The Rasch model

The Rasch model is a simple logistic latent trait Item Response Theory (IRT) model concerned with the quality of outcome measures. Scales that fit the Rasch model have true interval level measurement. This is essential for adding a set of items to firm an unidimensional scale, calculating means and standard deviations and applying parametric statistical analyses. It is important to note that few available PROs collect data that fit the Rasch model.

How can Rasch analysis help with your study?

  • Rasch analysis should be used in the development of new PROs to identify sets of items that form unidimensional scales. The resulting measure will have good measurement properties from the outset.
  • Older scales can be tested to assess whether individual scale items fit the Rasch model, to see whether items are free from differential item functioning (DIF; bias due to extraneous variables), to check that the response options work appropriately and that the items cover a good spread of the underlying trait (ie. it produces low levels of basement and ceiling effects). Such analyses are crucial as so few PROs have been developed using Rasch analysis.
  • Rasch analysis can be used to improve existing measures. Alterations can be made to existing measures to improve the measurement properties of the scale. For example, items can be removed from a PRO and the response options altered in order to improve the measure’s validity and functioning and lead to the production of more sensitive and responsive data. Such analyses can be conducted on existing PRO datasets.
  • The production of Rasch based estimates also justifies the use of parametric analyses – increasing the power of a clinical trial. As the Rasch model produces interval level measurement questionnaire data can be used for the calculation of change scores in clinical trials, analysis of variance and regression analysis.
  • Rasch analysis can be used to investigate the comparability of data from different countries. Despite careful adaptation measures will differ in terms of how patients respond due to differences in culture. Investigating the DIF associated with country can identify and overcome such cultural differences.
  • The application of Rasch scoring of PRO data provides a valid method of dealing with missing responses to a PRO.
  • Rasch analysis can be employed to test the validity of datasets. By identifying centres or patients that respond in unexpected ways the quality of trial data can be improved.