Early Selection in Hungary A Possible Cause of

Early Selection in Hungary A Possible Cause of High Educational Inequality Daniel Horn research fellow Institute of Economics, Hungarian Academy of Sciences and Department of Economics, Eötvös University Budapest horn@econ. core. hu

Motivation • Age of selection is likely the best proxy for comprehensive schooling – The later the first age of selection the longer students will study in heterogeneous groups • Comprehensive schooling is said to be decreasing inequality (Meghir and Palme 2005, Pekkarinen, Uusilato and Kerr 2007) – and tracking correlates with higher inequality across countries (e. g. Hanushek and Woessmann 2005, Pfeffer 2008, Horn 2009) • Why? – Teachers matter (Hanushek, Rivkin and Kain 2005; Mc. Kinsey 2007) – Peers matter (Sacerdote 2001, Hanushek et al 2003. ) – … • So the longer one studies in selected groups the larger the difference will be • The Hungarian system is an „ideal” case for testing early selection

• Inequality in the Hungarian education system is especially high The system of education is certainly not comprehensive Source: OECD PISA 2009 Vol. II, p 44.

The Hungarian public education system Early-selective tracks HUNGARY 8 -yr-ac and 6 -yr-ac 2009/2010 ISCED 0 Grade ISCED 1 1 ISCED 2 1 st cycle level 2 nd cycle 2 3 4 5 6 7 ISCED 3 8 9 10 11 12 kindergarten general school academic secondary school track. ISCED 3 a 13 1 vocational secondary school track. ISCED 3 a (technikum) vocational training track ISCED 3 c Age 3 4 (ex-apprentice training) 5 6 7 8 9 10 11 12 13 14 15 16 17 1 18

Questions • Are early-selective tracks status selective? • Do early-selective tracks have a higher value-added? • Do longer, 8 -year-long academic tracks further increase differences, as compared to the 6 -year-long academic tracks? • Do others lose because of the early selection? If answers are affirmative the early selective tracks increase initial differences between students of different status. Thus, the Hungarian system is more unequal (ceteris paribus) because of the early selection.

Data • National Assessment of Basic Competencies (NABC) – Annually collected since 2006, PISA-like survey of all 6 th, 8 th and 10 th grade students (see below) – Reading and mathematical literacy – Approximate mean: 1500, sd: 200 (before 2010: 500/100) • Cross cohort and cross year comparable. Mean 1500 and sd 200 is only for 2008. 6 th grade • I standardize these, 0 mean 1 sd, for each year, cohort and subject – Detailed background questionnaires • generated Socio-economic status (SES) index – 0 mean 1 sd – Individual panel since 2008

The NABC cohorts 4 th grade 6 th grade 8 th grade 10 th grade 0 20 students from every school 0 20 students from each track from each school 0 20 students from every school 20 students from each track from each school full cohort every student from a sample of 195 schools full cohort 30 students from each track from each teaching site full cohort every student from a sample of 200 schools full cohort 30 students from each track from each teaching site every student from a 2008* sample of 200 schools full cohort every student from a 2009* sample of 200 schools full cohort every student from a 2010* sample of 200 schools full cohort 2003 2004 2006 2007 * Individual identification numbers available

Descriptive statistics

Percentage of students in different tracks HUNGARY 2009/2010 ISCED 0 Grade ISCED 1 1 ISCED 2 1 st cycle level 2 nd cycle 2 3 4 5 kindergarten general school 6 3, 8% 7 ISCED 3 8 9 10 11 12 3, 9% academic secondary school track. 3, 0% ISCED 3 a 5, 3% 13 4, 8% vocational secondary school track. ISCED 3 a 30, 3% vocational training track ISCED 3 c 96, 2% 90, 8% (technikum) 40, 2% 1 1 21, 7% Age 3 4 (ex-apprentice training) 5 6 7 8 9 10 11 12 13 14 15 16 17 18

