University of Southern California Dissertations and Theses

Evaluating social-cognitive measures of motivation in a longitudinal study of people completing New Year's resolutions to exercise

(USC Thesis Other)

Evaluating social-cognitive measures of motivation in a longitudinal study of people completing New Year's resolutions to exercise

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Running head: SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE Evaluating Social-Cognitive Measures of Motivation in a Longitudinal Study of People Completing New Year’s Resolutions to Exercise Andrew L. Larsen Department of Psychology University of Southern California Dissertation for the Degree Doctor of Philosophy (Psychology) May, 2015 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 2 Table of Contents Abstract ............................................................................................................................................4 Introduction ......................................................................................................................................7 Importance of self-regulation .......................................................................................................8 Social-cognitive models of self-regulation.................................................................................10 Improving the social-cognitive models of self-regulation .........................................................13 Current study ..............................................................................................................................22 Hypotheses .................................................................................................................................26 Methods..........................................................................................................................................30 Participants .................................................................................................................................30 Procedures ..................................................................................................................................33 Intervention ................................................................................................................................34 Measures .....................................................................................................................................35 Analysis ......................................................................................................................................44 Results ............................................................................................................................................53 Initial recruitment and first round of participant exclusion ........................................................53 Second round of participant exclusion .......................................................................................54 Tests of intervention effects .......................................................................................................55 Goal-structure analysis ...............................................................................................................71 Measurement invariance of motivation latent variable ..............................................................75 Predictability of motivation latent variable ................................................................................78 Discussion ......................................................................................................................................91 Intervention outcomes ................................................................................................................91 Goal-structure outcomes ..........................................................................................................101 Motivation latent variable measurement invariance outcomes ................................................106 Predictability of motivation latent variable ..............................................................................108 Overall summary ......................................................................................................................113 Limitations ...............................................................................................................................116 Conclusions ..............................................................................................................................117 References ....................................................................................................................................119 Tables ...........................................................................................................................................136 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 3 Figures..........................................................................................................................................177 Appendix A .................................................................................................................................202 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 4 Abstract Overweight and obesity are major problems in developed countries around the world, and exercise is strongly associated with weight loss and weight control. Social-cognitive models have been used to study self-regulation of health behaviors, but they suffer from several limitations, including the use of intentions as a sole measure of motivation. There are validated motivational constructs from literature outside the health domain that could potentially combine to form a latent variable of motivation that would benefit social-cognitive models of self-regulation. It would provide a more accurate assessment of motivation and provide direct avenues of intervention that have yet to be fully explored in health research. On an adjusted longitudinal sample of 713 participants attempting to increase their vigorous and/or moderate exercise as part of a New Year’s Resolution, the present study had four aims. First, it utilized a randomized controlled design to test a novel intervention designed to increase exercise behavior based on the motivational concept goal-structure. Second, it evaluated the Chulef, Read, and Walsh (2001) goal taxonomy for assessing participant goal-structure in regards to exercise. Third, it evaluated whether three motivational constructs—goal autonomy, goal commitment, and goal-structure— form a robust latent variable. Finally, it tested whether the latent variable of motivation predicted exercise behavior over and above intentions. There were three time-points of data collection that started in January and occurred at 4-week intervals. Intervention effects immediately post intervention were tested via independent samples t-tests, and changes over time were tested via multi-group latent change score models. The intervention successfully impacted participant goal- structure post-intervention, however these effects failed to spill over into other constructs. Additionally, both Intervention and Control groups significantly increased their physical activity throughout the course of the study, as well as their perceptions of their intentions for exercise, SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 5 goal commitment, goal autonomy, and a few goal-structure scores, however there were few significant differences observed across groups. The utility of the Chulef, Read, and Walsh (2001) goal taxonomy for measuring goal structure in relation to exercise was tested using correlations, regressions, and exploratory factor analyses on five sub-samples of data. These analyses revealed that two goal-structure scores—(1) the degree exercise facilitates achieving other life goals and (2) the rank of a heath goal-cluster in comparison to other goal-clusters—stood out as potential useful additions to a latent variable measuring motivation. The robustness of the motivation latent variable derived from the motivation constructs was evaluated using longitudinal measurement invariance modeling testing for stable latent variable factor loadings, indicator means, and error variances over time and across gender and study group. The tests of measurement invariance showed that a single latent variable comprised of four others (goal commitment, goal autonomy, facilitation of exercise, and the rank of the heath goal-cluster) exhibited a good fit to the data across three time-points of measurement. Finally, a latent variable path analysis was used to test if the motivation latent variable predicted exercise behavior over and above intentions. After adjustment for missing data, the motivation latent variable at Time 1 predicted vigorous exercise at Time 2, vigorous/moderate exercise at Time 2, and success in accomplishing vigorous/moderate exercise intentions at Time 2, and the motivation latent variable at Time 2 significantly predicted vigorous exercise at Time 3, vigorous/moderate exercise at Time 3, and success in accomplishing vigorous/moderate exercise intentions at Time 3. Although the intervention failed to create significant differences in exercise across study groups, the findings support using at least some of the motivation constructs within the traditional social-cognitive frameworks. Future research should continue exploring intervention options based on the motivation constructs, update the method for measuring goal-structure SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 6 using the Chulef, Read, and Walsh (2001) goal taxonomy, and continue testing the relationships under study. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 7 Introduction Overweight and obesity continue to be major problems in developed countries around the world (Flegal, Carroll, Kit, & Ogden, 2012; Flegal, Graubard, Williamson, & Gail, 2005; Ogden, Carroll, Kit, & Flegal, 2012). An increasing focus has been placed on understanding and improving people’s self-regulatory ability in order to treat obesity and other health problems stemming from accumulated effects of maladaptive behaviors over time (Bandura, 2005). Social- cognitive models of self-regulation have attempted to explain self-regulation from a rational thinker perspective, but they suffer from a number of limitations (Armitage, 2009; Cervone, Shadel, Smith, & Fiori, 2006; Leventhal & Mora, 2005; Maes & Karoly, 2005; Sheeran, Gollwitzer, & Bargh, 2012; Sniehotta, Presseau, & Araújo-Soares, 2014). One issue in particular is the use of behavioral intentions as the sole measure of a person’s underlying motivation for self-regulation (Rhodes & Dickau, 2013). The present research seeks to improve the traditional social-cognitive models of self-regulation by developing a measure of motivation stemming from validated motivational constructs from literature outside of the health domain that more accurately gauges a person’s underlying motivation. Doing so has at least two practical implications. First, a measure of motivation that more accurately assesses who is and who is not likely to succeed during a behavior change attempt allows for intervention resources to be targeted to those who need them most (Armitage, 2009; Bandura, 2005; Knott, Muers, Aldridge, & Britain, 2008). Behavioral intentions, on the other hand, fail to capture several important aspects of motivation resulting in inaccurate measurements and relatively poor prediction of success over time. Secondly, measuring motivation with previously validated constructs that are amenable to direct intervention allows for the development of new behavior change interventions that have not been utilized in the social-cognitive literature to date. In comparison, behavioral SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 8 intentions provide no clear avenue for direct intervention on one’s motivation, which limits attempts at improving one’s motivation to intervening indirectly via other constructs. It is important to have effective guides for changing cognitive predictors of behavior in order to improve health outcomes (Bandura, 2005; Rothman, 2004). Having no clear avenue of intervention on intentions, the primary conceptualization of motivation, is a major weakness of social-cognitive models. Importance of self-regulation Self-regulation can be defined as: “a multi-component, multi-level, iterative, self-steering process that targets cognitions, affects, and actions, as well as features of the environment for modulation in the service of one’s goals” (Boekaerts, Maes, & Karoly, 2005, p. 150). Self- regulation is arguably ongoing at all times, and often operates outside of conscious attention (Bargh, 1997; Baumeister, Masicampo, & Vohs, 2011; Karoly, 1999). It is commonly recognized that all people fail to control their behavior in trying to achieve something they desire. Widespread failings of self-regulation are an interesting problem, as self-regulation implies one already has some knowledge of how to be successful. The problem is usually not learning a difficult skill or gaining important knowledge, it lies in motivating people to do behaviors that they already know how to do, and most interestingly, behaviors that they already want to do. For example, to be a healthy weight it is generally understood that one should eat healthy and exercise. However, many people who self-enroll in health clubs and purchase gym memberships or expensive diet programs have difficulty doing so. Furthermore, it is common for people to continue trying to lose weight repeatedly, even if they have experienced little success in the past (Polivy & Herman, 2002). Yet, in this context of consistent failure, there are thousands of people who successfully lose substantial amounts of weight and maintain their weight loss indefinitely SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 9 (Klem, Wing, McGuire, Seagle, & Hill, 1997; Thomas, Bond, Hill, & Wing, 2011; Wing & Phelan, 2005). Self-regulating health is particularly important in today’s society, as many costly diseases result from failing to undertake ongoing preventative behaviors such as eating a healthy diet and exercising (Flegal et al., 2012; Flegal et al., 2005; Silventoinen et al., 2004). These problems are further exacerbated because rates of childhood health problems are rising while people are simultaneously living longer lengthening the amount of time for minor health problems to develop into chronic diseases (Bandura, 2005; Ogden et al., 2012). Improving the nation’s health would have tremendous effects on the overall well-being of the population and the economy. It is well known that regular exercise is one of two major preventive behaviors for managing a person’s weight, the other being a healthy diet (Kyle, Zhang, Morabia, & Pichard, 2006). The Center for Disease Control and Prevention (CDC) recommends adults exercise at a vigorous-intensity (heart beating rapidly, sweating) for 75 minutes a week, or at a moderate- intensity (not exhausting, light perspiration) for 150 minutes a week (see, http://www.cdc.gov). About 90% of participants in the National Weight Loss Registry, a group of people who have lost at least 10% of their body weight and kept it off for over a year, reported that they included exercise as a major component of their weight loss regime (Klem et al., 1997; Thomas et al., 2011). Successful weight loss maintainers also report expending close to 2,571 kcal during moderate-to-vigorous physical activity per week in order to maintain their weight loss (Klem et al., 1997; Thomas et al., 2011). Exercise has also been linked to numerous health and psychological benefits beyond weight loss/management alone, including reduced risk of cancers, increased life-expectance (regardless of weight), and treating depression, to name a few(Blair &Morris, 2009; Fox, 1999; Warburton, Nicol, & Bredin, 2006).For example, approximately 25% SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 10 of cancer cases can be attributed to excess weight and physical activity, and it is estimated that physical activity has a median risk reduction up to 30% for certain kinds of cancers (Physical Activity Guidelines Committee, 2008; Vainio & Bianchini, 2002). Yet, the majority of the United States adult population is generally inactive (Nelson, Story, Larson, Neumark-Sztainer, & Lytle, 2008). This inactivity is probably not due to a lack of awareness of the importance of exercise—as evidenced by a multi-billion dollar a year market for exercise and weight loss related products and services. It is instead a problem in which people fail to self-regulate their behavior. Regular leisure-time exercise (i.e., exercise undertaken during one’s free time) requires ongoing self-regulation. People must schedule a time to complete it, prepare for it in advance, push themselves when they get tired, motivate themselves on days they lack energy, and sometimes recover from injuries. It is clearly a difficult endeavor as large numbers of people frequently attempt to increase their physical activity—for example, making a New Year’s Resolution to do so at the beginning of the year—and only a few people succeed (Klem et al., 1997; Polivy & Herman, 2002; Thomas et al., 2011). These characteristics of regular leisure-time exercise make it a suitable behavior for studying the nature of self-regulation and motivation. Social cognitive models of self-regulation Social cognitive models are a common theoretical approach used to study self-regulation. The basic concept shared among these models is that social behavior is best understood by evaluating peoples’ beliefs, perceptions, and representations of themselves in a social context (Rutter & Quine, 2002). Furthermore, these cognitions are believed to mediate many other determinants of behavior, including demographics, social factors, cultural values, self-esteem, and others (Conner & Norman, 2005). It is assumed that social cognitions are more amenable to SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 11 intervention than most other determinants of behavior, which is a primary reason they are frequently studied (Conner & Norman, 2005). Over the last 50 years, several social cognitive models have been closely examined in the literature, including the Health Belief Model (Janz & Becker, 1984; Rosenstock, 1966, 2000), Protection Motivation Theory (Maddux & Rogers, 1983; Rogers, 1975, 1983), the Theory of Reasoned Action and Theory of Planned Behavior (Ajzen, 1991; Ajzen & Fishbein, 1980; Fishbein & Ajzen, 1975), Social Cognitive Theory (Bandura, 1997; Bandura, 1982), and the Transtheoretical Model of Change (Prochaska & DiClemente, 1984). These models can be broken down into two broad categories: stages of change models and continuum models. The Transtheoretical Model of Change is a stage of change model that proposes a set of qualitatively different and discrete stages of behavior change. The other models listed are continuum models that view behavior change as a linear process in which the same cognitive variables predict behavior over time (Rutter & Quine, 2002). These models are typically evaluated based on two criteria: (a) their ability to predict behavioral intentions, which is used as a measure of overall motivation, and (b) their ability to predict actual behavior. Larsen (2013, unpublished report) provides a comprehensive review of the theoretical and empirical research on social cognitive models of self-regulation, and concludes that across most models and types of health contexts studied, three variables stand out as the best predictors of behavior: behavioral intentions, self-efficacy, and outcome expectations (also referred to as attitudes). In more general terms, the three variables correspond to: (a) a conscious decision to perform a behavior (intentions), (b) the degree one feels they are capable of performing that behavior (self-efficacy), and (c) the person’s anticipated consequences/results for performing the behavior (outcome expectations). The social cognitive models typically have SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 12 strong effect sizes for predicting intentions, but they tend to have moderate-to-weak effect sizes for predicting actual behavior. The disjunction between the motivation to act and the initiation of behavior is referred to as the intention-behavior gap (Rhodes & de Bruijn, 2013; Rhodes & Dickau, 2012; Rhodes, Plotnikoff, & Courneya, 2008). The intention-behavior gap is one of the strongest and most common criticisms of social cognitive models (Gebhardt & Maes, 2001; Maes & Karoly, 2005; Rhodes & de Bruijn, 2013). Intentions and behavior are estimated to have a correlation of approximately 0.50, which amounts to intentions explaining 25% of the variance in behavior (Armitage & Conner, 2001; McEachan, Conner, Taylor, & Lawton, 2011; Symons Downs & Hausenblas, 2005). Meta- analyses have found that a large change in intentions results in only small-to-moderate changes in behavior (Rhodes & Dickau, 2012; Webb & Sheeran, 2006). Additionally, only 54% of people who intend to exercise follow through on their intentions (Rhodes & de Bruijn, 2013). Research has attempted to close the gap through evaluating various mediators and moderators of intentions, including stability of intentions, planning, past behavior, and conscientiousness, as well as others. Rhodes and Dickau, (2013) observed that stability of intentions was the best moderator of the intention-behavior gap. Furthermore, stability of intentions accounted for other moderators of the intention-behavior gap such as anticipated regret, intention certainty, past behavior, self-schema, attitudinal control, and education. Rhodes and Dickau’s (2013) findings indicate that at least part of the intention-behavior gap may result from the inferiority of behavioral intentions as a sole measure of motivation. The researchers suggest that measures of motivation should go beyond the one-item response typically used to measure behavioral intentions. A one-item measure can result in low reliability and low validity by failing to assess the important dynamic properties of motivation (Barlow & SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 13 Proschan, 1975). For example, single items fail to account for how an intention relates to other important goals a person has (Karoly, 1999), which can either facilitate or hinder a person’s goal pursuit (Austin & Vancouver, 1996; Karoly, 1999; Presseau, Tait, Johnston, Francis, & Sniehotta, 2013). Intentions also fail to account for the qualitative nature of a person’s motivation, which has been found to influence the strength and consistency of a person’s motivation (Deci & Ryan, 2004, 2008; Sheldon & Elliot, 1998). Improving the social cognitive models of self-regulation Perhaps the most immediate and drastic improvement to the models that could aid in predicting health behavior is the creation of a measure that more accurately and more robustly gauges a person’s underlying motivation than the traditional measure of intentions (Rhodes & Dickau, 2013). Research on people’s personal goals provides important insights into how a new measure of motivation could be designed. Three constructs in particular—goal-structure, goal commitment, and goal autonomy—could be combined to form a latent variable to accomplish this objective (Deci & Ryan, 2004; Hollenbeck, Williams, & Klein, 1989; Karoly, 1999; Locke & Latham, 1990, 2002). Over the past 50 years, cognitive psychologists have studied various components of goals (for review, see Austin & Vancouver, 1996), and they have outlined the potential contributions that this research could make to the traditional social-cognitive models (Maes & Karoly, 2005; Strecher et al., 1995). Goals are “internal representations of desired states, where states are broadly construed as outcomes, events, or processes.” (Austin & Vancouver, 1996, p. 338). Goals are believed to aid behavior change through at least four processes: (1) channeling the person’s attention to important factors for goal achievement; (2) mobilizing energy for goal pursuit when necessary; (3) sustaining performance of goal pursuit over time; and (4) stimulating SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 14 the use of strategies for goal achievement (Locke & Latham, 1990, 2002). However, very little goals research has been included as part of the traditional social cognitive models of self- regulation (Maes & Karoly, 2005; Sniehotta et al., 2014). Karoly’s (1999) Goal Systems Theory adequately summarizes the various components of goal setting based on previous research on goals (Austin & Vancouver, 1996; Emmons, 1986, 2003; Ford, 1992; Little, 1983). Goal systems theory breaks down the characteristics of goals into 14 facets that describe the ‘why’, ‘what’, ‘where’, ‘how’, and the subjective experiences of goal pursuits. The 14 facets can be reorganized into six components of the self-regulation process: prior conditions, goal construal, anticipation of goal pursuit, feedback and failure monitoring, active goal pursuit, and goal facets that span the entire process and are integral at each level of goal pursuit. Prior conditions to goal pursuit include social context effects and biological-neurological context effects. The goal construal process (i.e., the ‘what’ and ‘why’ of goal pursuit) consists of goal content, goal topography, and goal structure. Anticipation of goal pursuit (i.e., mental preparation for goal pursuit) includes goal process representation, temporal- developmental effects, and goal formalization effects. Feedback and failure monitoring consists of intentional mindset effects, procedural predispositions, and accountability-responsibility heuristics. Active goal pursuit includes goal dynamics, which connects information from the feedback and failure monitoring facets and the anticipation of goal pursuit facets to continually update the person’s representation of their goals. Finally, the emotion interface and experiential dimensions span the entire process and are innately integral at each step of goal pursuit. Although several of the goal facets could be useful additions to the traditional social cognitive models of self-regulation, two goal facets in particular can immediately inform the measure of SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 15 intentions and underlying motivation: Goal-Structure and Goal Topography (Emmons, 1986; Little, 1983; Maes & Karoly, 2005). The first of three goal constructs that could be a useful addition to the social cognitive models is goal-structure (Austin & Vancouver, 1996; Chulef, Read, & Walsh, 2001; Maes & Karoly, 2005). A shortcoming of the standard measures of behavioral intentions, as well as the social cognitive models in general, is that they fail to account for the effects of multiple goals people are continuously pursuing (Gebhardt & Maes, 2001). There are many goals that people consider to be important, but due to limited time, energy, and resources people are unable to accomplish them all. Perhaps ironically, this means that two positive behaviors could have negative influences on each other if one must prioritize their limited resources to each behavior. Goal structure outlines how a particular goal interacts with other goals, including the prioritization of the goal (compared to other important goals) and the degree achieving the goal conflicts or facilitates achieving other goals (Austin & Vancouver, 1996; Karoly, 1999). Goal progress is enabled when the behaviors to reach that goal either facilitate or do not conflict—for time, energy, or monetary resources—with the achievement of other important goals (Gebhardt & Maes, 2001).Goal conflict/facilitation has successfully predicted a variety of behaviors across various contexts and methods of measurement, such as retirement decisions (Brougham &Walsh, 2005, 2007), diet and exercise (Presseau et al., 2013; Lee, Talevich, Lee, Larsen, Read, & Walsh, unpublished data), physician behavior (Presseau, Francis, Campbell, & Sniehotta, 2011), and academic success (Elliot & McGregor, 2001; Karoly, 2010; Okun, Fairholme, Karoly, Ruehlman, & Newton, 2006). Relatively recent research has focused on the degree that SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 16 exercising conflicts with or facilitates achieving a person’s other goals, and both goal conflict and facilitation have exhibited significant effects (Presseau et al., 2011). There are a variety of ways to measure goal-structure (Austin & Vancouver, 1996; Chulef et al., 2001; Ford & Nichols, 1991). A common method to assess goal structure is to have people list their important goals and subsequently rate the degree that they conflict with one another (Emmons & King, 1988; Karoly, 1999). This method is idiographic due to participants creating their own list of goals, as opposed to nomothetic in which all participants rate the same set of goals. Austin and Vancouver (1996) argue that perhaps the best way to measure goal structure would be to combine idographic and nomothetic approaches—which they refer to as an idiothetic approach. The Chulef, Read, and Walsh (2001) goal taxonomy is an example of an idiothetic measure of goal-structure. This measure groups a large sampling of goals into “goal-clusters” that are shared by the majority of the population (a nomothetic component), but allows individuals to rank the goal-clusters in order of importance (an idiographic component). A measure of goal structure can then be taken by measuring the degree that the time, energy, and other resources necessary for performing a particular behavior (e.g., exercising) conflict with or facilitate people in achieving each of the goal-clusters, and multiplying the conflict and facilitation scores by the overall importance rating each person gives to the goal-cluster. Therefore, the amount of conflict and facilitation the person perceives for each goal-cluster gets weighted based on how important that goal-cluster is to the person (relative to the other goal- clusters). Having little conflict or high facilitation with the most important goal-clusters indicates that the behavior is highly compatible with their existing goal-structure. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 17 The Chulef, Read, and Walsh (2001) goal taxonomy has been used in various ways (e.g., predicting retirement decisions; see, Brougham & Walsh, 2005, 2007), and has been updated (Read, Talevich, Walsh, Chopra, & Iyer, 2010). Most recently, versions of the taxonomy have been used to predict body mass index (Lee, Talevich, Lee, Larsen, Read, & Walsh, unpublished data). However, it has not been determined how to best use the taxonomy for gauging goal- structure related to specific weight management behaviors such as exercise. Composite scores for goal conflict and facilitation could conceivably be calculated in a number of ways, and research should empirically test how best to calculate these scores in relation to exercise. The other facet of Goal Systems Theory that could immediately aid in measuring underlying motivation is goal topography. Goal topography consists of stable characteristics describing the person’s goal, including: goal difficulty, goal specificity, goal importance, and goal autonomy. These stable characteristics can be more predictive of goal pursuit outcomes than the content of the goals themselves (Karoly, 1999; Little, 1983). There are numerous established measures for the various components of goal topography that could be used to improve the measure of intentions and underlying motivation (Deci & Ryan, 2008; Hollenbeck, Williams, et al., 1989; Locke & Latham, 2002; Sheldon & Elliot, 1998). A few in particular stand out as useful additions that—along with a measure of goal structure—could be combined to form a latent variable of motivation designed to replace the intentions construct in social cognitive models. Goal difficulty and goal specificity are perhaps the most commonly referenced dimensions of goal topography, primarily due to research from Locke and Latham’s (1990, 2002) goal setting theory. Setting specific difficult goals is commonly found to lead to better SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 18 performance. This may not be appropriate for health related goals, however. Less discussed in the literature is that goal setting theory finds difficult and specific goals to be detrimental in certain instances (Locke & Latham, 2006; Seijts & Latham, 2005). For example, if the goal involves a complex or ambiguous task, difficult goals can be detrimental unless the person has the necessary skills and knowledge for effectively achieving the goal. It is likely that many health related behaviors fall into the category of either complex or ambiguous tasks. In fact, research shows that high expectations (possibly indicating more difficult goals) when starting a weight loss program can be negatively correlated with weight loss (Teixeira, Going, Sardinha, & Lohman, 2005). Also, in a review of weight loss interventions by Powell, Calvin, and Calvin, (2007), a main conclusion was that simple and non-specific weight loss goals led to more success during interventions. Therefore, it may not be appropriate to include measures of difficulty and specificity as part of a motivation construct at this time. Importance of one’s goal is another aspect of goal topography. Goal importance is perhaps the most vital aspect of goal topography because a person needs to view a goal as important before the other aspects of goal topography have any effect on goal pursuit (Locke & Latham, 2002). That is, if a goal is unimportant to the person, then it will carry little motivational weight regardless of other characteristics of the goal. Goal importance can be measured as part of the overall prioritization of a health goal, as measured by the Chulef, Read, and Walsh (2001) goal structure measure discussed above, as well as goal commitment (Hollenbeck, Williams, et al., 1989; Klein, Wesson, Hollenbeck, Wright, & DeShon, 2001). Goal commitment is defined as one’s determination to reach a goal. In goal theory, commitment is hypothesized to be a prerequisite condition for goal attainment, because a goal SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 19 can have no motivational effect if there is no commitment to the goal (Locke & Latham, 1990). A strong commitment to a goal leads a person to exert high effort towards the goal and persist over difficulties and over time (Hollenbeck & Klein, 1987). Goal commitment is believed to be primarily a combination of the overall importance of a goal and the person’s self-efficacy for completing that goal (Locke & Latham, 2002), and it has been advocated as an avenue of research that should be explored further in the self-regulation literature (Strecher et al., 1995). Goal commitment has been found to be positively associated with behavior in a variety of contexts and for a variety of goals (Klein, Wesson, Hollenbeck, & Alge, 1999; Porter & Latham, 2013; Seijts & Latham, 2011). A final construct from the goal topography facet that could aid in the measurement of overall motivation is autonomous motivation, a cornerstone of Self-Determination Theory (Deci & Ryan, 2004, 2008). Autonomous motivation, or goal autonomy, is motivation that resides within the person resulting from the person’s goal aligning with their interests or core values (Deci & Ryan, 2004, 2008; Sheldon & Elliot, 1998). Autonomous motivation is opposed to controlled motivation (also called extrinsic motivation), which is motivation residing outside the person, such as motivation resulting from receiving a monetary reward or pressure from peers (Deci & Ryan, 2004). Autonomous motivation is considered a stronger and more stable form of motivation than extrinsic motivation because the main motivator—the person’s interests and core values—are less likely to change overtime than external factors that influence behavior. In empirical studies, autonomous motivation has shown correlations ranging from 0.14-0.35 for various health behaviors, and correlations of 0.25-0.45 for other core constructs from the main SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 20 social cognitive models (Judge, Bono, Erez, & Locke, 2005; Levesque et al., 2007; Sheldon & Elliot, 1998, 1999; Sheldon & Houser-Marko, 2001). Research on autonomous motivation shows interesting empirical results that could help explain the intention-behavior gap. In a study investigating goal attainment in college students, Sheldon and Elliot (1998) used measures of controlled and autonomous motivation to predict initial intended effort at the start of an academic semester, effort exerted towards achieving the goals at the middle of the semester, and goal attainment at the end of the semester. They found that both autonomous motivation and controlled motivation significantly predicted initial intended effort. However, only autonomous motivation predicted later effort and actual goal attainment. Sheldon and Elliot (1998) concluded that both forms of motivation predicted initial intended effort because being high in either form of motivation represents a high quantity of motivation, but only autonomous motivation sustained over time because it is a more stable form of motivation. If intentions, as measured in the traditional social cognitive models, is similar to intended effort, as measured by Sheldon and Elliot (1998), it could help explain the intention- behavior gap. That is, only a subset of the people reporting strong intentions will be autonomously motivated, and the people who are not may be creating the instability in the relationship between intentions and behavior over time. Therefore, it may be beneficial to consider this qualitative characteristic when measuring motivation that the traditional measures of behavioral intentions cannot account for. The three goal constructs described above could perhaps be combined to form a latent variable that can then supplement behavioral intentions as a measure of motivation in the social cognitive models. Each construct conceivably gauges a different aspect of motivation. Goal structure gauges motivation resulting from the person’s overall prioritization of a particular goal SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 21 combined with the degree that specific behaviors related to achieving that goal either conflict with or facilitate achieving other important goals. The unique component of goal structure is that it gauges motivation in relation to other important goals the person holds. Goal commitment, on the other hand, gauges the overall importance of one’s health goal, as opposed to the importance of the goal in relation to other goals. Finally, goal autonomy gauges the quality of a person’s motivation, with the underlying assumption that autonomous motivation is a more persistent form of motivation than controlled motivation. Furthermore, goal-structure, goal commitment, and goal autonomy would presumably fit nicely into the traditional social cognitive models. Locke and Latham’s (1990; 2002) goal setting theory proposes that increasing self-efficacy will increase goal commitment, and Deci and Ryan’s (2008; 1985) self-determination theory proposes that self-efficacy (which they term competency) is one of three psychological needs that are necessary for autonomous motivation. Research has found positive correlations between self-efficacy and goal commitment (Seijts & Latham, 2011) and between self-efficacy and goal autonomy (Sheldon & Elliot, 1998). The relationship between self-efficacy and goal structure has been less researched, but it is possible that this relationship is similar to the one hypothesized by Social Cognitive Theory between self- efficacy and environmental barriers and facilitators. That is, the higher a person’s self-efficacy, the less likely they will view a particular behavior as conflicting with their other important goals because that behavior seems more manageable. The relationship between outcome expectations and each of the goal constructs has also been less researched in the literature, but would likely exhibit positive associations between the constructs. The more positive a person’s outcome expectations are for a behavior the more motivated they will be to perform it. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 22 Current study The relationships of goal-structure, goal commitment, and goal autonomy have not been tested together simultaneously, let alone with the other core social cognitive variables. Additionally, relatively few behavior change interventions have been designed based on constructs stemming from people’s personal goals (beyond simple planning exercises, such as implementation intentions), which is especially true in the area of health research (Maes &Karoly, 2005; Presseau, Sniehotta, Francis, & Little, 2008; Sniehotta et al., 2014). The present research conducts four sets of analyses in order to test two main aims of the research. The first aim of the study, and the first set of analyses, will test whether a short online intervention designed to intervene on participants’ goal-structure can improve the success of people attempting to increase their moderate-to-vigorous exercise behavior. The intervention will attempt to increase the degree that participants perceive their goal—increasing their moderate-to- vigorous exercise for health purposes indefinitely—will facilitate them in achieving their other important life goals, and it will attempt to decrease the degree that their goal will conflict with their other important life goals. While there is recent evidence of the importance of goal conflict and goal facilitation in self-regulating health behavior, there are very few known interventions designed based on these principles (Presseau et al., 2013). The second aim of the study (which includes the second, third, and fourth sets of analyses) is to explore whether the three constructs from the personal goals literature (goal- structure, goal commitment, and goal autonomy) can be combined in the form of a latent variable representing motivation, and whether that variable adds predictability to traditional social cognitive models of self-regulation. Latent variables are unobserved variables that are calculated from the interrelationships of measured variables. Latent variables in SEM offer several SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 23 advantages over traditional statistical methods that allow for increased power for detecting relationships, more easily interpretable findings, and more flexibility in analyses (McArdle, 2009; McArdle & Nesselroade, 1994). However, several sub-goals must be achieved before a latent variable of motivation can be used in research. First, it must be determined how best to use the Chulef, Read, and Walsh (2001) goal taxonomy to measure goal-structure. There are conceivably multiple methods for calculating goal-structure scores (see below), and the taxonomy has not been used specifically for measuring goal structure for exercise. It must be determined which variable(s) calculated from the taxonomy may potentially combine with goal commitment and goal autonomy to form a latent variable of motivation. Second, it must be shown that calculating a latent variable exhibits better fit to the data than calculating simple composite scores, otherwise calculating latent variables adds unnecessary complexity to analysis (McDonald, 1999). The appropriateness of a latent variable can be tested by comparing the model fit of a model that holds all factor loadings equal to one (representing a simple summed composite score) to a model that allows factor loadings to differ for the observed variables (i.e., a latent variable model). The latter model must exhibit a better fit to the data to justify calculating a latent variable (McDonald, 1999). If calculating a latent variable is justifiable (i.e., model fit improves when a latent variable is calculated), the variable must then be shown to have robust relationships across various contexts in order to be suitable for use in analysis(McArdle & Nesselroade, 1994; McCall & Appelbaum, 1973). That is, the observed variables comprising the latent variable—in this case, goal commitment, goal autonomy, and goal-structure—must also have consistent relationships over time and across various groups of people. This is referred to as “measurement SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 24 invariance” (Meredith, 1993). Measuring the same variables over time does not always mean that the relationships between those variables will remain stable, which is important because a latent variable is unobservable by nature, and is therefore made up entirely of the relationships among the indicator variables (McArdle & Nesselroade, 1994). If the underlying relationships of the observed variables change across various contexts, then any observed differences in the variable can be interpreted as different latent variables being measured all together (McCall & Appelbaum, 1973). The robustness of a latent variable can be tested via measurement invariance analysis in SEM, which systematically restrains and releases certain parameters (factor loadings, indicator variable means, and indicator variable variances) across time and/or across groups to test whether the underlying relationships of a latent variable are consistent. Finally, the third sub-goal in accomplishing this aim is to test the utility of adding a latent variable of motivation comprised of the goal constructs to a typical social-cognitive model of self-regulation. Path analysis can be used to determine if the motivation latent variable significantly predicts behavior at a later time point over and above behavioral intentions. The relationships between the latent variable of motivation and other social-cognitive constructs can also be tested via structural path analysis (see, Duncan, 1977). The a priori model is shown in Figure 1. The model includes direct relationships between intentions, motivation, self-efficacy, and outcome expectations with behavior. Additionally, intentions are also predicted by motivation, self-efficacy, and outcome expectations. This is because intentions are believed to be the most proximal cognition in relation to behavior, as one must intend to do something before they pursue a conscious goal (Ajzen & Fishbein, 1980; Bandura, 2004; Fishbein et al., 2001; Fishbein & Ajzen, 1975). That is, outcome expectations, self-efficacy, and motivation lead to an intention, which ultimately leads to a behavior, and SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 25 therefore, the model includes indirect effects from each cognitive variable on behavior via intentions. Additionally, motivation includes relationships with self-efficacy and outcome expectations. Finally, a pathway is included from self-efficacy to outcome expectations, as hypothesized by Social Cognitive Theory (Bandura, 1997; Bandura, 2004), and has recently been supported with empirical evidence (Larsen, McArdle, Robertson, & Dunton, 2014). The present study utilized a sample of people who were (self-reportedly) beginning a New Year’s resolution to increase their moderate-to-vigorous exercise for health purposes for an indefinite period of time. New Year’s resolutions are goals people set for themselves at the beginning of each year (Cambridge Dictionaries Online). Estimates report that approximately 35% of people make New Year’s resolutions each year (Marlatt & Kaplan, 1972), with the most common resolutions made by adults relating directly to health behaviors (Marlatt & Kaplan, 1972; Norcross, Ratzin, & Payne, 1989). In a study by Koestner, Lekes, Powers, and Chicoine, (2002), in which participants were asked to list 3 resolutions that they were seeking to accomplish, 76% of participants listed at least one health goal. Studying a sample of participants making a New Year’s Resolution to increase their exercise provides an opportunity to study relatively normal people attempting to self-regulate their behavior (as opposed to studying a clinical population, which is common in health research and may not generalize to the rest of the population). The present research consists of four sets of analyses. The first analysis evaluates an online intervention targeting participant goal-structure. The second analysis adapts and evaluates the Chulef, Read, and Walsh (2001) goal-taxonomy for measuring goal-structure related to (a) exercise and (b) not exercising. The third analysis evaluates the robustness (i.e., measurement invariance) of a latent variable of motivation derived from goal-commitment, goal-autonomy, SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 26 and constructs from goal-structure that are observed to be consistently associated with exercise behavior and that factor with motivation constructs in exploratory factor analysis. The fourth analysis evaluates the utility of the motivation latent variable for predicting exercise behavior compared to (a) behavioral intentions, and (b) a more complete social-cognitive model of self- regulation including behavioral intentions, self-efficacy, and outcome expectations. The present research explores a number of hypotheses for each of the four analyses conducted. Hypotheses Hypotheses for the evaluation of the intervention on participant goal-structure are as follows. First, it is hypothesized that a significant manipulation effect will be observed in which the Intervention group will report more positive scores on the goal-structure scores immediately following the goal-structure intervention. Specifically, the Intervention group is expected to report greater degrees of goal-facilitation with exercise and goal-conflict with not exercising, while reporting less goal-conflict with exercise and goal-facilitation with not exercising compared to the Control group. Additionally, it is hypothesized that the Intervention Group will participate in greater levels of exercise following the intervention. More specifically, it is hypothesized that the Control Group will decrease their exercise over time, as new year’s resolutions are typically observed to have low success (Koestner et al., 2006, 2002; Polivy & Herman, 2002), while the Intervention group will either (a) increase over time, (b) remain consistent in their exercise over time, or (c) decrease their exercise over time to a lesser degree than the Control Group. Finally, the other cognitive variables measured (e.g., behavioral intentions, self-efficacy, goal-commitment, etc.,) are expected to mirror changes in reported exercise. That is, the Intervention group is expected to exhibit higher scores on these constructs compared to the Control group. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 27 Hypotheses for the adaption and evaluation of the Chulef, Read, and Walsh (2001) goal taxonomy for measuring goal-structure to exercise are as follows. First, it is expected that several goal-structure scores measured via the Chulef, Read, and Walsh (2001) goal taxonomy will predict exercise and be associated with the other motivational constructs. The overall rank of the health goal-cluster has not been used as an independent variable in analyses, but it is hypothesized that higher importance rankings will be associated with greater amounts of exercise. Goal-facilitation with exercise is expected to be positively correlated with exercise based on past research in the exercise domain (Presseau et al., 2013; Presseau et al., 2010). Goal- conflict with exercise, on the other hand, has shown mixed results for its ability to predict exercise behavior. However, it is hypothesized that goal-conflict will exhibit negative correlations with exercise behavior based on past research utilizing the Chulef, Read, and Walsh (2001) goal taxonomy to measure goal-conflict in other contexts (Brougham & Walsh, 2005, 2007). Associations between exercise behavior and goal-facilitation with not exercising and goal-conflict with not exercising have not been evaluated to date. However, these two constructs have been found to be significantly correlated with behavior when measured with the Chulef, Read, and Walsh (2001) goal-taxonomy (Brougham & Walsh, 2005, 2007) and similar findings are expected in the present study. Specifically, it is expected that goal-conflict with not exercising will positively correlate with exercise behavior and that goal-facilitation with not exercising will negatively correlate with exercise behavior. It is also hypothesized that goal- conflict and goal-facilitation with exercising will exhibit larger associations with exercise behavior compared to goal-conflict and goal-facilitation with not exercising. Furthermore, goal conflict and facilitation with exercise are expected to be more likely to enter as significant predictors in step-wise multiple regressions predicting exercise compared to goal-conflict and SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 28 goal-facilitation with not exercising. Finally, it is hypothesized that the goal-structure constructs that are consistently associated with exercise will likely factor with the other motivation constructs measured in the study (e.g., goal-commitment, goal-autonomy) during exploratory factor analyses. Goal-facilitation is hypothesized to be the most likely goal-structure construct to factor with the other motivation constructs based on past research showing consistent relationships with exercise behavior, followed by goal-conflict with exercise, and followed again by goal-conflict and goal-facilitation with not exercising. The hypotheses for the third analysis evaluating the robustness of a latent variable of motivation are as follows. It is hypothesized that the motivation latent variable will exhibit measurement invariance over time and across gender and intervention groups. That is, tests of model fit will indicate non-significant worsening of model fit after subsequently restraining factor loadings, indicator means, and error variances to be invariant over time and across groups indicating that the latent variable is suitable for use in analyses. Finally, the hypotheses for the fourth set of analyses evaluating the utility of the motivation latent variable for predicting exercise behavior are as follows. First, the motivation latent variable is hypothesized to significantly predict exercise behavior over and above behavioral intentions alone, as well as over and above a more complete social-cognitive model of self-regulation. Specifically, the motivation latent variable is expected to exhibit positive associations with exercise that are at least similar in size compared to behavioral intentions. Additionally, it is hypothesized that the motivation latent variable will exhibit large positive associations with behavioral intentions, which in turn could lead to significant indirect effects of the motivation latent variable on exercise behavior via behavioral intentions. Finally, it is SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 29 hypothesized that the motivation latent variable will exhibit positive associations with both self- efficacy and outcome expectations. