Edited by: Deborah L. Feltz, Michigan State University, USA
Reviewed by: Ben Jackson, University of Western Australia, Australia; Khaled Said Mokhtar Hegazy, Karlsruhe Institute of Technology, Germany; Brandon Irwin, Kansas State University, USA
*Correspondence: Darko Jekauc, Department for Sport Psychology, Faculty of Humanities and Social Sciences, Institute for Sport Science, Humboldt University of Berlin, Unter den Linden 6, Berlin 10099, Germany e-mail:
This article was submitted to Movement Science and Sport Psychology, a section of the journal Frontiers in Psychology.
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
In the processes of physical activity (PA) maintenance specific predictors are effective, which differ from other stages of PA development. Recently, Physical Activity Maintenance Theory (PAMT) was specifically developed for prediction of PA maintenance. The aim of the present study was to evaluate the predictability of the future behavior by the PAMT and compare it with the Theory of Planned Behavior (TPB) and Social Cognitive Theory (SCT). Participation rate in a fitness center was observed for 101 college students (53 female) aged between 19 and 32 years (
Participation in physical activity (PA) is associated with a variety of health benefits and a reduction in chronic diseases (Warburton et al.,
According to Rothman (
Before effective interventions can be developed the determinants of PA maintenance should be known. In this regard, theory-based interventions are more effective than approaches lacking a theoretical underpinning (Michie and Abraham,
According to the TPB, behavior is a function of a person's intention to perform behavior, which reflects the level of motivation toward performing behavior and represents the most proximal predictor of behavior. Intention is supposed to be determined by attitude, subjective norm, and perceived behavioral control. There is considerable empirical evidence that supports the applicability of the TPB to explain PA intentions and behavior (Godin and Kok,
Likewise, widely recognized and frequently applied in the field of PA behavior, the SCT describes factors influencing behavior (Bandura,
The PAMT can be seen as further development and a specification of the SCT in the context of the physical activity maintenance process. Considerably less attention was devoted to the PAMT which was originally developed as a theoretical framework for supporting PA maintenance interventions. Nigg et al. (
To our knowledge, to date there is no empirical study testing the assumptions of the PAMT. However, the single components of the PAMT have been empirically tested. The effectiveness of self-efficacy as a predictor of PA maintenance has been shown above. In several studies, goal setting was also shown to be a significant predictor of PA maintenance (e.g., Annesi,
The literature on determinants of PA maintenance reveals several limitations. First, a great part of the reviewed studies lack theoretical underpinning and separately analyze the determinants. This shortcoming impedes an understanding of the mechanisms at work. Second, there are only a few longitudinal studies with more than two measurement occasions, which are necessary for the analysis of PA patterns over time (cf. Armitage,
The purposes of the present study were (i) to describe the development of PA attendance, (ii) to predict PA participation patterns by the TPB, SCT and PAMT, and (iii) to compare the predictive power across the three theories. We assume that the predictive power of the PAMT will be superior to the TPB and the SCT as this theory is specified on two levels: (a) specified for the context of physical activity and (b) specified for the maintenance process as compared to the TPB and the SCT.
Participants were 101 (48 males and 53 females) college students and members of a fitness center. Age ranged from 19 to 32 years with an average age of 23.6 years (SD ± 2.9). Fitness center membership and, consequently, the possession of a magnetic card for the devices in the fitness center were inclusion criteria for this study. One week in advance of the start of this observational study, all members of the fitness center were informed by email about the study. A questionnaire measuring socio-demographic and psychological variables was distributed in the first two weeks of the semester. Participants were asked to complete the paper-and-pencil-questionnaire directly at the information center which is located at the entrance of the fitness center. Fitness center attendance was registered electronically for 20 consecutive weeks. Ethical approval for the study was provided by the University and all participants signed an informed consent form at the beginning of the study.
Participation frequency was assessed electronically by a magnetic register system which was activated when the participants used the machines. Each participant had to use the magnetic card when using the fitness center. The data were summarized as the frequency of the weekly participation. The week was defined from Monday to Sunday. The maximal score for 1 week could be 7 (when a participant visited the fitness center every single day of the week) and the minimal score 0 (when a participant did not even visit the fitness center once during the week).
All questionnaires assessing the constructs of the TPB were based on guidelines provided by Ajzen and Madden (
The development of the fitness center attendance over 20 weeks was analyzed using a latent class zero-inflated Poisson growth curve model (Long,
Maximum Likelihood was used for parameter estimation. Because no-shows were analyzed as part of the model, the dependent variable contains no missing values. For the independent variables, the proportion of missing data was 1.08% (range from 0.00 to 3.10% across variables). Parameters and standard errors (SEs) were estimated for initial status (i.e., the latent factor at baseline) and change trajectories (i.e., linear and nonlinear trends).