Average math test scores HUNGARY 2009/2010 ISCED 0 Grade ISCED 1 1 ISCED 2 1 st cycle level 2 nd cycle 2 3 4 5 6 1673 1675* kindergarten general school 7 1491 12 11 1756 10 9 8 1693* 1807 1704 1615 1598* vocational training track ISCED 3 c (technikum) 1608 Age 3 4 1 1452 (ex-apprentice training) 1 vocational secondary school track. ISCED 3 a 1450* 13 academic secondary school track. 1821 ISCED 3 a 1782 ISCED 3 5 6 7 8 9 10 11 12 13 14 15 16 17 18

Average SES index HUNGARY 2009/2010 ISCED 0 Grade ISCED 1 1 ISCED 2 1 st cycle level 2 nd cycle 2 3 4 5 6 7 ISCED 3 8 kindergarten general school 0, 85 0, 83 9 10 11 12 academic secondary school track. 0, 8 ISCED 3 a 13 0, 8 0, 39 vocational secondary school track. ISCED 3 a vocational training track ISCED 3 c -0, 059 (technikum) -0, 1 1 1 -0, 81 Age 3 4 (ex-apprentice training) 5 6 7 8 9 10 11 12 13 14 15 16 17 18

1 st question ARE EARLY-SELECTIVE TRACKS STATUS SELECTIVE? YES! (DETAILS IN THE PAPER, DUE TO TIME CONSTRAINT – SORRY!)

2 nd and 3 rd question DO EARLY-SELECTIVE TRACKS HAVE A HIGHER VALUE-ADDED? DO LONGER, 8 -YEAR-LONG ACADEMIC TRACKS FURTHER INCREASE DIFFERENCES, AS COMPARED TO THE 6 -YEAR-LONG ACADEMIC TRACKS? (6 TH TO 8 TH GRADE)

The model • Model: where, i-individual, s – school (track), t – time Y – test score, X – student charactersitics – SES, Gender Z – school characteristics – track type, ed. provider (FE) • Endogenity problem – if Zst = Zs(t-1) – Thus if students do not change tracks during observations, track has an effect on Y(t-1) as well – students start general tracks six years, and 8 -yr-ac tracks two years earier than we measure them. Thus 6 th grade test scores reflect the quality of the given track • Note: 6 -yr-ac value-added is unbiased

Endogenity problem • Sollution: Instrumental variable estimation – instrument: distance from home to nearest 6 -yr-ac and 8 -yr-ac – assumption: distance has an effect on the chance to enter an early-selective track, but it does not affect test scores • Problem with instrument: – correlates with average SES • establishment of early selective tracks was demand driven – correlates with average teacher quality • teachers are also selected; just as students • Easing the problem – Spliting the sample decreases the bias • 8 -yr-ac vs. 6 -yr-ac and 6 -yr-ac vs. general – Track effects in IV estimation are underestimated: the stronger the correlation bw. distance and teacher quality the more effect the instument „absorbs”.

OLS VARIABLES General school Standardized reading score, 6 th grade Standardized math score, 6 th grade SES female Education provider FE. Constant Observations R-squared (8) Math 0. 0221 (0. 0208) 0. 138*** (0. 0304) 0. 565*** (0. 00496) 0. 165*** (0. 00555) 0. 230*** (0. 00493) 0. 0832*** (0. 00379) 0. 222*** (0. 00493) y -0. 100*** (0. 0148) 0. 630*** (0. 00645) 0. 0597*** (0. 00441) -0. 131*** (0. 00537) y 0. 0227 (0. 0215) 82, 211 0. 638 8 -yr-ac (7) Read -0. 0383*** (0. 0145) 0. 0732*** (0. 0188) 82, 210 0. 596 Robust site clustered standard errors in parentheses, *** p<0. 01, ** p<0. 05, * p<0. 1

