INTRODUCTION

Regular physical exercise has many well-known physical and psychological benefits (^{Nogueira et al., 2017}; ^{Reiner et al., 2013}). However, exercise can also be detrimental when it reaches pathological or compulsive levels: individuals may train at higher intensity or frequency levels than recommended or continue training despite injury (^{Pope et al., 2000}). Indeed, compulsive exercise is a frequent symptom of several disorders, such as eating disorders (ED) or muscle dysmorphia (MD), and has received several names, such as exercise dependence, exercise addiction or compulsive exercise (^{Cunningham et al., 2016}; ^{Melissa et al., 2020}; ^{Nogueira et al., 2017}; ^{Olave et al., 2021}).

Compulsive exercise is a repetitive and rigid behavior whose function is to relieve anxiety and avoid feared situations in response to obsessions (mostly related to body appearance), and which is difficult to stop practicing despite its consequences (^{Taranis et al., 2011}). Compulsive exercise, both in non-clinical samples and in clinical samples of ED has been consistently associated with disordered eating to control body weight, mood dysregulation, compulsivity, perfectionism, and rigidity (^{Meyer et al., 2011}; ^{Trott et al., 2020}). Compulsiveness has also been associated with excessive exercise in samples of community adolescents (^{Goodwin et al., 2011}; ^{Orrit, 2019}), community exercisers (^{Gulker et al., 2001}), ED patients (^{Naylor et al., 2011}; ^{Young et al., 2018}) and males diagnosed with MD (^{Murray et al., 2012}).

Despite this, the psychological relationship that the person maintains with exercise (e.g., anxiety about the possibility of not training) provides better information as a predictor of psychopathology than the amount of physical exercise performed (^{Scharmer et al., 2020}).

Considering the negative effects and the disorders associated with compulsive exercise, it is important to have appropriate measurement instruments with which to assess it. The Compulsive Exercise Test (CET; ^{Taranis et al., 2011}) is one of the most widely used multidimensional measures of pathological or compulsive exercise. In the original validation carried out in a sample of 367 university female exercisers, the final scale comprised 24-item using 6-point Likert self-report response alternatives, and five factors were identified through an exploratory factor analysis: (1) Avoidance and rule-driven behavior (8 items), (2) Weight control exercise (5 items), (3) Mood improvement (5 items), (4) Lack of exercise enjoyment (3 items) and (5) Exercise rigidity (3 items). The CET showed good internal consistency (range: α = .71 to .81), and good convergent and concurrent validity with other measures of pathological exercise and eating psychopathology (^{Taranis et al., 2011}).

The factorial structure of the CET has been explored in several samples with different results. The original five-factor structure has been supported in a sample of 356 women with ED and 360 healthy women aged 16 to 60 years (^{Meyer et al., 2016}), in a sample of 104 adolescents with ED (7% males) (^{Formby et al., 2014}) and in a community sample of 1012 adolescents aged 12 to 14 years (45.3% males) (^{Goodwin et al., 2011}). However, ^{Swenne (2016)}, in a clinical sample of 212 adolescents with ED, found a four-factor solution more appropriate. In contrast, a three-factor model was found to be more suitable in a sample of 689 competitive athletes of both sexes (37.4% males) (^{Plateau et al., 2014}). Thus, the underlying structure of the scale appears to be a source of confusion as contradictory results emerge.

^{Sauchelli et al. (2016)} carried out the first validation of the Spanish version of the CET in a mixed sample of 157 patients with an ED and 128 healthy university students, in which only 57 (20.3%) were men and 43 of them belonged to the control group. The CET showed good internal consistency, between .79 to .96 for the whole sample, its scores were associated with ED symptoms and general psychopathology measures, and the five-factor solution was supported through a confirmatory factor analysis. However, regarding the males in the sample, the validation used what can be considered as a small sample, thus there is reason to believe that the factor structure may not have been adequately confirmed for males.

Men tend to perform more physical exercise than women, both recreationally and competitively (^{Piercy et al., 2018}), and may be more vulnerable to engaging in pathological behaviors related to physical exercise (^{Pope et al., 2000}). These behaviors in many cases are oriented to modify body weight, increasing it (i.e., via muscle mass) or decreasing it. A recent study with more than a thousand male university students showed that more than 44% were not satisfied with their weight and would like to modify it in some way (^{Gorbeña et al., 2021}). On the other hand, practicing sports, especially when competing, can generate stress and psychological demands that require personal coping resources (^{Almeida Pereira et al., 2020}). Likewise, competitive sports may generate high anxiety, associated more with sports that require closed skills, such as athleticism or swimming, compared to other sports that require open skills (e.g., football) (^{Pons et al., 2020}). In this vein, studies that have been carried out amongst competitive athletes have also shown high levels of body dissatisfaction, higher striving for perfection (^{Atienza et al., 2020}), vulnerability to ED and substance use (^{Souter et al., 2018}), including anabolic steroids (^{Pope et al., 2000}). Surprisingly, recreational exercisers (non-competitive) have also reported a high weekly training frequency, such as in the study by ^{Freire et al. (2021)} who examined 159 running or crossfit practitioners, concluding that weekly training frequency appeared to be an intervening factor in the degree of compulsive exercise among recreational exercisers, explaining an unexpected 57 % of the variance. On the contrary, other studies that have been carried out in male athlete samples have not found differences between competitive athletes and recreational exercisers in the use of extreme weight control behaviors (^{Gorrell et al., 2019}) and levels of exercise dependence (^{Menczel et al., 2017}).

