Combining a Local Comparison Group, a Pretest Measure, and Rich Covariates: How Well Do They Collectively Reduce Bias in Nonequivalent Comparison Group Designs?
{"title":"Combining a Local Comparison Group, a Pretest Measure, and Rich Covariates: How Well Do They Collectively Reduce Bias in Nonequivalent Comparison Group Designs?","authors":"Seth Brown, Mengli Song, T. Cook, M. Garet","doi":"10.3102/00028312221136565","DOIUrl":null,"url":null,"abstract":"This study examined bias reduction in the eight nonequivalent comparison group designs (NECGDs) that result from combining (a) choice of a local versus non-local comparison group, and analytic use or not of (b) a pretest measure of the study outcome and (c) a rich set of other covariates. Bias was estimated as the difference in causal estimate between each NECGD and a carefully appraised randomized experiment with the same intervention, outcome, and estimand. Results indicated that bias generally declined with the number of design elements in an NECGD, that combining all three sufficed to eliminate bias but was not necessary for it, and that this pattern of results was largely replicated across five different replication factors.","PeriodicalId":48375,"journal":{"name":"American Educational Research Journal","volume":null,"pages":null},"PeriodicalIF":3.5000,"publicationDate":"2022-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"3","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"American Educational Research Journal","FirstCategoryId":"95","ListUrlMain":"https://doi.org/10.3102/00028312221136565","RegionNum":1,"RegionCategory":"教育学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"EDUCATION & EDUCATIONAL RESEARCH","Score":null,"Total":0}
引用次数: 3
Abstract
This study examined bias reduction in the eight nonequivalent comparison group designs (NECGDs) that result from combining (a) choice of a local versus non-local comparison group, and analytic use or not of (b) a pretest measure of the study outcome and (c) a rich set of other covariates. Bias was estimated as the difference in causal estimate between each NECGD and a carefully appraised randomized experiment with the same intervention, outcome, and estimand. Results indicated that bias generally declined with the number of design elements in an NECGD, that combining all three sufficed to eliminate bias but was not necessary for it, and that this pattern of results was largely replicated across five different replication factors.
期刊介绍:
The American Educational Research Journal (AERJ) is the flagship journal of the American Educational Research Association, featuring articles that advance the empirical, theoretical, and methodological understanding of education and learning. It publishes original peer-reviewed analyses that span the field of education research across all subfields and disciplines and all levels of analysis. It also encourages submissions across all levels of education throughout the life span and all forms of learning. AERJ welcomes submissions of the highest quality, reflecting a wide range of perspectives, topics, contexts, and methods, including interdisciplinary and multidisciplinary work.