Perceptions of Change and Certainty Regarding ... - Semantic Scholar

Journal of Sport & Exercise Psychology, 2011, 33, 710-733 © 2011 Human Kinetics, Inc.

Perceptions of Change and Certainty Regarding the Self-as-Exerciser: A Multistudy Report Sean P. Mullen University of Virginia The purpose of these studies was to examine the relationship between perceptions of exercise-related changes (i.e., perceived mastery and physical change) and certainty with regard to the self-as-exerciser. It was hypothesized that seeing “change” would be associated with more favorable levels of exercise self-certainty and behavior relative to “no change.” Online surveys were repeatedly administered across 4 months (Study 1) and 4 weeks (Study 2) to 196 university students (Mage = 20.17), and 250 community dwellers (Mage = 38.44), respectively. Data were analyzed via latent variable modeling procedures. Consistent with hypotheses, latent classes (i.e., subgroups) reflecting interindividual differences in levels and trajectories of perceived change were associated with distinct patterns of selfcertainty and exercise behavior. The findings suggest that adults who experience mastery of skills and physiological changes also have greater self-certainty and exercise more regularly than those who do not see progress or feel as certain of their exercise identity. Keywords: perceived change, self-certainty, life-span study, latent profile analysis, growth mixture modeling

The average adult member of U.S. health clubs visits their club about 2 days per week (International Health, Racquet & Sportsclub Association [IHRSA], 2005), yet the American College of Sports Medicine (ACSM) recommends exercising most days of the week (Haskell et al., 2007). Interestingly, about a third of all members cancel their membership yearly, yet regular gym goers (who average four visits per week) are twice as likely to renew as those who go less frequently (IHRSA, 2005). In general, it takes 6 continuous months to observe significant exerciseinduced changes, particularly in body composition (McArdle, Katch, & Katch, 2010), although about 50% of people discontinue their programs in this same time frame (Dishman, 1991). It is possible that people do not exercise enough to realize noticeable progress and motivation subsequently declines when one’s efforts fail Sean P. Mullen was with the Department of Leadership, Foundations, and Policy, Curry School of Education, University of Virginia, Charlottesville, VA, and is now with the Department of Kinesiology and Community Health, University of Illinois at Urbana-Champaign, Urbana, IL. 710

Perceptions Regarding the Self-as-Exerciser 711

to produce immediate and tangible changes. Stopping exercise for just 1 week can cause the loss of physiologic and performance adaptations (Mujika & Padilla, 2001), so there is much to be lost from even temporary lapses in motivation. This study was designed to determine whether self-assessments of progress are systematically related to exercise participation over time in adults already engaging in exercise. According to Rothman (2000), people continue behaviors they find satisfying, and satisfaction is derived from an ongoing process of self-assessments that consider required effort in light of acquired outcomes. Bryan, Hutchison, Seals, and Allen (2007) have suggested that favorable evaluations of exercise experiences lead to a more positive outlook and “ultimately higher levels of exercise behavior” (p. 3). Supporting these ideas, Baldwin, Rothman, and Jeffery (2009) found that improvements in satisfaction with weight loss over 16 months are associated with increased effort and decreased frustration. Umstattd, Wilcox, and Dowda (2010) found that positive change in satisfaction with body appearance and function was associated with increases in physical activity levels assessed as part of a behavior change program for older adults. Both of these interventions, however, involved previously sedentary adults, and little is known about the subjective experiences of prototypical exercisers, such as their thoughts about whether their efforts are leading to meaningful progress. Inferences about perceptions of change primarily have been based on measures designed for other purposes, with few exceptions (Fleig, Lippke, Pomp, & Schwarzer, 2011; McAuley, Wraith, & Duncan, 1991; O’Connor, Rousseau, & Maki, 2004). For example, measures of perceived overall benefits (Ruppar & Schneider, 2007) and importance ratings of expectancies (Brassington, Atienza, Perczek, DiLorenzo, & King, 2002) have been purported as assessments of “interpretations of the exercise experience,” yet all of these constructs are conceptually distinct and likely have different implications for motivation. Most people can identify common benefits and possible outcomes associated with exercise, but they may not identify with them. One might attribute importance to and expect certain physical changes (e.g., weight loss, strength gain) and yet never perceive any improvements or receive any indication from others that improvements are being made toward this end. Changes in self-reported mood and effort exerted (Bryan et al., 2007) and frequency of felt pain (Fuchs, Göhner, & Seelig, 2011) have also been used to infer perceptions regarding experienced change, but one may overlook occasional downsides of exercise, such as hard work and muscle soreness, if they see themselves making progress over time. Thus, perceived change assessments should explicitly require respondents to reflect on the changes they perceive to be resulting from their own exercise efforts. When examining change constructs, researchers must also justify the time of measurement, and use measures that have characteristics demonstrating stability over time (i.e., temporal invariance). Indeed, “seeing physical change” has been reported as a prime reason by adults for maintaining their health club membership (Mullen & Whaley, 2010), and self-reported experiences of change have been associated with greater adherence to physical rehabilitation among college students (Fisher, Mullins, & Frye, 1993) and cardiac patients (Fleig et al., 2011). O’Connor et al. (2004) found that individuals who nearly matched ACSM’s guidelines tended to report the most favorable “experienced bodily changes,” whereas those who exceeded them or did not exercise at all reported minimal and negative changes, respectively. In a

712 Mullen

similar vein, McAuley et al. (1991) showed that “perceived success” relative to improvement in physical condition (following a 10-week aerobic dance class) was positively associated with enjoyment and perceived competence. Together, these findings suggest a connection between perceptions of exercise-related change and frequency of exercise behavior, but it remains unknown if perceived change systematically varies with exercise behavior over time. One psychological determinant of exercise behavior that has been extensively studied is self-efficacy (McAuley & Blissmer, 2000). Exercise self-efficacy is the belief in one’s own capabilities to adhere to an exercise regimen (McAuley, 1993). Bandura (1997) identified four sources of self-efficacy, namely, mastery experiences, social persuasion, vicarious experiences, and physiological responses. Mastery is believed to be the most powerful source of efficacy information, but this assumption has never been empirically tested. Repeated exposure to single exercises—the bench press—or in combination with feedback has been shown to increase efficacy (Wise, Posner, & Walker, 2004; Wise & Trunnell, 2001), but researchers have ignored the role that the physical experience and observation of others may have, and how all sources may be perceived outside of these contrived conditions. In the context of exercise trials, where efficacy may be derived from multiple sources, self-efficacy often declines at trial end, and fluctuates considerably over the course of the trial. According to McAuley et al. (2011), this occurs when participants must face the prospect of exercising without professional guidance. If that is the case, then the variation in self-efficacy that may also exist among exercisers working out in community-based facilities could be substantial. Such individuals are rarely, if ever, monitored by professionals. In the real-world fitness club environment, exercisers must rely on their own beliefs as to whether their routine is working for them. Some researchers (Gao, Xiang, Lee, & Harrison, 2008; McAuley, 1993) have speculated that the extent to which people rely on particular sources of efficacy changes with development. Gao et al. (2008) suggested that once individuals learn basic movements and how to adjust machinery, other sources may become more salient, such as seeing physiological changes. However, repeated self-evaluations of physiological changes may not serve as a reliable confidence booster, as actual changes exhibit nonlinear trajectories and vary by, among other things, age and fitness level (McArdle et al., 2010) and are thereby difficult to control and predict. Such ambiguity may contribute to motivational instability. When diagnostic information is unavailable or ambiguous, it can cause uncertainty, a mental state typically associated with psychological (Mishel, 1999) and physiological distress (e.g., high blood pressure; Holt-Lunstad, Uchino, Smith, Olson-Cerny, & Nealey-Moore, 2003). Feelings of uncertainty are exhibited by people with chronic health conditions, such as HIV/AIDS and organ failure (see Martin, Stone, Scott, & Brashers, 2010), who must deal with constant unknowns regarding the progression of their conditions. The same is true for people in ambivalent social relationships (HoltLunstad et al., 2003). In the context of physical activity, Frey and Ruble (1990) theorized that uncertainty stems from plateaus in progress or “stagnation.” They found that marathon runners who were no longer improving were more likely to consider a new sport than runners who were improving or declining. Thus, lack of information indicating change appears to open the door to uncertainty. Paradoxically, stagnation in improvement, at least initially, actually signifies that progress


