The quality and type of exercise motivation play a vital role in the success or failure of healthy behavioral change. Self-determination theory (SDT) provides a framework for explaining human behavior and motivation in sports and exercise (Deci and Ryan, 1985, 2000). According to SDT, there are three main types of human motivation in the field of exercise—namely, amotivation, extrinsic motivation, and intrinsic motivation. These three motivations are distributed on the motivation continuum according to the degree of self-determination. Amotivation and intrinsic motivation are located at the two extremes of the continuum, representing the most controlled and autonomous forms of motivation, respectively. Extrinsic motivation, situated between amotivation and intrinsic motivation on the continuum, refers to performing activities for instrumental reasons or achieving outcomes separate from the behavior itself. In the sub-theory of SDT, organic integration theory (Ryan and Deci, 2000), external motivation is further conceptualized into four types—namely, external regulation, introjected regulation, identified regulation, and integrated regulation. In the physical exercise domain, these types differ in their relative autonomy. External regulation occurs when individuals engage in exercise to fulfill external demands, obtain rewards, or avoid punishment, such as exercising in order to please others. Introjected regulation occurs when exercise is performed to avoid feelings such as guilt or shame or to enhance ego and feelings of self-worth, such as exercising in order to lose weight or improve body shape. These two external motives represent those described as controlled forms in SDT, which are not or are only partially internalized. Although they sometimes regulate (or motivate) short-term exercise behavior, they do not sustain maintenance over time (Deci and Ryan, 1985). In contrast, identified regulation and integrated regulation represent a more autonomous form of exercise regulation. Identified regulation exists when an individual participates in exercise not because of the fun and satisfaction of the behavior itself but instead because of its recognized health value and utility (Ryan et al., 2009). Integrated regulation is a type of regulation in which an individual regards exercise not only as individually essential but also as congruent with deeply held values and sense of self. However, exercise undertaken to achieve an external goal—creating or confirming one’s identity—is still considered external regulation.

Deci and Ryan (2008) suggested that more self-determined regulation is associated with greater persistence, enhanced performance, and better well-being. Similar to behavioral regulation, different types of motives may differ in their degree of internal or external orientation (Markland and Ingledew, 2007). Within the exercise domain, external goals (e.g., weight loss, appearance) are always considered to be associated with less self-determined regulation, while internal goals (e.g., health, affiliation) are always associated with more self-determined regulation (Gillison et al., 2006; Ingledew and Markland, 2008). In the MPAM-R, exercise for health and fitness is classified as controlled-oriented motives, and exercise for social interaction and competition is classified as autonomous motivation. Only the latter can significantly predict subsequent PA. However, unlike behavioral regulation, PA motives deal with the goal content—that is, they focus on what exercise-related goals people want to pursue (Deci and Ryan, 2000). On the contrary, behavioral regulation is more concerned with the why of goal pursuit—that is, the autonomous and controlled motivations that guide people’s efforts to achieve a certain goal.

This research project sought to develop and validate a Chinese University Students’ Physical Activity Motivation Scale (CUSPAMS) to measure various forms of motivational content geared toward performing healthy behaviors in the SDT paradigm. Figure 1 uses the form of a continuum to show the antecedents, perceived autonomy, and internalization of these different content motivations. In addition, it also describes the differences and connections between the MPAM-R, PALMS, and CUSPAMS in measuring motivational content and highlights the extent of relative autonomy of these motivations. The following hypotheses were put forward: In terms of structural validity, we assumed that the CUSPAMS would have nine factors and that the motivation goals mentioned in SDT would receive responses by the sample of Chinese university students (H1). As for criterion validity, two hypotheses were formulated: We expected that the CUSPAMS would be correlated with the selected scales: the revised version of the Behavioral Regulation in Exercise Questionnaire (BREQ-2; Markland and Tobin, 2004) and the Self-Efficacy for Exercise (SEE) Scale (Resnick and Jenkins, 2000). Specifically, we expected that the factors in the CUSPAMS, such as enjoyment, mastery, affiliation, and competition, would be highly positively correlated with the autonomous form of motivational regulation (e.g., identified and intrinsic motivation), while appearance, policy-related motives, and others’ expectations would be highly positively correlated with controlled forms of motivational regulation (e.g., amotivation, external and introjected regulation; H2). Based on the literature report (McAuley et al., 1991c1994; Kavussanu and Roberts, 1996; Slovinec D’Angelo et al., 2014), we also expected that in the CUSPAMS, more autonomous motivation factors would have a higher positive correlation with exercise self-efficacy, while controlled motivation factors would have a lower positive correlation or no correlation (H3).

