loss for reading and one-tenth of the size of the learning loss in math. Looking at the effects of temperature at the lower end of the temperature distribution on students’ performance, we do not find a clear pattern of temperature effects. We refrain from drawing strong conclusions from effects of lower temperatures, as extremely cold temperatures are rather rare in the underlying sample (see Figure D.1). 5.4.2 Heterogeneous impact of temperature We also explore heterogeneity in the causal effect of temperature on student performance on the test day. To avoid the possibility of omitted variable bias using interaction terms as pointed out by Feigenberg et al. (2023), we inspect heterogeneity by sample splits. We focus on student-level heterogeneity regarding gender, paternal income, and student ability. Results of these heterogeneity analyses can be found in Figures D.4, D.5, and D.6, where we report no heterogeneous effects of temperature on students’ reading and math performances. Furthermore, we perform a heterogeneity analysis regarding the construction year of the school building. Since detailed information on how school buildings are insulated and equipped with heating, ventilation, and air conditioning (HVAC) systems, we use construction year as a proxy for school building quality.7 Based on auxiliary data that we could not match directly with our estimation sample, Figure D.3 demonstrates a clear positive correlation between the construction 7HVAC systems can mitigate the effects of temperature on learning (e.g., PorrasSalazar et al., 2018; Sorensen et al., 2024; Wargocki & Wyon, 2007). However, many Dutch schools rely on natural ventilation rather than mechanical systems. A study by Ding et al. (2023) found that one-third of Dutch secondary schools use only natural ventilation. 127
RkJQdWJsaXNoZXIy MjY0ODMw