has been made (McArdle et al., 2010), something which many exercisers may not be aware of. When adults have greater clarity about what they want out of their exercise participation, such as stronger expectations of attaining physical, social, and psychological rewards, they tend to show more positive self-efficacy beliefs (Wójcicki, White, & McAuley, 2009). Self-certainty (Pelham, 1991), a construct that has received little attention in the exercise psychology literature, refers to the clarity one has toward beliefs about the self. I propose that certainty regarding the selfas-exerciser contributes to more focused exercise goal pursuits, because a clearer sense of self should facilitate intention enactment, via better planning and decision making—an idea echoed by self-concordance theorists (Sheldon & Elliot, 1999) who view self-consistency and psychological authenticity as vital for optimal psychological functioning. Indeed, multiple studies have shown that individuals with greater self-certainty report higher self-esteem and respond more quickly to “like me” and “not like me” on trait-similarity tasks (see DeMarree, Petty, & Briñol, 2007). Interestingly, Sheeran and Abraham (2003) found that intention certainty significantly moderated the relationship between exercise intentions and behavior. Here, certainty is viewed as a hierarchically arranged construct, as people may be certain of their intentions to exercise, but less certain they wish to be identified as an “exerciser.” Certainty regarding the self-as-exerciser may be thought of then as a higher-order meta-judgment that requires more general self-reference than intention certainty. Self-identity consistently predicts exercise and other health behaviors over and above intentions, even after statistically adjusting for past behavior (Rise, Sheeran, & Hukkelberg, 2010). Seeing oneself as an “exerciser” provides a behavioral standard that motivates one to behave in ways that are identity consistent (Kendzierski, 1988; Strachan, Brawley, Spink, & Jung, 2009). Supporting this idea, Strachan et al. (2009) found that people who perceived their recent exercise behavior as more consistent with their identity showed more positive and less negative affect. Stronger identification with exercise is associated with greater exercise intentions, selfefficacy, and more regular behavior (e.g., Strachan, Brawley, Spink, & Glazebrook, 2010; Strachan et al., 2009). Similar patterns have also been found for individuals classified as “exerciser schematic,” who view exercise as extremely important and descriptive of their self-image (Kendzierski, 1988). Thus, any change related to the strength or conviction in exercise self-identities may have important cognitive, affective, and behavioral consequences. For example, individuals without exerciser self-identities who perceive improvement may begin to redefine themselves as exercisers via self-persuasion and, consequently, exercise more regularly. In contrast, self-assessments reflecting no change may induce feelings of uncertainty, destabilize psychological functioning, and increase behavioral inconsistency. The purpose of the studies herein was to examine whether interpretations of change (i.e., improvement, stability, and decline) relate to certainty regarding the self-as-exerciser. I hypothesized that perceived mastery change (Study 1) and perceived physical change (Study 2) would be positively associated with self-certainty and exercise behavior over time. The investigation was prospective and expressly designed to study exercisers in their natural setting who should vary according to level and rate of change in mastery and physical outcomes over time. In addition, both studies focus on time-structured change that can occur within well-established

714 Mullen

time frames associated with changes in exercise performance and physiological outcomes (McArdle et al., 2010), and exercise adherence (Dishman, 1991), and the theoretical implications that perceptions of such outcomes may have for behavior within these time frames. Relevant background variables were also assessed for the purpose of replicating previous research showing interrelationships among perceived change, exercise participation, fitness, and age (e.g., O’Connor et al., 2004).

Study 1 The purpose of this study was to test the psychometric properties of novel self-report measures and to examine patterns of perceived mastery change (PMC), exerciser self-certainty (EXSC), exercise self-identity (EXID), perceived exercise ability (ABIL), and self-reported engagement in aerobic exercise, including days per week (AEROD) and minutes per session (AEROM). Contrary to the recent approaches used to study motivational profiles (McAuley et al., 2011; Sweet, Tulloch, Fortier, Pipe, & Reid, 2011), I employed a latent class profiling method here to analyze patterns of change across multiple variables simultaneously.

Method Participants and Procedure All procedures were approved by a university institutional review board. Participants were recruited from university-level physical education elective courses at the start of the semester. I was primarily interested in examining fitness facility members (likely to already be engaging in some level of exercise) during a time period when new skill and knowledge acquisition relevant to exercise was highly probable. Therefore, students were recruited from introductory-level physical activity education classes that involve teaching sport skills and physical training strategies. Data collection coincided with the beginning and end of the fall semester. Thus, the targeted sample was known to be relatively homogeneous in terms of age, education, and skill level. All students had unrestricted access to five stateof-the-art fitness facilities. One hundred ninety-six adults (147 women, 49 men), 18–31 years old (Mage = 20.17, SD = 2.12) consented to participate in the longitudinal study. Online surveys were administered via e-mail on three separate occasions (T1 = baseline, T2 = 8 weeks, T3 = 16 weeks); 1.0% and 8.2% failed to complete any assessments at T2 and T3, respectively. The sample consisted mostly of women (75%) at the university’s undergraduate level (93% vs. 5.7% graduate level; 1.3% missing) who self-identified as White/Non-Hispanic (75.5%). The sample was also represented by those who identified as Asian (9.3%), Black/African-American (3.3%), Multiracial (2.6%), Latino/Latina (1.3%), and others (7.9%), who did not report their race/ethnicity.

Measures Demographics. Participants were asked to report their age, gender, education,

and race/ethnicity.


Self-Reported Exercise Behavior. The Physical Activity Recall Questionnaire

(Blair et al., 1985), modified for online implementation, was used to assess the extent of participants’ exercise involvement specific to the fitness facility at which they were a member. Specifically, participants were asked to estimate the mean frequency (the number of days) and mean duration (the number of minutes per session) they engaged in aerobic exercise (e.g., treadmill running/brisk walking, cardio machines) at a moderate-to-vigorous intensity (“strenuous enough to make you breathe hard and or sweat”) over the past week. Participants were asked to distinguish exercise frequency and duration at their respective fitness facility from physical activity accumulated from class, sports, and other recreational activities. The original Physical Activity Recall Questionnaire has been validated with objective measures of physical activity including accelerometers and oxygen uptake (Hayden-Wade, Coleman, Sallis, & Armstrong, 2003).

Perceived Mastery Change. Perceived mastery change was assessed with four items. Participants were asked, “Please think about your current level of exercise knowledge and skills. Then, rate whether you have perceived any change over the past 2 months in response to each of the following questions.” Two items addressed knowledge/expertise (i.e., “How would you rate your current level of knowledge about exercise?” and “How would you rate your current level of exercise expertise/ proficiency?”) and two items addressed skill/performance (i.e., “How would you rate your current level of exercise performance?” and “How would you rate the status of your current exercise skill-level overall?”). Items were rated on a 5-point Likert scale (–2 = rapidly declining, –1 = declining, 0 = stable/no change, +1 = improving, +2 = rapidly improving). Scores were summed and then averaged. Exercise Self-Certainty and Self-Identity. Based on research on intention

certainty (Sheeran & Abraham, 2003) and overall self-certainty (DeMarree et al., 2007), EXSC was assessed with four identical 11-point Likert items (1 = not at all, 11 = very much), “I am certain of this self-description,” that corresponded with three EXID items (e.g., “Someone who exercises regularly”) from the Exercise SelfSchema Questionnaire (Kendzierski, 1988) and an additional item (“I am someone who will always be an exerciser”) used to tap perceptions of one’s future self-asexerciser (for a similar approach, see Strachan et al., 2010). Composite scores for each subscale were calculated by summing and averaging items.

Perceived Exercise Ability. Three items, closely based on items from other

relevant work (Kendzierski & Morganstein, 2009; McAuley et al., 1991) were used to assess ABIL (i.e., “I am very good/capable at exercise,” “I am very skilled at exercising,” “I am very competent with regard to exercise”). Items were assessed on a 5-point Likert scale (1 = strongly disagree, 5 = strongly agree). Scores were summed and averaged.

