Using graphical loglinear rasch models to investigate the construct validity of the perceived stress scale

Abstract

The Rasch model has been generalized and extended into what is now known as the class of Rasch models. In this chapter, we will explain in nontechnical terms the extension known as graphical loglinear Rasch models (GLLRM), which can be used to test model with departures from the pure Rasch model in terms of uniform local dependence (LD) or uniform differential item functioning (DIF). To demonstrate the utility of these models, we display the psychometric properties of the perceived stress scale (PSS) in two studies conducted in Australia and Denmark. Although the studies differed in cultural context, nature of the sample (nationally representative n = 3,857 and higher education students n = 1,552), and version of the PSS used (PSS-14 and PSS-10), consistent results were found. The analysis showed that the PSS consists of two subscales (Perceived Stress and Perceived Lack of Control), which is congruent with previous CFA literature. In addition, in both countries Items 7 and 10 were locally dependent and Items 1 and 3 displayed DIF by gender. For the Australian nationally representative sample, targeting was poor for both subscales, while for the Danish sample of higher education students targeting was excellent. Implications regarding the application of the PSS are discussed.