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 30 Methods Participants Participants were recruited during January of 2014. The majority of the sample (~90%) were recruited through Amazon Mechanical Turk (Mturk; www.mturk.com), an online crowdsourcing website that enables the collection of survey data. Mturk has become more popular in recent years, and studies find survey results from the website to be at least as reliable as other similar forms of data collection (Mason & Suri, 2012). The rest of the participants were students from the University of Southern California recruited through the psychology department’s subject pool. Mturk offers a feature that screens potential participants so that only people who fit certain criteria can view surveys. These features were set so that participants must live in the United States and have an approval rating of 75% or higher (based on other surveys they completed on the website). The study description also asked participants to be at least 18 years of age and to be willing to complete all three surveys. Finally, a goal of the study was to limit the type of exercise people were trying to accomplish in order to standardize the behavior under study. The study asked for only participants who recently began an attempt to increase their moderate or vigorous recreational exercise indefinitely for health reasons. The researchers felt that this sample would be specific, in that it would only contain people planning to do bouts of extended physical activity for health reasons, while being sufficiently broad enough to recruit a relatively large sample. The study recruited participants in January in order to take advantage of people planning to start New Year’s resolutions to exercise. Study instructions read: “In order to qualify for participation, you must have either started to increase your exercise activity in recent weeks, or intend to begin increasing your exercise activity in the next week… SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 31 We are only interested in exercise that lasts at least 10 minutes per session, is moderate or vigorous in nature (e.g., brisk walking, jogging, dance classes, weight lifting, etc.), and is for the purpose of improving your health. In other words, we are not interested in minor changes in exercise, such as using the stairs at work or walking a bit more by parking farther away from your destination.” The study used several methods to encourage participants to complete surveys 2 and 3. An increasing pay-scale was used for the Mturk sample that paid $0.50 for Survey 1, $1.00 for Survey 2, and $1.50 for Survey 3. An increasing pay-scale could not be used for the student sample; students were rewarded a half-hour of research credit for completing each survey. Study staff sent up to three reminder emails at two day intervals after the initial email invitation (four weeks after the first survey for Survey 2, and four weeks after the second survey for Survey 3). For the Mturk sample, two reminder emails were sent to the email address provided by participants through the survey website (Qualtrics.com), and one reminder email was sent directly through the Mturk website. All reminder emails for the student sample were sent through Qualtrics.com. Finally, surveys 2 and 3 included a question inquiring about participant motivation for returning to complete the surveys. These questions were coded and analyzed in order to gauge whether participants in the Intervention and Control groups had different motivations for returning (e.g., for the reward, because they are interested in health, because they agreed to, etc.). A number of methods were used to ensure the legitimacy of the data collected. Most importantly, the entire survey was timed to ensure that participants didn’t complete it too quickly. Additionally, each individual section of the survey was timed and the number of clicks the participant made on each page was recorded automatically in order to determine if SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 32 participants completed all tasks that were requested of them. There were also test questions included in each survey that explicitly stated which answer option the participants should select, which tested whether or not they were reading directions. Additionally, as part of the Mturk payment process, these participants were given a randomly generated password at the end of the study which they then had to copy and enter on a separate page, which was checked for accuracy by the researchers. Finally, the data were checked to ensure there were no duplicate email addresses provided, which would indicate the same people completed the survey using multiple accounts. The final sample was determined via two rounds of elimination. The first elimination had highly liberal inclusion criteria and was completed prior to inviting participants to take Survey 2. The goal of the first elimination was to remove participants who clearly did not take the study seriously, as we did not want to invite them to take the second and third surveys. Participants were removed from the study for: (1) quitting before finishing the survey; (2) completing the survey in less than five minutes; (3) incorrectly answering the test questions; (4) completing the survey twice; or (5) not entering the correct completion code (Mturk sample only). The second round of elimination was conducted after all data were collected. The purpose of the second elimination was to explore the data to create stricter inclusion criteria to remove participants who did not complete the survey honestly. Participants who did not meet the criteria for Survey 1 were excluded from all three surveys. Participants who did not meet the criteria for Survey 2 or 3 only had their data removed for that particular survey. The logic was that if participants did not provide quality data to start the study then they should not be included in the study at all. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 33 Inclusion criteria for the second elimination were as follows: participants must (1) report that they are not part of a university sport (student sample only); (2) take longer than 7 minutes to complete Survey 1 or longer than six minutes to complete Surveys 2 or 3; (3) take longer than 20 seconds to complete the ranking procedure for Survey 1 and longer than 15 seconds for Surveys 2 and 3; (4) click at least once during the ranking procedure; (5) answer the exercise questions at the beginning of the survey; and (6) in Survey 1, they must have reported a future exercise goal that was greater than what they reported as the amount of exercise they completed before beginning an intention to exercise (i.e., they must have reported wanting to do more exercise in the future than what they did in the past). The specific cut-points for minutes and seconds were chosen by subtracting ~15% from the fastest time the researchers could complete the survey themselves while following all directions. The authors consider these cut-points relatively conservative. Procedures A randomized controlled design with three time points of measurement at baseline, four weeks, and eight weeks was used to test the study aims. Approximately 50% of participants were assigned to the Intervention group via the survey website’s (Qualtrics.com) random assignment software. Participants were blind to which condition they were assigned, and neither group was aware an intervention had taken place. Participants were given one week to complete the second and third survey after initial contact from the research staff. Each survey included questions pertaining to participant exercise behaviors, motivational constructs, and demographics items. Participants also provided a unique code and an email address at the beginning of each survey for data matching and communication purposes. Additionally, the first survey included the intervention manipulation following the ranking procedure of the goal-structure measure and SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 34 retrospective estimates of vigorous, moderate, and mild exercise completed prior to initiating an attempt to increase exercise. The study was approved by the ethics committee of the Institutional Review Board at the University of Southern California as exempt research. An information sheet was presented to participants at the beginning of the first survey, but signed consent was deemed unnecessary. Intervention The online intervention was developed by the authors and was designed to intervene on components of goal-structure (Austin & Vancouver, 1996; Karoly, 1999; Presseau et al., 2013) in a relatively short amount of time. The aim of the intervention was to: (a) increase the degree that participants perceived their New Year’s resolution to increase their exercise would facilitate their three most important life goals; and (b) to decrease the degree it would conflict with their three most important life goals. First, all participants ranked the importance of nine life goals from most to least important to them using the Chulef, Read, and Walsh (2001) goal taxonomy as part of the goal-structure measure (all Control and Intervention group participants completed this task; the nine goal-clusters are displayed in Table 1). The Intervention started immediately following the ranking procedure (Appendix A includes a complete description of the tasks and instructions). Intervention group participants viewed a screen explaining the concepts of goal- structure in lay terms. To paraphrase, the instructions explained: (a) it is important to understand how beginning a new behavior interacts with existing behaviors; (b) people generally consider all nine of the goals from the ranking procedure to be very important, but due to limited resources (e.g., time, energy) they must prioritize the goals they pursue; and (c) it is therefore important to understand how their goal of increasing their exercise will facilitate them in achieving the three goals they ranked as most important. On the next page, the intervention tool SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 35 repeated the goal-cluster that each person listed as most important to them during the ranking procedure and provided an example of how someone might feel exercise would facilitate them in achieving that goal. Participants were then instructed to provide at least one example of their own indicating how they felt increasing their exercise could help them accomplish this goal. The same procedure was repeated for the person’s second and third most important goals on the following two pages. The Intervention group then viewed a screen explaining the importance of setting specific plans for exercise in order to (a) minimize the degree that exercise would conflict with their three most important goals, or (b) maximize the degree it would facilitate those goals. Participants reported (1) where they planned to exercise; (2) when they planned to exercise; (3) the type of exercise they planned to do; and (4) how long they planned each exercise session to be. On the following three screens, participants described how the plan they just created would help them to minimize the degree exercising would conflict with their three most important goal-clusters or maximize the degree it would facilitate them in achieving those goal-clusters. The intervention tool repeated the plans participants made on each page for their convenience. The planning procedure was the last task of the intervention, and after its completion, the Intervention group completed the rest of the study as normal. Measures Exercise. Exercise was measured using the Godin Leisure-Time Exercise Questionnaire (GLTEQ; Godin & Shephard, 1985). This measure has been validated with physiological and anthropometric measures (VO2 max and body fat; Godin & Shephard, 1985; Jacobs, Ainsworth, Hartman, & Leon, 1993). The GLTEQ measures vigorous-intensity (described as heart beating rapidly, sweating), moderate-intensity (described as not exhausting, light perspiration), and mild- SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 36 intensity exercise (described as minimal effort, no perspiration). Participants reported the number of exercise sessions they completed and the duration of those exercise sessions: (a) prior to starting an intention to increase their exercise (“prior” exercise; measured at Time 1 only); (b) during the last week (i.e., “current” exercise); and (c) the amount of exercise they planned to complete four weeks in the future (i.e., future exercise goals). The instructions specified to include only activities that are not part of daily work routines and that last at least 10 minutes in duration, as bouts of physical activity of at least 10 minutes have been shown to positively affect numerous health indices related to weight and physical activity(Donnelly, Jacobsen, Heelan, Seip, & Smith, 2000; Murphy, Nevill, Neville, Biddle, & Hardman, 2002).The total minutes of exercise (number of sessions X duration) was calculated for vigorous and moderate intensity levels and converted to metabolic equivalent (MET) minutes so that different intensity levels of exercise could be combined to create composite scores for vigorous exercise and vigorous/moderate exercise. Multipliers of 7.5 for vigorous and 4.0 for moderate were used in conversions (Plotnikoff et al., 2007). An additional exercise variable was calculated, described as participant success in accomplishing their intentions for exercise. Participants’ MET minutes of exercise at Time 2 and Time 3 were divided by their future exercise goals in MET minutes from the previous time point to determine how successful they were in accomplishing their exercise goals for that particular four-week window (e.g., minutes of exercise measured at Time 2 / future exercise goals measured at Time 1). Success in accomplishing intentions for exercise was calculated for vigorous/moderate exercise only. Success in accomplishing vigorous exercise was not tested because a substantial proportion of participants intended to do no vigorous exercise at Time 2 or Time 3, which substantially reduced the sample size (~12% at Time 2 and ~20% at Time 3). SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 37 Therefore, the final list of exercise variables tested in the present study were: vigorous exercise at Time 1, Time 2, and Time 3; vigorous/moderate exercise at Time 1, Time 2, and Time 3; and success in accomplishing intentions for vigorous/moderate exercise at Time 2 and Time 3. Square root transformations were used to correct for positive skew observed in all exercise variables. Habits. The habits measure is described here for completeness but was not included in analyses. Habits were measured by evaluating the stability of the environment in which participants typically exercise (e.g., see, Danner, Aarts, & Vries, 2008; Wood, Tam, & Witt, 2005), which was then multiplied by the frequency of which people exercise based on GLTEQ scores. The combination of environment stability and frequency of behavior is based on the theory that strong habits form when participants frequently perform a behavior and generally perform it in the same environment (Wood & Neal, 2007). The environment is defined as the location, time of day, and circumstances (e.g., weather, other people, etc.,) in which the behavior is performed. There will be one item each for vigorous-, moderate-, and mild-intensity exercise. The questions read: “We are interested in how stable the environment is that you currently exercise in. A stable environment is the degree that you perform the behavior in the same location, at the same time of day, and under the same circumstances (e.g., weather, other people, etc.). If these three aspects are always different (i.e., you exercise in different places, at different times, and under different circumstances each time you exercise), then it would be an unstable environment. However, if these three aspects remain the same, then it would be a stable environment.” The response options ranged from 0-9 with 0 corresponding to not exercising at all, 1 corresponding to a highly unstable environment, and 9 corresponding to a highly stable environment. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 38 Goal-structure. Goal-structure was measured using a version of the Chulef, Read and Walsh (2001) goal taxonomy at the level of nine goal-clusters. The nine goal-clusters were: (1) being a moral and virtuous person (e.g., sticking to personal morals, helping others, being highly regarded); (2) Religion and spirituality; (3) Self-fulfillment and being open to new experiences (e.g., gaining wisdom, appreciating beauty, embracing life); (4) Avoiding negative social experiences (e.g., self-protection, avoiding rejection or conflict); (5) Good social relationships (e.g., intimacy, belonging, influencing others); (6) Good family relationships (e.g., close to parents, being a good family member); (7) Being intelligent and skillful (e.g., intellectual growth, being autonomous, competent); (8) Having financial and occupational success (e.g., financial freedom, wealth, a respected job); and (9) Being physically healthy (e.g., being active, capable of daily tasks, physically fit). Participants sorted the nine goal-clusters based on how important each goal-cluster was to them. They then rated the degree that the time, energy, or other resources required to do the amount of exercise they intended would either conflict with or facilitate them in achieving each goal-cluster. Finally, they rated the degree that not exercising at all would conflict with or facilitate them in achieving each of the goal-clusters. Conflict and facilitation ratings occurred on the same 11-point scale ranging from -5 to 5. Scores from -5 to -1 indicated conflict, scores from 1-5 indicated facilitation, and a score of 0 indicated that exercise (or not exercising) neither conflicted nor facilitated the completion of that goal-cluster. Goal conflict and facilitation have been measured on bipolar scales in the past (e.g., Brougham & Walsh, 2005, 2005), as was done here, but have also been measured using two separate unipolar scales for goal facilitation and conflict (Riediger, 2007). A bipolar scale was chosen here because it is the only method that has been used with the Chulef, Read, and Walsh (2001) goal taxonomy to date. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 39 Seven scores were calculated and tested to determine which method(s) of scoring were most useful in predicting exercise and the other motivation constructs. First, the simple rating of the health goal-cluster was used (i.e. the position the participant placed this goal during the ranking procedure; from first to ninth most important to them). The ranking alone serves as a test of how the importance of health in general (in relation to other life goals) is associated to exercise behavior. The ranking of the health goal-cluster was reverse scored so that higher values correspond with higher importance. The rest of the scores were some variation of a weighted and summed composite score for conflict and/or facilitation with exercising or not exercising. That is, the conflict/facilitation score for each goal-cluster was multiplied by a weighted version of the importance rank each participant gave that goal-cluster. The weights of the importance ranks were as follows: the goal- cluster ranked most important received a weight of 5, the goal-clusters ranked second and third most important received a weight of 4, the goal-clusters ranked fourth through sixth received a weight of 3, the goal-clusters ranked seventh and eighth received a weight of 2, and the goal- cluster ranked last received a weight of 1. This method of weighting creates a relatively normal distribution of weights that should have a higher test-retest reliability than using non-weighted ranks alone. As long as the participant ranks a goal-cluster near where they ranked it the first time, it will receive the same or close to the same weight as before. This is particularly true for goal-clusters ranked in the middle of the continuum, which are of average importance and probably more likely to fluctuate over time compared to goal-clusters ranked extremely high or low. Minor fluctuations in ranks have less influence on the overall conflict/facilitation score with this method. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 40 Two composite scores (one each for exercising and not exercising) represented the weighted and summed conflict/facilitation scores and are referred to as aggregated conflict/facilitation with exercise and aggregated conflict/facilitation with not exercising. Conflict and facilitation scores for exercising (and not exercising) were aggregated so that conflict with one goal compensated for facilitation for another goal. In other words, conflict and facilitation were treated as opposites on the same bipolar scale, as opposed to two separate constructs on unipolar scales. Therefore, someone could presumably have a score of zero in which they perceived exactly the same amount of conflict and facilitation with the nine goal- clusters resulting from exercising (or not exercising). The last four composite scores represented only conflict or facilitation for exercising or not exercising. For example, in calculating the degree exercising conflicts with a person’s goals, any goal in which the person indicated exercise facilitates them in accomplishing that goal is scored as zero. Only goals that conflict with exercising get included in the goal conflict score, and any goals that facilitate exercise have no influence. The four composite scores were (all weighted and summed, as described above): (1) conflict with exercise, (2) facilitation with exercise, (3) conflict with not exercising, and (4) facilitation with not exercising. In summary, there were seven scores tested for measuring goal structure. The scores were: (1) the overall importance rank of the health goal-cluster from the ranking procedure; (2 and 3) aggregated conflict/facilitation scores for (a) exercise and (b) not exercising; and (4-7) conflict scores for (c) exercise and (d) not exercising, and facilitation scores for (e) exercise and (f) not exercising. Goal commitment. Goal commitment was measured using the Hollenbeck et al., (1989) measure, which was later adapted by Klein Wesson, Hollenbeck, Wright, and Deshon (2001). SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 41 The Klein et al., (2001) measure consists of five of the nine original items from the Hollenbeck et al., (1989) measure. The instructions read: “Please indicate how strongly you agree or disagree with the following statements in regards to your intention to exercise over the next 4 weeks:” Example items include: “I am strongly committed to my intentions to increase my exercise.” and “I think intending to increase my exercising is a good goal to shoot for.” Response options range from 1-5 with anchors I strongly disagree and I strongly agree. Items 1, 2, and 4 were reverse scored, and the five items were summed to create a composite score for goal commitment (average α = .765). Goal autonomy. Goal autonomy was measured using the Sheldon and Elliot (1998) goal concordance scale. The goal concordance scale consists of four items, two items corresponding to autonomous motivation and two items corresponding to controlled motivation. The directions read: “Using the following answer options, please indicate how much each of the following four reasons motivates you to increase your exercise.” An example item for autonomous motivation is: “Because of the fun and enjoyment which exercising provides you. While there may be many good reasons for exercising, the primary ‘reason’ is simply your interest in the experience itself.” Response options ranged from 1-9 with anchors Not at all because of this reason and completely because of this reason. The two autonomous motivation items and the two controlled motivation items were summed to create composite scores of each construct. Autonomous motivation had an average α = .572 and controlled motivation had an average α = .461. Behavioral Intentions. Intentions were measured using three suggested items from Schwarzer (2008) asking participants to indicate how strongly they intended to perform exercise over the next four weeks. The three items are designed to map onto the three types of exercise intensity measured in the GLTEQ: vigorous-, moderate-, and mild-intensity. The directions read: SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 42 “We are interested in how strong your intentions are for performing leisure time exercise. Please indicate on the following scale the degree that you feel each statement is true for your intentions to perform each type of exercise over the next 4 weeks.” An example item is: “I intend to perform vigorous-intensity exercise (i.e., heart beats rapidly, sweating, unable to hold conversation) over the next four weeks.” Response options ranged from 1-6 with anchors Not at all true and Absolutely true. Composite scores for intentions for vigorous exercise and vigorous/moderate exercise were calculated using the same MET multipliers that were used for exercise variables. This was done to keep the behavioral intentions variables consistent with the exercise variables. Self-efficacy. Action self-efficacy (two items), maintenance self-efficacy (two items), and recovery self-efficacy (three items) were measured using suggestions from Schwarzer (2008). Action self-efficacy is a person’s confidence in their ability to begin a behavior, maintenance self-efficacy is a person’s confidence in their ability to perform a behavior regularly, and recovery self-efficacy is a person’s confidence in their ability to continue performing a behavior following a set-back such as an injury. The directions read: “We are interested in how confident you are in your ability to accomplish your intentions of increasing your exercise. Using the following scale, please indicate the degree that you feel each of the following statements are true regarding your confidence in your ability to increase your exercise.” Example items are: “I am certain that I can meet my intentions for exercise on a regular basis, even if it is difficult. [action self-efficacy],” and “I am confident that I am able to resume meeting my intentions for exercise regularly, even after failures to pull myself together. [recovery self-efficacy].” Participants responded on a six-point scale with anchors Not at all true and Highly true. Exploratory factor analysis with an oblique rotation revealed that all seven SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 43 items loaded strongly onto a single factor explaining ~74% of the variance in the data with an average α = .942. Therefore, the seven items were summed to form single composite scores for self-efficacy at each time point in order to simplify analyses. Outcome expectations. Outcome expectations were measured using a shortened version of the Multidimensional Outcome Expectations for Exercise Scale (Wójcicki, White, & McAuley, 2009). The measure contained six items designed to map onto the three types of outcome expectations specified by Bandura (1997): physical, social, and self-evaluative expectations. The directions read: “We are interested in how you feel exercising would influence other aspects of your life. Using the following scale, please indicate how strongly you agree with each of the statements.” Example items were: “If I increase my exercise then it will increase my mental alertness [self-evaluative],” and “If I increase my exercise then it will improve my social standing [social].” The response options ranged from 1-5 with anchors Strongly disagree and Strongly agree. Exploratory factor analysis was used to determine whether to calculate scores for each type of outcome expectation or to create a single outcome expectations composite score. A single factor solution accounted for ~50% of the variance in the data and had an average α = .770. Therefore, for simplification purposes, all six outcome expectations were summed to form single composite scores at each time point. Planning. Planning was measured using four items suggested by Schwarzer (2008) gauging the degree participants plan when, where, and how they will do their exercise. The directions read: “We are interested in the degree that you plan your exercise in advance. Using the scale provided, please indicate the truthfulness of each of the following statements in regard to how detailed you plan your exercise activities.” The items were: “I usually plan which type of exercise I will perform in advance (e.g., walking, biking).”, “I usually plan where I will exercise SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 44 in advance (e.g., the park, the gym).”, “I usually plan which days of the week I will exercise.”, and “I usually plan for how long I will exercise.” Response options range from 1-4 anchors Not at all true and Exactly true. The four items were summed to form single composite scores at each time point (average α = .803). Demographics. Demographic questions assessed participant age, sex, education level, race, the region of the country they live (Mturk sample only), and whether they are part of a university sponsored program involving large amounts of physical activity (e.g., sports, band; student sample only). Analysis Descriptive statistics, correlation, regression, and exploratory factor analyses were conducted in SPSS version 17.0. Structural equation modeling was done in R version 2.15.2 using the lavann package (Rosseel, 2011). Missing data were handled using the Full Information Maximum Likelihood (FIML) function in the lavann package, an estimation procedure that utilizes all available data to reduce bias from missing data. Model fit was determined using a combination of the minimum function chi-square/degrees of freedom (𝜒 2 /df), the Comparative Fit Index (CFI), the Tucker-Lewis Index (TLI), and the Root Mean Square Error of Approximation (RMSEA). Conventional standards for good fit are values above 0.95 for the CFI and TLI and values below 0.06 for the RMSEA (Hu & Bentler, 1999). Change in model fit was evaluated using change in chi-square/change in degrees of freedom (Δ𝜒 2 /Δdf). Intervention effects. Independent samples t-tests were used to compare Intervention and Control groups on the social-cognitive (behavioral intentions, self-efficacy, outcome expectations, and planning) and motivation variables (goal commitment, autonomous motivation, controlled motivation, and goal-structure) at Time 1. These constructs at Time 1 were measured SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 45 immediately following the intervention, and therefore represent an initial test of immediate intervention effects. Exercise variables were not included because they were measured prior to the intervention. Changes over time and across groups in all variables were tested via a series of multi- group latent change score (LCS) models in SEM. LCS models can test the same hypotheses as more traditional methods, such as ANOVA, MANOVA, or hierarchical linear modeling (HLM), but offer several advantages (McArdle, 2009; McArdle & Nesselroade, 1994; McCall & Appelbaum, 1973). For instance, (a) ANOVA and MANOVA rely on several improbable assumptions that LCS models in SEM do not (e.g., spherecity, compound symmetry; McCall & Appelbaum, 1973); (b) LCS models allow researchers to specify and test specific a priori hypotheses and relationships beyond mean group differences (e.g., differences in variances or covariances); and (c) SEM offers clearer and more comprehensive analysis of relationships and changes. LCS models also allow for modeling both intra-individual and inter-group changes over time, while ANOVA and MANOVA can only test mean changes in groups and treat within- group differences as error (Voelkle, 2007). While ANOVA and MANOVA are restricted to the use of categorical predictors because of the sole focus on mean changes, LCS models can test any combination of categorical and/or continuous predictors. This opens up the possibility of highly complex longitudinal models that can test relationships between two or three sets of variables over time (e.g., dual change score models; bivariate and trivariate cross-lagged latent change score models). Steps for constructing LCS and dual change score models are shown in Figures 2 through 6. Following guidelines for SEM diagrams (McArdle, 2005), observed variables are drawn as squares and latent variables are drawn as circles. All model parameters are represented by one- SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 46 headed or two-headed arrows. One-headed arrows deriving from latent variables and pointing to observed variables represent factor loadings. Two-headed arrows connecting two variables represent covariances. Two-headed arrows with both heads connecting to the same variable represent variances (for latent variables) or error terms (for observed variables). A fixed variable set at 1 and drawn as a triangle represents mean scores, and arrows deriving from this fixed variable represent group means. Classical test theory states that every measured variable (i.e., observed score) is a combination of an unobserved (latent) True score (z[1]) and some error (e[1]), depicted in Figure 2 (Nunnally, Bernstein, & Berge, 1967; Spearman, 1904). The error terms are assumed to have a mean of zero, a non-zero variance that is invariant over time, and to lack associations with any other scores in the model. With multiple measures of the same variable on the same individuals over time, these restrictive assumptions can be enforced in the SEM framework allowing for true scores to be modeled as latent variables (McArdle, 2001). LCS models can then be modeled starting with a series of auto-regressions. A single auto-regression is shown in Figure 3.a, in which a score on variable z at time 2 (z[2]) is regressed on variable z at time 1 (z[1]) creating an error term (e[2]) associated with the prediction of z[2]. This model can be altered to a change score model by setting the path from z[1] to z[2] equal to 1, implying that some portion of z2 is exactly equal to z1 (Figure 3.b). The resulting residual, formally an error term, can now be directly interpreted as a latent change score representing the degree variable z changed from time N to time N + 1 (Bollen, 2002; McArdle & Nesselroade, 1994). Doing so makes change scores explicit parameters that can be manipulated and directly tested in SEM. N time points of data can be regressed in this manner, creating N-1 latent change scores (Figure 4). With multiple latent change scores in the model (minimum of three time points of measurement), group-level latent SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 47 variables can be added for the intercept (mean baseline score; I[0]) and slope (mean linear change over time; Figure 5). Finally, regressing latent change scores on the previous score (β) adds an element of proportional change to the model (e.g., being high at previous time may be related to less future growth), resulting in a dual change score model (Figure 6).The linear change estimate (a slope) combined with a proportional change element (β) allows for modeling any form of curvature in change over time. However, it is not necessary to include both forms of change in the model. For example, it is possible to have a best fitting model that includes only a slope without proportional change (i.e., a linear change), or a model that includes proportional change and no slope (i.e., a proportional change). The best model for the data can be empirically tested in SEM by comparing fit of nested models that restrict certain parameters. A multi-group approach in SEM can test the same hypothesis as more traditional methods, but does so in a different manner. Rather than using dummy-coded predictor variables in a regression framework (e.g., Control group = 0, Intervention group = 1), a multi-group approach tests similarly structured models for multiple groups simultaneously (Figure 7; Sörbom, 1974, 1978). This approach provides more flexibility in analysis because any parameter can be constrained across groups in one model and then released in a subsequent model to test for differences between groups in said parameter. For example, mean change over time can be held equal for both groups in model A and subsequently released in Model B. If Model B fits the data better than Model A, it can be concluded that the group means differ. The same can be done for testing variances, covariances, regression pathways, or any other parameter. A series of eight models were run to test for changes over time, group differences in changes over time, group differences in the covariance between changes over time and baseline averages, and group differences in the variance of changes over time between Intervention and SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 48 Control groups for each variable under study. The first model tested was the null model, which holds all parameters invariant across groups and constrains change over time to zero (i.e., the slope and proportional change were restricted to zero). The second model continued holding all parameters invariant across groups, but allowed the proportional change parameter to be estimated to test for proportional change over time (a proportional change model).The third model held parameters invariant across groups, restricted the proportional change back to zero, and allowed the linear change element to be freely estimated (a linear change model). The fourth model was the first dual change score model, allowing both the slope and proportional change parameters to be freely estimated, but again all parameters were held invariant across groups. A comparison in fit of the null model to the proportional change, linear change, and the dual change models constitutes a test for a main effect of time on change. If any of the models improve model fit compared to the null model, then it can be concluded that scores significantly changed over time. The fifth model then released constraints on the slope across groups, while the sixth model released constraints on the proportional change across groups. If the fifth or sixth model improves model fit compared to models constraining these parameters to be invariant across groups, it signifies a significant difference in change over time across groups. Finally, the seventh model released constraints on the covariance between the slope and the intercept across groups, and the eight model released group constraints on the variances of the intercept and slope across groups. Adapting the Chulef, Read, and Walsh (2001) goal taxonomy for predicting exercise behavior. A primary aim of the present work was to determine how to adapt the Chulef, Read, and Walsh (2001) goal taxonomy for optimally predicting exercise behavior. In other words, several different methods for calculating composite scores for goal-structure were tested to SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 49 determine which method(s) worked best. As described above, seven goal-structure scores were calculated: (1) the overall rank of the health goal-cluster, (2) aggregated conflict/facilitation with exercise (3) aggregated conflict/facilitation with not exercising, (4) conflict with exercise, (5) facilitation with exercise, (6) conflict with not exercising, and (7) facilitation with not exercising. The usefulness of the seven goal structure scores were tested in several ways to determine which scores, if any, may be useful additions to a latent variable measuring motivation. Additionally, each test was completed on five different samples as a way to replicate and validate significant associations found with the exercise variables and the other motivation variables. The five samples used were: (1-3) three different random samples of 50% of the overall study sample, (4) a complete-cases sample, and (5) the entire sample. The degree that the samples overlapped is presented in Table 2. The logic of this analysis was that if a relationship between any of the goal structure scores and other variables is robust, then the relationship should be observed in multiple samples of data. The criterion for the present analysis was that significant findings must occur in at least three of the five samples to be considered robust. Pearson r correlations and multiple regression were used to test associations between the seven goal- structure scores and the exercise constructs. Exploratory factor analyses with a direct oblimin (oblique) rotation were used to evaluate relationships between the goal-structure scores and the other motivation constructs. Test of measurement invariance: the existence of a robust latent variable measuring motivation. Measurement invariance analysis testing for the existence of a robust latent variable measuring motivation were conducted on composite scores for the goal commitment scale, the goal autonomy scale, and the chosen goal-structure scores measured from the Chuleff, Read, and Walsh (2001) goal taxonomy (based on the outcome of the goal-structure analysis).The model SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 50 tested is displayed in Figure 8. Measurement invariance analysis tests for a consistent factor by progressively constraining the factor loadings, observed variable means, and the observed variable variances over time and across various sub-groups, and comparing model fit and change in model fit after each step (Horn & McArdle, 1992). The logic is that if the model continues to fit the data well after adding each set of constraints, then the latent variable is said to be stable across those parameters. The latent variable can be considered robust, and therefore used in analysis, if the model continues to fit the data well after adding constraints over time and after adding constraints across sub-groups. The first model tested was a Rasch model forcing all loadings equal to 1.00 (McDonald, 1999). The Rasch model tests the fit of a model that represents a simple summed composite score of the items. The next model tested was a “weak invariance” model, which allows loadings to be different across items, but forces loadings of each item to be equivalent over time (Meredith, 1993). If the weak invariance model fails to improve upon the fit of the Rasch model, then there is no need to calculate a latent variable because a simple summed composite score would suffice. The third model tested was a “strong invariance” model, which forces the latent variable means to be equivalent overtime in addition to the factor loading constraints of the weak invariance model (Meredith, 1993). The fourth model was a “strict invariance” model which forces the observed variable variances (i.e., latent variable errors) to be equal over time in addition to the factor loadings and observed variable means(Meredith, 1993). Finally, the last series of models tested for strict invariance across study group and gender (Intervention vs. Control groups and males vs. females) by constraining and subsequently releasing constraints on model parameters of a strict invariance model across sub-groups. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 51 Testing predictability of the motivation latent variable. The utility of the motivation latent variable was tested via two series of path analyses for each of the dependent variables (vigorous exercise at Time 2 and Time 3; vigorous/moderate exercise at Time 2 and Time 3; and success in accomplishing intentions for vigorous/moderate exercise at Time 2 and Time3). Cognitions at Time 1 were used to predict exercise behavior at Time 2, and cognitions at Time 2 were used to predict exercise behavior at Time 3. The first series of models included only an exercise variable, motivation for exercise, and intentions for exercise matched to the intensity level of exercise in the model (e.g., when predicting vigorous exercise, intentions for vigorous exercise was used; when predicting vigorous/moderate exercise, intentions for vigorous/moderate exercise was used). The second series of models included self-efficacy and outcome expectations into the existing model to assess how relationships changed after including a more complete social cognitive model of self-regulation, and to observe the relationships between the variables within that model. An example model is shown in Figure 9. The first model (A) restricted the pathways between exercise and intentions and exercise and motivation to zero, representing the baseline model. Model B included a pathway from intentions to exercise, but not from motivation to exercise. Model C included a pathway from motivation to exercise, but not from intentions to exercise. Model D included both pathways from intentions to exercise and motivation to exercise. Finally, Model E removed the latent variable from the model allowing the indicator variables to be associated with the other variables in the model on their own. This series of models tests (1) whether intentions or motivation predict exercise on their own, (2) whether both variables predict exercise when included in the model together, and (3) the utility of including a latent variable in the model compared to the indicator variables on their own. Additionally, all SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 52 models include a pathway from motivation to intentions, allowing for an indirect effect of motivation on exercise to be calculated via intentions. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 53 Results Initial recruitment and first round of participant exclusion A total of 1,053 participants started Survey 1 (49.2% Intervention Group; 62.7% female; 90.0% Mturk sample). Participants were removed from the sample for the following reasons: 100 quit the survey almost immediately upon starting it; 28 more finished in less than five minutes; 37 incorrectly answered the test questions; three people completed the survey twice; and 40 Mturk workers did not provide a matching completion code. The sample invited to take Survey 2 and Survey 3 consisted of 845 participants (48.9% Intervention group; 62.8% female; 90.1% Mturk sample). Seven-hundred-four participants started Survey 2 (83.6% of the total emailed). Participant data were removed for the following reasons: 18 quit the survey almost immediately upon starting it; 12 completed the survey twice, so only their first attempt was included; one was removed from Survey 1 but was accidentally sent the invitation email; and two answered using only anchors of scales for all questions. Survey 2 had a final sample size of 671 (79.4% of the total people emailed). All 845 participants from Survey 1 were invited to complete Survey 3 regardless of whether they completed Survey 2. Six-hundred-fourteen participants started Survey 3 (72.7% of the total people emailed). Participant data were removed for the following reasons: 22 people quit the survey almost immediately upon starting it; one person completed the survey twice, so only their first attempt was included; and one person answered all questions with anchors of scales. Survey 3 had a final sample of 590 (69.8% of the total people emailed). The final count for each survey after the first round of elimination was 845 for Survey 1, 671 for Survey 2, and 590 for Survey 3. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 54 Second round of participant exclusion The second round of elimination was conducted after all data were collected. One- hundred-thirty-two participants were removed from Survey 1, and subsequently the rest of the study, 51 participants were removed from Survey 2, and 71 participants were removed from Survey 3. Chi-square tests of independence revealed that participants removed at any survey were more likely to be removed at each of the other surveys (p< .05), indicating these participants continued to provide poor data throughout the study. Additionally, chi-square tests of independence revealed that removed participants were more likely to be outliers on several of the main dependent variables (p < .05). The final sample was 713 participants (student = 73), 514 of whom completed Survey 2, 433 of whom completed Survey 3, and 386 of whom completed all three surveys. Participant flow is presented in Figure 10, and demographic characteristics are presented in Table 3. The Intervention group contained 349 participants (48.9%). The sample was approximately 63% female and had a mean age of 31.76 (SD = 11.23) years. The intervention took a median time of approximately seven minutes and 30 seconds. The types of exercise people completed at each time point are presented in Table 4. Walking, running/jogging, and weight lifting were consistently the most frequently selected exercise options. Analysis of missing data. Dummy-coded variables for missingness were created for each time point by coding participants missing surveys as 1 and participants completing surveys as 0. Chi-square tests and logistic regression were used to predict which participants were likely to miss Survey 2 and Survey 3 (p < .05). Younger participants were more likely to be missing for Survey 2 and Survey 3. Male participants, and people reporting doing more vigorous exercise prior to attempting an increase of their exercise were more likely to be missing for Survey 2. The SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 55 student sample and people reporting doing less vigorous exercise during the time of Survey 1 were more likely to be missing at Survey 3. Additionally, the question at the end of Surveys 2 and 3 inquiring about the participant’s main motivation for continuing the study was coded, and a chi-square test of independence was used to test if the study groups had different reasons for continuing the study. No differences were observed between groups (p> .05). All variables predicting missingness were included in statistical models to account for missing data using FIML (age, gender, student vs. Mturk sample, and vigorous exercise prior to beginning their New Year’s resolution). Group differences prior to randomization. Chi-square tests of independence and independent samples t-tests were used to test for differences in groups prior to randomization (p < .05). The majority of the survey took place following the intervention component, and therefore testing differences in groups on these measures does not provide information about how groups differed prior to intervention. Only demographic questions (age, gender, student or Mturk sample, race, education, and region of country) and questions related to when they started an intention to increase exercise, their previous exercise, their current exercise, their future exercise goals, and their exercise habits were reported prior to the intervention. The Control group had a higher proportion of the student sample (65.8%, p = .008). No differences were observed in the other variables. Tests of intervention effects Means, medians, standard deviations, and quintiles for the exercise variables are presented in Table 5, and the means, medians, and standard deviations for other variables are presented in Table 6. Means, medians, and standard deviations for all variables split by intervention group are presented in Tables 7 and 8. The social-cognitive variables (behavioral SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 56 intentions, self-efficacy, outcome expectations, and planning) and motivation variables (commitment, autonomous motivation, controlled motivation, and the goal-structure scores) at Time 1 were measured after the intervention component. Independent samples t-tests were used to test for mean differences between groups in these variables at Time 1 to gauge the immediate impact of the intervention. Results are presented in Table 9. Significant differences were observed in all of the goal-structure scores except for the overall rank of the health goal-cluster (Ps < .01). All affects were in the hypothesized direction, in that the Intervention Group exhibited higher facilitation with exercise, higher conflict with not exercise, lower conflict with exercise, and lower facilitation with not exercising compared to the Control Group. Additionally, the Intervention group reported higher levels of self-efficacy, outcome expectations, planning, intentions for vigorous exercise, commitment, and autonomous motivation, but none of these differences were significant at the p < .05 level. Changes in intervention variables over time were tested using multi-group latent dual change score models testing for: (a) proportional change over time; (b) linear change over time; and (c) differences in study groups over time (freeing slope mean, proportional change, slope- intercept covariance, and slope and intercept variances across groups). The model tested is shown in Figure 7, which can be written as 𝑦 𝑔 ,𝑡 = 𝑌 𝑔 ,𝑡 −1 + 𝛥𝑦 [𝑡 ] 𝑔 ,𝑛 where participants’ score at any time point (𝑌 𝑔 ,𝑡 ) is a combination of the score at the previous time point (𝑌 𝑔 ,𝑡 −1 ) and change over time (𝛥𝑦 [𝑡 ] 𝑔 .