Analyzing Psychopathology Items: A Case for Nonparametric Item Response Theory Modeling.

Abstract

The authors discuss the applicability of nonparametric item response theory (IRT) models to the construction and psychometric analysis of personality and psychopathology scales, and they contrast these models with parametric IRT models. They describe the fit of nonparametric IRT to the Depression content scale of the Minnesota Multiphasic Personality Inventory-2 (J. N. Butcher, W. G. Dahlstrom, J. R. Graham, A. Tellegen, & B. Kaemmer, 1989). They also show how nonparametric IRT models can easily be applied and how misleading results from parametric IRT models can be avoided. They recommend the use of nonparametric IRT modeling prior to using parametric logistic models when investigating personality data.Recently, several authors have introduced and discussed the advantages of applying item response theory (IRT) models (e.g., Embretson & Reise, 2000) to construct personality 1 scales and to explore the structure of personality data sets. For example, Waller, Tellegen, McDonald, and Lykken (1996) contrasted the use of IRT with principalcomponent factor analysis, and Reise and Waller (2003) discussed the choice of an IRT model to analyze psychopathology-test data. That is, they compared the fit of the two-parameter and three-parameter logistic models (PLMs) on 15 unidimensional factor scales from the Minnesota Multiphasic Personality Inventory-Adolescent (MMPI-A; Butcher et al., 1992). Most studies apply parametric IRT models (in particular, the 2PLM and 3PLM) to investigate the quality of personality and psychopathology tests (e.g., Panter, Swygert, Dahlstrom, & Tanake, 1997;Robie, Zickar, & Schmitt, 2001;Steinberg, 1994;Waller, Thompson, & Wenk, 2000).The aim of the present study was to illustrate the usefulness of nonparametric IRT (NIRT) to construct and to analyze psychopathology and personality scales and tests. In our opinion, the use of NIRT has been underexposed in the recent personality literature (for an exception, see Santor & Ramsay, 1998). We show that these models are very suitable to exploring the psychometric properties of personality data. Interesting in this context is a study by Chernyshenko, Stark, Chan, Drasgow, and Williams (2001), who explored the use of NIRT modeling in personality measurement. Chernyshenko et al. fitted the 2PLM, the 3PLM, a graded response model, and Levine's nonparametric maximum-likelihood formula scoring models to dichotomous and polytomous data of the Sixteen Personality Factor Questionnaire (Conn & Rieke, 1994). They concluded that the nonparametric model provided the best fit of the models considered. Chernyshenko et al. and also Reise and Waller (2003) concluded that the response process underlying personality measurement is less well-understood than the response process in the cognitive domain. This being the case, we argue that using NIRT models based on exploring the simple covariance structure between items and based on nonparametric regression will lead to useful information that (a) can be interpreted very easily by practitioners, (b) avoids forcing the data ...

Estimating the level of functional ability of children identified as likely to have an intellectual disability

“…Meijer and Baneke (2004) have argued for the utility of non-parametric IRT models, in particular for the analysis of psychopathology scales. Unlike parametric IRT models, non-parametric models do not impose a specific structure on the relation between item responses and the underlying latent trait (the 'item response function' or IRF).…”

“…This is an advantage when assumptions about the form of the IRF, for example the logistic function, are unrealistic for empirical data. Meijer and Baneke (2004) note that non-parametric IRT models can provide useful information about the performance of items and scales without the need to make these assumptions. In addition, samples in clinical studies tend to be of only modest size due to the relative infrequency of clinical disorders in the population and/or recruitment difficulties.…”

“…When the assumptions of MHM hold, it is possible to infer stochastic ordering on the latent trait, that is, that higher test scores are probabilistic indications of a higher level on that trait. Although scale scores are frequently assumed to possess this property, it is important to explicitly test this assumption (Meijer & Baneke, 2004). In practical terms, evidence that MHM holds for a given scale provides some justification for the use of the scale in clinical practice for tasks that require an ordering of individuals based on severity of the trait of interest (e.g., referral for treatment).…”

Exhaustion, depression and hopelessness in cardiac patients: a unidimensional hierarchy of symptoms revealed by Mokken scaling

“…5 Indeed, factor analytic studies which have attempted to determine the dimensionality of depression and vital exhaustion have been inconclusive, finding either one or multiple factors. [8][9][10] However, this may simply reflect that the scales used assessed differing symptoms, 5 or the fact that factor analysis is not appropriate for binary or highly skewed items, 11,12 and such items are typically seen in vital exhaustion and depression interviews or questionnaires.…”

“…Item response theory states that you can order items along levels of a latent trait, and that you can predict the answers to some items from answers to other items. 11,12 For example, someone who reports feeling hopeless will also probably endorse a 'less difficult' item indicating depressed mood. We used a Mokken scaling in a sample of cardiac patients to determine the dimensionality, and hierarchy, of symptoms of depression and vital exhaustion.…”