Best tests of treatment effect for matching with multiple controls
Abstract
This thesis searches for the best tests of treatment effect for matching with multiple controls in which each treated subject may be matched with one or more controls. First the paired t-test is extended to the more general situations where each matched set may contain more than one treated subjects when the underlying distribution is normal. Based on the generalized t-test the issue of match quality vs. design efficiency is also discussed. In the non-normal case a number of rank tests are compared using the asymptotic relative efficiency as a criterion. Then the Hodges-Lehmann (H-L) estimate of the treatment effect and the confidence intervals based on the H-L estimate are developed.
Subject Area
Statistics|Biostatistics
Recommended Citation
Zhang, Hao, "Best tests of treatment effect for matching with multiple controls" (1994). Dissertations available from ProQuest. AAI9521150.
https://repository.upenn.edu/dissertations/AAI9521150