.C., Dietz, T., Abel, T., Guagnano, G.A., & Kalof, L. 1999. A value-beleif-norm theory of support for social movements: The case of environmentalism. Human Ecology Review, 6, 81-97. Stern, P.C., Dietz, T., Kalof, L., & Guagnano, G.A. 1995. Values, beliefs, and proenvironmental action: Attitude formation toward emergent attitude change. Journal of Applied Social Psychology, 25, 1611-1636. Stott, P.A., Stone, D.A., & Allen, M.R. (2004). Human contribution to the European heatwave of 2003. Nature, 432, 610-614. Stutzman, T.M., & Green, S.B. 1982. Factors affecting energy consumption: Two field tests of the Fishbein-Ajzen model. Journal of social psychology, 117, 183-201. 153 Sundblad, E-L., Biel, A., & Garling, T. 2007. Cognitive and affective risk judgments related to climate change. Journal of Environmental Psychology, 27, 97-106. Sutton, S., 2000. Interpreting cross-sectional data on stages of change. Psychological Health, 15, 163-171. Syme, G.J., Nancarrow, B.E., & Seligman, C. (2000). The evaluation of information campaigns to promote voluntary household water conservation. Evaluation Review, 24, 539-578. Triandis, H.C. 1980. Values, attitudes, and interpersonal behavior. In H. Howe ed., Nebraska symposium on motivation, pp. 195-259. Lincoln, University of Nebraska Press. US Energy Information Administration 2010. Retrieved on June 17, 2010 from http://www.eia.doe.gov/aer/consump.html University of Southern California (2008). USC energy statistics. Retrieved on April 13, 2008, from http://www.usc.edu/fms/dept_energy_stats.shtml Uziel, L. 2010. Rethinking social desirability scales: From impression management to interpresonally oriented self-control. Perspectives on Psychological Science, 5, 243-262. Van Houwelingen, J.H., & F.W. Van Raaij. 1989. The effect of goal setting and daily electronic feedback on in-home energy use. Journal of Consumer Research, 16, 98-105. Van Liere, K., & R. Dunlap. 1978. Moral norms and environmental behavior: An application of Schwartz’s norm-activation model to yard burning. Journal of Applied Social Psychology, 8, 174-188. Wainer, H., & Kiely, G. L. 1987. Item clusters and computerized adaptive testing: A case for testlets. Journal of Educational Measurement, 24, 185-201. Winett, R. A., & Ester, P. 1983. Behavioral science and energy conservation. Conceptualizations, strategies, outcomes, energy policy applications. Journal of Economic Psychology 3: 203-229. Winett, R.A., Hatcher, J.W., Foret, T.R., Leckliter, I.N., Love, S.Q., Riley, A.W., & Fishback, J.F. (1982). The effects of videotape modeling and daily feedback on residential electricity conservation, home temperature and humiity, perceived comfort, and clothing worn: Winter and Summer. Journal of Applied Behavior Analysis, 15, 381- 402. 154 Winnett, R.A., Kagel, J.H., Battalio, R.C., & R.C. Winkler. 1978. Effects of monetary rebates, feedback, and information on residential electricity conservation. Journal of Applied Psychology, 63, 73-80. Winett, R.A., Neale, M.S., & Grier, H.C. 1979. Effects of self-monitoring and feedback on residential electricity consumption. Journal of Applied Behavior Analysis, 12, 173- 184. Wodarski, J.S. 1982. National and state appeals for energy conservation: A behavioral analysis of effects. Behavioral Engineering, 7, 119-130. Zerbe, W.J., & Paulhus, D.L. 1987. Socially desirable responding in organizational behavior: A reconception. Academy of Management Journal, 12, 250-264. 155 Appendix 1A: Appeals Posters 156 157 Appendix 1B: Focus Group Questions Focus groups were dynamic in nature and the material below provided a basic structure for the conversations that took place. Based on participant responses, discussion related to the material below was added or ignored. Opening question: Relate to some universal interest that might generate responses related to environment/sustainability. EX: what is one place you’d like to visit in your lifetime? Student interests in environmental issues • There is a lot of talk about environmental issues today. What types of things do you consider to be environmental issues? Intervention website [describe briefly] • Tell us about your perceptions of the Ecolympics website. Dorm cohesion • Tell us about any sense of dorm affiliation or cohesion. When did that develop, and what did it look like? • In the competition, we attempted to strengthen this. Did you notice an effect? How could this be done most effectively? Recruitment Strategy • How did you hear about the study? • We did a variety of things to encourage study participation, like talking with RAs, RCs, and USC professors, presenting at USRC meetings, emailing eligible people directly, using Facebook, and partnering with the USC Sustainability office to encourage survey participation. How did you feel about those things? • Did you notice the study on Facebook? If not, how could Facebook be better leveraged to encourage participation? • What are some ways to encourage somebody to complete the survey without being intrusive? • How did the pizza party prize influence participation in the survey? In the energy reduction challenge? How about the raffle prizes? • What are other ways that we could use to encourage survey participation or to encourage you to check out the intervention website? • How could this competition be marketed/advertised to promote student interest? Student preferences for content • What content would you like to see featured on the website? • What did/would you (other students) find most useful to learn about? 