𝑛 ). Furthermore, change over time can be written as 𝛥𝑦 [𝑡 ] 𝑔 ,𝑛 = α * 𝑠 𝑔 ,𝑛 + β 𝑔 ,𝑛 * 𝑦 [𝑡 – 1] 𝑔 ,𝑛 + 𝑧 [𝑡 ] 𝑔 ,𝑛 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 57 where participants’ change at any given time point is a combination of a fixed slope (α * 𝑠 𝑔 .𝑛 ), a proportional relationship to the previous time point (β 𝑔 ,𝑛 * 𝑦 [𝑡 – 1] 𝑔 .𝑛 ), and an additive error term (𝑧 [𝑡 ] 𝑔 .𝑛 ). The above equation can be separated by group for multi-group analysis and written as Intervention group: 𝛥𝑦 [𝑡 ] 𝑖 ,𝑛 = α * 𝑠 𝑖 ,𝑛 + β 𝑖 ,𝑛 * 𝑦 [𝑡 – 1] 𝑖 ,𝑛 + 𝑧 [𝑡 ] 𝑖 ,𝑛 Control group: 𝛥𝑦 [𝑡 ] 𝑐 ,𝑛 = α * 𝑠 𝑐 ,𝑛 + β 𝑐 ,𝑛 * 𝑦 [𝑡 – 1] 𝑐 ,𝑛 + 𝑧 [𝑡 ] 𝑐 ,𝑛 Exercise variables. All exercise variables were converted to MET units. Vigorous exercise and Vigorous/Moderate exercise were the main dependent variables for this analysis. Model fit indices are presented in Table 10, and the parameter estimates of the best fitting model for each variable are presented in Table 11. Vigorous exercise and Vigorous/Moderate exercise showed (mostly) the same pattern of results. The proportional change models(Models 1b and 2b) improved fit of the data compared to the null models (ΔΧ 2 (Δdf) = 43.238 (1), p = <.001 for vigorous exercise; ΔΧ 2 (Δdf) = 33.028 (1), p = <.001 for vigorous/moderate exercise), the linear change models(Models 1c and 2c) further improved fit compared to both the null models (ΔΧ 2 (Δdf) = 49.083 (3), p = <.001for vigorous exercise; ΔΧ 2 (Δdf) = 54.448 (3), p = <.001for vigorous/moderate exercise) and the proportional change models (although, only marginally for vigorous exercise; ΔΧ 2 (Δdf) = 5.845 (2), p = .054 for vigorous exercise; ΔΧ 2 (Δdf) = 21.420 (2), p<.001 for vigorous/moderate exercise), and the dual change models(Models 1d and 2d) fit the data better than the other models (compared to the linear change model, ΔΧ 2 (Δdf) = 20.225 (1), p = <.001 for vigorous exercise; ΔΧ 2 (Δdf) = 30.199 (1), p = <.001 for vigorous/moderate exercise). No differences were observed across groups in any of the parameters (based on non-significant changes in model fit shown in Table 10). The SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 58 best fitting model for both variables was the dual change score model with all parameters held invariant across groups, with parameter estimates 𝛥𝑣𝑒𝑥 [𝑡 ] 𝑛 = α * 𝑠 𝑛 + β * 𝑣𝑒𝑥 [𝑡 – 1] 𝑛 = 20.309 + -0.942 * 𝑣𝑒𝑥 [𝑡 – 1] 𝑛 for vigorous exercise, and 𝛥𝑣 /𝑚𝑒𝑥 [𝑡 ] 𝑛 = α * 𝑠 𝑛 + β * 𝑣 /𝑚𝑒𝑥 [𝑡 – 1] 𝑛 = 24.650 + -0.873 * 𝑣 /𝑚𝑒𝑥 [𝑡 – 1] 𝑛 for vigorous/moderate exercise. The positive slope parameters and negative proportional change indicate increases in exercise over time that eventually asymptote at the point in which the proportional change element equals the constant change (which is approximately 21-28 time points of measurement, greatly outside the present data). Behavioral intentions. The behavioral intentions variables included intentions for vigorous exercise and intentions for vigorous/moderate exercise. Model fit indices are presented in Table 12, and parameter estimates for the best fitting model are shown in Table 13. The proportional change models (Models 3b and 4b) failed to significantly improve model fit over the null model for both the intentions for vigorous exercise (ΔΧ 2 (Δdf) = 0.556 (1), p = .456) and intentions for vigorous/moderate exercise variables (ΔΧ 2 (Δdf) = 1.614 (1), p = .204). The linear change models(Models 3c and 4c) greatly improved model fit for both variables compared to the null and proportional change models (intention for vigorous exercise: ΔΧ 2 (Δdf) = 45.216 (2), p <.001; intention for vigorous/moderate exercise: ΔΧ 2 (Δdf) = 59.056 (2), p < .001), and the dual change score models (Models 3d and 4d) improved model fit compared to the linear change model (intention for vigorous exercise: ΔΧ 2 (Δdf) = 18.918 (1), p< .001; intention for vigorous/moderate exercise: ΔΧ 2 (Δdf) = 28.389 (1), p< .001). No differences SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 59 were observed across groups in any of the parameters (based on non-significant changes in model fit shown in Table 12). The best fitting model for the strength of intentions variables was the dual change score model with all parameters held invariant across groups, with parameter estimates 𝛥𝑖𝑣𝑒𝑥 [𝑡 ] 𝑛 = α * 𝑠 𝑛 + β * 𝑖𝑣𝑒𝑥 [𝑡 – 1] 𝑛 = 27.524 + -0.745 * 𝑖𝑣𝑒𝑥 [𝑡 – 1] 𝑛 for intention for vigorous exercise, and 𝛥𝑖𝑣 /𝑚𝑒𝑥 [𝑡 ] 𝑛 = α * 𝑠 𝑛 + β * 𝑖𝑣 /𝑚𝑒𝑥 [𝑡 – 1] 𝑛 = 47.818 + -0.868 * 𝑖 𝑣 /𝑚𝑒𝑥 [𝑡 – 1] 𝑛 for intention for vigorous/moderate exercise. The positive slopes combined with the negative proportional change parameters indicate a curve that increases over time and eventually asymptotes (well outside the bounds of the data). Social-cognitive variables. The social-cognitive variables included self-efficacy, outcome expectations, and planning. Model fit indices are presented in Table 14, and the parameter estimates of the best fitting model for each variable are presented in Table 15. For self-efficacy, the proportional change model (Model 5b) failed to improve model fit compared to the null model (ΔΧ 2 (Δdf) = 1.852 (1), p = .174). The linear change model (Model 5c) improved model fit compared to the proportional change model (ΔΧ 2 (Δdf) = 25.167 (2), p< .001) and the null model (ΔΧ 2 (Δdf) = 27.019 (3), p< .001), and proved to be the best fitting model for self-efficacy, as none of the other models improved model fit compared to the linear change model (as indicated by non-significant changes shown in Table 14). The parameters of the best fitting models for self-efficacy are shown in Table 15, which can be written as 𝛥𝑆𝐸 [𝑡 ] 𝑛 = α * 𝑠 𝑛 + β * 𝑆𝐸 [𝑡 – 1] 𝑛 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 60 = -0.706 + 0 * 𝑆𝐸 [𝑡 – 1] 𝑛 The negative slope parameter, with no proportional change element, indicates a linear decrease in self-efficacy over time. However, although adding the linear-change element to the model significantly improved model fit compared to the null model, the slope estimate of -0.706 is not statistically significant by itself (p = .202). Therefore, while there may be some indication that self-efficacy decreased throughout the study, this decrease was minor. Outcome expectations exhibited negative variances for the slope when those parameters were estimated, so the slope variance for this variable was restricted to zero in order to run the models without error (which, additionally, requires the slope-intercept covariance to be set to zero as well). This change leads to the outcome expectations models having different numbers of degrees of freedom compared to the other models, and the last two models (freeing the covariance across groups and freeing the intercept and slope variances across groups) were unable to be tested. Both the proportional change model (Model 6b) and the linear change model (Model 6c) failed to improve model fit compared to the null model (ΔΧ 2 (Δdf) = 0.006 (1), p = .938; ΔΧ 2 (Δdf) = -5.069 (1); respectively). The linear change model and the proportional change model are not directly comparable because they contain the same number of degrees of freedom. The dual change model (Model 6d) significantly improved model fit compared to the proportional change and linear change model, but failed to improve model fit compared to the null model (ΔΧ 2 (Δdf) = 1.835 (2), p = .400). Finally, freeing the slope across groups in the dual change model (Model 6e) improved model fit compared to holding all parameters of the dual change model invariant, but this too was insignificant compared to the null model (ΔΧ 2 (Δdf) = 6.721 (3), p = .081). Therefore, the best fitting model was the null (no growth) model, with a baseline average (intercept) of 25.884 and a variance of 28.047. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 61 For planning, the proportional change model (Model 7b) also failed to improve model fit compared to the null model (ΔΧ 2 (Δdf) = 0.728 (1), p = .394). Additionally, the linear change model (Model 7c) failed to improve model fit compared to the proportional change model (ΔΧ 2 (Δdf) = 4.446 (2), p = .108) and the null model (ΔΧ 2 (Δdf) = 5.174 (3), p= .160). However, the dual change score model (Model 7d) significantly improved model fit compared to the other three models (ΔΧ 2 (Δdf) = 18.516 (4), p< .001 compared to the null model; ΔΧ 2 (Δdf) = 17.788 (3), p< .001 compared to the proportional change model; ΔΧ 2 (Δdf) = 13.342 (1), p< .001 compared to the linear change model). Although the linear change and the proportional change elements failed to improve model fit individually, including both in the model significantly improved fit. No differences in parameters were observed across groups (indicated by non-significant changes in model fit shown in Table 14). Therefore, the best fitting model for planning was the dual change score model with all parameters held invariant across groups, with parameter estimates 𝛥𝑝𝑙𝑎𝑛𝑛𝑖𝑛𝑔 [𝑡 ] 𝑛 = α * 𝑠 𝑛 + β * 𝑝𝑙𝑎𝑛𝑛𝑖𝑛𝑔 [𝑡 – 1] 𝑛 = 12.386 + -1.028 * 𝑝𝑙𝑎𝑛𝑛𝑖𝑛𝑔 [𝑡 – 1] 𝑛 where the positive slope and negative proportional change parameters indicate a curve that increases over time and eventually asymptotes. Motivation variables. Motivation variables included commitment, autonomous motivation, and controlled motivation. Model fit indices are presented in Table 16, and the parameter estimates of the best fitting model for each variable are presented in Table 17. The proportional change models (Models 8b, 9b, and 10b) failed to improve model fit compared to the null model for all three variables (commitment: ΔΧ 2 (Δdf) = 0.077 (1), p = .718; SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 62 autonomous motivation: ΔΧ 2 (Δdf) = 3.791 (1), p = .052; controlled motivation: ΔΧ 2 (Δdf) = 0.957 (1), p = .328). For commitment, the linear change model (Model 8c) significantly improved model fit compared to the null and proportional change models (ΔΧ 2 (Δdf) = 40.163 (3), p< .001; and ΔΧ 2 (Δdf) = 40.086 (2), p< .001, respectively), and the dual change score model (Model 8d) significantly improved model fit compared to the linear change model (ΔΧ 2 (Δdf) = 25.904 (1), p< .001). Additionally, there were no significant differences observed across groups. For autonomous motivation, the linear change model (Model 9c) failed to improve fit compared to the null model (ΔΧ 2 (Δdf) = 5.295 (3), p = .15), however the dual change model (Model 9d) significantly improved fit compared to the linear change (ΔΧ 2 (Δdf) = 7.803 (1), p = .005) and null models (ΔΧ 2 (Δdf) = 13.098 (4), p = .011). There were no significant differences observed across groups. The best fitting model for both commitment and autonomous motivation was the dual-change score model with all parameters held invariant across groups. The parameter estimates of the best fitting models for both variables are shown in Table 17, which can be written as 𝛥𝑐𝑜𝑚𝑚𝑖𝑡 [𝑡 ] 𝑛 = α * 𝑠 𝑛 + β * 𝑐𝑜𝑚𝑚𝑖𝑡 [𝑡 – 1] 𝑛 = 14.004 + -0.796 * 𝑐𝑜𝑚𝑚𝑖𝑡 [𝑡 – 1] 𝑛 for commitment, and 𝛥𝑎𝑢𝑡 _𝑚 [𝑡 ] 𝑛 = α * 𝑠 𝑛 + β * 𝑎𝑢𝑡 _𝑚 [𝑡 – 1] 𝑛 = 11.755 + -1.047 * 𝑎𝑢𝑡 _𝑚 [𝑡 – 1] 𝑛 for autonomous motivation. The positive slope and negative proportional change parameters for both models indicate an increase in commitment that eventually asymptotes over time. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 63 For controlled motivation, the linear change model (Model 10c) significantly improved model fit compared to the proportional change model (ΔΧ 2 (Δdf) = 7.614 (2), p = .022) and null model (ΔΧ 2 (Δdf) = 8.571 (3), p = .036). The dual change score model (Model 10d) failed to improve model fit compared to the linear change model (ΔΧ 2 (Δdf) = 2.637 (1), p = .104), indicating that the data for controlled motivation appear to include a linear change element but not a proportional change element. None of the models allowing parameters to differ across groups significantly improved model fit compared to the linear change model alone. For example, releasing constraints on the slope-intercept covariance across groups (Model 10f) improved model fit compared to Model 10e (ΔΧ 2 (Δdf) = 4.855 (1), p = .028), but it did not improve fit compared to the linear change model with all parameters invariant across groups (ΔΧ 2 (Δdf) = 8.821 (4), p = .065). Therefore, the best fitting model is the linear change model holding all parameters invariant across groups, with parameter estimates 𝛥𝑐𝑜𝑛 _𝑚 [𝑡 ] 𝑛 = α * 𝑠 𝑛 + β * 𝑐𝑜𝑛 _𝑚 [𝑡 – 1] 𝑛 = -0.346 + 0 * 𝑐𝑜𝑛 _𝑚 [𝑡 – 1] 𝑛 including a negative slope indicating that controlled motivation decreased over time. Although a linear change model improved the fit of the data, the slope (-0.346) was not significantly different from zero (p = .232). Therefore, caution should be taken when interpreting these results, as any changes in controlled motivation were minimal. Goal structure variables. The goal structure variables included (a) the overall rank of the health goal-cluster; (b) the composite scores for goal facilitation/conflict with exercise and goal facilitation/conflict with not exercising; and (c) the four individual conflict and facilitation scores (facilitation with exercise, conflict with exercise, facilitation with not exercising, and conflict with not exercising). SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 64 Model fit indices for the rank of the health goal, conflict/facilitation with exercise, and conflict/facilitation with not exercising are presented in Table 18, and the parameter estimates of the best fitting model for each variable are presented in Table 19. The rank of the health goal- cluster and conflict/facilitation with exercise both showed no change over time, as none of the models improved fit compared to the null model. For example, for conflict/facilitation with exercise, the dual change score model (Model 12d) significantly improved model fit compared to the linear change model (Model 12c; ΔΧ 2 (Δdf) = 5.474 (1), p = .019), however the dual change score model failed to improve fit compared to the proportional change model (Model 12b; ΔΧ 2 (Δdf) = 7.477 (3), p = .058) and the null model (ΔΧ 2 (Δdf) = 8.951 (4), p = .062). The parameters for the no growth models were a baseline average of 4.968 with a variance of 1.770 for the rank of the health goal-cluster and a baseline average of 49.164 with a variance of 444.534 for conflict/facilitation with exercise. For conflict/facilitation with not exercising, the proportional change model (Model 13b) failed to improve model fit compared to the null model (ΔΧ 2 (Δdf) = 2.366 (1), p = .124,), and the linear change model (Model 13c) failed to improve model fit compared to the proportional change model (ΔΧ 2 (Δdf) = 5.083 (2), p = .079). However, the dual change model (Model 13d) significantly improved model fit compared to the linear change, proportional change, and null models (linear change: ΔΧ 2 (Δdf) = 5.572 (1), p = .018; proportional change: ΔΧ 2 (Δdf) = 10.655 (3), p = .014; null model: ΔΧ 2 (Δdf) = 13.021 (4), p = .017). Relaxing the constraints across groups for the slope (Model 13e) and the proportional change (Model 13f) failed to improve model fit compared to the dual change score model with these parameters invariant (ΔΧ 2 (Δdf) = .392 (1), p = .531; ΔΧ 2 (Δdf) = .694 (2), p = .707; respectively). However, allowing the covariance to be freely estimated across groups (Model 13g) did improve SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 65 model fit compared to the dual change score model (ΔΧ 2 (Δdf) = 10.396 (3), p = .016). Additionally, this improvement in model fit did not require the slope and proportional change parameters to vary across groups, as the significant improvement in fit upon relaxing constraints on the covariance between the slope and intercept remained when the slope and proportional change were held invariant across groups (ΔΧ 2 (Δdf) = 10.130 (1), p < .001; fit indices not shown in Table 18). Therefore, the best fitting model is a dual change score model that allows the slope and intercept covariance to be freely estimated across groups while holding all other parameters invariant. The parameter estimates in Table 19 describing the change over time (which were equal for both groups) can be written as 𝛥𝑐𝑓 _𝑛𝑜 _𝑒𝑥 [𝑡 ] 𝑛 = α * 𝑠 𝑛 + β * 𝑐𝑓 _𝑛𝑜 _𝑒𝑥 [𝑡 – 1] 𝑛 = -23.847 + -0.974 * 𝑐𝑓 _𝑛𝑜 _𝑒𝑥 [𝑡 – 1] 𝑛 Where the negative slope combined with the negative proportional change element indicate a curve that decreases exponentially over time. The conflict/facilitation with not exercise composite score was a summed score in which negative values indicate conflict with not exercising and positive scores indicate facilitation with not exercising. The sample decreased over time, indicating that the sample reported greater degrees of conflict with not exercising and/or lesser degrees of facilitation with not exercising over time. Additionally, the Intervention group had a slope-intercept covariance of 878.242 while the Control group had a covariance of 604.523, indicating changes in the Intervention group were more strongly associated with their baseline averages. That is, people with low starting values were more likely to decrease over time in the degree they perceived not exercising would conflict/facilitate them in achieving their most important life goals, and this relationship was stronger among the Intervention group participants. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 66 Model fit indices for conflict with exercise, facilitation with exercise, conflict with not exercising, and facilitation with not exercising are presented in Table 20, and parameter estimates for the best fitting model are presented in Table 21. Conflict with exercise and facilitation with not exercising both exhibited negative slope variances for several of the models, which required these parameters to be constrained to zero in later models, and thus, the last two models were unable to be tested. For conflict with exercise, the proportional change model (Model 15b) significantly improved model fit compared to the null model (ΔΧ 2 (Δdf) = 7.666 (1), p = .006), however the linear change model (Model 15c; which required the slope variance to be constrained to zero, and was not directly comparable to the proportional change model because they contained the same number of degrees of freedom) was not significantly different from the null model (ΔΧ 2 (Δdf) = 3.362 (1), p = .079). The dual change score model (Model 15d; which DID run correctly when the slope variance was estimated) significantly improved model fit compared to the proportional change model (ΔΧ 2 (Δdf) = 11.995 (3), p = .007). No differences were observed across groups in any of the parameters (based on non-significant changes in model fit shown in Table 20). Therefore, the best fitting model for conflict with exercise was the dual change score model constraining all parameters invariant across groups, with parameter estimates 𝛥𝑐𝑜𝑛 _𝑒𝑥 [𝑡 ] 𝑛 = α * 𝑠 𝑛 + β * 𝑐𝑜𝑛 _𝑒𝑥 [𝑡 – 1] 𝑛 = 5.805 + -1.607 * 𝑐𝑜𝑛 _𝑒𝑥 [𝑡 – 1] 𝑛 Where the positive slope and the negative proportional change parameters indicate a curve that increases over time but asymptotes rather quickly (compared to change curves for other variables that show this same pattern). SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 67 For facilitation with not exercising, the proportional change model (Model 16b) significantly improved model fit compared to the null model (ΔΧ 2 (Δdf) = 10.297 (1), p < .001). None of the other models improved model fit compared to the proportional change model. Releasing the proportional change element across groups (Model 16f) significantly improved model fit compared to the dual change score model (Model 16d; ΔΧ 2 (Δdf) = 6.092 (2), p = .048), but this change was not significant compared to the proportional change model holding all parameters invariant across groups (ΔΧ 2 (Δdf) = 7.229 (3), p = .065). The parameter estimates for the proportional change model can be written as 𝛥𝑓𝑎𝑐 _𝑛𝑜 _𝑒𝑥 [𝑡 ] 𝑛 = α * 𝑠 𝑛 + β * 𝑓𝑎𝑐 _𝑛𝑜 _𝑒𝑥 [𝑡 – 1] 𝑛 = 0 + -0.090 * 𝑓𝑎𝑐 _𝑛𝑜 _𝑒𝑥 [𝑡 – 1] 𝑛 Which shows a slightly decreasing trend over time, indicating that participants perceived not exercising at all to be less facilitating of the goal-clusters over time. The conflict with not exercising models ran correctly without altering the variables/models. The proportional change model (Model 17b) and linear change model (Model 17c) failed to improve model fit compared to the null model (ΔΧ 2 (Δdf) = 030 (1), p = .863; ΔΧ 2 (Δdf) = 1.778 (3), p = .620; respectively). The dual change score model (Model 17d) significantly improved model fit compared to the linear change (ΔΧ 2 (Δdf) = 13.067 (1), p < .001), proportional change (ΔΧ 2 (Δdf) = 14.845 (3), p = .002), and null models (ΔΧ 2 (Δdf) = 14.875 (4), p = .005). None of the models releasing parameters across groups improved model fit compared to the dual change score model. Releasing the proportional change in addition to the slope across groups (Model 17f) significantly improved fit compared to releasing the slope alone (Model 17e ΔΧ 2 (Δdf) = 4.632 (1), p = .031), however this improvement in fit was not significant compared to the dual change score model with all parameters equal (ΔΧ 2 (Δdf) = SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 68 4.660 (2), p = .097). Therefore, the best fitting model was the dual change score model constraining all parameters to be invariant across groups, with parameter estimates 𝛥𝑐𝑜𝑛 _𝑛𝑜 _𝑒𝑥 [𝑡 ] 𝑛 = α * 𝑠 𝑛 + β * 𝑐𝑜𝑛 _𝑛𝑜 _𝑒𝑥 [𝑡 – 1] 𝑛 = 6.332 + -1.042 * 𝑐𝑜𝑛 _𝑛𝑜 _𝑒𝑥 [𝑡 – 1] 𝑛 Where the positive slope and negative proportional change indicate a curve that increases slightly and asymptotes relatively quickly. The facilitation with exercise models also ran correctly without further adjusting the variables/models. The proportional change model (Model 14b) failed to improve fit compared to the null model (ΔΧ 2 (Δdf) = .003 (1), p = .956), and the linear change model (Model 14c) failed to improve model fit compared to the proportional change model (ΔΧ 2 (Δdf) = 4.874 (2), p = .087). The dual change score model (Model 14d) significantly improved model fit compared to the linear change, proportional change, and null models (linear change: ΔΧ 2 (Δdf) = 4.962 (1), p = .026; proportional change: ΔΧ 2 (Δdf) = 9.846 (3), p = .020; null model: ΔΧ 2 (Δdf) = 9.836 (4), p = .043). Additionally, relaxing the slope parameter across groups (Model 14e) improved model fit compared to the dual change score model (ΔΧ 2 (Δdf) = 4.104 (1), p = .043) as well as the null model (ΔΧ 2 (Δdf) = 13.943 (5), p = .016). None of the other models significantly improved model fit. Therefore, the best fitting model was the dual change model allowing the slope parameters to vary across groups, with parameter estimates Intervention group: 𝛥𝑓𝑎𝑐 _𝑒𝑥 [𝑡 ] 𝑖 ,𝑛 = α * 𝑠 𝑖 ,𝑛 + β * 𝑓𝑎𝑐 _𝑒𝑥 [𝑡 – 1] 𝑖 ,𝑛 = 46.061 + -0.828 * 𝑓𝑎𝑐 _𝑒𝑥 [𝑡 – 1] 𝑖 ,𝑛 Control group: 𝛥𝑓𝑎𝑐 _𝑒𝑥 [𝑡 ] 𝑐 ,𝑛 = α * 𝑠 𝑐 ,𝑛 + β * 𝑓𝑎𝑐 _𝑒𝑥 [𝑡 – 1] 𝑐 ,𝑛 = 49.564 + -0.828 * 𝑓𝑎𝑐 _𝑒𝑥 [𝑡 – 1] 𝑐 ,𝑛 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 69 where the positive slope and negative proportional change parameters indicate curves that increase over time and eventually asymptote (well outside the bounds of the data). Additionally the Control group had larger gains in the degree that they perceived exercise would facilitate them in achieving their other important life goals compared to the Intervention group, as shown by the differences in linear change over time (49.564 and 46.061, respectively). Future exercise goals. The last set of variables analyzed were people’s goals for the amount of exercise they planned to complete four weeks in the future. These variables included future vigorous exercise goals and future vigorous/moderate exercise goals (all in MET units). These variables were standardized using the mean and standard deviation from the first time point of measurement for each variable in order to run the models without error. Goodness of fit indices are shown in Table 22, and the parameter estimates for the best fitting models are shown in Table 23. Additionally, the slope-intercept covariance and the slope variances were required to be set to zero in order to run several of the models for both variables. Therefore, the last two models were unable to be tested, and several of the models had differing degrees of freedom compared to the other variables. For future vigorous exercise goals, the proportional change model (Model 18b) significantly improved model fit compared to the null model (ΔΧ 2 (Δdf) = 19.373 (1), p < .001). The linear change model (Model 18c), which required the slope-intercept covariance and slope variance restricted to zero to run correctly, failed to improve model fit compared to the null model (ΔΧ 2 (Δdf) = 2.557 (1), p = .110), and is not directly comparable to the proportional change model due to equal degrees of freedom. The dual change model (Model 18d) significantly improved fit compared to the null (ΔΧ 2 (Δdf) = 26.439 (2), p <.001), proportional SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 70 change (ΔΧ 2 (Δdf) = 7.066 (1), p = .008), and linear change models (ΔΧ 2 (Δdf) = 23.88 (1), p < .001). None of the other models significantly improved model fit compared to the linear change model (as indicated by non-significant changes in model fit shown in Table 22; the last model was not estimated because the program failed to converge correctly). Therefore, the best fitting model was the dual change model holding all parameters invariant across groups, with parameter estimates 𝛥𝑓𝑢𝑡 _𝑖𝑛𝑡 _𝑣𝑖𝑔 [𝑡 ] 𝑛 = α * 𝑠 𝑛 + β * 𝑓 𝑢 𝑡 _𝑖𝑛𝑡 _𝑣𝑖𝑔 [𝑡 – 1] 𝑛 = -.224 + .198 * 𝑓𝑢𝑡 _𝑖𝑛𝑡 _𝑣𝑖𝑔 [𝑡 – 1] 𝑛 where the negative slope and positive proportional change estimates indicate a curve that decreases slightly at the second time point before beginning to increase at later time points. For future vigorous/moderate exercise goals, the proportional change model (Model 19b) significantly improved model fit compared to the null model (ΔΧ 2 (Δdf) = 30.492 (1), p < .001), while the linear change model (Model 19c; which also required slope-intercept covariance and slope variances restricted to zero) failed to improve model fit compared to the null model (ΔΧ 2 (Δdf) = .789 (2), p = .374). The linear change and proportional change models were not directly comparable due to equal degrees of freedom. The dual change model (Model 19d) failed to improve fit compared to the proportional change model (ΔΧ 2 (Δdf) = 3.007 (1), p = .083). Finally, none of the other models improved fit compared to the proportional change model. Therefore, the best fitting model was the proportional change model holding all parameters invariant across groups, with parameter estimates 𝛥𝑓𝑢𝑡 _𝑖𝑛𝑡 _𝑣𝑖𝑔𝑚𝑜𝑑 [𝑡 ] 𝑛 = α * 𝑠 𝑛 + β * 𝑓𝑢𝑡 _𝑖𝑛𝑡 _𝑣𝑖𝑔𝑚𝑜𝑑 [𝑡 – 1] 𝑛 = 0 + .248 * 𝑓𝑢𝑡 _𝑖𝑛𝑡 _𝑣𝑖𝑔𝑚𝑜𝑑 [𝑡 – 1] 𝑛 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 71 Where the positive proportional change element indicates a curve that increases proportionally over time. Goal-structure analysis Means, medians, and standard deviations for the seven goal-structure scores are shown in Table 6. Goal conflict with exercise and not exercising were coded so that high values correspond to high levels of conflict (as opposed to negative values corresponding to high levels of conflict). Due to the method for calculating the last four conflict and facilitation scores, in which the conflict or facilitation is calculated alone (i.e., the non-aggregated scores), these variables are typically skewed. Square-root transformations to make variables more closely resemble a normal distribution were used for conflict with exercise, conflict with not exercising, and facilitation with not exercising. Relationships with exercise variables. Analyses used Pearson r correlations and multiple regression to test associations between the seven goal-structure scores and the exercise constructs. Exercise constructs were: (1) vigorous exercise (at Time 1, Time 2, and Time 3); (2) vigorous/moderate exercise (at Time 1, Time 2, and Time 3); and (3) success in accomplishing intentions for vigorous/moderate exercise (at Time 2 and Time 3). Correlations were tested for goal-structure scores at each time point with past, concurrent, and future exercise. Associations with past exercise (e.g., goal facilitation at Time 2 with exercise at Time 1) were included because it is possible that the amount of exercise someone completes influences their perception of conflict/facilitation, as opposed to these perceptions causing exercise, so all associations were tested. The above correlations were tested on each of the five samples. Significant correlations occurring in at least three of the samples were considered robust and were averaged and reported in Table 24. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 72 Four of the seven goal-structure scores correlated with very few or none of the exercise variables. The combined conflict/facilitation with not exercising failed to correlate with any exercise variable consistently, conflict with exercise exhibited only one consistent correlation, facilitation with not exercising exhibited only two consistent correlations, and conflict with not exercising exhibited only three consistent correlations. Of the remaining four goal-structure scores, the aggregated conflict/facilitation with exercise and facilitation with exercise scores were most frequently correlated with exercise variables. However, it seems that facilitation with exercise was driving the relationships between exercise variables and the aggregated conflict/facilitation with exercise score, as these two variables had almost identical findings and facilitation with exercise generally had slightly larger associations. The overall rank of the health goal-cluster was significantly correlated with several exercise variables, but only with the scores for the degree the person succeeded in accomplishing their intentions to exercise. Based on the above findings, the aggregated conflict/facilitation with exercise and conflict/facilitation with not exercising scores were removed from subsequent analyses. Aggregated conflict/facilitation with not exercising failed to consistently correlate with any of the exercise variables at any time point. The aggregated conflict/facilitation score with exercise was significantly correlated with many exercise variables, but this relationship appeared to be driven by facilitation with exercise alone as the two variables had almost identical relationships. Because facilitation with exercise was correlated with almost all of the same variables as the aggregated score, and because the other conflict and facilitation scores may potentially add predictability in other analyses, the aggregated scores were deemed unsatisfactory. Therefore, subsequent analyses used only five goal-structure scores: (1) the overall rank of the health goal- SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 73 cluster, (2) conflict with exercise, (3) facilitation with exercise, (4) conflict with not exercising, and (5) facilitation with not exercising. Multiple regression with exercise variables regressed onto goal-structure scores were used to determine if (a) the associations observed in the correlation analyses remained when all five of the goal-structure scores are included in analyses and (b) if multiple goal-structure scores significantly predict exercise variables. Due to the assumed causal relation in regression, only associations with concurrent and future exercise scores were tested (i.e., goal-structure at Time 1 predicting exercise at Time 1, Time 2, and Time 3; goal structure at Time 2 predicting exercise at Time 2 and Time 3; etc.,). The same method for determining robust relationships for correlations was used: only significant finding across at least three out of the five samples were included. Table 25 reports results of the regression analyses. It is clear that facilitation with exercise is the strongest and most frequent predictor of exercise behavior, which is consistent with the results from the simple correlations. Facilitation with exercise significantly predicted vigorous and vigorous/moderate exercise at Time 2 and at Time 3, as well as the degree the person met their vigorous/moderate intentions for exercise at Time 3. The overall rank of the health goal at Time 2 significantly predicted the degree the person met their vigorous/moderate exercise intention at Time 2. Finally, conflict with exercise at Time 2 significantly predicted the degree the person met their vigorous/moderate exercise at Time 3. It’s also interesting to note that multiple goal-structure scores did not significantly predict exercise consistently in any of the analyses (while there were a few times this occurred within individual sub-samples, but never consistently). SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 74 Relationships with other social-cognitive and motivational constructs. The motivation constructs were: autonomous motivation, controlled motivation, goal commitment, and strength of intention for completing vigorous and moderate exercise. Analyses used the same five samples as above. Pearson r correlations were tested between the five remaining goal- structure scores and the motivation constructs. However, there were so many significant correlations that interpretation was difficult. Therefore, exploratory factor analysis with a Direct Oblimin (oblique) rotation was utilized to evaluate relationships between the goal-structure scores and motivation constructs. The five goal-structure scores and motivation constructs at each time point were entered together (i.e., all Time 1 scores together, all Time 2 scores together, and all Time 3 scores together). The motivation constructs were goal commitment, autonomous motivation, controlled motivation, and behavioral intentions for vigorous and vigorous/moderate exercise. Factor analyses were repeated on each of the five sub-samples. Results of the factor analyses were fairly consistent across samples and time points. Based on the eigenvalue greater than one rule, the factor analyses typically concluded a five- factor solution. However, the factors were inconsistent, ambiguous, and the scree plot indicated far fewer factors (between 2-3 factors). Forcing a two factor solution accounted for approximately 38.5% of the variance in the data, and forcing a three factor solution accounted for approximately 50%. A typical two factor solution is shown in Table 26. In general, conflict with exercise, conflict with not exercising, and facilitation with not exercising tended to factor together as the second factor, while facilitation with exercise (always) and the overall rank of the heath goal- cluster (almost always) factored with the other motivation variables. Another finding was that SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 75 controlled motivation—often considered a poorer form of overall motivation (Deci & Ryan, 2004; Sheldon & Elliot, 1998)—typically failed to load highly on either factor (i.e., < 0.30). Forcing a three factor solution was less consistent as the two factor solution, but a typical pattern of results is shown in Table 27. The most consistent finding was that conflict with exercise, conflict with not exercising, and facilitation with not exercising continued loading together. Additionally, facilitation with exercise and the overall rank of the heath goal-cluster continued loading with the other motivation constructs. The main difference between the two factor and three factor solutions was that the motivation constructs (including facilitation with exercise and the overall rank score) consistently split into two factors, however the groupings of the two factors were inconsistent across time points and samples. In summary, whether a two or three factor solution is desired, goal facilitation and the overall rank of the health goal-cluster tended to factor with other motivation constructs. The conflict with exercise, conflict with not exercising, and facilitation with not exercising scores tended to factor together separately from the other motivation constructs. These findings suggest that facilitation with exercise and the overall rank of the health goal-cluster may be useful additions to a latent variable composed of several motivation constructs. Measurement invariance of motivation latent variable Correlations among commitment, autonomous motivation, facilitation with exercise, and the overall rank of the health goal-cluster are shown in Table 28. The measurement model tested is shown in Figure 11 with values for Model 20d (strict invariance across the whole sample). Facilitation with exercise was standardized using the Time 1 mean and variance in order to run the multi-group analyses without error due to large variances of the unstandardized scores. First, a Rasch model (Model 20a) forcing all factor loadings to 1.00, representing a summed composite SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 76 score, was compared to a weak invariance model (Model 20b) allowing factor loadings to differ across items but be constrained across time within items, representing a latent variable (McDonald, 1999; Meredith, 1993). Fit indices are presented in Table 29. The weak invariance model yielded a significantly better fit to the data than the Rasch model (𝛥 𝜒 2 /Δdf= 218.160 / 3, p< .001), indicating that calculating a latent variable is a better fit to the data than simple summation. The next series of models tested for measurement invariance over time by subsequently constraining observed variable means (“strong invariance”) and observed variable variances (“strict invariance”) over time on the complete sample and comparing change in model fit after each step. The weak invariance model yielded a good fit to the data with 𝜒 2 /df = 76.149 / 45, CFI = .984, TLI = .977, RMSEA =.031. The strong invariance model (Model 20c), while yielding a good fit of the data with 𝜒 2 /df = 106.981 / 51, CFI = .971, TLI = .963, RMSEA = .039, significantly worsened fit compared to the weak invariance model (𝛥 𝜒 2 /Δdf= 30.832 / 6, p< .001). The worsening in fit after constraining observed variable means indicates potential issues of scalar invariance, which would make tests of mean differences in the latent variable over time problematic (discussed further below). The strict invariance model (Model 20d) also yielded good fit to the data with 𝜒 2 /df = 119.12/59, CFI = .969, TLI = .966, RMSEA = .038, and was an insignificant change in model fit compared to the strong invariance model (𝛥 𝜒 2 /Δdf= 12.139 / 8, p = .145). Autonomous motivation had the highest loading (1.093) followed by commitment (1.00), the rank of the health goal-cluster (.276), and facilitation with exercise (.117). The next series of models tested for measurement invariance across study group and gender (Intervention vs. Control and males vs. females). Fit indices for both tests are shown in SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 77 Table 30. The latent variable means and variances were held invariant across groups and the latent variable covariances were restrained to zero in order to run the models without error. These constraints led to models that fit the data poorly, but measurement invariance can still be tested by evaluating changes in model fit when forcing parameters invariant across groups and subsequently releasing the constraints. There was a non-significant difference in model fit when relaxing these constraints across gender (Model 21a compared to 21b; 𝛥 𝜒 2 /Δdf= 17.163 / 10, p = .075), indicating that the motivation latent variable was invariant for males and females. However, relaxing the constraints across study group (Model 22a compared to 22b) yielded a significant improvement in model fit (𝛥 𝜒 2 /Δdf= 34.127 / 10, p< .001), indicating that the motivation latent variable may not be invariant for the Intervention and Control groups. Models relaxing one parameter constraint at a time (i.e., factor loadings, observed variable means, and observed variable variances) were used to explore the source of the invariance across study groups. These models found that the source of invariance was due to relaxing the observed variable means over time, which significantly worsened model fit compared to all parameters invariant across groups (𝛥 𝜒 2 /Δdf= 22.493 / 3, p< .001). Further models that released the observed variable means of each item individually indicated that facilitation with exercise (𝛥 𝜒 2 /Δdf= 10.095 / 1, p = .002) and autonomous motivation (𝛥 𝜒 2 /Δdf= 16.391 / 1, p< .001) were the main sources of invariance across groups. Measurement Model Results Summary. All latent variable models (“weak,” “strong,” and “strict” invariance) yielded good fits to the data. The strong invariance model significantly worsened model fit indicating a potential violation of scalar invariance, which appeared to be due to differences across study groups over time in two variables (facilitation with exercise and autonomous motivation). However, because the latent variable models fit the data well, and the SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 78 worsening of fit when testing scalar invariance was relatively minor, it is concluded that the latent variable was suitable for use in analyses evaluating the utility of including the motivation latent variable within a social cognitive model of self-regulation. Predictability of motivation latent variable. In all models, exercise dependent variables (vigorous exercise, vigorous/moderate exercise, and success in accomplishing vigorous/moderate exercise goals) were predicted by cognitions measured at the previous time point. That is, exercise variables at Time 2 were predicted by cognitions measured at Time 1, and exercise dependent variables at Time 3 were predicted by cognitions measured at Time 2. Two sets of models were used to test the predictability of the latent variable of motivation for predicting exercise behavior. The first series of models, shown in Figure 12, included only an exercise variable, strength of intentions, and the motivation latent variable. The second series of models, shown in Figure 13, included self- efficacy and outcome expectations as covariates in addition to the previously mentioned variables to observe how relationships changed after including a more complete social-cognitive model in the analyses. The progression of tests were the same for the two series of models. The first model tested for no relationship between exercise and strength of intentions or motivation(Figures 12a & 13a), followed by models testing the relationship between exercise and strength of intentions (but not motivation; Figures 12b & 13b), the relationship between exercise and motivation (but not strength of intentions; Figures 12c & 13c), the relationships between exercise and both motivation and strength of intentions(Figures 12d & 13d), and finally, the relationships between exercise and strength of intentions and the individual observed variables comprising the motivation latent variable(Figures 12e & Figures 13e). The predictability of motivation was SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 79 tested by observing whether model fit was improved after adding the regression parameter between exercise and motivation, as well as the significance of said regression parameter. These models also allowed for comparing the relative effects of behavioral intentions and motivation for predicting exercise by comparing the regression coeffecients of these variables. After testing changes in model fit for each model, the a priori model (allowing both intentions and motivation to predict exercise) is discussed in more detail for each set of models for consistency purposes regardless of whether it provided the best fit of the data. Predicting Exercise at Time 2. The path diagrams including only exercise, intentions, and motivation with parameter estimates predicting Time 2 vigorous exercise are shown in Figure 14, and model fit indices are shown in Table 31. The null model (Model 23a) had a 𝜒 2 /df = 120.006 / 10, CFI = .660, TLI = .490, RMSEA =.124. Adding the intentions pathway (Model 23b) significantly improved model fit compared to the null model (𝛥 𝜒 2 /Δdf= 79.226 / 9, p< .001), as did adding the motivation pathway (Model 23c; 𝛥 𝜒 2 /Δdf= 51.643 / 9, p< .001). Including both pathways together (Model 23d) significantly improved model fit compared to the next best fitting model (𝛥 𝜒 2 /Δdf= 12.506 / 1, p< .001), while removing the latent variable (Model 23e) significantly and substantially worsened model fit compared to all other models (𝜒 2 /df = 177.939 / 6, CFI = .468, TLI = -.330, RMSEA =.200). Therefore, the a priori model was the best fitting model (Model 23d), which included both regression pathways. The model explained 16.9% of the variance in Time 2 vigorous exercise and 11.7% of the variance of intentions. Intentions were significantly predicted by motivation (β = .341, p < .001, [.243, .439]), and Time 2 vigorous exercise was significantly predicted by Intentions (β = .294, p < .001, [.209, .381]) and motivation (β = .204, p < .001, [.090, .318]). There was also a significant indirect effect of motivation predicting Time 2 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 80 vigorous exercise via intentions (β = .100, p < .001, [.061, .139]), creating a total effect of motivation of .304 (p < .001, [.192, .416]). The pattern of results for the fit indices were mostly the same after including the other social-cognitive variables (Figure 15). The null model (Model 24a) had a model fit of 𝜒 2 /df = 210.180 / 16, CFI = .783, TLI = .619, RMSEA =.130. Adding the intentions (Model 24b) and motivation pathways (Model 24c) significantly improved model fit compared to the null model (𝛥 𝜒 2 /Δdf= 60.215 / 1, p< .001; 𝛥 𝜒 2 /Δdf= 53.078 / 1, p< .001; respectively), and adding both pathways together (Model 24d) improved model fit compared to the next best fitting model (𝛥 𝜒 2 /Δdf= 5.031 / 1, p = .025). However, removing the latent variable in this model (Model 24e) significantly improved model fit (𝛥 𝜒 2 /Δdf= 87.952/ 8, p< .001), which was an unexpected finding. This improvement in model fit seemed to be a result of relationships between outcome expectations and the other variables in the model, as post-hoc analyses revealed that dropping outcome expectations from the model resulted in a worsening of fit after removing the latent variable (𝜒 2 /df = 177.951 / 6, CFI = .746, TLI = .112, RMSEA =.200; fit indices not reported in Table 31). The a priori model (Model 24d) that includes regressions from intentions and motivation will be discussed in detail although it technically did not provide the best fit to the data. Model 24d explained 18.4% of the variance in Time 2 vigorous exercise and 14.8% of the variance of intentions. Intentions were significantly predicted by motivation (β = .327, p = .020, [.184, .470]), outcome expectations (β = .116, p= .038, [.006, .226]), but not self-efficacy (β = -.017, p= .885, [-.252, .218]). Time 2 vigorous exercise was significantly predicted by intentions (β = .295, p < .001, [.205, .385]), motivation (β = .361, p = .020, [.184, .470]), but not by outcome expectations (β = -.109, p= .083, [-.232, .014]) or self-efficacy (β = -.117, p = 389, [- SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 81 .385, .150]). There was also a significant indirect effect of motivation predicting Time 2 vigorous exercise via intentions (β = .096, p = .017, [.055, .137]), creating a total effect for motivation of β = .455 (p = .014, [.284, .626]). The path diagrams with parameter estimates predicting Time 2 vigorous/moderate exercise with only intentions and motivation are shown in Figure 16, and model fit indices are shown in Table 32. The null model (25a) exhibited poor fit to the data with a 𝜒 2 /df = 137.782 / 10, CFI = .645, TLI = .467, and RMSEA =.134. Adding the intentions pathway (Model 25b) significantly improved model fit compared to the null model (𝛥 𝜒 2 /Δdf= 68.535 / 9, p< .001), as did adding the motivation pathway (Model 25c; 𝛥 𝜒 2 /Δdf= 76.319 / 9, p< .001). Including both pathways together (Model 25d) significantly improved model fit compared to the next best fitting model (𝛥 𝜒 2 /Δdf= 17.519 / 1, p< .001), while removing the latent variable (Model 25e) significantly and substantially worsened model fit compared to all other models (𝜒 2 /df = 177.893 / 6, CFI = .522, TLI = -.194, RMSEA =.200). Therefore, the a priori model (25d) was the best fitting model, which included both regression pathways. The model explained 20.2% of the variance in Time 2 vigorous/moderate exercise and 18.2% of the variance of intentions. Intentions were significantly predicted by motivation (β = .426, p < .001, [.326, .526]), and Time 2 vigorous/moderate exercise was significantly predicted by intentions (β = .215, p < .001, [.119, .311]) and motivation (β = .314, p < .001, [.191, .437]). Motivation also exhibited a significant indirect relationship predicting Time 2 vigorous/moderate exercise via intentions (β = .092, p <.001, [.049, .135]) creating a total effect of motivation of β = .406 (p < .001, [.294, .518]). The pattern of results for the fit indices were mostly the same after including the other social-cognitive variables (Figure 17). The null model (Model 26a) had a fit of 𝜒 2 /df = 208.257 / SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 82 16, CFI = .796, TLI = .643, RMSEA =.130. Adding the intentions (Model 26b) and motivation pathways (Model 26c) significantly improved model fit compared to the null model (𝛥 𝜒 2 /Δdf= 43.311 / 1, p< .001; 𝛥 𝜒 2 /Δdf= 48.837/ 1, p< .001; respectively), and adding both pathways together (model 26d) improved model fit compared to the next best fitting model (𝛥 𝜒 2 /Δdf= 7.434 / 1, p = .006). However, removing the latent variable in this model (Model 26e) significantly improved model fit compared to the next best fitting model (𝛥 𝜒 2 /Δdf= 94.990 / 8, p< .001), which was again an unexpected finding. Like the case above, this improvement in model fit seemed to be a result of relationships between outcome expectations and the other variables in the model, as post-hoc analyses revealed that dropping outcome expectations from the model resulted in a worsening of fit after removing the latent variable (𝜒 2 /df= 177.911 / 6, CFI = .735, TLI = .071, RMSEA =.200; fit indices not reported in Table 32). Again, the a priori model (Model 26d) will be discussed in detail even though it technically did not provide the best fit to the data. Model 26d explained 23.4% of the variance in Time 2 vigorous/moderate exercise and 20.9% of the variance of intentions. Intentions were significantly predicted by motivation (β = .400, p = .004, [.257, .543]), but not outcome expectations (β = .098, p = .083, [-.011, .208]) or self-efficacy (β = .002, p = .987, [-.241, .245]). Time 2 vigorous/moderate exercise was significantly predicted by intentions (β = .221, p < .001, [.115, .327]), motivation (β = .589, p = .001, [.413, .765]), and outcome expectations (β = -.153, p = .030, [-.292, -.014]), which interestingly was a negative relationship, but not by self-efficacy (β = -.249, p = .108, [-.555, .057]). There was also a significant indirect effect of motivation predicting Time 2 vigorous/moderate exercise via intentions (β = .088, p = .002, [.059, .117]), creating a total effect for motivation of β = .677 (p < .001, [.503, .851]). SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 