Graded response modeling of the Quality of Life Interview

A graded response model [Samejima, F. (1969). Ps)Psychometrika Monograph Su pplement, 17; Samejima, F. (1972). Psychometrika Monograph Supplement, 18; Samejima, F. (1997). Handbook of Modern Item Reponse Theory (pp. 85-100). N ew York: Springer Verlag] was outlined and applied to an aggregated data se t (n = 1805) from four studies involving the subjective items from the Qual ity of Life Interview (QOLI) [Lehman, A. F. (1988). Evaluation and Program Planning, 11, 51-62]. Detailed item analyses was performed through use of t he MULTILOG program [Thissen, D. (1991). MULTILOG (Version 6). Moorsville, IN: Scientific Software]. Item parameters, their standard errors, item and scale weighted information and subject latent trait (theta) levels were obt ained. The relationship between IRT information, predicted reliability r(<( theta)over cap>(1),<(theta)over cap>(2)) [Samejima, F. (1994). Applied Psyc hological Measurement, 18, 229-244] and coefficient alpha was demonstrated for this data set. Based on the results, including information and reliabil ity levels, several short, customized QOLI scales were created. This graded response model may be used to meet important scale development needs and i s appropriate for scales involving ordered, graded categories often used in mental health research. (C) 1999 Elsevier Science Ltd. All rights reserved .