Uncertainty Analysis for Demographic Microsimulation

Douglas A. Wolf, Syracuse University

Demographic microsimulation "samples" from a future population, first sampling from some baseline population and then sampling from the distribution of future values to which the baseline will evolve. Because microsimulation entails sampling, summary statistics based on its output should be viewed as uncertain, reflecting the inevitable errors associated with sampling. Yet little attention has been paid to this type of variability. Uncertainty about microsimulation-based forecasts arises from several sources, including parameter uncertainty, classical sampling error, Monte Carlo errors, and data imputation errors. This paper attempts to quantify prediction uncertainty in the context of a simple projection model of kin networks. The approach used permits me to partition uncertainty into several components, with particular attention to the influence of variability in model parameters. Initial results indicate that standard errors grow about 1-10 percent, in most cases, when we account for the various sources of variability uniquely attributable to the microsimulation methodology.