Data Analytic Strategy All modeling procedures were performed using Mplus 6.0 (Muthén & Muthén, 2011). No significant baseline differences were found between study “dropouts” vs. “completers”; therefore, missing data were assumed to be at random and maximum likelihood robust estimation was employed. Before meaningful interpretations of

716 Mullen

“change” could be made regarding the novel PMC, EXSC, EXID, and ABIL measures, longitudinal invariance testing was conducted with single-factor models of each measure across all three measurement occasions. Invariance testing followed conventions regarding sequential comparisons of nested models, namely, configural, metric, scalar, and residual invariance models (Cheung & Rensvold, 2002; Vandenberg & Lance, 2000). More restrictive models were assumed to be invariant if nonsignificant differences were found between the scaled chi-square statistics, and if differences between the comparative fit index (CFI) were less than .01 (see Table 1). Model-based reliability coefficients were calculated with standardized estimates using McDonald’s (1999) ω1 coefficient, which measures the common variance in the scale as proportional to the total variance (Zinbarg, Yovel, Revelle, & McDonald, 2006). Linear growth curve models with fixed time scores (0, 1, 2) were subsequently conducted. Each was evaluated using multiple criteria. Traditionally, evidence of correct model specification has been based on significant p values associated with χ2, in accordance with root mean square error of approximation (RMSEA) values of < .06, and CFI values ≥ .95 (Hu & Bentler, 1999), and RMSEA and CFI have been shown to perform better than χ2 in the context of linear growth curve models (Wu & West, 2010). However, fit indices are currently a source of great debate and all are sensitive to sample size and model type (Chen, Curran, Bollen, Kirby, & Paxton, 2008; Marsh, Hau, & Wen, 2004). I have adopted a conservative stance for evaluating and reporting models here, relying on multiple criteria, as recommended by the majority of psychometricians (Hu & Bentler, 1999; Jackson, Gillaspy, & Purc-Stephenson, 2009; Marsh et al., 2004; Vandenberg & Lance, 2000). Table 1 Longitudinal Invariance (Study 1) Δ Scaled χ2

Δ df

p Value

Δ CFI

1. Configural vs. Metric 2. Metric vs. Scalar 3. Scalar vs. Residual Exerciser Self-Certainty (ESC)

0.145 6.511 1.850

3 3 4

.988 .089 .772

.007 .009 .005

1. Configural vs. Metric 2. Metric vs. Scalar 3. Scalar vs. Residual Exerciser Self-Identity (EXID)

2.917 3.331 12.066

3 3 4

.576 .308 .283

.000 .000 .001

1. Configural vs. Metric 2. Metric vs. Scalar 3. Scalar vs. Residual Perceived Exercise Ability (ABIL)

0.320 3.124 13.801

3 3 4

.963 .396 .049

.001 .000 .006

1. Configural vs. Metric 2. Metric vs. Scalar 3. Scalar vs. Residual

0.243 2.938 8.363

2 2 3

.884 .232 .142

.000 .000 .002

Model Comparisons Perceived Mastery Change (PMC)


Upon determining overall 16-week curves for each variable, I conducted a latent profile analysis of standardized change scores (i.e., based on the difference between baseline and 8 weeks) to test whether change patterns across multiple self-perceptions (i.e., PMC, EXSC, and EXID) were associated with differential changes in AEROD, AEROM and ABIL. I theorized that heterogeneity in rates of change, particularly during the initial skill-development phase of the semester, could be masked by linear growth curve models. Latent profile analysis resembles cluster analysis (O’Connor et al., 2004), but it is a more person-centered classification, which maximizes differences between empirically derived subgroups and minimizes differences within those subgroups. I assumed missingness to be at random, but AEROD, AEROM, and ABIL were included as an auxiliary variable to reduce the potential of bias (Enders, 2008) and to test for substantive class differences. Posterior probability-based multiple imputations with 1 df were used for the pairwise tests of the equality of means. Decisions regarding the number of classes that best represent the data should be based on theory, fit indices, and substantive value (Samuelson & Mitchell Dayton, 2010). I theorized that at least three different subgroups could exist within the data, representing individuals who perceived no change, negative change, and positive change. I used the Bayesian information criterion (BIC), sample-size-adjusted BIC (ABIC), and the parametric bootstrap likelihood ratio test (BLRT) to determine the proper class solution. Lower BIC and ABIC values suggest a better fitting model, and significant p values (< .05) associated with the BLRT indicate that the solution with one fewer class can be rejected in favor of the target solution. In simulation studies (Nylund, Asparouhov, & Muthén, 2007; Tofighi & Enders, 2008), the BLRT, BIC, and ABIC were the best performing fit statistics, particularly for models with small sample sizes (≤ 250) and continuous outcomes. I also used entropy values and average posterior probabilities for most likely class membership to assess the quality of class solutions. I considered values approximating .90 to be high a priori, but no cutoffs have been established in the literature. I only interpreted best-fitting models.

Results and Discussion Descriptives Internal reliability coefficients (ω1) were as follows: PMC (.78, .77, 83), EXSC (.87, .88, .91), EXID (.90, .90, .91), and ABIL (.91, .90, .92). Results of the invariance testing are reported in Table 1. See Table 2 for bivariate relationships among all study variables. Nearly half (42.9%) the sample indicated that they spent at least 3 days per week engaging in aerobic activity at an average of 47.95 min per session at T1. Exerciser self-certainty shared only a moderate amount of variation with EXID (r = .33–.41), and PMC positively correlated with perceived ability (.27–.34), providing evidence of convergent validity.

Linear Growth Models No significant change or significant slope variance was detected in any of the constructs across 16 weeks, with the exception of PMC. Specifically, the PMC model fit the data well (χ2 = .87 (3), p = .87, RMSEA = .00 [90% CI = .00–.06], CFI = 1.00) and the significant slope (M = –.06, p < .01) indicated a negative trajec-

718

.15*

.56**

PMC2

PMC3

EXSC1

EXSC2

EXSC3

EXID1

EXID2

EXID3

ABIL1

ABIL2 ABIL3

4

5

6

7

8

9

10

11

12

13 14

.16* .22**

.25**

.19*

.18*

.28**

.13

.02

.08

.03

–.02

.11

—

2

.31** .23**

.29**

.39**

.39**

.42**

.13

.13

.14†

.48**

.50**

—

4

.27** .20*

.08

.33**

.32**

.13

.27**

.24**

.11

.60**

—

5

.24** .34**

.18*

.35**

.26**

.19*

.30**

.23**

.09

—

6

.31** .27**

.30**

.31**

.34**

.41**

.50**

.51**

—

7

.27** .34**

.19**

.32**

.33**

.25**

.68**

—

8

.24** .37**

.21**

.40**

.32**

.24**

—

9

.59** .53**

.66**

.83**

.84**

—

10

.62** .57**

.59**

.88**

—

11

.60** .59**

.57**

—

12

.79** .69**

—

13

— .70**

14

Note. AEROD = aerobic training days per week; AEROM = aerobic minutes per session; PMC = perceived mastery change; EXSC = exercise self-certainty; EXID = exercise self-identity; ABIL = perceived ability. **p < .01; *p < .05; †p = .05.

.35** .36**

.38**

.54**

.52**

.11

.18*

.28**

.21**

.34**

PMC1

3

.45**

AEROM

1 —

2

Variables 1 AEROD

Table 2 Correlations Among Study Variables for Study 1


tory from baseline (intercept M = .56, p < .01). Although average change (MΔ) was not observed in constructs across the entire 16 weeks, interindividual differences in intraindividual change were predicted to be masked by the growth modeling approach. Just-fit models involving an increment factor with mean and variance set to zero were conducted for scores across each 8-week time frame, and, as predicted, significant variability was found across all self-perceptions. These findings imply that there are individual differences in perceptions across measurement periods.

