SMABS 2004 Jena University
SMABS 2004 Home Organization About Jena Sponsors Links Imprint / Contact SMABS Home

European Association of Methodology

Department of methodology and evaluation research

Jena University

Contributions: Abstract

Construction of a short-form of an assessment instrument for rehabilitation patients by Mixed Rasch Analysis

Markus Wirtz
Universit├Ąt Freiburg

The IRES-Questionnaire for patients is one of the most frequently used assessment tools in medical rehabilitation in Germany. The short-form IRES-24 was developed by means of Item Response Theory, based on the data of 1840 patients, who took part in a stationary orthopedic rehabilitation treatment. The short-form of the questionnaire was developed to assess the four dimensions Psychic condition, Functionality in every day life, Somatic health, and Burden of pain.

Following missing-data-diagnosis and imputing missing data by EM-algorithm, items were selected with the purpose to ensure that data on each scale could be modelled by one latent dimension and ordered thresholds. The item fit index Q achieved very good and non-significant values for all items.

A crucial assumption for any Rasch-Scale is the homogenity of persons. As this assumption was not met for the whole sample, mixed Rasch analysis was used to identify the sources of heterogenity. Distinguishing among different classes of patients led to class-specific solutions which fulfill all Rasch criteria for the first three scales.

For each of the two scales Psychic condition and Functionality in every day life, classes differed only regarding to the response sets tendency to the mean and tendency to extreme values. For Somatic health, different latent variables are assessed in different classes. For a small class of patients on the scale Burden of pain, not all criteria for a Rasch-scale are met.

Using Latent Class Analysis, a strong association of class membership across scales was found, proving in particular the stability of response sets across scales. For each scale, class membership can be predicted very well from the variability of scores of the underlying scale items, the number of extreme values and the item mean values. For clinical assessment, a two stage diagnosis is recommended: in the first stage class membership should be determined, followed by the specification of class-specific person parameters.