Method: The TFEQ was administered to 395 overweight and obese adults (66.8% women; BMI = 34.7 ± 4.6 kg/m2), aged 18 to 65 years (age = 43.8 ± 10.1 years) at baseline of an 18 month weight management intervention trial. The TFEQ is comprised of 51-items over three subscales (Cognitive Restraint, Disinhibition, Hunger).
Analysis/Results: Rasch analysis was conducted separately for each Subscale of the TFEQ using the Winsteps software. The dimensionality, item-model fit (Infit and Outfit statistics ≥ 0.6 or ≤ 1.4 logits), item difficulty, and item-person map were assessed to determine the construct validity of each Subscale. Rasch separation indexes were calculated to determine the reliability (equivalent to traditional Cronbach’s Alpha) of each subscale. The results indicated that all three subscales demonstrated unidimensionality. Nineteen out of 21 Restraint subscale items (difficulty -2.51 to 2.03 logits), twelve out of 16 Disinhibition subscale items (difficulty -1.65 to 1.79 logits), and twelve out of 14 Hunger subscale items (difficulty -2.41 to 1.40 logits) were identified as good fit items. Ability levels (e.g., ability to consciously restrict food to control weight) of a substantial number of participants (Restraint ~12%; Disinhibition ~32%; Hunger ~50%) were not covered by item difficulty levels at the lower end of the person-item map. The separation index for item measures was 8.67 logits for Restraint, 6.58 logits for Disinhibition, and 7.23 logits for Hunger with a reliability of 0.98 to 0.99.
Conclusions: Rasch analysis supported construct validity of the TFEQ after removing the misfit items, although internal consistency reliabilities were marginal. Modifications to the TFEQ (e.g., adding more easier-to-endorse items) are needed so that the large variation of the dietary eating habits among this population can be adequately assessed.