The CUSPAMS was designed to assess types of motivation among university students engaged in PA. In the current study, the scale development process included the following procedures. First, a comprehensive iterative review and in-depth content analysis were conducted in order to identify and describe the different aspects of university students’ motivation to participate in PA. Depending on common agreement among studies on PA motivation and drawing on studies of the motivation of Chinese scholars, nine dimensions were proposed: (1) stress management, (2) competition, (3) appearance, (4) affiliation, (5) enjoyment, (6) others’ expectations, (7) mastery, (8) health benefits, and (9) policy intervention.

Second, once the dimensions were identified, item pools were generated using the “deductive” method recommended by Raykov and Marcoulides (2011)—that is, through a literature review and assessment of existing scales and indicators in the field of sports and exercise. The authors of this study invited five experts from the fields of PE, sport psychology, linguistics, and Chinese school PE policy to conduct three rounds of online meetings to discuss and determine the items in the new scale. In the first meeting round, the expert panel reviewed the previous 16 motivational scales and concluded that the items in the MPAM-R and PALMS were a suitable basis for item development. There were three reasons for this conclusion: First, the two scales are derived from classic SDT as the theoretical basis for developing the structure of the scale; second, the items in these two scales describe specific motivational goals in the PA domain; third, these two scales have demonstrated good psychometric characteristics in different cultures. In the second meeting round, the panel compared the verbal expression of both scales to extract a series of items that reflected the same factors but used different terminologies in these two scales. For example, the statement “Because I want to be physically fit” in the MPAM-R is similar to “be physically fit” in the PALMS, as both reflect the dimension of health motivation—therefore, this item was retained. Items that did not reflect the actual context of Chinese university students’ PA participation, such as “to earn a living” and “because I get paid to do it” in the PALMS, were deleted after discussion between the experts. Besides, with reference to the views put forward by Chinese scholars (Chen et al., 2008; He and Yang, 2018), the authors drafted four policy intervention motivation items. Then, the expert panels, especially policy experts and linguists, discussed and reviewed the content and sentences of the four items. In the third meeting round, the panel members conducted a comprehensive evaluation of the conceptual scope, semantic equivalence, clarity, readability, relevance, and conciseness of the initial items. Three items with low clarity were deleted in this round. After all panel members agreed that none of the remaining items required further modification, 41 items were generated from three extraction methods (joint extraction of two scales, item extraction from a single scale, and self-developed items from the Chinese literature; Table 3), together comprising the first version of the CUSPAMS.

Finally, the face validity of the new scale items was obtained through extensive online comments by 20 Chinese university students (face validity). The purpose was to determine whether they understood the items and were able to respond to them. Their feedback indicated that no further modifications to the scale were needed.