83 The path diagrams with parameter estimates predicting success in accomplishing intentions for vigorous/moderate exercise at Time 2 with only intentions and motivation are shown in Figure 18, and model fit indices are shown in Table 33. The null model (27a) exhibited a poor fit to the data with𝜒 2 /df = 61.402 / 10, CFI = .819, TLI = .728, and RMSEA =.085. Adding the intentions pathway (Model 27b) failed to improve model fit compared to the null model (𝛥 𝜒 2 /Δdf= 1.565 / 1, p = .211), while adding the motivation pathway (Model 27c) significantly improved model fit compared to the null model (𝛥 𝜒 2 /Δdf= 16.390 / 1, p< .001). Including both pathways (Model 27d) failed to improve model fit compared to the motivation pathway alone (𝛥 𝜒 2 /Δdf= 1.349 / 1, p= .246), and removing the latent variable (Model 27e) significantly and substantially worsened model fit compared to all other models (𝜒 2 /df = 177.893 / 6, CFI = .393, TLI = -.516, RMSEA =.200). Again, the a priori model (Model 27d) will be discussed in more detail even though it did not provide the best fit to the data. Model 27d explained 6.5% of the variance in success in accomplishing intentions for vigorous/moderate exercise at Time 2 and 19.1% of the variance of intentions. Intentions were significantly predicted by motivation (β = .437, p < .001, [.339, .536]), and success in accomplishing intentions for vigorous/moderate exercise at Time 2 was significantly predicted by motivation (β = .275, p < .001, [.140, .410]), but not by intentions (β = -.061, p = .248, [-.165, .043]). The indirect effect of motivation on success in accomplishing intentions for vigorous/moderate exercise was not significant (β = -.027, p = .289, [-.074, .020]). The pattern of results for the fit indices were mostly the same after including the other social-cognitive variables (Figure 19). The null model (Model 28a) had a model fit of 𝜒 2 /df = 160.297 / 16, CFI = .834, TLI = .710, RMSEA =.112. Adding the intentions pathway (Model 28b) failed to improve model fit compared to the null model (𝛥 𝜒 2 /Δdf= .277 / 1, p= .599), while SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 84 adding the motivation pathway (Model 28c) significantly improved model fit compared to the null model (𝛥 𝜒 2 /Δdf= 10.283/ 1, p< .001). Adding both pathways together (Model 28d) failed to improve model fit compared to the motivation pathway alone (𝛥 𝜒 2 /Δdf= .666 / 1, p = .414). However, removing the latent variable in this model (Model 28e) significantly improved model fit compared to the next best fitting model (𝛥 𝜒 2 /Δdf= 92.352 / 8, p< .001), which was again an unexpected finding. Same as above, this improvement in model fit seemed to be a result of relationships between outcome expectations and the other variables in the model, as post-hoc analyses revealed that dropping outcome expectations from the model resulted in a worsening of fit after removing the latent variable (𝜒 2 /df = 177.911 / 6, CFI = .735, TLI = .071, RMSEA =.200; fit indices not reported in Table 33). Again, the a priori model (Model 28d) will be discussed in more detail even though it did not provide the best fit to the data. Model 28d explained 11.7% of the variance in success in accomplishing intentions for vigorous/moderate exercise at Time 2 and 21.1% of the variance of intentions. Intentions were significantly predicted by motivation (β = .408, p = .004, [.265, .551]), but not outcome expectations (β = .097, p = .086, [-.013, .207]) or self-efficacy (β = -.005, p = .971, [-.250, .240]). Success in accomplishing intentions for vigorous/moderate exercise at Time 2 was significantly predicted by motivation (β = .579, p = 002, [.389, .769]) and outcome expectations (β = -.232, p = .002, [-.377, -.087]), which interestingly was again a negative relationship, but not by intentions (β = -.045, p = .427, [-.157, .067]) or self-efficacy (β = -.234, p = .155, [-.557, .089]). The indirect effect of motivation predicting success in accomplishing intentions for vigorous/moderate exercise at Time 2 was not significant (β = -.018, p = .507, [-.045, .009]). SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 85 Predicting Exercise at Time 3. The path diagrams with parameter estimates predicting Time 3 vigorous exercise with only intentions and motivation at Time 2 are shown in Figure 20, and model fit indices are shown in Table 34. The null model (Model 29a) had a 𝜒 2 /df = 194.553 / 10, CFI = .723, TLI = .475, RMSEA =.114. Adding the intentions pathway (Model 29b) significantly improved model fit compared to the null model (𝛥 𝜒 2 /Δdf= 125.964 / 1, p< .001), as did adding the motivation pathway (Model 29c; 𝛥 𝜒 2 /Δdf= 96.764 / 1, p< .001). Including both pathways together (Model 29d) significantly improved model fit compared to the next best fitting model (𝛥 𝜒 2 /Δdf= 13.962 / 1, p< .001), while removing the latent variable (Model 29e) significantly and substantially worsened model fit compared to all other models (𝜒 2 /df = 192.782/ 6, CFI = .705, TLI = -.770, RMSEA =.209). Therefore, the a priori model (Model 29d) was the best fitting model, which included both regression pathways. The model explained 34.0% of the variance in Time 3 vigorous exercise and 31.9% of the variance of intentions. Intentions were significantly predicted by motivation (β = .518, p < .001, [.426, .610]), Time 3 vigorous exercise was significantly predicted by Intentions (β = .399, p < .001, [.291, .507]) and motivation (β = .228, p < .001, [.108, .348]). There was also a significant indirect effect of motivation predicting Time 3 vigorous exercise via intentions (β = .207, p < .001, [.146, .268]), creating a total effect of motivation of .435 (p < .001, [.331, .539]). The pattern of results for the fit indices was mostly the same after including the other social-cognitive variables (Figure 21). The null model (Model 30a) had a model fit of 𝜒 2 /df = 183.897 / 16, CFI = .860, TLI = .693, RMSEA =.094. Adding the intentions (Model 30b) and motivation (Model 30c) pathways significantly improved model fit compared to the null model (𝛥 𝜒 2 /Δdf= 75.196/ 1, p< .001; 𝛥 𝜒 2 /Δdf= 59.790 / 1, p< .001; respectively), and adding both SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 86 pathways together (Model 30d) improved model fit compared to the next best fitting model (𝛥 𝜒 2 /Δdf= 4.329 / 1, p = .038). However, removing the latent variable in this model (Model 30e) significantly improved model fit (𝛥 𝜒 2 /Δdf= 45.518 / 8, p< .001), which was an unexpected finding. This improvement in model fit seemed to be a result of relationships between outcome expectations and the other variables in the model, as post-hoc analyses revealed that dropping outcome expectations from the model resulted in a worsening of fit after removing the latent variable (𝜒 2 /df = 192.790 / 8, CFI = .817, TLI = -0.369, RMSEA = .209). Again, the a priori model (Model 30d) will be explained in more detail even though it did not provide the best fit to the data. Model 30d explained 34.1% of the variance in Time 3 vigorous exercise and 32.6% of the variance of intentions. Intentions were significantly predicted by motivation (β = .464, p < .001, [.342, .586]), but not by outcome expectations (β = -0.049, p = .280, [-.137, .039]) or by self-efficacy (β = .088, p = .429, [-.130, .306]). Time 3 vigorous exercise was significantly predicted by intentions (β = .396, p< .001, [.290, .502]), motivation (β = .272, p = .042, [.133, .411]), but not by outcome expectations (β = .002, p = .971, [-.092, .096]) or self-efficacy (β = - .046, p = .679, [-.264, .172]). There was also a significant indirect effect of motivation predicting Time 3 vigorous exercise via intentions (β = .184, p < .001, [.135, .233]), creating a total effect for motivation of β = .456 (p = .001, [.315, .597]). The path diagrams with parameter estimates predicting Time 3 vigorous/moderate exercise with only intentions and motivation are shown in Figure 22, and model fit indices are shown in Table 35. The null model (31a) exhibited poor fit to the data with a 𝜒 2 /df = 152.418 / 10, CFI = .671, TLI = .506, and RMSEA =.141. Adding the intentions pathway (Model 31b) significantly improved model fit compared to the null model (𝛥 𝜒 2 /Δdf= 104.564 / 1, p< .001), as SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 87 did adding the motivation pathway (Model 31c; 𝛥 𝜒 2 /Δdf= 97.751/ 1, p< .001). Including both pathways together (Model 31d)significantly improved model fit compared to the next best fitting model (𝛥 𝜒 2 /Δdf= 18.542 / 1, p< .001), while removing the latent variable (Model 31e)significantly and substantially worsened model fit compared to all other models (𝜒 2 /df = 193.006 / 6, CFI = .567, TLI = -.081, RMSEA =.209). Therefore, the a priori model (Model 31d) was the best fitting model, which included both regression pathways. The model explained 28.7% of the variance in Time 3 vigorous/moderate exercise and 29.7% of the variance of intentions. Intentions were significantly predicted by motivation (β = .545, p < .001, [.449, .641]). Time 3 vigorous/moderate exercise was significantly predicted by intentions (β = .315, p < .001, [.201, .429]) and motivation (β = .294, p < .001, [.161, .427]). Motivation also exhibited a significant indirect relationship predicting Time 2 vigorous/moderate exercise via intentions (β = .172, p < .001, [.107, .237]) creating a total effect of motivation of β = .466 (p < .001, [.356, .576]). The pattern of results for the fit indices were similar after including the other social- cognitive variables (Figure 23). The null model (Model 32a) had a fit of 𝜒 2 /df = 149.515 / 16, CFI = .856, TLI = .749, RMSEA =.108. Adding the intentions (Model 32b) and motivation pathways (Model 32c) significantly improved model fit compared to the null model (𝛥 𝜒 2 /Δdf= 54.462 / 1, p< .001; 𝛥 𝜒 2 /Δdf= 42.197 / 1, p< .001; respectively). Adding both pathways together (Model 32d) also significantly improved model fit compared to the next best fitting model (𝛥 𝜒 2 /Δdf= 4.583 / 1, p = .032). Removing the latent variable in this model (Model 32e) significantly improved model fit compared to the next best fitting model (𝛥 𝜒 2 /Δdf= 40.277/ 8, p< .001), which was again an unexpected finding. Same as above, this improvement in model fit seemed to be a result of relationships between outcome expectations and the other variables in SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 88 the model, as post-hoc analyses revealed that dropping outcome expectations from the model resulted in a worsening of fit after removing the latent variable (𝜒 2 /df = 193.009 / 6, CFI = .775, TLI = .212, RMSEA =.209). The a priori model (Model 32d) will be discussed in more detail even though it did not provide the best fit to the data. Model 32d explained 28.2% of the variance in Time 3 vigorous/moderate exercise and 29.4% of the variance of intentions. Intentions were significantly predicted by motivation (β = .407, p < .001, [..285, .529]), but not outcome expectations (β = .038, p = .393, [-.048, .124]) or self-efficacy (β = .138, p = .228, [-.085, .361]). Time 3 vigorous/moderate exercise was significantly predicted by intentions (β = .334, p <.001, [.224, .444]) and by motivation (β = .335, p = .020, [.190, .480]), but was not significantly predicted by outcome expectations (β = .032, p = .515, [-.064, .128]), or self-efficacy (β = -.091, p = .464, [-.334, .152]). However, there was also a significant indirect effect of motivation predicting Time 3 vigorous/moderate exercise via intentions (β = .136, p = .001, [.095, .177]), creating a total effect for motivation of β = .471 (p = .001, [.326, .616]). The path diagrams with parameter estimates predicting success in accomplishing intentions for vigorous/moderate exercise at Time 3 with only intentions and motivation are shown in Figure 24, and model fit indices are shown in Table 36. The null model (33a) exhibited a poor fit to the data with 𝜒 2 /df = 50.782/ 10, CFI = .877, TLI = .816, and RMSEA =.076. Adding the intentions pathway (Model 33b) improved model fit compared to the null model (𝛥 𝜒 2 /Δdf= 4.527 / 1, p = .033) and adding the motivation pathway (Model 33c) also significantly improve model fit compared to the null model (𝛥 𝜒 2 /Δdf= 22.489 / 1, p< .001). Including both pathways together (Model 33d) failed to improve model fit compared to the next best fitting model (𝛥 𝜒 2 /Δdf= 1.117 / 1, p= .291). Finally, removing the latent variable (Model 33e) SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 89 significantly and substantially worsened model fit compared to all other models (𝜒 2 /df = 194.855 / 6, CFI = .431, TLI = -.422, RMSEA =.210). The a priori model (Model 33d) is discussed in more detail even though it did not provide the best fit to the data. Model 33d, which included both pathways in the model, explained 9.9% of the variance in success in accomplishing intentions for vigorous/moderate exercise at Time 3, and it explained 30.0% of the variance in intentions. Intentions were significantly predicted by motivation (β = .548, p < .001, [.452, .644]). Success in accomplishing intentions for vigorous/moderate exercise at Time 3 was significantly predicted by motivation (β = .351, p < .001, [.278, .616]), but not by intentions (β = -.078, p = .352, [-.208, .074]). The indirect effect of motivation on success in accomplishing intentions for vigorous/moderate exercise via intentions was also not significant (β = -.036, p = .280, [-.116, .044]). The pattern of results for the fit indices were mostly the same after including the other social-cognitive variables (Figure 25). The null model (Model 34a) had a model fit of 𝜒 2 /df = 94.177 / 16, CFI = .906, TLI = .835, RMSEA =.083. Adding the intentions pathway (Model 34b) failed to improve model fit compared to the null model (𝛥 𝜒 2 /Δdf = .005 / 1, p= .944), while the motivation pathway significantly improved model fit compared to the null model (Model 34c; 𝛥 𝜒 2 /Δdf 6.731 / 1, p = .010). Adding both pathways together (Model 34d) failed to improve model fit compared to the best fitting model (𝛥 𝜒 2 /Δdf= .876 / 1, p = .349). Finally, removing the latent variable in this model (Model 34e) significantly improved model fit compared to the next best fitting model (𝛥 𝜒 2 /Δdf= 36.898 / 8, p< .001), which was again an unexpected finding. Same as above, this improvement in model fit seemed to be a result of relationships between outcome expectations and the other variables in the model, as post-hoc analyses revealed that dropping outcome expectations from the model resulted in a worsening of fit after removing the latent SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 90 variable (𝜒 2 /df = 194.857 / 6, CFI = .741, TLI = .095, RMSEA =.210). Again, the a priori model (Model 34d) is discussed in more detail even though it did not provide the best fit to the data. Model 34d explained 10.0% of the variance in success in accomplishing intentions for vigorous/moderate exercise at Time 3 and 30.2% of the variance of intentions. Intentions were significantly predicted by motivation (β = .413, p < .001, [.293, .533]), but not outcome expectations (β = .034, p = .441, [-.052, .120]) or self-efficacy (β = .141, p = .210, [-.079, .361]). Success in accomplishing intentions for vigorous/moderate exercise at Time 3 were significantly predicted by motivation (β = .447, p = .007, [.278, .616]), but not by intentions (β = -.067, p = .352, [-.208, .074]), outcome expectations (β = .010, p = .870, [-.106, .126]), or self-efficacy (β = -.133, p = .358, [-.415, .149]). The indirect effect of motivation predicting success in accomplishing intentions for vigorous/moderate exercise at Time 3 via intentions was also not significant (β = -.028, p = .372, [-.059, .003]). SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 91 Discussion Intervention outcomes Significant differences at Time 1 were observed in all of the goal-structure scores except for the overall rank of the health goal-cluster. The Intervention Group reported higher levels of facilitation with exercise, higher levels of conflict with not exercising, lower levels of conflict with exercise, and lower levels of facilitation with not exercising compared to the Control Group. These findings indicate that the intervention successfully affected the most proximally targeted cognitions. However, these effects failed to spill over to the other social-cognitive and motivation variables in the study. The latent change score models indicated that most variables changed significantly over time, although relatively few differences were observed across groups. Exercise changes over time. The results testing for intervention effects on the exercise variables were fairly consistent for vigorous and vigorous/moderate exercise. A dual change score model improved the fit over the null, proportional change, and linear change models, indicating that exercise changed over time. Additionally, both variables showed a large positive linear change combined with a small negative proportional change, indicating a positive increase in exercise that eventually asymptotes over time (but well outside the range of the data). There was no evidence of differences in exercise across groups. Behavioral intention changes over time. Both behavioral intentions variables (for vigorous and vigorous/moderate exercise), exhibited the same pattern of change over time. The best fitting model was the dual change score model with a large positive slope and a small negative proportional change. Similar to exercise, these parameters indicate an increase in SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 92 behavioral intentions over time that eventually asymptotes well outside the bounds of the data, and there were no differences across groups. Motivation variables changes over time. The three motivation variables—commitment, autonomous motivation, and controlled motivation—exhibited differing patterns of results. Both commitment and autonomous motivation were best fit by a dual change model holding all parameters invariant across groups. Both variables exhibited a larger positive slope combined with a smaller negative proportional change, indicating increases over time that eventually asymptote outside the bounds of the data. Controlled motivation, on the other hand, was best fit by a linear change model with a negative slope indicating a slight decrease in controlled motivation over time. However, the slope estimate itself was not significant, indicating that this variable changed very little over time, even though the linear-change model significantly improved fit compared to the null model. These findings are interesting, as controlled motivation is often considered a “lesser quality” of motivation compared to autonomous motivation, in that it is fleeting over time (Deci & Ryan, 2004, 2008; Sheldon & Elliot, 1999). Social-cognitive variable changes over time. The social cognitive variables—self- efficacy, outcome expectations, and planning—had a diverse set of findings. Planning was the only variable that was fitted best by a dual change score model with a positive slope and negative proportional change and all parameters invariant across groups. For self-efficacy, the linear change model with all parameters held invariant across groups was the best fitting model, and this model exhibited a negative slope indicating a slight decrease over time. However, the slope estimate by itself was not significant, indicating that this variable changed very little over time. Outcome expectations was one of the few variables for which no model significantly improved fit compared to the null model, indicating that outcome expectations did not change SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 93 over time. However, due to complications estimating parameters (negative values for slope variances), the slope variance and slope-intercept covariance were restrained to zero in order to run the models without error. The last two models (freeing the slope-intercept covariance and the slope and intercept variances across groups) were unable to be tested. It is possible that in a different sample in which these models could be estimated correctly significant changes would be observed. Another possibility is that the measure used was inappropriate or ineffective, as it was an abridged version of a previously developed scale that was adapted for exercise behavior (Wójcicki et al., 2009). Outcome expectations is often an overlooked construct in the social- cognitive literature, thus there is a limited amount of research related to measuring this construct (Larsen, et al., 2015; Rhodes & Nigg, 2011). Goal-structure score changes over time. The seven goal-structure scores also exhibited a diverse pattern of findings. The conflict/facilitation with not exercising composite score, facilitation with exercise, conflict with exercise, and conflict with not exercising all had some version of the dual change score model as the best fitting model. Conflict with not exercising and conflict with exercise both had a dual change score model with all parameters held invariant across groups as the best fitting model. Both of these models also exhibited positive slopes and negative proportional change parameters, indicating growth over time that eventually asymptotes. However, these parameters were relatively small compared to some of the other variables in the study indicating a small degree of changing over time (improvements in chi- square of ~2.75 per degree of freedom, compared to improvements in chi-square of ~17.25 per degree of freedom for the exercise variables, for example). The best fitting model for facilitation with exercise was the dual change score model with the slope freely estimated across groups, indicating a significant difference in change between the Intervention and Control groups. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 94 Specifically, the Control group saw a slightly larger increase in facilitation with exercise over time compared to the Intervention group (49.564 compared to 46.061, respectively). The best fitting model for facilitation with not exercising was the proportional change model with all parameters invariant across groups. This model showed a slight decrease in facilitation with not exercising over time (which was the expected direction in relation to an increase in exercise). Two of the goal structure variables—the overall rank of the health goal-cluster and conflict/facilitation with exercise composite—did not change over time. The rank of the heath goal may have suffered from a lack of variability (compared to the other variables in the study) as this variable ranged from 1-9. The conflict/facilitation with exercise variable may not have changed over time due to the two underlying variables comprising the composite score (facilitation with exercise and conflict with exercise) both increasing over time, thus partially canceling out the effects of each variable. This is a potential problem when measuring conflict and facilitation on the same scale (i.e., using a bimodal response option), as doing so implies that those variables are opposites of the same construct. However, the fact that conflict with exercise and facilitation with exercise both increased doesn’t support this notion, but rather that the two variables are distinct constructs (Presseau et al., 2011, 2013; Riediger, 2007). If this is the case, two unimodal scales, rather than one bimodal scale, should be used when measuring these constructs. This concept is addressed further in the next section discussing the results of the goal- structure analysis. Changes in future exercise goals. Finally, the best fitting model for future vigorous exercise goals was a dual change model with all parameters held invariant across groups and a slightly decreasing slope combined with a positive proportional change estimate, indicating a curve that dips at first and then starts to increase by the third time point of measurement. The SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 95 best fitting model for future vigorous/moderate exercise goals was a proportional change model with a positive proportional change element. Therefore, both variables measuring future exercise goals tended to increase, proportionally for vigorous/moderate exercise, and following the second time point for vigorous exercise. Considering the status quo for most New Year’s resolution attempts is failure (Koestner et al., 2006, 2002; Polivy & Herman, 2002), one might expect decreases of future exercise goals over time, as one might expect people would have their highest goals (i.e. most desirable behavioral expectations) at the beginning of their attempt. It could be possible that: (a) the study motivated participants (of both groups, which is explained in more detail below) to stick to their original plans for the amount of exercise they planned to do in the future; (b) the study was too short to pick up change in this variable, and a longer study might see decreases of future exercise goals over time as people began quitting their New Year’s Resolutions; or (c) people who actually decreased their future exercise goals dropped out of the study, and therefore were not accounted for. The findings in relation to these variables should be interpreted with caution, as the measure was novel and therefore not previously validated, and the variables themselves exhibited a large degree of positive skew, which required the variables to be standardized in the models in order for the models to run without error. Summary of intervention effects. The intervention successfully impacted the goal- structure scores at Time 1 (except the overall rank of the health goal-cluster), and the majority of the study variables showed some amount of change over time. However, there were few differences in changes over time across groups. Only outcome expectations, the overall rank of the health goal-cluster, and conflict/facilitation with exercise remained constant over time (indicated by no model improving fit compared to the null model). The most common best fitting model was the dual change score model with a positive slope and a negative proportional change, SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 96 in which the slope was substantially larger than the proportional change. This pattern of findings indicates a curve that increases overtime and eventually asymptotes. However, the asymptote generally occurred well outside the bounds of the data (indicated by the positive slope being substantially larger than the negative proportional change), so caution should be taken when generalizing these results beyond the time frame of the present study. From a practical standpoint, this kind of pattern of results indicates that growth occurred faster between the first and second time points, and then tapered off between the second and third time points. The variables that fit this pattern of change were: vigorous and vigorous/moderate exercise, behavioral intentions for vigorous and vigorous/moderate exercise; planning; commitment; autonomous motivation; conflict/facilitation with not exercising; facilitation with exercise; conflict with exercise; and conflict with not exercising. The dual change model described here might be what one would expect for changes to the variables in the present context. These behaviors/cognitions would not be expected to increase indefinitely, if one were to expect these variables to increase at all. It would be impossible to increase these behaviors continuously, so any initial increase would at some point be followed by a slow-down in growth or a decrease. This is in contrast to contexts in which people can increase or decrease indefinitely over the window of measurement—such as learning new skills, child cognitive development, or even weight gain—whereas there are clear limits to how much exercise people can do. Several variables changed in the opposite direction expected. Both self-efficacy and controlled motivation decreased slightly, as indicated by a linear change model with a negative slope exhibiting the best fit to the data. However, the slope estimate itself was insignificant for both variables despite the improvement in model fit (p > .05). Therefore, it seems that there was SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 97 more likely a lack of change rather than a decrease in these two variables over time. Additionally, there was a significant increase in conflict with exercise as indicated by a dual change score model with a positive slope and negative proportional change exhibiting the best fit to the data. However, this was a small change in model fit as well. There were relatively few differences in changes over time observed across groups. There was a significant difference observed for perceived facilitation of exercise in the slope parameter of the dual change score model. The Control group had a larger linear increase (49.564) compared to the Intervention Group (46.061), indicating the Control group increased in their perceptions of the degree that exercise facilitated their other life goals more than the Intervention group. However, the Intervention Group actually exhibited slightly higher mean perceived facilitation in exercise throughout the study due to higher scores at Time 1. Also, the conflict/facilitation with not exercising composite score showed group differences in the intercept-slope covariance parameter across groups, in that the Intervention had a stronger positive covariance (878.242) compared to the Control group (604.523), indicating there was a stronger relationship between Intervention Group participants’ baseline scores and linear change over time compared to the Control Group. This could indicate the intervention had an effect on participants holding higher perceptions of facilitation with not exercising at baseline. The lack of group differences observed in exercise following the intervention were disappointing. However, the intervention was minimalist, taking a median time of 7.5 minutes, and was done only once at the beginning of the study. Additionally, the approach—intervening on the degree participants perceived their exercise plans would conflict with or facilitate them in achieving their most important life goals—had not been previously attempted (Presseau et al., 2013), and the intervention components themselves were newly created for this study. It is SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 98 possible that the intervention was either simply ineffective, or too weak to create change in the variables that were not directly targeted. However, there are also several more optimistic explanations for the lack of group differences in changes over time. For instance, it could be possible that these types of perceptions are less amenable to alteration at this point in the behavior change process. That is, increasing the degree participants perceive exercise will facilitate them in achieving their most important life goals, and decreasing the degree participants perceive exercise will conflict with achieving their most important life goals, may not have an effect on people who have already decided to change their behavior (Arden & Armitage, 2008; Armitage, 2006, 2009; Prochaska & DiClemente, 1984). In other words, people who have decided to increase their exercise behavior may already exhibit relatively high levels of motivation. This kind of informational/motivational intervention may be more effective for people at the preparation stage of behavior change rather than the action stage (Armitage, 2009; Armitage, Sheeran, Conner, & Arden, 2004; Prochaska & DiClemente, 1984). There was also a planning element of the study, making it more than merely an informational/motivation intervention. Planning interventions have been shown to positively affect behavior change outcomes (Gollwitzer, 1993; Gollwitzer & Sheeran, 2006; Schwarzer, 1992; Wiedemann, Lippke, Reuter, Ziegelmann, & Schüz, 2011). However, other studies have failed to show significant effects of planning interventions as well (Koestner et al., 2002). For instance, Koestner et al., (2002), found no effect of implementation intentions on the degree participants’ succeeded in pursuing their New Year’s Resolutions. It’s possible that participants in the Control Group had created plans for implementing their exercise on their own, which would nullify much of the effect of having Intervention Group participants complete a planning exercise. For instance, Gollwitzer (1999) reports that people spontaneously make these kinds of SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 99 plans 67% of the time, and people beginning New Year’s Resolutions may do so at an even higher rate because they are aware the odds of success are stacked against them (Koestner et al., 2002). There is some evidence to support this notion in the present study, as no differences in planning were observed across groups at Time 1 even though the Intervention Group partook in the planning exercise. Successful New Year’s Resolution attempts are often considered improbable, particularly for weight-related resolutions (Koestner et al., 2006, 2002; Polivy & Herman, 2002). It is odd, then, that both the Intervention and Control Groups significantly increased their exercise across the two-month study, which would indicate that both groups were on average successful in increasing their exercise during that time period. In comparison, Koestner et al., (2006) found that participants pursuing New Year’s Resolutions rated themselves on average 2.57 out of 9 on the degree they were making progress towards achieving their goals six months after starting. While the current study is only two months long, which may have been too short of a window to observe differences in groups and large group failures in exercising, other studies have observed failure in shorter periods of time. In a similar study evaluating participants’ perceived goal progress, Koestner et al., (2002) observed that participants rated themselves on average only 3.79 out of 9 after only one month. Also, a community-based study observed large-scale failure of New Year’s Resolutions after only one month (Norcross et al., 1989). While the literature on New Year’s Resolution success is relatively sparse, the results of the present study seem to go against the typical finding of failure, even after only two months of measurement. A potential explanation for finding improvements in frequency of exercise for both study groups was that having the Control Group complete the goal-structure measure—which the intervention was based off of—was enough to spark the intended effects of the intervention. That SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 100 is, participants in the Control Group still ranked the nine goal-clusters from most to least important to them and gauged the degree exercising or not exercising at all would conflict with or facilitate them achieving each of the goal-clusters. Therefore, the Control group may have effectively partook in a portion of the Intervention unintentionally. Simply measuring cognitions has been shown to have an effect on the cognitions that are measured. This is referred to as the question-behavior effect—also known as the mere measurement effect, the self-prophecy effect, or self-generated validity— and is a well-known phenomenon in which questioning people about their attitudes or perceptions about a behavior increases the likelihood that they do that behavior because those perceptions become more salient (Dholakia, 2010; Sprott et al., 2006; Wood, Conner, Sandberg, Godin, & Sheeran, 2014). The question behavior effect has been observed for a variety of weight related behaviors, including diet (Levav & Fitzsimons, 2006; Williams, Fitzsimons, & Block, 2004), and exercise (Godin, Bélanger-Gravel, Amireault, Vohl, & Pérusse, 2011; Sandberg & Conner, 2011; Spence, Burgess, Rodgers, & Murray, 2009). Future studies could test this hypothesis by recruiting a third study group that would completely skip the goal- structure portion of the study. The present study failed to do this because (a) the possibility that measuring goal-structure could effectively mimic the intervention wasn’t considered, and (b) the other goals of the study—evaluating the Chulef, Read, and Walsh goal taxonomy for measuring goal-structure for exercise, and using goal-structure scores to form a latent variable of motivation for predicting exercise—required large sample sizes. Therefore, goal-structure was measured for all participants in order to have adequate power for the complex models used to test these other hypotheses. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 101 Goal-structure outcomes The analysis of the Chulef, Read, and Walsh (2001) goal taxonomy tested the degree the seven goal-structure scores predicted exercise variables and the degree the goal-structure scores were associated with the other motivation constructs. The goal-structure scores were: (1) the overall rank of the health goal-cluster compared to the other 8 goal-clusters; (2) aggregated conflict/facilitation with exercising; (3) aggregated conflict/facilitation with not exercising; (4) conflict with exercise; (5) facilitation with exercise; (6) conflict with not exercising; and (7) facilitation with not exercising. Analyses indicated that the scores for facilitation with exercise and the overall rank of the health goal-cluster are potentially useful additions to a latent variable measuring motivation that can be added to common social-cognitive models of self-regulation. Analyses revealed that those two goal-structure scores were: (1) significantly correlated with several of the exercise dependent variables; (2) consistently factored with the other motivation constructs in exploratory factor analysis; and (3) the goal facilitation with exercise score was the most common predictor of exercise in multiple regression analyses. The above analyses were conducted for each time point of data (i.e., Time 1, Time 2, and Time 3), and on five separate samples of data (three random samples of 50% of the data, a complete-cases sample, and the total sample). Analyses concluded that the aggregated conflict/facilitation with exercise and conflict/facilitation with not exercising scores were inadequate. The conflict/facilitation with not exercising score failed to robustly correlate with any of the exercise variables. The conflict/facilitation with exercise score robustly correlated with many exercise variables, however it held the same number of relationships as the facilitation with exercise score, but the aggregate score had weaker relationships over all. Therefore, the un-aggregated scores appeared SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 102 to be better predictors of exercise behavior. These results indicate that goal conflict and goal facilitation may be unique constructs, as opposed to opposites on a bipolar scale. Both methods for measuring goal conflict and facilitation have been used in past research (Riediger, 2007). It is becoming more common to conceptualize goal conflict and facilitation as different entities, and therefore to utilize separate scores for conflict and facilitation. The results of the present work support the use of separate scores for the two constructs. Using un-aggregated scores for conflict and facilitation have implications for using the Chulef, Read, and Walsh (2001) goal taxonomy in research. Traditionally, this goal taxonomy has utilized a bipolar scale for measuring the degree that a behavior (e.g., exercising) influences (i.e., conflicts with or facilitates) achieving the goal-clusters. One benefit of a bipolar scale is that it requires half as many questions as two unipolar scales, which can make a substantial difference when measuring multiple constructs repeatedly over time. However, if goal conflict and goal facilitation are qualitatively different constructs—which is supported by recent research (Presseau et al., 2011, 2013; Riediger, 2007), as well as the present work—then a unipolar scale would be a conceptually more appropriate way for measuring these constructs. A bipolar scale for goal conflict/facilitation creates several conceptual problems (Riediger, 2007; Riediger & Freund, 2004). A score of zero on a bipolar scale is ambiguous. A score of zero could mean that perceived conflict and facilitation with a behavior are exactly equal, or it could mean that the behavior in question is completely unrelated to that particular goal-cluster. A bipolar scale also implies that a behavior cannot conflict with and facilitate achieving a goal simultaneously. For example, a person may perceive exercise as facilitating the goal-cluster having financial and occupational success in that exercise may help them think clearly at work, give them more energy throughout the day, or help them avoid missing work due SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 103 to illness. However, that same person may also perceive some degree of conflict between exercise and financial and occupational success due to time limitations. A bipolar scale requires mental energy to contrast the degree a behavior conflicts with and facilitates a goal-cluster, which could be mentally straining for participants and could lead to inaccuracies. This could be particularly problematic if there are biases in how people mentally compare conflict and facilitation, as goal facilitation may be more salient than goal conflict in certain situations (Presseau et al., 2011, 2013). Therefore, when measuring goal-structure via conflict/facilitation scores, researchers should weigh the pros and cons of minimizing the number of items in a survey by using a bipolar scale or using a conceptually more accurate unipolar scale. Additionally, it may be time to experiment with unipolar scales for measuring conflict and facilitation when using the Chulef, Read, and Walsh (2001) goal taxonomy. The overall ranking of the health goal was determined to be a potentially useful addition to a latent variable measuring motivation. However, the overall rank should be tested in a population that is not initiating an attempt to increase their exercise. The rank of the health goal in a population that has recently decided to increase their exercise may be biased towards greater importance than a more representative sample of people. It may also be likely to fluctuate if the person just recently started considering that goal to be more important to them. That is, if the health goal recently shifted upwards in importance, it may soon shift back downwards. The overall rank of the health goal-cluster may be a stronger predictor in a sample that resembles the general population. However, causality of the overall rank is still an important question, and without established causality it may be problematic to use these associations as arguments for designing interventions attempting to alter a person’s importance rankings. Future interventions could explore this relationship by emphasizing the importance of prioritizing a new goal (such as SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 104 exercise) over existing goals people feel are also important, as opposed to simply fitting the new goal into a person’s already established schedule of resource allocation. In addition to the present study being the first to test the Chuleff, Read, and Walsh, (2001) goal taxonomy to measure goal-structure for completing exercise, this study is perhaps the first to explore how conflict and facilitation with not exercising influences exercise behavior. However, the results were not particularly informative. The two scores, conflict with not exercising and facilitation with not exercising, failed to correlate with many exercise variables and consistently factored into their own group of constructs, along with conflict with exercise, during the exploratory factor analyses. However, as the present work is one of the few/only studies testing conflict/facilitation with not exercising, it may be premature to dismiss these constructs altogether. Future studies should continue testing the usefulness of these perceptions, and they should utilize a unipolar scale for potentially better measures of conflict/facilitation with not exercising. Another finding to note was that the significant relationships between facilitation with not exercising and exercise behavior were positive, which is the opposite direction one might expect. However, similar findings (i.e., opposite relationships) have been observed between conflict with exercising and exercise behavior (Presseau et al., 2013). These counterintuitive results have been explained to be a result of highly ambitious people who are more likely to pursue and accomplish many goals. The more goals someone strives for, the more those goals will likely conflict with one another. Studies in which participants are asked to list the number of goals they are currently pursuing (as opposed to using a pre-constructed taxonomy, like the present study) can control for the number of goals people are currently striving for (Presseau et al., 2013). SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 105 Conflict with exercise was another construct that failed to predict most of the exercise variables. It also typically factored with the other conflict/facilitation scores instead of the motivation constructs in exploratory factor analyses. These findings may seem odd in comparison to some theories and empirical evidence. For example, Image Theory (Beach & Mitchell, 1987; Beach, 1998) posits that goal conflict is a more powerful motivator than facilitation. Research using the Chuleff, Read, and Walsh (2001) goal taxonomy has found that conflict with a behavior significantly predicts decisions to retire (Brougham & Walsh, 2005, 2007) and BMI (Lee et, al., in preparation). Additionally, conceptually similar constructs to goal conflict, such as stressors and daily hassles, have also been found to significantly predict behavior (O’Connor, Jones, Conner, McMillan, & Ferguson, 2008; Presseau et al., 2013). However, empirical evidence continues to indicate that perceptions of facilitation with exercise tend to be stronger predictors of exercise behavior than perceptions of conflict with exercise (Li, et al., 2008; Presseau, Sniehotta, Francis, & Gebhardt, 2010; Presseau et al., 2013). Despite these findings, conflict with exercise may still be an important motivational construct. Presseau et al., (2013) observed that only perceived facilitation predicted exercise behavior when using perceptions of goal conflict/facilitation with exercise. However, when objectively measuring the degree exercise conflicted with or facilitated other important goals via daily diaries (as opposed to measuring perceptions alone), only conflict predicted exercise behavior. To add to this confusion, once objective measures of conflict were added to the statistical model, perceptions of facilitation with exercise were no longer a significant predictor. Whether goal conflict or facilitation predict behavior may depend on how salient these perceptions are in the context under study. Goal conflict may predict behavior more strongly in ‘simpler’ contexts. For example, Presseau et al., (2011) observed that perceptions of goal conflict SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 106 and goal facilitation predicted the likelihood physicians gave patients medical advice about exercise. Physicians are frequently encouraged to give patients exercise advice, but time restraints make it difficult for physicians to do so. Presseau et al., (2011) hypothesized that perceptions of goal conflict were likely more salient for the physicians because the behavior in question involved a single context (work) and primarily a single competing resource (time constraints). It appears that perceptions of facilitation may be more salient than perceptions of conflict in certain contexts, which leads perceptions of goal facilitation to be better predictors of some behaviors such as exercise. However, actual goal-conflict—for example, operationalized as time spent pursuing conflicting goals—may be more influential of a person’s behavior. Perceptions of goal conflict may also be more salient and therefore more powerful predictors of behavior in less complex environments (e.g., physician behavior at work dealing with time constraints). Clearly, more research is needed to fully understand how conflicting and facilitating goals influence behavior, and ultimately in order to design interventions attempting to manipulate these constructs. Motivation latent variable measurement invariance outcomes The goal-structure analysis revealed that the overall rank of the health goal-cluster and facilitation with exercise were potentially useful additions to a latent variable model measuring motivation, as these two constructs consistently predicted exercise and tended to factor with other motivation constructs in exploratory factor analysis. Measurement invariance testing was used to determine if the rank of the health goal-cluster, facilitation with exercise, autonomous motivation, and commitment formed a robust latent variable that remained stable over time and across sub-groups of the overall sample. These models systematically constrained and released SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 107 certain parameters and tested changes in model fit to determine if measurement invariance was violated. The “weak” invariance model holding loadings invariant across time significantly improved model fit compared to a Rasch model holding all loadings equal to one, indicating that a latent variable was a better fit to the data than a summed composite score. The “weak,” “strong,” and “strict” invariance models all yielded good fit to the data. However, there was a significant worsening of fit when comparing the strong invariance model to the weak invariance model, indicating a potential violation of scalar invariance over time. Multi-group analyses testing for measurement invariance across gender and study groups indicated that the violation of scalar invariance was due to differences in observed variable means across Intervention and Control groups of the two constructs facilitation with exercise and autonomous motivation. Therefore, there may have been an aspect of the intervention that caused the violation of scalar invariance. While there were no observed differences in autonomous motivation across groups following the intervention, the Intervention group reported significantly higher scores for goal facilitation with exercise at Time 1 which may have contributed to the scalar invariance of this variable. Tests of scalar invariance evaluate the degree the observed variable means are equal (Meredith, 1993). A lack of latent variable scalar invariance makes mean differences difficult to interpret because any differences observed may be due entirely to different starting values. Take for example, a latent variable of depression that includes an item measuring the frequency people cry. It is likely that women cry more frequently than men whether or not they are depressed (i.e., women would have a higher mean score on this item regardless of depression). If this violation of scalar invariance isn’t corrected for, women would consistently appear to be more depressed SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 108 than men due to pre-existing differences on the observed variable, even if their levels of depression are in fact equivalent. In essence, two different latent variables are being compared because the underlying relationships are not invariant across the two groups. Because all models yielded good fit to the data, despite the worsening of fit of the strong invariance model, it was concluded that the latent variable was suitable for use in the later analyses evaluating the utility of the latent variable of motivation. Furthermore, tests of mean differences on the latent variable were not conducted, and therefore, were not an issue. However, the factor structure of the motivation latent variable should continue to be tested in future studies. Perceptions of facilitation with exercise exhibited the lowest factor loading and also contributed to violations of scalar invariance. Removing this score from the latent variable may improve model fit. However, it is possible that a more appropriate measure of facilitation with exercise would satisfy scalar invariance requirements. It may also be that a new sample would do so as well without altering any measures, as the changes in model fit were relatively minor, albeit significant. Predictability of motivation latent variable The motivation latent variable was observed to significantly predict exercise over and above intentions for most of the models. It also exhibited larger direct associations with exercise than intentions in some models, and even exhibited significant relationships with exercise when intentions failed to do so. Both intentions and motivation significantly improved model fit in most of the models, including Time 2 vigorous exercise, Time 3 vigorous exercise, Time 2 vigorous/moderate exercise, and Time 3 vigorous/moderate exercise. However, intentions failed to improve