Variance estimation in longitudinal microsimulation / Milorad S. Kovacevic and Gurupdesh S. Pandher. : CS11-613/93-10E-PDF

"As a policy development tool, longitudinal microsimulation (MS) modelling is widely used in researching the impact of various policy scenarios on certain characteristics of interest in a hypothetical but representative population. In this context, microsimulation output has analytical and inferential uses rather than merely providing data to construct simple point estimates for descriptive purposes. In general, MS output is autocorrelated so that the direct application of simple variance estimation techniques is inappropriate. The present work addresses the question of how the correlated nature of observations 'collected' during long-run simulation can be overcome. The emphasis is on the design of the MS experiment and the proper choice of the variance estimator. Two designs are considered, the independent runs and the batched run design, and three variance estimators are compared regarding their statistical properties. These ideas are applied to sequences of data produced by the Canadian Population Health Model (POHEM)"--Abstract.

Lien permanent pour cette publication :
publications.gc.ca/pub?id=9.837379&sl=1

• Format MARC XML
• Format MARC HTML

Demander des formats alternatifs