Measuring eating disorder attitudes and behaviors: a reliability generalization study

I used reliability generalization procedures to determine the mean score reliability of the Eating Disorder Inventory (EDI), the Eating Attitudes Test (EAT), and the Bulimia Test (BULIT). Reliability generalization is a type of meta-analysis used to examine the mean score reliability of a measure across studies and to explore study factors that influence mean score reliability. Score reliability estimates were included in 28.67% of 293 studies using the EDI, 36.28% of 215 studies using the EAT, and 41.46% of 41 studies utilizing the BULIT. For the EDI, mean Cronbach?s alphas for the subscales ranged from .52 to .89 and the mean estimate for the total score was .91. For the EAT-40 and EAT-26, mean estimates of internal consistency were .81 and .86 respectively. Mean estimates of internal consistency for the EAT-26 subscales ranged from .56 to .80. The mean estimate of internal consistency for the BULIT-R was .93. Overall, the mean reliability of scores on all three measures and their subscales/factors was acceptable except for the Asceticism subscale of the EDI and the Oral Control factor on the EAT-26, which had mean internal consistency estimates of .52 and .56 respectively. For the EDI, the majority of the subscales that measure specific eating disorder attitudes and behaviors, such as Bulimia and Perfectionism displayed higher score reliability in clinical eating disorder samples than in nonclinical samples. This difference was not found in the Drive for Thinness and Body Dissatisfaction subscales, perhaps because these attitudes are common in both eating disorder and nonclinical samples. Score reliability information for the EAT and BULIT was primarily reported for nonclinical samples; therefore, it is difficult to characterize the effect of type of sample on these measures. There was a tendency for mean score reliability for all the measures to be higher in the adult samples than in adolescent samples and in the female samples compared to the male samples. This study highlights the importance of assessing and reporting internal consistency every time a measure is used because reliability is affected by characteristics of the participants being examined.