Olivier De Groote and Koen Declercq, "Tracking and Specialization of High Schools: Heterogeneous Effects of School Choice", Journal of Applied Econometrics, Vol. 36, No. 7, 2021, pp. 898-916. Program files are provided in the zip file school_choice.zip. Some of these are discussed at the bottom of this file. The empirical analysis in this article combines datasets from different sources. 1. Data on secondary education The main data used in the article are administrative data about all students enrolled in secondary education in Flanders. This is a confidential data source administered by the department Agodi from the Flemish Ministry of Education and Training. Permission to use these data should be requested from: Agentschap voor onderwijsdiensten Koning Albert II-laan 15 B-1210 Brussel Belgium agodi@vlaanderen.be This dataset contains 72,292 and 72,581 students who started for the first time in the first year of secondary education in 2003 and 2004, respectively. We observe students in administrative records until 2012 but we do not use the final year for the first cohort so that we observe students in both cohorts during nine consecutive years. We keep only students who start at the comprehensive program. In the first year, students can also start at a pre-vocational program. Students starting at this program often did not successfully finish their secondary education and therefore could not enroll in elite schools. This reduces our sample to 60,754 students in 2003 and 60,992 students in 2004. We also remove double entries, reducing the sample by 26 observations. We remove four students who entered both an elite and a non-elite school. Next, we remove students who do not live in one of the Flemish provinces. This reduces our sample to 59,092 and 59,824 students, respectively. Removing the observations with missing information about individual characteristics or detailed location reduces the sample to 55,527 and 56,252 observations, respectively. Note that the same source data was used by the KU Leuven research center HIVA (https://hiva.kuleuven.be/) who helped us to acquire and interpret it such that it could be used for research. They could only do this after we received the permission from the Flemish Ministry of Education and Training for our project. Variables: * Student characteristics anoniem_persoonsid Anonymized ID number to follow students during multiple years male Indicator = 1 if student is male toelage_SO Indicator = 1 if student received a study grant in secondary education no_ned_thuis Indicator = 1 if student does not speak dutch at home opl_moe1 Indicator = 1 if the mother did not graduate from secondary education opl_moe2 Indicator = 1 if the highest obtained degree of the mother is a degree in secondar education opl_moe3 Indicator = 1 if the highest obtained degree of the mother is a degree in higher education geboortedatum date of birth nationaliteit Code for nationality niscode Postal code of student sector Statistical sector of student repeated Student repeated at least one year before starting high school schooljaar Academic year * School characteristics hoofdnr School ID vplnr Number of the campus of the school hs Type of education (secondary education) stamnummer Code of study program nummer_admrgr Code of study program scho_ASO School only offers academic programs (treatment) * Student - school characeristics d_scho_ASO Distance to closest elite school d_scho_noASO Distance to closest school that does not offer the academic track d_scho_mixed Distance to closest school that offers both the academic and at least one other track * Outcome variables nodropout_strict Indicator = 1 if the student obtained a degree within at most 9 years of studying nodropout_soft Indicator = 1 if the student obtained a degree within at most 9 years of studying or is still in school after 3 years of study delay nodropout_wounknown Indicator = 1 if the student obtained a degree within at most 9 years of studying, students still in school after 3 years of study delay are omitted degreeontime Indicator = 1 if the student obtained a degree within 6 years of studying downgrade Indicator = 1 if the student downgraded at least once during secondary education 2. Data on neighborhood characteristics We use data on neighborhood characteristics at the municipal and statistical sector level provided by ADSEI (now Statbel). From this data source, we use median income, average educational level, share of inhabitants with Belgian nationality, and population density. Most neighborhood characteristics are measured in 2001, before students enroll in secondary education. Median income is measured in 2004 because this is not available in the census of 2001. These data are available on the following website: While census 2011 data are easily available on this website: https://statbel.fgov.be/nl/open-data , several older versions of this dataset need to be requested. Variables: niscode Postal code sectorcode Code statistical sector mediaaninkomen Median income in statistical sector mediaaninkomen_nis Median income in municipality gemiddeldinkomen Average income in statistical sector gemiddeldinkomen_nis Average income in municipality bev Population density at the statistical sector level bev_nis Population density of the municipality totaalbelgrel Fraction of inhabitants with Belgian nationality in the statistical sector totaalbelgrel_nis Fraction of inhabitants with Belgian nationality in the municipality hogeronderwijsrel Fraction of inhabitants with a degree in higher education in the statistical sector hogeronderwijsrel_nis Fraction of inhabitants with a degree in higher education in the municipality sechogerrel Fraction of inhabitants with a degree in secondary education in the statistical sector sechogerrel_nis Fraction of inhabitants with a degree in secondary education in the municipality More information on statistical sectors and data on their coordinates can also be obtained from Statbel. We use this to calculate distances. The data cleaning is acticated by the do file "do files/All_datacleaning". The analysis is activated by the do file "do files/All_estimation.do". All other do files in this folder are called by one of these 2. We also included a folder "do files/Appendix_C" which contains additional do files to create the results found in that appendix. Note that 2 do files are placed in the root directory as this was needed to work well with a user-written STATA command (parallel). We clarified where in the do files we gather information we used for figures and Tables. Note that some of the do files are called more than once after choosing a new macro such that we can look at alternative outcomes and samples with the same code. The tables and figures we mention in these do files are the ones where we use the main outcome and sample.