《PSM_SSWR_2004》由会员分享，可在线阅读，更多相关《PSM_SSWR_2004（43页珍藏版）》请在金锄头文库上搜索。
1、Introduction to Propensity Score Matching: A New Device for Program EvaluationWorkshop Presented at the Annual Conference of the Society for Social Work ResearchNew Orleans, January, 2004Shenyang Guo, Ph.D., Richard Barth, Ph.D. , and Claire Gibbons, MPH Schools of Social Work and Public Health Univ
2、ersity of North Carolina at Chapel HillNSCAW data used to illustrate PSM were collected under funding by the Administration on Children, Youth, and Families of the U.S. Department of Health and Human Services. Findings do not represent the official position or policies of the U.S. DHHS. PSM analyses
3、 were partially funded by the Robert Wood Johnson Foundation and the Childrens Bureaus Child Welfare Research Fellowship. Results are preliminary and not quotable. Contact information: sguoemail.unc.eduOutline Overview: Why Propensity Score Matching? Highlights of the key features of PSM Example: Do
4、es substance abuse treatment reduce the likelihood of child maltreatment re-report?Why Propensity Score Matching? Theory of Counterfactuals The fact is that some people receive treatment. The counterfactual question is: “What would have happened to those who, in fact, did receive treatment, if they
5、had not received treatment (or the converse)?” Counterfactuals cannot be seen or heardwe can only create an estimate of them. PSM is one “correction strategy” that corrects for the selection biases in making estimates.Approximating Counterfactuals A range of flawed methods have long been available t
6、o us: RCTs Quasi-experimental designs Matching on single characteristics that distinguish treatment and control groups (to try to make them more alike)Limitations of Random Assignment Large RCTs take a long time and great cost to generate answersanalysis of existing data may more timely, yet accepta
7、bly accurate RCTs are not feasible when variables cannot be manipulatede.g., some events in child welfare are driven by legal mandates Prior analysis of the need for withholding treatment should be done before RCTs are deemed necessary.Limitations of Quasi-Experimental Designs Selection bias may be
8、substantial Comparison groups used to make counterfactual claims may have warped counters and failing factuals, leading to intolerably ambiguous findingsLimitations of Matching If the two groups do not have substantial overlap, then substantial error may be introduced: E.g., if only the worst cases
9、from the untreated “comparison” group are compared to only the best cases from the treatment group, the result may be regression toward the mean makes the comparison group look better Makes the treatment group look worse.Propensity Score Matching Employs a predicted probability of group membershipe.
10、g., treatment vs. control group-based on observed predictors, usually obtained from logistic regression to create a counterfactual group Propensity scores may be used for matching or as covariatesalone or with other matching variables or covariates.PSM Has Many Parents In 1983, Rosenbaum and Rubin p
11、ublished their seminal paper that first proposed this approach. From the 1970s, Heckman and his colleagues focused on the problem of selection biases, and traditional approaches to program evaluation, including randomized experiments, classical matching, and statistical controls. Heckman later devel
12、oped “Difference-in-differences” methodPSM Has Skeptics, TooHoward Bloom, MDRC Sees PSM as a somewhat improved version of simple matching, but with many of the same limitations Inclusion of propensity scores can help reduce large biases, but significant biases may remain Local comparison groups are
13、bestPSM is no miracle maker (it cannot match unmeasured contextual variables) Short-term biases (2 years) are substantially less than medium term (3 to 5 year) biasesthe value of comparison groups may deteriorateMichael Sosin, University of Chicago Strong assumption that untreated cases were not tre
14、ated at random Argues for using multiple methods and not relying on PSMLimitations of Propensity Scores Large samples are required Group overlap must be substantial Hidden bias may remain because matching only controls for observed variables (to the extent that they are perfectly measured)(Shadish,
15、Cook, & Campbell, 2002)Criteria for “Good” PSM Identify treatment and comparison groups with substantial overlap Match, as much as possible, on variables that are precisely measured and stable (to avoid extreme baseline scores that will regress toward the mean) Use a composite variablee.g., a propen
16、sity scorewhich minimizes group differences across many scoresRisks of PSM They may undermine the argument for experimental designsan argument that is hard enough to make, now They may be used to act “as if” a panel survey is an experimental design, overestimating the certainty of findings based on the PSM.A Methodological Overview Reference list The crucial diff
