![Sample our Education journals, sign in here to start your access, latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/5932/TOC-Subject-Sample-2021-Education.jpg"})
Abstract
If school admission committees use alphabetically sorted lists of applicants in their evaluations, one’s position in the alphabet according to last name initial may be important in determining access to selective schools. Jurajda and Münich (2010) ‘Admission to Selective Schools, Alphabetically. Economics of Education Review, 29 (6): 1100–1109’ provide evidence consistent with this hypothesis based on graduation exams taken in grade 13 in the Czech Republic: ‘Z’ students in selective schools had higher exam scores than ‘A’ students. In this paper, we use the Trends in International Mathematics and Science Study & Progress in International Reading Literacy Study test scores of 4th graders and the Programme for International Student Assessment (PISA) test scores of 8th and 9th graders in the Czech Republic to provide evidence on how the alphabetical sorting outcome uncovered in Jurajda and Münich (2010) ‘Admission to Selective Schools, Alphabetically. Economics of Education Review, 29 (6): 1100–1109’ arises during early tracking into selective schools. Using the PISA data, we also provide corresponding evidence for Denmark, where sorting into selective schools happens in higher grades.
Acknowledgements
CERGE-EI is a joint workplace of the Center for Economic Research and Graduate Education, Charles University, and the Economics Institute of the Academy of Sciences of the Czech Republic. Both authors are Research Affiliates at CEPR, London; Jurajda is also Research Fellow at IZA, Bonn. Miroslava Federičová provided excellent research assistance. While working on this paper, Š. Jurajda has been supported by the Karel Janeček Foundation. D. Münich gratefully acknowledges support from the Czech Science Foundation (grant P402/12/G130).
Disclosure statement
No potential conflict of interest was reported by the authors.
Notes
1. Jurajda and Münich (Citation2010) discuss a more general, if still simple, model of student–school matching, where there is a group of indistinguishable applicants on the margin of admission thanks to either noisy admission exams or discrete support of the admission ‘score’ measure, and where the distribution of ability is assumed to be independent of one’s position in the alphabet.
2. One could also not rule out the possibility that studying in Czech secondary schools somehow favours the educational outcomes of ‘Z’ students. Students are not seated according to alphabetical order in Czech classrooms and we are aware of no mechanism that could generate such alphabetical learning biases.
3. The national system was in fact not fully introduced until 2011 and has been subjected to several design changes since then.
4. Asking hundreds of schools about admission procedures they used several years ago was not feasible. Jurajda and Münich (Citation2010) cite anecdotal evidence where a Czech university’s admission procedure openly used the alphabetical order to break ties among applicants with identical admission test scores.
5. Jurajda and Münich (Citation2010) illustrate that the repeated use of alphabetical sorting at the entry to both secondary and tertiary education can lead to efficiency losses in matching students to schools.
6. Only about 3% of inhabitants were born outside of the country, according to the Czech Statistical Office. We use the available indicators in the data to exclude students from immigrant families from our analysis. In the Czech Republic, there are also no types of last names related to a history of family wealth, such as ‘van’ or ‘von’ (Moldanová Citation2004).
7. 5th graders can submit an unlimited number of applications to eight-year ‘gymnazia’ programmes, but they are constrained by the geographical availability of such programmes with a typical NUTS-4 district offering at most two. Some of these programmes are private. The application process typically consists of a written exam administered independently by each school. Competition for entry is fierce, as only about half of applicants to these programmes were admitted in 2010, according to data collected by the Ministry of Education of the Czech Republic. See Filer and Münich (Citation2013) for a detailed description of the Czech education system.
8. Historically, ‘Efterskoler’ catered to pupils who had encountered academic or personal problems in public education; however, this image has changed dramatically and these schools are now attended by an increasing number of teenagers who desire (or whose parents desire) a year or more away from home.
9. Whether a 15-year-old is observed in the PISA data in grade 8 or 9, i.e. before or after the selection into boarding schools in Denmark, depends on the age at elementary school entry. We assume that age at school entry is independent of one’s last name initial. This assumption is supported by analysis of Czech 4th graders’ test scores in Sections 3 and 4.
10. The TIMSS&PIRLS data collection is based on a two-stage random sample design, with a sample of schools drawn first and classes of students selected from each of the sampled schools.
11. The longitudinal data collection effort at the Faculty of Education of Charles University, Prague is supported by a grant from the Czech Science Foundation No. P402/12/G130.
12. The share of observations with name initials on the initial sample, before any exclusion of data, is identical. In total, we have 3573 test scores with name information available.
