Estimating endogenous treatment effects in retrospective data analysis.

Treatment effect estimation is one of the mainstays of the field of outcomes research. It is, for example, a key component in analyzing the cost-effectiveness of a proposed qualitative intervention. Some outcomes researchers are hesitant to use retrospective data for treatment effect estimation because of the potential endogeneity of the treatment variable. This is unfortunate, given the abundance and other advantages of retrospective data. Others who have used retrospective data have ignored the endogeneity problem, or have not recognized its potential for causing bias in their estimates. In this paper, an econometric method that is unbiased in the presence of endogeneity and therefore broadens the potential for use of retrospective data in the estimation of treatment effects is proposed. This two-stage method is also designed to accommodate nonlinearity in the relationship between the treatment variable and the outcome. An easy to apply GAUSS implementation of the estimator is offered.