Many economic factors (such as international trade, immigration, or differential fertility) and non-economic factors (such as climate change, conflict, or persecution) have led to increasing cultural, ethnic, and socioeconomic diversity in many countries (Duflo 2007, Pekkala Kerr and Kerr 2019, Dustmann et al. 2023). A common political and policy concern related to these changes revolves around the interactions between the minority and the majority groups; and especially between the minority and the majority children in schools. As a result, a burgeoning literature on minority peer effects has emerged.

However, these studies face many methodological challenges related to causal identification (Manski 1993, Angrist 2014) and, to date, their findings are largely inconclusive. Some studies find negative effects of ‘exposure’ to racial minorities (e.g. Hoxby 2000, Hoxby and Weingarth 2005, Hodler et al. 2022) or immigrants (e.g. Schneeweis and Winter-Ebmer 2007, Gould et al. 2009) while other studies find no such relationship (e.g. Angrist and Lang 2004, Antecol et al. 2016). There is even work suggesting positive effects of exposure to immigrants for native students (Figlio, et al. 2024).

What is different across these papers? First, only Hoxby and Weingarth (2005) and Antecol et al. (2016) rely on variation that is (close to) a gold standard of a randomised control trial. Second, many of the papers consider the effects of heterogeneous groups of students, while no prior work jointly considers the responses of students, parents, and teachers – all of whom contribute to the education production function.

In a recent paper (de Gendre et al. 2024), we overcome many of the limitations existing in prior work and ask whether exposure to minority peers in the classroom affects student achievement in Taiwanese middle schools. We probe two plausible channels through which the peer effect could operate: (1) characteristics correlated with being a minority and (2) changes in the behaviour of students, parents, and teachers that contribute to better educational outcomes.

Taiwan is a useful case study because of the institutionally mandated random allocation of children to classrooms which naturally replicates a randomised controlled trial that assigns peers to students, and is therefore ideal for causal identification. We also have access to extensive survey data that includes information not only about students but also about their parents and teachers. The minority group we consider are the Indigenous Peoples of Taiwan (about 4% of the population) who are the native inhabitants of the island of Taiwan; we define the majority as the Han Chinese (about 95% of the population). Similar to minority groups studied in other contexts, the Indigenous Peoples of Taiwan have been historically discriminated against and are still heavily disadvantaged across virtually any dimension. In our data, we further document that Indigenous students are perceived as lower ability by their teachers, even after accounting for a high-quality measure of their actual ability.

To discipline our empirical analysis and reconcile our findings with the existing literature, we first develop a multi-agent model of the educational production function, where the investments of students, parents, and teachers (which we observe in our data) are linked to the presence of minority students in the classroom (who are randomly allocated to classrooms in our setting). We make two assumptions in the model: (i) that inputs of the three aforementioned agents are q-complements (i.e. more of one input makes each unit of other inputs more productive); and (ii) that endowments of Indigenous students (actual or perceived) are lower than those of majority students. More Indigenous students lower classroom endowments and, as a result, decrease inputs from students, parents, and children, ultimately decreasing academic achievement.

Bringing the theory to the data, we find that exposure to Indigenous students lowers the test scores of majority students in Taiwanese classrooms. The negative effect is modest in magnitude, with one additional Indigenous child (in a class of 35) lowering the achievement of the remaining students by 1.2% of a standard deviation. We do not find differential effects on Indigenous versus non-Indigenous students. This same exposure also lowers students’ effort (by 2.2% of a standard deviation), parents’ investments (by 2.1% of a standard deviation), and teachers’ effort (by 7.2% of a standard deviation). Both findings are consistent with our proposed theory.

We then ask to what extent characteristics correlated with Indigeneity vis-à-vis the aforementioned endogenous responses could explain the negative peer effect on achievement. Figure 1 shows how much of the negative effect of Indigenous peers on test scores is explained by other peer characteristics correlated with having more Indigenous peers. Indigenous students are very disadvantaged compared to Han Chinese students, which means that having more Indigenous peers also means having more disadvantaged peers along many dimensions. The dimensions we consider are:

- test scores (e.g. Indigenous students have lower test scores by 72% of a standard deviation)
- socioeconomic status (e.g. parents of Indigenous students are 20 percentage points more likely to be in the lowest income bracket)
- language skills (Indigenous students have 4 percentage points lower fluency in Mandarin, the language of instruction)
- parental investments (Indigenous students are 27 percentage points less likely to have had private tutoring before junior high school)
- disruptiveness (Indigenous students are 12 percentage points more likely to display truant behaviour)

Our results show that only test scores and socioeconomic status explain part of the negative peer effect, and that all factors together explain less than one-third of the negative coefficient. We therefore conclude that the negative effect of having more Indigenous peers is not solely explained by the accompanying decrease in peer quality.