The initial version of the CUSPAMS was completed by 641 university students (194 males and 447 females), with a mean age of 20.23 ± 1.46 years, from five universities in five different provinces located in central China. Of these students, 301 were freshmen (46.9%), 176 were sophomores (27.4%), 56 were juniors (8.7%), and 88 were seniors (13.7%). Various majors were represented by these students. The most common physical activities reported were walking (n = 200), jogging (n = 175), and basketball (n = 32). The participants reported an average of 1.68 ± 1.66 h of PA per week. For subsequent data analysis, the participants were randomly split into two subsamples: Sample 1 (n = 311; male = 89, female = 222; M = 20.17 ± 1.39) was used to develop the initial scale and to examine its underlying factor structure by exploratory factor analysis (EFA); Sample 2 (n = 330; male = 105, female = 225; M = 20.28 ± 1.53) was used to test for data fit by confirmatory factor analysis (CFA). To test the model stability and criterion validity, 574 university students (Sample 3; Mage = 19.21 ± 1.23, male = 314) from a public university in Henan Province voluntarily completed a survey, including demographic information, and three scales at the same time. To measure the test–retest reliability of the CUSPAMS, each participant in Sample 3 was assigned a unique questionnaire code, and each was advised to refill in the same questionnaire using their unique code 21 days later, i.e., to complete the survey a second time. Finally, 177 participants were randomly selected from Sample 3 for test–retest process.

In this study, we established the following inclusion criteria for the participants: (1) had to be a registered university student, (2) had to be at least 18 years old, (3) had to be non-sport/PE major students, (4) had to be Chinese, (5) had to have a strong comprehension of the Chinese language in terms of reading, speaking, and writing, and (6) had to regularly participate in physical activities at least once per week for the past 3 months. And, finally, since the questionnaire was conducted online, the participants (7) had to be smartphone users. The exclusion criteria were as follows: (1) cannot participate in regular exercise due to physical disability, or (2) unwilling to participate in the study.

Hall et al. (2020) mentioned that the ongoing impact of the COVID-19 pandemic is changing people’s exercise behavior. A long-term home isolation lifestyle has led to more physical inactivity and increased sedentary behavior. To reduce these impacts as much as possible, the online questionnaire was distributed in August 2020, at which time it had been at least 3 months since the participants’ lives had returned to normal in China. University students had also returned to their respective universities. The questionnaire was distributed through “Questionnaire Star,” a mobile application specializing in online surveys that is widely used in China. Prior to online survey administration, ethical approval was obtained from the Ethics Committee of the University of Malaya (UM. TNC2/UMREC-976). Since the actual data collection was in China, research permission was also obtained from the Student Affairs Office of the five targeted Chinese universities. The researchers then contacted administrators and PE lecturers and explained the objectives and other vital information related to the study. With their approval, they shared a QR code that included all online questionnaires to WeChat (a popular social chat application in China, similar to WhatsApp) groups of students from different study years and majors. These WeChat groups were created by administrators and PE lecturers after the students had enrolled in the university with the purpose of sharing physical class or academic-related information. Online consent to participate was obtained when the participants volunteered to scan the QR code and complete the online questionnaire. The participants were fully informed that they could withdraw from the study at any time during the process of completing the online questionnaires and that all of the information they provided would remain confidential. The online questionnaire took approximately 3–6 min to complete. After the participants had completed the CUSPAMS online questionnaire, their responses were automatically saved in the “Questionnaire Star” application, after which the researchers downloaded the collected data directly from the application.

Data collection for Sample 3 occurred in November 2021 using a method similar to that for Sample 1 and Sample 2. In addition to completing the CUSPAMS, Sample 3 also participated in two other instruments to test the criterion validity of the CUSPAMS, including the BREQ-2 developed based on SDT and the SEE Scale. In previous studies, behavioral regulation and self-efficacy were demonstrated to have a strong correlation with motivation types (Mullan and Markland, 1997; Ryan and Deci, 2000; Teixeira et al., 2012).

The participants reported key demographic information, including gender, age, university affiliation, major/field of study, and number of years in university. They also reported the frequency of their regular weekly PA, its duration, and its intensity during the past 3 months.

