Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

The labour market effects of the polish educational reform of 1999

The labour market effects of the polish educational reform of 1999 We estimate the effect of the 1999 education reform in Poland on employment and earnings. The 1999 education reform in Poland replaced the previous 8 years of general and 3/4/5 years of tracked secondary education with 9 years of general and 3/3/4 years of tracked upper-secondary education. The reform also introduced new curricula, national examinations, teacher standards, and a transparent financing scheme. Our identification strategy relies on a differ - ence-in-differences approach using a quasi-panel of pooled year-of-survey and age-of-respondent observations from the Polish sample of the EU-SILC database. The results indicate that the reform has increased employment probability (by around 3 percentage points) and earnings (by around 4%). Keywords: Education reform, Returns to education, Poland, Detracking, Labour market, Difference-in-differences JEL Classification: I21, I24, I26, J24 time, the earnings of those with highly educated parents 1 Introduction decreased, but the average effect was positive. In this study, we look at the Polish education reform of The reform in Finland in the 1970s delayed the tracking 1999 that, among others, has increased general education of students from age 11 to 16 by extending comprehen- for all students, introduced a national core curriculum, sive education from 4 to 9  years. Pekkarinen, Uusitalo, standardised examinations at the end of each education and Kerr tested the effects of this reform on the income stage, introduced new teacher qualification require - elasticity (Pekkarinen et  al., 2009) and the average test ments, and adopted a transparent financing scheme. scores (Kerr et  al., 2013) and concluded that it had only Previous results on the effect of education reforms a small but overall positive effect on both dimensions. In on labour market outcomes are mixed. Comprehensive Norway, a school reform implemented between 1960 and education reforms in Scandinavia are usually hailed for 1972 increased compulsory education from 7 to 9 years. their positive effects on labour market outcomes. Swe - The 9  years of comprehensive education consisted of den carried out a major education reform in the 1950s, 6  years of primary school and a 3  year lower second- increasing compulsory education from 7/8 to 9  years, ary school. Lower secondary schools existed before the abolishing tracking (i.e. sorting students into general reform, too, but attending them was voluntary and they and vocational tracks) based on academic achievement were only available in some municipalities. This reform after the sixth grade, and introducing a national curricu- also improved the quality of education by standardising lum. Meghir and Palme (2005) showed that this reform the curriculum. Aakvik et al. (2010) find that the reform increased educational attainment and the future earn- increased the level of education in the country, and it also ings of children with low educated parents. At the same increased the returns to upper secondary and tertiary qualifications. However, some papers find no-returns to extending compulsory education in Germany (Pischke *Correspondence: drucker_luca-flora@phd.ceu.edu and von Wachter 2008) or France (Grenet 2013), and Central European University, Budapest, Hungary the story is similar if we look at education reforms that Full list of author information is available at the end of the article © The Author(s) 2022. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/. 13 Page 2 of 21 L. F. Drucker et al. increased the time of education particularly for voca- shortened by one year so that the overall length of pri- tional students: no (substantial) effects were found for mary and secondary education remained 12 years, except the Netherlands (Oosterbeek and Webbink 2007). in the basic vocational track, which was 3  years before These differences in the results might stem from the the reform and remained 3 years. differences in the type of education extended. While in Besides these structural changes, the reform changed Germany and France, the increase in education was after other major parts of the education system: Poland intro- tracking, in the Anglo-Saxon (Grenet 2013) and Scan- duced a national core curriculum, standardised examina- dinavian countries, where a positive effect of additional tions at the end of each education stage, introduced new years of schooling was found, it was before tracking. teacher qualification requirements, and adopted a trans - We argue that the reason for this is that “before track- parent financing scheme. Using PISA data, Jakubowski ing reforms” not only extended education but also (2015) showed that the between-school variance of test improved the quality of inputs, such as the curriculum, scores decreased after the reform. Jakubowski et  al. teachers, or peers, as these are crucial in the education (2016) also showed, using difference-in-differences production function (e.g. Chetty et al. 2014; Rivkin et al. approach adjusted with propensity score matching, that 2005; Sacerdote 2011). When additional years of edu- the fast increase in Polish students’ PISA test scores was cation are "inserted" before students are selected into driven by potential vocational students, who after the tracks, the effect is different from when they are already 1999 reform were in the comprehensive lower second- selected into academic or vocational tracks. ary school instead of already having been tracked into the Offering increased general training to vocational stu - low-tier vocational schools. They argue that the increased dents does not affect teachers and peers, as these inputs resources of an additional year of better teachers, peers, are unlikely to change due to the increased length of and general curriculum have improved these students’ vocational education or increased general content. test scores. They also show that the improvements in However, if the system is ’de-tracked’, i.e. the age of selec- student achievement at the age of 15 are still substantial tion is increased, the composition of teachers and peer at the age of 16 or 17 when students are already tracked groups will remain as they were before selection in that to different secondary school programs. These results “inserted” year, and thus–at least for the later vocational confirm our expectation that improved inputs result not track students–a higher peer and teacher effect might only in better achievement but provide a stronger basis improve their long term outcomes. for further development of foundational skills among stu- The 1999 reform of Poland, which we analyse in this dents who continue education in vocational tracks. study, was a comprehensive one like the Scandinavian Additional years of education translate into higher reforms. However, it was unique in the sense that besides wages (Psacharopoulos, Patrinos, 2018). This effect is extending comprehensive education, it altered all of the partly explained by the human capital theory, which above-mentioned essential inputs in the education pro- assumes that better education translates into skills that duction function. This reform replaced the former 8 year increase our productivity (Becker, 1975). An alternative general primary school with a 6 year general primary and explanation is that by passing through