Figure
Age | 24.3 | 2.7 | |||||||||
Gender (0 = m; 1 = w) | 0.49 | 0.50 | |||||||||
Age | 22.9 | 3.3 | |||||||||
Gender (0 = m; 1 = w) | 0.57 | 0.49 | |||||||||
1 | Exercise frequency | 0.99 | 0.65 | −0.18 | 0.09 | 0.21 |
0.28 |
||||
2 | Intention | 5.35 | 2.50 | −0.09 | 0.35 |
0.47 |
|||||
3 | Subjective norm | 17.14 | 3.24 | −0.01 | −0.13 | ||||||
4 | Attitude | 9.20 | 3.21 | 0.12 | |||||||
5 | PBC | 10.81 | 3.83 | ||||||||
1 | Exercise frequency | 0.00 | −0.02 | −0.11 | 0.30 |
−0.30 |
|||||
2 | Self-efficacy | 35.92 | 6.72 | 0.18 | 0.12 | −0.14 | −0.26 |
||||
3 | Outcome expect. | 53.42 | 7.85 | 0.11 | 0.18 | 0.23 |
|||||
4 | Soc. Support family | 10.89 | 4.45 | 0.21 |
0.03 | ||||||
5 | Soc. Support friends | 14.17 | 5.30 | 0.13 | |||||||
6 | Barriers | 15.23 | 3.48 | ||||||||
1 | Exercise frequency | 0.08 | 0.12 | 0.01 | −0.04 | 0.09 | 0.16 | 0.27 |
−0.21 |
||
2 | Satisfaction | 34.42 | 5.76 | 0.88 |
0.70 |
0.21 |
0.57 |
0.36 |
0.16 | −0.02 | |
3 | Attainment | 33.93 | 5.58 | 0.77 |
0.25 |
0.63 |
0.36 |
0.13 | 0.07 | ||
4 | Commitment | 36.30 | 6.00 | 0.31 |
0.58 |
0.27 |
0.07 | 0.08 | |||
5 | Self-motivation for maintenance | 37.31 | 5.99 | 0.26 |
0.40 |
0.05 | −0.08 | ||||
6 | Self-motivation pros/cons | 21.19 | 5.11 | 0.31 |
0.20 | 0.21 | |||||
7 | Self-efficacy barrier | 20.01 | 4.28 | 0.40 | −0.09 | ||||||
8 | Self-efficacy relapse | 11.44 | 2.18 | −0.14 | |||||||
9 | Life stress | 5.69 | 5.97 |
To identify distinct categorical participation patterns, a latent class zero-inflated Poisson growth curve analysis was carried out. Preliminary analyses showed that a second order polynomial growth curve model provided a good description of the participation rate over the 20 weeks. Based on this preliminary analysis, three different models were estimated in which the variance of the intercept, linear slope, and, quadratic slope in the zero-part, as well as in the count-part, were constrained to zero: a one-class model (1), a two-class model (2), and a three-class model (3). Model comparisons are presented in Table
Free parameters | 6 | 13 | 20 |
AIC | 3509.40 | 3212.68 | 3150.05 |
BIC | 3525.10 | 3246.67 | 3202.25 |
Sample adjusted BIC | 3506.14 | 3205.61 | 3139.18 |
Entropy | 0.92 | 0.90 | |
Vuong, Lo, Mendell, Rubin | n.a. | ||
Model test | 2 v 1 | 3 v 2 | |
−2LL difference | 310.73 | 76.63 | |
0.00 | 0.52 | ||
Lo, Mendel, Rubin adjusted | n.a. | ||
Model test | 2 v 1 | 3 v 2 | |
−2LL difference | n.a. | 301.40 | 74.33 |
n.a. | 0.00 | 0.52 | |
c1 = 101 | c1 = 46; | c1 = 37; | |
c2 = 55 | c2 = 20; | ||
c3 = 44 |
The estimates for the latent class zero-inflated Poisson growth curve model for the two-class solution is presented in Table
Intercept | 1.55 | 0.11 | 5.19 | 0.00 | ||
Lin traj | 0.02 | 0.02 | 0.67 | 0.51 | ||
Quadr. traj. | 0.00 | 0.00 | −1.32 | 0.19 | ||
Intercept | −0.70 | 0.43 | −1.63 | 0.10 | 0.50 | 0.33 |
Lin. traj. | 0.85 | 0.48 | 1.76 | 0.08 | ||
Quadr. traj. | −0.57 | 0.18 | −3.10 | 0.00 | ||
Intercept | −0.24 | 0.22 | −1.09 | 0.28 | ||
Lin. traj. | 0.21 | 0.07 | 2.97 | 0.00 | ||
Quadr. traj. | −0.02 | 0.01 | −4.14 | 0.00 | ||
Intercept | 1.01 | 0.49 | 2.07 | 0.04 | 2.73 | 0.73 |
Lin. traj. | −0.01 | 0.28 | −0.04 | 0.97 | ||
Quadr. traj. | −0.12 | 0.05 | −2.51 | 0.01 |
For Class 2, the attendance rate for the first week is not significantly different from zero. However, both the linear and quadratic trajectories are significant, indicating that the participants' attendance rate changes over the consecutive 20 weeks. In the first week, the attendance rate is very low but increases continuously in the consecutive weeks. The maximum is reached in the seventh week, after that the attendance rate only decreases continuously. The probability of not attending the fitness center in the first week is 73%. The quadratic trajectory of the likelihood for not attending the fitness center is with a value of 0.12, significantly different from zero, suggesting a u-shaped function for the likelihood of non-attendance. The participants in this class can be seen as intermittent exercisers.
