Date of Award

1-1-2016

Document Type

Dissertation

Degree Name

Ph.D.

Organizational Unit

Morgridge College of Education, Research Methods and Information Science, Research Methods and Statistics

First Advisor

Antonio Olmos, Ph.D.

Second Advisor

Kathy Green

Third Advisor

Kellie Keeling

Keywords

Multiple treatment, Propensity score, Monte Carlo simulation

Abstract

The application of propensity score techniques (matching, stratification, and weighting) with multiple treatment levels are similar to those used in binary groups. However, given that the application of propensity scores in multiple treatment groups is new, factors affecting the performance of matching, stratification, and weighting in multiple treatment groups are less explored. Therefore, this study was conducted to determine the performance of different propensity score techniques with multiple treatment groups under various circumstances. Specifically, the study focused on examining how the three propensity score corrective techniques perform in estimating treatment effects under (1) overt and (2) hidden types of selection bias. In this study, the performance of propensity score matching, stratification, and weighting techniques were tested under three different sample sizes and three levels of overt and hidden bias.

A Monte Carlo simulation was used to generate data with specific sample sizes and levels of overt and hidden bias. A total of 54 data conditions with 1000 replications for each condition was generated to compute the average treatment effect (ATE). The difference between the pre-specified ATE and estimated ATE was calculated to evaluate the performance of propensity score techniques. Two 3x3x3x2 analyses of variance were conducted to assess the effect of propensity score technique, level of bias, sample size, and type of treatment effect on the amount of bias in estimating the treatment effect under overt and hidden bias conditions.

The results provided four key findings of information about the application of propensity score analysis in multiple treatment groups. The first key finding is that the treatment effect estimate will be underestimated after imposing propensity score adjustments. Second, the treatment effect estimates are affected by the level of overt bias. Third, propensity score analysis does not account for hidden bias. The fourth finding is that the propensity score techniques performed differently in a small sample size condition. Overall, these four key findings provide cautionary notes to the users of propensity score analysis in multiple treatment groups. The study is concluded with the limitations of this study and the recommendations for future research.

Keywords: Propensity score, multiple treatment

Publication Statement

Copyright is held by the author. User is responsible for all copyright compliance.

Rights Holder

Priyalatha Govindasamy

Provenance

Received from ProQuest

File Format

application/pdf

Language

en

File Size

436 p.

Discipline

Statistics

Download



Share

COinS