Latent Profile Analysis The results of the alternative latent profile analysis revealed subgroups exhibiting patterns of changes across multiple variables that were differentially related to exercise behavior. Specifically, I obtained class solutions for one, two, three, and four classes. I rejected the four-class solution in favor of the three-class solution based on a nonsignificant BLRT (8.05; p = .33). A significant BLRT (16.23, p < .001; based on 100 bootstrap draws) indicated that the two-class solution could be rejected in favor of the three-class solution. The BIC and ABIC values revealed conflicting information for the two- and three-class solutions (1600.42 and 1568.74 vs. 1605.26 and 1560.91, respectively); however, the three-class solution had higher entropy (.87 vs. .91). Class membership probabilities were similar (ranging from .87 to .99). Finally, the best log-likelihood value (–765.76) for the three-class solution was replicated using multiple sets of starting values and seed points, and I retained this solution for interpretation.1 Based on the PMC, EXSC, and EXID patterns of change, the three classes of participants were characterized as no mastery change (C1, n = 2), negative mastery change (C2, n = 20), and positive mastery change (C3, n = 172). The C1 group represented individuals who experienced no significant change in PMC (MΔ = –.45, p = .53), and at the same time they exhibited a steep negative trend in EXSC (MΔ = –3.74, p < .001) coupled with a nonsignificant decline in EXID (MΔ = –.10, p = .83). The C2 group exhibited a negative trajectory for PMC (MΔ = –1.34, p < .001), EXSC (MΔ = –1.12, p < .01), and EXID (MΔ = –1.43, p < .001), whereas the C3 group exhibited a slight positive trend across all three: PMC (MΔ = .19, p < .05), EXSC (MΔ = .21, p < .01), and EXID (MΔ = .19, p < .05). Classes were also differentially associated with AEROD and ABIL, but not AEROM. Specifically, the C3 group reported significantly (p < .05) more positive change in AEROD (.08) relative to C2 (–.44), and greater change in ABIL (.08) relative to C1 (–1.23) and C2 (–.45). The results of this study provide initial evidence of strong psychometric properties for all of the novel self-report measures. Specifically, all measures were fully longitudinally invariant. Construct validity was also demonstrated via bivariate relationships. Moreover, the findings provide initial support for the theory-based prediction that PMC is related to exercise participation and that the perceptions of change (be it positive or negative) are better than perceptions of no change at all. Together, these results suggest that the more participants exercised, the more mastery of skill and knowledge change they perceived, and the greater certainty they felt regarding their exerciser identities. These preliminary results provide a strong foundation for continuing to explore the interrelationships among exercise behavior, self-assessments of progress, and certainty regarding the self-as-exerciser.

720 Mullen

Study 2 Extending the work of O’Connor et al. (2004) on experienced physical changes resulting from exercise, Study 2 was intended to examine change in perceived physical change (PPC) over time. A secondary objective was to examine a relatively heterogeneous sample of adult fitness club members over a shorter 4-week interval to increase the likelihood of capturing variability in interindividual differences in intraindividual change. To capitalize on heterogeneity in PPC, I used growth mixture modeling to examine latent classes (Jung & Wickrama, 2008; Ram & Grimm, 2009). I hypothesized that subgroups exist in the data that differ in terms of their level and rate of change in PPC. Specifically, I predicted that latent classes would reflect improvement, decline, and stability in PPC over time, mirroring profiles reported by O’Connor et al. (2004). Further, I hypothesized that more favorable levels and PPC trajectories would be associated with higher aerobic and resistance training, more positive certainty, identity, and self-efficacy scores. Due to established relationships among PPC, age, and fitness (O’Connor et al., 2004), participants’ age, experience, and nonstop continuous exercise were also predicted to differ across latent trajectory classes.

Method Participants and Procedure Once again, all procedures were approved by a university institutional review board. A separate sample consisting of 250 (198 female, 52 male) adult members of 24-hr fitness facilities were recruited to participate in this study. I determined this sample size cutoff a priori, as this size has been found to have sufficient power for the planned analysis (Tofighi & Enders, 2008). During recruitment, 151 additional survey respondents submitted multiple blank survey pages (completed 80% missing data) were subsequently dropped from the analysis. Participants identified as being White/ Non-Hispanic (90.8%), Asian/Pacific-Islander (3.6%), Black/African-American (2.8%), Hispanic/Latino (1.6%), or Other (.8%). The highest level of education reported was a graduate degree (33.6%), followed by a college degree (31.2%), some college (30.0%), and a high school diploma (3.6%) (.8% missing). The average age of the entire sample was 38.44 years (SD = 18.36). Three sample-defined age groups were constructed to examine age differences across the young (n = 136; age range = 18–34 years; M = 24.51, SD = 4.17), midlife (n = 68; 35–54 years; M = 45.15, SD = 5.85), and oldest adults (n = 46; 55–94 years; M = 69.67, SD = 10.50). Participants were recruited from across the state of Virginia primarily by informational posters and electronic advertisements. Participants with computer access were asked to either (a) contact an e-mail address account at which they would receive an automated response containing a direct link to the online survey or (b) click on a URL link within an online advertisement they received that redirected


them to the survey; a “drop box” was made available to those who opted for hard copies of the survey in the main office of their fitness facility. Surveys were administered at equidistant time points across 4 weeks (T1 = baseline, T2 = 2 weeks, T3 = 4 weeks), but all participants had varying starting points. All participants were offered a small participation incentive: three chances to win a free personal training consultation valued at $70.

Measures Demographics. Participants were asked to indicate their age, gender, education, race/ethnicity, and their previous exercise experience (i.e., one 12-point Likert item, “How much experience do you have doing structured exercise within a fitness facility that includes some combination of aerobic, strength, and flexibility training?” 1 = none, 12 = more than 10 years). Nonstop Exercise Record. Participants’ on-going record of nonstop continuous

exercise was assessed via one item on a 13-point Likert scale (“How long have you been exercising continuously?” 1 = I have never intentionally exercised, 13 = more than 1 year). Directions indicated that the word “continuous means that you have not taken more than 3 consecutive weeks off and you have been doing at least 3 days per week of moderate to vigorous physical activity.”

Self-Reported Exercise Behavior. Following the same format as Study 1, the modified Physical Activity Recall Questionnaire was used to assess participants’ aerobic exercise involvement. In addition, participants were asked to report the average days and minutes they spent engaging in strength (STRGD, STRGM) and flexibility training (FLEXD, FLEXM). Examples of strength training activities included weight machine and pull-ups, whereas flexibility training included stretching and yoga. Days and minutes for each activity were averaged across time points for the modeling analyses. Perceived Physical Change. The Experienced Bodily Changes Scale (O’Connor

et al., 2004) was used to assess participants’ perceptions of physical changes (PPC) relevant to the exercise experience. Participants were asked to reflect “over the past two weeks” and subsequently rate the extent to which each change had been experienced on an 11-point Likert scale (–5 to +5 change). O’Connor et al. reduced the scale from eight to five items. All items were used in this study, but only five were retained (i.e., strength, breathing, reflexes/reaction times, energy, and appearance) based on CFA and invariance testing. A PPC composite was computed by summing and averaging scores.

Exercise Self-Certainty and Self-Identity. The modified Exercise Self-Schema

Questionnaire measure was used, and EXSC and EXID variables were calculated as previously described, with the exception that the “future self” items were omitted in an attempt to reduce participant burden.

Exercise Self-Efficacy. An abbreviated version (first four items) of McAuley’s (1993) Exercise Self-Efficacy Scale was used to assess participants’ confidence in accumulating 40 min or more per exercise session, per week, for 4 weeks. Items are rated on an 11-point Likert scale (0% = not at all confident, 100% = highly confident), and they were summed and averaged.

722 Mullen

Data Analytic Strategy Again, all models were conducted using Mplus 6.0 (Muthén & Muthén, 2011), and employment of data screening and estimation procedures were identical to Study 1. Initially, invariance testing was conducted with the PPC measure (see Table 3). Growth mixture modeling (Jung & Wickrama, 2008; Ram & Grimm, 2009) was subsequently conducted. Similar to latent profile analysis in Study 1, growth mixture modeling involves a systematic extraction of k + 1 classes until a final solution is reached. In contrast to Study 1, a linear growth curve for PPC was estimated and unobserved latent classes (c) were extracted based only on interindividual differences in average PPC scores at Week 4 (intercept) and rate of PPC change (slope). Class differences were compared across multiple auxiliary variables, including age, experience, nonstop exercise, AEROD, AEROM, STRGD, STRGM, FLEXD, FLEXM, EXSC, EXID, and exercise self-efficacy (EXSE). Again, decisions regarding the number of classes that best represent the data were based on theory and previous work (O’Connor et al., 2004), in combination with traditional fit indices used in latent class analyses.