the education sys- a 3  year general lower secondary school. Therefore, the tem, one signals its innate ability and employers reward age of first selection was postponed from age 15 to 16. it with higher wages (see Page, 2010, for a review). I our The length of tracked upper-secondary education was case, both theories imply that the Polish reform should increase wages as it improved skills but also educational attainment. Research also shows that at the country level the improvement of cognitive skills translates into higher A counterexample is presented by Malamud and Pop-Eleches (2010), who economic growth (Hanushek, Woessmann, 2012). More- find no effect before tracking. over, better skills bring non-economic benefits (Heckman Improving only the curriculum, however, might not be sufficient. In Swe- et al., 2018). den in the 1990’s, academic content in all vocational tracks was increased. This change reduced curriculum differences between the academic and The empirical research documents well that the Pol- vocational tracks at the upper secondary level, also allowing vocational ish reform was beneficial for students’ school perfor- graduates to apply to university. The reform was preceded by a pilot mance, but we are aware of only a couple of attempts scheme, introducing the new system in selected schools throughout the country. Using this time and spatial variance Hall tested the effect of this to assess its long-term impacts on labour market out- reform on tertiary enrolment and wages (Hall 2012) and on unemployment comes. Liwiński (2020) applies a regression discon- chances (Hall 2016). The results of these analyses have shown no difference tinuity design (RDD) focusing on the differences in between pre- and post-treatment cohorts in outcomes. Hall argues that a potential reason for the non-effect is the increased dropout rate of the voca- employment probability and wages among adults with tional track students induced by the academic content. basic vocational degrees. He found that the reform At least in the short run, and thus, the typically utilised regression-discon- improved hourly wages by 13%. However, as the tinuity approach shows no effects. The labour market effects of the polish educational reform of 1999 Page 3 of 21 13 reform has also changed patterns of selection into dif- 1.1 The educational reform of 1999 in Poland ferent upper secondary tracks, with a decreasing num- The 1999 educational reform was one of the four ber of students attracted to basic vocational schools, reforms—social security, healthcare, public administra- the results comparing just adults with basic vocational tion, and education—implemented by the government training might provide a biased picture of the over- elected in 1997. The three main goals of the education all effect of the reform. Strawinski and Broniatowska reform were to increase the level of education in soci- (2021) applied a similar RDD approach to the Polish ety, provide equal educational opportunities to every- LFS data, but limiting their sample to people with sec- one, and improve the quality of education (Bialecki et al. ondary education degree only. Their result suggests a 2002). While the other three reforms were implemented rather small but positive overall impact of the reform. simultaneously, they similarly affected all students and They provide additional results showing larger ben- all adults in the population. The education reform was efits for vocational school graduates in rural areas. rolled out so that the older cohorts followed the old sys- However, these results might be biased for the same tem, while only the younger cohorts, born from 1986, fol- reasons as the analysis of Liwiński (2020) as a larger lowed the new system with the new structure, curricula, number of students in the post-reform cohorts contin- and standardised national exams. ued education into tertiary level. The reform consisted of many distinct parts, of which These papers use RDD and rely on the assumption we concentrate mainly on the structural changes but that cohorts before and after the reform are as if they briefly discuss all parts here (see Jakubowski, 2021, for a were randomly allocated. This is a strong assumption. more detailed discussion of the reform components). In Typically, RDD studies use high-density data around sum, the 1999 reform (1) extended comprehensive gen- the cut-off, for these studies, this is clearly not the case eral education by one year, and (2) changed the struc- (but see Grenet 2013 for a similar approach). Thus, ture of general education by dividing the previous 8 year using a quasi-DID identification strategy could offer a primary school into 6  years of primary and 3  years of strong robustness check to the RDD results. This study general lower secondary school. It also (3) shortened aca- aims at filling this gap by adding more evidence on demic and technical upper secondary education by one the labour market outcomes of the reform. We show year, but the basic vocational school remained 3-years- that the reform had a significant positive effect on the long. Besides these structural changes, the reform (4) earnings and employment chances of the post-reform introduced a core curriculum, which guided teachers in cohorts. We argue that these results come from the determining their syllabi, giving them more autonomy, comprehensive nature of the reform, i.e. general educa- and it (5) introduced the teacher professional attainment tion was extended, and its quality was improved before ladder with four levels and strong incentives for profes- students were tracked to upper secondary schools. sional development and obtaining master degrees for all Unlike the Scandinavian reforms, the Polish reform teachers. The third line of changes concerned the testing did not have a pilot stage; instead, it was introduced and admission system: (6) new, standardised tests were simultaneously in the whole country. Thus, our esti- introduced at the end of each education stage (primary, mations cannot exploit any time or spatial variation. lower secondary, and upper secondary), and parallel to Due to the lack of high-frequency micro-data before that, (7) the admission system to each stage shifted from and after the reform, it is impossible to use a regres- entrance examinations to using the results of these end- sion discontinuity design. However, by pooling sev- of-stage standardised exams. Finally, (8) the reform also eral years of cross-sectional surveys, we can generate transferred ownership of nearly all schools to local gov- a quasi-panel of year-of-survey and age brackets, ernments and introduced a new per pupil formula-based which we will use to estimate difference-in-differences financing system. models. Unlike in the Scandinavian studies, the vari- Figure 1 shows the structural changes in the Polish edu- ance does not come from the time of implementation cation system. Most importantly, the newly established but from the time (year) of observation. This way, we comprehensive lower secondary schools (gimnazjum will directly compare the employment chances and in Polish) had to follow the same general curriculum real earnings of pre-reform (control) and post-reform and admit all students from their catchment area with- (treatment) cohorts. In Sect.  2, we discuss the 1999 out any additional requirements. They could accept reform in more detail. Section  3 presents the data and additional applicants from other areas to the remaining descriptive analyses. In Sect.  4, we present our esti- places. Lower secondary schools were larger, and there mation methodology and baseline results. Section  6 were fewer of them than primary schools. They opened shows robustness checks, and Sect. 7 concludes. 