Class membership was predicted using logistic regression analysis. The same two-class-model as described above was applied separately for all three prediction theories. For each theory, a similar developmental pattern of participation was found in both classes. In the class of regular attenders, the intercept differed significantly from zero and non-significant linear and quadratic trajectories of participation rate were observed. In the class of intermittent exercisers, significant linear and quadratic trajectories for participation rate were found.
In the logistic regression presented in Table
Intercept | 0.25 | 0.24 | 1.06 | 0.15 | |
Intention | −0.26 | 0.70 | −0.94 | 0.83 | 0.77 |
Subj. norm | −0.27 | 0.23 | −1.2 | 0.88 | 0.76 |
Attitude | 0.42 | 0.24 | 1.76 | 0.04 | 1.52 |
PBC | 0.55 | 0.26 | 2.14 | 0.02 | 1.74 |
Nagelkerke's pseudo |
|||||
Intercept | 0.25 | 0.24 | 1.02 | 0.16 | |
Goals | 0.32 | 0.24 | 1.31 | 0.10 | 1.38 |
Self-efficacy | 0.47 | 0.28 | 1.66 | 0.04 | 1.61 |
Outcome expectancies | −0.11 | 0.24 | −0.29 | 0.67 | 0.90 |
Barriers | −0.85 | 0.28 | −2.96 | 0.00 | 0.43 |
Social support family | −0.07 | 0.24 | −0.29 | 0.62 | 0.93 |
Social support friends | 0.47 | 0.25 | 1.87 | 0.03 | 1.60 |
Nagelkerke's pseudo |
|||||
Intercept | −0.43 | 0.29 | −1.51 | 0.07 | |
Commitment | −0.25 | 0.36 | −0.70 | 0.75 | 0.78 |
Attainment | 0.92 | 0.59 | 1.56 | 0.06 | 2.50 |
Satisfaction | −0.67 | 0.54 | −1.25 | 0.89 | 0.51 |
Self-motivation PA maint. | −0.16 | 0.27 | −0.58 | 0.72 | 0.86 |
Self-motivation pros/cons | 0.14 | 0.32 | 0.44 | 0.33 | 1.15 |
Self-efficacy barrier | 0.27 | 0.30 | 0.89 | 0.19 | 1.31 |
Self-efficacy relapse | 0.24 | 0.26 | 0.90 | 0.38 | 1.27 |
Life stress | −0.92 | 0.36 | −2.54 | 0.01 | 0.40 |
Nagelkerke's pseudo |
The results of the logistic regression (see Table
In the PAMT, only one variable significantly predicted membership in both classes: life stress. Increasing perceived life stress by one standard deviation led to a decrease in the odds of being a member of the class of regular attenders by 2.5 (=1/0.40) (Nagelkerke's pseudo
TPB as well as SCT are among the most widely used theoretical frameworks to explain and predict exercise behavior. Considerably less attention was devoted to the PAMT, which was explicitly developed to explain PA maintenance. The purposes of the present study were (i) to describe the development of PA participation over time, (ii) to predict development of PA participation by the TPB, SCT, and PAMT, and (iii) to compare the predictive power of the three theories.