Results and Discussion Descriptives The PPC measure was not fully invariant across time, but its properties were deemed sufficient for further testing (ω1 = .76, .76, .80). The average days per week spent engaging in aerobic, strength, and flexibility training were positively associated with PPC scores (see Table 4), although strength training (r = .31–.36) showed the strongest relationships with PPC, followed by aerobic (r = .25–.27) and flexibility training (r = .15–.18). Interestingly, average strength training durations were correlated with PPC (r = .20–.25), whereas aerobic training durations showed an inconsistent relationship, and no relationship was found between flexibility training durations and PPC. The majority of the sample (82% and 48.8%) reported spending at least 3 days per week engaging in 30 or more minutes of aerobic training and

Table 3 Longitudinal Invariance Testing (Study 2) Model Comparisons Perceived Physical Change 1. Configural vs. Metrica 2. Metrica vs. Scalar 3. Scalar vs. Residual

Δ Scaled χ2

Δ df

p Value

Δ CFI

4.246 9.196 5.272

3 4 5

.345 .056 .828

.001 .008 .011

aThe S-B corrected χ2 difference between the configural and metric models (i.e., after constraining factor loadings to be equal across time) was significant (p = .04). Therefore, partial metric invariance was tested involving equivalent loadings for four of the five items (breathing, strength, energy, appearance), and loadings for reflexes were allowed to vary across occasions. Although CFI change was > .01 in the last invariance test, the difference in RMSEA was < .015 (a cutoff recommended by Chen, 2007), and overall, the change between models is negligible.

723

.01

.11

.27**

—

5

.14* .15*

.44** .39** .29** .39** .33** .21** .19**

.47** .40** .36** .40** .34** .25** .22**

13 EXID1

.05 .72**

—

10

—

11

—

13

—

.50** .29** .42** .36** .27**

.23** .21** .23** .29** .35** .36** .50** .58** .58** .74**

—

18

Note. AEROD = average aerobic training days per week; AEROM = average aerobic minutes per session; STRGD = average strength training days per week; STRGM = average strength minutes per session; FLEXD = average flexibility training days per week; FLEXM = average flexibility minutes per session; PPC = perceived physical change; EXSC = exercise self-certainty; EXID = exercise self-identity; EXSE = exercise self-efficacy. **p < .01; *p < .05; †p = .06.

.16*

.28** .19** .17** .35** .38** .36** .57** .57** .57**

.53** .30** .48** .40** .28** .20** .24** .25** .34** .35** .40** .45** .51** .57** .63** .73** .85**

—

17

17 EXSE2

—

16

18 EXSE3

.15*

.17** .50** .54** .56** .86** .93**

15

.49** .40** .39** .41** .32** .24** .24**

—

14

.42** .34** .34** .40** .22**

.45** .54** .46** .90**

.53** .51** .47**

—

12

15 EXID3

.05

–.04

.20** .78** .78**

.12*

.05

—

9

16 EXSE1

.03

.14*

.10

14 EXID2

.14*

.10

.12†

.27** .19** .24** .26** .24** .20**

.15*

12 EXSC3

.14*

.19**

.17** .24**

—

.48** .67**

.51**

8

.27** .18** .26** .25** .25** .19**

.16*

.01 .07

—

7

10 EXSC1

.36** .22** .18**

.04

—

6

11 EXSC2

.06

.25**

.26** .09** .35** .25** .18**

8 PPC2

9 PPC3

.27** .18** .31** .20** .15**

.04

.08

—

4

7 PPC1

.11

.41**

6 FLEXM

.12

.27** .36** .52**

.40**

4 STRGM

—

3

5 FLEXD

—

.46**

.54** .20**

2 AEROM

2

3 STRGD

1 —

Variables 1 AEROD

Table 4 Correlations Among Study Variables for Study 2

724 Mullen

strength training at T1. The frequency most often cited for stretching was 0 days (32.4%) and 0 minutes (31.7%). Again, the evidence suggested that EXSC was distinct from other well-established self constructs, including EXID (r = .53–.56) and EXSE (.35–.45).

Growth Mixture Model Initially, a linear growth curve model was applied to determine the overall shape of the PPC curve. The model with fixed time scores (0, 1, 2) was not tenable (χ2 = 9.37 (1), p = .002, CFI = .96, RMSEA = .18). Upon visual inspection of the means, the PPC trajectory appeared to have a quadratic shape, but, with only three data points, it was not possible to test a nonlinear curve. Therefore, the second time score was freed and residual variances were constrained to be equal. This model fit the data well (χ2 = 3.40 (2), p = .18, CFI = .99, RMSEA = .05). In this case, the slope represents the overall change that occurred between T1 and T3 and the second loading is the proportion of change that occurred between T1 and T2 (and between T2 and T3). The slope was –.21 (p < .01) and the second loading was .60, indicating that the average PPC change between T1 and T3 was –.21 unit, and that 60% of that change occurred between T1 and T2 (leaving, on average, 40% between T2 and T3). Next, a growth mixture model with identical time scores was conducted. The overall model’s time scores (0, *, 1) and estimated T2 slope loading changed slightly (.50), but can be interpreted similarly to the linear growth curve model. Specifically, 50% of the change occurred between T1 and T2, and 50% occurred between T2 and T3. However, in this case, there was heterogeneity within the sample with regard to rates of change. I obtained class solutions for one, two, three, and four classes. I rejected the four-class solution in favor of the three-class solution based on a failure to replicate the best log likelihood. I rejected the two-class solution in favor of the three-class solution based on a significant BLRT (115.20, p < .001; based on 100 bootstrap draws), and lower BIC and ABIC values (11,191.96 and 11,096.86 vs. 11,126.46 and 11,002.82, respectively). Furthermore, relative to the two-class solution, the three-class solution had higher entropy (.74 vs. .86, respectively) and larger probabilities for accurate class assignment based on most likely class membership (.90 to .94 vs. .92 to .96). Finally, the best log-likelihood value (–5455.56) for the three-class solution was replicated using multiple sets of starting values and seed points. The three-class model also led to the most substantive interpretation of the data (see Table 5 for Ms and SDs). Based on the PPC patterns in conjunction with selfreported outcome variables, the three classes of participants were characterized as no change, no trend (C1), moderate positive change, declining trend (C2), and high positive change, no trend (C3). The C1 group’s 4-week PPC scores and plateaushaped trend indicated no degree of change at all, which was coupled with relatively low to moderate levels of exercise and other self-perceptions. The C2 group had moderately positive PPC baseline scores and a significant declining trend combined with a moderately high exercise pattern and psychological profile; note that this group’s scores remained in the positive range of the scale (albeit approaching “no change”). The C3 group had a fairly stable trend of high, positive PPC scores over time, in addition to the most exercise and most favorable psychological profile.

725

1.95

31.41

.38

5.71

.96

11.90

37.49

7.86

8.15

8.62

7.16

6.96

AEROD

AEROM

STRGD

STRGM

FLEXD

FLEXM

AgeT1

ExperienceT1

Nonstop ExerciseT1

EXSCT3

EXIDT3

EXSET3

.38

.32

.25

.48

.49

1.99

2.05

.18

1.39

.09

3.37

.22

.09

SD .14

44.64

9.24

9.13

—c

9.41

—b N = 154

10.21

—c

9.09

—c

39.52

—b

18.38

—a

2.00

—c —c

33.70

—c

2.23

3.27

—c —c

–.25

—c

M .74

.19

p .55

.19

.15

.15

.24

.27

1.63

1.66

.17

1.77

.11

1.57

.12

.11

SD .09

—b

— a†

—b

— a,b

—a

—a

—b

—b

—b

—b

—b

—b

2 months, < 3 months) and “10” (i.e., > 4 months, < 5 months). a,b,c Significant mean differences (p < .05); identical notation = not significantly different; †marginally significant (p = .06).

