A major interest in the most recent years has been devoted to STEM disciplines, for increasing job demand and higher wages associated to this field of studies. In this perspective particular attention has been directed to persistent gender inequalities in the field of STEM studies, which is often the main responsible of the widening gender gap in work career and life opportunities. The data about Italian situation show a sharp unbalanced composition by gender at university: among STEM graduates the male proportion is dominant, reaching 59%, whereas among non-STEM graduates, women prevail: they are almost two out of three.

Previous research has showed that there was an important influence of cultural status of the student's family, captured by educational attainment of the mother, and a crucial influence both of teachers marks and performance at INVALSI test in math, with a slightly higher impact of the former, on the probability to choose a highly scientific course of study at the time of enrolment at university. However, the academic career and eventually the obtaining of the final degree are a much relevant outcome to consider in terms of success into a STEM path for students. Therefore, with this contribution we aim to analyze the university career into STEM, using INVALSI test score as predictors for success.

We used as a base an original dataset built on the combination of different data sources: MIM (Ministry of Education and Merit), INVALSI (National Institute for the Evaluation of the Education and Training System) and University Register of students. Data available at the University level and with a very specific course classification, which can be recode into a binary variable (STEM / non-STEM).

Analyses about the development of career of students into scientific tracks have been carried on by looking at similarities and differences between boys and girls in the speed and the average mark they obtain in scientific courses, associated with their personal and contextual characteristics. Having a wide set of socio-demographic and contextual variables from INVALSI data it has been possible to estimate the impact of the different characteristics on progression through the university career for boys and girls, keeping as a key variable t the "mathematic skills" measured as a continuous variable (WLE score) or as a categorical variable based on the WLE score levels (ordinal scale from 1 to 5) during the INVALSI test of level 13.

Preliminary results give an indication that, although less represented, girls obtain on average higher scores at exams and proceed more rapidly into university STEM career compared to their male colleagues. With this research we intend to further disentangle how contextual factors (family and school as the main agents) and personal factors (both academic and non-academic skills) shape the probability of success in university career for students who have chosen a STEM degree, and particularly for girls.

Students’ solving strategies and their probability of successfully encountering an item are contingent upon several factors (Son & Watanabe, 2017). Among these, the medium, i.e. the way in which an achievement test is administered to students, can play an important role (e.g., Gu, Drake & Wolfe, 2006).

In the present study, we focused on missing answers to Mathematics items. More specifically, by analysing data collected for the Trends in International Mathematics and Studies (TIMSS) at grade 8 in 2019, we compared students' responses to the same mathematics item administered in paper- and computer-based assessment (PBA and CBA, respectively) and compared the number of missing responses for each item.

We focused on open-ended mathematics items, which tend to have the highest number of missing answers. A preliminary investigation showed that (i) the difference between missing answers in the achievement tests administered in PBA and those administered in CBA is usually small and less than 5%, (ii) the items with the largest difference (i.e. equal or greater than 10%) in terms of missing values between PBA and CBA are those belonging to two cognitive domains, namely 'reasoning' and 'applying', and finally (iii) students are more likely to give an answer to a Mathematics item in CBA.

TIMSS includes items designed to assess three cognitive domains: 'knowing', 'applying' and 'reasoning'. The first domain, knowing, covers the facts, concepts and procedures that students need to know. The second, applying, focuses on students' ability to apply knowledge and conceptual understanding to solve problems or answer questions. The third domain, reasoning, goes beyond the solution of routine problems to include unfamiliar situations, complex contexts and multi-step problems. Students' answers to the 'knowing' items are thus clearly linked to teaching practices and to the Brousseau while both 'reasoning' and 'applying' refer, to varying degrees, to each student's ability to use logical and systematic thinking, including intuitive and inductive reasoning. In particular, 'reasoning' items - those where the difference of missing answers between PBA and CBA is greater - include intuitive and inductive reasoning based on patterns and regularities that can be used to find solutions to problems set in novel or unfamiliar situations. Such problems may be purely mathematical or may have real-life settings.