The newly developed 41-item CUSPAMS was used to identify the perceived reasons for participating in PA. The participants were asked to respond to questions such as the following: “Why do you participate in physical activity?” The online scale included motives for nine factors. For example, “Because I want to be physically fit” is an example of an item from the health benefits factor (six items); “Because I like to engage in activities that physically challenge me” is an example of an item from the mastery factor (six items); “Because I want to compete with other people around me” is an example of an item from the competition factor (four items). All items were based on a 7-point Likert scale ranging from 1 (not at all true for me) to 7 (very true for me), indicating the degree to which each motive was personally true for each participant with respect to primary PA.

The BREQ-2 (Markland and Tobin, 2004) was employed to measure behavioral regulation in exercise and has been found to be reliable and valid among Chinese university students (Liu et al., 2015). The C-BREQ-2 consists of 18 items with responses on a 5-point Likert scale ranging from 0 (not true for me) to 4 (very true for me). Five subscales are included in this scale: amotivation, with four items (e.g., “I do not see why I should have to exercise”), external regulation, with four items (e.g., “I exercise because others will not be pleased with me if I do not”), introjected regulation, with three items (e.g., “I feel like a failure when I have not exercised in a while”), identified regulation, with three items (e.g., “It’s important to me to exercise regularly”), and intrinsic regulation, with four items (e.g., “I get pleasure and satisfaction from participating in exercise”). A higher score on the identified regulation and intrinsic motivation subscale indicated that the exerciser’s motivational form tended to be more autonomous. The Cronbach’s α for the C-BREQ-2 was 0.90.

The SEE Scale was designed to evaluate people’s confidence to continue exercising in the face of barriers to exercise (Resnick and Jenkins, 2000). The reliability and validity of the Chinese version of the SEE (SEE-C) was provided by Lee et al. (2009). The SEE-C comprises nine items with ratings on a 10-point Likert scale ranging from 0 (not confident) to 10 (very confident). Higher scores indicate higher levels of self-efficacy. The Cronbach’s α for the SEE-C was 0.91.

Exploratory factor analysis and CFA are widely used in measurement applications for scale development and construct validation (Ye et al., 2018a; Chang et al., 2020; Fung and Fung, 2020). The data from the initial samples were randomly divided into two groups using SPSS, i.e., Sample 1 for EFA and Sample 2 for CFA. Descriptive statistics of the items of the CUSPAMS and the bivariate correlations between items were computed using data from Sample 1 (n = 311). Kolmogorov–Smirnov, skewness, and kurtosis tests were employed to check for data normality. The data were shown to be normally distributed. Before further analysis, Kaiser-Meyer-Olkin (KMO) and Bartlett’s sphericity tests were run to measure for sampling adequacy. Subsequently, the factorial structure of the CUSPAMS was tested using EFA, and internal consistency was estimated. There is currently no consensus on the sample size standard for variants of factor analysis but, according to suggestions by Pearson and Mundform (2010), a sample of 200 people was determined to be a sufficient minimum. The communality for each item was set to be greater than or equal to 0.40 (Leimeister, 2010) to confirm that each item shared some common variance with other items. Theoretical understanding and parsimony are considered when an item is loaded onto different factors (Kline, 2015). The reliability of the CUSPAMS was obtained by measuring the Cronbach’s α and the McDonald’s omega (ω; McDonald, 1999). This is because the latter can overcome some of the shortcomings of using α and thus represents one of the best measures of reliability (Yang and Green, 2011; Goodboy and Martin, 2020). Reliability measures less than 0.70 are considered moderate, 0.70–0.80 are considered sufficient, and more than 0.80 are considered good (Evers et al., 2010). In addition, we calculated the Pearson correlation coefficients between the CUSPAMS and the CUSPAMS completed after the 3-week interval to check test–retest reliability. The intraclass correlation coefficient value equal to or above 0.70 was considered acceptable (Weir, 2005).

