A quantitative model for gender gap in G8 standardized Mathematics tests in Italian schools

The gender gap in Mathematics, i.e. the different performances of male and female students, is a well-known and well-documented phenomenon. Testing from OCSE-PISA, in particular, highlights how the gap in Italy is much larger than the international average. The didactic component of this gap has been investigated in the literature through one of two broad strategies: either large-scale, statistical analysis of test results, or item-level analysis of very few selected items with the theory of the didactic contract. Object and research hypothesis. Object of this work is to identify which kinds of items, or which properties of items, lead to a gender gap. In particular, our hypothesis is that it is possible to predict an item’s discrimination by classifying it according to appropriately-defined categories. Data used. We use data from INVALSI standardized testing, grade 8, years 2009/2017, for a total of 8 tests each with approx. 500,000 samples, and more than 340 different items. We select this grade because the gender gap increases with the grade, and this is the latest school year where all students follow the same curriculum - with high school choice highly correlated to gender. Methods. We group the eight tests in two sets of four: the model construction set and the model validation set. All items in both sets are scored with a discrimination metric, based on the Differential Item Functioning. We use highly-discriminating items from the model construction set to identify 16 categories, such as “Explain your reasoning”, “Multiple-choice item”, “Redundant information” or “Asymmetric distractors”. We then classify all items from both sets according to these categories. Finally, we compute the discrimination scores of the categories using a least-squares method on the model construction set, then test them on the control set.