The role of peers in promoting energy conservation among Chinese university students

China’s carbon neutrality objective requires greater promotion of consumer-side pro-environmental behavior, yet most current measures focus on production. University students, as future leaders in the energy transition and highly susceptible to peer influence, represent a key population. The study asks whether a student’s energy-saving behavior is positively influenced when peers in the same major exhibit active energy-conservation behaviors. Leveraging the strong social context of Chinese universities—where students of the same major attend classes and activities together—the paper investigates the presence, magnitude, and mechanisms of peer effects on energy-saving behavior among university students, along with implications for policy design.

The paper reviews evidence on peer effects arising from social norms across diverse domains (education, environmental compliance, well-being, prosociality, mental health, misconduct, consumption, waste separation, DER adoption). Manski (1993) distinguishes endogenous, exogenous (contextual), and correlated effects and highlights identification challenges. Random assignment of peers (e.g., roommates) has been used but may capture only part of social influence, and findings on academic outcomes are sometimes fragile. Energy-conservation behavior research identifies psychological, informational, and socioeconomic determinants, and intervention strategies (games, messaging, stickers, information nudges) show varying effectiveness. Peer influence has been studied more in adoption of solar PV and vehicles, where proximity and visibility intensify effects. Few studies focus on peer effects within the broader student-major context on routine energy-saving behaviors, leaving a gap this paper addresses by examining peers within the same major and hypothesizing positive contagion of energy-saving behavior.

Data come from the University Students' Cognition of Energy and Environment Survey (CSCES), administered face-to-face at Xiamen University (Dec 2020–Apr 2021) with computer-aided support. The questionnaire has 240+ items; content validity steps and a 100-sample pre-survey were conducted. A total of 2993 effective samples were collected. Peer groups are defined by same education stage, grade, school, and major; students cannot choose these peers, approximating random assignment. The minimum reference group size is set to 5 (following Alvarez-Cuadrado et al., 2016). After applying group-size criteria, 1617 valid observations remain; reference group sizes range 5–33 (mean 12.58; median 10). Sampling was stratified by school with simple random sampling within schools. The dependent variable (sav) is a summed score (range 5–25) across five Likert items reflecting routine energy-saving behaviors (habit of buying energy-saving products; adjusting AC/heater temperatures to save energy; turning off AC/lights when unused; minimizing refrigerator opening time). Higher scores indicate more active energy saving. Reliability is good (Cronbach’s alpha 0.78; construct reliability 0.78). The core explanatory variable (peer) is the mean of peers’ energy-saving behavior (excluding individual i). Controls include individual characteristics (gender, birth year, religion, marital status, political affiliation, leadership role, school, education stage, grade, major) and family characteristics (province, hukou, living type, house ownership, local income level, number of properties, car ownership, parents’ education and political status). Variance inflation factors fall in [1.03, 6.77]. The classical linear-in-means model (Manski, 1993) is estimated via OLS with standard errors clustered at the major level, including fixed effects for education stage, school, grade, and major. Model diagnostics include White test (cannot reject homoskedasticity at 5%) and Ramsey RESET (p=0.7019). Robustness checks: instrumental variables (2SLS and limited-information maximum likelihood) using peers’ families’ low-carbon expectations and behaviors as instruments (strong first-stage; Kleibergen-Paap statistics indicate relevance; over-identification test p=0.7337; DWH tests suggest limited endogeneity); placebo tests assigning pseudo peers within schools to address correlated effects; varying minimum peer-group sizes (2–14); controlling exogenous peer characteristics (peer means of responsibility, eco-attitudes, intention, parents’ education); restricting to first-year students; alternative estimators (ordered probit/logit) for the ordered dependent variable. Mechanism analysis regresses mediators (energy-saving knowledge, willingness, perceived importance, social responsibility) on peer behavior with the same controls.

