The Effects of School Start Time on Educational Outcomes: Evidence from the 9 O’clock Attendance Policy in South Korea

In the Appendix, I first show that the distribution of middle school attendance time before and after the implementation of the 9 O’clock Attendance Policy in Gyeonggi and Seoul in Table A1. I then report results of several robustness tests. First, I estimate the model that additionally includes standardized test results that the students achieved in the 9th grade. All 9th and 11th graders took the NAEA exam from 2008 to 2016, and 6th graders took the exam from 2008 to 2012. Therefore, 11th graders also took the test when they were 9th graders two years ago. Because the composition of students can vary by year and region, it is possible that higher test scores of 11th graders in Gyeonggi after the policy may just reflect initially different academic abilities of students. Since achievement measures of 9th graders in the NAEA by region-year are available from 2008 to 2016, they are included as regressors to control the initial academic ability of each region-year cohort. For example, the ratios in 2008 are matched to 11th graders in the corresponding province in 2010. More specifically, the ratio of 9th graders with a normal academic background or higher, the ratio of 9th graders with a basic academic background, and the ratio of 9th graders with less than a basic academic background in math, Korean, and English tests for each year-and-province are the information available during the period. When a student understands a large part of the basic content expected to be achieved by students in the grade, the student is evaluated as having a normal academic background. When a student has a partial understanding of the basic content that students in the grade expect to achieve, the student is evaluated as having a basic academic background.

Table A2 presents the estimation results for the model, including the academic performance in the 9th-grade NAEA by province-year. The results show that the estimates are robust to the inclusion of the measures of academic achievement in the 9th grade. Rather, the estimates become greater, and the confidence intervals become tighter mostly when the measures of academic achievement in the 9th grade are controlled.

Conversely, as a falsification test, I test whether 11th graders in Gyeonggi province after the 9 o’clock attendance policy had better achievement in the NAEA when they were in the 9th or 6th grade. In the first analysis, the dependent variable is the log of the proportion of students with a normal academic background or higher in each province-year. Provincial fixed effects and year fixed effects are only controlled in column (1), and regional-specific time trends are additionally controlled in column (2). Panel (a) of Table A3 shows the estimation results. All estimates are small and not statistically significant. In the second analysis, the dependent variable is the average score in each province-year. The average score for each subject in the 9th grade is available only in 2009, 2010, 2013, and 2014 NAEA. Panel (b) of Table A3 presents the estimation results. All estimates are negative, which means that 11th graders in Gyeonggi after the 9 o’clock attendance policy had lower average scores when they were 9th graders, and statistically insignificant. Finally, I investigate whether the 9 o’clock attendance policy has any impact on test results when students were 6th graders. Test results for 6th graders are available for the years 2008–2012. However, the year in which 6th graders who took the test in 2012 turn into 11th graders is 2017. The 2017 results are excluded from the sample because it is not in the sample period. The estimation results in Panel (c) of Table A3 show that the estimates are small and statistically insignificant. The results in Table A3 show that the academic performance of the students affected by the policy was not particularly different in earlier grades.

As another robustness analysis, I construct alternative control groups made up of provinces more similar to Gyeonggi and estimate the policy impact for the alternative control groups. The first alternative control group consists of provinces with the progressive superintendent. In consideration of the possibility that similar types of educational policies would be implemented in provinces with similar types of superintendents of education, provinces with liberal superintendents such as the Gyeonggi superintendent consists of the first alternative control group. The second alternative control group consists of metropolitan cities. Gyeonggi Province is a region in the capital area, including Seoul and Incheon, and is composed of several large cities. The second alternative control group is constructed, excluding provinces where include many rural areas. The third alternative control group consists of provinces except Seoul. Considering that Seoul may have been partially affected by the policy of delaying school start times, the alternative control group is formed except Seoul. Table A4 presents the estimation results, and they show that the results are robust when using the alternative control groups. In particular, the positive effect of the 9 o’clock attendance policy on the math score is common in all results.