期刊名称：Practical Assessment, Research and Evaluation
印刷版ISSN：1531-7714
电子版ISSN：1531-7714
出版年度：2003
卷号：8
出版社：ERIC: Clearinghouse On Assessment and Evaluation
摘要：When randomized field trials are impossible or impractical, researchers in education and the social sciences more broadly must use observational data, such as standardized test scores or responses to survey questions, to quantitatively evaluate the effects of policies or programs. A potential pitfall with these analyses, however, is that units of observation are not randomly assigned to participate; rather, they self-select to introduce the policy or program of interest. This becomes a problem for estimation and inference if the decision to institute the policy is correlated with the outcome measure—e.g. if states that are more likely to introduce high-stakes testing are also more likely to have a larger gain in test scores. In the econometrics literature, statistical techniques used to analyze these data are often referred to as "treatment effects" models (Goldberger, 1972; Maddala, 1983), where the policy of interest is the "treatment."
