pact of measurement error on our results. In addition, we correct for differences in test timing by including year fixed effects in our analyses. We enrich the LVS data with data from the Netherlands Cohort Study on Education (NCO). These data incorporate register data on students’ backgrounds from Statistics Netherlands and information at the school level. The NCO dataset distinguishes itself from other datasets in multiple ways. The data combine information on all Dutch students over multiple years, and enable us to follow a large group of students throughout their whole primary school period. Furthermore, because of their administrative nature, the data do not suffer from low response rates, resulting in problems with data quality, non-representative samples, recall bias, etc. 3.3.2 Descriptives The estimation sample consists of data of 31,830 students.7,8 Table 3.1 provides the descriptive statistics of the skill scores of these students. The non-standardized skill scores increase across grades for all skills. On average, students obtain a reading skill score of 121 at the end of grade one and 207 in grade six. For spelling, the average skill scores are 166 and 375 in mid grades one and six, respectively. For math, 7Table B.2 shows that our estimation sample is roughly representative for the Dutch student population. 8We opt to use a rather fixed sample of students across grades. For this reason, we only keep the students of which we have at least one skill score available in every grade. Some students deviate from their normal time of six years in primary education if they repeat or skip a grade. We drop students that skipped a grade during primary education. We add a dummy for students that repeat a grade to our analysis and drop the second observation in that specific grade. 51
RkJQdWJsaXNoZXIy MjY0ODMw