Practical Significance of Item Response Theory Model Misfit:Much Ado About Nothing?

This thesis centers around practical applications of psychological, educational, and health assessment. In psychometric testing, complex statistical equations are often used to approximate individuals’ levels of the measured characteristic, based on their response patterns. These levels (or scores) are then used to draw conclusions and to make decisions regarding the tested individuals. For instance, a psychologist might calculate a child’s score on an attention problems test to assess whether the child experiences attention deficits; a university might use the scores on an admission exam in order to select students; or a clinician might use patients’ scores on a physical functioning test in order to assess their overall health. More often than not, the statistical models that are used to calculate the scores do not describe the reality (i.e., individuals’ response patterns) very well. This, in turn, might affect the accuracy of the calculated scores. The overarching question that I strived to answer in this thesis was: Are the decisions that are made based on these scores (e.g., classifying a child as suffering from attention deficit, admitting certain applicants into a study program, or classifying a person as physically disabled) affected by the fallacy of the used model? And if so, to what extent? This question is important for test constructors and researchers who wish to create and use high-quality psychometric instruments. The findings showed that, in general, decisions were only marginally influenced by the imperfect fit of the models used. The research presented in this thesis have important implications for everyone involved in psychometric testing, from test constructors to test takers