Model sensitivity to prior selection in replication studies

When doing a Bayesian Analysis for a replication study, selecting priors is a widely discussed issue. On one hand, we could argue that an informative prior specified by previous research is preferable because we have some knowledge and expectations regarding the phenomena. However, when the goal is to replicate findings from previous research, we do not want to use prior findings to influence results of the replication study; that is, for a replication study, we should use a non-informative prior, which would maximize the utility of current data. By analyzing a replication research for a widely cited psycholinguistics paper (Fine, Jaeger, Qian, & Farmer, 2013), this thesis aims to provide insight as to how a replication researcher might go about selecting priors for analyzing replication studies within a Bayesian framework. By using sensitivity analyses, posterior predictive checking, and information criteria, researchers can start with a more reasonable prior setting that eventually leads to more valid confirmation or non-confirmation of previous research.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Educational Psychology

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2019

Author dc:creator

• Liu, Ella

Contributors dc:contributor

• Anderson, Carolyn J

• Replication Study, Bayesian Analysis, Prior Selection

Statement dc:rights

• Copyright 2019 Qiawen Liu

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/105756

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/105756