A1202
Title: A simulation study of inconsistency in mortality modeling in the field of econometrics
Authors: Stuart Gilmour - St. Lukes International University (Japan) [presenting]
Abstract: In recent years, econometric methods have been applied to mortality data to assess health policy. The purpose is to review policy assessments from NBER and to identify seven different mortality modeling specifications. These models are applied to simulated mortality data across a range of scenarios alongside the (correct) Poisson specification and compared for bias and accuracy. The models are also compared with the correct Poisson specification for assessing the age profile of mortality in Japanese population data. All specifications were biased and, in many scenarios, were not even able to correctly identify the sign of a policy effect. They were also unable to correctly estimate the Japanese mortality profile, with coefficients in some cases being wrong by orders of magnitude. It is found that models used for mortality studies in econometrics are inconsistent, incoherent, and have no basis in statistical theory. It is likely that most mortality studies in the econometrics literature are wrong and give incorrect policy recommendations. Most published papers should be retracted, and policy recommendations re-evaluated. In the future, this field needs to adopt the standard, correct Poisson regression framework for modeling mortality and stop using or teaching arbitrary transformations in an OLS framework that is biased and inconsistent.