Later, the structure of the CUSPAMS was further examined using CFA performed in SPSS AMOS version 22, based on data from Sample 2 (n = 330). CFA provides further evidence regarding the fitness of the suggested model with regard to the structure of the factors identified via EFA. The model parameters were estimated using the maximum likelihood function. Goodness of fit is a measure indicating how well a specified model reproduces the covariance matrix among the indicator variables (Hair et al., 2010). Multiple goodness of fit tests was used to evaluate the model’s fit to the data. To evaluate the fit of the models, we considered four indices of model fit: chi-square/degree of freedom (χ²/df), the comparative fit index (CFI), the Tucker-Lewis index (TLI), and the root mean square of error approximation (RMSEA). As the only true inference statistic that represents the model test (Markland, 2007), χ² has been identified as potentially problematic due to sample size sensitivity, but its value was still reported. Marsh and Hocevar (1985) have recommended that a χ²/df value between 2.0 and 5.0 indicates a reasonable macro structure. As for the CFI and TLI, they should be ≥0.90 to indicate a good fit (Hair et al., 1998). The RMSEA value should be <0.05, as it can be said to indicate a convergent fit to the analyzed data of the model, while it also indicates a fit close to good when it produces a fit value between 0.05 and 0.08 (Browne and Cudeck, 1992).

Initially, the factorability of the 41 items was examined. Consistent with Field’s (2013) suggestion, the bivariate correlation matrix of all items was analyzed, and no items with a bivariate correlation score greater than 0.80 were found, thus avoiding the occurrence of multi-collinearity among items. The KMO value was 0.90, and Bartlett’s test of sphericity was significant (χ² = 7,201.45, df = 820, p < 0.05), indicating that the samples met the criteria for factor analysis (Hair et al., 1998). However, item 24 (“Because it was prescribed by a doctor or physiotherapist”) and item 12 (“Because I want to define my muscles so that I look better”) showed communalities below 0.40. Osborne et al. (2008) stated that if communalities for a particular variable are low (between 0.00 and0.40), then that variable may struggle to load significantly on any factor. Thus, these two items were carefully observed in the subsequent EFA to determine whether to exclude or retain them. EFA was computed on the 41 items in the CUSPAMS for the 311 participants.

We follow the method provided by Tabachnick et al. (2007) to select the rotation of the factors. After requesting the oblique rotation, if the correlation between the factors exceeds 0.32, indicating that the variances between the factors partially overlap, it is more reasonable to use the oblique rotation, unless there is a convincing reason for the orthogonal rotation. Given that the correlation between the eight factors is between 0.20 and 0.40, a principal axis analysis with Promax rotation was performed. In addition, items that were found to cross-load or factor loadings lower than 0.40 were removed (Field, 2009), as were items that loaded ambiguously (i.e., the literal meaning of an item was clearly different from those of other items in the same factor). The remaining items underwent subsequent analyses, and inspections of the loadings were conducted. This iteration process was repeated until each item loaded significantly on only one factor with no cross-loadings. For example, item 24 (“Because it was prescribed by a doctor or physiotherapist”), item 22 (“Because I want to meet new people”), and item 14 (“Because it is fun”) were deleted due to factor loadings lower than 0.40, while item 16 was deleted due to ambiguity. Thirty-seven items remained following this procedure. After rerunning EFA (Table 4), eight factors emerged with eigenvalues of 1.00 (Nunnally, 1978), accounting for 67.03% of the variance, exceeding the minimum acceptable target of 60% for scale development (Hinkin, 1998). The examination of the factor structure revealed eight clearly distinct sets of items reflecting psychological feelings, sport/physical education policy, others’ expectations, appearance, affiliation, mastery, health benefits, and competition.

