Using moderated nonlinear factor analysis to separate, and estimate, treatment effects and DIF

Recent work explores the affordances of studying the impact of interventions at the item level in addition to at the test level. In that spirit, this paper demonstrates that moderated nonlinear factor analysis (MNLFA) models can be specified and estimated to disentangle main effects on the construct from item-specific deviations from this effect without a priori anchor item selection. I present the proposed MNLFA-based approach in contrast to another method for handling item-level deviations from a main treatment effect, item-level heterogeneous treatment effect (IL-HTE) analysis. I demonstrate the types of DIF that do and do not degrade the models’ performance and discuss the circumstances under which each is appropriate: IL-HTE models are highly confirmatory in the sense that they assume a distribution of DIF across all items, so the more exploratory MNLFA approach may make sense as a precursor or alternative to IL-HTE analysis. A reanalysis of data previously studied using IL-HTE methods paints a rather different picture of the extent of treatment-based DIF in the outcome measure, but leaves the overall conclusion largely unchanged.