Wednesday, March 1, 2017 at 12:30 p.m.
Room 203, Graduate Student Centre (6371 Crescent Road), UBC Point Grey Campus
Supervisor: Dr. Bruno Zumbo
Supervisory Committee: Dr. Amery Wu
University Examiners: Dr. Sterett Mercer, and Dr. Jeremy Biesanz (Psychology)
External Examiner: Dr. Craig Wells (University of Massachusetts-Amherst)
Title: “On Monte Carlo Simulation Algorithms for Research in Psychomatrics”
ABSTRACT
Monte Carlo simulations have become the workhorse of the modern methodologist aimed at providing both novel statistical insights and to guide data analysis practice. In spite of its widespread use, familiarity with data-generating algorithms is rare among users and consumers of simulation-based research, making the process appear as a “black box” of sorts. Without a good understanding of these algorithms, design flaws can appear in Monte Carlo studies which can influence the recommendations offered to applied researchers. In order to address these potential problems, this dissertation will highlight three issues in three separate papers related to the process of simulation as well as potential recommendations to deal with them. The first paper (chapter 2) focuses on the importance of matching the population model with the simulation design underlying the researcher’s hypothesis. It takes the Spearman rank correlation as a case study and documents the impact that potential disparities between simulation design and methodology can have on the conclusions derived from computer studies. The second paper (chapter 3) investigates a popular data-generating method within the social sciences, the Vale-Maurelli algorithm, and compares its results to a second one, the Headrick method, in terms of the kind of data they can generate and how this influences simulation results within a Structural Equation Modelling framework. The third paper (chapter 4) takes a closer look at the both the univariate (Fleishman) and multivariate (Vale-Maurelli) versions of the 3rd-order polynomial transformation to generate correlated, nonnormal data and documents the impact that its multiplicity of solutions has on simulation study results. In conclusion, this dissertation has the ultimate goal to help illuminate the process of simulation to psychometricians and social scientists alike in order to help create better study designs and promote a critical evaluation of Monte Carlo studies among methodologists and applied researchers alike.