日本行動計量学会 岡山地域部会 第69回研究会
テーマ：事前・事後テストデザインにおける処遇効果の評価：事前テストの測定誤差の影響
日 時：2018年12月23日(日・祝) 15:00～17:30
場 所：岡山理科大学 Ａ１号館１階 プレゼンテーションルーム
プログラム：
座長：山田剛史（岡山大学）
15:00 Yasuo Miyazaki (Virginia Tech, USA)
Bias in Treatment Effect Estimates when Pretest has a measurement error in Pretest-Posttest Design ANCOVA Analysis
17:00 総合討論
17:30 閉会
Abstract:
Consequences of measurement error in pretest on the treatment effects’ estimate are investigated in the context of program evaluation. In program evaluation, pretest-posttest design is a frequently employed research design in evaluating the treatment effects. In such settings, pretest is used as a covariate to adjust for the pre-existing differences in an analysis of covariance (ANCOVA) model. One of the key assumptions of the ANCOVA model is that the pretest as a covariate is measured without error. However, in the real-world settings, almost all the psychological measurements involve some amount of measurement errors. Therefore, we wanted to know how measurement error in pretest influenced the treatment effect estimates. A simulation study was conducted in various conditions of the size of the measurement error represented as the reliability coefficient and the size of the pre-existing difference in pretest scores between two conditions. The results indicated that it could be overestimated or underestimated, and in certain conditions, positive treatment effects can be estimated as negative effects. This possibility was confirmed by deriving the closed-form expression of the bias.