A nonparametric visual test of mixed hazard models

Dato: December, 2013

Skrevet af: Jaap Spreeuw (Cass Business School, City University London), Jens Perch Nielsen (Cass Business School, City University London) & Søren Fiig Jarner (ATP)

Udgivet i: Statistics and Operations Research Transactions (2013)

Abstract (på engelsk): 

We consider mixed hazard models and introduce a new visual inspection technique capable of detecting the credibility of our model assumptions. Our technique is based on a transformed data approach, where the density of the transformed data should be close to the uniform distribution when our model assumptions are correct. To estimate the density on the transformed axis we take advantage of a recently defined local linear density estimator based on filtered data. We apply the method to national mortality data and show that it is capable of detecting signs of heterogeneity even in small data sets with substantial variability in observed death rates.