When do missed prompts matter? A simulation and empirical sensitivity study of structured nonresponse in intensive longitudinal assessment.

Inference in intensive longitudinal assessment depends on whether observed prompts adequately represent the construct or phenomenon under scrutiny. When missed prompts are systematically related to unobserved momentary states, response rate summaries and missing data-adjusted estimates may underestimate clinically meaningful outcomes. Using suicidal ideation as a high-stakes exemplar, this project evaluated structured nonresponse in two linked studies. Study 1 used Monte Carlo simulation to examine recovery of complete-schedule momentary suicidal ideation prevalence across response mechanism (missing completely at random, missing at random, symptom-dependent missing not at random [MNAR], disengagement-driven MNAR), target response rate, and volatility-dependent missingness. Simulation parameters were determined based on empirical research and calibration to varied MNAR assumptions. Six estimators were evaluated: observed response mean, last observation carried forward, G-computation, inverse probability weighting, augmented inverse probability weighting, and targeted maximum likelihood estimation. Under missing completely at random and missing at random, estimators were approximately unbiased or modest with coverage deterioration in lower response and volatility-dependent conditions. Under MNAR mechanisms, all estimators underestimated complete-schedule prevalence. Study 2 applied a delta-shift sensitivity analysis to an open ecological momentary assessment data set. Even with 90.3% item observability, stronger outcome-enrichment assumptions for missed prompts increased estimated complete-schedule prevalence from 5.94% to 7.24%. Findings indicate that high response rates and robust estimators do not guarantee valid intensive longitudinal assessment when nonresponse is structured by unobserved risk states. Simulation and empirical illustration results should be considered in the context of study parameters and calibration decisions. Design-stage planning and delta-shift sensitivity analyses can help researchers evaluate, reduce, and transparently report potential bias from structured nonresponse. (PsycInfo Database Record (c) 2026 APA, all rights reserved).