Grading practices, gender bias and educational outcomes: evidence from Italy

ABSTRACT

We study if the Italian school system suffers from gender bias when judging students. To this aim, we use a differences-in-differences approach that compares the teachers' assessments and the standardized test scores that the students receive during the school year. We have census data for all Italian fifth and sixth graders in two different subjects, math and language, that include a rich set of additional controls. Our evidence reveals that, since primary school, boys are graded less favourably than girls in both math and language. This result is also confirmed for middle school students (sixth graders), and it holds even when (a) we separate the analysis between the most and least developed Italian regions, (b) we control for possible gender-specific attitude towards cheating and teachers' manipulation and (c) we introduce class and school fixed effects in the models. Comparing the results obtained across different levels of schooling and subjects, we cannot clearly identify the role of specific mechanisms in determining the gender bias. Overall the analysis suggests further study on the role of teachers' characteristics.

KEYWORDS:

• Grading practices
• gender bias
• schooling outcomes
• Italy

JEL CLASSIFICATION:

• I21
• I24
• O15

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 The recent OECD averages tell that 51% of 25–34-year-old women (34% in Italy) hold a tertiary qualification against only 39% of men (22% in Italy) OECD 2020.

2 Data show that students who repeat a year are more likely to engage in high-risk behaviour and/or dropout. On this, see also OECD 2014; Lyche 2010.

3 On the exclusion criteria and missing data, see Section 1.3 in the Online Appendix.

4 It is a written exam, nationally set and it is marked by an external teacher (for a sub-sample of students) or by another teacher of the same school. More on this in Section 3.

5 See Section 2 for more details.

6 An exception is Hanna and Linden 2012.

7 Bonesrønning 2008 does not use the simple teachers' scores in the analysis. The focus is on the teachers' grading practices that are not directly observable, and are estimated using a model with the teachers' grade as dependent variable.

8 In this case, the more ‘masculine’ a subject is thought to be (such as math or physics), the more favoured are the female candidates, while the opposite is true for the more ‘feminine’ subjects Breda and Ly 2015. Cobb-Clark and Moschion (2017) find that girls in low and middle socio-economic status families have an advantage in reading, while boys in high socio-economic status families show an advantage in numeracy.

9 See also Angelo and Reis 2021 with data on several cohorts of Portuguese students, Cobb-Clark and Moschion 2017 with data on Australian third graders and, even if the main focus of this study is on teachers' bias across ethnic minority and white pupils, Burgess and Greaves 2013 use five annual censuses of all state school pupils in England.

10 Carlana 2019 conducted a survey to measure the teachers' gender stereotypes with IAT on a sample of 145 middle schools from 5 provinces in the North of Italy, and finds that stereotypes of literature teachers have no effect on reading performance of boys or of girls, while math teachers stereotypes have a negative influence on girls and no effect on boys.

11 On the presence of manipulation and cheating behaviour during the standardized test procedures in the Italian school system, see Bertoni, Brunello, and Rocco 2013; Paccagnella and Sestito 2014.

12 Dee 2005 also identifies an active teacher effect described as biases in the teacher's ‘… prior expectations of and interactions with students who have different demographic traits’, with mechanisms similar to those described above. See Dee 2005, p. 159.

13 Hvidman and Sievertsen 2021 identify a different reaction between Danish boys and girls to a change in their high-stakes GPA caused by the implementation of a new grading system.

14 See also Resh 2009; Prøitz 2013 for a survey of the empirical studies of teachers grading behaviour and the nurture/nature debate to explain the gender differences in education.

15 Carey et al. 2019 provide a literature review into the relationship between maths anxiety and performance and also report recent evidence on both Italian and UK students.

16 See The Knowledge Promotion reform in 2006 and the revision of the Norwegian educational act, particularly the chapter on assessment and final grading amended in 2009 Prøitz 2013. On this, see also Bonesrønning 2008.

17 More on this in the Online Appendix.

18 For more on this see the Online Appendix and Lucifora and Tonello 2015.

19 First generation are students born abroad of foreign-born parents, while second-generation students are native-born children of foreign-born parents.

20 The social capital indicator is from Cartocci (2007), while the level of criminality is estimated by Fiaschi, Gianmoena, and Parenti (2011).

21 Prior to 2000, they could also teach without a Master degree qualification. More on this in the Online Appendix.

22 The INVALSI test results give grades on a scale from 0 to 100 while the teachers' assessment has a 0-10 range.

23 That is, for both math and language, the data set is a stacked file including the teachers' and standardized tests' scores and the number of observations in our regressions will be twice the number of students.

24 For each level of schooling and subject, we use the same reduced sample of students with no missing values in the full set of controls. See Tables A5.1, A5.2, A6.1 and A6.2.

25 ISTAT (Italian national institute of statistics), 2020.

26 For this sample the inspectors verify all the steps of the procedure within the school, including the check of the tests results.

27 This analysis uses data for the academic year 2012-13 and for different grades.

28 It is worth noting that, in our data, we observe only one class per school in 25% of classes in 5th grade and 22% in 6th grade, where class and school fixed effect would be overlapping.

29 A similar result was found by Bonesrønning 2008, while evidence of cross-disciplinary effects for sixth-grade Italian students has also been found in Meroni and Abbiati 2016.

30 In detail, Panel A only includes students who strongly agree with the statement I am proficient in math/language. In panel B, the results are obtained using the sub-sample of students that strongly disagree with the same statement. For more on this, see Tables 3 and 4.