2011 HSR&D National Meeting Abstract
2011 — Psychometric Overview for Proper Development, Validation, and Evaluation of Self-Report Measures
Voils CI (Durham VA Medical Center), Maciejewski ML
(Durham VA Medical Center), Yancy Jr. WS
(Durham VA Medical Center)
Self-reported measures are often used in health services research studies, but many measures are not developed and validated consistent with psychometric theory. Of particular import is the distinction between two different types of scale items (indicators): effect indicators and causal indicators. Effect indicators derive their value from the latent construct they represent, so item responses change as the underlying construct changes. Causal indicators, in contrast, are partial, additive components of a latent construct. The distinction between effect and causal indicators is important because it determines the type of reliability and validity analyses that are appropriate. Conducting inappropriate psychometric analyses may lead to erroneous conclusions regarding reliability and validity, thereby hampering the ability to detect statistically significant relationships and limiting the evidence base. The purpose of this workshop is to provide researchers an overview of psychometric theory of effect indicator models and causal indicator models and the appropriate reliability statistics for each type of model so that researchers can properly develop, validate, and evaluate self-report measures.
The presenters will present causal and effect indicator-based models of latent constructs; describe the types of psychometric analyses that are appropriate to provide evidence of reliability and validity for each model; and demonstrate these principles using existing measures of constructs commonly assessed in health services research (e.g., medication nonadherence and quality of life).
This workshop is targeted to researchers interested in instrument development and validation for a range of health behaviors, and to individuals interested in applying validated, theory-based self-report measures.
Assumed Audience Familiarity with Topic:
No familiarity with the topic will be required.