In summary, the CET is a multidimensional questionnaire that apparently presents adequate psychometric properties, albeit with inconsistencies, especially regarding its factor structure. Although CET has been studied in female and mixed samples, in the latter the percentage of males has been limited (e.g., ^{Goodwin et al., 2011}; ^{Plateau et al., 2014}). To our knowledge, the psychometrics of the scale in a large and arguably adequate sample of males has not yet been examined in any country or language. Hence, this study has attempted to fill this apparent gap in compulsive exercising research. Thus, the main purpose of our study was to examine the psychometric properties of the CET in a sample of young males. The aims were: (a) to assess the factor structure of the scale, (b) to evaluate its measurement invariance as a function of the competitive level, (c) to evaluate its reliability, (d) to examine the concurrent validity with MD symptomatology and obsessive-compulsiveness, and (e) to examine the convergent validity with the amount of vigorous physical activity. We hypothesized that: (1) the original five-factor solution would be supported in the whole sample, (2) the scale would not be invariant for competitive and non-competitive exercisers, (3) the test would show good reliability and (4) greater levels of compulsive exercise would be associated with greater obsessive-compulsiveness and MD symptomatology, and with a greater amount of vigorous physical activity.

MATERIAL AND METHOD

RESULTS

Confirmatory factor analysis and exploratory structural equation model

Fit statistics for each model are presented in Table 1. The five-factor model structure proposed by ^{Taranis et al. (2011)} showed poor fit to our data, as our fit indexes were far from recommended cut-off points. Furthermore, this model showed convergence problems, indicating that this model was not adequate. A more thorough investigation showed that two items were problematic: Item 8 (“I do not exercise to be slim”) and item 21 (“I do not enjoy exercising”). Both items had low inter-item correlations (below .30) and low corrected item-rest correlations (r_8X^c = .03; r_21X^c = -.06). These items have shown poor performance in other CET validations (e.g., ^{Formby et al., 2014}; ^{Plateau et al., 2014}; ^{Sauchelli et al., 2016}). Thus, they were removed from further analyses.

Table 1. Fit index values for the tested models (n = 750). 

A five-factor model with 22 items was then tested. This new model was also problematic, with a non-positive definite latent variable covariance matrix. Several factor correlations were extremely large, above .90 or even 1.00 (the standardized correlation between Lack of exercise enjoyment and Mood improvement was 1.43). Thus, this model was not adequate for our data. Then, we tested the same model using the ESEM approach with Target rotation (^{Asparouhov & Muthén, 2009}), but fit was not adequate. Furthermore, there were several significant zero-target cross-loadings.

Given the previous results for the five-factor model and considering that several modification indexes suggested an overlap between Lack of exercise enjoyment and Mood improvement factors, a four-factor solution (Avoidance of negative affect, Weight control exercise, Mood improvement and Exercise rigidity) with 22 items was tested, where Mood improvement included 7 items (5 from Mood improvement plus 2 from Lack of exercise enjoyment). However, this model showed also poor fit to our data

Therefore, we tested the three-factor model (Avoidance of negative affect, Weight control exercise and Mood improvement) proposed by ^{Plateau et al. (2014)}. This model showed the best fit to our data (see Table 1), which could be considered adequate. Thus, this last three-factor model was retained (see Figure 1).

Figure 1. Retained factor model (three-factor 15-item) for the Compulsive Exercise Test (CET) in Spanish males (n = 750). 

Measurement invariance

Although the three-factor model of 15-items appeared to be adequate for the total sample, ^{Plateau et al. (2014}) proposed this model only for competitive athlete populations. Since our sample included both non-competitive and competitive exercisers, we first tested independently the internal structure of the CET in the two subsamples, to examine if the same factor structure was supported for both groups. Only the three-factor model with 15 items was tested (see Table 2).

Table 2. Fit index values for the three-factor model of CET. 

