Date of Award

2018

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Graduate Group

Statistics

First Advisor

Shane Jensen

Second Advisor

Dylan Small

Abstract

Matching allows us to estimate the effect of a chosen variable, providing highly interpretable inference without parametric assumptions. When matching, finding good controls is where nearly all the difficulty lies. We develop a theoretical framework and a methodology to generate a set of matches, evaluate them and select a best match given the input variables. We apply this method to a problem of interest, urban data in Philadelphia. In this setting, we also outline our full data collection pipeline in order to encourage replication. In a separate time series setting, we propose a latent model in order to generate probabilities at each time point; these form the basis of an interrupted time series match.

Files over 3MB may be slow to open. For best results, right-click and select "save as..."

Share

COinS