N = 67

–.12

Slope

Class Size

M .08

Variable Intercept

C1 = No Change, Stable

Table 5 Latent Class Descriptive Statistics (Study 2)

726 Mullen

Mirroring findings reported in Study 1, perceptions of change were strongly linked to EXSC and exercise behavior. Here, change perceptions were identified in two sample-derived subgroups (i.e., Classes 2 and 3), both of whom reported overall improvements. However, Class 3 experienced steady improvement whereas Class 2 experienced significantly less positive changes every 2 weeks. A third subgroup, Class 1, perceived no change, consistently, across time. Findings were irrespective of age, yet profiles indicating “change” (mean levels and trajectories) were associated with longer periods of nonstop exercise relative to profiles indicating “no change.” Interestingly, participants in Class 2 also had the most experience, >2 years, although only significantly more than Class 1. Again, these results suggest that the more participants exercised, the more physical changes they interpreted, and the greater certainty they ascribed to their exerciser identities, although the direction of these relationships is unknown.

General Discussion The purpose of this study was to test the hypothesis that perceived change is systematically related to exercise behavior. I theorized that perceptions of change act as a source of information through which one can gauge his or her progress and that seeing no change should be associated with lower levels of self-certainty and exercise behavior. Therefore, I hypothesized that subgroups (i.e., latent classes) of adults would differ in their perceptions of change, and that perceived change would be positively associated with EXSC and exercise behavior. In both studies, subgroups reflecting substantive differences in perceptions of change were found. Moreover, two types of change (PMC and PPC) were linked to higher EXSC and more exercise behavior than perceptions reflecting no noticeable change. These findings are consistent with Frey and Ruble’s (1990) work on marathon runners’ performance status, and they extend the work of other researchers (Fuchs et al., 2011; O’Connor et al., 2004) who have found greater physical activity levels among those who perceive the most change. The underlying theme of this multistudy report is that adults draw from multiple sources of information and they do not perceive exercise-related change at the same time, rates, levels, or in the same direction. I have attempted to shed light on this rather large gap in the literature by developing novel assessments of two types of change to examine these individual differences. The take-home message here is that more noticeable change corresponds with more favorable exercise outcomes. It remains to be determined which of these information sources are most perceptible, meaningful, and important for exercise behavior maintenance. The commonly held belief that approximately 50% of exercisers who begin an exercise program drop out within 6 months (Dishman, 1991) was further substantiated herein, as 49.6% of the sample in Study 2 reported less than six full months of nonstop exercise. Interestingly, those who perceived overall positive physical change (C2 and C3) had a similar 4- to 5-month record for nonstop exercise, and they exercised for similar durations across each activity type, and yet the more experienced group (C2) exercised fewer days relative to the less experienced group (approx. 2 vs. 5 days per week). It is difficult to speculate the reason for this finding. However, given that the results merely represent a snapshot of each groups’ perceived activity level over the course of 4 weeks, one cannot assume that these


exercisers remained at this level. The “veterans” may even be transitioning out of rigorous exercise program. Although no age differences were found, studies have shown that when performing the same exercise intensity, older adults perceive greater overall exertion and reduced mood compared with younger adults (e.g., Focht, Knapp, Gavin, Raedeke, & Hickner, 2007), and this could influence their interpretation of change and subsequent behavior. Perhaps of most concern is that in Study 1, 10% of the sample exhibited a negative trend in PMC associated with exercise across 2 months while involved in a physical activity skill-development program, whereas in Study 2, 62% exhibited a negative trend while exercising on their own. Whether declining or leveling off, the members of each group might serve as prime candidates for tailored interventions designed to reduce staleness and deficits in their current routines (e.g., lack of strength training, too few days spent engaging in aerobic activity). Interventions based on Rothman’s framework and designed to increase “satisfaction” have, in large part, been successful at increasing exercise behavior among previously sedentary adults (Baldwin et al., 2009; Fleig et al., 2011; Umstattd et al., 2010). Although satisfaction with exercise experiences may be necessary for a person to initially follow a prescribed exercise regimen, it may not be sufficient to sustain exercise behavior. Unfortunately, satisfaction was not examined in this study, but previous work found a moderately strong correlation (r = .52) between a measure of satisfaction and exercise experiences (Fleig et al., 2011), although this relationship would likely attenuate if one’s goals were not being met. Supporting this idea, Kalarchian and colleagues (2011) have shown that satisfaction with weight control is moderated by one’s goals within the first 6 months of tailored interventions (e.g., focusing on appearance, health, or both), and it has no relationship with weight loss outcomes a year later. It is also possible that one may perceive no or negative change as a result of exercise, and still be satisfied. Future research should examine changes in these constructs together in response to exercise training. Bandura (1997) theorized that four sources of information contribute to selfefficacy. McAuley et al. (2011) have shown that sedentary older adults tend to recalibrate (lower) their self-efficacy after just 3 weeks of an exercise intervention, but we do not know what sources of information are used in this process. This study suggests that adults already engaging in exercise use two sources of efficacy. It still remains undetermined whether exercisers attribute more importance to information indicating mastery or physiological change when judging their overall exercise selfefficacy. Interestingly, the correlations between exercise behavior and sources were similar in magnitude across studies. But, as McAuley (1993) and Gao et al. (2008) have suggested, the reliance on one particular source over another is likely to occur at different stages of development. Much more work is needed to further illuminate the connections among sources of efficacy in natural and experimental settings. It is possible that many of the adults in this study identified themselves as “exercisers” and that people were just behaving in ways that were consistent with their identities, as Strachan and her colleagues (2009, 2010) and others (Kendzierski, 1988; Rise et al., 2010) have shown. Although the correlations between EXSC and EXID ranged from .32 to .56 across studies, these appear to be separate constructs that operate independently, as evidenced by the differential relationships across subgroups. To date, no one has explored whether exercise identity development is “bottom up” or “top down,” but it is possible that an accumulation of efficacy

728 Mullen

information and self-certainty contribute to this process. It is also possible that hoping for such an identity (Whaley & Ebbeck, 2002) may cause interpretations of change to be more optimistic, as O’Connor et al. (2004) found that individuals with the highest levels of optimism also perceived the most physical change. This study has several strengths. It is the first of its kind to investigate interpretations of change related to mastery of skill and knowledge, and physiological responses to exercise. Furthermore, few (Sheeran & Abraham, 2003) have examined certainty in the exercise context, a construct with great relevance for the field owing to its ties with actual physiological outcomes (Holt-Lunstad et al., 2003), cognitive performance (Holland, Verplanken, & van Knippenberg, 2003), and the automaticity of self-related judgments (Buckley & Cameron, 2011; DeMarree et al., 2007; Kendzierski, 1988). In addition, the study’s repeated-measures design and powerful analytic techniques are strengths. Mixture modeling, in particular, has been used only in a handful of studies on exercise motivation and behavior (McAuley et al., 2011) and was used in the current study as a means of theory testing, as opposed to data exploration. In addition, participants in this study were purposively sampled in an effort to test theoretically based predictions with a relatively homogenous (Study 1) and heterogeneous sample (Study 2) of nonsedentary adults. Despite its strengths, several of the study’s limitations should be noted. Some researchers have argued that hybrid mixture modeling techniques, such as those employed in this study, may capitalize on “outliers” as opposed to real subgroup differences (Bauer & Curran, 2003). In addition, inferences made about very small subgroups (n = 2) may not be generalizable to larger populations. It is my view that outliers reflect real people with extreme values that occur every day in practice (Muthén, 2003). Moreover, this study’s methodology was theoretically driven, matched the research question, and abided by every guideline offered by psychometricians advocating their use (Nylund et al., 2007; Ram & Grimm, 2009; Samuelson & Mitchell Dayton, 2010). It is possible that people who responded to the online survey represent a biased sample to some extent; however, they are likely no different from typical respondents to more traditional survey-based methods (Dillman, 2007). In addition, I should reiterate that all measures were based on selfreport and it would be prudent to track objective measures of change (i.e., physical activity, fitness, functioning, and physique) in harmony with perceptions of change. Given that no self-report measures were available, I used proxy measures for mastery and physiological responses, and further psychometric testing is warranted to test for invariance across groups and varying time frames. More data points may also better capture change, and may reveal that patterns are nonlinear in nature. Finally, although the demographic characteristics match member characteristics reported by the fitness industry (IHRSA, 2005), it is unknown whether self-assessments of change are consistent across all groups. There are several directions for future research. This study focused on timestructured change and interindividual differences regarding change. Within-person interpretations of change may be even more variable over the course of time and may fluctuate considerably depending on the time of measurement. In addition, much more work is needed to further understand the combined use of each source of efficacy and how they each respond to exercise training across various populations. The two efficacy sources not targeted here, social persuasion and vicarious experi-