In light of these differences, and as with previous studies in mathematics education, we hypothesise that personal factors, such as students' gender (e.g., Ethington, 1990; Hyde, Fennema & Lamon, 1990; Else-Quest, Hyde, & Linn, 2010; Leder, 2019), may be associated with the number of missing answers and, in particular, that girls are less likely than boys to answer open-ended items that are not directly related to classroom practice and the Brousseau.

The aim of the present study is to test such a hypothesis and to see whether the medium plays a role in such an association.

Quality education and lifelong learning opportunities for all are central to ensuring a full and productive life to all individuals and to the realization of sustainable development. The Targets of Goal 4 Of Agenda 2030 concern different dimensions: access for all to education of all levels (primary, secondary and tertiary), the quality of education, the possession of knowledge and skills for employment and sustainable development; the elimination of gender disparities in education and equal access for the most vulnerable; monitoring of school facilities, so that they are suitable for everyone’s needs.

The 2030 Agenda also aims to link the Targets of a specific goal to those of another goal. In this case, the study of gender differences and differences due to socioeconomic background can be assessed through the indicators found in Goal 4 (Quality of education,) Goal 5 (Gender inequality) and Goal 8 (Decent work and economic growth).

The family background has a great impact on the capacity of achieving a certain educational level. As known, a relation exists between lower education and labor market opportunities. Indeed, low education implies greater difficulty in entering the labor market, in finding good quality and stable jobs, and in realizing the full potential of an individual. Although the family background does not seem to have a direct impact on the employment for young people having the same low educational level, having a great impact on the education poverty, causes also future socio-economic disadvantages. In the long run, the lack of educational opportunities increases the likelihood of being at an economic hardship when adult. Such disadvantage for the future work career can be transmitted from generation to generation, triggering a vicious circle of education poverty.

Women in Italy are more educated than men. Furthermore, gender differences are increasing. However, the female advantage in education does not translate into an employment advantage.

The analysis of the indicators present in Goal 4 combined with the indicators present in the other Goals can provide information about

• the influence of family socio-economic context and gender on educational opportunities

• how much the educational poverty of the original family is transmitted between generations

Introduction

In recent years, there has been growing interest in the quantitative study of the mechanisms underneath inequalities’ reproducation in the school system. From a theoretical standpoint, the concept of "tertiary effects" (Esser 2016) has been conceptualized, referring to the influence of the school system and its actors on inequalities in learning and educational pathways. At the same time, empirical studies focused on various school mechanisms reinforcing the reproduction processes of educational inequalities, both in Europe and in Italy (Argentin and Pavolini 2020). This evolution has been made possible by the creation and dissemination of large administrative datasets in European countries, built for evaluative purposes. Therefore, it is useful to reflect on these issues at the conference promoted by INVALSI.

As often happens, secondary analyses of information coming from standardized assessmets display disavantages too. Indeed, collecting information about the malfunctions of the school system by public evaluation institutions rasises several issues. The purpose of our contribution is to describe an attempt to integrate administrative data with data collected ad hoc by the research group, in order to enable new insights about educational inequalities and the mechanisms generating them.

Data and methods

A pilot online survey was conducted on a random national sample of 100 Italian lower secondary schools, aiming to interview school principals, teachers, and parents of students. The autonomy of this data collection, compared to the institutional one developed for evaluative purposes, provided us with ample freedom in designing new questions, scales and items. We tested them with the different targets involved in the survey.

Less than two-thirds of the schools invited to particpate responded to the survey, leading to selected samples of principals, teachers, and parents. These smaples are not large enough to allow inferences about the related populations, but sufficient to validate the measurement tools we used.

Results

The proposed contribution develops two separate analyses.

Firstly, we investigate the sample self-selection process of schools participating in the survey by relating administrative data on the original sample to administrative data on the subset of respondents. It is thus observed that the self-selection of schools does not appear to be neutral concerning characteristics of the schools/students’ populations.

Secondly, we validate the most innovative scales used in the questionnaires for principals, teachers, and parents. We show that the research group's creative effort has yielded mixed results: some scales have proven to be robust and reproducible in future surveys, while others require deep rethinking.

Beyond the presentation of these pilot survey results, we draw implications for future large-scale surveys on inequalities in the school system.