158 Appendix 1C: Exploratory Analyses with Group Identification Table 1. Results of univariable regression models predicting each outcome from group identification score (n=172). Time 1 group identification PEB Electricity use Consumer behavior Time 1 Group Identification F (p) Beta R² 6.34 (p<.02) 0.19* .036 1.54 (p<.22) 0.09 .009 3.68 (p<.06) 0.15 .024 Time 2 Group Identification F (p) Beta R² 2.62 P<.11 0.12 .015 3.36 (p<.07) 0.14 0.14 0.02 P<.90) 0.01 .0001 *p<.05. Table 2. R² values associated with adding additional predictors to each model for general Pro-environmental Behavior (PEB), electricity use, and consumer behavior outcomes (n=172). PEB Electricity use Consumer behavior Model I Time 1 score 58.2 46.6 44.6 Model II Year of enrollment 0.5 0.0 0.0 Model IIIa Model IIIb Group identification 2.0 0.01 1.0 2.1 1.8 0.2 Model IV a Model IV b Intervention status (yes=1) 0.0 0.001 0.8 0.3 2.2 1.0 Model V a Model V b Intervention status x Time 1 score interaction 0.0 0.0 0.0 0.0 0.2 0.2 Total a Total b 60.7 59.9 48.2 49.0 48.8 46.0 159 Table note. To calculate sequential R², variables were added in a stepwise manner using the order listed. Models III a, IV a, and V a include time 1 group identification scores. Models III b, IV b, and V b include time 2 group identification scores. Table 3a. Standardized regression coefficients for final models with time 1 group identification (n=172). PEB Electricity use Consumer behavior F (p) 51.32 (p<.0001) 30.84 (p<.0001) 30.34 (p<.0001) Time 1 score 0.74*** 0.67*** 0.69*** Year of enrollment 0.04 -0.03 0.0002 T1 group identification 0.14** 0.09 0.15* Intervention status (yes=1) 0.02 0.09 0.07 Intervention status x Time 1 score interaction 0.01 -0.01 -0.20 *p<.05. *p<.001. ***p<.0001. Table 3b. Standardized regression coefficients for final models with time 2 group identification (n=172). PEB Electricity use Consumer behavior F (p) 49.52 (p<.0001) 31.86 (p<.0001) 28.25 (p<.0001) Time 1 score 0.75*** 0.68*** 0.70*** Year of enrollment 0.05 -0.03 0.01 T2 group identification 0.10* 0.14* 0.07 Intervention status (yes=1) 0.01 0.06 0.10 Intervention status x Time 1 score interaction 0.02 0.001 -0.22 *p<.05. *p<.001. ***p<.0001. 160 Appendix 1D: Total Sample Demographic Characteristics Sample demographic characteristics based on survey 1 data. Total sample (n=291) Participated in Survey 1 only (n=75) Participated in Surveys 1 and 2 (n=216) Mean age (SD) Range 19.0 (1.2) 18-25 n=287 18.8 (1.2) 18-22 n=75 19.0 (1.2) 18-25 n=212 Sex (% female) 75.2 n=282 68.1 n=72 77.6 n=210 High school GPA >4.7 4.2-4.7 3.6-4.1 3.0-3.5 Below 3.0 7.0 39.7 41.4 11.2 0.7 n=285 8.2 34.3 41.1 13.7 2.7 n=73 6.6 41.5 41.5 10.4 0.0 n=212 Maternal education No high school diploma High school graduate Some college Technical school 4-year college degree Graduate degree 4.6 7.7 18.2 3.2 37.5 28.8 n=285 4.1 14.9 18.9 1.4 32.4 28.4 n=74 4.7 5.2 18.0 3.8 39.3 28.9 n=211 Paternal education No high school diploma High school graduate Some college Technical school 4-year college degree Graduate degree 4.6 11.5 12.6 1.8 24.8 44.8 n=286 6.8 10.8 14.9 0.0 28.4 39.2 n=74 3.8 11.8 11.8 2.4 23.6 46.7 n=212 Urbanicity of place of origin Rural Suburban Urban 3.5 74.9 21.6 n=287 2.7 64.0 33.3 n=75 3.8 78.8 17.5 n=212 International student (% yes) 12.4 18.7 10.2 University class status Freshman Sophomore Junior 53.3 28.8 11.6 71.2 19.2 2.7 47.2 32.1 14.6 161 Senior 6.3 n=285 6.9 n=73 6.1 n=212 Number of semesters in residence at current residence hall 1 2 3 4 5+ 76.9 2.8 14.0 1.8 4.6 n=286 84.0 2.7 8.0 2.7 2.7 n=75 74.4 2.8 16.1 1.4 3.6 n=211 Facebook.com account ( % yes) 97.2 n=286 97.3 n=75 97.2 n=211 Facebook log-in frequency At least once daily A few times per week Weekly A few times per month Less than monthly 87.2 9.7 2.2 1.8 1.4 n=279 87.7 6.9 2.7 2.7 0.0 n=73 86.9 7.8 1.9 1.5 1.9 n=206 162 Appendix 2A: Measures Schultz PEB Scale Stem question How often have you done each of the following in the past 2 weeks? Response options: 1: Never 2: Rarely 3: Sometimes 4: Fairly Often 5: Very Often Items: 1. Recycled paper 2. Looked for ways to reuse things 3. Volunteered time to help an environmental group 4. Wrote a letter or email, or endorsed a petition on a website in support of an environmental issue 5. Encouraged friends or family to recycle *6. Turned up the heat on thermostat when I felt cold or turned on the air conditioner when I felt hot 7. Composted food scraps 8. Used cold water to wash clothes *9. Left lights on when I was not in a room 10. Recycled cans or bottles 11. Purchase products in reusable containers 12. Donated money to an environmental group 13. Picked up litter that was not my own *14. Thrown away leftover food on my plate that I didn’t eat 15. Conserved gasoline by walking or bicycling *16. Let the water run while I was brushing my teeth 17. Shut off my computer when I was not using it * Reverse-coded item. The content of these 17 items was used in the Intentions Scale, Expected Outcome Scale, Perceived Control Scale, Social Norms Scale, Personal Norms Scale, and Ascribed Responsibility Scale. The stem questions are response options for these scales are listed below. 163 Intentions Scale stem question I intend to ___ in the next few weeks: Response options: 1: very unlikely 2: unlikely 3: undecided 4: likely 5: very likely Expected Outcome Scale stem question If I did each of the following behaviors in the next few weeks, it would be: Response options: 1: Very bad 2: bad 3: poor 4: neither good nor bad 5: fair 6: good 7: very good Perceived Control Scale stem question How sure are you that you could do each of the following if you wanted to? Response options: 1: completely unsure 2: somewhat unsure 3: don’t know 4: somewhat sure 5: completely sure Note: All of the items on this scale were positively keyed due to them stem question. 164 Social Norms Sale stem question Stem question: Most people who are important to me would support my decision to ____ Response Options: 1: Very unlikely 2: Unlikely 3: Likely 4: Very likely Note: All of the items on this scale were positively keyed due to them stem question. Sundblad Global Warming Quiz (assessed awareness of consequences; Sundblad et al., 2007). Correct answers are in bold. 1. The global average temperature in the air has increased approximately 1.0 degrees Fahrenheit during the last 1:0:0: years. 1: true 2: false 2. The 1990: decade had a normal average temperature compared to other decades during the last 100 years. 1: true 2: false 3. The global change in temperature in the past 100 years is the largest change during the past 1000 years. 1: true 2: false 4. In the past 100 years, precipitation has increased in most areas in the middle and northern parts of the northern hemisphere. 1: true 2: false 5. Climate change is mainly caused by increased concentration of greenhouse gases. 1: true 2: false 6. Climate change is mainly caused by the ozone hole. 1: true 2: false 165 7. The carbon dioxide concentration has increased between 20% and 30% in the atmosphere during the last 250 years. 1: true 2: false 8. The increase of greenhouse gases is mainly caused by air pollution from industry. 