Saturday, November 12
Data Quality and Measurement Error
Sat, Nov 12, 4:00 PM - 5:25 PM
Orchid AB
Statistical Methods to Assess Data Quality

Geographically Weighted Item Response Theory (IRT) for Detection of Spatially Varying Item Functionality (303314)

*Samantha Elizabeth Robinson, University of Arkansas 

Keywords: Item Response Theory, Differential Item Functioning, Measurement Invariance, Spatial Analysis, Geographically Weighted Modeling

Mappings of spatially-varying Item Response Theory (IRT) parameters are proposed, allowing for visual investigation of potential Differential Item Functioning (DIF) based upon geographical location without need for pre-specified groupings and prior to any confirmatory DIF testing. This proposed model merges Geographically Weighted Regression (GWR) and IRT methods, with current emphasis being on a 1PL/Rasch model. This geographically weighted approach to IRT modeling and DIF detection provides a flexible framework, with various extensions discussed. Applications to simulated examination data, utilizing a boxcar kernel weighting scheme with fixed and adaptive bandwidths on both regular and irregular lattices, illustrates this method’s benefit and practical value as a pretesting method for questionnaire design and assessment, especially when comparisons are made to traditional DIF techniques. There is not only practical value with this method but also visual appeal when initial attempts to consider measurement invariance are being made across national, state, or other political boundaries. This approach, making use of three-dimensional surface mappings of estimated item difficulty parameters, serves to detect DIF across space without a priori groupings, thereby identifying regional disparities and latent spatial trends in item functionality that may be unobservable on a global level.