Date of Award:

Spring 2017

Document Type:

Thesis

Degree Name:

Master of Science (MS)

Department:

Psychology

Advisor/Chair:

Eric N. Reither

Abstract

For the last four decades, scholars have made significant efforts to develop statistical techniques to estimate the independent contributions of three temporal dimensions (i.e. age, period, and cohort) to population health and other outcomes. These efforts have been challenged by the “identification (ID) problem,” a statistical conundrum that occurs due to an algebraic dependence between the three temporal terms. Hierarchical Age-Period-Cohort (HAPC) modeling and Intrinsic Estimator (IE) methods, which are two of the most recent and important innovations in Age-Period-Cohort (APC) analysis, provide unique model specifications to address the ID problem. However, recent critiques have questioned the validity of these two methods in properly addressing the ID problem by presenting evidence that both have limitations, including potentially invalid estimation of age, period, and cohort effects. In this dissertation, I test the argument advanced by proponents of HAPC and IE methods that each of them provides unbiased estimation of parameter values when the data structure satisfies model assumptions. In Chapters 2 and 3, I conduct a series of simulation analyses to assess the validity of these claims, as well as the usefulness of preliminary analyses (i.e., descriptive and model selection statistics) in identifying data structures that are compatible with APC models. In Chapter 4, I provide a step-by-step demonstration of the HAPC method to empirical data to study how age, period and cohort contribute to educational inequalities in health in United States. The results from these analyses indicate that descriptive and model selection statistics are useful in identifying temporal data structures prior to the application of HAPC and IE models, and that these methods tend to provide unbiased estimates when the data structures are three-dimensional. Furthermore, even when the data structures and corresponding “best models” were ambiguous, it was possible to utilize APC methods by cross-validating nested models.

Checksum

03939b9c3e98d407ec01042c7df90968

Share

COinS