IV – split sample VARIABLES (1) First stage 8 -yr-ac vs. 6 -yr-ac (2) (3) Second stage Read Math 0. 228*** (0. 0831) general Standardized reading score, 6 th grade -0. 545** (0. 226) 8 -yr-ac 0. 0810 (0. 0600) (4) First stage General 6 -yr-ac vs. general (5) (6) Second stage Read Math 0. 0114 (0. 311) 0. 00694 0. 545*** 0. 188*** -0. 00970*** 0. 561*** 0. 165*** (0. 0143) (0. 0119) (0. 0138) (0. 00178) (0. 00557) (0. 00636) -0. 0232 0. 217*** 0. 702*** -0. 0282*** 0. 215*** 0. 625*** (0. 0185) (0. 0106) (0. 0138) (0. 00359) (0. 00840) (0. 0113) SES -0. 0183 0. 0483*** 0. 0744*** -0. 0194*** 0. 0720*** 0. 0591*** female (0. 0120) 0. 0284 (0. 00842) 0. 190*** (0. 00876) -0. 244*** (0. 00240) -0. 00673** (0. 00594) 0. 219*** (0. 00797) -0. 127*** (0. 0174) (0. 0138) (0. 0145) (0. 00283) (0. 00545) (0. 00593) y 0. 391* (0. 220) y 0. 0305 (0. 301) 78, 743 0. 618 78, 744 0. 584 Standardized math score, 6 th grade distance bw. home and closest 8 -yr-ac -0. 0169*** (0. 00155) distance bw. home and closest 6 -yr-ac Ed. provider FE Constant Observations R-squared y 0. 515*** (0. 0598) y -0. 0240 (0. 0299) y -0. 0421 (0. 0435) 0. 00198*** (0. 000238) y 0. 949*** (0. 00602) 7, 986 0. 131 7, 989 0. 536 7, 986 0. 644 78, 744 0. 186

Estimated differences bw. tracks, reference 6 -yr-ac HUNGARY 2009/2010 ISCED 0 Grade ISCED 1 1 ISCED 2 1 st cycle level 2 nd cycle 2 3 4 5 6 kindergarten general school ISCED 3 7 8 9 10 11 12 M: -0, 226** R: n. s. vocational secondary school track. ISCED 3 a M: 0, 228*** M: 0, 17** academic secondary school track. ISCED 3 a R: n. s. M: n. s. R: -0, 545** Age 3 4 1 (technikum) vocational training track ISCED 3 c (ex-apprentice training) 13 5 6 7 8 9 10 11 12 13 14 15 16 17 1 18

4 th question DO OTHERS LOSE BECAUSE OF THE EARLY SELECTION? YES, IN MATH. Students in general schools, where early-selective tracks are available perform worse in math bw. 6 th and 8 th grade, compared to students in general schools with no early-selective track around.

Conclusion

Conclusion • Early selective tracks are status selective (6 -yr-ac tested) • Early selective tracks have a higher value added – between 6 th and 8 th grade in reading – between 8 th and 10 th grade in math • 8 -yr-ac performs better than 6 -yr-ac in math each year • Students left in general tracks lose Early selection increases the gap between students of different social background

Thank you for your attention! Comments welcome! horn@econ. core. hu

Number of sites Track type combinations within sites 2463 228 162 95 91 82 78 72 52 50 43 23 23 23 15 14 8 8 8 7 7 5 3 3 2 2 2 1 1 1 general 1 0 0 1 0 1 1 1 1 0 1 1 0 0 1 1 1 8 -yr-ac 0 0 0 0 1 0 0 0 0 1 1 1 6 -yr-ac 0 0 0 1 1 0 0 0 0 0 1 1 1 0 0 1 4 -yr-ac 0 0 0 1 1 1 0 0 0 1 1 0 1 1 1 0 0 1 1 1 technikum voc. train. 0 1 1 1 0 0 0 1 1 1 0 1 0 0 1 1 0 0 0 0 1 1 0 0 0 1 0 1 1 1 0 0 1 1 0

Spatial distribution of early selective tracks