Adaptive Regression Models for Modeling and Understanding Evolving Populations

Abstract

When regression analysis is carried out with a prediction purpose, an evolution in the modeled phenomenon between the training and the prediction stages forces the statistician to start a new analysis. Similarly, when regression aims to explain the modeled phenomenon, a new regression model must be estimated whenever the phenomenon or its study conditions change. In this paper we show how the previous regression analysis can be used for the estimation of the regression model in the new situation, saving a new expensive data collection effort. Two case studies are considered: first a regression model of house prices versus house and household features is adapted from a city in the Southern United States (Birmingham, AL) to another city of the United States west coast (San Jose, CA). In the second case the link between CO2 emissions and gross national product in 1999 is analyzed by using a previous analysis dating from 1980.