Survey rating scale data can be assessed for validity and internal consistency, and, if found of satisfactory quality, can be transformed from ordinal to interval data, by fitting them to a probabilistic conjoint (Rasch) measurement (PCM) model. PCM models aid in assessing, improving, and linearizing ordinal rating scale data. Fitting data to a PCM model's requirements tests the hypothesis that the variable measured is quantitative, revealing anomalies and inconsistencies in the data that could otherwise obscure or confound the object of measurement, making it possible to correct or remove them. When the quantitative hypothesis is not falsified, and data fit the chosen model, several scientific and statistical advantages follow, including equal interval units of measurement (logits); individualized error terms, used in calculating reliability; several kinds of individualized model fit statistics, used in analyzing construct validity; and the ability to account for missing data. The latter means that 1) more data can be retained and made useful when items are skipped; 2) instruments can be adapted to people, instead of vice versa, by allowing people to skip irrelevant items; 3) different instruments measuring the same variable can be equated to measure in the same unit; and 4) an instrument can be revised without making existing data incommensurable with new data. This presentation will describe 1) dichotomous, polytomous, partial credit, and multifaceted fundamental measurement models; 2) approximative and unconditional maximum likelihood methods for parameter estimation; and 3) mean square information-weighted and outlier-sensitive model fit statistics.
Learning Objectives: At the conclusion of this presentation, the participant learner will be able to: 1. recognize mathematical models that test data for adherence to the requirements of objective inference; 2. identify at least two methods for estimating parameters for these models; and 3. discuss the value of evaluating data quality via model fit statistics
Keywords: Biostatistics, Outcome Measures
Presenting author's disclosure statement:
Organization/institution whose products or services will be discussed: None
I do not have any significant financial interest/arrangement or affiliation with any organization/institution whose products or services are being discussed in this session.
The 128th Annual Meeting of APHA