Can Teachers Reduce Student Social Isolation?

This study asks whether giving teachers detailed maps of their students’ classroom social networks can reduce social isolation and antisocial behavior. Using data from an RCT in 46 Italian primary schools, the researchers analyzed friendship nominations, incentivized games measuring sabotage and cooperation, and surveys on student well-being. The intervention reduced the share of children reporting no classroom friends by 1.5 percentage points (a 50% relative decrease) and cut sabotage behavior by 7.6 percentage points (an 11% reduction). It also increased friendship nominations by about 0.10 standard deviations. Effects were strongest for students who were isolated at baseline.

The Policy Scientist's Perspective

This article addresses a policy issue of broad significance: whether schools can actively reshape peer dynamics to reduce social isolation, a factor linked to long-term educational and economic outcomes. The study’s randomized controlled trial provides high-quality causal evidence, supported by rich network data, well-designed behavioral measures, and careful implementation. Its contributions are timely given rising concerns about student mental health and classroom climate. The results appear credible and potentially generalizable to similar primary-school settings, offering a meaningful addition to the literature.

Full Citation and Link to Article

Alan, S., Carlana, M., & Leone, M. (2025). Inclusive teaching: Spotting social isolation in the classroom. American Economic Journal: Applied Economics, forthcoming. Official article link: https://www.aeaweb.org/articles?id=10.1257/app.20230821

Central Research Question

The study investigates whether providing teachers with detailed information about the social structure of their classrooms—specifically, friendship networks and indicators of social isolation—can reduce social isolation and antisocial behavior among primary-school students. At its core, the research asks whether teachers, when equipped with accurate and actionable data about peer relationships, can modify classroom practices in ways that strengthen social inclusion and diminish harmful behaviors. The authors seek to determine not only whether isolation decreases, but also whether behavioral spillovers emerge, such as reduced sabotage, improved socioemotional well-being, or shifts in prosocial behavior. This question reflects a broader interest in whether low-cost information interventions can alter peer dynamics in early schooling, a setting where teachers exert substantial influence but often have incomplete visibility into the social experiences of individual students.

Previous Literature

The study builds on several strands of literature: research on peer effects, social preferences, the impacts of antisocial behavior, and teacher influence on socioemotional development. Prior work shows that social isolation in childhood is associated with poor mental health, increased inflammation in adulthood, weaker socioemotional adjustment, and disengagement from school. Studies in economics and psychology (e.g., Brown and Taylor 2008; Carrell et al. 2018; Leigh-Hunt et al. 2017) document long-run consequences of bullying, exclusion, and exposure to antisocial peers. Research in social psychology highlights the mechanisms through which exclusion shapes expectations, behavior, and later outcomes. A second relevant literature examines teachers’ roles in shaping norms, classroom climate, and identity formation. Work by Carlana (2019), Algan et al. (2013), and Chetty et al. (2014) shows that teachers influence both academic outcomes and social preferences. Related educational interventions—such as perspective-taking treatments or cooperative-learning structures—have demonstrated that classroom practices can shift relations among peers, though few studies provide direct causal evidence on the effects of teacher-facing information about social networks. The authors’ study contributes to this literature by embedding detailed peer-network data into a randomized controlled trial and testing whether information alone is sufficient to generate behavioral change.

Data

The data originate from a cluster-randomized controlled trial involving 46 Italian public primary schools across Lombardy, Piedmont, and Lazio. The sample includes 2,200+ students and 239 teachers in 135 third-grade classrooms. Baseline data were collected in 2021–2022 when students were in grade 2, followed by the delivery of the intervention in late 2022 and endline data collection in spring 2023. The data consist of: (1) friendship nominations used to map classroom social networks; (2) incentivized behavioral games measuring antisocial behavior (a sabotage game), prosocial behavior (a dictator game and a cooperation game), and economic waste generated by sabotage; (3) socioemotional outcomes assessed through standardized item-response questionnaires on mental health, sense of belonging, and perceptions of teacher support; (4) cognitive ability (Raven’s Progressive Matrices); (5) administrative records from the Ministry of Education and INVALSI, including test scores and demographic variables; and (6) teacher characteristics, including emotional intelligence scores, classroom practices, and experience. The breadth of data—behavioral, administrative, cognitive, and socioemotional—allows the authors to triangulate mechanisms while preserving strong internal validity.

