[学习分享] 【独家发布】A Diagnostic Procedure to Detect Departures From Local Independence in Ite [推广有奖]

Abstract Item response theory (IRT) is a widely used measurement model. When considering its use in education, health outcomes, and psychology, it is likely to be one of the most impactful psychometric models in existence. IRT has many advantages over classical test theory-based measurement models. For these advantages to hold in practice, strong assumptions must be satisfied. One of these assumptions, local independence, is the focus of the work described here. Local independence is the assumption that, conditional on the latent variable(s), item responses are unrelated to one another (i.e., independent). Stated another way, local independence implies that the only thing causing items to covary is the modeled latent variable(s). Violations of this assumption, quite aptly titled local dependence, can have serious consequences for the estimated parameters. A new diagnostic is proposed, based on parameter stability in an item-level jackknife resampling procedure. We review the ideas underlying the new diagnostic and how it is computed before covering some simulated and real examples demonstrating its effectiveness. Translational Abstract Almost everyone has taken a test or responded to a survey at some point in their life. In social science research, one of the primary uses of tests and surveys is to collect information about aspects of a person that are not directly observable. Unlike height and weight, which can be viewed directly, things like intelligence, depression, and quality of life cannot (necessarily) been seen just by looking at someone. Many of these constructs (our generic term for these unobservable qualities) are critical to psychology— both in practice and in research. We use a variety of statistical models to relate the information we can see (item responses) to the constructs we hope are driving those responses. One popular model for this kind of work is item response theory (IRT). IRT, like all statistical models, makes some assumptions about the world and the data being analyzed. In this paper, we present a new method that can detect violations of one particular assumption in IRT. We show that this new method, called the Jackknife Slope Index (JSI), works as well or better than the existing ways we have to look for these kinds of violations. We conclude by discussing some of the limitations of our study and highlighting some additional research on the JSI that could further improve its performance.
Keywords: item response theory, local dependence, local independence, diagnostics












如有侵权，请联系删帖，谢谢合作！