1: true 2: false 9. The number of storms and flooding has increased prominently in the past 100 years. 1: true 2: false 10. Global precipitation is expected to decrease in the next 100 years. 1: true 2: false 11. The global sea level has risen approximately 8 inches the past 100 years. 1: true 2: false 12. Climate change will increase the risk for diseases transferred by water (e.g., diarrhea) the next 100 years. 1: true 2: false 13. Negative health impacts caused by global climate change will affect humans in the countryside more than humans living in cities. 1: true 2: false 166 New Ecological Paradigm (assessed attitudes; Dunlap et al., 2000) Stem Question: Please indicate the extent to which you agree or disagree with the following items: Response Options: 1: strongly disagree 2: disagree 3: neither agree nor disagree 4: agree 5: strongly agree Items: 1. We are approaching the limit to the number of people the earth can support. *2. Humans have the right to modify the natural environment to suit their needs 3. When humans interfere with nature it often produces disastrous consequences *4. Human ingenuity will insure that we do not make the earth unlivable 5. Humans are severely abusing the environment *6. The earth has plenty of natural resources if we just learn how to develop them *7. Plants and animals exist primarily to be used by humans *8. The balance of nature is strong enough to cope with the impacts of modern industrial nations 9. Despite our special abilities, humans are still subject to the laws of nature *10. The so-called “ecological crisis” facing humankind has been greatly exaggerated 11. The earth is like a spaceship with very limited room and resources *12. Humans were meant to rule over the rest of nature 13. The balance of nature is delicate and easily upset *14. Humans will eventually learn enough about how nature works to be able to control it 15. If things continue on their present course, we will soon experience a major ecological disaster *Indicates reverse coded item. 167 Garling Ascribed Responsibility Scale Response options: 1: strongly disagree 2: disagree 3: somewhat disagree 4: neither agrees nor disagree 5: somewhat agree 6: agree 7: strongly agree Items: 1. I am not concerned about the environment. 2. Every citizen must take responsibility for the environment. 3. Authorities rather than the citizens are responsible for the environment. Clark Ascribed Responsibility Scale Response options: 1: strongly disagree 2: disagree 3: somewhat disagree 4: neither agree nor disagree 5: somewhat agree 6: agree 7: strongly agree Items: 1: Households like mine should not be blamed for environmental problems caused by energy production and use. 2: My responsibility is to provide only for my family and myself. 3: My personal actions can greatly improve the well being of people I don’t know. Author’s Ascribed Responsibility Scale stem question How much would your participating in each of the following activities help reduce your personal contribution to future global warming? (If you already do something listed below, then rate how much you believe this reduces your personal contribution to global warming currently.) Response options: 1: would not help at all 2: 3: 4: would help somewhat 5: 6: 7: would help a lot 168 Items: Same as PEB scale; no items were reverse-coded Garling Personal Norms Scale Response options: 1: strongly disagree 2: disagree 3: somewhat disagree 4: neither agree nor disagree 5: somewhat agree 6: agree 7: strongly agree 1: I feel a moral obligation to protect the environment. 2: I feel that I should protect the environment. 3: I feel it is important that people in general protect the environment. 4: Our environmental problems cannot be ignored. Clark Personal Norms Scale Response options: 1: strongly disagree 2: disagree 3: somewhat disagree 4: neither agree nor disagree 5: somewhat agree 6: agree 7: strongly agree Items 1: I worry about conserving energy only when it helps to lower my utility bills. 2: The individual alone is responsible for his or her satisfaction in life. 3: It is my duty to help other people when they are unable to help themselves. 169 Author’s Personal Norms Scale Stem question I feel a moral obligation to do the following Response options: 1: strongly disagree 2: disagree 3: somewhat disagree 4: neither agree nor disagree 5: somewhat agree 6: agree 7: strongly agree Items: Same as PEB scale; no items were reverse-coded 170 Appendix 2B: Fitting Alternative TPB Models Note: Shaded column represents final TPB model. Table 1a. TPB baseline and alternative models. Baseline model (Figure 2.2) Model I Baseline plus direct path from PNor→ PEB Model II Baseline plus direct SNor → PEB path Model III Baseline plus direct ExOut → PEB path Model IV Baseline plus direct PerCont → PEB path Model V Baseline plus direct Att→PEB path χ² (df) 91.4 (12) 91.4 (11) 91.3 (11) 90.8 (11) 88.3 (11) 91.2 (11) R² PEB Int .70 .29 .70 .29 .70 .29 .70 .29 .70 .29 .70 .29 CFI .83 .83 .83 .83 .83 .83 RMSEA .18 .19 .19 .19 .19 .19 Table 1b. TPB baseline and alternative models. Model VI Baseline plus direct PNor and SNor →PEB paths Model VII Baseline plus direct ExOut → Int path. Model VIII Baseline plus direct SNor → Int path Model IX Direct paths from all variables to PEB χ² (df) 91.3 (10) 89.0 (11) 90.6 (11) 84.8 (7) R² PEB Int .83 .29 .83 .28 .83 .30 .83 .29 CFI .20 .19 .19 .23 RMSEA .70 .70 .70 .71 171 Table 2. Testing the Relative Importance of Int Compared to Other Variables in the TPB Framework. Model X Baseline model but remove direct paths to Int and add direct paths from PNor, Att, PerCont → PEB. Model XI Same as model V but remove Int from model and all paths → Int Model XII Same as model V but keep direct paths → Int χ² (df) 112.9 (9) 368.2 (13) 87.7 (9) R² PEB Int .72 -- .21 -- .70 .29 CFI .75 .23 .83 RMSEA .24 .36 .21 Table 3. Comparing select TPB Models from tables 1 and 2 across intervention and control participants. Table 1, Model I Table 1b, Model IX Table 2, Model X Table 2, Model XI Table, 2 Model XII Con- strained a Free Con- strained Free Con- strained Free Con- strained Free Con- strained Free χ² (df) 123.9 (30) 119.9 (24) 117.1 (25) 104.7 (15) 138.2 (25) 129.3 (17) 104.6 (20) 101.3 (13) 121.0 (30) 115.1 (21) Resid PEB Pnor Att Int C/I .26/.26 .93/.70 .83/.86 .55/.52 C/I .26/.26 .93/.71 .83/.86 .54/.50 C/I 24/.26 .93/.70 .83/.86 .54.52 