model fit when predicting success in accomplishing intentions for vigorous/moderate exercise at Time 2 and at Time 3 (while motivation improved model fit for both variables). SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 109 The models exhibited small-medium effect sizes for predicting Time 2 and Time 3 exercise (r-squares ranging from .169 – .234, and .282 – .341, respectively). The models exhibited small effects for predicting success in accomplishing intentions for vigorous/moderate exercise at Time 2 and Time 3 (r-squares ranging from.065 – .117, and .099 - .100, respectively; Cohen, 1977). Motivation typically exhibited similar, if not larger, relationships with exercise in comparison to intentions. The overall size of the relationship between motivation and exercise was frequently boosted slightly after adding in the other social cognitive covariates. Vigorous and vigorous/moderate exercise at Time 2 were predicted by intentions exhibiting beta coefficients ranging from .215-.295 and by motivation with beta coefficients ranging from .204- .589. Vigorous and vigorous/moderate exercise at Time 3 were predicted by intentions exhibiting beta coefficients ranging from .315-.339 and by motivation with beta coefficients ranging from .228-.335. Success in accomplishing intentions for vigorous/moderate exercise at Time 2 and Time 3 was only predicted by motivation with beta coefficients ranging from .275-.579 and .351 - .447, respectively. Additionally, motivation consistently exhibited a significant indirect relationship with exercise via intentions, and therefore, had a total effect that was consistently larger than intentions. The level of predictability exhibited by the motivation latent variable was impressive. To put it in perspective, there are a wide variety of other social-cognitive predictors that were left out of the present study because they fail to predict behavior consistently (e.g., risk perceptions, subjective norms, perceived severity; Conner & Norman, 2005). The motivation latent variable fully mediated the relationships between self-efficacy and outcome expectations, which are considered to be the most consistent predictors of behavior in all of the social-cognitive models SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 110 of self-regulation (Conner & Norman, 2005; Plotnikoff et al., 2010, 2008, Schwarzer et al., 2008). It also exhibited similarly strong relationships with exercise as did behavioral intentions, and behavioral intentions are believed to be the most proximal cognition to behavior (Ajzen & Fishbein, 1980; Bandura, 2004; Fishbein et al., 2001; Fishbein & Ajzen, 1975). For the most part, outcomes were the same between models that evaluated the predictability of intentions and motivation in isolation and models that included self-efficacy and outcome expectations as covariates. However, an unexpected and consistent finding after adding the social-cognitive covariates was that model fit improved after removing the latent variable from the model and allowing the indicator variables to be associated with all other variables individually. In contrast, when the social cognitive variables were left out of the model, removing the latent variable consistently and substantially worsened model fit. This result seemed to be caused by relationships with outcome expectations, as post hoc analyses removing outcome expectations from the model resulted in a worsening of model fit upon removing the latent variable. Outcome expectations consistently showed a significant relationship ~.259 with the motivation latent variable, but was only significantly associated with facilitation for exercise and autonomous motivation, not goal commitment or the overall rank of the health goal. There were some other interesting results after including self-efficacy and outcome expectations, particularly that motivation altered the relationships of self-efficacy and outcome expectations in some interesting ways. For instance, self-efficacy typically exhibited a significant positive relationship with the exercise variables in the models that did not include a pathway between exercise and the motivation latent variable. However, after the motivation pathway was included, the remaining relationship between exercise and self-efficacy was negative in all but one model. A similar finding occurred for outcome expectations as well, although outcome SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 111 expectations rarely significantly predicted exercise in the models. Therefore, not only did motivation fully mediate the relationship between the other social cognitive variables and exercise, the residual variances left over were negatively associated with exercise. These patterns of results resemble Simpson’s paradox (Kievit, Frankenhuis, Waldorp, & Borsboom, 2013; Pearl, 2014), or an inconsistent mediation (MacKinnon, Fairchild, & Fritz, 2007), in which the direction of an association switches (e.g., goes from positive to negative) after including a mediator variable. These results may help explain mixed findings in the literature in relation to self-efficacy and outcome expectations (Teixeira, Going, Sardinha, & Lohman, 2005). There are known situations where self-efficacy and outcome expectations are inaccurate, for instance when a person is over-confident in their abilities or hold unrealistically high expectations (Bandura, 1997; Foster, Wadden, Vogt, & Brewer, 1997; Larsen, Koritzky, &Walsh, under review). Perhaps the motivation variables are mediating the positive relationships between self-efficacy, outcome expectations, and exercise, while underlying negative relationships between these variables are leftover. In the present models, the method for measuring goal facilitation may have been problematic, as previously indicated in the discussion of the goal-structure evaluation. Goal facilitation had the weakest factor loading of the four motivation indicator variables, and contributed to violations of scalar invariance. Future research should use a more appropriate measure of goal facilitation to determine if it is useful in the model. It was also consistently associated with outcome expectations, which in turn was consistently associated with the latent variable. These findings could indicate that outcome expectations and goal facilitation may be similar constructs. This would make sense conceptually, as both variables are essentially SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 112 anticipated consequences of behavior. There has been limited comparisons between goal facilitation and outcome expectations in the literature. Presseau et al., (2010) and Presseau et al., (2011) tested whether goal facilitation and goal conflict added predictability to the Theory of Planned Behavior when predicting physical activity and physician behavior, respectively. In both studies, simple correlation tables showed that goal facilitation exhibited its single highest association with attitudes (a conceptually similar construct to outcome expectations; correlations of .34 and .60, respectively; Bandura, 2005; Conner & Norman, 2005). However, in both studies, the researchers failed to include attitudes and perceptions of conflict/facilitation in the same model predicting behavior. In both studies, goal facilitation added predictability to the model, but it would be interesting to see how that might change had attitudes been included. It is questionable why attitudes weren’t included in the model to begin with in studies evaluating whether or not goal facilitation and conflict improved predictability of the Theory of Planned Behavior when attitudes are a core component of the theory. It may be more appropriate to objectively measure facilitation and conflict via diary studies as opposed to retrospective or anticipatory measures because of potential difficulty and biases of people accurately judging facilitation and conflict (Presseau et al., 2011). However, these methods are relatively time-, cost-, and labor-intensive compared to other methods, so the benefit of doing so must be clear when considering the costs of performing the research. Furthermore, the study showing that objectively measured goal conflict/facilitation predicts physical activity behavior above people’s perceptions did not include a measure of attitudes or outcome expectations within the model, which again could be problematic considering the consistent shared variance among these constructs. If outcome expectations taps a similar SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 113 construct, as the present evidence partially suggests, collecting data on goal facilitation may not be warranted. Overall summary The present study utilized a sample of 713 participants attempting New Year’s Resolutions to increase their vigorous and/or moderate physical activity for health purposes indefinitely, and measured them three times at four-week intervals. The study had four main objectives. First, it tested the efficacy of a novel intervention designed to increase exercise behavior based on the concept of goal-structure. Models of goal-structure hypothesize that people must prioritize their goals and understand how attributing time, energy, and monetary resources to one goal affects their other goals. The goal of the intervention was to reduce the degree the person’s exercise resolution would conflict with their three most important life goals, and to increase the degree it would facilitate them in achieving those goals. Tests of the goal- structure constructs at Time 1 (immediately following the intervention) confirmed that the intervention successfully altered all of these variables in the hypothesized direction. However, the latent change score analysis determined that both the Intervention and Control group increased their exercise behavior and many of the social-cognitive and motivation variables over time, but there were few significant differences across groups. These findings were unexpected for two reasons. The first reason was that it was hypothesized the Intervention group would participate in more vigorous and/or moderate exercise after taking part in the Intervention. The second reason was that the Control group was expected to decrease over time, as New Year’s Resolutions are famous for being unsuccessful (Koestner et al., 2006, 2006; Polivy & Herman, 2002). One explanation is that an intervention based on goal-structure is simply ineffective. A more optimistic explanation is that it could be possible that an intervention based on goal- SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 114 structure would be more appropriate for someone who has not yet started a behavior change (Armitage, 2009; Armitage et al., 2004; Prochaska & DiClemente, 1984), as someone who’s already started one may already view the new behavior as cohesive with their other life goals. While the study did also include a planning element, which has been shown to be effective in changing behavior (Gollwitzer, 1993; Gollwitzer & Sheeran, 2006; Wiedemann et al., 2011), it may have been nullified by the fact people may spontaneously form plans on their own before beginning a behavior change (Gollwitzer, 1993), particularly for New Year’s Resolutions (Koestner et al., 2002). Future research could test the effects of an intervention designed on goal- structure on a sample of participants merely considering a behavior change, rather than a sample who have already decided to attempt one. Finally, there may also have been an unintended question-behavior effect (aka mere-measurement effect) caused by the Control group completing the goal-structure measure the intervention was based off of (Dholakia, 2010; Sprott et al., 2006; Wood et al., 2014). Future studies could test this empirically by including a third group of participants as an extra Control group who do not complete the goal-structure measure at all, or at least not at the beginning of the study. The second objective of the study was to adapt and evaluate the Chulef, Read, and Walsh (2001) goal taxonomy for use in measuring goal-structure specifically related to exercise behavior. The study utilized 5 sub-samples of the overall sample in order to test for consistent relationships between the seven goal-structure scores, physical activity, and the other motivation constructs measured in the study. The analysis revealed that two goal-structure scores, the rank of the health goal-cluster and the degree that exercise facilitated other important goals, were potentially useful additions as indicator variables in a latent variable of motivation predicting exercise. However, a major finding of this analysis was that conflict and facilitation may be SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 115 measured inappropriately by the standard Chulef, Read, and Walsh (2001) goal taxonomy. Evidence supported treating conflict and facilitation as different constructs, as opposed to opposites on the same continuous scale, and thus should be measured separately (Riediger, 2007). Future research using the Chulef, Read, and Walsh (2001) could potentially benefit from using separate scales for measuring conflict and facilitation. The third objective of the present study was to determine if a robust motivation latent variable could be formed from previously validated measures of motivation stemming from other areas of literature, specifically, goal commitment, goal autonomy, and goal-structure (i.e., rank of the health-goal cluster and facilitation with exercise). Longitudinal measurement invariance analysis was used to determine if the motivation latent variable had robust factor loadings, indicator means, and error variances over time and across gender and study group. The analysis revealed that all models testing measurement invariance exhibited a good fit to the data, however there was evidence that scalar invariance was violated over time and across study group. The motivation latent variable was deemed suitable for use in the present study, but future research should continue testing the variable with updated motivation measures. Specifically, a more appropriate measure of goal-structure may exhibit a better outcome, as goal facilitation had the lowest factor loading and was found to contribute to violations of scalar invariance. However, it may also be the case that a different sample would find complete measurement invariance among the four motivation constructs studied as the significant worsening in model fit was relatively small (or perhaps just a smaller sample would find measurement invariance, as chi-square is highly sensitive to large samples). The fourth objective of the present research was to evaluate the utility of the motivation latent variable by determining whether or not it predicted exercise behavior over and above SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 116 intentions. Two series of path analyses revealed that the motivation latent variable consistently predicted exercise, and sometimes exhibited larger relationships with exercise compared to intentions, and even predicted exercise when intentions failed to do so in some models. The motivation latent variable also exhibited strong relationships with intentions, which resulted in a significant indirect relationship of motivation on exercise via intentions. Furthermore, the motivation latent variable frequently fully mediated the effects of self-efficacy and outcome expectations on both exercise and intentions. However, there were again potential issues related to goal facilitation, in that goal facilitation and outcome expectations may be conceptually similar constructs, which was supported statistically. Because of the inconclusive evidence supporting the goal-facilitation measures, and its conceptual overlap with outcome expectations, it would be interesting to reanalyze the data without it in the model. Doing so would reduce the complexity of the model and reduce the demands placed on participants because the goal- structure measures are relatively time-intensive regardless of how they are measured. As cognition is just one of several important factors influencing behavior, it would be prudent to avoid spending extra time and resources measuring unnecessary constructs. Future research should continue exploring the utility of goal facilitation and goal conflict, particularly in relation to outcome expectations (or attitudes), as the current literature in this area is lacking. Limitations The present research had a number of limitations. The study was conducted online and utilized self-report questionnaires, which are subject to bias. However, the majority of the measures used had been previously validated, and numerous methods were used to ensure participants were answering the questions honestly. Because the study was online, there was no way of knowing whether participants were the same at each time point (i.e., multiple people SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 117 sharing an Mturk account). There was also some attrition in the study: 16.4% of the sample failed to complete the second survey and 27.3% failed to compete the third. Several methods were used for minimizing the attrition rate, including an increasing pay scale and numerous reminder emails via multiple platforms, and these attrition rates are lower than comparable online interventions (Shahab & McEwen, 2009). Additionally, several variables measured at Time 1 were associated with missingness, and these variables were included in models utilizing fiml estimation to reduce possible bias associated with missingness. Study groups also differed significantly on a few variables prior to randomization, and these variables were also controlled for in models. Finally, some of the latent change score models used to test for intervention effects failed to run correctly, which could potentially be improved with more time-points of measurement. Conclusions While there were few differences between the Intervention group and the Control group, the total sample had numerous positive outcomes, including increased vigorous and/or moderate exercise and increased motivation throughout the study. Goal facilitation and the rank of the health goal-cluster could potentially be useful additions to a latent variable that measures motivation. Furthermore, the latent variable of motivation comprised of goal commitment, goal autonomy, goal facilitation, and rank of the health goal-cluster was fairly robust over time and across sub-groups. However, the latent variable also exhibited potential measurement issues, some of which stemming from the goal-structure measures. Future research should test an improved measure of goal-structure to determine if more appropriate measurement would mitigate these issues. Overall, the motivation latent variable exhibited impressive predictability of exercise behavior in comparison to the typical social cognitive model variables. This evidence SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 118 supports the utility of including the motivation latent variable within the social cognitive models of self-regulation. The motivation latent variable constitutes an important step in closing the infamous intention-behavior gap associated with these models, and it opens new avenues of intervention based on the indicator variables that have yet to be fully explored in the health domain. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 119 References Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human Decision Processes, 50(2), 179–211. Ajzen, I., & Fishbein, M. (1980). Understanding attitudes and predicting social behaviour. Retrieved from http://www.citeulike.org/group/38/article/235626 Arden, M. A., & Armitage, C. J. (2008). Predicting and explaining transtheoretical model stage transitions in relation to condom-carrying behaviour. British Journal of Health Psychology, 13(4), 719–735. Armitage, C. J. (2006). Evidence that implementation intentions promote transitions between the stages of change. Journal of Consulting and Clinical Psychology, 74(1), 141. Armitage, C. J. (2009). Is there utility in the transtheoretical model? British Journal of Health Psychology, 14(2), 195–210. doi:10.1348/135910708X368991 Armitage, C. J., & Conner, M. (2001). Efficacy of the theory of planned behaviour: A meta- analytic review. British Journal of Social Psychology, 40(4), 471–499. Armitage, C. J., Sheeran, P., Conner, M., & Arden, M. A. (2004). Stages of change or changes of stage? Predicting transitions in transtheoretical model stages in relation to healthy food choice. Journal of Consulting and Clinical Psychology, 72(3), 491. Austin, J. T., & Vancouver, J. B. (1996). Goal constructs in psychology: Structure, process, and content. Psychological Bulletin, 120(3), 338. Bandura, A. (1982). Self-efficacy mechanism in human agency. American Psychologist, 37(2), 122. Bandura, A. (1997). Self-efficacy: The exercise of control. Worth Publishers. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 120 Bandura, A. (2004). Health promotion by social cognitive means. Health Education & Behavior, 31(2), 143–164. Bandura, A. (2005). The Primacy of Self-Regulation in Health Promotion. Applied Psychology, 54(2), 245–254. Bargh, J. A. (1997). The automaticity of everyday life. Lawrence Erlbaum Associates Publishers. Retrieved from http://psycnet.apa.org/psycinfo/1997-08492-001 Barlow, R. E., & Proschan, F. (1975). Statistical theory of reliability and life testing: probability models. DTIC Document. Retrieved from http://oai.dtic.mil/oai/oai?verb=getRecord&metadataPrefix=html&identifier=ADA00639 9 Baumeister, R. F., Masicampo, E., & Vohs, K. D. (2011). Do conscious thoughts cause behavior? Annual Review of Psychology, 62, 331–361. Blair, S. N., & Morris, J. N. (2009). Healthy hearts—and the universal benefits of being physically active: physical activity and health. Annals of Epidemiology, 19(4), 253–256. Boekaerts, M., Maes, S., & Karoly, P. (2005). Self-Regulation Across Domains of Applied Psychology: Is there an Emerging Consensus? Applied Psychology, 54(2), 149–154. Bollen, K. A. (2002). Latent variables in psychology and the social sciences. Annual Review of Psychology, 53(1), 605–634. Brougham, R. R., & Walsh, D. A. (2005). Goal expectations as predictors of retirement intentions. The International Journal of Aging and Human Development, 61(2), 141–160. Brougham, R. R., & Walsh, D. A. (2007). Image theory, goal incompatibility, and retirement intent. The International Journal of Aging and Human Development, 65(3), 203–229. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 121 Cervone, D., Shadel, W. G., Smith, R. E., & Fiori, M. (2006). Self-Regulation: Reminders and Suggestions from Personality Science. Applied Psychology, 55(3), 333–385. Chulef, A. S., Read, S. J., & Walsh, D. A. (2001). A hierarchical taxonomy of human goals. Motivation and Emotion, 25(3), 191–232. Cohen, J. (1977). Statistical power analysis for the behavioral sciences (rev. Lawrence Erlbaum Associates, Inc. Retrieved from http://psycnet.apa.org/psycinfo/1987-98267-000 Physical Activity Guidelines Advisory Committee (2008). Physical activity guidelines advisory committee report, 2008. Washington, DC: US Department of Health and Human Services, 2008. Conner, M., & Norman, P. (2005). Predicting health behaviour. open university Press. Retrieved from http://books.google.com/books?hl=en&lr=&id=YjvuX4Q9s_wC&oi=fnd&pg=PP1&dq= predicting+health+behavior+a+social+cognition+approach&ots=QLPPD5Ul0l&sig=LU mSpWjftp5CQtaskB6bOS-0bFc Danner, U. N., Aarts, H., & Vries, N. K. (2008). Habit vs. intention in the prediction of future behaviour: The role of frequency, context stability and mental accessibility of past behaviour. British Journal of Social Psychology, 47(2), 245–265. Deci, E. L., & Ryan, R. M. (2004). Handbook of self-determination research. Univ of Rochester Pr. Deci, E. L., & Ryan, R. M. (2008). Self-determination theory: A macrotheory of human motivation, development, and health. Canadian Psychology, 49(3), 182–185. Dholakia, U. M. (2010). A critical review of question–behavior effect research. Review of Marketing Research, 7, 145–197. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 122 Donnelly, J. E., Jacobsen, D. J., Heelan, K. S., Seip, R., & Smith, S. (2000). The effects of 18 months of intermittent vs. continuous exercise on aerobic capacity, body weight and composition, and metabolic fitness in previously sedentary, moderately obese females. International Journal of Obesity and Related Metabolic Disorders: Journal of the International Association for the Study of Obesity, 24(5), 566–572. Elliot, A. J., & McGregor, H. A. (2001). A 2$\times$ 2 achievement goal framework. Journal of Personality and Social Psychology, 80(3), 501. Emmons, R. A., & King, L. A. (1988). Conflict among personal strivings: immediate and long- term implications for psychological and physical well-being. Journal of Personality and Social Psychology, 54(6), 1040. Fishbein, M., & Ajzen, I. (1975). Belief, attitude, intention and behavior: An introduction to theory and research. Retrieved from http://trid.trb.org/view.aspx?id=1150648 Fishbein, M., Triandis, H. C., Kanfer, F. H., Becker, M., Middlestadt, S. E., & Eichler, A. (2001). Factors influencing behavior and behavior change. Handbook of Health Psychology, 3–17. Fiske, S. T., & Taylor, S. E. (1991). Social cognition, 2nd. NY: McGraw-Hill. Flegal, K. M., Carroll, M. D., Kit, B. K., & Ogden, C. L. (2012). Prevalence of obesity and trends in the distribution of body mass index among US adults, 1999-2010. Jama, 307(5), 491–497. Flegal, K. M., Graubard, B. I., Williamson, D. F., & Gail, M. H. (2005). Excess deaths associated with underweight, overweight, and obesity. JAMA: The Journal of the American Medical Association, 293(15), 1861. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 123 Ford, M. E., & Nichols, C. W. (1991). Using goal assessments to identify motivational patterns and facilitate behavioral regulation and achievement. Advances in Motivation and Achievement, 7, 51–84. Foster, G. D., Wadden, T. A., Vogt, R. A., & Brewer, G. (1997). What is a reasonable weight loss? Patients’ expectations and evaluations of obesity treatment outcomes. Journal of Consulting and Clinical Psychology. Vol. 65(1), 65(1), 79–85. Fox, K. R. (1999). The influence of physical activity on mental well-being. Public Health Nutrition, 2(3a), 411–418. Gebhardt, W. A., & Maes, S. (2001). Integrating social-psychological frameworks for health behavior research. American Journal of Health Behavior, 25(6), 528–536. Godin, G., Bélanger-Gravel, A., Amireault, S., Vohl, M.-C., & Pérusse, L. (2011). The effect of mere-measurement of cognitions on physical activity behavior: a randomized controlled trial among overweight and obese individuals. International Journal of Behavioral Nutrition and Physical Activity, 8(2). Retrieved from http://www.biomedcentral.com/content/pdf/1479-5868-8-2.pdf Godin, G., & Shephard, R. J. (1985). A simple method to assess exercise behavior in the community. Canadian Journal of Applied Sport Sciences. Journal Canadien Des Sciences Appliquees Au Sport, 10(3), 141–146. Gollwitzer, P. M. (1993). Goal achievement: The role of intentions. European Review of Social Psychology, 4(1), 141–185. Gollwitzer, P. M., & Sheeran, P. (2006). Implementation intentions and goal achievement: A meta-analysis of effects and processes. Advances in Experimental Social Psychology, 38, 69–119. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 124 Hollenbeck, J. R., & Klein, H. J. (1987). Goal commitment and the goal-setting process: Problems, prospects, and proposals for future research. Journal of Applied Psychology, 72(2), 212. Hollenbeck, J. R., Williams, C. R., & Klein, H. J. (1989). An empirical examination of the antecedents of commitment to difficult goals. Journal of Applied Psychology, 74(1), 18. Horn, J. L., & McArdle, J. J. (1992). A practical and theoretical guide to measurement invariance in aging research. Experimental Aging Research, 18(3), 117–144. Hu, L., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling: A Multidisciplinary Journal, 6(1), 1–55. Jacobs Jr, D. R., Ainsworth, B. E., Hartman, T. J., & Leon, A. S. (1993). A simultaneous evaluation of 10 commonly used physical activity questionnaires. Medicine and Science in Sports and Exercise, 25(1), 81–91. Janz, N. K., & Becker, M. H. (1984). The health belief model: A decade later. Health Education & Behavior, 11(1), 1–47. Judge, T. A., Bono, J. E., Erez, A., & Locke, E. A. (2005). Core self-evaluations and job and life satisfaction: the role of self-concordance and goal attainment. Journal of Applied Psychology, 90(2), 257. Karoly, P. (1999). A goal systems–self-regulatory perspective on personality, psychopathology, and change. Review of General Psychology, 3(4), 264. Karoly, P. (2010). Goal Systems and Self-Regulation. Handbook of Personality and Self- Regulation, 218–242. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 125 Kievit, R. A., Frankenhuis, W. E., Waldorp, L. J., & Borsboom, D. (2013). Simpson’s paradox in psychological science: a practical guide. Frontiers in Psychology, 4. Retrieved from http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3740239/ Klein, H. J., Molloy, J. C., & Brinsfield, C. T. (2012). Reconceptualizing workplace commitment to redress a stretched construct: Revisiting assumptions and removing confounds. Academy of Management Review, 37(1), 130–151. Klein, H. J., Wesson, M. J., Hollenbeck, J. R., & Alge, B. J. (1999). Goal commitment and the goal-setting process: conceptual clarification and empirical synthesis. Journal of Applied Psychology, 84(6), 885. Klein, H. J., Wesson, M. J., Hollenbeck, J. R., Wright, P. M., & DeShon, R. P. (2001). The assessment of goal commitment: A measurement model meta-analysis. Organizational Behavior and Human Decision Processes, 85(1), 32–55. Klem, M. L., Wing, R. R., McGuire, M. T., Seagle, H. M., & Hill, J. O. (1997). A descriptive study of individuals successful at long-term maintenance of substantial weight loss. The American Journal of Clinical Nutrition, 66(2), 239. Knott, D., Muers, S., Aldridge, S., & Britain, G. (2008). Achieving Culture Change: A Policy Framework: a Discussion Paper by the Strategy Unit. Cabinet Office, Strategy Unit. Koestner, R., Horberg, E. J., Gaudreau, P., Powers, T., Di Dio, P., Bryan, C., Salter, N. (2006). Bolstering implementation plans for the long haul: The benefits of simultaneously boosting self-concordance or self-efficacy. Personality and Social Psychology Bulletin, 32(11), 1547–1558. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 126 Koestner, R., Lekes, N., Powers, T. A., & Chicoine, E. (2002). Attaining personal goals: self- concordance plus implementation intentions equals success. Journal of Personality and Social Psychology, 83(1), 231. Kyle, U. G., Zhang, F. F., Morabia, A., & Pichard, C. (2006). Longitudinal study of body composition changes associated with weight change and physical activity. Nutrition, 22(11), 1103–1111. Larsen, A. L., McArdle, J. J., Robertson, T., & Dunton, G. (2014). Nutrition self-efficacy is unidirectionally related to outcome expectations in children. Appetite. Retrieved from http://www.sciencedirect.com/science/article/pii/S0195666314004930 Lee, G., Talevich, J., Lee, S., Larsen, A.L., Read, S.J., Walsh, D. A., (in preparation). The impact of goal perceptions on health behaviors and successful weight management. Levav, J., & Fitzsimons, G. J. (2006). When Questions Change Behavior The Role of Ease of Representation. Psychological Science, 17(3), 207–213. Leventhal, H., & Mora, P. A. (2005). Is there a science of the processes underlying health and illness behaviors? A comment on Maes and Karoly. Applied Psychology, 54(2), 255–266. Levesque, C. S., Williams, G. C., Elliot, D., Pickering, M. A., Bodenhamer, B., & Finley, P. J. (2007). Validating the theoretical structure of the Treatment Self-Regulation Questionnaire (TSRQ) across three different health behaviors. Health Education Research, 22(5), 691–702. Li, K.-K., Chan, D. K., & others. (2008). Goal conflict and the moderating effects of intention stability in intention-behavior relations: Physical activity among Hong Kong Chinese. Journal of Sport & Exercise Psychology, 30(1), 39. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 127 Locke, E. A., & Latham, G. P. (1990). A theory of goal setting & task performance. Prentice- Hall, Inc. Retrieved from http://psycnet.apa.org/psycinfo/1990-97846-000 Locke, E. A., & Latham, G. P. (2002). Building a practically useful theory of goal setting and task motivation: A 35-year odyssey. American Psychologist, 57(9), 705. MacKinnon, D. P., Fairchild, A. J., & Fritz, M. S. (2007). Mediation analysis. Annual Review of Psychology, 58, 593. Maddux, J. E., & Rogers, R. W. (1983). Protection motivation and self-efficacy: A revised theory of fear appeals and attitude change. Journal of Experimental Social Psychology, 19(5), 469–479. Maes, S., & Karoly, P. (2005). Self-Regulation Assessment and Intervention in Physical Health and Illness: A Review. Applied Psychology, 54(2), 267–299. Marlatt, G. A., & Kaplan, B. E. (1972). Self-initiated attempts to change behavior: A study of New Year’s resolutions. Psychological Reports, 30(1), 123–131. Mason, W., & Suri, S. (2012). Conducting behavioral research on Amazon’s Mechanical Turk. Behavior Research Methods, 44(1), 1–23. McArdle, J. J. (2005). The development of the RAM rules for latent variable structural equation modeling. Contemporary Psychometrics: A Festschrift for Roderick P. McDonald, 225– 273. McArdle, J. J. (2009). Latent variable modeling of differences and changes with longitudinal data. Annual Review of Psychology, 60, 577–605. McArdle, J. J., & Nesselroade, J. R. (1994). Using multivariate data to structure developmental change. Retrieved from http://doi.apa.org/psycinfo/1994-97263-010 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 128 McCall, R. B., & Appelbaum, M. I. (1973). Bias in the analysis of repeated-measures designs: Some alternative approaches. Child Development, 401–415. McDonald, R. P. (1999). Test theory: A unified treatment. Psychology Press. Retrieved from http://books.google.com/books?hl=en&lr=&id=acww1KB6BdEC&oi=fnd&pg=PP1&dq =mcdonald+test+theory+a+unified+treatment&ots=V- a9SXNUfM&sig=NF1NOie45Ar7FJqDa2S2y-VsjNs McEachan, R. R. C., Conner, M., Taylor, N. J., & Lawton, R. J. (2011). Prospective prediction of health-related behaviours with the theory of planned behaviour: A meta-analysis. Health Psychology Review, 5(2), 97–144. Meredith, W. (1993). Measurement invariance, factor analysis and factorial invariance. Psychometrika, 58(4), 525–543. Murphy, M. H., Nevill, A. M., Neville, C., Biddle, S., & Hardman, A. E. (2002). Accumulating brisk walking for fitness, cardiovascular risk, and psychological health. Medicine & Science in Sports & Exercise, 34(9), 1468–1474. Nelson, M. C., Story, M., Larson, N. I., Neumark-Sztainer, D., & Lytle, L. A. (2008). Emerging adulthood and college-aged youth: an overlooked age for weight-related behavior change. Obesity, 16(10), 2205–2211. Norcross, J. C., Ratzin, A. C., & Payne, D. (1989). Ringing in the New Year: The change processes and reported outcomes of resolutions. Addictive Behaviors, 14(2), 205–212. Nunnally, J. C., Bernstein, I. H., & Berge, J. M. ten. (1967). Psychometric theory (Vol. 226). McGraw-Hill New York. Retrieved from http://rds.epi- ucsf.org/ticr/syllabus/courses/46/2005/10/20/Lecture/readings/Psychometric%20Theory.p df SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 129 O’Connor, D. B., Jones, F., Conner, M., McMillan, B., & Ferguson, E. (2008). Effects of daily hassles and eating style on eating behavior. Health Psychology, 27(1S), S20. Ogden, C. L., Carroll, M. D., Kit, B. K., & Flegal, K. M. (2012). Prevalence of obesity and trends in body mass index among US children and adolescents, 1999-2010. Jama, 307(5), 483–490. Okun, M. A., Fairholme, C., Karoly, P., Ruehlman, L. S., & Newton, C. (2006). Academic goals, goal process cognition, and exam performance among college students. Learning and Individual Differences, 16(3), 255–265. Pearl, J. (2014). Comment: Understanding Simpson’s Paradox. The American Statistician, 68(1), 8–13. Plotnikoff, R. C., Lippke, S., Reinbold-Matthews, M., Courneya, K. S., Karunamuni, N., Sigal, R. J., & Birkett, N. (2007). Assessing the validity of a stage measure on physical activity in a population-based sample of individuals with type 1 or type 2 diabetes. Measurement in Physical Education and Exercise Science, 11(2), 73–91. Polivy, J., & Herman, C. P. (2002). If at first you don’t succeed: False hopes of self-change. American Psychologist. Vol. 57(9), 57(9), 677–689. Porter, R. L., & Latham, G. P. (2013). The effect of employee learning goals and goal commitment on departmental performance. Journal of Leadership & Organizational Studies, 20(1), 62–68. Presseau, J., Francis, J. J., Campbell, N. C., & Sniehotta, F. F. (2011). Goal conflict, goal facilitation, and health professionals’ provision of physical activity advice in primary care: an exploratory prospective study. Implement Sci, 6, 73. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 130 Presseau, J., Sniehotta, F. F., Francis, J. J., & Gebhardt, W. A. (2010). With a little help from my goals: Integrating intergoal facilitation with the theory of planned behaviour to predict physical activity. British Journal of Health Psychology, 15(4), 905–919. Presseau, J., Sniehotta, F. F., Francis, J. J., & Little, B. R. (2008). Personal project analysis: opportunities and implications for multiple goal assessment, theoretical integration, and behaviour change. Retrieved from http://aura.abdn.ac.uk/handle/2164/276 Presseau, J., Tait, R. I., Johnston, D. W., Francis, J. J., & Sniehotta, F. F. (2013). Goal conflict and goal facilitation as predictors of daily accelerometer-assessed physical activity. Health Psychology, 32(12), 1179. Prochaska, J. O., & DiClemente, C. C. (1984). The transtheoretical approach: crossing boundaries of traditional therapy. Dow Jones-Irwin, Homewood/Illinois. Read, S. J., Talevich, J., Walsh, D. A., Chopra, G., & Iyer, R. (2010). A comprehensive taxonomy of human motives: a principled basis for the motives of intelligent agents. In Intelligent Virtual Agents (pp. 35–41). Retrieved from http://link.springer.com/chapter/10.1007/978-3-642-15892-6_4 Rhodes, R. E., & de Bruijn, G.-J. (2013). How big is the physical activity intention–behaviour gap? A meta-analysis using the action control framework. British Journal of Health Psychology, 18(2), 296–309. Rhodes, R. E., & Dickau, L. (2012). Experimental evidence for the intention–behavior relationship in the physical activity domain: A meta-analysis. Health Psychology, 31(6), 724. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 131 Rhodes, R. E., & Dickau, L. (2013). Moderators of the intention-behaviour relationship in the physical activity domain: a systematic review. British Journal of Sports Medicine, 47(4), 215–225. Rhodes, R. E., & Nigg, C. R. (2011). Advancing physical activity theory: a review and future directions. Exercise and Sport Sciences Reviews, 39(3), 113–119. Rhodes, R. E., Plotnikoff, R. C., & Courneya, K. S. (2008). Predicting the physical activity intention–behavior profiles of adopters and maintainers using three social cognition models. Annals of Behavioral Medicine, 36(3), 244–252. Riediger, M. (2007). Interference and facilitation among personal goals: Age differences and associations with well-being and behavior. Personal Project Pursuit: Goals, Action, and Human Flourishing, 119–143. Riediger, M., & Freund, A. M. (2004). Interference and facilitation among personal goals: Differential associations with subjective well-being and persistent goal pursuit. Personality and Social Psychology Bulletin, 30(12), 1511–1523. Rogers, R. W. (1975). A Protection Motivation Theory of Fear Appeals and Attitude Change1. The Journal of Psychology, 91(1), 93–114. Rogers, R. W. (1983). Cognitive and physiological processes in fear appeals and attitude change: A revised theory of protection motivation. Social Psychophysiology, 153–176. Rosenstock, I. M. (1966). Why people use health services. The Milbank Memorial Fund Quarterly, 44(3), 94–127. Rosenstock, I. M. (2000). Health Belief Model. Retrieved from http://psycnet.apa.org/books/10519/035 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 132 Rosseel, Y. The ‘lavaan’ package in R. Retrieved from http://cran.r-roject.org/web /packages/lavaan/index.html. 2011. Rothman, A. J. (2004). International Journal of Behavioral Nutrition and Physical Activity. International Journal of Behavioral Nutrition and Physical Activity, 1, 11. Rutter, D., & Quine, L. (2002). Social cognition models and changing health behaviours. Changing Health Behaviour: Intervention and Research with Social Cognition Models, 1–27. Sandberg, T., & Conner, M. (2011). Using self-generated validity to promote exercise behaviour. British Journal of Social Psychology, 50(4), 769–783. Schwarzer, R. (1992). Self-efficacy in the adoption and maintenance of health behaviors: Theoretical approaches and a new model. Hemisphere Publishing Corp. Retrieved from http://psycnet.apa.org/psycinfo/1992-97719-010 Schwarzer, R. (2008). Modeling health behavior change: How to predict and modify the adoption and maintenance of health behaviors. Applied Psychology, 57(1), 1–29. Seijts, G. H., & Latham, G. P. (2011). The effect of commitment to a learning goal, self-efficacy, and the interaction between learning goal difficulty and commitment on performance in a business simulation. Human Performance, 24(3), 189–204. Shahab, L., & McEwen, A. (2009). Online support for smoking cessation: a systematic review of the literature. Addiction, 104(11), 1792–1804. Sheeran, P., Gollwitzer, P. M., & Bargh, J. A. (2012). Nonconscious processes and health. Retrieved from http://psycnet.apa.org/psycinfo/2012-21504-001/ SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 133 Sheldon, K. M., & Elliot, A. J. (1998). Not AM Personal Goals Are Personal: Comparing Autonomous and Controlled Reasons for Goals as Predictors of Effort and Attainment. Personality and Social Psychology, 24(5), 546. Sheldon, K. M., & Elliot, A. J. (1999). Goal striving, need satisfaction, and longitudinal well- being: the self-concordance model. Journal of Personality and Social Psychology, 76(3), 482. Sheldon, K. M., & Houser-Marko, L. (2001). Self-concordance, goal attainment, and the pursuit of happiness: Can there be an upward spiral? Journal of Personality and Social Psychology, 80(1), 152. Silventoinen, K., Sans, S., Tolonen, H., Monterde, D., Kuulasmaa, K., Kesteloot, H., & Tuomilehto, J. (2004). Trends in obesity and energy supply in the WHO MONICA Project. International Journal of Obesity, 28(5), 710–718. Sniehotta, F. F., Presseau, J., & Araújo-Soares, V. (2014). Time to retire the theory of planned behaviour. Health Psychology Review, 8(1), 1–7. Sörbom, D. (1974). A general method for studying differences in factor means and factor structure between groups. British Journal of Mathematical and Statistical Psychology, 27(2), 229–239. Sörbom, D. (1978). An alternative to the methodology for analysis of covariance. Psychometrika, 43(3), 381–396. Spearman, C. (1904). The proof and measurement of association between two things. The American Journal of Psychology, 15(1), 72–101. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 134 Spence, J. C., Burgess, J., Rodgers, W., & Murray, T. (2009). Effect of pretesting on intentions and behaviour: A pedometer and walking intervention. Psychology and Health, 24(7), 777–789. Sprott, D. E., Spangenberg, E. R., Block, L. G., Fitzsimons, G. J., Morwitz, V. G., & Williams, P. (2006). The question–behavior effect: What we know and where we go from here. Social Influence, 1(2), 128–137. Strecher, V. J., Seijts, G. H., Kok, G. J., Latham, G. P., Glasgow, R., DeVellis, B., … Bulger, D. W. (1995). Goal setting as a strategy for health behavior change. Health Education & Behavior, 22(2), 190–200. Symons Downs, D., & Hausenblas, H. A. (2005). Exercise behavior and the theories of reasoned action and planned behavior: A meta-analytic update. Journal of Physical Activity and Health, 2, 76–97. Teixeira, P. J., Going, S. B., Sardinha, L. B., & Lohman, T. G. (2005). A review of psychosocial pre-treatment predictors of weight control. Obesity Reviews, 6(1), 43–65. Thomas, J. G., Bond, D. S., Hill, J. O., & Wing, R. R. (2011). THE NATIONAL WEIGHT CONTROL REGISTRY: A Study of“ Successful Losers.” ACSM’s Health & Fitness Journal, 15(2), 8. Vainio, H., & Bianchini, F. (2002). Weight control and physical activity. IARC. Voelkle, M. C. (2007). Latent growth curve modeling as an integrative approach to the analysis of change. Psychology Science, 49(4), 375. Warburton, D. E. R., Nicol, C. W., & Bredin, S. S. D. (2006). Health benefits of physical activity: the evidence. Canadian Medical Association Journal, 174(6), 801–809. doi:10.1503/cmaj.051351 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 135 Webb, T. L., & Sheeran, P. (2006). Does changing behavioral intentions engender behavior change? A meta-analysis of the experimental evidence. Psychological Bulletin, 132(2), 249. Wiedemann, A. U., Lippke, S., Reuter, T., Ziegelmann, J. P., & Schüz, B. (2011). The more the better? The number of plans predicts health behaviour change. Applied Psychology: Health and Well-Being, 3(1), 87–106. Williams, P., Fitzsimons, G. J., & Block, L. G. (2004). When consumers do not recognize “benign” intention questions as persuasion attempts. Journal of Consumer Research, 31(3), 540–550. Wing, R. R., & Phelan, S. (2005). Long-term weight loss maintenance. The American Journal of Clinical Nutrition, 82(1), 222S. Wójcicki, T. R., White, S. M., & McAuley, E. (2009). Assessing outcome expectations in older adults: the multidimensional outcome expectations for exercise scale. The Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 64(1), 33–40. Wood, C., Conner, M., Sandberg, T., Godin, G., & Sheeran, P. (2014). Why does asking questions change health behaviours? The mediating role of attitude accessibility. Psychology & Health, 29(4). Retrieved from http://www.tandfonline.com/eprint/42syGNBje8APs64kYdbw/full Wood, W., & Neal, D. T. (2007). A new look at habits and the habit-goal interface. Psychological Review, 114(4), 843–863. Wood, W., Tam, L., & Witt, M. G. (2005). Changing Circumstances, Disrupting Habits. Journal of Personality and Social Psychology, 88(6), 918. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 136 Tables Table 1 Goal clusters used from the Chulef, Read, and Walsh (2001) goal taxonomy Cluster description 1 Being a moral and virtuous person (e.g., sticking to personal morals, helping others, being highly regarded). 2 Religion and spirituality 3 Self-fulfillment and being open to new experiences (e.g., wisdom, appreciating beauty, embracing life) 4 Avoiding negative social experiences (e.g., self-protection, avoiding rejection or conflict). 5 Good social relationships (e.g., intimacy, belonging, holding power over others). 6 Good family relationships (e.g., close to parents, being a good family member). 7 Being intelligent and skillful (e.g., intellectual growth, being autonomous, competence). 8 Having financial and occupational success. (e.g., financial freedom, wealth, respected job). 9 Being physically healthy (e.g., being active, capable of daily tasks, physically fit). SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 137 Table 2 Similarities of 5 samples: Sample sizes and percentage of participants shared across samples Whole sample (n = 713) Complete cases (n = 386) 50% sample 1 (n = 357) 50% sample 2 (n = 378) 50% sample 3 (n = 358) Whole sample X 54.1% 50.1% 53.0% 50.2% Complete cases X 27.5% 26.8% 27.3% 50% sample 1 X 25.4% 25.1% 50% sample 2 X 27.9% 50% sample 3 X SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 138 Table 3 Participant demographics Participants (N = 713) N Intervention group 349 (48.9%) Percent Mturk Sample 90% Percent Female 62.8% Mean Age (SD) 31.76 (11.23) Mean days since starting an intention to increase exercise 7.97 (7.47) Race/ethnicity % White/Caucasian 74.9% Asian 9.3% African American 7.7% Latino 4.8% Other 3.4% Education % </= high school degree 9.8% Enrolled in college 45.1% 4 year college degree 32.4% Post-graduate degree 12.6% Mturk sample location (N = 640) % North East 27.7% South East 24.7% Midwest 24.6% South West 12.7% North West 9.2% Alaska/Hawaii 1.1% SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 139 Table 4 Percent of people reporting types of exercise they planned to do at Time 1, Time 2, and Time 3 Type of Exercise Time 1 Time 2 Time 3 Walking 63.8% 73.7% 78.8% Running/Jogging 60.9% 57.4% 58.9% Cycling 23.4% 26.1% 27.5% Swimming 11.6% 12.1% 13.2% Weight lifting 53.7% 49.6% 50.1% Exercising with friends 26.2% 24.7% 22.9% Recreational activities 14.9% 14.2% 13.6% Joining an exercise club (e.g., cross fit) 14.2% 16.5% 16.2% Joining an exercise program (e.g., personal trainer) 24.5% 24.7% 24.5% Other 21.7% 19.5% 13.9% SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 140 Table 5. Descriptive statistics for minutes per week of vigorous exercise, vigorous/moderate exercise, and success in accomplishing future exercise goals prior to initiating an intention to increase exercise, Time 1, Time 2, and Time 3. Mean Median Sd 20 th percentile 40 th percentile 60 th percentile 80 th percentile Prior exercise Vigorous 22.55 0.00 48.37 0.00 0.00 0.00 35.00 Vigorous/moderate 62.77 30.00 89.89 0.00 20.00 50.00 110.00 Time 1 exercise Vigorous 53.51 20.00 75.48 0.00 10.00 40.00 100.00 Vigorous/moderate 122.56 90.00 125.30 20.00 63.00 120.00 195.00 Time 2 exercise Vigorous 69.83 30.00 108.58 0.00 15.00 60.00 120.00 Vigorous/moderate 163.26 120.00 171.67 45.00 90.00 150.00 240.00 Vigorous/moderate success 0.75 0.76 0.44 .42 .67 .85 1.01 Time 3 exercise Vigorous 72.91 35.00 112.11 0.00 20.00 60.00 120.00 Vigorous/moderate 168.74 120.00 192.98 39.00 90.00 155.00 260.00 Vigorous/moderate success 0.76 0.82 0.41 .45 .73 .89 1.04 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 141 Table 6 Descriptive statistics for social-cognitive predictors at all three time points (entire sample used, untransformed variables) Variable Time 1 Time 2 Time 3 Median Mean SD Median Mean SD Median Mean SD Strength of intentions for exercise Vigorous 37.50 33.74 12.95 37.50 31.61 13.66 37.50 32.47 13.74 Vigorous/Moderate 57.50 54.13 13.99 55.50 51.27 16.41 57.00 52.02 16.53 Social-Cognitive Variables Self-Efficacy 33.00 31.97 6.83 32.00 30.90 7.42 32.00 31.02 7.97 Outcome-Expectations 25.00 24.86 3.01 24.00 24.66 3.12 25.00 24.99 3.06 Planning 13.00 12.66 2.58 12.00 12.30 2.71 12.00 12.31 2.69 Motivation Variables Commitment 21.00 20.99 3.12 20.00 20.06 3.54 20.00 20.08 3.65 Autonomous Motivation 13.00 12.40 3.49 12.00 12.15 3.69 14.00 12.47 3.58 Controlled motivation 7.00 6.70 3.49 6.00 6.57 3.16 7.00 7.03 3.49 Goal Structure variables Ranking 4.00 3.89 1.99 4.00 3.82 1.95 4.00 3.81 2.06 Facilitation/conflict exercise 55 56.43 32.58 48.00 40.96 31.93 49.00 52.74 34.99 Facilitation/conflict no exercise -33.00 -30.76 42.47 -31.00 -29.72 38.93 -29.50 -29.33 38.25 Conflict exercise 9.00 12.12 6.69 9.00 12.59 8.27 9.00 12.04 7.66 Facilitation exercise 61.00 63.42 28.25 54.00 48.75 26.59 55.00 59.81 29.58 Conflict no exercise 40.00 44.83 26.38 38.00 42.40 24.74 37.00 41.89 24.56 Facilitation no exercise 9.00 17.42 20.25 16.06 9.00 18.89 9.00 15.88 17.46 Future exercise goals Minutes of Vigorous 90.00 131.50 151.54 80.00 123.02 171.17 90.00 127.11 164.98 Minutes of Vigorous/Moderate 210.00 278.06 251.99 210.00 271.50 249.45 220.00 281.65 282.72 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 142 Table 7 Intervention Group descriptive statistics for all variables at Time 1, Time 2, and Time 3 Variable Time 1 Time 2 Time 3 Median Mean SD Median Mean SD Median Mean SD Exercise variables (MET units) Minutes of vigorous 20.00 54.66 75.60 40.00 73.44 120.66 40.00 74.28 107.86 Minutes of vigorous/moderate 90.00 123.25 128.02 130.00 163.81 167.33 120.00 160.75 187.58 Vigorous/moderate success .76 .74 .44 .79 .72 .39 Strength of intentions for exercise Vigorous 37.50 34.26 12.67 37.50 32.17 13.62 37.50 31.72 14.17 Vigorous/Moderate (MET units) 57.50 54.28 13.88 54.00 51.41 16.47 53.50 50.66 17.03 Social-Cognitive Variables Self-Efficacy 33.00 32.01 6.78 32.00 31.15 7.08 32.00 31.34 7.99 Outcome-Expectations 25.00 25.11 2.95 24.00 24.44 3.25 25.00 24.88 3.13 Planning 13.00 12.69 2.59 12.00 12.24 2.73 12.00 12.32 2.81 Motivation Variables Commitment 21.00 20.85 3.11 20.00 20.16 3.43 20.00 20.29 3.55 Autonomous Motivation 12.00 12.38 3.43 12.00 11.95 3.56 13.00 12.64 3.27 Controlled motivation 7.00 6.87 3.46 6.00 6.33 3.20 7.00 6.81 3.38 Goal Structure variables Ranking 4.00 3.91 2.01 4.00 3.76 2.04 3.00 3.61 2.08 Facilitation/conflict exercise 60.00 61.19 30.85 49.50 52.18 29.92 49.00 52.18 35.52 Facilitation/conflict no exercise -40.00 -39.93 38.64 -32.00 -33.78 37.35 -33.00 -32.76 38.10 Conflict exercise 9.00 11.08 5.25 9.00 11.91 6.91 9.00 12.03 8.78 Facilitation exercise 64.00 66.60 27.47 55.00 59.26 25.44 55.00 59.35 29.84 Conflict no exercise 45.00 49.17 28.50 39.00 44.07 27.19 38.00 43.35 26.72 Facilitation no exercise 9.00 13.53 13.73 9.00 13.84 13.97 9.00 14.24 15.33 Future exercise goals Minutes of Vigorous 100.00 129.01 137.46 90.00 132.56 206.27 90.00 123.71 148.43 Minutes of Vigorous/Moderate 195.00 260.00 206.48 222.50 266.13 257.02 205.00 253.78 235.67 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 143 Table 8 Control Group descriptive statistics for all variables at Time 1, Time 2, and Time 3 (untransformed variables) Variable Time 1 Time 2 Time 3 Median Mean SD Median Mean SD Median Mean SD Exercise variables Minutes of vigorous 20.00 52.42 75.45 30.00 66.68 96.88 30.00 71.72 115.89 Minutes of vigorous/moderate 90.00 121.90 122.81 112.50 162.77 175.68 130.00 175.67 197.68 Vigorous/moderate success .73 .76 .45 .82 .79 .43 Strength of intentions for exercise Vigorous 37.50 34.20 12.50 37.50 30.88 13.74 37.50 33.12 13.36 Vigorous/Moderate 57.50 54.46 13.79 54.00 50.74 16.24 57.50 53.20 16.03 Social-Cognitive Variables Self-Efficacy 32.00 31.09 7.06 32.00 30.45 7.51 32.00 30.73 7.96 Outcome-Expectations 25.00 24.85 2.99 24.00 25.00 3.00 25.00 25.10 3.00 Planning 13.00 12.53 2.62 12.00 12.25 2.69 12.00 12.30 2.58 Motivation Variables Commitment 21.00 20.42 3.35 20.00 19.74 3.57 20.00 19.90 3.74 Autonomous Motivation 13.00 12.13 3.63 12.00 12.18 3.62 13.00 12.32 3.83 Controlled motivation 7.00 7.16 3.73 7.00 6.90 3.34 7.00 7.22 3.59 Goal Structure variables Ranking 4.00 3.89 1.97 4.00 3.88 1.88 4.00 3.98 2.03 Facilitation/conflict exercise 49.50 51.87 33.57 46.00 49.89 33.62 49.50 53.23 34.59 Facilitation/conflict no exercise -28.00 -22.07 44.19 -30.00 -26.18 39.99 -28.00 -26.36 38.22 Conflict exercise 9.00 13.13 7.70 9.00 13.18 9.26 9.00 12.04 6.56 Facilitation exercise 57.00 60.37 28.69 52.00 58.31 27.60 54.50 60.21 29.40 Conflict no exercise 37.00 40.68 23.47 37.00 40.94 22.33 37.00 40.62 22.49 Facilitation no exercise 9.00 21.16 24.39 9.00 17.99 22.15 9.00 17.30 19.03 Future exercise goals Vigorous 75.00 114.65 132.98 75.00 114.65 132.88 90.00 130.06 179.76 Vigorous/Moderate 210.00 295.38 288.21 210.00 276.20 243.01 225.00 305.80 316.43 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 144 Table 9 Tests of Mean Differences Between Intervention and Control Groups in Social-Cognitive and Motivation Variables at Time 1 (n = 711-713). Variables Int Con Independent Samples T-Test results Mean Mean Mean Diff S P 95% CI Strength of intentions for exercise Vigorous 34.26 34.20 .05 .94 .956 [-1.80, 1.90] Vigorous/Moderate 54.28 54.46 -.18 1.04 .86 [-2.21, 1.86] Social-Cognitive Variables Self-Efficacy 32.01 31.09 .92 .52 .076 [-.10, 1.94] Outcome Expectations 25.11 24.85 .25 .22 .254 [-.18, .69] Planning 12.69 12.53 .16 .20 .428 [-.23, .54] Motivation Variables Commitment 20.85 20.42 .43 .24 .080 [-.05, .90] Autonomous Motivation 12.38 12.13 .25 .26 .336 [-.26, .77] Controlled motivation 6.87 7.16 -.28 .27 .295 [-.81, .25] Goal Structure variables Ranking 3.91 3.89 .02 .15 .903 [-.27, .31] Facilitation/conflict exercise 61.19 51.87 9.32 2.42 <.001*** [4.57, 14.06] Facilitation/conflict no exercise -39.93 -22.07 -17.77 3.11 <.001*** [-23.88, -11.65] Conflict exercise 11.08 13.13 -.205 .50 <.001*** [-3.02, -1.07] Facilitation exercise 66.60 60.37 6.24 2.11 .003** [2.10, 10.37] Conflict no exercise 49.17 40.68 8.48 1.95 <.001*** [4.65, 12.32] Facilitation no exercise 13.53 21.16 -7.63 1.49 <.001*** [-10.56, -4.70] Note. Int = Intervention Group; Con = Control Group; S = standard error. