Impact of cluster sampling on scale psychometrics: simulation study and application to mental health survey
Loading...
Date
2018
Authors
Chen, Xuan
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Cluster sampling designs are frequently used in mental health surveys and prevention studies. The overall purpose of this thesis research is to investigate the impact of cluster sampling on scale psychometric properties and the psychometrics of a mental health assessment tool in Canadian culture.
We conducted the simulation study to examine the impact on scale psychometrics of ignoring the non-independence of subjects within cluster. Results indicated that: (a) as the dependence among observations (i.e., ICC) increases, the model goodness of fit become worse or even not acceptable if we specified a single-level model for a multilevel data; (b) Single-level reliability estimates would consistently estimate reliability at both levels if the true reliability at both levels was the same or ICC is low; (c) Single-level reliability estimates would fall in the interval of true reliability at individual level and the true reliability at the school level.
We also used data from Manitoba provincial Grade 5 mental health survey to examine the psychometrics of Strength and Difficulty Questionnaire (SDQ) as well as the influence of cluster sampling. Results indicate that the 5 factor structures identified in other cultures fit the Canadian sample well and the estimates of psychometrics (e.g., reliabilities) fell into reasonable range if we use the single level model.
The study provides guidance for estimation of psychometrics with cluster sampling. Empirical analyses of psychometric properties of the Canadian SDQ provide supports for the usefulness of the SDQ as a screening tool for mental health of children and youth in the general Canadian population.
Description
Keywords
Cluster sampling, Multilevel CFA, Scale psychometrics