Applying Item Response Theory to the OPD Structure Questionnaire: Identification of a Unidimensional Core Construct and Feasibility of Computer Adaptive Testing

Abstract

Recent developments in the dimensional assessment of personality functioning have made the implementation of latent measurement models increasingly attractive. In this study, we applied item response theory (IRT) to a well-established personality functioning instrument (the OPD Structure Questionnaire) to identify a unidimensional latent trait and to evaluate the feasibility of computer adaptive testing (CAT). We hypothesized that the use of IRT could reduce the test burden – compared to a fixed short form – while maintaining high precision over a wide range of the latent trait. The OPD-SQ was collected from 1235 patients in a psychosomatic clinic. IRT assumptions were fulfilled. A 9-factor model yielded sufficient fit and unidimensionality in exploratory factor analysis with bifactor rotation. Items were iteratively reduced, and a graded-response IRT model was fitted to the data. Simulations showed that a CAT with approximately 7 items was able to capture an OPD-SQ global severity score with an accuracy similar to that of a fixed 12-item short form. The final item bank and CAT yielded satisfactory content validity. Strong correlations with depression and anxiety replicated previous results on the OPD-SQ. We concluded that IRT applications could be useful to reduce the test burden of personality functioning instruments.

Acknowledgments

We thank Sarah Goldstein from the Center of Self-Report Science, University of Southern California, Los Angeles, for her great support in proofreading the manuscript.

Data availability statement

The data that support the findings of this study are available from the corresponding author, A.O., upon reasonable request.

Notes

1 Development process and Results on psychometric properties of the OPD-SQ are described in detail here as the available literature is predominantly in German.

2 Excluded due to residual correlations: OPD09, OPD14, OPD28, OPD42, OPD43, OPD77, OPD78, OPD84

3 Excluded due to non-discriminative item-response curves: OPD3, OPD6, OPD10, OPD11, OPD12, OPD13, OPD17, OPD18, OPD19, OPD20, OPD25, OPD29, OPD30, OPD31, OPD32, OPD34, OPD35, OPD36, OPD38, OPD40, OPD46, OPD47, OPD52, OPD53, OPD54, OPD55, OPD56, OPD57, OPD59, OPD60, OPD62, OPD63, OPD64, OPD66, OPD68, OPD69, OPD70, OPD71, OPD72, OPD79, OPD83, OPD86, OPD87, OPD89, OPD91, OPD93, OPD94, OPD95

4 Excluded due to bad fit: OPD15, OPD51, OPD85

5 One count corresponds to one standard deviation

Additional information

Funding

This work was supported by a research fellowship (OB 437/2-1, recipient: A.O.) by the German Research Foundation (Deutsche Forschungsgemeinschaft, DFG).