Multilevel measurement models: Improving assessment of time-varying constructs with a limited number of observations per person
Open Access
- Author:
- McAleavey, Andrew Athan
- Graduate Program:
- Psychology
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- April 24, 2015
- Committee Members:
- Louis Georges Castonguay, Dissertation Advisor/Co-Advisor
Louis Georges Castonguay, Committee Chair/Co-Chair
Stephen Jeffrey Wilson, Committee Member
Jeffrey Hayes, Committee Member
Benjamin D Locke, Committee Member
Peter Cm Molenaar, Committee Member - Keywords:
- Multilevel Structural Equation Modeling
Psychotherapy outcome monitoring - Abstract:
- The most common uses of factor analysis (FA) in psychological measurement may leave uncertainty regarding general versus domain-specific factors, frequently cannot distinguish between traits and states, and do not account for differences between persons when estimating the structure of underlying factors. In part, these problems are due to the use of a single timepoint of assessment for individuals rather than multiple observations. Under circumstances of limited but multiple observations per person, it is proposed that multilevel factor analysis (MLFA) may provide a viable alternative to traditional FA, capable of more reliably distinguishing states and traits, and informing the general vs. specific factor question. In addition, an extension of the MLFA model incorporating randomly-varying measurement parameters may provide an acceptable estimate of idiographic and nomothetic factor structures in large samples, with a relatively small number of observations per person. In two studies, these models are investigated and found to be potentially useful, though also with clear weaknesses in their complexity and computational requirements. In particular, it is suggested that the benefits of such models may most outweigh the costs in routine treatment outcome assessment in applied clinical settings, in which each subject presents with different concerns, theoretical factors of interest are thought to be time-varying, and regular but not intensively repeated observations are made within persons (with as few as 3-10 observations per person being common). Though the studies here remain preliminary, the findings suggest that accounting for multiple observations within persons in psychological measurement may be an important direction for future analyses.