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 145 Table 10 Model fit indices for intervention effects on vigorous and vigorous/moderate exercise Variable Model # Models Goodness of fit indices Change statistics compared to prior model Χ 2 DF p-value CFI/TLI RMSEA ΔΧ 2 (Δdf) p-value of Δ Vigorous exercise MET units 1a No change 97.741 31 <.001 .888/.892 0.078 1b Add β 54.503 30 0.003 .957/.956 0.05 43.238 (1) <.001*** 1c Add slope 48.658 28 0.009 .965/.963 0.045 5.845 (2) .054 1d Add β and slope 28.433 27 0.389 .998/.997 0.012 20.225 (1) <.001*** 1e Free slope across groups 28.337 26 0.342 .996/.995 0.016 0.096 (1) .757 1f Free β across groups 28.336 25 0.293 .994/.993 0.019 0.001 (1) .975 1g Free cov across groups 28.218 24 0.251 .993/.991 0.022 0.118 (1) .731 1h Free slope & intercept variance across groups 28.009 22 0.215 .992/.989 0.025 0.209 (2) .901 Vigorous/ moderate exercise MET units 2a No change 120.274 31 <.001 .832/.837 0.09 2b Add β 87.246 30 .001 .892/.892 .073 33.028 (1) <.001*** 2c Add slope 65.826 28 <.001 .929/.924 0.062 21.42 (2) <.001*** 2d Add β and slope 35.627 27 0.124 .984/.982 0.03 30.199 (1) <.001*** 2e Free slope across groups 35.432 26 0.103 .982/.989 0.032 0.195 (1) .659 2f Free β across groups 34.327 25 0.101 .982/.979 0.032 1.105 (1) .293 2g Free cov across groups 32.809 24 0.108 .983/.979 0.032 1.518 (1) .218 2h Free slope & intercept variance across groups 32.565 22 0.089 .982/.977 0.034 0.244 (2) .885 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 146 Table 11 Parameter estimates for the best fitting model for intervention effects on vigorous and vigorous/moderate exercise Vigorous Exercise MET Vigorous/Moderate exercise MET Model description Groups combined; Dual change model (Model 1d) Groups combined; Dual change model (Model 2d) Parameters M S M S Mean 𝐼 0 15.301 1.561 22.967 1.645 Mean SL 20.309 2.937 24.650 3.025 B -0.942 .126 -0.873 .095 𝐼 0 ↔SL 41.026 7.555 40.205 8.215 𝐼 0 Variance 28.712 10.514 58.239 10.319 SL Variance 93.124 19.355 98.203 15.845 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 147 Table 12 Strength of intentions model fit indices Variables Model # Models Goodness of fit indices Change statistics compared to prior model Χ 2 DF p-value CFI/TLI RMSEA ΔΧ 2 (Δdf) p-value of Δ Intentions Vigorous Exercise MET Units 3a No change 97.33 31 <.001 .908/.911 0.077 3b Add β 96.774 30 <.001 .908/.908 0.079 0.556 (1) .456 3c Add slope 51.558 28 0.004 .967/.965 0.049 45.216 (2) <.001*** 3d Add β and slope 32.64 27 0.209 .992/.991 0.024 18.918 (1) <.001*** 3e Free slope across groups 32.639 26 0.173 .991/.989 0.027 0.001 (1) .975 3f Free β across groups 31.94 25 0.16 .990/.988 0.028 0.699 (1) .403 3g Free cov across groups 31.574 24 0.138 .990/.987 0.030 0.366 (1) .545 3h Free slope & intercept variance across groups 29.614 22 0.161 .992/.988 0.028 1.96 (2) .375 Intentions Vigorous/ Moderate Exercise MET Units 4a No change 161.944 31 <.001 .861/.866 0.109 4b Add β 120.018 30 <.001 .852/.852 0.092 1.614 (1) .204 4c Add slope 60.962 28 <.001 .946/.942 0.057 59.056 (2) <.001*** 4d Add β and slope 32.573 27 0.212 .991/.990 0.024 28.389 (1) <.001*** 4e Free slope across groups 32.52 26 0.176 .989/.988 0.027 0.053 (1) .818 4f Free β across groups 31.426 25 0.175 .989/.987 0.027 1.094 (1) .296 4g Free cov across groups 31.02 24 0.153 .988/.986 0.029 0.406 (1) .524 4h Free slope & intercept variance across groups 29.553 22 0.163 .989/.986 0.028 1.467 (2) 4h Free slope & intercept variance across groups 28.535 22 0.196 .990/.986 0.026 0.447 (2) .480 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 148 Table 13 Strength of intentions parameter estimates for best fitting model Strength Intentions Vigorous Exercise Strength Intentions Vigorous/Moderate exercise Model description Groups combined; Dual change model (Model 3d) Groups combined; Dual change model (Model 4d) Parameters M S M S Mean 𝐼 0 41.200 1.555 60.219 1.766 Mean SL 27.524 4.671 47.818 6.200 B -0.745 0.112 -0.868 0.101 𝐼 0 ↔SL 57.718 10.809 93.914 13.499 𝐼 0 Variance 91.324 7.840 106.279 10.552 SL Variance 76.381 17.493 146.465 27.177 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 149 Table 14 Model fit indices for self-efficacy, outcome expectations, and planning Variables Model # Models Goodness of fit indices Change statistics compared to prior model Χ 2 DF p-value CFI/TLI RMSEA ΔΧ 2 (Δdf) p-value of Δ Self- Efficacy 5a No change 59.688 31 0.001 .952/.953 0.051 5b Add β 57.836 30 0.002 .953/.953 0.051 1.852 (1) .174 5c Add slope 32.669 28 0.248 .992/.992 0.022 25.167 (2) <.001*** 5d Add β and slope 29.743 27 0.326 .995/.995 0.017 2.926 (1) .087 5e Free slope across groups 29.559 26 0.286 .994/.993 0.020 0.184 (1) .668 5f Free β across groups 29.558 25 0.241 .992/.991 0.023 0.001 (1) .975 5g Free cov across groups 29.333 24 0.208 .991/.989 0.025 0.225 (1) .635 5h Free slope & intercept variance across groups 28.876 22 0.184 .990/.987 0.027 0.457 (2) .796 Outcome Expectations (variance of slope restricted to zero) 6a No change 52.246 31 0.01 .963/.964 0.044 6b Add β 52.24 30 0.007 .961/.961 0.046 0.006 (1) .938 6c Add slope 57.315 30 0.002 .952/.952 0.051 -5.075 (0) X 6d Add β and slope 50.081 29 0.009 .963/.962 0.045 7.234 (1) .007** 6e Free slope across groups 45.525 28 0.02 .969/.967 0.042 4.556 (1) .033* 6f Free β across groups 44.017 27 0.021 .970/.967 0.042 1.508 (1) .219 6g Free cov across groups X X X X X X X 6h Free slope & intercept variance across groups X X X X X X X Planning 7a No change 47.592 31 0.029 .963/.964 0.039 7b Add β 46.864 30 0.026 .962/.962 0.04 0.728 (1) .394 7c Add slope 42.418 28 0.04 .968/.966 0.038 4.446 (2) .108 7d Add β and slope 29.076 27 0.357 .995/.995 0.015 13.342 (1) <.001*** 7e Free slope across groups 28.752 26 0.322 .994/.993 0.017 0.324 (1) .569 7f Free β across groups 27.964 25 0.309 .993/.992 0.018 0.788 (1) .375 7g Free cov across groups 27.27 24 0.292 .993/.991 0.020 0.694 (1) .405 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 150 7h Free slope & intercept variance across groups 27.152 22 0.25 .991/.988 0.023 0.118 (2) .943 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 151 Table 15 Parameter estimates for best fitting models for self-efficacy, outcome expectations, and planning. Self-efficacy Planning Outcome Expectations Model description Groups combined; Linear change model (Model 5c) Groups combined; Dual change model (Model 6d) Groups combined; No growth model (Model 7a) Parameters M S M S M S Mean 𝐼 0 28.343 0.881 11.981 0.342 25.884 .358 Mean SL -0.706a 0.554 12.386 1.959 X X B X X -1.028 0.160 X X 𝐼 0 ↔SL 1.305a 1.217 3.820 0.693 X X 𝐼 0 Variance 28.047 2.528 3.934 0.399 5.472 .382 SL Variance 2.659 0.980 4.835 1.404 X X Note. a = p > .05; Self-efficacyslope p= .202; self-efficacy covariance p = .284; SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 152 Table 16 Model fit indices for commitment, autonomous motivation, and controlled motivation Variables Model # Models Goodness of fit indices Change statistics compared to prior model Χ 2 DF p-value CFI/TLI RMSEA ΔΧ 2 (Δdf) p-value of Δ Commit- ment 8a No change 105.69 31 <001 .865/.869 0.082 8b Add β 105.613 30 <.001 .863/.863 0.084 0.077 (1) .718 8c Add slope 65.527 28 <.001 .932/.927 0.061 40.086 (2) <.001*** 8d Add β and slope 39.623 27 0.056 .977/.975 0.036 25.904 (1) <.001*** 8e Free slope across groups 39.388 26 0.045 .976/.972 0.038 0.235 (1) .628 8f Free β across groups 39.125 25 0.036 .974/.969 0.04 0.263 (1) .608 8g Free cov across groups 38.541 24 0.03 .974/.967 0.041 0.584 (1) .445 8h Free slope & intercept variance across groups 36.545 22 0.036 .976/.968 0.041 1.996 (2) .369 Autono- mous motivation 9a No change 47.714 31 0.028 .970/.971 0.039 9b Add β 43.923 30 0.048 .975/.975 0.036 3.791 (1) .052 9c Add slope 42.419 28 0.04 .974/.972 0.038 1.504 (2) .471 9d Add β and slope 34.616 27 0.149 .986/.985 0.028 7.803 (1) .005** 9e Free slope across groups 32.984 26 0.163 .988/.986 0.027 1.632 (1) .201 9f Free β across groups 29.444 25 0.246 .992/.990 0.022 3.540 (1) .060 9g Free cov across groups 28.572 24 0.237 .992/.990 0.023 0.872 (1) .350 9h Free slope & intercept variance across groups 24.517 22 0.376 .997/.996 0.043 4.055 (2) .132 Controlled motivation 10a No change 53.135 31 0.008 .943/.945 0.045 10b Add β 52.178 30 0.007 .943/.943 0.046 0.957 (1) .328 10c Add slope 44.564 28 0.024 .958/.955 0.041 7.614 (2) .022* 10d Add β and slope 41.927 27 0.033 .962/.958 0.039 2.637 (1) .104 10e Free slope across groups 40.904 26 0.032 .962/.956 0.04 1.023 (1) .312 10f Free β across groups 40.598 25 0.025 .960/.952 0.042 0.306 (1) .580 10g Free cov across groups 35.743 24 0.058 .970/.962 0.037 4.855 (1) .028* SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 153 10h Free slope & intercept variance across groups 35.724 22 0.044 .967/.958 0.039 0.019 (2) .991 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 154 Table 17 Parameter estimates for best fitting models for commitment, autonomous motivation, and controlled motivation Commitment Autonomous Motivation Controlled Motivation Model description Dual change model; Groups invariant (Model 8d) Dual change model; Groups invariant (Model 9d) Linear change model; Groups invariant (Model 10c) Parameters M S M S M S Mean 𝐼 0 18.593 .423 10.633 0.451 9.635 .449 Mean SL 14.004 1.890 11.755 2.087 -0.341(a) .289 B -0.796 0.101 -1.047 0.188 X X 𝐼 0 ↔SL 4.604 0.726 6.591 1.386 -.754 .352 𝐼 0 Variance 6.591 0.586 7.064 0.680 6.854 .672 SL Variance 6.092 1.247 8.207 2.776 .760 .287 Note.a = p > .05; controlled motivation slope p = .238 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 155 Table 18 Model fit indices for rank of health goal, conflict/facilitation with exercise composite, and conflict/facilitation with not exercising composite Variables Model # Models Goodness of fit indices Change statistics compared to prior model Χ 2 DF p-value CFI/TLI RMSEA ΔΧ 2 (Δdf) p-value of Δ Rank of health goal cluster 11a No change 34.998 31 0.284 .989/.990 0.019 11b Add β 34.622 30 0.257 .987/.987 0.021 0.376 (1) .540 11c Add slope 29.637 28 0.381 .996/.995 0.013 4.985 (2) .083 11d Add β and slope 29.23 27 0.35 .994/.993 0.015 0.407 (1) .524 11e Free slope across groups 25.406 26 0.496 1.000/1.002 <.001 3.824 (1) .051 11f Free β across groups 24.516 25 0.49 1.000/1.002 <.001 0.89 (1) .346 11g Free cov across groups 23.927 24 0.466 1.000/1.000 <.001 0.589 (1) .443 11h Free slope & intercept variance across groups 22.673 23 0.48 1.000/1.001 <.001 1.254 (2) .534 Conflict/ facilitation with exercising 12a No change 57.466 31 0.003 .892/.895 0.049 12b Add β 55.992 30 0.003 .894/.894 0.049 1.474 (1) .225 12c Add slope 53.989 28 0.002 .894/.886 0.051 2.003 (2) .367 12d Add β and slope 48.515 27 0.007 .912/.902 0.047 5.474 (1) .019* 12e Free slope across groups 45.668 26 0.01 .920/.907 0.046 2.847 (1) .092 12f Free β across groups 45.463 25 0.007 .916/.900 0.048 0.205 (1) .651 12g Free cov across groups 42.878 24 0.1 .923/.904 0.047 2.585 (1) .108 12h Free slope & intercept variance across groups 39.547 23 0.017 .932/.912 0.045 3.331 (2) .189 Conflict/ Facilitation with not exercising 13a No change 84.386 31 <.001 .817/.823 0.07 13b Add β 82.02 30 <.001 .822/.822 0.07 2.366 (1) .124 13c Add slope 76.937 28 <.001 .832/.820 0.07 5.083 (2) .079 13d Add β and slope 71.365 27 <.001 .848/.831 0.068 5.572 (1) .018* 13e Free slope across groups 70.973 26 <.001 .846/.822 0.07 0.392 (1) .531 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 156 13f Free β across groups 70.671 25 <.002 .843/.812 0.072 0.302 (1) .583 13g Free cov across groups 60.969 24 <.001 .873/.842 0.066 9.702 (1) .002** 13h Free slope & intercept variance across groups 59.946 23 <.001 .873/.835 0.067 1.023 (2) .560 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 157 Table 19 Parameter estimates for best fitting model for rank of health goal, conflict/facilitation with exercise composite, and conflict/facilitation with not exercising composite Rank of health goal Conflict/facilitation with exercise Conflict/facilitation with not exercising Model description No growth model; Groups invariant (Model 11a) No growth model; Groups invariant (Model 12a) Dual change model; covariance free across groups (Model not in table 18) Control Group Intervention Group Parameters M S M S M S M S Mean 𝐼 0 4.968 .221 49.164 3.641 -23.355 5.646 = = Mean SL X X X X -23.847 6.074 = = B X X X X -0.974 .158 = = 𝐼 0 ↔SL X X X X 604.523 163.357 878.242 162.957 𝐼 0 Variance 1.770 .147 444.534 40.224 1080.203 107.209 = = SL Variance X X X X 765.385 252.127 = = SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 158 Table 20 Model fit indices for conflict with exercise, facilitation with exercise, conflict with not exercising, and facilitation with not exercise Variables Model # Models Goodness of fit indices Change statistics compared to prior model Χ 2 DF p-value CFI/TLI RMSE A ΔΧ 2 (Δdf) p-value of Δ Facilitation with exercise 14a No change 48.327 31 0.024 .943/.945 0.04 14b Add β 48.324 30 0.018 .940/.940 0.94 0.003 (1) .956 14c Add slope 43.45 28 0.031 .950/.946 0.039 4.874 (2) .087 14d Add β and slope 38.488 27 0.07 .962/.958 0.035 4.962 (1) .026* 14e Free slope across groups 34.384 26 0.126 .973/.968 0.03 4.104 (1) .043* 14f Free β across groups 34.016 25 0.108 .971/.965 0.032 0.368 (1) .544 14g Free cov across groups 33.898 24 0.087 .968/.960 0.034 0.118 (1) .731 14h Free slope & intercept variance across groups 33.863 23 0.067 .965/.954 0.036 0.035 (2) .983 Conflict with exercise 15a No change 114.746 31 <.001 0.00/.015 0.087 15b Add β 107.08 30 <.001 .051/.019 0.087 7.666 (1) .006** 15c Add slope 111.484 30 <.001 .010/.010 0.087 -4.404 (0) X 15d Add β and slope 95.085 27 <.001 .173/.081 0.084 16.399 (3) <.001*** 15e Free slope across groups 93.987 26 <.001 .174/.047 0.086 1.098 (1) .295 15f Free β across groups 93.894 25 <.001 .163/.004 0.088 0.093 (1) .760 15g Free cov across groups X X X X X X X 15h Free slope & intercept variance across groups X X X X X X X Facilitation with not exercising 16a No change 175.248 31 <.001 .217/.242 0.114 16b Add β 164.951 30 <.001 .267/.267 0.112 10.297 (1) .001** 16c Add slope 171.465 30 <.001 .232/.232 0.115 -6.514 (0) X 16d Add β and slope 163.814 29 <.001 .268/.243 0.114 7.651 (1) .006** 16e Free slope across groups 162.666 28 <.001 .269/.217 0.116 1.148 (1) .284 16f Free β across groups 157.722 27 <.001 .290/.212 0.117 4.944 (1) .026* 16g Free cov across groups X X X X X X X SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 159 16h Free slope & intercept variance across groups X X X X X X X Conflict with not exercising 17a No change 66.441 31 <.001 .898/.901 0.057 17b Add β 66.411 30 <.001 .895/.895 0.058 0.030 (1) .863 17c Add slope 64.663 28 <.001 .895/.887 0.061 1.748 (2) .417 17d Add β and slope 51.596 27 0.003 .929/.922 0.051 13.067 (1) <.001*** 17e Free slope across groups 51.568 26 0.002 .927/.915 0.053 0.028 (1) .867 17f Free β across groups 46.936 25 0.005 .937/.924 0.05 4.632 (1) .031* 17g Free cov across groups 44.19 24 0.007 .942/.928 0.049 2.746 (1) .098 17h Free slope & intercept variance across groups 43.96 23 0.005 .940/.921 0.051 0.23 (2) .891 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 160 Table 21 Parameter estimates for best fitting model for conflict with exercise, facilitation with exercise, conflict with not exercising, and facilitation with not exercise Facilitation with exercise Conflict with exercise Facilitation with not exercising Conflict with not exercising Model Description Dual changes model; Linear change free across groups (Model 14e) Dual change model; Groups invariant (Model 15d) Proportional change model; Groups invariant (Model 16b) Dual change model; Groups invariant (Model 17d) Control Intervention Parameters M S M S M S M S M S Mean 𝐼 0 58.383 3.771 = = 3.604 .099 4.227 .205 6.206 .261 Mean SL 49.564 16.095 46.061 15.848 5.805 .885 6.332 .837 B -.828 .259 = = -1.607 .235 -.090 .027 -1.042 .132 𝐼 0 ↔SL 282.372 105.494 = = .222 .056 1.851 .325 𝐼 0 Variance 414.605 48.563 = = .106 .039 1.182 .124 2.434 .224 SL Variance 285.645 153.998 = = .521 .156 2.231 .559 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 161 Table 22 Model fit indices for future vigorous exercise goals and future vigorous/moderate exercise goals Variables Model # Models Goodness of fit indices Change statistics compared to prior model Χ 2 DF p-value CFI/TLI RMSEA ΔΧ 2 (Δdf) p-value of Δ Future Vigorous Exercise goals (MET Units) 18a No change 85.702 31 <.001 .929/.932 0.07 18b Add β 66.329 30 <.001 .953/.953 0.058 19.373 (1) <.001*** 18c Add slope 83.143 30 <.001 .931/.931 0.07 -16.814 (0) x 18d Add β and slope 59.263 29 0.001 .961/.960 0.054 23.88 (1) <.001*** 18e Free slope across groups 58.504 28 0.001 .961/.958 0.055 0.759 (1) .384 18f Free β across groups 58.457 27 <.001 .959/.955 0.057 0.047 (1) .838 18g Free cov across groups X X X X X X X 18h Free slope & intercept variance across groups X X X X X X X Future Vigorous/ Moderate Exercise Goals (MET Units) 19a No change 103.277 31 <.001 .876/.880 0.081 19b Add β 72.785 30 <.001 .927/.927 0.063 30.492 (1) <.001*** 19c Add slope 102.488 30 <.001 .876/.876 0.082 -29.703 (0) X 19d Add β and slope 69.778 29 <.001 .930/.928 0.063 32.71 (1) <.001*** 19e Free slope across groups 66.628 28 <.001 .934/.929 0.062 3.15 (1) .076 19f Free β across groups 66.567 27 <.001 .932/.925 0.064 0.061 (1) .805 19g Free cov across groups X X X X X X X 19h Free slope & intercept variance across groups X X X X X X X SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 162 Table 23 Parameter estimates for best fitting model for future vigorous exercise goals and future vigorous/moderate exercise goals Intentions for Future Vigorous Exercise Intention for Future Vigorous/Moderate Exercise Model description Dual change model; Groups invariant (Model 18d) Proportional change model; Groups invariant (Model 19b) Parameters M S M S Mean 𝐼 0 .579 .108 .279 .092 Mean SL -.224 .085 X X B .198 .044 .248 .050 𝐼 0 ↔SL X X X X 𝐼 0 Variance .283 .032 .277 .035 SL Variance X X X X Note. All variables standardized in order to run the models (using first time point mean and SD). Variances and covariance were restricted to zero for future vigorous exercise goals and future vigorous/moderate exercise goals. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 163 Table 24 Goal structure scores robustly correlated with exercise variables Significant relationships 3 out of 5 samples. Measure Time 1 Exercise Variables Time 2 Exercise Variables Time 3 Exercise Variables Ranking of health goal cluster -- Success V/M 2 (-.149) Success V/M 3(-.154) Success V/M 2 (-.167) Success V/M 2 (-.156) Facilitation/conflict exercise -- V 2 (.155) V 3 (.168) VM 2 (.207) VM 3 (.186) Success V/M 2 (.176) V 2 (.132) V 3 (.219) VM 1 (.146) VM 2 (.191) VM 3 (.227) Success V/M 2 (.173) Success V/M 3 (.175) Facilitation/conflict no exercise -- -- -- Conflict exercise -- Success V/M 3 (.188) -- Facilitation exercise VM 2 (.147) V 2 (.157) V 3 (.182) VM 2 (.196) VM 3 (.201) Success V/M 2 (.173) V 1 (.175) V 2 (.145) V 3 (.230) VM 1 (.168) VM 2 (.205) VM 3 (.238) Success V/M 2 (.143) Conflict no exercise VM 2 (.122) -- V 1 (.129) VM 1 (.135) Facilitation no exercise -- VM 3 (.134) Success V/M 2 (.109) -- Note. Numbers in parenthesis are averaged Pearson r correlations from significant relationships within the five samples of data (three 50% random samples, complete-cases sample, and the total sample). All variables in MET units. V = vigorous exercise; VM = vigorous/moderate exercise; Success V/M = success in accomplishing vigorous/moderate intentions for exercise; dashes SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 164 indicate no consistent correlations observed for that cell; subscripted numbers indicate time point in which variable was measured. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 165 Table 25 Results of regression analysis with exercise variables regressed onto the five remaining goal-structure scores (overall rank of health goal-cluster, facilitation with exercise, conflict with exercise, facilitation with not exercising, and conflict with not exercising) Significant relationships in 3 out of 5 samples Measures Time1 Goal-Structure Variables Time2 Goal-Structure Variables Time3 Goal-Structure Variables Time 1 exercise variables Vigorous -- XX XX Vigorous/moderate -- XX XX Time 2 exercise variables Vigorous -- Facilitation exercise (.166) XX Vigorous/moderate -- Facilitation exercise (.186) XX Success vigorous/ moderate intentions -- Rank of health goal-cluster (-.128) XX Time 3 exercise variables Vigorous -- Facilitation exercise (.202) Facilitation exercise (.275) Vigorous/moderate -- Facilitation exercise (.185) Facilitation exercise (.271) Success vigorous/ moderate intentions -- Conflict exercise (.200) Facilitation exercise (.158) Note. Numbers in parenthesis represent averaged standardized beta coeffecients across significant relationships from five samples (three 50% random samples, a complete-cases sample, and the total sample). Dashes indicate consistent significant relationships were not observed for that cell; XX indicates that the relationship was never tested due to temporal order of measurement. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 166 Table 26 A representative example of an exploratory factor analysis with an oblique rotation forcing a two-factor solution with goal-structure scores and motivation constructs Measures Factor loadings Factor 1 Factor 2 Autonomous motivation .723 Commitment .667 Facilitation exercise 𝑎 .524 Rank of health goal-cluster 𝑎 .490 Intention for vigorous exercise .465 Intention for vigorous/moderate exercise .314 Facilitation no exercise 𝑎 .821 Conflict no exercise 𝑎 .349 -.739 Conflict exercise 𝑎 .411 Controlled motivation Note. Loadings < .30 suppressed. 𝑎 Goal-structure scores. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 167 Table 27 A representative example of an exploratory factor analysis with an oblique rotation forcing a three-factor solution with goal-structure scores and motivation constructs Measures Factor Loadings Factor 1 Factor 2 Factor 3 Facilitation exercise 𝑎 .730 Autonomous motivation .683 Intention for vigorous exercise .517 Facilitation no exercise 𝑎 .909 Conflict no exercise 𝑎 .413 -.679 Conflict exercise 𝑎 .432 Commitment -.733 Controlled motivation .357 .650 Intention for vigorous/moderate exercise -.503 Rank of health goal cluster 𝑎 -.408 Note. Loadings < .30 suppressed. 𝑎 Goal-structure scores. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 168 Table 28 Correlations and descriptive statistics for measurement model analysis Mean SD Correlations Com 1 Com 2 Com 3 Aut 1 Aut 2 Aut 3 Fac 1 Fac 2 Fac 3 Rank 1 Rank 2 Rank 3 Com 1 20.63 3.24 -- .575 ** .492 ** .301 ** .324 ** .361 ** .119 ** .059 .166 ** .210 ** .183 ** .178 ** Com 2 19.94 3.51 -- .716 ** .361 ** .445 ** .476 ** .172 ** .190 ** .208 ** .273 ** .225 ** .193 ** Com 3 20.08 3.65 -- .355 ** .438 ** .454 ** .120 * .196 ** .204 ** .238 ** .192 ** .231 ** Aut 1 12.25 3.53 -- .581 ** .574 ** .273 ** .223 ** .268 ** .221 ** .180 ** .140 ** Aut 2 12.07 3.59 -- .684 ** .247 ** .297 ** .353 ** .179 ** .155 ** .143 ** Aut 3 12.47 3.58 -- .261 ** .239 ** .299 ** .187 ** .119 * .138 ** Fac 1 63.42 28.25 -- .454 ** .468 ** .090 * .150 ** .108 * Fac 2 58.75 26.59 -- .532 ** .103 * .103 * .024 Fac 3 59.81 29.58 -- .156 ** .161 ** .123 * Rank 1 6.10 1.99 -- .497 ** .459 ** Rank 2 6.18 1.95 -- .542 ** Rank 3 6.19 2.06 -- SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 169 Table 29 Numerical Results for Motivation Latent Variable Measurement Model Fitted to Whole Sample Across Three Time Points (n = 713) Rasch model (Model 20a) Weak invariance (Model 20b) Strong invariance (Model 20c) Strict invariance (Model 20d) M S M S M S M S Fixed parameters Mean LV1 0.00 — 0.00 — 20.446 .117 20.444 .118 Mean LV2 0.00 — 0.00 — 19.951 .136 19.946 .135 Mean LV3 0.00 — 0.00 — 20.141 .141 20.128 .140 Random parameters Variance LV1 .383 .057 3.095 .493 3.469 .514 3.662 .518 Variance LV2 .409 .064 4.120 .715 4.736 .763 4.655 .755 Variance LV3 .479 .077 4.391 .789 5.061 .845 5.115 .864 Factor loadings invariant over time for each model Commitment 1.00 — 1.000 — 1.000 — 1.000 — Autonomous 1.00 — 1.275 .153 1.123 .129 1.093 .126 Facilitation w/ exercise 1.00 — .185 .024 .176 .023 .177 .024 Rank of health goal 1.00 — 2.89 .043 .269 .039 .276 .039 Goodness of fit indices 𝜒 2 /df 294.309/48 76.149/45 106.981/51 119.120/59 CFI/TLI .875/.827 .984/.977 .971/.963 .969/.966 RMSEA .085 .031 .039 .038 Change in fit compared to previous model Δ𝜒 2 /Δdf — 218.160/3* 30.832/6* 12.139/8 M = maximum likelihood estimate; S = standard errors; LV = motivation latent variable. *indicates significant change in model fit (p < .05; ΔCFI > .01) SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 170 Table 30 Numerical Results for Motivation Latent Variable Measurement Model Fitted Across Study Group and Gender (n = 713) Split by gender Split by study group Model 21a Model 21b Model 22a Model 22b Males (n = 259) Females (n = 454) Groups invariant Intervention (n = 349) Control (n = 364) Groups invariant M S M S M S M S M S M S Factor loadings (invariant over time) Commitment 1.00 — 1.00 — 1.00 — 1.00 — 1.00 — 1.00 — Autonomous motivation .837 .405 .349 .312 .615 .271 .929 .209 .109 .137 .655 .265 Facilitation of exercise .153 .078 .153 .124 .191 .086 .258 .062 .047 .046 .196 .082 Rank of health goal-cluster .183 .120 .171 .109 .208 .095 .254 .088 .114 .087 .235 .099 Indicator means Commitment 0 -- 0 -- 0 -- 0 -- 0 -- 0 -- Autonomous motivation -4.602 8.207 5.096 6.324 -.224 5.486 -6.571 4.246 9.983 2.777 -1.044 5.368 Facilitation of exercise -3.213 1.581 -3.145 2.501 -3.930 1.740 -5.240 1.250 -1.071 .938 -4.037 1.665 Rank of health goal-cluster 2.302 2.438 2.733 2.211 1.913 1.917 1.010 1.784 3.774 1.770 1.367 1.997 Error terms Commitment 8.542 .955 10.775 1.009 10.215 .655 9.112 .712 10.852 .807 10.308 .631 Autonomous motivation 10.735 1.017 12.421 .756 11.831 .617 10.024 .768 13.287 .814 11.761 .615 Facilitation of exercise .860 .064 .987 .065 .934 .047 .832 .058 1.014 .057 .932 .046 Rank of health goal-cluster 3.948 .274 3.852 .202 3.886 .164 4.016 .244 3.767 .215 3.875 .165 Goodness of fit indices 𝜒 2 /df 480.786 / 130 497.949 / 140 486.758 / 130 520.885 / 140 CFI/TLI .822/.820 .819/.829 .822/.819 .810/.821 RMSEA .087 .085 .088 .087 Note. Facilitation with exercise standardized using Time 1 mean and variance. Latent variable means and variances invariant across groups. Latent variable covariances constrained to zero. M = maximum likelihood estimate; S = standard errors. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 171 Table 31 Path analysisusing Time 1 cognitions for predicting Time 2 vigorous exercise with and without self-efficacy and outcome expectations in the model No regression Intentions regression Motivation regression Both regression No latent variable Models without self-efficacy and outcome expectations Model 23a Model 23b Model 23c Model 23d Model 23e Goodness of fit indices 𝜒 2 /df 120.006 / 10 40.780 / 9 68.363 / 9 28.274 / 8 177.939 / 6 CFI/TLI .660 / .490 .902 / .836 .816 / .694 .937 / .882 .468 / -.330 RMSEA .124 .070 .096 .060 .200 Models including self-efficacy and outcome expectations Model 24a Model 24b Model 24c Model 24d Model 24e Goodness of fit indices 𝜒 2 /df 210.18 / 16 149.965 / 15 157.102 / 15 144.934 / 14 56.982 / 6 CFI/TLI .783/.619 .849/.717 .841 / .702 .843 / .706 .943 / .733 RMSEA .130 .112 .115 .115 .109 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 172 Table 32 Path analysis using Time 1 cognitions for predicting Time 2 vigorous/moderate exercise with and without self-efficacy and outcome expectations in the model No regression Intentions regression Motivation regression Both regression No latent variable Models without self-efficacy and outcome expectations Model 25a Model 25b Model 25c Model 25d Model 25e Goodness of fit indices 𝜒 2 /df 137.782 / 10 69.247 / 9 61.463 / 9 43.944 / 8 177.893 / 6 CFI/TLI .645 / .467 .883 / .721 .854 / .757 .900 / .813 .522 / -.194 RMSEA .134 .097 .009 .079 .200 Models including self-efficacy and outcome expectations Model 26a Model 26b Model 26c Model 26d Model 26e Goodness of fit indices 𝜒 2 /df 208.257 / 16 164.946 / 15 159.420 / 15 151.986 / 14 56.996 / 6 CFI/TLI .796/.643 .841/.703 .847 / .714 .853 / .707 .946 / .747 RMSEA .130 .118 .116 .118 .109 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 173 Table 33 Path analysis using Time 1 cognitions for predicting Time 2 success in accomplishing vigorous/moderate exercise intentions with and without self-efficacy and outcome expectations in the model No regression Intentions regression Motivation regression Both regression No latent variable Models without self-efficacy and outcome expectations Model 27a Model 27b Model 27c Model 27d Model 27e Goodness of fit indices 𝜒 2 /df 61.402 / 10 59.837 / 9 45.012 / 9 43.663 / 8 177.893 / 6 CFI/TLI .819 / .728 .821 / .701 .873 / .788 .874 / .764 .393 / -.516 RMSEA .085 .089 .075 .079 .200 Models including self-efficacy and outcome expectations Model 28a Model 28b Model 28c Model 28d Model 28e Goodness of fit indices 𝜒 2 /df 160.297 / 16 610.020 / 15 150.014 / 15 149.348 / 14 56.996 / 6 CFI/TLI .834/.710 .833/.689 .845 / .710 .844 / .689 .941 / .727 RMSEA .112 .116 .112 .116 .109 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 174 Table 34 Path analysis using Time 2 cognitions for predicting Time 3 vigorous exercise with and without self-efficacy and outcome expectations in the model No regression Intentions regression Motivation regression Both regression No latent variable Models without self-efficacy and outcome expectations Model 29a Model 29b Model 29c Model 29d Model 29e Goodness of fit indices 𝜒 2 /df 194.553 / 10 68.589 / 9 97.789 / 9 54.627 / 8 192.782 / 6 CFI/TLI .723/.475 .920/.840 .874/.748 .941/.874 .705/-.770 RMSEA .114 .063 .079 .056 .209 Models including self-efficacy and outcome expectations Model 30a Model 30b Model 30c Model 30d Model 30e Goodness of fit indices 𝜒 2 /df 183.897 / 16 108.701 / 15 124.107 / 15 104.372 / 14 58.854 / 6 CFI/TLI .860/.693 .926/.830 .912/.799 .929/.829 .954/.575 RMSEA .094 .070 .076 .070 .111 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 175 Table 35 Path analysis using Time 2 cognitions for predicting Time 3 vigorous/moderate exercise with and without self-efficacy and outcome expectations in the model No regression Intentions regression Motivation regression Both regression No latent variable Models without self-efficacy and outcome expectations Model 31a Model 31b Model 31c Model 31d Model 31e Goodness of fit indices 𝜒 2 /df 152.418 / 10 47.854 / 9 54.667 / 9 29.312 / 8 193.006 / 6 CFI/TLI .671/.506 .910/.850 .894/.824 .951/.908 .567/-.081 RMSEA .141 .078 .084 .061 .209 Models including self-efficacy and outcome expectations Model 32a Model 32b Model 32c Model 32d Model 32e Goodness of fit indices 𝜒 2 /df 149.515 / 16 95.053 / 15 107.318 / 15 90.47 / 14 50.193 / 6 CFI/TLI .856/.749 .914/.839 .901/.815 .918/.835 .952/.778 RMSEA .108 .087 .093 .088 .102 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 176 Table 36 Path analysis using Time 2 cognitions for predicting Time 3 success in accomplishing intentions for vigorous/moderate exercise with and without self-efficacy and outcome expectations in the model No regression Intentions regression Motivation regression Both regression No latent variable Models without self-efficacy and outcome expectations Model 33a Model 33b Model 33c Model 33d Model 33e Goodness of fit indices 𝜒 2 /df 50.782 / 10 46.255 / 9 28.293 / 9 27.176 / 8 194.855 / 6 CFI/TLI .877/.816 .888/.813 .942/.903 .942/.892 .431/-.422 RMSEA .076 .076 .055 .058 .210 Models including self-efficacy and outcome expectations Model 34a Model 34b Model 34c Model 34d Model 34e Goodness of fit indices 𝜒 2 /df 94.177 / 16 94.172 / 15 87.446 / 15 86.570 / 14 50.518 / 6 CFI/TLI .906/.835 .905/.822 .913/.837 .913/.825 .946/.750 RMSEA .083 .086 .082 .085 .102 SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 177 Figures Figure 1. A priori path analysis comparing the motivation latent variable to intentions with self- efficacy and outcome expectations included in the model. Motiv = motivation latent variable; facil with exercise = facilitation with exercise. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 178 Figure 2. Representation of classical test theory, in which all measured variables (Z[1])are a combination of an unobserved (latent) true score (z[1]) and an error term (e[1]). SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 179 Figure 3. An auto-regression model (Figure 3.a) in which a score on Variable z at time 2 (z[2]) is regressed on variable z at Time 1 (z[1]) creating an error term (e[2]) associated with the prediction of z[2], which can then be altered to a change score (Figure 3.b) by setting the path from z[1] to z[2] equal to one. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 180 Figure 4. Three time points of data z[t] auto-regressed creating latent change scores at Time 2 (DZ[2]) and Time 3 (DZ[3]) with associated error terms (DE[2] and DE[3]). SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 181 Figure 5. Adding group-level latent variables into the change-score model representing the intercept (i.e., the baseline mean; I[0]) and linear change over time (slope). SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 182 Figure 6. Adding a proportional change element to the model (β) by regressing change scores on the score from the previous time point, resulting in a dual change score model calculating linear change (slope) and proportional change over time. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 183 Figure 7. A multi-group dual-change score model testing differences across Intervention and Control groups. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 184 Figure 8. Measurement model testing robustness of motivation latent variable across three time points. Com = commitment; Aut = autonomous motivation; Rank = overall rank of the health goal-cluster; Facil = Goal facilitation with exercise; subscripted numerals indicate time-point of measurement. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 185 Figure 9. Example path analysis testing the relationships between exercise, intentions, motivation, self-efficacy, and outcome expectations. Figure 1a restricts pathways from intentions and exercise and from motivation and exercise to zero. Figure 1b allows the pathway from intentions and exercise to be estimated, while restricting the pathway between motivation and exercise to zero. Figure 1c allows the pathway from motivation and exercise to be estimated, while restricting the pathway between intentions and exercise to zero. Model 1d allows both pathways from intentions and motivation to exercise to be estimated. Model 1e removes the motivation latent variable from the model allowing the indicator variables to be directly associated with the other variables in the model. Motiv = motivation latent variable; Facil with exercise = facilitation with exercise; Vig Ex Time 2 = Vigorous exercise at time 2. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 186 Figure 10. Flow diagram of participants through the recruitment, screening, and assessment phases of the intervention. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 187 Figure 11. Measurement model with values for strict invariance on whole sample (Model 20d). Com = commitment; Aut = autonomous motivation; Rank = overall rank of the health goal cluster; Facil = facilitation with exercise; subscripted numerals indicate time point of measurement. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 188 Figure 12. Series of models tested including only motivation, intentions, and exercise.Motiv = motivation latent variable; Facil with exercise = facilitation with exercise. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 189 Figure 13. Path analysis including social cognitive variables in addition to intentions, motivation, and exercise. Motiv = motivation latent variable; Facil with exercise = facilitation with exercise. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 190 Figure 14. Path analysis using Time 1 cognitions for predicting Time 2 vigorous exercise, without self-efficacy or outcome expectations in the model.Motiv = motivation latent variable; Facil with exercise = facilitation with exercise; Vig Ex Time 2 = Vigorous exercise at time 2. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 191 Figure 15. Path analysis using Time 1 cognitions for predicting Time 2 vigorous exercise with self-efficacy and outcome expectations in the model. Motiv = motivation latent variable; Vig ex 2 = Time 2 vigorous exercise; Facil with exercise = facilitation with exercise. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 192 Figure 16. Path analysis using Time 1 cognitions for predicting Time 2 vigorous/moderate exercise, without self-efficacy or outcome expectations in the model. Motiv = motivation latent variable; VM ex 2 = Time 2 vigorous/moderate exercise; Facil with exercise = facilitation with exercise. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 193 Figure 17. Path analysis using Time 1 cognitions for predicting Time 2 vigorous/moderate exercise with self-efficacy and outcome expectations in the model. Motiv = motivation latent variable; VM ex 2 = Time 2 vigorous/moderate exercise; Facil with exercise = facilitation with exercise. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 194 Figure 18. Path analysis using Time 1 cognitions for predicting success in accomplishing intentions for vigorous/moderate exercise at Time 2, without self-efficacy or outcome expectations in the model. Motiv = motivation latent variable; VM success 2 = success in accomplishing intentions for vigorous/moderate exercise at Time 2; Facil with exercise = facilitation with exercise. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 195 Figure 19. Path analysis using Time 1 cognitions for predicting success in accomplishing intentions for vigorous/moderate exercise at Time 2 with self-efficacy and outcome expectations in the model. Motiv = motivation latent variable; VM Success 2 = success in accomplishing intentions for vigorous/moderate exercise at Time 2; Facil with exercise = facilitation with exercise. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 196 Figure 20. Path analysis using Time 2 cognitions for predicting Time 3 vigorous exercise, without self-efficacy or outcome expectations in the model. Motiv = motivation latent variable; Vig ex 3 = Time 3 vigorous exercise; Facil with exercise = facilitation with exercise. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 197 Figure 21. Path analysis using Time 2 cognitions for predicting Time 3 vigorous exercise with self-efficacy and outcome expectations in the model. Motiv = motivation latent variable; Vig ex 3 = Time 3 vigorous exercise; Facil with exercise = facilitation with exercise. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 198 Figure 22. Path analysis using Time 2 cognitions for predicting Time 3 vigorous/moderate exercise, without self-efficacy or outcome expectations in the model. Motiv = motivation latent variable; VM ex 3 = Time 3 vigorous/moderate exercise; Facil with exercise = facilitation with exercise. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 199 Figure 23. Path analysis using Time 2 cognitions for predicting Time 3 vigorous/moderate exercise with self-efficacy and outcome expectations in the model. Motiv = motivation latent variable; VM ex 3 = Time 3 vigorous/moderate exercise; Facil with exercise = facilitation with exercise. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 200 Figure 24. Path analysis using Time 2 cognitions for predicting success in accomplishing intentions for vigorous/moderate exercise at Time 3, without self-efficacy or outcome expectations in the model. Motiv = motivation latent variable; VM success 3 = success in accomplishing intentions for vigorous/moderate exercise at Time 3; Facil with exercise = facilitation with exercise. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 201 Figure 25. Path analysis using Time 2 cognitions for predicting success in accomplishing intentions for vigorous/moderate exercise at Time 3 with self-efficacy and outcome expectations in the model. Motiv = motivation latent variable; VM Success 3 = success in accomplishing intentions for vigorous/moderate exercise at Time 3; Facil with exercise = facilitation with exercise. SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 202 Appendix A Ranking procedure The instructions for the ranking procedure were as follows: Below is a list of 9 major life goals. Not all of these goals are equally important to every person. We want you to use your mouse to drag the goals into a rank order that shows their relative importance to you! Find the goal that is most important and drag it to the top of the list. Drag the goal that is second most important to you, to the second position in the list, etc. Once you have a final ranking for how important these 9 goals are to you, then click the continue button. The nine goal goals participants ranked were: (1) being a moral and virtuous person (e.g., sticking to personal morals, helping others, being highly regarded); (2) Religion and spirituality; (3) Self-fulfillment and being open to new experiences (e.g., gaining wisdom, appreciating beauty, embracing life); (4) Avoiding negative social experiences (e.g., self-protection, avoiding rejection or conflict); (5) Good social relationships (e.g., intimacy, belonging, influencing others); (6) Good family relationships (e.g., close to parents, being a good family member); (7) Being intelligent and skillful (e.g., intellectual growth, being autonomous, competent); (8) Having financial and occupational success (e.g., financial freedom, wealth, a respected job); and (9) Being physically healthy (e.g., being active, capable of daily tasks, physically fit). Goal structure explanation The explanation of goal-structural in lay terms was as follows: When beginning a new behavior such as exercise, psychologists find that it is important to understand how that behavior interacts with the other things in your life that you SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 203 devote attention, time, and effort to. That is, how will exercising affect you in accomplishing the other major life goals that you sorted on the previous screen? In general, people consider most of the nine broad goals as highly important. However, people have limited amounts of time, energy, and resources to accomplish all the things that they care about. Therefore, psychologists find that the three goals people choose as their most important often take up the majority of people’s time and effort. Research finds it is very important to consider how starting a new behavior like exercise will influence you in achieving the three goals that you ranked as most important to you. On the next few pages, we’ve listed the three goals that you ranked number 1, number 2, and number 3. For each goal, please give AT LEAST one example of how increasing your exercise will help you in accomplishing that goal. Feel free to provide more examples, typically the more the behavior will enable you in achieving your most important goals, the more likely you will be able to make the behavior a permanent part of your life. Examples provided to participants 1. The goal you ranked as most important was: "Being a moral and virtuous person (e.g., sticking to personal morals, helping others, being highly regarded)." For example, other people who consider this an important life goal report that: "Being physically fit and physically able is important to me because it is my duty to take care of SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 204 my body, and doing so will gain respect of others and give me the ability and energy to help others." 2. The goal you ranked as most important was: "Religion and spirituality" For example, other people who consider this an important life goal report that: "Being physically fit and healthy through regular exercise will give me peace of mind and stamina to pursue my religious beliefs and spirituality." 3. The goal you ranked as your second most important was: "Self-fulfillment and being open to new experiences (e.g., wisdom, appreciating beauty, embracing life)" For example, other people who consider this an important life goal report that: "Being healthy and physically fit through regular exercise will allow me to engage in any life experiences that come my way so that I don’t miss out on interesting activities or opportunities." 4. The goal you ranked as your second most important was: "Avoiding negative social experiences (e.g., self-protection, avoiding rejection or conflict)." For example, other people who consider this an important life goal report that: "Being physically fit through exercise will help discourage social aggression from others like being ostracized or criticized for my appearance, and it will help me to protect myself against others if necessary." 5. The goal you ranked as your second most important was: "Good social relationships (e.g., intimacy, belonging, holding power over others)." For example, other people who consider this an important life goal report that: "Being physically fit through exercise will make me attractive and desirable to others, and overall will make me a more respected and influential person." SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 205 6. The goal you ranked as your second most important was: "Good family relationships (e.g., close to parents, being a good family member)." For example, other people who consider this an important life goal report that: "Being physically fit through regular exercise will help me interact with and support my close family members without limitations." 7. The goal you ranked as your second most important was: "Being intelligent and skillful (e.g., intellectual growth, being autonomous, competence)." For example, other people who consider this an important life goal report that: "Being physically fit through regular exercise will allow me to think clearly, function independently, and accomplish the life tasks I care about." 8. The goal you ranked as your second most important was: "Having financial and occupational success. (e.g., financial freedom, wealth, respected job)." For example, other people who consider this an important life goal report that: "Being healthy through regular exercise will give me more energy to work hard, help me avoid missing work due to illness, and be more respected by my colleagues, all of which allowing me to generate wealth and succeed at my career." 9. The goal you ranked as your second most important was: "Being physically healthy (e.g., being active, capable of daily tasks, physically fit)." For example, other people who consider this an important life goal report that: "Regular exercise will make me a healthy and physically fit person capable of completing the life tasks I undertake, and it will reduce the chances I will suffer from consequences associated with bad health." SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 206 Planning portion of intervention The instructions for the planning portion of the intervention were as follows: Perhaps the best way to minimize the degree that the time and energy required to exercise conflicts with achieving your most important goals is to make specific plans on how you will implement your exercise. Specific plans include determining where you will exercise, when you will exercise, how long you will exercise, and what kind of exercise you will do. Planning your exercise in advance allows you to minimize the degree that it conflicts with your other important goals, and it can perhaps even help you in accomplishing those goals as well depending on the plans that you make. Research finds that the more plans you make, and the more specific your plans are, the more likely you will be able to incorporate exercise into your life permanently. Keeping in mind the three goals that you chose as your most important life goals, we would like you to create plans for how to implement your exercise in order to: (a) minimize the degree your exercise will conflict with those goals, and (b) maximize the degree your exercise will facilitate those goals. Remember, the more detailed and specific your plans are, the better they will be able to help you accomplish your intentions to increase your exercise. Participants then answered the following four questions specifying “where,” “when,” “what type,” and “how long” they planned to complete their exercise in order to minimize the SOCIAL-COGNITIVE MEASURES OF MOTIVATION TO EXERCISE 207 degree their exercise would conflict with their three most important goals, or maximize the degree it would facilitate them in accomplishing those goals. 1. WHERE do you plan to exercise (e.g,. at home, at a gym, multiple places, etc.,)? 2. WHEN do you plan to do your exercise (e.g., before work on weekdays, immediately after work as many days as I can, etc.,)? 3. WHAT TYPE OF EXERCISES do you plan to do (e.g., walk, lift weights, play a sport, dance classes, etc.,)? 4. HOW LONG do you plan to exercise (e.g., at least 30 minutes a day, 45 minutes when walking and 25 minutes when jogging, etc.,)? Finally, the plans participants specified were repeated for their convenience while they stated how their plans would conflict with or facilitate them in achieving their three most important goals (each goal, one at a time). For example, “How will this plan help you minimize conflict or maximize facilitation of the goal: Being a moral and virtuous person (e.g., sticking to personal morals, helping others, being highly regarded).”

