The generalized graded unfolding model (GGUM) is developed. This model allows for either binary or graded responses and generalizes previous item response models for unfolding in two useful ways. First, it implements a discrimination parameter that varies across items, which allows items to discriminate among respondents in different ways. Second, the GGUM permits response category threshold parameters to vary across items. A marginal maximum likelihood algorithm is implemented to estimate GGUM item parameters, whereas person parameters are derived from an expected a posteriori technique. The applicability of the GGUM to common attitude testing situations is illustrated with real data on student attitudes toward abortion. Index terms: attitude measurement, generalized graded unfolding model, graded unfolding model, item response theory, Likert scale, marginal maximum likelihood, Thurstone scale, unfolding model.Several researchers have recently argued that binary or graded agree-disagree responses to attitude statements generally result from an ideal point process (Coombs, 1964) in which a person endorses an attitude statement to the extent that it matches the person's opinion (This argument implies that an unfolding (i.e., proximity) model that implements a single-peaked response function would be best for analyzing agree-disagree responses, including both binary and graded responses. In the context of item response theory (IRT), an unfolding model suggests that a person will agree with a statement to the extent that the person and the statement are located near each other on an underlying continuum-a latent continuum that spans the two poles of negative and positive affect.Several unidimensional item response models are available for unfolding agree-disagree responses to attitude statements. Some of these models are appropriate for binary responses, whereas others allow for either binary or graded data. Models for binary data use parametric approaches (Although nonparametric models are practical because they make fewer assumptions about the specific form of the item response function, correctly specified parametric models offer additional measurement advantages. Attitude estimates from parametric models are invariant to the items used to calibrate the estimates. Additionally, estimates of item locations are invariant to the distribution of attitudes in the sample. These two qualities emerge from the fact that if an unfolding model correctly and completely specifies response category probabilities, given only the item and the

Although the belief has been expressed that performance assessments are intrinsically more fair than multiple‐choice measures, some forms of performance assessment may in fact be more likely than conventional tests to tap construct‐irrelevant factors. The assessment of differential item functioning (DIF) can be helpful in investigating the effect on subpopulations of the introduction of performance tasks. In this study, two extensions of the Mantel‐Haenszel (MH; 1959) procedure that may be useful in assessing DIP in performance measures were explored. The test of conditional association proposed by Mantel (1963) seems promising as a test of DIF for pofytomous items when the primary interest is in the between‐group difference in item means, conditional on some measure of ability. The generalized statistic proposed by Mantel and Haenszel may be more useful than Mantel's test when the entire response distributions of the groups are of interest. Simulation results showed that, for both inferential procedures, the studied item should be included in the matching variable, as in the dichotomous case. Descriptive statistics that index the magnitude of DIP were also investigated.