In order to describe the development of exercise behavior over 20 weeks a latent class zero-inflated Poisson growth curve analysis was carried out. The procedure identified two different participation patterns. In the class of regular attenders, participants exercised on average 1.55 times in the first week with a probability of not attending the fitness center of 33%. In the 19 following weeks, their participation rate did not significantly change indicating that they maintained their physical activity at a comparable level. On the contrary, strong fluctuations in exercise behavior were observed in the class of intermittent exercisers. At the beginning of the semester, their participation rate was not different from zero with a probability of not attending the fitness center of 73%. However, they increased their attendance rate during the semester reaching a maximum in the middle of the semester (i.e., the seventh week). Thereafter, their participation rate decreased continuously and the probability of no attendance increased at the same time. At the beginning of the semester, the students probably dispose of more time to devote to exercise and can therefore increase their attendance rate until the middle of semester. However, as the end of the semester approaches and the exam stress increases, the intermittent exercisers fail to exercise continuously.
The difference between these two behavior patterns were explained in terms of three theories: for the TPB, it could be shown that the regular participation pattern was associated with a positive attitude toward attending the fitness center and higher perceived behavioral control. However, these results are not fully in line with the assumptions of the TPB because behavioral intention—as an immediate and most important predictor—could not differentiate between regular attenders and intermittent exercisers. These results are consistent with findings of the study of Armitage (
In the SCT, three predictors were identified to significantly predict exercise behavior. In accordance with several studies on maintenance (Plotnikoff et al.,
PAMT is a rather new theory and was specifically developed to explain maintenance in PA behavior. The only significant predictor in the PAMT is life stress indicating that participants with high levels of life stress have higher probability to belong to the group of intermittent exercisers. Life stress might lead to distractions from regular PA behavior and impede the habituation process. However, all other predictors such as goal setting, self-motivation, and self-efficacy were not significant predictors of the exercise attendance. The variables of this theory are considerably correlated with each other showing a Nagelkerke's
When comparing the three theories, SCT had the greatest predictive power with Nagelkerke's
The results of this study suggest that impeding aspects such as life stress and barriers might have the greatest predictive power in the process of exercise maintenance. In the SCT as well as in PAMT, impeding aspects were the strongest predictors of exercise patterns. Presumably, such external factors play a more important role in the process of PA maintenance than predicted by the presented models. Several studies have shown that barriers might be associated with sedentary lifestyle and prevent exercise maintenance (e.g., Salmon et al.,
Additionally, when compared to social support from family members, social support from friends might be more important for college students as they normally leave their parental home for studying and have not yet started their own families. Interestingly, subjective norms in the TPB did not contribute to the prediction of exercise attendance. It could be shown consistently for all three theories that goal setting and intentions were not predictive of exercise pattern. It might be that all participants who start to exercise have intentions to exercise regularly but the intention changes over a short period of time as shown by Conroy et al. (
This study has a number of limitations. First, the sample size (101 participants) is relatively small to detect more complex participation patterns. Possibly, a greater sample size would facilitate identification of more detailed participation patterns. Second, predictor variables were measured only at baseline, meaning that week-by-week changes of predictors were not assessed. Third, the sample of the study is highly selective as it contains only college students. This aspect might impair the generalizability of this study. Forth, some scales showed only sufficient internal consistency. This aspect could negatively affect the analyzed effects.
Nonetheless, this study has several merits. First, to our knowledge, the present study is the first study to empirically evaluate PAMT and to compare it to SCT, and TPB in the context of exercise attendance. The dependent variable attendance (rate) was objectively measured by a magnetic card system, which is a reliable and valid assessment of exercise attendance. Finally, sound statistical methods as zero-inflated Poisson growth curve analyses were used to describe participation patterns, which are more informative than the simple prediction of the total number of fitness center attendances.
This study identified two patterns of exercise attendance: regular attenders and intermittent exercisers. SCT showed greatest predictive power compared to PAMT and TPB. Our hypothesis that the more specified PAMT would predict the exercise patterns more precisely than the rather global theories could not be confirmed. Impeding aspects such as life stress and barriers were the strongest predictors of the participation patterns suggesting that overcoming barriers might be an important aspect in the process of exercise maintenance. Future studies should examine PA maintenance and put stronger focus on these impeding external variables. Self-efficacy, perceived behavioral control and social support were shown to differentiate between regular attenders and intermittent exercisers whereas intentions and goals were not significant predictors. Due to the modest predictive power of all three theories an extension of the theories seems inevitable. Accordingly, the past behavior or affective determinants of behavior, among others, should be taken into account.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.