ences, also deserve future attention in terms of measure development. It would also be of great value to follow up with people exhibiting each type of change profile found here to determine if or when they drop out. Delineating the antecedents of behavioral change will help understand who might be more likely to return after a break and who might not set foot into a fitness club again. There are several practical implications that can be drawn from these studies. Fitness professionals should teach their clients how to change their routine to yield different results upon reaching personal plateaus, such as using periodization training (Bompa, 1999). In addition, based on Study 2, it appears that the ACSM recommendations (Haskell et al., 2007) for combining strength with aerobic training may be beneficial in terms of being more likely to “see change.” A large percentage of people did absolutely no stretching exercises. Therefore, when exercise clients disproportionately rely on seeing physiological changes for motivation, their focus should be redirected toward functional improvements, namely, flexibility, or to other benefits of exercise such as social and psychological rewards (Wójcicki et al., 2009). Most importantly, novice exercisers may have the misconception that a plateau is a bad thing rather than indicating a physiological adaptation; consequently, all beginners should be advised of when changes do and do not take place, as a function of exercise training. In conclusion, perceived change is positively related to self-certainty and aerobic and strength training. Pelham (1991, p. 519) claimed that self-certainty “should insulate people from a number of sources of self-concept change.” The findings reported here support his perspective but further suggest that, in the context of exercise, perceptions of no change may reduce self-certainty. Thus, continuity in perceiving change or that one’s effort is meaningfully productive, may be critical in maintaining motivational stability.

Note 1. A reviewer suggested that I focus on a two-class solution, given that Class 1 was “miniscule” in size. Other researchers may also side with this perspective, such as Bauer and Curran (2003), who might refer to these two individuals as “outliers,” or a product of “over-extraction” of classes. One may also feel that drawing conclusions about this group is unlikely to generalize to other populations. I concede that this is possible, and yet I believe it is far more likely that this small group represents a substantive interindividual difference (stability), and it would be a mistake to knowingly misclassify a group of people when there are theoretical, empirical, and practical reasons for maintaining a distinction among classes.

Acknowledgments This study was partially supported by a doctoral dissertation award from the Center of Advanced Study of Teaching and Learning (CASTL) at the University of Virginia. I would like to thank my wife, Jessica Mullen, for serving as my project coordinator; Diane Whaley, Courtney Lyder, Nancy Deutsch, and Tonya Moon for their feedback with regard to the study’s design; Deborah Kendzierski for sharing her insight regarding theoretical issues; James Peugh for statistical advice; and Edward McAuley for reviewing an early draft of the manuscript.

730 Mullen

References Baldwin, A.S., Rothman, A.J., & Jeffery, R.W. (2009). Satisfaction with weight loss: examining the longitudinal covariation between people’s weight-loss-related outcomes and experiences and their satisfaction. Annals of Behavioral Medicine, 38(3), 213–224. Bandura, A. (1997). The anatomy of stages of change. American Journal of Health Promotion, 12(1), 8–10. Bauer, D.J., & Curran, P.J. (2003). Distributional assumptions of growth mixture models: Implications for overextraction of latent trajectory classes. Psychological Methods, 8(3), 338–363. Blair, S., Haskell, W., Ho, P., Paffenbarger, R., Jr., Vranizan, K., Farquhar, J., et al. (1985). Assessment of habitual physical activity by a sevenday recall in a community survey and controlled experiments. American Journal of Epidemiology, 122(5), 794–804. Bompa, T. (1999). Periodization theory and methodology of training. Champaign, IL: Human Kinetics. Brassington, G.S., Atienza, A.A., Perczek, R.E., DiLorenzo, T.M., & King, A.C. (2002). Intervention-related cognitive versus social mediators of exercise adherence in the elderly. American Journal of Preventive Medicine, 23(2), 80–86. Bryan, A., Hutchison, K.E., Seals, D.R., & Allen, D.L. (2007). A transdisciplinary model integrating genetic, physiological, and psychological correlates of voluntary exercise. Health Psychology, 26(1), 30–39. Buckley, J., & Cameron, L.D. (2011). Automatic judgments of exercise self-efficacy and exercise disengagement in adults experienced and inexperienced in exercise selfregulation. Psychology of Sport and Exercise, 12, 324–332. Chen, F.F. (2007). Sensitivity of goodness of fit indices to lack of measurement invariance. Structural Equation Modeling, 14, 464–504. Chen, F., Curran, P.J., Bollen, K.A., Kirby, J., & Paxton, P. (2008). An empirical evaluation of the use of fixed cutoff points in RMSEA test statistic in structural equation models. Sociological Methods & Research, 36(4), 462–494. Cheung, G.W., & Rensvold, R.B. (2002). Evaluating goodness-of-fit indexes for testing measurement invariance. Structural Equation Modeling, 9(2), 233–255. DeMarree, K.G., Petty, R.E., & Briñol, P. (2007). Self-certainty: Parallels to attitude certainty. International Journal of Psychology & Psychological Therapy, 7(2), 159–188. Dillman, D.A. (2007). Mail and Internet surveys: The tailored design method. Hoboken, NJ: John Wiley & Sons Inc. Dishman, R.K. (1991). Increasing and maintaining exercise and physical activity. Behavior Therapy, 22(3), 345–378. Enders, C.K. (2008). A note on the use of missing auxiliary variables in full information maximum likelihood-based structural equation models. Structural Equation Modeling, 15(3), 434–448. Fisher, A.C., Mullins, S.A., & Frye, P.A. (1993). Athletic trainers’ attitudes and judgments of injured athletes’ rehabilitation adherence. Journal of Athletic Training, 28(1), 43–47. Fleig, L., Lippke, S., Pomp, S., & Schwarzer, R. (2011).Exercise maintenance after rehabilitation: How experience can make a difference. Psychology of Sport and Exercise, 12, 293–299. Focht, B.C., Knapp, D.J., Gavin, T.P., Raedeke, T.D., & Hickner, R.C. (2007). Affective and self-efficacy responses to acute aerobic exercise in sedentary older and younger adults. Journal of Aging and Physical Activity, 15(2), 123–138. Frey, K.S., & Ruble, D.N. (1990). Strategies for comparative evaluation: Maintaining a sense of competence across the life span. In R.J. Sternberg & J. Kolligian, Jr. (Eds.), Competence considered (pp. 167–189). New Haven, CT: Yale University Press. Fuchs, R., Göhner, W., & Seelig, H. (2011).Long-term effects of a psychological group intervention on physical exercise and health: The MoVo concept. Journal of Physical Activity and Health, 8, 794–803.


