The Role of Microsimulation in Longitudinal Data Analysis

The term “microsimulation” has been linked to a range of tools and techniques that are finding growing use in empirical social science applications. This paper considers one such area, namely the potential for microsimulation to serve the needs of the data analyst, in contrast to the more common use of microsimulation by the model user. Furthermore, the focus is on longitudinal rather than cross-sectional data analysis. The paper identifies several types of longitudinal data modeling approaches in which microsimulation is particularly relevant, suggesting algorithms with which to conduct such microsimulations. Microsimulation can be used to extend the range of inferences that can be drawn from the estimated parameters of a model, can help to solve certain types of defective-data problems, and can fill gaps in available data. A relatively underdeveloped area is that of quantifying the uncertainty inherent in summary statistics based on data produced by a microsimulation program. I argued that due to strong parallels between the multiple imputation methodology and the structure and procedural aspects of many microsimulation exercises, the multiple imputation methodology provides a natural framework with which to develop estimates of the variances, and therefore the confidence intervals, that accompany estimates based on simulated data