Addressing Missing Data From Prompt Noncompliance: A Practical Guide to Auxiliary Variables in Ecological Momentary Assessment

Missing data from noncompliance and missed prompts are pervasive in ecological-momentary-assessment (EMA) research. Although researchers often examine variables associated with prompt nonresponse, these variables are rarely incorporated into analytic models, leaving standard multilevel analyses to rely on missing-at-random (MAR) assumptions that are often implausible given the available data. In this tutorial, we introduce an accessible strategy for integrating such variables, known as “auxiliary variables,” directly into multilevel models to bolster the plausibility of the MAR assumption. We outline a practical framework for identifying promising auxiliary variables in EMA data sets (e.g., structural-design features, lagged self-reports, passive metadata) and provide step-by-step instructions for (a) preparing EMA data sets, (b) evaluating the utility of candidate auxiliary variables, and (c) incorporating them into multilevel models of EMA data. To facilitate adoption, we supply open-source R code for each step and introduce an R function ( EMAuxiliary ) that streamlines the estimation of multilevel models with auxiliary variables using freely available Blimp software. A fully worked example using real EMA data illustrates the approach. Our aim is to equip applied researchers with a practical, widely applicable method for reducing bias from missing prompts and strengthening inferences in EMA studies.