the temperature at the municipality of the student’s school above that of the student’s home for two reasons. First, students take the test at school. Second, we argue that learning mostly takes place at school. Therefore, we presume that especially the temperature at school days are relevant, which is also suggested by previous literature. Park et al. (2020) find that weekend and summer temperature have only limited impact on student performance and Park et al. (2021) draw a similar conclusion. 5.2.5 Other data We combine the student performance and weather data with data from Netherlands Cohort Study on Education (NCO) to obtain information at the student level. We also use data on addresses of school buildings from Dienst Uitvoering Onderwijs (DUO) combined with information on the construction year of school buildings from Basisregistratie Adressen en Gebouwen (BAG). 5.2.6 Descriptives Our final dataset contains information on 777,063 students of 2,929 Dutch primary schools. The data have 3,923,943 observations of 674,211 unique students for reading, whereas this is 5,228,360 observations of 775,666 unique students for math.4 Consequently, we have around five and seven observations per student in the estimation sample for reading and math, respectively. 4The number of observations in our analyses deviate from the total number of observations solely due to dropped singleton observations. This is the case for all analyses in the remainder of the chapter. 119
RkJQdWJsaXNoZXIy MjY0ODMw