The revealing structure of concepts: R-based 6-step analysis for health psychology research

Abstract

Background: Questionnaire-based research is prevalent in health psychology, and commonly employs several related questions (scales) to assess individual concepts. Therefore, the quality of evidence depends substantially on scale quality. Yet, scale validation is often insufficient, or it is performed in separate studies. Recent developments within the R environment make psychometric analyses easier to perform together with statistical analyses for substantive research questions. I present a 6-step procedure that gives an overview of scale quality using various R functions and psychometric theory. Methods: The 6-step analysis examines item distributions (descriptive statistics), item properties (item response theory; IRT), scale structure (factor analysis), scale reliability (classical test theory), and clustering of respondents (cluster analyses), and applies decision rules for item selection. I illustrate the procedure on a 24-item behavioural measure of health status, the Sickness Impact Profile, administered within a survey on living with chronic pain in the United Kingdom. Findings: Items reflected the impact of chronic pain on the 222 respondents differently (4.5% to 87.2% endorsement rates). IRT analyses identified an 18-item unidimensional ordinal scale with good reliability and distribution (H=0.47; α=.88[.84-.91]; ω=.89[.86-.91]; mean(SD)=10.23(4.61); range 3-18) that showed monotonicity and invariant item ordering. Interval scaling assumptions were not met. Factor analyses partially converged with IRT findings. No distinct respondent clusters emerged. Discussion: Examining scale structure enables an in-depth understanding of the phenomena studied. I discuss the implications of the example findings, and the possibilities of adapting the 6-step analysis for various research needs and thus improving the use of health psychology concepts.