Should we use Multilevel Regression and Post-stratification when simulating area-level population outcomes?

Estimating unknown outcomes at small-area population level is a routine task in GIScience. We ask whether Multilevel Regression and Poststratification (MRP), an approach to simulating public opinion, might overcome deficiencies in spatial microsimulation (SPM), the de facto approach in applied spatial analysis. Using microdata from Health Survey for England and 2021 Census tables, we evaluate MRP and SPM at estimating a known, groundtruthed, health outcome. When parameterised with few constraints, there are only slight differences in estimation between the two approaches. With more, and geographic, constraints, often desired by spatially-inclined researchers, errors begin to accumulate in SPM estimates that do not appear in those arrived at via MRP.