Understanding why immigrant children underperform: evidence from Italian compulsory education

ABSTRACT

We aim to investigate the extent to which children of immigrants achieve lower levels of academic proficiency in reading and mathematics compared to native students in compulsory education in Italy. Advancing the current literature, we investigate in a more comprehensive way the importance of a variety of individual characteristics in accounting for children of immigrants’ penalties. In particular, we examine the role of family structure, parents’ socio-economic resources, parents’ cultural and educational resources and students’ school-related attitudes and behaviour. The empirical analysis makes use of a unique dataset collected by the National Institute for the Evaluation of the Italian School System (INVALSI) on the whole population of students enrolled in primary school (5th grade) and lower secondary education (6th grade) in 2012. We apply Kitagawa-Blinder–Oaxaca decomposition models to identify the net contribution of each characteristic to the disadvantages faced by immigrant offspring. We found partial support for the composition hypothesis (socio-economic resources) and culturalist explanations (especially language spoken at home), but pupils’ school-related attitudes – which received less attention in the previous literature – also contribute to explaining the gaps, especially in lower secondary school.

KEYWORDS:

• Migrants
• education
• academic competencies
• inequalities
• test scores

Acknowledgements

This paper was presented at the 49th Scientific meeting of the Italian Statistical Society at the University of Palermo (June 20–22, 2018). We would like to thank participants for useful insights on previous versions of this manuscript. As usual, all errors remain ours.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The datasets analysed during the current study are available on request in the INVALSI repository, https://www.invalsi.it/invalsi/index.php.

Notes

1 OECD/EU (2018) – StatLink https://doi.org/10.1787/888933844161 (10.11.19).

2 ISTAT Immigrants.Stat data, available at: http://stra-dati.istat.it/Index.aspx?lang=en&SubSessionId=dfb99504-ff40-453a-a3c5-730683d7648f (10.11.2019).

3 MIUR (2018) provides detailed information on the distribution of nationalities by educational level, which, however, largely overlap with that computed on the whole population. The main difference to the latter is a greater incidence of Indians (5th nationality at primary and pre-primary school, corresponding to 3.5% of all non-Italian students at each level) and students from the Philippines (5th nationality at upper and lower secondary school; respectively 4.1% and 3.8% of enrolled allochthones).

4 We have also conducted our analysis also on a subsample of schools in which an external examiner was present at the moment of the test and which are unaffected by the most visible and common sources of cheating. The results are analogous to the ones presented in this article.

5 The Rasch model is the standard Item Response Theory (IRT) model to analyse the results from standardised tests, which takes into account the respondent's abilities and each item difficulty on the basis of the pattern of respondents’ answers. For a broad overview on this approach, see Lord (2012).

6 When the information was not available, we instead relied on information about the citizenship in order to exclude the smallest number of cases. This affects only 3% of the cases and does not lead to changes in the substantive results.

7 Since results of the school with an external observer are used to correct the test scores for potential cheating in the whole population of schools, it is important to include this variable among the control variables.

8 As cultural assets, we consider the number of books at home and the language spoken at home for the majority of the time. In the data at hand, the language spoken with parents and the one spoken with siblings are not distinguishable. However, if a student uses the destination country language with her siblings and origin country language with her parents, the variable will take the value 0 only if the student at home interacts for the majority of the time with siblings and only less frequently with parents. In this case, the variable will not distinguish her from her Italian peers, in line with the idea that no substantial cultural differences regarding this aspect are to be recorded.

9 Unfortunately, our variable for ‘non-intact families’ cannot distinguish between divorced/separated families and transnational families. This is an important distinction according to previous studies (Smeekens, Stroebe, and Abakoumkin 2012), which could help to better understand the possibly heterogeneous consequences of different family situations for children’s outcomes.

10 It is important to note that educational resources at home are only weakly correlated to socioeconomic background variables in our data.

11 On the basis of the findings from the first regression models, students from mixed couples were excluded from the decomposition analysis.

12 The second component captures all potential effects of differences in unobserved variables, but also an unobserved component that makes the returns to individual characteristics in terms of achievement vary across students’ categories (e.g., talent).

13 One should note, however, that we can only attach a descriptive interpretation to results of the Blinder-Oaxaca decomposition, since it does not take into account the potential endogeneity of specific individual characteristics included in the model and their potential reciprocal relationships (Huber 2015).

14 These numbers are obtained after averaging the estimates of the percentage of the achievement gap accounted for by each characteristic reported in Figure 3. For each characteristic, the average, minimum and maximum are computed by considering the six values estimated across school levels, subjects and migration background.

15 We talk about the net contribution because it could be that social background affects achievement also by influencing pupils’ attitudes and behaviour, making its total contribution to the gap larger.

16 Results reported in Table A4 in the Online Appendix.