Flexible and Nonparametric Item Response Models

Flexible and Nonparametric Item Response Models

A common finding in the educational and psychological measurement literature is the failure of simple IRT models to have good fit for certain items. In other words, an assumption regarding the functional form of the relationship between the latent construct and the item responses is violated. This observation sometimes occurs for psychopathology data, data with mixed populations, and for multiple choice educational tests where there is some effect of guessing. Some alternatives are notorioiusly difficult to estimate (3- and 4-parameter logistic models), and some nonparametric methods may not neatly fit the demands of operational testing programs in which policy and high-stakes decisions are made (e.g., patient-reported outcomes, academic entrance and licensure exams, large-scale international educational assessments). In such testing programs, it is often required that a candidate model is able to be used with multiple groups of participants (such as demographic groups, in order to investigate item or test bias) and planned missing data designs (in which participants do not complete all test items), can be estimated relatively quickly, and can be used in conjunction with a computer adaptive test. In addition, retaining items that would otherwise not fit is highly desirable as such items can cost thousands of dollars to develop. Thus, the overarching goal of this line of research is to develop more flexible modeling approaches that can more readily meet these operational demands.

Avatar
Carl F. Falk
Associate Professor of Quantitative Psychology

Publications

Purpose In developing item banks for patient reported outcomes (PROs), nonparametric techniques are often used for investigating …

Large-scale assessments often use a computer adaptive test (CAT) for selection of items and for scoring respondents. Such tests often …

We present a monotonic polynomial graded response (GRMP) model that subsumes the unidimensional graded response model for ordered …

One flexible approach for item response modeling involves use of a monotonic polynomial in place of the linear predictor for commonly …

We present a semi-parametric approach to estimating item response functions (IRF) useful when the true IRF does not strictly follow …

We present a logistic function of a monotonic polynomial with a lower asymptote, allowing additional flexibility beyond the …