Chapter 5. Heat and Learning in a Moderate Climate: Temperature Effects on Primary School Students in the Netherlands dinschool s is regressed on the daily temperature (TEMPds).We control with a set of weather conditions (W′ds) to ensure that our results for temperature are not confounded by the influence of other weatherrelated factors. More specifically, we consider wind speed, precipitation, and relative humidity. We include a student fixed effect (αi) controlling for unobserved individual heterogeneity. Furthermore, we include a year-fixed effect (γy(d)) to capture all time-variant year-specific factors and a grade-test period-school-fixed effect (σps) to capture all time-invariant factors specific for a grade-test period within a school. As described in Section 5.2.3, a grade-test period refers to the combination of a grade and the moment at which the test has been taken (mid-term or end-of-term test) within that grade. Therefore, the gradetest period-school-fixed effect controls, for instance, for the fact that a school decides to cover certain course materials in reading and math at the beginning of second grade, or that schools have certain policies regarding reading frequency within a grade. Last, we add a calendar month-fixed effect (ωy(m)), which controls for month-specific factors. On the one hand, the later the test is conducted, the more learning time students have to develop their skills. It follows that students presumably score higher on these later tests. On the other hand, later tests are normally taken under higher temperatures, as they are administered more closely to the summer. These reasons justify the inclusion of the calendar month-fixed effect in our model. εipds is the error term. We cluster standard errors at the municipality level.6. 6We cluster standard errors at the level of variation in our treatment variable. It is likely that error terms may exhibit correlation within the same municipality, both across students and over time. There are also reasons to suspect that errors are correlated across municipalities. However, including year-fixed effects will absorb away the common shock and there will be no within-cluster correlation among the remaining errors (Cameron & Miller, 2015) 124
RkJQdWJsaXNoZXIy MjY0ODMw