Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Human rights --- Human rights --- Minorities --- Minorities --- Legal status, laws, etc --- Legal status, laws, etc

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

The aim of my dissertation research is to examine possible methods for indicating sources of individual differences in binary test data. The data to concentrate on in the dissertation are binary person by item data sets of a size that is not very large in terms of persons and items, which is rather common in psychology, in contrast with educational sciences. The research is restricted to binary data, because in many psychological domains when tests are used, and also in experiments, this type of outcome is rather common. The sources of individual differences are studied in association with item covariates. Item covariates are features of the items that may have an  effect on the success rate. For example, in a test with a design (Embretson, 1985), these features are varied as in an experiment, and can be considered as factors in an experimental design. In other cases, the item covariates are hidden and often are obtained from judgments of the items by experts. The effect of covariates can differ from person to person. For example, a negation tends to make the item more difficult, but perhaps the size of this effect is not equal for all persons. The term heterogeneity is used here for individual differences in the effect of the item covariates. Heterogeneity implies sources of variation beyond the fixed effects of the items and their covariates. Person-based heterogeneity implies individual differences which play a major role in psychological research, where predictions are made about achievement, psychiatric problems, etc. Individual differences are especially important in the educational context, in personality, work and social psychology. For prediction as well as for explanation, individual differences are highly relevant. In order to approach heterogeneity, a random effects concept will be used, meaning that the individual differences are considered to stem from an underlying distribution, which will always be the normal distribution in this dissertation when data are generated for a simulation study. The detection and modeling of random effects may lead to a better understanding of data through a more appropriate modeling, and the possibility to link the individual differences to external variables. For the purpose of modeling, the Item Response Theory (IRT) framework provides a wide variety of models that can be applied. Especially relevant a model formulation for binary data with item covariates linked to random effects is offered by the Random Weights Linear Logistic Test Model (RWLLTM; Rijmen & De Boeck, 2002). The detection approach suggested in the title means that not a hypothesis-testing approach is followed in this dissertation, concentrated on testing a predefined set of heterogeneity sources. The reasons for this kind of detection approach can be theoretical, because even when one is not really interested in these sources of heterogeneity, one would still want to take them into account to avoid misspecifications and distortion effects on some crucial estimates. The reasons can be also practical, for example, if there are many potential sources, and one wants to include only those that make a chance to be substantial. Finally, the reasons can be also of a general explorative kind. In the studies described in the four chapters, different diagnostic procedures are investigated for the detection of heterogeneity linked to latent or manifest item covariates. As explained earlier, the item covariates do not need to be apriori features of the items, they can, for example, also stem from an aposteriori judgment of items by experts. The aim of the dissertation is related to detecting multidimensionality and detecting local item dependence. Therefore, the concept of multidimensionality and local item dependence provide alternative approaches to the research problem of the dissertation.