The proposed research is intended to demonstrate the potential Probabilistic Conjoint Measurement (PCM) models offer for measuring health-related variables with a rigor and convenience equivalent to the measurement of physical variables. Measurement of this kind is required for creating the common currencies needed for exchanging quantitative values representing more and less health. Given the ostensible goal of health care reform, to pay for health and not service, the calibration of universally-accepted reference standards for health-related variables ought to be in great demand. Five studies are proposed to be conducted over a period of six months with the aim of demonstrating PCM's advantages for research employing surveys and Likert rating scales in health care, using the 23,767 cases from, AHCPR's Medical Expenditure Panel Survey (MEPS). PCM models test data for properties that enable the construction of non-arbitrary quantitative (interval or ratio) measures from ordinal observations, with many benefits supporting scientific generalization. Use of these models in health service research is growing, but few standards are available to guide their implementation, analysis, interpretation, or reporting. To begin addressing these issues, AHCPR's Round 2 health status and access to care data from the MEPS were downloaded from the AHCPR web site and prepared for analysis. A preliminary analysis of MEPS survey items involving families' perceptions of health care quality (Q-USC) were completed, with promising results. The proposed research will build on this analysis to 1) test the hypothesis that the Q-USC and BEP variable were quantitative, by fitting the data to PCM models; 2) illustrate item analysis, which involves testing data fit to measurement models requiring unidimensionality, and interpreting the item-scale hierarchy; and 3) test the invariance and robustness of the items' scale values and the person measures by calibrating subsamples of items on all respondents, and measuring subsamples of persons using all items, and comparing the resulting sets of calibrations and measures. After establishing a precise measurement system for the Q-USC variable, 4) quality will be compared across three different types of health-care recipients: Private HMO, Public HMO, and Medicare. Finally, 5) results from the prior four studies will be compared with results from the application of other psychometric methods, such as the Likert summated ratings method, and IRT models, in order to clarify misconceptions regarding the requirements of objectivity in measurements. The proposal concludes with suggestions for a multi-center pilot of interlaboratory trials and reference standard implementations in live clinical settings.