13 Page 4 of 21 L. F. Drucker et al. Fig. 1 Pre- and post-reform structure of the Polish education system. The figure shows the structure of the Polish education system before and after the reform. The width of the specific upper-secondary level tracks approximates the ratio of students attending these tracks in 2000 (pre-reform) and 2004 (post-reform). Source of data: Polish Statistical Office in larger settlements, which was especially important in comprehensive education, the end of compulsory educa- rural areas, where one gimnazjum collected the children tion increased to age 18. from neighbouring villages who previously attended dif- Finally, Poland introduced a new curriculum and stand- ferent local primary schools. So, students spent their last ardised final exams to measure whether students achieve 3  years of general education in a comprehensive institu- the curricular goals at the end of the 6th and 9th grades. tion intended to give the same high-quality education for The 6th-grade test was low-stakes and provided informa - students all over the country. tion on student and school performance. The 9th-grade Before the reform, there were three upper secondary tests were high-stakes and were used as an entrance exam tracks: a 4  year academic secondary school (liceum), a to the upper secondary stage. These exams were first 5  year technical secondary school (technikum), and a launched in 2002 for the first cohorts that followed the new 3  year basic vocational school. Only liceum and techni- curriculum in the new comprehensive lower secondary kum ended with a maturity exam, which did not change schools. For the same cohort, the new standardised matu- with the reform. After the reform, academic and techni- rity exams were introduced in 2005 and replaced higher cal secondary schools became one year shorter, but the education admission exams. The standardised examina - basic vocational schools remained 3  years long. There - tions strengthened the impact of the extended general fore, lower secondary graduates and those who chose the education as they were compulsory for all students. vocational track received one additional year of compre- Overall, the reform revolutionised all parts of the hensive education. A new institution was operating for school system. However, one can reasonably expect a short time, the so-called profiled academic secondary that these changes differently affected various groups of school, but it was abolished after a few years. In general, students and had different impacts over time. One can the upper secondary education curricula and programs expect a delayed effect on student outcomes of changes were not changed substantially. In academic and techni- in curriculum, professional teacher standards, or system cal upper secondary schools the curricula were adjusted governance structure. Even the reformers assumed that where necessary as parts of the material taught in first several years were needed to benefit from these changes. grades were now moved to general lower secondary However, one year’s general education extension schools. Before 1999, education was compulsory until age 17 Compulsory age of schooling was increased from 17 to 18 in 1997 in arti- cle 70 of the Polish Constitution. This change affected all cohorts already in with the possibility of part-time education after finish - school, i.e. both pre- and post-treatment cohorts (see Jung-Miklaszewska ing the 8  year primary school. With the extension of 2003). The labour market effects of the polish educational reform of 1999 Page 5 of 21 13 immediately provided students with one more year of and health status data at the personal and household the general curriculum. The new accountability system level. The data consists of private households with all established by new standardised examinations provided household members surveyed, but only above age 16 are additional incentives to students and teachers to cover interviewed personally for the income data. In this paper, the general curriculum. However, that made a difference we use the cross-sectional database of EU-SILC. only to students who would otherwise have gone to the To generate a balanced “quasi-panel”, we pool the cross- basic vocational schools. Before the reform, the other sectional datasets between 2005 and 2013 and restrict groups had to follow the academic curriculum and pass the sample to participants between ages 20 and 27. This secondary school entrance examinations. allows us to compare the pre-reform and post-reform The evaluation of changes in student achievement con - groups at the same age: in 2005, the youngest control firms that the main group that the reform affected imme - group members were 20 years old, and in 2013, the oldest diately was the students of basic vocational schools. treatment group members were 27 (see Table  1 below). Jakubowski et  al. (2016) demonstrate that the immediate This means we have 16 cohorts in the sample, eight in the impact of the reform on 15-year-olds’ achievement was close treatment (T86 to T93–where 86, 87, etc., signal the birth to one standard deviation for students whose socioeconomic year of the cohort) and eight in the control group (C78 background is identical to former students of basic voca- to C85). These people were born between 1978 and 1993 tional schools. Moreover, additional comparisons for 16- and (see Tables 8 and 9 in the Appendix). There are around 48 17-year-old students suggest that this effect is long-lasting. 500 observations in the sample, with 23 500 in the con- After 1–2  years of basic vocational education, students still trol group and 25 000 in the treatment group. In Poland, show improved reading and mathematics outcomes close to the school starting age is 7, and the threshold is January an equivalent of one year of instruction. A similar achieve- 1. In the sample, everyone born after January 1 1986 is ment effect for students in the academic track was negligible considered to be treated—to have studied in the new sys- even 6 years after the reform was implemented. tem—and everyone born until December 31 1985, is con- As the new comprehensive lower secondary schools sidered to be in the control group. There is a scope for started to operate already in 1999/2000, the first cohort that misclassification around the threshold because of grade started their 7th grade in the new system is the 1986 cohort. retention or skipping a grade. However, grade repetition So, we categorise everyone born after 1986 as a member of was always rare in Polish schools and was mainly limited the treated group. We argue that structural changes imme- to students with special needs. The earliest data on grade diately affected every cohort from 1986 onwards, while the repetition are available for 2005, and they show that in other changes were beneficial in the long run, affecting both primary schools, only 0.6% of students were repeating a pre- and post-treatment cohorts, although to a slightly dif- grade (GUS, 2006). ferent extent. So, if we compare a few cohorts before and For educational attainment, we rely on the ISCED clas- after the reform, the potential differences in their labour sification : those