Behavioural Validation and Structural Sensitivity in Dynamic Microsimulation: A Diagnostic Study of Employment and Health Transitions in the UK SimPaths Model

Dynamic microsimulation models represent individual life-course trajectories through stochastic transitions across multiple domains. Their credibility depends on the accuracy and robustness of the behavioral mechanisms that govern persistence and lag dependence. While the published SimPaths model for the United Kingdom demonstrates strong external validity at the aggregate level, reproducing observed trends between 2011 and 2019, further diagnostic validation is required to assess whether its behavioral structure reproduces the conditional dynamics observed in reality.

This study introduces a framework for conditional and structural validation within SimPaths, focusing on employment and health transitions. Using data from Understanding Society (UKHLS), we derive empirical transition probabilities and persistence measures conditional on key socio-demographic factors. The analysis proceeds in three stages. First, matched observed–simulated datasets are compiled for key life-course variables describing employment, self-reported health, and household composition. Second, model outputs are compared with empirical transition patterns and persistence measures across age, sex, and region, quantifying the extent to which simulated dynamics replicate observed life-course behaviour. Third, controlled sensitivity experiments vary the parameters governing behavioral persistence and lag dependence within plausible ranges, tracing how these changes propagate through employment, health, and household outcomes, including inequality indicators such as the Gini coefficient and the 80/20 income ratio.

The analysis is guided by two assumptions. First, employment transitions are expected to be more sensitive to persistence and lag-structure specifications than health transitions, reflecting stronger state dependence in labour-market attachment. Second, misspecification of employment persistence is expected to systematically bias projected income distributions, inflating measures of inequality and in-work poverty over the life-course.

Together, these steps test both the empirical validity and structural stability of SimPaths. Substantively, the analysis highlights which life-course domains are most sensitive to behavioral specifications, helping to prioritize future calibration and validation efforts. Methodologically, it offers a reproducible procedure for module-level testing that can be replicated across countries or policy domains, promoting transparent and cumulative model development. The broader aim is to integrate systematic behavioral validation into the standard development cycle of dynamic microsimulation models, ensuring that projected outcomes reflect both credible behavioral dynamics and empirical regularities observed in longitudinal data.