How wrong could parameter estimates be? Statistical consequences of fitting the wrong model to human mortality data

Laszlo Nemeth, Max Planck Institute for Demographic Research
Trifon I. Missov, Max Planck Institute for Demographic Research

In this paper we check for and quantify the systematic bias regarding mortality parameter estimates when the fitted model is misspecified. We focus on the Gompertz, Gompertz-Makeham, gamma-Gompertz, gamma-Gompertz-Makeham models, which are commonly applied when studying adult human mortality. We simulate lifespans from each of these models, calculate the corresponding age-specific death counts and exposures, and fit the correct and a set of misspecified models to each resulting dataset. We check whether or not the known parameters are captured correctly by the estimation procedure and quantify the bias if it exists.

  See extended abstract

Presented in Poster Session 2