with ISCED 2 (lower secondary) qualifi - market outcomes are most likely caused by the additional cation or below are considered low-educated, those with year of general education before the delayed tracking. ISCED 3 (upper secondary) or 4 (post-secondary non- tertiary) are at the medium level, and those with ISCED 5 1.2 Da ta and descriptive statistics (tertiary) are highly educated. We see the highest educa- We use the EU Statistics on Income and Living Condi- tion level attained in the data and the year when it was tions (EU-SILC) data between 2005 and 2013. The EU- achieved, but unfortunately, we do not observe the exact SILC contains detailed income and labour, education, type of school the person graduated from. So, in the case of upper secondary education, we see how old the per- son was when she finished the upper secondary level, but School enrollment cut-off date in Poland is January 1. we do not see whether it was a liceum, a technikum, or a 6 th For example, the first 6 grade standardised exams took place in 2002, so basic vocational school. Figure  2 shows the distribution the 1986 and 1987 cohorts did not have to take them yet. of the ages when each level of education was achieved, Previous studies looking at the effect of the Polish education reform of 1999 have used the Polish LFS data (Liwiński 2020, Strawinski and Bronia- separately for the treatment and control group. The dis - towska 2021). We have also planned our initial analysis on the Polish LFS, tribution of finishing ages is different for all three educa - however when we asked the Polish Statistical Office for the official data- tion levels for the two groups. The most visible change is base, their offer has vastly exceeded our budget. So we decided to use the EU-SILC instead, which was freely available through the Eurostat. Admit- tedly, the national LFS is better in its level of education measure as the SILC. However—as we point out below—education is a ‘bad control’ in this reform. We also admit that the number of observations is higher in the LFS 8 The summary of the variables can be found in the Appendix (Table 7). than in the SILC, however, earnings are much better recorded in the EU- Note that each figure shows only those people who have that particular SILC. us we b Th elieve that the SILC is not an inferior option compared to education level as the highest finished level at the date of the survey. LFS for this analysis. 13 Page 6 of 21 L. F. Drucker et al. Table 1 Distribution of treatment and control group cohorts by age and year of survey Year of survey 2005 2006 2007 2008 2009 2010 2011 2012 2013 Age 20 C85 T86 T87 T88 T89 T90 T91 T92 T93 21 C84 C85 T86 T87 T88 T89 T90 T91 T92 22 C83 C84 C85 T86 T87 T88 T89 T90 T91 23 C82 C83 C84 C85 T86 T87 T88 T89 T90 24 C81 C82 C83 C84 C85 T86 T87 T88 T89 25 C80 C81 C82 C83 C84 C85 T86 T87 T88 26 C79 C80 C81 C82 C83 C84 C85 T86 T87 27 C78 C79 C80 C81 C82 C83 C84 C85 T86 The table shows the distribution of the treatment and control cohorts in our sample. The numbers in the cells indicate the birth year of the cohort that falls into that year-of-survey-age cell. The letters C and T show if the cohort belongs to the control group or the treatment group For the number of observations, see the Appendix, Table 8 in the median finishing age of the lowest educated: the cohorts, but this average zero effect masks a significant median age increased from 15 to 16, mainly because of composition effect: the low educated stay about 0.9 years the one additional year of comprehensive education. longer in school, while the average upper secondary The difference for upper secondary graduates (ISCED graduate stays a little over 1  month (0.09  years) longer 3) is not so salient, but we also see the distribution shift in school. This average effect for the medium educated to the right, driven by the one extra year for vocational is probably due to the ca. 15% of students in basic voca- students. The distribution of tertiary (ISCED 5) gradu - tional tracks, who stay about one year longer in school. ation age became bimodal, as, in the Bologna system, a On the other hand, higher educated people finish educa - BA degree also counts as a tertiary qualification, while tion about 0.7 years earlier, which is probably due to the earlier only long-cycle programmes existed. Note that previously non-existent BA degree. When looking only studying the differences between treated and non-treated at those who are currently not pursuing any education– cohorts in this dimension is not straightforward. Poland which means that they have finished their educational signed the Bologna Declaration in 1999 and 28 other career, at least for a while–the pattern is similar. How- European countries, which introduced the typical three- ever, effect sizes are slightly different: low educated stay level system of tertiary education—bachelor, master, and in school 0.7  years longer, medium level educated over doctorate. The three-cycle structure of higher education 2 months longer and higher educated about 0.8 years less. was implemented gradually. In 2003 Poland was already The two main outcome variables we are interested in implementing the Bologna structure (Kwiek 2014). By are employment status and earnings. EU-SILC classifies the school year 2004/2005 10% of state higher educa- activity status into four categories: at work, unemployed, tion institutions had already adopted the 2-cycle model in retirement or early retirement, and other inactive. The (Bachelor and Master) in all fields of study and 50% of the first category covers those who work either full-time or institutions in at least 50% of the fields of study (see Euro - part-time or are self-employed full-time or part-time. pean Commission 2005). In 2008 all tertiary students When looking at employment chances, we compare were enrolled in the Bologna system (see Kwiek 2014). employed people to unemployed, as we find that the Consequently, the first bachelor-level graduates of share of the active population compared to the inactive the new system entered the labour market in around did not change during the sample period. We drop those 2006. Therefore, the 1984 and 1985 cohorts also had the in retirement or in early retirement as there are only 55 of opportunity to study in the Bologna system, depending these people in our sample. on their institution and field of study–however, the first Income data is collected as gross current monthly full "Bologna cohort" was the 1989 cohort (who finished earnings before the deduction of taxes and social insur- in 2011). This coincidence makes it hard for us to study ance contributions. Income is given in Euros and cur- the effect of the reform on the upper end of the education rent prices, so we converted this data to Polish Złoty in distribution. 