Turning Hazard Models Inside Out

David Melamed
Chivon Fitch, University of Tampa

Regression modeling is a workhorse demographic method. Traditionally, the focus of a regression model is on the variables and occasionally how they are related to or moderated by one another. Regression Inside Out (RIO) is a recent extension to traditional regression techniques that puts emphasis on the cases that constitute the model. Specifically, RIO shows how each case or subsets of them contributes to the estimated regression coefficients. Here, we extend RIO to the Cox Proportional Hazard Model and introduce a generalization that permits any model in the Generalized Linear Model to be turned “inside out.” Moreover, we introduce a resampling scheme in the context of RIO that permits uncertainty estimates around the contributions of individuals cases to overall model coefficients. We illustrate RIO applied to Hazard models with a reanalysis of a published example, showing differential mortality by sex, race and ethnicity, and the novel substantive conclusions that can be gleaned through the use of RIO.

See extended abstract

 Presented in Session P8. Demographic Trends, History, Data and Methods