C/I .23/.25 .93/.71 .83/.86 .54/.50 C/I .24/.26 .9/.702 .83/.86 -/- C/I .23/.25 .93/.70 .83/.86 -/- C/I .66/.63 .92/.70 .83/.86 -/- C/I .65/.62 .92/.70 .83/.86 -/- C/I .25/.26 .54/.52 .93/.70 .83/.86 C/I .25/.25 .54/.50 .93/.71 .83/.86 R² PEB Pnor Att Int C/I .70/.68 .15/.15 .11/.07 .30/.26 C/I .69/.70 .14/.16 .12/.06 .32/.31 C/I .72/.68 .15/.15 .11/.07 .30/.27 C/I .73/.71 .14/.15 .12/.06 .32/.31 C/I .73/.70 .14/.16 .11/.08 -/- C/I .73/.72 .14/.16 .12/.07 -/- C/I .22/.20 .14/.15 .11/.08 -/- C/I .22/.26 .14/.16 .12/.07 -/- C/I .70/.68 .30/.27 .14/.16 .11/.08 C/I .70/.70 .32/.31 .14/.16 .12/.07 CFI .80 .80 .81 .81 .73 .73 .56 .54 .81 .80 RMSEA .17 .20 .14 .14 .21 .25 .20 .26 .17 .21 Table note. Resid=residual variance. C=Control group, I=Intervention group. a In Constrained models, regression paths were equated across intervention and control groups, but means and variances were permitted to vary across groups. In Free models, regression paths were freed. 172 Appendix 2C: Fitting Alternative NAM Models Note: Shaded column represents NAM final model. Table 1. NAM baseline and alternative models. NAM Model I Baseline model (figure 2.1) Model II Baseline model plus AC main effect Model III Baseline model plus AR main effect Model IV Baseline model plus AR & AC main effects Model V Baseline model plus AR & AC main effects, remove interaction χ² (df) 10.4 (2) 13.8 (3) 74.0 (3) 80.4 (4) 68.8 (3) R² PEB .21 .22 .19 .20 .20 CFI .90 .87 .51 .49 .53 RMSE A .14 .13 .34 .30 .33 Table 2. Comparing select NAM models from table 1 across intervention and control participants. Table 1, Model I Table 1, Model IV Table 1, Model V Constraine d Free Constraine d Free Constrained Free χ² (df) 17.3 (7) 14.0 (4) 97.4 (13) 93.8 (8) 77.5 (10) 73.5 (6) Resid PEB PNor C/I .75/.65 .93/.70 C/I .74/.63 .93/.70 C/I .73/.66 .92/.70 C/I .72/.63 .92/.70 C/I .32/.65 .92/.70 C/I .72/.63 .92/.70 R² PEB PNor C/I .22/.20 .14/.15 C/I .17/.29 .14/.15 C/I .20/.18 .15/.15 C/I .14/.30 .15/.15 C/I .20/.18 .15/.15 C/I .14/.29 .15/.15 CFI .88 .88 .46 .46 .52 .52 RMSEA .12 .16 .25 .32 .26 .33 Table note. Resid=residual variance. C=Control group, I=Intervention group. In Constrained models, regression paths were equated across groups, but means and variances were permitted to vary across groups. In Free models, regression paths were freed. Shaded column represents the final NAM model 173 Appendix 2D: Fitting Alternative Hybrid Models Note: Shaded column represents final Hybrid model. Table 1. Hybrid Model baseline and alternative models based on a 3-way interaction between PNor, AR, AC → PEB. Model I Baseline model (figure 2.5) Model II Baseline model but remove PNor→Int path Model III Baseline model plus direct AR →PEB path Model IV Model III plus AC→PEB path Model V Model II plus AC →PEB path Model VI Model IV plus PerCont→PEB path Model VII Model IV plus SNor→PEB path χ² (df) 107.4 (17) 141.6 (18) 225.0 (23) 242.4 (29) 125.1 (23) 121.8 (22) 125.0 (22) R² PEB Int .70 .29 .71 .21 .70 .29 .70 .29 .70 .29 .70 .29 .70 .29 CFI .81 .74 .60 .58 .79 .79 .78 RMSEA .16 18 .21 .19 .15 .15 .15 Table 2. Hybrid Model baseline and alternative models based on a 3-way interaction between Int, AR, AC PEB. Model I Baseline model (TPB baseline model with 3- way interaction between Int, AR, AC PEB) Model II Baseline model plus AR and AC main effects →Int Model III Model II plus AR and AC main effects →Int Model IV Model III minus AC main effect Model V Model III minus AR main effect Model VI Model V plus direct PerCont → PEB path Model VII Model V plus direct PNor → PEB path Model VIII Model V plus direct SNor → PEB path χ² (df) 123.0 (18) 250.3 (30) 166.2 (24) 158.9 (21) 129.9 (21) 126.8 (20) 129.9 (20) 129.9 (20) R² PEB Int .70 .29 .70 .29 .70 .29 .70 .29 .70 .29 .70 .29 .70 .29 .71 .21 CFI .78 .58 .73 .73 .78 .78 .77 .70 RMSEA .17 .19 .17 .18 .16 .16 .14 .19 174 Table 3. Hybrid Model baseline and alternative models based on a 3-way interaction between 3-way interaction between PNor, AR, AC → Int. Model I Baseline model (TPB baseline model with 3- way interaction between PNor, AR, AC → Int). Model II Same as Model I above plus AR and AC main effects →Int Model III Model II minus AC main effect Model IV Model II minus AR main effect Model V Model IV plus direct PerCont →PEB path Model VI Model IV plus direct PNor→P EB path Model VII Model IV plus direct SNor→PE B path χ² (df) 107.7 (18) 241.0 (30) 224.4 (24) 124.5 (24) 121.4 (23) 124.5 (23) 124.4 (23) R² PEB Int .70 .29 .69 .26 .69 .26 .70 .29 .70 .29 .70 .29 .70 .29 CFI .81 .58 .60 .79 .79 .79 .79 RMSEA .16 .19 .20 .14 .14 .15 .15 Table 4. Comparing select Hybrid Models from tables 1, 2, and 3 across intervention and control participants. Table 1, Model I Table 2, Model I Table 3, Model I Con- strained Free Con- strained Free Con- strained Free χ² (df) 363.5 (73) 356.1 (62) 153.1 (47) 141.9 (36) 166.4 (47) 158.8 (36) Resid PEB Int PNor Att C/I .26/.26 .55/.52 .93/.70 .83/.86 C/I .25/.26 .54/.50 .93/.70 .83/.86 C/I .26/.26 .56/.53 .94/.72 .84/.88 C/I .25/.26 .54/.50 .94/.72 .84/.88 C/I .26/.26 .56/.53 .94/.72 .84/.88 C/I .25/.26 .55/.51 .94/.72 .84/.88 R² PEB Int PNor Att C/I .70/.68 .30/.27 .14/.16 .11/.08 C/I .70/.71 .32/.31 .14/.16 .12/.06 C/I .71/.68 .30/.27 .14/.15 .11/.08 C/I .71/.71 .34/.35 .14/.15 .12/.07 C/I .71/.69 .30/.27 .14/.15 .11/.08 C/I .71/.72 .32/.32 .14/.15 .12/.07 CFI .57 .56 .78 .78 .76 .75 RMSEA .20 .22 .15 .14 .16 .18 Table note. Resid=residual variance. C=Control group, I=Intervention group. In Constrained models, regression paths were equated across groups, but means and variances were permitted to vary across groups. In Free models, regression paths were freed. Shaded column represents the final NAM model 175 Appendix 2E: Sample Bias Analyses Table 1. Means, standard deviations, and correlations among PEB and Int variables at Time 1 and time 2 among different sample sizes. All individuals with T1 PEB Individuals with T1 and T2 PEB Complete data on all key measures at t1 and t2 n 278 206 179 PEB1↔Int1 .86 .84 .83 PEB2↔Int2 .87 N=203 .87 N=203 .90 N=179 PEB1 49.44 (8.48) 48.84 (8.05) 48.31 (7.91) PEB2 50.53 (8.47) N=206 50.53 (8.47) 50.13 (8.23) Int1 55.62 (9.24) 55.20 (8.81) 54.85 (8.59) Int2 55.14 (9.20) N=206 55.14 (9.20) 54.76 (9.20) Time 1 correlation matrices for the three groups Correlations among all key time 1 variables were computed within each of the three possible samples. Pairs of correlations were then compared across the samples and differences of greater than .05 were indicated in the correlation matrices below. Of 108 comparisons, a total of 3 pairs of correlations differed by more than .05. The difference was .06 among these three pairs. This indicates that the set of relationships between key variables among sample selected for analysis was roughly equivalent to the set of relationships among other possible samples. We therefore concluded that results would be unlikely to differ if based on alternative sample sizes. 