Selective Attrition and Longitudinal Consistency in Dynamic Microsimulation: Evidence from Understanding Society

Dynamic microsimulation models rely extensively on longitudinal household data to project long-term socio-economic, demographic, and policy outcomes. While these dynamic frameworks serve as powerful tools for economic projections, their predictive reliability depends fundamentally on the representativeness of the data used for parameterisation. Utilising 14 waves of the UK Household Longitudinal Study (UKHLS) spanning 2010 to 2024, this paper evaluates sample representativeness and attrition bias against external national benchmarks from the UK Census. We find that over time, the survey increasingly overrepresents older, white, highly educated, and partnered individuals, in a way that is only partially corrected by the provided survey weights. Probit regression models confirm a pronounced "survival bias," demonstrating that incoming entry and replacement mechanisms fail to symmetrically balance selective exit rates. Consequently, we conclude that relying solely on standard survey weights is insufficient to maintain longitudinal consistency, emphasising the necessity of integrating external alignment strategies within dynamic microsimulation frameworks.