Skip to main content
eScholarship
Open Access Publications from the University of California

Essays in psychometrics and behavioral statistics

  • Author(s): Gochyyev, Perman
  • Advisor(s): Wilson, Mark
  • et al.
Abstract

This dissertation consists of three chapters. The main focus of the first chapter is on Lord’s paradox. Lord’s paradox arises from the conflicting inferences obtained from two alternative approaches that are typically used in evaluating the treatment effect using a pre-post test design. The chapter is designed as a guide to researchers who are using this research design. As an example, I investigate whether the treatment—a new mathematics curriculum—had an effect on student-level outcomes using both approaches. I demonstrate that Lord’s paradox can occur even when the two approaches are accounting for the measurement error in variables.

Ordinal response data obtained from surveys and tests are often modeled using cumulative, adjacent-category, or continuation-ratio logit link functions. Instead of using one of these specifically designed procedures for each of these formulations of logits, we can modify the structure of the data in such a way that methods designed for dichotomous outcomes (i.e., binary logistic regression) allow us to achieve the targeted polytomous contrasting (cumulative, adjacent-category, or continuation-ratio). Thus, one can implement procedures designed for dichotomous outcomes on appropriately expanded data. The techniques presented in the second chapter, which I refer to as data expansion techniques, represent this approach.

The third chapter aims to contribute to the estimation and interpretation of multidimensional item response theory (MIRT) models within the field of psychometrics and latent variable modeling. The main goal of the chapter is to advance the use of the second-order Rasch model. A second-order Rasch model assumes an overall dimension as a second order factor that explains the covariance between the first-order (component) dimensions. The main contribution of the chapter is to suggest ways of using the model by still preserving the advantages of the Rasch model. Historically, the main challenge in the use of such models were (1) computationally intensive estimation and (2) availability of software. In addition, it is difficult to obtain reliable and meaningful estimates in cases when a variance of one of the dimensions is low relative to other dimensions. In such cases, one first needs to re-assess if the multidimensional structure is appropriate. One, then, can use alternative parameterization of the model to avoid difficulties in the estimation, and guidelines in this chapter provide recommendations on how to achieve such parameterizations with the Rasch model.

Main Content
Current View