176 Table 2. Correlations among key time 1 variables based on n=206. PEB Int PNor SNor Att Ex Out Per Cont AR AC PEB 1.0 Int .84 1.0 PNor .46 .55 1.0 SnNr .25 .31 .37 1.0 Att .26 .28 .39 .35 1.0 ExOut .33 .42 .54 .38 .31 1.0 PerCont .41 .41 .46 .39 .22 .49 1.0 AR .32 .40 .61 .40 .35 .55 .35 1.0 AC .08 .11 .06 .23 . 28 .05 .07 .17 1.0 Table note. Table 2-Table 3 differences of more than .05 are shown in green text. Table 2-table 4 differences of more than .05 are shown in red text. Table 3. Correlations among key time 1 variables based on n=278. PEB Int PNor SNor Att Ex Out Per Cont AR AC PEB 1.0 Int .86 1.0 PNor .49 .56 1.0 SnNr .27 .33 .38 1.0 Att .26 .30 .42 .36 1.0 ExOut .37 .44 .48 .35 .32 1.0 PerCont .39 .37 .43 .41 .22 .43 1.0 AR .34 .39 .63 .43 .34 .51 .38 1.0 AC .05 .08 .07 .18 .27 .10 .04 .16 1.0 Table note. Table 2-Table 3 differences of more than .05 are shown in green text. Table 3-table 4 of more than .05 differences are shown in yellow highlighting. 177 Table 4. Correlations among key time 1 variables based on n=179. PEB Int PNor SNor Att Ex Out Per Cont AR AC PEB 1.0 Int .83 1.0 PNor .46 .54 1.0 SnNr .26 .31 .38 1.0 Att .27 .29 .40 .36 1.0 ExOut .35 .44 .54 .38 .30 1.0 PerCont .40 .40 .47 .39 .22 .47 1.0 AR .34 .42 .60 .42 .34 .56 .39 1.0 AC .08 .11 .10 .21 .25 .11 .05 .21 1.0 Table note. Table 2-Table 4 differences of more than .05 are shown in red text. Table 3-table 4 of more than .05 differences are shown in yellow highlighting. 178 Appendix 2F: Validity Verification To examine the performance of each scale as a single factor, 1-factor confirmatory factor analyses were performed. For each scale, raw item scores were loaded onto a single factor. A constrained model was run first with item loadings and residuals constrained to be equal. A second model was then run with freed loadings and residuals. Several model fit indices were considered in evaluating how well each measure was represented by a single factor, including χ², CFI, and RMSEA. Item R², item residual variance, and factor loadings were also considered in evaluating factor structure. Comparing fit indices and model parameters across the Constrained and Free models for each measure served as another basis for evaluating how well the items performed together as a scale. Model fit indices and standardized model parameters are reported in the tables below. Model fit indices indicated that overall, the 1-factor models had considerable misfit, with χ² and RMSEA values greater than 360 and .90, and CFI values less than .70. The Free models generally performed slightly better than the Constrained models, but item R² values did not improve considerably in the Free relative to the Constrained models. This suggests that permitting item loadings to vary did not increase the explained variance in the items beyond what was explained given equal items loadings, as in an observed scale score. The model misfits are consistent with the fact that the measures were designed to capture a variety of PEB. Across all scales, item R² values in the Constrained models ranged from 11-.78. This suggests that overall, 1-factor structures 179 with equal item loadings performed reasonably well in explaining item variances. Therefore, we proceeded with observed scale scores as model constructs. Note: all confirmatory factor analysis model parameters are standardized. Modified Schultz PEB Scale, 17 items Constrained Free χ² (df) 796 (151) 339.52 (120) CFI 0 .61 RMSEA .14 .09 Item loadings .37 -.08-1.0 Item residuals 1.12 .35-2.04 Item R² .11-.47 .002-.53 Intentions Scale, 17 items Constrained Free χ² (df) 632.75 (151) 360.19 (120) CFI .34 .67 RMSEA .12 .10 Item loadings .46 -1.0 .04-1.0 Item residuals 1.05 .62-1.81 Item R² . 17-.49 .001-.53 Social Norms Scale, 17 items Constrained Free χ² (df) 1432.97 (151) 1100.35 (120) CFI .52 .63 RMSEA .21 .20 Item loadings .79-1.0 .63-1.0 Item residuals .29 .11-.50 Item R² .68-.78 .58-.86 180 Personal Norms Scale, 17 items Constrained Free χ² (df) 1405 (151) 1080 (120) CFI .59 .68 RMSEA .21 .20 Item loadings 1.03 .83-1.15 Item residuals .96 .32-1.74 Item R² .51-.53 .28-.78 Perceived Control Scale, 17 items Constrained Free χ² (df) 1421.42 (151) 784.59 (120) CFI .29 .63 RMSEA .21 .17 Item loadings .73-1.0 .64-1.0 Item residuals .61 .21-1.49 Item R² .47-.62 .22-.83 Expected Outcome Scale, 17 items Constrained Free χ² (df) 2148.89 (151) 1122.83 (120) CFI .18 .59 RMSEA .25 .20 Item loadings .82-1.0 .28-1.18 Item residuals 1.08 0.19-3.02 Item R² .39-.48 .03-.79 New Ecological Paradigm, 15 items Constrained Free χ² (df) 549.83 (118) 447.86 (91) CFI .45 .54 RMSEA .14 .14 Item loadings .54 .24-1.0 Item residuals .70 .45-1.0 Item R² .30-.59 .06-.57 181 Ascribed Responsibility Scale, 17 items Constrained Free χ² (df) 1682.83 (151) 1286.09 (120) CFI .59 .69 RMSEA .23 .22 Item loadings 1.04-1.0 .86-1.13 Item residuals .80 .31-1.91 Item R² .58-.56 .29-.79 Sundblad Climate Change Quiz, 13 items Constrained Free χ² (df) 121.84 (43) 113.94 (38) CFI 0.0 .62 RMSEA .09 .06 Item loadings 0.27-.41 -.55 - .71 Item residuals .93-.83 -- Item R² .07 .008-.51 Personal Norms and Ascribed Responsibility construct validity Initial plans included creating latent factors to represent the personal norms and ascribed responsibility constructs. The personal norms construct was assessed using two additional existing scales beyond the one created by the first author. A 4-item measure developed by Garling et al. (2003) assessed the extent to which participants felt morally obligated to protect the environment in general. Responses were provided on a 7-point Likert scale ranging from “strongly disagree” to “strongly agree”. As well, Clark’s 3-item scale (2003) asked respondents to rate statements about the extent to which they felt a moral duty to help with general societal and environmental problems. Responses were given on a 7-point Likert scale ranging from “strongly disagree” to “strongly agree”. 