Psychometric Methods to Develop and to Analyze Clinical Measures: A Comparison and Contrast of Rasch Analysis and Classical Test Theory Analysis of the PedsQL 4.0 Generic Core Scales (Parent-report) in a Childhood Cancer Sample

Abstract

<p>Traditionally, measures have been developed using Classical Test Theory (CTT). Modern psychometric methods (e.g. Rasch analysis) are being applied to increase understanding of item-level statistics and to aid in interpreting rating scale scores. This thesis aims to compare and contrast psychometric findings for the PedsQL^TM 4.0 Generic Core Scales using CTT and Rasch analysis to determine if a Rasch approach provides information that furthers our understanding of scale scores. The assumptions, advantages and limitations of each psychometric paradigm are presented.</p> <p>Issues that arise when measuring quality of life are discussed to set the stage for a psychometric analysis of the PedsQL^TM in a childhood cancer sample. The PedsQL^TM measures child health in terms of physical, social, emotional and school function. The parent-report version was used in a Canadian study of 385 parents of children aged 2 to 17 years on active cancer treatment and data was re-analyzed for this thesis. CTT analysis was performed using PASW Statistics and Rasch analysis was performed using Rumm2030.</p> <p>Internal consistency reliability was higher using CTT (a = 0.93) than Rasch analysis (Person Separation Index = 0.78). Rasch analysis item curves showed respondents did not discriminate between response categories and a 3 point scale (vs. 5) was preferred. Item curves also indicated most items were free of bias. There are no equivalent visual representations in CTT of how respondents use response categories or of whether items display bias. Both approaches indicate a large ceiling effect associated with the overall score.</p> <p>Results challenge internal consistency reliability of the PedsQL^TM 4.0. Rasch analysis permits detailed and visually pleasing examination of item-level statistics more effectively than CTT. Research is needed to determine which testing circumstances render Rasch analysis useful and justify time and resources to use both paradigms as complementary tools to maximize understanding of rating scale scores.</p>