FIM measurement properties and Rasch model details

To summarize, we take issue with the criticisms of Dickson & Köhler for two main reasons: 1. Rasch analysis provides a model from which to approach the analysis of the FIM, an ordinal scale, as an interval scale. The existence of examples of items or individuals which do not fit the model does not disprove the overall efficacy of the model; and 2. the principal components analysis of FIM motor items as presented by Dickson & Köhler tends to undermine rather than support their argument. Their own analyses produce a single major factor explaining between 58.5 and 67.1% of the variance, depending upon the sample, with secondary factors explaining much less variance. Finally, analysis of item response, or latent trait, is a powerful method for understanding the meaning of a measure. However, it presumes that item scores are accurate. Another concern is that Dickson & Köhler do not address the issue of reliability of scoring the FIM items on which they report, a critical point in comparing results. The Uniform Data System for Medical Rehabilitation (UDSMRSM) expends extensive effort in the training of clinicians of subscribing facilities to score items accurately. This is followed up with a credentialing process. Phase 1 involves the testing of individual clinicians who are submitting data to determine if they have achieved mastery over the use of the FIM instrument. Phase 2 involves examining the data for outlying values. When Dickson & Köhler investigate more carefully the application of the Rasch model to their FIM data, they will discover that the results presented in their paper support rather than contradict their application of the Rasch model! This paper is typical of supposed refutations of Rasch model applications. Dickson & Köhler will find that idiosyncrasies in their data and misunderstandings of the Rasch model are the only basis for a claim to have disproven the relevance of the model to FIM data. The Rasch model is a mathematical theorem (like Pythagoras') and so cannot be disproven by empirical data once it has been deduced on theoretical grounds. Sometimes empirical data are not suitable for construction of a measure. When this happens, the routine fit statistics indicate the unsuitable segments of the data. Most FIM data do conform closely enough to the Rasch model to support generalizable linear measures. Science can advance!