182 Two additional existing scales were included to assess ascribed responsibility. Garling et al.’s measure (2003) included 3 items that asked about feelings of obligation to protect the environment. Responses were given on a 7-point Likert scale ranging from “strongly disagree” to “strongly agree”. Clark’s 3-item scale (2003) assessed ascribed responsibility for general societal issues. Responses were given on a 7-point Likert scale ranging from “strongly disagree” to “strongly agree”. Time 1 scores among the three scales assessing personal norms and the three scales assessing ascribed responsibility were highly intercorrelated (see table 1 below). A series of factor analyses were therefore conducted to examine construct validity among the scales. Items included in each scale are listed at the end of this Appendix. Table 1. Pearson correlations among scales assessing personal norms and ascribed responsibility (n=183). Personal norms PN Clark PN Garling ARZ AR Clark AR Garling Personal norms 1.00 0.28 0.63 0.63 0.36 0.40 PN Clark 1.00 0.41 0.30 0.54 0.42 PN Garling 1.00 0.55 0.51 0.67 ARZ 1.00 0.37 0.43 AR Clark 1.00 0.57 AR Garling 1.00 Table note. P<.0001 for all. First, 1-factor confirmatory factor analyses were performed for each scale, following the exact procedures described above. Model fit indices and standardized model parameters are reported in the tables below. Model fit indices indicated that 183 overall, the 1-factor models had considerable misfit, with χ² and RMSEA values greater than 360 and .90, and CFI values less than .70. The Free models generally performed slightly better than the Constrained models, but item R² values did not improve considerably in the Free relative to the Constrained models. This suggests that permitting item loadings to vary did not increase the explained variance in the items beyond what was explained given equal items loadings, as in an observed scale score. The model misfits are consistent with the fact that the measures were designed to capture a variety of PEB. Across all scales, item R² values in the Constrained models ranged from 11 -.78. This suggests that overall, 1-factor structures with equal item loadings performed reasonably well in explaining item variances. Therefore, we proceeded with observed scale scores as model constructs. Garling Personal Norms Scale, 4 items Constrained Free χ² (df) 30.5 (8) 4.02 (3) CFI .95 .99 RMSEA .12 .04 Item loadings .75-1.0 .64-1.0 Item residuals .28 .16-.33 Item R² .67-.78 .55-.84 Clark Personal Norms Scale, 3 items Constrained Free χ² (df) 1.50 (3) 0.70 (1) CFI 1.0 1 RMSEA 0.00 0 Item loadings .37-1.0 .46-1.0 Item residuals 1.34-2.42 1.27-2.37 Item R² .05-.43 .08-.40 184 Garling Ascribed Responsibility Scale, 3 items Constrained Free χ² (df) 54 (4) 1.1 (1) CFI .20 .99 RMSEA .25 .02 Item loadings .71 .53-1.0 Item residuals 1.1 .61-1.9 Item R² .32-.48 .13-.61 Clark Ascribed Responsibility Scale, 3 items Constrained Free χ² (df) 27.7 (4) 4.9 (1) CFI .20 .87 RMSEA .17 .14 Item loadings .59-1.0 .34-1.0 Item residuals 1.25 .99-1.65 Item R² .22-.44 .10-.50 Partialing out method variance Items that formed the Personal Norms, Clark PN, Clark AR, Garling PN, and Garling AR scales shared the same 7-point (strongly disagree- strongly agree) response format, which may account for shared variance. A 4-factor structure was specified whereby variance attributable to shared response format was partitioned out. Items sharing the same response format were loaded onto one latent factor; items from the Ascribed Responsibility Scale, which had a different response format, were loaded onto another. Item 2 from the Clark Personal Norms Scale was dropped due to lack of association with other items (intercorrelations with 94% of other items ranged from -.10 – .13; three intercorrelations ranged from .16 - .32). Two additional latent factors were created to represent the ascribed responsibility and personal norms constructs. See figure 1 below for a visual representation of the 4-factor model. Factor loadings were 185 Response format 1 Response format 2 PNG items 1-4 Ascribed Responsibility Scale items 1-17 Personal Norms Scale items 1-17 ARC items 1-3 ARG items 1-3 PNC items 1-3 Personal Norms Ascribed Responsibility constrained to be equal within each of the four factors in the model, and correlations between the method factors and other latent factors in the model were set to zero. The results of the confirmatory factor analysis indicated that the 4-factor model resulted in considerable misfit to the data (χ² (1073) =5056.67, CFI=0.65, RMSEA=0.12). Freeing the item loadings improved the fit of the 4-factor model (χ² (944) =3010.6.56, CFI=0.76, RMSEA=.11). The latent constructs correlated at .62. Specifying a 2-factor model that represented just the latent theoretical constructs PNor and AR but did not account for shared response format resulted in a decrement in fit when loadings were constrained (χ² (1076) =4973.48, CFI=0.55, RMSEA=0.14), and freed (χ² (1033) =4423.38, CFI=.61, RMSEA=.13). Figure 1. 4-factor model of ascribed responsibility and personal norms with method variance partialled out. 186 Exploratory factor analyses A set of follow-up exploratory factor analysis was conducted including all of the items from the six scales (except the dropped item from the Clark Personal Norms Scale). Eigenvalues ranged from 19.13 for the first factor to 1.06 for the seventh factor, with values below 1 for additional factors. As our interests were in distinguishing between the personal norms and ascribed responsibility constructs, we examined estimates for the 2- factor model. The misfit was considerable (χ² (944) = 4139, CFI=0.63, RMSEA=0.13). Items from all three of the personal norms scales loaded onto the first factor with Geomin rotated item loadings ranging from 0.28-0.89. As well, all but one item each from the Garling and Clark Ascribed Responsibility scales loaded onto the first factor, with loadings ranging from 0.27-0.73. One item each from the AR Garling (#3) and AR Clark (#1) scales did not load onto either factor (loadings less than .20). The items from the author’s Ascribed Responsibility Scale loaded strongly