Given that final models are often complex in fixed-effects IDA because of the need to control for the main effects of covariates and their interactions with study membership, we recommend using a model building strategy that includes the practice of model trimming. As is common in single-study analysis, we begin with models that test study differences in the associations between control variables and outcomes to establish a baseline model. We then add to the baseline model theoretical predictors of our outcomes, typically in a sequence of models defined by our substantive theory (e.g., adding unique effect predictors or interactions among theoretical predictors later in the sequence). In each step of the model building sequence, we test interactions between all variables (including higher order interaction terms) with study membership.