2005 prices. The database contains data on experience, Appendix Table 11 expl or es the age of finishing educa - expressed as the number of years spent as an employee or tion in a regression framework. Treated cohorts, on aver- self-employed since the respondent first started a regular age, tend to stay just as long in education as the control job. As low educated people studied for one more year The labour market effects of the polish educational reform of 1999 Page 7 of 21 13 Fig. 2 Distribution of the age when the highest educational level was attained. The figures show the distributions of the age when the respondents finished their highest education level. Each figure corresponds to a subsample with a specific education level: those with at most lower secondary education (ISCED 2), upper secondary (ISCED 3), and tertiary education (ISCED 5) 13 Page 8 of 21 L. F. Drucker et al. after the reform, and this change was immediate, there Table 2 Ages when control and treatment group members started their first job was a year (2001) when, in theory, no one graduated from the lowest education level. There is a similar gap year Age Control Treated Difference (t-stat.) (2004) for vocational graduates because their studies also a. Full sample became longer by one year. This, and other differences 20 18.14 18.52 0.38 in labour market characteristics make it very important (2.79) to control for labour market entry in our estimations. 21 18.87 19.33 0.46 There is data on the year of starting the first job, which, of (4.85) course, is only available for those who have ever worked. 22 19.52 19.71 0.19 To correct for this, we generated a variable called ’labour (2.27) market entry year’ that equals the year of starting the 23 20.00 20.03 0.03 first job, or, when it is not available, the year when the (0.35) respondent finished their highest education level (pro - 24 20.33 20.30 −0.03 vided in the EU-SILC database). We corrected this vari- (−0.36) able so that the labour market entry age of respondents is 25 20.81 20.88 0.07 at least 18, as this is the legal age when young people can (0.70) start to work full time in Poland. 26 21.12 21.21 0.09 Table 2a–c show when each age group of pre- and post- (0.75) reform cohorts started their first job. Since post-reform 27 21.22 21.38 0.16 cohorts finished their education later, they started work - (0.95) ing later (at 20.47 vs 20.35  years, p < 0.01 ). The differ - b. ISCED 2 or below ence is significant for the younger cohorts and disappears 20 16.90 18.14 1.24 in the older cohorts. Surprisingly, later job starting does (3.27) not go together with less experience: treated people of 21 18.26 18.81 0.55 the same age tend to have higher years of experience than (1.83) the control group (2.87 vs 2.58 years on average, p < 0.01, 22 18.75 19.22 0.47 see Table  3a–c for more detail). This could be due to (1.53) different employment chances. If treated cohorts have 23 19.02 19.34 0.33 higher employment chances than the control cohorts, (0.90) then–on average–even if they start working later, they might secure more stable jobs and gain more experience 24 19.00 19.75 0.75 over a shorter period. Table  4a–c seem to underline this (1.90) assumption, as we see a higher share of employed peo- 25 19.28 20.25 0.96 ple in younger cohorts among the treatment group than (2.47) in the control group, and the differences are even higher 26 19.75 20.19 0.44 within the lowest educated people. Across all age groups (0.79) in the full sample the means are 79 percent vs 77 percent 27 19.91 19.28 −0.62 (p < 0.01). (−0.93) Figure  3 shows the real earnings distribution of the c. ISCED 3 pre- and post-reform cohorts. Treated people tend to 20 18.34 18.62 0.28 earn more on average (PLN 1664 vs PLN 1388), mainly (2.03) because fewer people are at the bottom of the earn- 21 18.92 19.38 0.46 ings distribution in the treatment group. That is, the (4.49) earnings distribution is shifted to the right, moving 22 19.55 19.72 0.18 (2.03) 23 19.94 19.95 0.01 10 th The last primary school graduates finished 8 grade in 2000, and the first (0.07) th lower secondary graduates finished 9 grade in 2002. 24 20.16 20.07 −0.09 The results are essentially the same if we change the minimum working (−0.88) age to 15 for the control and 16 for the treatment cohorts. The pre-reform 25 20.26 20.33 0.07 cohorts could start working in part-time jobs at the age of 15. Along with the extension of comprehensive education until the age of 16, the legal job- (0.57) starting age for part-time jobs was also extended to 16. 26 20.30 20.26 −0.04 Means are weighted by the inverse of the size of each age group in the (−0.24) sample. The labour market effects of the polish educational reform of 1999 Page 9 of 21 13 Table 2 (continued) Table 3 Mean years of experience by age Age Control Treated Difference (t-stat.) Age Control Treated Difference (t-stat.) a. Full sample 27 20.24 20.32 0.08 20 0.79 0.85 0.06 (0.44) (0.69) Tables 2a–c show the mean age when treated and control group members 21 0.99 1.11 0.12 started their first regular job, separately by age groups. The third column shows the differences between the two groups at each age with t-statistics in the (1.78) parentheses. 2a presents the data of the whole sample, Table 2b for those who 22 1.36 1.64 0.28 have at most a lower secondary qualification, and 2c for those with an upper (4.55) secondary qualification 23 1.83 2.22 0.38 (5.70) 24 2.36 2.84 0.48 those at the bottom of the distribution to the middle. (6.21) This shift is apparent in the full sample and for the low 25 2.79 3.23 0.45 educated. (4.90) From the raw data and from the research before us, 26 3.44 3.81 0.36 we assume that Poland’s 1999 comprehensive educa- (3.08) tion reform had a non-negligible and positive effect 27 4.23 4.40 0.17 on the Polish labour market. We believe that it was (1.04) especially young people (where the level of educa- b. ISCED 2 or below tion and skills gained matter most) and those at the 20 1.02 1.01 −0.01 bottom of the education distribution, who stayed one (−0.05) more year in school, who benefited the most from the 21 0.83 1.32 0.50 reform. (2.75) 22 1.79 1.70 −0.08 1.3 M ethodology and baseline results (−0.36) We now turn to our causal estimates. Year of birth deter- 23 2.33 2.47 0.14 mined the assignment into treatment, which means that (0.50) self-selection into the treatment or control group was 24 2.97 2.95 −0.03 impossible. However, there are cohort-specific differ - (−0.08) ences, as the treatment and control group members were 25 3.09 3.18 0.10 born in different years. To handle this issue, we control (0.26) for age fixed-effects. There are also differences between 26 3.54 3.83 0.28 the years each survey was taken, so we control for sur- (0.41) vey year fixed effects. We assume that in the absence 27 4.40 5.78 1.38 of the treatment, in both the treatment and the control (1.34) group, changes in outcomes between two consecutive c. ISCED 3 survey years would have been the same for all ages, and 20 0.75 0.81 0.06 vice versa: changes between two consecutive ages would (0.57) have been the same for all survey years (parallel trends 21 1.02 1.09 0.07 assumption). For this reason, as a baseline, we opted for (1.00) a difference-in-differences method, where age and year- 22 1.36 1.63 0.27 of-survey act as the two dimensions of the estimation (4.16) 23 1.86 2.27 0.41 (5.36) We also included labour-market entry fixed effects in some regressions, 24 2.48 3.06 0.58 which further diminish the problem of cohort-specific differences. See below (6.03) for more detail. 25 3.14 3.74 0.60 For instance, suppose the only available survey years are 2005-2006. Then the average change between 2005 and 2006 for age groups ranging from 21 (5.05) to 27 is a counterfactual change for age 20 (the only age group from which 26 3.96 4.65 0.69 we have both treated and control observations in the sample). Now, suppose (4.31) we have only 2006 and 2007, the assumption is that average change between 27 4.85 5.13 0.27 2006 and 2007 for cohorts 22-27 and 20 is a counterfactual for change at age 21 etc. And similar logic applies for the other dimension. See Appendix B (1.16) for further discussion. 