Abstract
Within the Item Response Theory, although a lot of different methods and techniques have been proposed to assess item fit, this topic rises up relevant questions in relation to which not completely satisfactory answers have been given (e.g., Hattie 1984, 1985; Embretson and Reise 2000, etc.), especially for big data (Gustafson 1980). In this paper, fit control was based on the graphical inspection of Item Characteristic Curves (ICCs), estimated by using the Rasch model (one of the most used tool in educational research to estimate students’ ability) that allows observing deviations between observed and expected values for specific ability levels. This can be particularly useful in order to formulate specific hypotheses to understand and pick out at least some possible causes of violations. In order to do this, we analyzed data collected from 2009 to 2015 by the Italian National Institute for the Evaluation of Educational System (INVALSI), by administering Math achievement test to more than 30000 students attending the 2nd grade level of high school, per year. Data analysis allowed picking out, catalogue and classify all deviations between observed and expected values, fostering the emergence of a systematic relationship between item characteristics and violation types, some of which could suggest interpretations based on some of the most popular didactic paradigms.In this paper, we presented a close examination of a particular violation type, i.e. the over-discrimination that allows highlighting some characteristics of items that, by a side, clarify the functionality of the employed statistical model and, by the other side, offer some interesting avenues to explore regarding the nature of the phenomenon.