Data Loss from Pretest to Posttest as a Sample Selection Problem

When estimating regression models of educational achievement with pre- and posttest data, researchers have overlooked a sample selection bias that may occur even where initial assignment to the control and experimental groups is random. The bias arises because students who take the pretest but do not take the posttest are excluded from the regression analysis. Using data from a nationally normed test of high school student knowledge of economics, adjustment for this bias is shown to influence the estimated effectiveness of programs aimed at increasing student learning of economics. Copyright 1990 by MIT Press.