Data analyses

3.3.1 Examination of muscle dysmorphia in male weightlifters and university students

3.3.1.1 Analyses of the examination of muscle dysmorphia symptoms, eating disorder variables, and body attitudes

To explore the differences of MD symptoms, eating disorder related psychopathological characteristics, and body attitudes between weightlifters and undergraduates, a series of independent sample t-tests and Mann-Whitney U-tests were performed. Effect size was calculated using Cohen’s d. An effect size of more than .2 was considered to be a small effect size. An effect size of .5 or more signified a medium-size effect, and an effect size of .8 or more indicated a large effect size (Cohen, 1992). Group differences were explored using multiple linear regression analysis adjusted to age and BMI.

3.3.1.2 Analyses to explore the explanatory variables of muscle dysmorphia

To investigate explanatory variables that are associated with MD among male weightlifters, multiple linear regression analyses were performed, adjusted to age and the strength and direction of relationships between the dependent variable (levels of MD symptoms based on the MASS scores). Furthermore, explanatory variables (group participation –weightlifter or undergraduate–, body weight, desired body weight, EDI subscales: body dissatisfaction, ineffectiveness, perfectionism, interpersonal distrust, interoceptive awareness, maturity fears; body attitudes, and age) were assessed by determining regression coefficients (adjusted β), and t-test statistics. The proportion of the variance in the dependent variable explained by the explanatory variables (adjusted R²) has also been determined.

3.3.1.3 Analyses of the study variables in self-reported steroid users and non-users Comparing the self-reported steroid users and non-users in the weightlifter group, a series of independent sample t-tests and Mann-Whitney U-tests were performed and the

means (SD) of the different study variables were compared. Effect size was calculated using Cohen’s d. Group differences were explored using multiple linear regression analysis adjusted to age and BMI. The SPSS 16.0 statistical software package was used for statistical analyses.

3.3.2 Adaptation of the Muscle Appearance Satisfaction Scale in Hungary

3.3.2.1 Analyses of the factorial structure of the Hungarian version of the Muscle Appearance Satisfaction Scale

We examined the factorial structure of the Hungarian version of the MASS with exploratory factor analysis (EFA) with MPLUS 6.0. Based on the fact that the distribution of answers to the items showed considerable floor or ceiling effects, we treated the answers as an ordinal scale, and the robust weighted least squares estimation was used in EFA (Brown, 2006). We applied GEOMIN rotation in EFA, which is an oblique type of rotation; therefore correlations between factors are allowed (Browne, 2001). In the EFA, goodness of fit is characterized by the root-mean-square error of approximation (RMSEA), its 90% confidence interval (90% CI) and its closeness of fit.

Closeness of model fit using RMSEA (CFit of RMSEA) is a statistical test (Browne &

Cudek, 1993), which evaluates the statistical deviation of RMSEA from the value .05.

Nonsignificant probability values (p > .05) indicate acceptable model fit. The further fit index reported in this study is comparative fit index (CFI) which is expected to be above .95.

3.3.2.2 Analyses of the reliability of the Muscle Appearance Satisfaction Scale

The scale score reliability for the MASS total score and subscale scores were calculated using Cronbach’s alpha coefficient. To examine the test-retest reliability, intraclass correlation is sometimes required; however, some authors argue for the use the product-moment correlation, if we do not want to take systematic error into account (Rousson, Gasser, & Seifert, 2002). In our analysis, we reported the intraclass correlations based on a two-way random model, and we also reported Pearson’s correlations in order to be comparable with previous studies.

3.3.2.3 Analyses of the construct validity of the Muscle Appearance Satisfaction Scale To test the construct validity multivariate regression analyses were estimated with the MASS subscales as observed outcome variables in both groups. The MASS subscales’

scores were calculated with the sum of the items belonging to one factor for the sake of the comparability with previous and following studies. In the weightlifter sample the model was estimated with the factors of the MASS as observed variables, and age, body mass index, self-esteem, drive for thinness, current use of steroids, and current use of food supplements as explanatory variables. This model is presented in Figure 6.

Figure 6. Multivariate model to explain Muscle Appearance Satisfaction Scale subscales in the weightlifter sample.

In this fully saturated model, the error covariances of Muscle Appearance Satisfaction Scale subscales are not presented for the sake of clarity.

Similarly to the previous model, in the undergraduate sample the model was estimated with the factors of the MASS identified in this group as observed variables, and age, body mass index, self-esteem, drive for thinness, and weightlifting activity as explanatory variables. The above models were saturated models, therefore the level of fit was not estimated due to the perfect fit. The SPSS 16.0 and MPLUS 6.0 statistical software packages were used for statistical analyses.

