Summed versus estimated factor scores: Considering uncertainties when using observed scores.

The Intuitive Eating Scale-3: Development and psychometric evaluation

“…Scholars who have the ability (e.g., based on sample size/power, access to relevant statistical software) to compute a bifactor-ESEM model and use this model within their analyses should do so. Indeed, for all scales in the social sciences, latent variable models (i. e., the factorsderived from factor analysisthat account for variation and covariation in a set of items) control for unreliability but also provide empirically optimal weights and variance separation, and are therefore more accurate than manifest (or observed) scores (i.e., summing or averaging scale/subscale items) (for discussions, see Liu & Pek, 2024;Swami, Maïano, & Morin, 2023). For those who wish to use latent scores, we recommend the use of the G-factor rather than the individual S-factors, as a portion of the item variance that loaded onto the G-factor would be missing if only S-factors were used.…”