A Look at Quantile Regression
Joseph C Cappelleri, Pfizer Inc
Keywords: quantile regression, ordinary least squares regression, health policy, health services, health economics, resource utilization
Research in health services research commonly use multivariate regression techniques such as ordinary least squares regression to measure the relationship of health service outcomes (as dependent variables) with clinical characteristics, demographic factors, and policy changes (as explanatory or independent variables). Ordinary least squares regression involves a difference between groups at their means on the outcome variable. Doing so assumes the regression coefficients are constant across different values of the outcome. There are times, however, when researchers, policymakers, and clinicians may be interested in group differences of an explanatory variable across the entire distribution of a dependent variable rather than only at its mean. Quantile regression relaxes this assumption of common regression slope and therefore allows for group differences to be examined at lower and higher portions of the outcome distribution. In this presentation, quantile regression will be contrasted with ordinary least squares regression and illustrated with several applications. Statistical attributes of quantile regression will be highlighted.