Classic Test Theory (CTT) offers important statistical tools for the assessment of tests (Barbaranelli & Natali, 2005), the analytical studies presented in this paper are mainly based on the more modern Item Response Theory. This latter solution makes use of various mathematical models to measure latent variables and allows us to overcome the principal limitations of CTT, such as the dependence between estimated student ability and item difficulty.
In this context, we will consider the simplest IRT model: the Rasch model (Rasch, 1960) that is a one-parameter logistic model, and thus the simplest of the IRT models. It allows us to calculate the probability of correct response to a determined item, according to the ability of the student and the psychometric characteristics of the item itself (particularly, the item’s difficulty). From a strictly statistical point of view, it could be expected then that a higher level of student ability correlates with a higher percentage of correct answers for an item and, simultaneously, a lower percentage of wrong answers. The percentage of wrong answers given always decrease with the students’ ability but for some items it is possible to see answers’ trends which are not strictly decreasing. we call this phenomenon “humped performance”. Analysis of this phenomenon is complex as various interactive factors come into play: students with varying levels of ability may encounter different obstacles when faced with a task, supply wrong answers for different reasons, and favour one wrong answer over another as a result of different approaches and problems.
In this research, results will be analysed from various school levels (from primary to high school) which display good measurement properties and in which at least one option of response demonstrates a “humped performance” that may be linked to teaching factors. In particular, in the following examples, one of the main constructs that can supply a key to reading statistical results of this type at a systemic level is the didactic contract (Brousseau, 1988; EMS-EC, 2012).