Age

Explanatory variables MASS subscales as observed

variables

3.3.3 Muscle dysmorphia among Hungarian male weightlifters

3.3.3.1 Analyses to explore the prevalence of muscle dysmorphia

To set out the estimated prevalence rate of MD and in order to identify a group of weightlifters with the common features of MD, latent class analysis (LCA) was conducted, by which a homogenous group –a subgroup of males with MD– can be distinguished within a heterogeneous group (male weightlifters). LCA is a statistical method for identifying unmeasured class membership among subjects using categorical and/or continuous observed variables. Therefore, LCA creates subgroups on the basis of previously defined variables, by which the associations between the groups can be reduced. Several studies have used LCA to find distinct diagnostic categories given the presence/absence of several symptoms or types of attitudes (e.g., Demetrovics et al., 2012; Fink et al., 2004; Keel et al., 2004). Moreover, a recent study also used LCA to reveal a MD group within male weightlifters (Hildebrandt, Schlundt, Langenbucher, &

Chung, 2006). Ten variables were involved in the analysis: (1–5) five subscales of the MASS, (6) Exercise Addiction Inventory, (7) quantity time X frequency of exercise as continuous variable, and (8) supplement use, (9) current AAS use, (10) lifetime AAS use as categorical variables (see Table 14). Variable 7 (QF of exercise) indicates quantity-frequency of exercise, that had been calculated the amount of time in hours spent weightlifting per day multiplied by the number of days spent with work out per week (Hildebrandt, Langenbucher, & Schlundt, 2004; Hildenbrandt et al., 2006). The variables that have been involved in the LCA are associated with the criteria for MD (Pope et al., 1997). Latent class analysis was performed with 2 to 4 classes with the full sample (n = 304) with MPLUS. The LCA (Collins & Lanza, 2010; Vermunt &

Magidson, 2002) is a latent variable analysis with a dichotomous latent variable –i.e., men with MD– and continuous indicators such as scores of MASS and also with categorical indicators such as AAS use. To determine the number of latent classes and the relative goodness of fit of the models, the Bayesian Information Criteria (BIC) parsimony index was used, with the minimization of cross-classification probabilities, entropy and the interpretability of clusters. Lower BIC value indicates a better-fitting model, and higher entropy value indicates better classification quality. Furthermore, the

likelihood-ratio difference test (Lo-Mendell-Rubin Adjusted LRT Test) was also used during the final determination of the number of latent classes, which compares the estimated model with a model having one less class than the estimated model (Muthén

& Muthén, 1998). A low p value (< .05) indicates that the estimated model is fitting better than the model with one less class.

3.3.3.2 Analyses to determine the cut-off score for the MASS

To define the tentative cut-off score of the MASS a sensitivity analysis was carried out.

As there is no existing ‘gold standard’ available for the measurement of MD symptoms in Hungary, a sensitivity analysis based on the membership in the MD group in the latent class analysis was conducted. Thus, considering the MD group as a gold standard, sensitivity and specificity values were calculated for several MASS cut-off points. The accuracy of the MASS can be assessed by calculating the proportion of cases who are classified as belonging in the MD group versus the non-MD cases. Sensitivity (i.e., the proportion of true positives that are correctly identified by the MASS) and specificity (i.e., the proportion of true negatives that are correctly identified by the MASS) were defined according to the suggestion of Altman and Bland (1994a) and Glaros and Kline (1988). To explore the probability that the MASS will give the correct “diagnosis”, the positive predictive values, the negative predictive values, and the accuracy values were calculated for several MASS cut-off points. Positive predictive value (PPV) was defined as the proportion of individuals with positive test results who were correctly diagnosed (Altman & Bland, 1994b; Glaros & Kline, 1988). Negative predictive value (NPV) was defined as the proportion of participants with negative test results who were correctly diagnosed (Altman & Bland, 1994b; Glaros & Kline, 1988).

3.3.3.2 Analyses to assess the psychological correlates of muscle dysmorphia

Comparisons of the three groups resulted from the LCA (i.e., normal weightlifters, low risk MD group, and high risk MD group) with respect to continuous variables were performed by one way analysis of variance (ANOVA; F-value), followed by post-hoc Tukey-Kramer pairwise comparison of means. If homogeneity of variance was violated, robust Welch ANOVA (W-value) was applied, followed by post-hoc Games-Howell pairwise comparison of means. In case of non-normally distributed variables and

homogeneity of variance, Kruskal-Wallis test (H-value) was conducted. If the homogeneity of variance was violated, adjusted rank Welch test (rW3-value) was applied. Tests of pairwise stochastic equalities was used as post-hoc test.

3.3.3.3 Analysis to assess the characteristics of anabolic-androgenic steroid users For the comparison of psychological correlates of steroid non-users, past- and current steroid users, one-way analysis of variance (ANOVA; F-value) was used, followed by post-hoc Tukey-Kramer pairwise comparison of means. If homogeneity of variance was violated, Robust Welch ANOVA (W-value) was applied, followed by post-hoc Games-Howell pairwise comparison of means. In case of non-normally distributed variables Kruskal-Wallis test (H-value) or adjusted rank Welch test (rW3-value) were applied, followed by post-hoc tests of pairwise stochastic equalities.

3.3.3.4 Analysis to explore the risk factors of lifetime AAS use

To investigate the predictors of lifetime AAS use among male weightlifters, binary logistic regression analysis was performed adjusted to age and level of education. The SPSS 16.0, MPLUS 6.0, and ROPstat statistical software packages were used for statistical analyses.