From the output of the EFA, two issues emerged to which we paid attention. First, all items clearly loaded on different factors. However, the originally designed two factors related to “stress management” and “enjoyment” converged into one factor, a result that was somewhat different from the expected factor structure. From a definition point of view, enjoyment is a positive emotion experienced by an individual after a satisfying activity (Abraham, 1943). While stress management implies a more specific process of cognitive appraisal to determine whether an individual believes he or she has the resources to respond effectively to the challenges of a stressor or a change (Folkman and Lazarus, 1988), it must be mentioned that both factors fall under the umbrella of psychological feelings. Therefore, we temporarily named this factor “psychological condition” in the eight-factor model in the subsequent analysis. However, this factor structure based on the EFA contradicted the traditional SDT classification of motivation types. In the early developmental stage of SDT, Deci (1975) proposed that people engage in an activity because they enjoy the activity and categorized this type of motivation as part of intrinsic motivation. Subsequently, this type of motivation is common among people from different social systems, cultures, and nationalities (Frederick and Ryan, 1993; Chen et al., 2006; Kueh et al., 2019). Stress management, on the other hand, is a type of extrinsic motivation related to the body (Rogers and Morris, 2003). Interestingly, items that originated from two different motivation types converged into one factor in the EFA. We speculate that the reason for this is that the EFA involved a preliminary exploration to identify the underlying relationships between measured variables without imposing any preconceived structure on the outcome (Child, 1990).

Therefore, it was necessary to conduct a follow-up detailed review of the eight overlapping items. These eight items were rerun in EFA with the extraction of two fixed factor numbers to investigate whether the loadings on each factor made any sense. The final EFA output showed that four items related to stress management and four items related to enjoyment loaded onto two factors separately, which is consistent with the proposed structure. Hence, two form models (eight-factor model and nine-factor model) were proposed and later tested using CFA to produce the final structural model.

Second, it should be noted that only two items representing others’ expectations were generally considered weak or unstable. However, Eisinga et al. (2013) explained that it is not uncommon for a questionnaire to have no more than two indicators to gage a particular self-assessment. This situation might be caused by two reasons. One is that due to resource and survey time constraints, only a limited number of items were available to assess a specific structure. The other is that poor-quality items were removed from the limited item pool, resulting in a small number of items in the scale, with only two items occasionally remaining. Given the high factor loading of these two items, i.e., item 6 (“Because my friends want me to”) and item 15 (“Because people tell me I need to”), we retained them for subsequent analysis.

Summary statistics for the CFA of the tested eight-factor and nine-factor models are presented in Table 5. The factor “psychological condition,” composed of eight items in the eight-factor structure model, was split into two factors—“enjoyment” and “stress management”—in the nine-factor structure model. The factor loading of six items was increased (except item 32), which indicated that these items better reflected their respective latent variables and were highly correlated. By comparing the model matching index, it was obvious that all of the indices in the nine-factor model were higher than those in the eight-factor model, thereby demonstrating that the nine-factor model fit the data better. Considering the goodness of fit and the interpretability of solutions, the nine-factor model was the most appropriate for the current data.

Although the values for the RMSEA and the χ²/df in the initial nine-factor model were acceptable, the values on the CFI and TLI were below the minimum acceptable levels. Therefore, based on recommendations by Hair et al. (2010), items with loadings below 0.70 were screened. After removing item 9 (“Because I want to meet the physical activity time required by the university”), item 12 (“Because I want to define my muscles so that I look better”), item 2 (“Because I like to engage in activities that physically challenge me”), item 19 (“Because I want to improve my cardiovascular fitness”), and item 1 (“Because I want to be physically fit”) due to low factor loading values that were considered insignificant, the CUSPAMS was reduced further, from 37 items to 32 items, while maintaining sufficient representation of the original variables with lower loadings. For the other six items (items 6, 15, 27, 29, 40, and 41), because their loadings were close to 0.70, and because, if deleted, their corresponding factors would only have two items, which would affect the subsequent internal consistency analysis (Hair et al., 2006), we decided to retain them. The modified CUSPAMS with 32 items revealed a data fit of χ²/df = 2.359, CFI = 0.91, TLI = 0.89, and RMSEA = 0.06, with adequate factor loadings as shown in the final nine-factor model. Further investigation improved the modified model by correlating the items’ residuals within the same latent variable. According to the modification indices, covariance for correlated items’ residuals was added to three pairs of items (items 8 and 17, items 18 and 27, and items 20 and 36) to produce the final version. The CFI and TLI values in the final model were all greater than 0.90, while the RMSEA was less than 0.08, which indicated an adequate fit for the expected nine-factor model for the Chinese sample (Table 5). Standardized factor loadings on all factors of the CUSPAMS ranged from 0.53 to 0.98, which were statistically significant (p < 0.001). This model also obtained better-fitting data in Sample 3 (χ² = 1,723.84, df = 419, p < 0.001, CFI = 0.91, TLI = 0.90, RMSEA = 0.07).