Abstract (if available)

Abstract Overweight and obesity are major problems in developed countries around the world, and exercise is strongly associated with weight loss and weight control. Social‐cognitive models have been used to study self‐regulation of health behaviors, but they suffer from several limitations, including the use of intentions as a sole measure of motivation. There are validated motivational constructs from literature outside the health domain that could potentially combine to form a latent variable of motivation that would benefit social‐cognitive models of self‐regulation. It would provide a more accurate assessment of motivation and provide direct avenues of intervention that have yet to be fully explored in health research. On an adjusted longitudinal sample of 713 participants attempting to increase their vigorous and/or moderate exercise as part of a New Year’s Resolution, the present study had four aims. First, it utilized a randomized controlled design to test a novel intervention designed to increase exercise behavior based on the motivational concept goal‐structure. Second, it evaluated the Chulef, Read, and Walsh (2001) goal taxonomy for assessing participant goal‐structure in regards to exercise. Third, it evaluated whether three motivational constructs—goal autonomy, goal commitment, and goal‐structure—form a robust latent variable. Finally, it tested whether the latent variable of motivation predicted exercise behavior over and above intentions. There were three time‐points of data collection that started in January and occurred at 4‐week intervals. Intervention effects immediately post intervention were tested via independent samples t‐tests, and changes over time were tested via multi‐group latent change score models. The intervention successfully impacted participant goal‐structure post‐intervention, however these effects failed to spill over into other constructs. Additionally, both Intervention and Control groups significantly increased their physical activity throughout the course of the study, as well as their perceptions of their intentions for exercise, goal commitment, goal autonomy, and a few goal‐structure scores, however there were few significant differences observed across groups. The utility of the Chulef, Read, and Walsh (2001) goal taxonomy for measuring goal structure in relation to exercise was tested using correlations, regressions, and exploratory factor analyses on five sub‐samples of data. These analyses revealed that two goal‐structure scores—(1) the degree exercise facilitates achieving other life goals and (2) the rank of a heath goal‐cluster in comparison to other goal‐clusters—stood out as potential useful additions to a latent variable measuring motivation. The robustness of the motivation latent variable derived from the motivation constructs was evaluated using longitudinal measurement invariance modeling testing for stable latent variable factor loadings, indicator means, and error variances over time and across gender and study group. The tests of measurement invariance showed that a single latent variable comprised of four others (goal commitment, goal autonomy, facilitation of exercise, and the rank of the heath goal-cluster) exhibited a good fit to the data across three time‐points of measurement. Finally, a latent variable path analysis was used to test if the motivation latent variable predicted exercise behavior over and above intentions. After adjustment for missing data, the motivation latent variable at Time 1 predicted vigorous exercise at Time 2, vigorous/moderate exercise at Time 2, and success in accomplishing vigorous/moderate exercise intentions at Time 2, and the motivation latent variable at Time 2 significantly predicted vigorous exercise at Time 3, vigorous/moderate exercise at Time 3, and success in accomplishing vigorous/moderate exercise intentions at Time 3. Although the intervention failed to create significant differences in exercise across study groups, the findings support using at least some of the motivation constructs within the traditional social‐cognitive frameworks. Future research should continue exploring intervention options based on the motivation constructs, update the method for measuring goal‐structure using the Chulef, Read, and Walsh (2001) goal taxonomy, and continue testing the relationships under study.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