**Figure 1 **Indigenous peer effects on test scores explained by other peer characteristics

As mentioned above, in our study we find that more Indigenous peers lower students’ effort, parents’ investments, and teachers’ effort. In Figure 2, we explore how much of the negative effect of Indigenous students on test score is explained by these responses. We find that, together, changes in the behaviours of the three agents account for 39% of the negative peer effect. This makes sense, given that student effort (Anaya and Zamarro 2024), parental investments (Hsin 2009), and teacher inputs (Papageorge et al. 2020) have all been shown to influence student achievement.

**Figure 2** Indigenous peer effect on test scores mediated by changes in inputs from other students, parents, and teachers

In our final exercise, presented in Figure 3, we jointly account for correlated peer characteristics and endogenous responses in explaining the negative Indigenous peer effect. Although neither channel separately can fully explain the negative peer effect (both red bars are statistically significant at the 10% level), when we consider them jointly, the point estimate shrinks from 1.2% of a standard deviation to 0.03% of a standard deviation and it is no longer statistically significant at conventional levels. This suggests that endogenous responses are at least as good an explanation for the negative Indigenous peer effect on test scores as the differences in peer quality, and that both these explanations account for virtually the entire negative Indigenous peer effect.

**Figure 3 **Indigenous peer effects on test scores

We conclude that exposure to minority children in a classroom: (1) leads to modest declines in test scores of the remaining students; (2) part of the effect is due to lower endowments of minority students; but that (3) another important part is due to negative effort responses of students, parents, and teachers; and finally that (4) together, these two explanations nearly account for the entire negative Indigenous peer effect. Thus, from a policy perspective (Domnisoru et al. 2018, Fershtman and Pavan 2020, List et al. 2023), it appears critical to understand what and who is driving the reduced-form peer effects that are commonly estimated in the extant literature.

## References

Anaya, L, and G Zamarro (2024), “The role of student effort on performance in PISA: Revisiting the gender gap in achievement”, *Oxford Economic Papers* 76(2): 533–60.

Angrist, J, and K Lang (2004), “Does school integration generate peer effects? Evidence from Boston’s Metco programme”, *American Economic Review* 94(5): 1613–34.

Angrist, J (2014), “The perils of peer effects”, *Labour Economics* 30: 98–108.

Antecol, H, O Eren, and S Ozbeklik (2016), “Peer effects in disadvantaged primary schools: Evidence from a randomized experiment”,* Journal of Human Resources *51(1): 95–132.

de Gendre, A, K Karbownik, N Salamanca, and Y Zenou (2024), “Integrating minorities in the classroom: The role of students, parents, and teachers”, CEPR Discussion Paper 19041.

Domnisoru, C, L Taylor, and D Card (2018), “Invest in public education to increase intergenerational mobility”, VoxEU.org, 6 October.

Duflo, E (2007), “Successful assimilation of immigrants”, VoxEU.org, 4 December.

Dustmann, C, Y Kastis, and I Preston (2023), “Inequality and immigration”, VoxEU.org, 1 July.

Fershtman, D, and A Pavan (2020), “Soft affirmative action and minority recruitment”, CEPR Discussion Paper DP15075.

Figlio, D, P Giuliano, R Marchingiglio, U Özek, and P Sapienza (2024), “Diversity in schools: Immigrants and the educational performance of US born students”, *Review of Economic Studies*, forthcoming.

Gould, E, V Lavy, and D Paserman (2009), “Does immigration affect the long-term educational outcomes of natives? Quasi-experimental evidence”, *Economic Journal* 119(540): 1243–69.

Hodler, R, B Eugster, S Balestra, C Chuard, and A Aerne (2022), “How (not) to school ethnic minority students”, VoxEU.org, 2 November.

Hoxby, C (2000), “Peer effects in the classroom: Learning from gender and race variation”, NBER Working Paper 7867.

Hoxby, C, and G Weingarth (2005), “Taking race out of the equation: School reassignment and the structure of peer effects”, unpublished.

Hsin, A (2009), “Parent’s time with children: Does time matter for children’s cognitive achievement?”, *Social Indicators Research *93(1): 123–16.

List, J, F Momeni, M Vlassopoulos, and Y Zenou (2023), “The influence of neighbourhoods on early childhood development”, VoxEU.org, 14 June.

Manski, C (1993), “Identification of endogenous social effects: The reflection problem”, *Review of Economic Studies* 60(3): 531–42.

Papageorge, N, S Gershenson, and K Kang (2020), “Teacher expectations matter”, *Review of Economics and Statistics *102(2): 234–51.

Pekkala Kerr, S, and W Kerr (2019), “Immigrant networking”, VoxEU.org, 1 March.

Schneeweis, N, and R Winter-Ebmer (2007), “Peer effects in Austrian schools”, *Empirical Economics* 32: 387–409.