The combined results of EFA and CFA led to the emergence of a model of nine motivational goals corresponding to the model framework proposed by the authors. But unfortunately, during the factor extraction process of the EFA, the two factors of enjoyment and stress management loaded onto one factor. We thus compared the model fitting indices of the two models through CFA and obtained the final nine-factor model version. Therefore, H1 was only partially confirmed (Figure 2).

The internal consistency of the CUSPAMS was assessed using Cronbach’s α and McDonald’s ω. The Cronbach’s α for the whole scale was 0.93 and, for the nine factors, varied from 0.79 to 0.88; meanwhile, McDonald’s ω ranged from 0.79 to 0.89. These results indicate high reliability of the nine factors of the CUSPAMS (Table 6). The internal consistency coefficient of the “others’ expectations” factor of only two items was not ideal, which was attributable to the sensitivity of Cronbach’s α to the number of items in the short scale. In this case, it may be more appropriate to report the mean inter-item correlation for the items (Pallant, 2010). After calculation, the inter-item correlation between item 6 and item 15 was 0.43, which is close to the optimal range of between 0.20 and 0.40 proposed by Briggs and Cheek (1986). The test–retest correlation coefficient for the total score of the CUSPAMS was 0.753, indicating that the CUSPAMS had higher stability.

The convergent validity of the measurement model was established by calculating the average variance extracted (AVE) and composite reliability (CR). These indicators can reflect the degree of shared variance between latent variables. Fornell and Larcker (1981) suggested that if the value of the AVE was greater than 0.70, it should be considered very good, whereas a level of 0.50 is acceptable; meanwhile, a CR value greater than or equal to 0.70 is considered acceptable. Based on the final nine-factor model, in addition to the others’ expectations factor, the CR values of the other eight factors ranged from 0.78 to 0.88, and the AVE values ranged from 0.49 to 0.70. Hence, we concluded that the convergent validity of the construct was adequate. The values of CR, AVE, and correlation coefficients are shown in Table 6. The factor correlations that were not significant were others’ expectations with appearance, and affiliation with appearance. However, other pairs were significant.

Recently, the heterotrait-monotrait ratio of correlations (HTMT) approach was proposed to assess discriminant validity (Henseler et al., 2015). HTMT is derived from the classic multitrait-multimethod (MTMM) matrix (Campbell and Fiske, 1959), which is an estimate of the correlation between constructs. Henseler et al. (2015) proposed that HTMT values lower than 0.90 show that the true correlation between the two constructs should differ. From the calculation results (Table 7), all correlations achieved the recommended value of below 0.90. This indicated that the final nine-factor model of the CUSPAMS showed good discriminant validity.

The correlation matrix between the CUSPAMS and C-BREQ-2 subscales showed that enjoyment, mastery, health benefits, stress management, appearance, competition and intrinsic motivation, identified regulation were highly positively correlated, while policies, others’ expectations and amotivation, external regulation were highly positively correlated (see Table 8). Slightly different from the expected hypothesis, H2, appearance was related to a more autonomous form of regulation. In addition, the CUSPAMS also correlated with the SEE-C. Except for policy intervention, others’ expectations and appearance were weakly correlated with the SEE-C, and the remaining six factors were significantly and moderately correlated with the SEE-C. These results confirm hypotheses H2 and H3.