The reduced three-factor model showed an acceptable fit in both non-competitive and competitive exercisers. Thus, we next examined the measurement invariance of this model using MG-CFA and testing configural, metric, and scalar invariance (see Table 3). The configural invariance model showed adequate fit to our data, supporting the configural invariance of the three-factor 15-item CET. The fit of the metric and scalar invariance models was also adequate. Furthermore, the ΔCFI and ΔRMSEA values were below the recommended cut-off points in both cases. Thus, scalar invariance for the CET retained model was supported for the male university sample.

Table 3. Fit indexes for measurement invariance models of CET with 15-items. 

Additionally, we carried out the alignment method to examine if there were some loadings or intercepts of particular items that were non-invariant between both groups. We initially used a FREE approach, but this analysis provided a standard error warning, so we used a FIXED approach (^{Asparouhov & Muthén, 2014}), with the non-competitive exercisers as the reference group. Results from this analysis showed that only the intercept of item 2 (I exercise to improve my appearance) was non-invariant, which had the higher absolute fit function contribution. All remaining parameters (loadings and intercepts) were deemed as invariant, further supporting the measurement invariance of the CET in our sample.

Descriptive statistics, internal consistency, convergent, and concurrent validity

Means, standard deviations and internal consistency for all measures, as well as correlations among all evaluated variables, are presented in Table 4. The CET-S and its subscales showed good to excellent internal consistency, with omega coefficient values above .80.

Table 4. Means (standard deviations), internal consistency and correlations among variables of CET-S with 15-items. 

^Notes:Omega coefficients are presented along the diagonal in bold. CET=Compulsive Exercise Test; ANA=Avoidance of Negative Affect; WCE=Weight Control Exercise; MI=Mood Improvement; MDDI = Muscle Dysmorphia Disorder Inventory; DFS = Drive for Size; AI = Appearance Intolerance; FI = Functional Impairment; OCI-R = Obsessive Compulsive Inventory Revised; IPAQ-S = International Physical Activity Questionnaire-Short Form; VA = Vigorous physical activity; BMI = Body Max Index. N.A.= Not Applied.

^*p<.05

^**p<.01

Regarding the convergent validity, the total CET-S score showed moderate significant correlations with the vigorous physical activity scale of the IPAQ-S. This was also the case for the ANA and the MI subscale. The WCE subscale showed significant but low correlations with those scores.

Regarding the concurrent validity, there was a positive low significant correlation between the OCI-R and the CET-S scores. On the other hand, the MDDI total score showed positive low to moderate significant correlation with the CET-S and its subscales. There were also positive moderate to high correlations between the MDDI-FI subscale and the CET-S, especially with the ANA subscale and the total CET-S score. There was also a moderate positive correlation between the WCE subscale and the MDDI-AI subscale. The MDDI-DFS subscale showed low significant correlations with the CET-S.

DISCUSSION

Our study aimed to examine the psychometric properties of the CET in a sample of Spanish male university students. To do so, we assessed its factor structure, measurement invariance, reliability and convergent and concurrent validity with obsessive-compulsiveness, MD symptomatology and exercise frequency and duration. We hypothesized that the original five-factor solution would be supported, the scale would not be invariant, and that the CET would show good reliability and convergent and concurrent validity.

Regarding the CET factor structure, CFA performed in our sample did not support the original five-factor model proposed by ^{Taranis et al. (2011)} and supported by ^{Sauchelli et al. (2016)} in Spain. On the contrary, the data showed adequate fit for the three-factor model suggested by ^{Plateau et al. (2014)}, indicating the multidimensional nature of physical exercise. Although the development and validation study by ^{Taranis et al. (2011)} was carried out in a community sample of physically active women, the conceptualization of the test was carried out with specialists and ED patients, as well as in the majority of the sample (men = 57, 20.3%) used by ^{Sauchelli et al. (2016)}. Therefore, the theoretical development of CET from a female-centered viewpoint of ED may cause discrepancies when applied to men, as has happened in body dissatisfaction questionnaires (^{Compte & Sepúlveda, 2014}). In men, rule-driven behavior, exercise rigidity and lack of exercise enjoyment components are not aspects that influence the maintenance of compulsive exercise, unlike the avoidance of negative affect, weight loss control exercise and mood improvement. Unlike women, men show the so-called "feeling and talking taboo", particularly if they are related to their body image, and being more likely to use potentially risk behaviors as an emotional regulation strategy (^{Pope et al., 2000}). Male body dissatisfaction is mainly oriented to increasing muscle mass and leads to risky behaviors related to diet and physical exercise (^{Murray et al., 2012}; ^{Orrit, 2019}), a central feature of MD and absent in ED of female predominance.

The invariance analysis performed did not support our initial hypothesis. The three-factor model of ^{Plateau et al. (2014)} was proposed for competing athletes. However, in our sample, the CET shows invariance, both from a traditional approach (MGCFA) and from the recent alignment optimization approach (^{Asparouhov & Muthén, 2014}), between the groups of competing and non-competing students. The fact that both techniques pointed in the same direction supports the robustness of the CET-S items and structure in males. ^{Sauchelli et al. (2016}) also performed an invariance analysis between men and women; however, its subsample of 57 men was very scarce to carry out this type of analysis, which affects the stability of the results.