- Baseline peer effect: With full controls and fixed effects, peer coefficient is 0.271 (SE 0.080; p=0.001; 95% CI [0.112, 0.430]). A one–standard deviation increase in peers’ behavior (SD peer=1.367) is associated with a 0.114 SD increase in individual behavior (SD sav=3.250). Without controls, the analogous standardized effect is 0.132 SD (coef 0.313). - Heterogeneity: • Gender: Significant for both males (coef 0.317, p=0.030) and females (0.236, p=0.008); no significant gender difference (interaction term ns; Bootstrap p=0.241; Fisher combined p=0.237). • Grade: Significant for high grades (≥3) (0.353, p=0.004), not significant for low grades (1–2) (0.205, p=0.137), suggesting peer effects strengthen over time. • Depression: Interaction of high depression with peer behavior is negative (inter_1 = -0.298, p=0.002), indicating weaker peer influence among recently depressed students. • Environmental orientation/concern: Significant for those with positive environmental values (0.264, p=0.001) and for those attentive to energy issues (0.239, p=0.003); not significant for general values (0.166, p=0.514) or those never attentive (0.212, p=0.111). • Peer composition differences: Greater dispersion in peers’ energy-saving behavior reduces the effect (inter_2 = -0.021, p=0.014). - IV estimates: 2SLS/LIML peer coefficients ~0.400–0.402 (SE ~0.099–0.105; p<0.001), indicating robust positive peer effects; tests indicate strong instruments and no evidence against exogeneity. - Robustness: Results persist when adding peer and peer-family characteristics; placebo tests show baseline coefficient exceeds 95% quantiles of simulated pseudo-peer effects; ordered probit/logit show significance at 1%; restricting to first-year students yields nonsignificant effect, consistent with limited exposure time. Varying minimum group size shows significance up to size 12–13, with loss of significance at larger minima likely due to reduced sample size and power. - Mechanism analysis: Peer behavior positively predicts mediators: energy-saving knowledge (0.062, p=0.047), willingness (0.064, p=0.002), perceived importance (0.045, p=0.010), and social responsibility (0.190, p=0.012), supporting both direct (knowledge diffusion) and indirect (values/attitudes) pathways.

Findings confirm that within-major peer groups exert a significant positive influence on individual students’ routine energy-saving behaviors, addressing the core research question and supporting the hypothesis of social contagion in pro-environmental behavior. The effect is robust across specifications and identification strategies, suggesting it is not driven by contextual or correlated effects. Heterogeneity results imply that peer influence strengthens with longer peer exposure (higher grades), is attenuated by recent depression, and is amplified among students with pro-environmental values and greater attention to energy issues; compositionally cohesive peer groups enhance contagion. Mechanism analyses indicate both direct knowledge transfer and indirect shifts in social responsibility, perceived importance, and willingness underpin the peer effect. These insights are relevant for designing interventions that leverage social networks to magnify policy impacts on energy conservation among youth, with potential multiplier effects through interpersonal interactions.

The study demonstrates significant peer effects on the energy-saving behavior of Chinese university students in the same major. A one–standard deviation increase in peers’ energy-saving behavior is associated with a 0.114–0.132 SD increase in individual behavior, depending on controls. Peer influence operates through both direct knowledge diffusion and indirect changes in social responsibility, perceived importance, and willingness. Policy implications include leveraging peer networks to amplify energy-conservation campaigns, targeting groups most susceptible to peer influence (e.g., higher-grade students, those with stronger environmental orientation, and contexts with cohesive peer compositions), and enhancing educational initiatives to build knowledge and values that facilitate contagion. Future research should test generalizability across institutions and regions, and extend analyses to other pro-environmental behaviors such as green consumption.

The dataset is from a single university, limiting external validity and raising questions about generalizability to other institutions or regions. The focus is on energy-saving behaviors; peer effects on other environmentally friendly behaviors (e.g., green consumption) were not studied. Although identification strategies and robustness checks mitigate endogeneity concerns, unobserved factors may remain. Sample reductions when increasing minimum peer group size reduce power, affecting significance in some robustness subsets.