Methods

The study employs a cluster-randomized design, with the school as the unit of randomization. Twenty-two schools received the intervention; the remaining schools serve as controls. Treatment teachers were provided with personalized reports containing detailed classroom social-network maps highlighting students who were socially isolated, statistics on exclusion, and information regarding the links between isolation, mental health, and long-run outcomes. An online meeting reinforced the content of the reports and provided guidance on inclusive classroom strategies. The authors then measure treatment effects using OLS regressions with strata fixed effects, enumerator fixed effects, baseline values of the outcome when available, and an extensive battery of controls, including student demographics, cognitive ability, immigrant status, parental education, teacher characteristics, and class composition. Standard errors are clustered at the school level, consistent with the unit of randomization. Composite indices for social isolation, antisocial behavior, prosocial behavior, and socioemotional well-being are constructed using standardized measures relative to the control group. The methods emphasize causal inference through randomization rather than statistical adjustment, aligning with best practices in experimental design.

Findings/Size Effects

The intervention produced several statistically meaningful effects. First, social isolation declined. The probability that a student reported having no friends, or was not nominated by any classmates, fell by roughly 1.5 percentage points—a 50 percent reduction relative to the control mean. Students in treatment classrooms received more friendship nominations, with an increase of approximately 0.10 standard deviations. The effects are strongest among students who were socially isolated at baseline, suggesting targeted benefits for the most vulnerable peers.

Second, antisocial behavior, measured through the sabotage game, declined substantially. Treatment students were 7.6 percentage points less likely to sabotage an anonymous classmate, an 11 percent decline relative to the baseline probability. The quantity of sabotage, when it did occur, also fell. These reductions in antisocial behavior translated into higher aggregate payoffs in the game and decreased inequality in earnings, demonstrating that behavioral improvements produced measurable economic consequences.

Third, the intervention produced no detectable increase in prosocial behavior in the cooperation or dictator games. This suggests that the mechanism is not a broad shift in altruism but rather a reduction in harmful acts, possibly due to altered expectations about peer interactions or enhanced teacher monitoring.

Fourth, socioemotional well-being improved modestly, though effects were less precisely estimated. Students reported better mental health and stronger belonging, although effect sizes here were smaller than those for social inclusion and antisocial behavior.

Mechanism analyses indicate that treated teachers changed classroom practices. Teachers were more likely to mix students of different ability levels during group work, a form of structured interaction that reduces segregated patterns linked to isolation. Given that academic ability correlates with isolation, socioeconomic status, and immigrant background, this practice likely promoted cross-group integration. Survey data also show that treated teachers expressed more inclusive attitudes and were more attentive to subtle signs of exclusion. These shifts are consistent with the hypothesis that information altered teacher beliefs about the prevalence and consequences of isolation, thereby influencing behavior.

Conclusion

The study demonstrates that a low-cost informational intervention delivered to teachers—consisting of social-network maps and evidence about the risks of social exclusion—can meaningfully reduce social isolation and antisocial behavior in primary-school classrooms. The causal evidence is strong due to the randomized design, and the rich multidimensional data reinforce confidence in the findings. Although the intervention blends two components (network feedback and general information), making it impossible to isolate their separate effects, the results show that even simple, scalable tools can shift classroom dynamics. The study also highlights limitations: it cannot identify the unique contribution of each component; teacher practices may differ in education systems with less classroom continuity; and administrative behavioral data were not available. Nonetheless, the findings have potentially broad applicability across systems where teachers play central roles in shaping peer interactions. The work expands the literature on the social dimensions of schooling and shows that teacher-facing data can serve as a powerful behavioral lever.