A measure of weight management strategies and evidence for its utility

PDF

A functional use of response time data in cognitive assessment

PDF

Latent change score analysis of the impacts of memory training in the elderly from a randomized clinical trial

PDF

A "second chance" program for vulnerable young adults: a program evaluation and a randomized controlled trial of adjunctive motivational interviewing to improve retention

PDF

Assessing the psychological correlates of belief strength: contributing factors and role in behavior

PDF

Goals and weight: the interplay of goal perceptions in weight management

PDF

Sources of stability and change in the trajectory of openness to experience across the lifespan

PDF

A closer look at intention to use condoms and risky sexual behaviors among homeless people: application of the theories of reasoned action and planned behavior

PDF

Longitudinal relationships between bereavement and physical health‐related quality of life in middle‐ to older‐aged women

PDF

Developmental etiology of reactive and proactive aggression, electrodermal reactivity and their relationships: a twin study

PDF

Using ecological momentary assessment to study the impact of social-cognitive factors on paretic hand use after stroke

Asset Metadata

Creator Larsen, Andrew L. (author)

Core Title Evaluating social-cognitive measures of motivation in a longitudinal study of people completing New Year's resolutions to exercise

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Psychology

Publication Date 04/28/2015

Defense Date 03/04/2015

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

Tag autonomous motivation,Exercise,Motivation,OAI-PMH Harvest,physical activity,self‐efficacy,self‐regulation

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Walsh, David A. (committee chair), McArdle, John J. (committee member), Rueda, Robert (committee member), Wood, Wendy (committee member)

Creator Email allarsen@usc.edu,Larsena3@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-561896

Unique identifier UC11299475

Identifier etd-LarsenAndr-3404.pdf (filename),usctheses-c3-561896 (legacy record id)

Legacy Identifier etd-LarsenAndr-3404.pdf

Dmrecord 561896

Document Type Dissertation

Format application/pdf (imt)

Rights Larsen, Andrew L.

Type texts

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

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA