*3.5. Discriminant Analysis*

Given that in almost all questionnaires and subcategories, significant differences were found between OB-BED, OB and CO groups, a linear discriminant analysis (LDA) was conducted to assess the relative contribution of different sets of psychopathological features in grouping and distinguishing the groups. Given the high internal consistencies of SEES (with happiness reverse-coded), DEBQ, and RSQ subscales (excluding wish for independence, which did not differ between groups), Cronbach's α 0.74, 0.71, and 0.66, respectively, mean total scores for these questionnaires were computed. Thus, eleven total scores for all questionnaires were entered as predictors in the LDA. Because of missing scores on at least one questionnaire, 116 cases were included in the analysis.

Results of the LDA revealed a two-dimensional solution (see Figure 2), with a significant function 1 accounting for 89.6% of the variance, χ<sup>2</sup> (22) = 174.84, *p* < 0.001, and a significant function 2 accounting for the remaining 10.4% of variance, χ<sup>2</sup> (10) = 30.49, *p* < 0.001. Group centroids (i.e., means in multivariate space) suggest a clear distinction between CO (−2.17), OB (0.69) and OB-BED (1.80) groups due to predictors associated with function 1, with predictors related to function 2 primarily distinguishing between OB (−0.68) and OB-BED (0.68) groups, with CO (0.21) in between. In order of importance, predictors primarily associated with function 1 (loadings in parentheses) were EDEQ (0.70), PSRS (−0.69), YFAS (0.49), BDI (0.41), CTQ (0.36), and RSQ (0.34). Function 2 predictors were FCTQ (0.71), SSES (0.64), SEES (0.63), DEBQ (0.45), and BIS-15 (0.24). Because Box's *M* statistic indicated a significant violation of the assumption of equal covariance at *p* < 0.001, which could render LDA results unstable, a secondary LDA with only OB-BED and OB groups was conducted to confirm the reliability of function 2 predictors. Meeting assumption checks (Box's *M p* = 0.30), the additional analysis yielded a similar solution for the distinction of OB-BED and OB groups, χ<sup>2</sup> (11) = 31.64, *p* < 0.001, with FCTQ (0.87), SSES (0.74), DEBQ (0.70), and SEES (0.69) emerging again as the most important predictors of group differences. Taken together, the LDA findings suggest that CO can be distinguished from OB, and OB-BED groups along a continuum of increasing problematic eating behaviors and attitudes, depression, and childhood adversities, with OB-BED further distinguishable from OB along a continuum of increasing emotional eating tendencies, and eating-related as well as general self-regulation impairments.

**Figure 2.** Cases and centroids of study groups on the two discriminant functions derived from the linear discriminant analysis (LDA) of questionnaire responses. Asterisks (\*) indicate the largest absolute correlation between each predictor and the discriminant functions. CO = control; OB = obese; OB-BED = obese with co-morbid Binge Eating Disorder.
