Conditional item dependence for testlet items

David C Nofer, Fordham University


With respect to analyzing test data, a condition assumed by standard item response theory (IRT) models is that, once proficiency (&thetas;) is accounted for, item responses will be unrelated to one another. In practice, item responses may remain interdependent after &thetas; is held constant. The presence of item-response interdependence, or conditional item dependence (CID), has been shown to bias both the standard errors (SEs) of &thetas; estimates and the estimates of discrimination ("a") parameters for the standard IRT models. Dealing with data for the multiple-choice (MC) sections—Verbal Reasoning (VR), Physical Sciences (PS), and Biological Sciences (BS)—of 2 forms of the Medical College Admission Test (MCAT), the present research was focused on CID for responses to items for a testlet. A testlet is defined as an item cluster based on a common stimulus, such as a reading passage. The SCORIGHT computer program, which models withintestlet CID and adjusts parameter/SE estimates accordingly, was employed. The 3 objectives were to: (a) quantify the within-testlet CID, (b) determine the extents to which certain IRT-based estimates are affected by within-testlet CID, and (c) identify variables that predict the observed levels of within-testlet CID. Within-testlet CID was confirmed. When within-testlet CID was ignored, the SEs of &thetas; estimates and a parameters were underestimated. The degree of within-testlet CID was found to be negatively associated with number of items in a testlet for the VR sections of both forms and the BS section of 1 form, to be positively associated with the order in which the testlet was administered for VR, and, for VR, to vary for testlets involving differing subject matters. SCORIGHT is suggested as one viable option available for evaluating MCAT data. ^

Subject Area

Psychology, Psychometrics

Recommended Citation

Nofer, David C, "Conditional item dependence for testlet items" (2007). ETD Collection for Fordham University. AAI3271271.