Gao, Z., Xiang, P., Lee, A.M., & Harrison, L. (2008). Self-efficacy and outcome expectancy in beginning weight training class: Their relations to students’ behavioral intention and actual behavior. Research Quarterly for Exercise and Sport, 79(1), 92–100. Haskell, W.L., Lee, I.M., Pate, R.R., Powell, K.E., Blair, S.N., Franklin, B.A., et al. (2007). Physical activity and public health: updated recommendation for adults from the American College of Sports Medicine and the American Heart Association. Medicine and Science in Sports and Exercise, 39(8), 1423–1434. Hayden-Wade, H., Coleman, K., Sallis, J., & Armstrong, C. (2003). Validation of the telephone and in-person interview versions of the 7-day par. Medicine and Science in Sports and Exercise, 35(5), 801–809. Holland, R.W., Verplanken, B., & van Knippenberg, A. (2003). From repetition to conviction: Attitude accessibility as a determinant of attitude certainty. Journal of Experimental Social Psychology, 39(6), 594–601. Holt-Lunstad, J., Uchino, B.N., Smith, T.W., Olson-Cerny, C., & Nealey-Moore, J.B. (2003). Social relationships and ambulatory blood pressure: structural and qualitative predictors of cardiovascular function during everyday social interactions. Health Psychology, 22(4), 388–397. Hu, L., & Bentler, P.M. (1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling, 6, 1–55. International Health, Racquet & Sportsclub Association [IHRSA]. (2005). American sports data health club trend report. Fort Mill, SC. Jackson, D.L., Gillaspy, J.A., & Purc-Stephenson, R. (2009). Reporting practices in confirmatory factor analysis: An overview and some recommendations. Psychological Methods, 14(1), 6–23. Jung, T., & Wickrama, K. (2008). An introduction to latent class growth analysis and growth mixture modeling. Social and Personality Psychology Compass, 2, 302–317. Kalarchian, M.A., Levine, M.D., Klem, M.L., Burke, L.E., Soulakova, J.N., & Marcus, M.D. (2011). Impact of addressing reasons for weight loss on behavioral weight-control outcomes. American Journal of Preventive Medicine, 40, 18–24. Kendzierski, D.A. (1988). Self-schemata and exercise. Basic and Applied Social Psychology, 9(1), 45–59. Kendzierski, D.A., & Morganstein, M.S. (2009). Test, revision, and cross-validation of the Physical Activity Self-Definition Model. Journal of Sport & Exercise Psychology, 31(4), 484–504. Marsh, H.W., Hau, K.T., & Wen, Z. (2004). In search of golden rules: Comment on hypothesis-testing approaches to setting cutoff values for fit indexes and dangers in overgeneralizing Hu and Bentler. Structural Equation Modeling, 11(3), 320–341. Martin, S.C., Stone, A.M., Scott, A.M., & Brashers, D.E. (2010). Medical, personal, and social forms of uncertainty across the transplantation trajectory. Qualitative Health Research, 20(2), 182–196. McArdle, W.D., Katch, F.I., & Katch, V.L. (2010). Exercise physiology: nutrition, energy, and human performance (7th ed.). Philadelphia: Wolters Kluwer/Lippincott Williams & Wilkins Health. McAuley, E. (1993). Self-efficacy and the maintenance of exercise participation in older adults. Journal of Behavioral Medicine, 16(1), 103–113. McAuley, E., & Blissmer, B. (2000). Self-efficacy determinants and consequences of physical activity. Exercise and Sport Sciences Reviews, 28(2), 85–88. McAuley, E., Mailey, E.L., Mullen, S.P., Szabo, A.N., Wójcicki, T.R., White, S.M., et al. (2011). Growth trajectories of exercise self-efficacy in older adults: Influence of measures and initial status. Health Psychology, 30(1), 75–83. McAuley, E., Wraith, S., & Duncan, T.E. (1991). Self-efficacy, perceptions of success, and intrinsic motivation for exercise. Journal of Applied Social Psychology, 21(2), 139–155. McDonald, R.P. (1999). Test theory: A unified treatment. Mahwah, NJ: Lawrence Erlbaum Associates Inc.

732 Mullen

Mishel, M. (1999). Uncertainty in chronic illness. Annual Review of Nursing Research, 17(1), 269–294. Mujika, I., & Padilla, S. (2001). Cardiorespiratory and metabolic characteristics of detraining in humans. Medicine and Science in Sports and Exercise, 33(3), 413–421. Mullen, S.P., & Whaley, D.E. (2010). Age, gender, and fitness club membership: factors related to initial involvement and sustained participation. International Journal of Sport and Exercise Psychology, 8(1), 24–35. Muthén, B. (2003). Statistical and substantive checking in growth mixture modeling: Comment on Bauer and Curran (2003). Psychological Methods, 8, 369–377. Muthén, L., & Muthén, B. (2011). Mplus user’s guide [Computer software and manual]. Los Angeles: Muthén & Muthén. Nylund, K., Asparouhov, T., & Muthén, B. (2007). Deciding on the number of classes in latent class analysis and growth mixture modeling: A Monte Carlo simulation study. Structural Equation Modeling, 14(4), 535–569. O’Connor, B.P., Rousseau, F.L., & Maki, S.A. (2004). Physical exercise and experienced bodily changes: The emergence of benefits and limits on benefits. International Journal of Aging & Human Development, 59(3), 177–203. Pelham, B.W. (1991). On confidence and consequence: The certainty and importance of self-knowledge. Journal of Personality and Social Psychology, 60(4), 518–530. Ram, N., & Grimm, K.J. (2009). Methods and measures: Growth mixture modeling: A method for identifying differences in longitudinal change among unobserved groups. International Journal of Behavioral Development, 33(6), 565–576. Rise, J., Sheeran, P., & Hukkelberg, S. (2010). The role of self-identity in the theory of planned behavior: A meta-analysis. Journal of Applied Social Psychology, 40(5), 1085–1105. Rothman, A.J. (2000). Toward a theory-based analysis of behavioral maintenance. Health Psychology, 19(1, Suppl) 64–69. Ruppar, T.M., & Schneider, J.K. (2007). Self-reported exercise behavior and interpretations of exercise in older adults. Western Journal of Nursing Research, 29(2), 140–157. Samuelson, K.M., & Mitchell Dayton, C. (2010). Latent class analysis. In G.R. Hancock & M.O. Mueller (Eds.), The reviewer’s guide to quantitative methods in the social sciences (pp. 173–184). New York: Taylor & Francis. Sheeran, P., & Abraham, C. (2003). Mediator of moderators: Temporal stability of intention and the intention-behavior relation. Personality and Social Psychology Bulletin, 29(2), 205–215. Sheldon, K., & Elliot, A. (1999). Goal striving, need satisfaction, and longitudinal wellbeing: The self-concordance model. Journal of Personality and Social Psychology, 76(3), 482–497. Strachan, S.M., Brawley, L.R., Spink, K., & Glazebrook, K. (2010). Older adults’ physicallyactive identity: Relationships between social cognitions, physical activity and satisfaction with life. Psychology of Sport and Exercise, 11(2), 114–121. Strachan, S.M., Brawley, L.R., Spink, K.S., & Jung, M.E. (2009). Strength of exercise identity and identity-exercise consistency: Affective and social cognitive relationships. Journal of Health Psychology, 14(8), 1196–1206. Sweet, S.N., Tulloch, H., Fortier, M.S., Pipe, A.L., & Reid, R.D. (2011). Patterns of motivation and ongoing exercise activity in cardiac rehabilitation settings: A 24-month exploration from the TEACH study. Annals of Behavioral Medicine, 42, 55–63. Tofighi, D., & Enders, C. (2008). Identifying the correct number of classes in growth mixture models. In G.R. Hancock & K.M. Samuelson (Eds.), Advances in latent variable mixture models (pp. 317–341). Charlotte, NC: Information Age Publishing. Umstattd, M.R., Wilcox, S., & Dowda, M. (2011). Predictors of change in satisfaction with body appearance and body function in mid-life and older adults: Active for Life. Annals of Behavioral Medicine, 41, 342–352.


Vandenberg, R.J., & Lance, C.E. (2000). A review and synthesis of the measurement invariance literature: Suggestions, practices, and recommendations for organizational research. Organizational Research Methods, 3(1), 4–70. Whaley, D., & Ebbeck, V. (2002). Self-schemata and exercise identity in older adults. Journal of Aging and Physical Activity, 10, 245–259. Wise, J.B., Posner, A.E., & Walker, G.L. (2004). Verbal messages strengthen bench press efficacy. Journal of Strength and Conditioning Research, 18(1), 26–29. Wise, J.B., & Trunnell, E.P. (2001). The influence of sources of self-efficacy upon efficacy strength. Journal of Sport & Exercise Psychology, 23(4), 268–280. 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, 64B(1), 33–40. Wu, W., & West, S.G. (2010). Sensitivity of fit indices to misspecification in growth curve models. Multivariate Behavioral Research, 45(3), 420–452. Zinbarg, R., Yovel, I., Revelle, W., & McDonald, R. (2006). Estimating generalizability to a universe of indicators that all have one attribute in common: A comparison of estimators for omega. Applied Psychological Measurement, 30, 121–144. Manuscript submitted: October 11, 2010 Revision accepted: June 26, 2011