13. The TIMSS&PIRLS study (as well as the PISA study; see OECD Citation2009) generates five versions of each test score, i.e. multiple estimates of the dependent variable of our analysis. These multiple estimates, which are sometimes referred to as ‘plausible values’, were developed to obtain consistent estimates of population characteristics in situations when rotated tests are used to increase the content coverage of tests without increasing the per-student testing time. The means of the dependent variables presented for our analysis-ready samples in Tables and are thus estimated using the pv routine (with full use of the sampling weights) developed for the analysis of ‘plausible values’ in TIMSS, PIRLS and PISA data in Stata by Kevin Macdonald.
14. The PISA study targets 15-year-old students attending educational institutions in grades 7 and higher and uses a two-stage stratified sample design where schools and students correspond to the first- and second-stage sampling units, respectively.
15. Specifically, we exclude the 83 Danish students in the 10th grade, the 22 students in the 7th grade, and the 19 students who already are in boarding continuation schools in the 8th grade.
16. We thus do not use Czech data on 27 pupils in the 7th grade and also on 2787 10th graders. We omit the large group in the 10th grade from the original PISA sample to maximise comparability between the two country analyses and also because these students are divided into five secondary school types leading to small samples and noisy estimates. We also drop 137 students who attend the special schools for students with learning or other disabilities.
17. Specifically, we exclude those students who report being born in another country or speaking a language at home that is different from the language of the test. In Denmark, where parents were also covered by the survey, we also exclude students whose parents report speaking a different language at home.
18. We maintain the assumption that students do not adjust their application strategy based on their position in the alphabet. We are not aware of any public discussion of the issue of alphabet sorting in admission procedures in the Czech Republic outside of the brief mention of our 2010 publication in reference to university admissions. It appears that neither the students nor the schools consider this issue important.
19. Both measures have a mean of 50 and a standard deviation of about 25.
20. The inference is based on robust standard errors, i.e. the Huber-White unconditional heteroscedasticity correction, and is not sensitive to clustering of residuals at the level of last name initial (i.e. to allowing for potential correlations of student unobservables within name initial groups).
21. We have also asked whether age of the surveyed 4th graders, which is determined by age at entry into elementary education, depends on one’s last name initial. Consistent with the assumption made in n. 9, the alphabet position parameters were equal to zero for all practical purposes and were not statistically significant.
22. In effect, we assume that PISA test scores reflect ability at the time of admission into the current programme.
23. Jurajda and Münich (Citation2010) found last name initials to be predictive of high school graduation test scores and also found that first name initials did not affect these test scores, which was reassuring for the interpretation of the evidence as corresponding to alphabetical effects in school admissions.
24. In contrast, the age and gender dummies have t ratios of six to seven in the mathematics test regression. Inference is again not affected by allowing for clustering of residuals at the level of name initials. The first name coefficients are large, but do not reach conventional levels of statistical significance. As was the case with Czech 4th graders in Table , boys are stronger in mathematics; next, older pupils have lower test scores, possibly due to class repetition or delayed school entry by weaker pupils.
25. Again, the unreported gender coefficient estimates indicate that boys score better in math than girls, while girls are better in reading in both countries. The age coefficient is small, negative and not statistically significant in most estimated specifications.
26. We have conducted several additional sensitivity checks, including estimating the relationships for each gender separately, and found little sensitivity. We have also asked whether the age of the tested students (and, hence, their age at school entry) is related to their alphabetical position and found no statistically significant effects consistent with the assumption made in note n. 9.
27. The Czech alphabetical sorting effect is roughly the same size as the gender effect on test scores.
28. Jurajda and Münich (Citation2010) also obtained one puzzling negative coefficient estimate of the first-name initial position effect, possibly corresponding to type-I inference error.
29. Again, estimates based on the cohort distribution measure of one’s alphabetical position are almost identical to those based on the numerical order in the alphabet. Again, the corresponding first name coefficients are small and never reach statistical significance levels.
30. Jurajda and Münich (Citation2010) provide direct evidence on the alphabetical effect at the margin of admission. Our data do not allow us to study the success of marginal applications to specialised schools.
31. Our measurement of the alphabetical sorting gap complements the extensive literature on educational performance gaps based on the PISA survey. For example, Botezat and Seiberlich (Citation2013) compare educational gaps in Finland to those in several Eastern European countries. Inequality in access to education is explored by Martins and Veiga (Citation2010) for the EU or by Gamboa and Waltenberg (Citation2012) for Latin America.
32. For example, the use of randomization within admission rules for New Delhi Nursery Schools is contested in courts (New York Times Citation2014).
33. On the other hand, covert randomisation makes it more difficult to use lotteries as sources of experimental assignment to identify the effect of attending an oversubscribed school on subsequent student outcomes, as, e.g. Zhang (Citation2013) or Engberg et al. (Citation2014) do.