To date, psychometric properties of the CET have not been assessed in such a large sample of males in any country or language. These results support the use of the CET in a generalized way in the population of adult males regardless of their sports level, extending the use of the short version proposed by ^{Plateau et al. (2014)}. Nevertheless, differences in the factorial structure of the CET between men and women reflect the sensitivity of the scale to gender bias, suggesting a flexible application with a reduced 15-item version for men and the original 24-item for women.

The omega coefficient better fits the nature of our data, so the comparison with Cronbach's alpha reliability indices from previous studies is limited. In any case, for our sample of university male students, the internal consistency of the CET-S was good, making it a reliable instrument (range: ω = .82 to .91).

Even though the CET-S implicitly carries a dimension of compulsivity, its relationship with the obsessive dimension in our sample, although significant, is limited. On the contrary, in line with the research hypothesis, an intense relationship is observed between the level of compulsive exercise and the symptoms of MD. The amount of compulsive exercise and its use to avoid negative emotional states generates higher levels of functional impairment in the person's life and increases the risk of developing serious disorders such as MD.

On the other hand, the relationship between CET-S and drive for size is limited despite its statistical significance. The search for greater muscle volume is obtained through the continuous practice of strength exercise, but, above all, through dietary alterations (e.g., increased amount of proteins) or supplement use. In this sense, the link between compulsive exercise and dietary alterations has been accredited in the literature, to the point that the development of CET occurs in the context of ED (^{Taranis et al., 2011}). However, its association with the dimensions of MD is limited, since the weight-oriented CET items refer to the fear of gaining weight and/or becoming thinner, without reference to bulking. Although MD is currently within the obsessive-compulsive disorder spectrum, this classification is not without criticism within the scientific community, which claims for its inclusion within the spectrum of an ED (^{Compte et al., 2019}; ^{Compte et al., 2018}; ^{Compte & Sepúlveda, 2014}; ^{Murray et al., 2012}). Data on the high relationship between CET-S and MD symptoms, on the one hand, and the low relationship with the obsessive spectrum, on the other, support this view.

In line with the research hypothesis, there is a relationship between the CET-S and its different subscales, and the amount of vigorous physical activity done by the male students, mainly with those subscales that explore the positive and negative emotional reinforcement of physical activity. In contrast, the relationship between the amount of vigorous physical activity with the weight control exercise subscale of the CET has a lower effect size. As mentioned above, the content of the CET was designed on the area of ED with mostly female populations in which the drive for thinness is a central aspect. However, the results suggest that in males, aspects related to emotional regulation play a more important role in intense physical activity than weight control in the terms referred to in the CET. In particular, the guilt or discomfort at the possibility of not complying with the exercise routine.

BMI is also found to be positively related to the different subscales of the CET. However, it is the weight control exercise subscale that shows a larger effect size. BMI is a parameter related to dimensions of body dissatisfaction such as compulsive exercise, ED, or MD. However, it is sometimes only reported as a description of participants and no analyses are performed on the validation studies (^{Taranis et al., 2011}; ^{Sauchelli et al., 2016}). Males also present a dislike of body fat and overweight, although they may want to obtain pathological levels of weight associated with muscle volume (^{Compte et al., 2018}; ^{Gorbeña et al., 2021}). The CET as an instrument would be limited in capturing these muscle-seeking aspects, which are frequent in the male population.

CONCLUSIONS

In summary, the Spanish three-factor-15-item version of the CET validated in male university students has good psychometric properties in male populations, as it is a valid and reliable compulsive exercise self-report questionnaire. Thus, the CET-S could be used as a screening measure of compulsive exercise, and it could help examine its prevalence and maintenance factors in male populations. However, more research is necessary to support the findings from this study and to understand the development and maintenance factors of excessive exercise in male populations.

IMPLICATIONS FOR PRACTICE

Re-examining the psychometric properties of measures designed and used mostly in female populations is relevant for both research and clinical treatment in order to avoid possible gender biases when applied to males. The original factorial structure of the questionnaire does not fit the reality of men and may underestimate the presence of compulsive exercise in this population when using the 24 items, five-factor version of the CET as a screening instrument. The proposal of a generalized use of the short three-factor version of the CET with only 15 items, regardless of sport level, provides to clinical professionals and researchers a tool that can be easily applied to explore compulsive exercise in Spanish-speaking men. Compulsive exercise is strongly related to the symptomatology of serious pathologies such as Muscle Dysmorphia, more prevalent in men. The use of CET as a screening instrument can be useful, along with other instruments, to detect symptoms of Muscle Dysmorphia among male populations, including sportsmen.