Longitudinal studies are vulnerable to participant dropout, intermittent nonresponse, and measurement gaps that bias estimates and reduce power. The relevance is broad: policy decisions, epidemiology, and social science rely on accurate change estimates over time. Foundational theory from Donald B. Rubin Harvard University and Roderick J.A. Little University of Michigan frames missingness as missing completely at random, missing at random, or missing not at random, which guides method choice and interpretation.
Core methods
When the assumption of missing at random is plausible, two widely recommended approaches are maximum likelihood and multiple imputation. Maximum likelihood implemented in mixed-effects models uses all available data to estimate parameters without deleting incomplete cases; this approach is supported in the statistical literature by Roderick J.A. Little University of Michigan and Donald B. Rubin Harvard University. Multiple imputation replaces each missing value with several plausible values drawn from a predictive model, combines analysis across imputations, and propagates uncertainty; this strategy is central to Rubin’s work and is endorsed by the National Research Council as a principled solution in clinical and longitudinal settings.
Inverse probability weighting and related methods address informative dropout by reweighting observed cases according to estimated participation probabilities. James Robins Harvard T.H. Chan School of Public Health developed many of the causal and weighting tools now used to correct for time-dependent confounding and nonrandom attrition. For data likely missing not at random, explicit modeling via selection models or pattern-mixture models offers sensitivity analysis paths rather than definitive corrections; Paul D. Allison University of Pennsylvania has discussed these approaches for longitudinal applications.
Choosing a method and consequences
Choice depends on the missingness mechanism, study size, and the social or territorial context that generates nonresponse. In surveys among marginalized groups, dropout may correlate with migration, employment, or cultural stigma, making the MAR assumption less plausible and increasing the importance of sensitivity analyses. Improper handling—such as complete-case analysis when data are not missing completely at random—can produce biased trajectories and misleading policy conclusions, with real-world consequences for resource allocation and health interventions.
Practically, combine principled methods (maximum likelihood or multiple imputation when reasonable), diagnostic checks, and targeted sensitivity analyses to assess robustness. Citing methodological authorities such as Donald B. Rubin Harvard University, Roderick J.A. Little University of Michigan, James Robins Harvard T.H. Chan School of Public Health, and the National Research Council helps ensure analyses follow established best practice.