onto the second factor, with Geomin rotated item loadings ranging from 0.44-0.95. Whereas, the author’s scales were well distinguished by the two factors, the Clark and Garling Ascribed Responsibility scales shared variance with the author’s Personal Norms Scale. This suggests that the Clark and Garling scales did not perform well with the authors’ scales. Evaluation of the Theory of Planned Behavior is sensitive to the level of specificity at which the measures assess the construct of interest. The Ascribed Responsibility and Personal Norms scales developed by the first author reflected the specific behaviors contained in the behavior outcome measure, whereas the Garling and Clark scales assessed constructs at a more general level. Combining items that assess at 187 the general and specific levels may be considered incompatible measurement of the construct and may not produce a meaningful factor for use in the TPB model. Prior work demonstrated that when TPB model constructs were not compatible, the model explained 3.2% of the variance in behavior compared to 50.2% when the constructs were compatible (Kaiser et al., 2007). We therefore elected to use the scales developed by the authors that corresponded to the items in the behavior outcome scale. 2-factor CFA with authors’ scales To ensure discriminant validity of the constructs, a confirmatory 2-factor analysis was conducted using items from the author’s Ascribed Responsibility and Personal Norms scales. Item loadings were allowed to vary. See the table below for additional details. Table 2. 2-factor confirmatory factor analysis of personal norms and ascribed responsibility using author-developed scale items. χ² (df) 3127 (527) CFI 0.64 RMSEA 0.16 Personal Norms Ascribed Responsibility Item loadings 0.82-1.14 0.87-1.14 Item residuals 0.33-1.72 0.32-1.90 Item R² 0.28-0.77 0.29-0.78 Factor correlation .61 188 Appendix 2G: Latent Difference Score Model Outcome [t2] ∆Outcome Outcome [t1] 1 1 1 1 α 0 σ 1d σ d 2 µ 1 σ 1 2 189 Appendix 2H: Testing for Differences in Model Parameters Across Intervention and Control Groups Based on Change Models Model I. Latent variable means and variances set equal across groups. All regression paths set equal across groups. Observed variable means and variances and correlations free across groups. Model II. Latent variable means and variances set equal across groups. Observed variable means and variances, correlations, and regression paths free across groups. Model III. All regression paths set equal across groups. Correlations free across groups. Observed and latent variable means and variances free across groups. Model IV. All parameters free across groups. Testing differences in model parameters across groups based on TPB final change model (Table 3d, Model IV). Model I Model II Model III Model IV χ² (df) 394.1 (159) 378.5 (147) 384.6 (146) 370.1 (134) Resid PEB ∆PEB Int ∆Int PNor ∆PNor Att ∆Att C/I .25/.26 .26 .55/.53 .39 .92/.70 .52 .80/.92 .49 C/I .25/.26 .26 .54/.52 .38 .93/.70 .52 .80/.92 .49 C/I .25/.25 .26/.22 .55/.52 .43/.35 .92/.70 .52/.52 .82/.88 .60/.34 C/I .25/.25 .26/.22 .55/.51 .42/.34 .92/.70 .52/.53 .82/.87 .60/.35 R² PEB ∆PEB Int ∆Int PNor ∆PNor Att ∆Att C/I .74/.71 .46 .38/.34 .11 .15/.19 .04 .11/.07 .002 C/I .72/.74 .43/.49 .40/.39 .15/.13 .14/.21 .04/.05 .11/.09 .001/.04 C/I .74/.72 .47/.47 .37/.35 .11/.11 .16/.28 .05/.03 .11/.08 .002/.003 C/I .72/.75 .45/.49 .39/.39 .15/.13 .15/.20 .04/.04 .11/.09 .00/.02 CFI .86 .86 .86 .86 RMSEA .12 .12 .13 .13 190 Testing differences in model parameters across groups based on NAM final change model (Table 4, Model V). Model I Model II Model III Model IV χ² (df) 169.5 (45) 159.8 (40) 138.1 (37) 128.2 (32) Resid PEB ∆PEB PNor ∆PNor C/I .82./73 .44 .92/.69 .52 C/I .80/.73 .43 .92/.69 .52 C/I .83/.73 .47/.39 .92/.69 .52/.52 C/I .81/.72 .46/.38 .92/.69 .52/.52 R² PEB ∆PEB PNor ∆PNor C/I .15/.13 .03 .15/.18 .04 C/I .09/.25 .01/.14 .14/.19 .03/.05 C/I .16/.12 .04/.04 .16/.16 .05/.03 C/I .09/.24 .02/.11 .15/.18 .04/.04 CFI .77 .78 .82 .83 RMSEA .16 .17 .16 .17 191 Appendix 3A: Behavior and Intentions Measures Behavior scale stem question How often have you done each of the following in the past 2 weeks? 1: Never 2: Rarely 3: Sometimes 4: Fairly Often 5: Very Often Intentions scale stem question I intend to ___ in the next few weeks: 1: very unlikely 2: unlikely 3: undecided 4: likely 5: very likely Expected outcome scale stem question If I did each of the following behaviors in the next few weeks, it would be ___: 1: very bad 2: bad 3: poor 4: neither good nor bad 5: fair 6: good 7: very good 192 Block 1 items (BEHAV1 and INTENT1) 1. Recycled Paper 2. Looked for Ways to reuse things 3. Volunteered time to help an environmental group 4. Wrote a letter or email, or endorsed a petition on a website in support of an environmental issue 5. Encouraged friends or family to recycle *6. Turned up the heat on thermostat when I felt cold or turned on the air conditioner when I felt hot 7. Composted food scraps 8. Used cold water to wash clothes *9. Left lights on when I was not in a room Block 2 items (BEHAV2 and INTENT2) 10. Recycled cans or bottles 11. Purchase products in reusable containers 12. Donated money to an environmental group 13. Picked up litter that was not my own *14. Thrown away leftover food on my plate that I didn’t eat 15. Conserved gasoline by walking or bicycling *16. Let the water run while I was brushing my teeth 17. Shut off my computer when I was not using it *Indicates reverse-coded item. Note: All intentions and expected outcomes items were phrased using future tense. Items were administered together for expected outcome scale (not in blocks). 193 Appendix 3B: Intentions-Behavior Correlations Pearson correlations between behavior and intention items with matching item content (see Appendix 3A for item content and numbering). Item number r 1 .53 2 .46 3 .61 4 .56 5 .70 6 .81 7 .69 8 .77 9 .54 10 .78 11 .70 12 .50 13 .73 14 .75 15 .75 16 .78 17 .80 p<.0001 for all correlations. Pearson correlations among behavior and intentions scales and subscales. Behavior BEHAV1 BEHAV2 Intentions .86 .77 .74 INTENT1 .77 .80 .54 INTENT2 .79 .59 .84 Table note. p<.0001 for all correlations.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Microsoft Word - $ASQ102980_supp_undefined_52801B5E-C057-11E0-8B9D-4B3E9E1A67F9.doc

