Social norms and climate-friendly behavior of adolescents

(see Section B.4 in S1 File). In addition, the regression models include the observations of our two HYPO conditions, in which the exemplary donation amounts were explicitly presented as hypothetical and fictitious. Anchoring and demand effects in HYPO should be at least as strong as in the LOW and HIGH conditions, so that the comparison will reveal whether the information describing peer behavior made a behavioral difference. In the following, we first describe the results from the treatments LOW and HIGH, and then the comparison to the HYPO conditions. Fig 2 plots the corresponding coefficients for LOW and HIGH as well as for HYPO LOW and HYPO HIGH. Alternative specifications without fixed effects on the school class level and with socio-demographic controls are reported in Section B.2–B.4 in S1 File. For the empirical expectations (upper left), we replicate the results from the non-parametric analysis: only the treatment LOW has a significant effect on the empirical expectations (p <0.01, CI: [-0.148, -0.066]). While LOW dampens the empirical expectations, HIGH increases them albeit in a non-significant way, so that a significant difference is also observed between the two treatment conditions (LOW vs. HIGH, p = 0.000). For the normative expecta- tions (bottom), in contrast to the non-parametric tests, we find that, for small donations (0–2 €), normative expectations in LOW are significantly different from those in CONTROL (p <0.05, CI: [0.017, 0.406]). This means that in the LOW condition students assumed their fel-

low schoolmates would consider small donations as more socially appropriate than students in CONTROL or HIGH (p = 0.059). Lastly, in the actual donations (upper right), the decreasing pull of LOW is visible, but the confidence intervals are generally wider [-0.189, 0.058] and the effect is not statistically significant. Turning to the HYPO conditions, we find that providing hypothetical donation amounts as a reference shifted the empirical expectations only slightly. Unlike LOW, HYPO LOW does not alter students’ empirical expectations in a statistically significant manner. Thus, it made a significant difference whether the information was hypothetical or described peer behavior Fig 2. Coefficient plots of regression models estimating the treatment effects. This figure presents the tobit regression coefficients and estimated 95% confidence intervals. The coefficient interpretation is that if an individual was part of one of the four treatment conditions: HIGH, LOW, HYPO HIGH or HYPO LOW, then their expectations or donation amount is expected to change by the respective regression coefficient in the censored tobit scale, holding everything else constant. Full model results are reported in Section B in S1 File. Additional controls were gender, age, migration background, students’ connectedness to nature (Inclusion with Nature Scale [58]), existing engagement in climate-friendly behaviors, and participation in Fridays for Future protests. The models also include fixed effects on school class level. https://doi.org/10.1371/journal.pone.0266847.g002 PLOS ONE

Social norms and climate-friendly behavior of adolescents PLOS ONE | https://doi.org/10.1371/journal.pone.0266847 April 27, 2022 9 / 16