PDF

Microsoft Word - $ASQ113905_supp_undefined_834F9CCE-E7FC-11E0-A829-C4119E1A67F9.doc

PDF

Microsoft Word - $ASQ114028_supp_undefined_08DDE1C4-F37A-11E0-B76D-D447F0E6BF1D.doc

PDF

Microsoft Word - $ASQ112217_supp_undefined_79E4D690-D4DD-11E0-B309-1C08D352ABB1.doc

PDF

Microsoft Word - $ASQ102667_supp_undefined_3CA250BE-B895-11E0-ACB2-80713012225A.doc

PDF

Microsoft Word - $ASQ116013_supp_665C9512-F2E8-11E0-ACCD-A3E19D1A67F9.docx

PDF

Microsoft Word - $ASQ102919_supp_undefined_20865796-B998-11E0-A65C-BB2C9E1A67F9.docx

PDF

Microsoft Word - $ASQ101738_supp_undefined_9EEDD572-A94A-11E0-81A6-334B9E1A67F9.docx

PDF

Microsoft Word - $ASQ103248_supp_undefined_559BEF38-B8A1-11E0-A10A-3F4DD352ABB1.doc

PDF

Microsoft Word - $ASQ102814_supp_undefined_27ABC756-B35B-11E0-9A05-5A68F0E6BF1D.docx

PDF

Microsoft Word - $ASQ102797_supp_undefined_C82BF10E-B24A-11E0-AD7F-D002D352ABB1.docx

PDF

Microsoft Word - $ASQ111440_supp_undefined_BA0AA86A-D11A-11E0-B6B3-961F3012225A.doc

PDF

Microsoft Word - $ASQ111152_supp_undefined_9F9527E2-D329-11E0-A416-F092F0E6BF1D.doc

PDF

Microsoft Word - $ASQ104889_supp_undefined_F56782CC-AD91-11E0-8111-0B1FF0E6BF1D.doc

PDF

Microsoft Word - $ASQ103071_supp_undefined_B6A045E2-B782-11E0-867D-2FDC9D1A67F9.rtf

PDF

Microsoft Word - $ASQ116369_supp_undefined_7757EFF4-F5CE-11E0-B683-4781F0E6BF1D.doc

PDF

Microsoft Word - $ASQ102640_supp_undefined_AC4743C2-EA07-11E0-BA63-ED45D352ABB1.doc

PDF

Microsoft Word - $ASQ100064_supp_220A1D06-9AC5-11E0-A04F-7A6C3012225A.docx

PDF

Microsoft Word - $ASQ117920_supp_undefined_36A1D492-099D-11E1-858F-A02A9E1A67F9.docx

PDF

Microsoft Word - $ASQ117979_supp_undefined_EEF05D74-0FED-11E1-BA93-74232E1BA5B1.doc

Asset Metadata

Core Title Microsoft Word - $ASQ100489_supp_undefined_9C841A88-B234-11E0-8B96-E32D9E1A67F9.doc

Tag OAI-PMH Harvest

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC1348304

Unique identifier UC1348304

Legacy Identifier etd-SintovNico-137