13 Page 10 of 21 L. F. Drucker et al. Table 3 (continued) Table 4 Share of employed people among those who are active (employed or unemployed) Tables 3a–c show the mean years of experience in the treatment and the control group, separately by age groups. The third column shows the differences Age Control Treated Difference (t-stat.) between the two groups at each age with t-statistics in the parentheses. 3 a presents the data of the whole sample, 3b for those who have at most a lower a. Full sample secondary qualification, and 3c for those with an upper secondary qualification 20 0.52 0.64 0.12 (the first differences) and the treatment variable as the (2.82) diff-in-diff (second difference) estimator. 21 0.60 0.70 0.10 The baseline specification of the multivariate model is (3.90) the following: 22 0.70 0.76 0.06 (3.04) Y = α + βTreat + ρX + γ + δ +μ + ε asri as asri a s asri, 23 0.77 0.79 0.02 (1.34) where Y is the outcome variable (current educational sta- 24 0.80 0.81 0.01 tus, employment status, or earnings) for each individual (0.76) (i). Treat is the treatment dummy, which can vary across 25 0.82 0.82 −0.01 ages (a) and year-of-survey (s). X are individual-level (−0.34) variables (gender and highest level of education, in some 26 0.83 0.87 0.03 specifications), and γ, δ, and μ are age, year of survey, (1.97) and region fixed effects. ε is the idiosyncratic error term, 27 0.85 0.82 −0.03 while α, β, and ρ are parameters to be estimated. For a similar estimation framework, see, for instance, Pischke (−1.41) (2007). b. ISCED 2 or below Table 5 shows the results of our baseline regression on 20 0.41 0.59 0.19 labour market outcomes. In columns 1 and 2 we estimate (1.89) linear probability models of the employment probability 21 0.44 0.59 0.15 on the sample of the active population. Columns 3 and (2.02) 4 show the results of regressions of log real earnings. We 22 0.58 0.71 0.13 estimate all models with age and year-of-survey fixed (1.84) effects as well as region fixed effects. While these should 23 0.54 0.68 0.14 take out much of the unobserved heterogeneity across (1.79) cohorts in educational outcomes, pre- and post-treat- 24 0.62 0.65 0.03 ment cohorts differ in their year of labour market entry, (0.33) too, which means they might have faced very different 25 0.60 0.58 −0.02 labour market conditions. This difference can easily bias (−0.19) the effect of the treatment on longer-run labour market 26 0.65 0.67 0.03 outcomes. Consequently, in columns 2 and 4 we also (0.26) include year-of labour-market-entry fixed effects. On 27 0.60 0.41 −0.18 the one hand, controlling for the year when one enters (−1.36) the labour market seems to be essential as different c. ISCED 3 demand-side factors can alter the entrants’ employment 20 0.54 0.65 0.11 probabilities and initial earnings. On the other hand, the (2.24) year of labour market entry might be considered a ’bad 21 0.62 0.71 0.09 control’ (see Angrist and Pischke 2008) as it correlates (3.25) well with years spent in schooling, which depend on the 22 0.71 0.76 0.05 reform. Moreover, year of labour market entry correlates (2.48) strongly with age and year-of-survey, which inflates the 23 0.78 0.80 0.02 (0.89) 24 0.80 0.81 0.01 The treatment variable is basically a simplified interaction term between (0.29) the age and the year-of survey, as it is shown in table 1. 25 0.82 0.85 0.03 We have tested for potential differences in composition across regions (see Bukowski 2019). Including regional fixed-effects in the regressions do (1.30) not change any of the results. 26 0.82 0.85 0.04 For people with no experience, we imputed their year of labour market (1.38) entry with their year of finishing highest education (see above). The labour market effects of the polish educational reform of 1999 Page 11 of 21 13 Table 4 (continued) we cannot compare the average finishing age for the full pre- and post-treatment sample. However, we can Age Control Treated Difference (t-stat.) compare the lowest educated sub-population, as there 27 0.84 0.82 −0.02 was no compositional change regarding this education (−0.59) group. Appendix Fig.  5 shows the first stage of our IV: Tables 4a–c show the share of employed respondents in the active population the age distribution for those with only ISCED 2 or less in the treatment and the control group, separately by age groups. The third for each pair of cohorts. The median finishing age for this column shows the differences between the two groups at each age with level was 15 until the 1985 cohort and became 16 with t-statistics in the parentheses. 4a presents the data of the whole sample, 4b for those who have at most a lower secondary qualification, and Table 4c for those the 1986 cohort. Columns 1 and 2 in Appendix Table  12 with an upper secondary qualification show that the age when the highest degree was obtained for this population has zero effect in itself on log of earn - ings and employment probability. 1st stage estimates in variance of the model. Nevertheless, substantial results of columns 3 and 5 show that the reform can act as a strong the models do not differ much with or without the year of instrument: treated cohorts are 0.4 or 0.6  years older labour market entry fixed effect. than the pre-treatment cohorts when they finish their The results in Table  5 show that the treated group is highest degree of schooling. In columns 4 and 6 we see about 3 percentage points more likely to be employed and that the 2SLS coefficients of finishing age on earnings and earn 4 to 5% higher earnings. Additional results pre- employment are high and positive, 13.6% and 12 percent- sented in Fig. 4a and b show how these average treatment age points, respectively. Unfortunately, they are insignifi - effects vary across age cohorts. The estimates are too cant, due most likely to the small power of our analysis, imprecise for employment probability to detect any sta- but they are of a very similar magnitude to the signifi - tistically significant and consistent age-related patterns. cant estimates of Table  6 in earnings. Nevertheless, However, people closer to their twenties benefited from both estimations show that the lowest educated popula- a 10 percentage points increase in their earnings after the tion would have been more likely to earn more had they reform compared to the pre-treatment people of similar attended comprehensive lower secondary schooling for age. This effect declines with age and disappears around an additional year. age 24–26. These results can be explained by arguing that the reform improved labour market entrance, but the 2 Conclusion effects disappear with age as experience becomes more We analysed the effects of the 1999 education reform on important than general skills learned at school. labour market outcomes, in particular on employment probability and earnings. This reform was comprehen - 1.4 Robustness checks sive, as it extended general education by one year before In a robustness check, we simplify our models and com- students were tracked into upper secondary education. It pare only two cohorts–right before and right after the also improved the quality of education for the low-track reform–with each other. This might decrease the power students before they were tracked and increased the gen- of our analysis substantially but also highlight the impor- eral skills of potential vocational students and those who tance of the “inserted” year of education before tracking, later chose the vocational track. We show that the reform as all other reform elements have impacted these two resulted in a 3 percentage point increase in employment cohorts similarly. Table  6 presents the same models as probability. The reform also increased earnings by 4–5% in Table  5 regressed only on the 1985 and 1986 sample, on average. The results suggest that the positive effects on so the cohorts born right before and right after the 1986 earnings decrease with age, so that the reform treatment January 1 cut-off. The effects on the employment prob - could have been beneficial in the labour market entrance, abilities are not significant; the 1986 cohort earns 9–11% but its long-term impact is relatively small. more than the 1985 cohort on average. Our research tries to answer the question of why there In another robustness check, we test the mechanisms are mixed results in the literature about the returns of by using the reform as an instrument for the finishing age education. Increasing the length of compulsory education of schooling. Unfortunately, the database does not con- or an additional year of education offered to vocational tain data on the years spent in education, but we know students usually do not help their employment chances the age when the highest educational level was attained. or increase their earnings. We show that a comprehensive As shown above, due mainly to the Bologna Process, 18 19 The effects on earnings are not different by gender. The effects on employ - As treated are only 0.4-0.6 years older than the pre-treatment cohorts, the ment probabilities are, on the other hand, greater for women than for men. causal estimate from the IV should be around 13.6*0.4 = 5.44% for earnings See Appendix Table 13 for the results by gender. and 12*0.58 = 6.96 percentage points for employment. 13 Page 12 of 21 L. F. Drucker et al. Fig. 3 Distribution of real earnings of treatment and control group in 2005 PLN. The figures show the distribution of real earnings in the treatment and the control group in 2005 Polish zloty. The first figure presents data for the whole sample, the second for those who have at most lower secondary qualification (ISCED 2), and the third for those who have upper secondary qualification (ISCED 3) The labour market effects of the polish educational reform of 1999 Page 13 of 21 13 Table 5 The effect of the reform on labour market outcomes, linear models Variables (1) (2) (3) (4) Employed vs unemployed Employed vs unemployed Log earnings Log earnings Treated 0.0314*** 0.0325*** 0.0361** 0.0514*** (0.0108) (0.0110) (0.0157) (0.0168) Female −0.0438*** −0.0517*** −0.183*** −0.183*** (0.00607) (0.00618) (0.00888) (0.00893) Constant 0.565*** 0.453*** 6.764*** 6.716*** (0.0218) (0.0457) (0.0323) (0.0672) Observations 23,274 23,274 15,499 15,499 R-squared 0.051 0.061 0.187 0.191 Age fixed-effect y y y y Year-of-survey fixed-effect y y y y Region fixed-effect y y y y Year of LM entry fixed-effect n y n y Robust standard errors in parentheses All models are linear regressions on the sample of the active population. Columns 1–2 show the results of an LPM on employment probability, while columns 3–4 show the results of the linear regression on earnings. In columns 2 and 4, labour market entry fixed effects are included, too ***p < 0.01, **p < 0.05, *p < 0.1 type of reform can successfully improve the labour mar- Finally, despite being successful in improving the edu- ket outcomes, and the effects are driven by the young and cation outcomes (as shown by the PISA study) and the the low educated (and likely by vocational students). This labour market outcomes of post-reform students (as study has its drawbacks: due to the lack of a proper edu- shown by this study, but also by Liwiński, (2020) and cation measure in the EU-SILC we can only estimate an Strawinski and Broniatowska (2021), using different average effect for all education levels. This tells us little data, methods and subgroups), after 18  years, the Pol- about the potential mechanisms driving the results, thus ish government reversed the reform in January 2017 we can only speculate that the increased general educa- and re-introduced the old system. So, from September tion and decreased tracking (better peers and teachers) 2017, students study again in the old 8 + 4/5 system. caused the effects. After 1999 students in Poland were Along with this reform, they try to improve the qual- forced to sit an additional year in less selected classes than ity of vocational education. Based on the current paper before and were taught by teachers who were less selected. and research before us, we would warn against a re- This change was likely beneficial for the low-track chil - tracking reform like this as it might not be the best way dren, as their composition of peers and teachers improved to improve the labour market conditions of vocational substantially. Our results suggest that the reform has students. reached its initial goal of decreasing inequalities. 13 Page 14 of 21 L. F. Drucker et al. Fig. 4 a The effect of the reform on employment probability and earnings by age groups. The figure shows the effects of the reform on each age cohort estimated by models 2 and 4 in Table 5, extended with treatment*age interactions. The left figure shows the treatment effect on the probability of being employed (interpreted as percentage points), while the right figure on the logarithm of earnings (interpreted as percentage change in the earnings). The controls include gender and age, year-of-survey, region, and year-of-labour-market-entry fixed effects. The bars show the average effects with 95% confidence intervals. b The effect of the reform on employment probability and earnings by age groups, without labour market entry fixed effects. The figure shows the effects of the reform on each age cohort estimated by models 1 and 3 in Table 5, extended with treatment*age interactions. The left figure shows the treatment effect on the probability of being employed (interpreted as percentage points), while the right figure on the logarithm of earnings (interpreted as percentage change in the earnings). The controls include gender and age, year-of-survey, and region fixed effects. The bars show the average effects with 95% confidence intervals The labour market effects of the polish educational reform of 1999 Page 15 of 21 13 Table 6 The effect of the reform on employment probability and earnings—cohorts 1985/1986 Variables (1) (2) (3) (4) Employed vs unemployed Employed vs unemployed Log earnings Log earnings Treated 0.0184 −0.00361 0.0876*** 0.107*** (0.0127) (0.0141) (0.0204) (0.0232) Female −0.0474*** −0.0493*** −0.171*** −0.167*** (0.0113) (0.0118) (0.0180) (0.0182) Constant 0.569*** 0.571*** 6.725*** 6.645*** (0.0400) (0.0530) (0.0961) (0.116) Observations 5082 5082 3439 3439 R-squared 0.049 0.059 0.223 0.245 Age fixed-effect y y y y Year-of-survey fixed-effect y y y y Region fixed-effect y y y y Year of LM entry fixed-effect n y n y Robust standard errors in parentheses All models are linear regressions on the sample of the active population born in 1985–1986. Columns 1–2 show the results of an LPM on employment probability, while columns 3–4 show the results of the linear regression on earnings. In columns 2 and 4, labour market entry fixed effects are included, too ***p < 0.01, **p < 0.05, *p < 0.1 Appendix A See Tables 7, 8, 9, 10, 11, 12, 13 Table 7 Summary of independent variables Variable Obs Mean Std. Dev Min Max Year of birth 48,557 1985.7 3.60 1978 1994 Female 48,557 0.48 0.50 0 1 Experience 26,697 2.27 2.26 0 12 Age when the first job began 23,186 20.28 2.47 8 27 Age 48,557 22.87 2.57 19 27 Level of educ.: low 43,057 0.18 0.39 0 1 Level of educ.: medium 43,057 0.65 0.48 0 1 Level of educ.: high 43,057 0.17 0.37 0 1 Treated 48,557 0.52 0.50 0 1 Gross real earning (PLZ) 16,762 1473 795 15 16,939 Labour market status At work 48,557 0.44 0.50 0 1 Unemployed 48,557 0.12 0.33 0 1 Retired 48,557 0.00 0.03 0 1 Inactive 48,557 0.43 0.50 0 1 13 Page 16 of 21 L. F. Drucker et al. Table 8 Number of observations in each age/year-of-survey cell Age Year of survey Total 2005 2006 2007 2008 2009 2010 2011 2012 2013 19 929 814 679 657 626 586 532 497 529 5849 20 843 821 732 623 572 569 538 499 485 5682 21 860 730 732 655 523 528 498 511 464 5501 22 926 773 687 655 546 539 518 496 486 5626 23 812 801 665 658 573 491 503 469 424 5396 24 746 708 732 618 582 523 495 489 420 5313 25 790 644 618 640 558 515 494 479 467 5205 26 708 676 587 555 567 534 499 495 465 5086 27 675 599 603 531 488 527 515 501 460 4899 Total 7289 6566 6035 5592 5035 4812 4592 4436 4200 48,557 Table 9 Number of observations in each year-of-birth/year-of-survey cell Year of birth Year of survey 2005 2006 2007 2008 2009 2010 2011 2012 2013 Total 1978 675 0 0 0 0 0 0 0 0 675 1979 708 599 0 0 0 0 0 0 0 1307 1980 790 676 603 0 0 0 0 0 0 2069 1981 746 644 587 531 0 0 0 0 0 2508 1982 812 708 618 555 488 0 0 0 0 3181 1983 926 801 732 640 567 527 0 0 0 4193 1984 860 773 665 618 558 534 515 0 0 4523 1985 843 730 687 658 582 515 499 501 0 5015 1986 929 821 732 655 573 523 494 495 460 5682 1987 0 814 732 655 546 491 495 479 465 4677 1988 0 0 679 623 523 539 503 489 467 3823 1989 0 0 0 657 572 528 518 469 420 3164 1990 0 0 0 0 626 569 498 496 424 2613 1991 0 0 0 0 0 586 538 511 486 2121 1992 0 0 0 0 0 0 532 499 464 1495 1993 0 0 0 0 0 0 0 497 485 982 1994 0 0 0 0 0 0 0 0 529 529 Total 7289 6566 6035 5592 5035 4812 4592 4436 4200 48,557 The labour market effects of the polish educational reform of 1999 Page 17 of 21 13 Table 10 Number of observations and row percentage of activity status and current education activity Basic activity status Current education activity Total Not in education In education Missing At work 13,958 4450 3047 21,455 % 65 21 14 100 Unemployed 5027 424 549 6000 % 84 7 9 100 In retirement or early retirement 18 36 1 55 % 32.73 65.45 1.82 100 Other inactive 3219 15,996 1832 21,047 15.29 76 8.7 100 Total 22,222 20,906 5429 48,557 % 45.76 43.05 11.18 100 Table 11 Linear regressions on the age of finishing education Variables (1) (2) (3) (4) Full sample Currently not in education Age of finishing ed Age of finishing ed Age of finishing ed Age of finishing ed Treated 0.0164 0.0900*** −0.0208 0.184*** (0.0480) (0.0285) (0.0674) (0.0426) Low ed −3.824*** −3.847*** (0.0490) (0.0518) High ed 3.946*** 4.297*** (0.0377) (0.0369) Treat* low ed 0.806*** 0.494*** (0.0590) (0.0757) Treat* high ed −0.785*** −0.956*** (0.0654) (0.0817) Female 0.548*** −0.0444*** 0.749*** −0.0142 (0.0247) (0.0148) (0.0377) (0.0232) Constant 17.62*** 18.95*** 17.68*** 18.77*** (0.0502) (0.0280) (0.0817) (0.0533) Observations 37,569 37,569 21,510 21,510 R-squared 0.188 0.724 0.138 0.731 Age fixed-effect y y y y Year-of-survey fixed-effect y y y y Robust standard errors in parentheses *** p < 0.01, **p < 0.05, *p < 0.1 13 Page 18 of 21 L. F. Drucker et al. Table 12 The effect of the reform on employment probability and earnings—IV estimation for people with ISCED2 or less Variables (1) (2) (3) (4) (5) (6) OLS 1st stage 2sls 1st stage 2sls log earning Employed log real earning Employed Age when highest education −0.00307 0.00633 0.136 0.120 was attained (0.0150) (0.00923) (0.194) (0.0768) Treated 0.401** 0.583*** (0.188) (0.123) Female −0.171*** −0.178*** 0.186 −0.202*** 0.166** −0.201*** (0.0615) (0.0298) (0.143) (0.0649) (0.0805) (0.0362) Constant 6.422*** 0.361** 15.46*** 4.829* 15.22*** −1.247 (0.247) (0.158) (0.295) (2.865) (0.189) (1.115) Observations 914 1917 914 914 1917 1917 R-squared 0.174 0.101 0.151 0.079 0.115 0.009 Region fixed-effect y y y y y y Age fixed-effect y y y y y y Year-of-survey fixed-effect y y y y y y Robust clustered standard errors in parentheses Columns 1–2 show the association of the age when the highest degree was obtained with log earning and with employment probability, respectively for those who finished at most ISCED 2 level. Columns 3–4 and 5–6 show the 2SLS estimations of log real earning and employment probability with the treatment used as an instrument for the age when they finished the ISCED 2 level *** p < 0.01, **p < 0.05, *p < 0.1 Table 13 The effect of the reform on employment probability and earnings separately by gender Variables (1) (2) (3) (4) Employed vs Employed vs log earnings log earnings unemployed unemployed Treated 0.0230* 0.0210* 0.0377** 0.0514*** (0.0121) (0.0121) (0.0165) (0.0175) Female −0.0506*** −0.0607*** −0.182*** −0.183*** (0.00710) (0.00715) (0.0116) (0.0116) Treated* Female 0.0184 0.0244* −0.00325 0.000110 (0.0136) (0.0136) (0.0187) (0.0187) Constant 0.569*** 0.457*** 6.763*** 6.716*** (0.0219) (0.0460) (0.0322) (0.0671) Observations 23,274 23,274 15,499 15,499 R-squared 0.051 0.061 0.187 0.191 Age fixed-effect y y y y Year-of-survey fixed-effect y y y y Region fixed-effect y y y y Year of LM entry fixed-effect n y n y Robust standard errors in parentheses ***p < 0.01, **p < 0.05, *p < 0.1 The labour market effects of the polish educational reform of 1999 Page 19 of 21 13 Figure 5 Fig. 5 Distribution of ages when highest educational level was attained by pairs of cohorts for students with ISCED 2 or below. The coloured bars show the distribution of the ages of the second birth cohort, while the empty bars show that of the first birth cohort in each figure. The median age of schooling was 15 for all birth cohorts until 1985 and became 16 for all birth cohorts starting from 1986 Figure  7 compares people of different ages in the same Appendix B survey year, so as we move to the right in the figure, the Unfortunately we cannot do a proper common trend cohorts presented become younger. Thus, none of these analysis with our data. As this is a quasi-panel—of age- graphs is informative enough alone. The only way to of-respondent and year-of-observation—, we could compare pre-and post-reform cohorts is by accounting either compare people of similar age in different survey for both dimensions in a regression framework—includ- years or people of different ages in the same survey year. ing both age and year-of-observation as fixed-effects; and Figure  6 compares age cohorts in different survey years. this is what we do in the paper. The increase in the log wages as we move to the right in Figures 6, 7 the graph is therefore partly caused by economic trends. 13 Page 20 of 21 L. F. Drucker et al. Fig. 6 Log wage by age groups before and after the reform Fig. 7 Log wage by survey-year groups before and after the reform The labour market effects of the polish educational reform of 1999 Page 21 of 21 13 Acknowledgements Jakubowski, M., Harry, P., Emilio, P., Jerzy, W.: The effects of delaying tracking in We would like to thank Pawel Bukowski, Adam Booij, Eva Holb, Karolina secondary school: evidence from the 1999 education reform in Poland. Kurpaska, Heiko Rachinger, Balázs Reizer and Ágnes Szabó-Morvai, and two Educ. Econ. 24(6), 557–572 (2016) anonymous referees as well as the participants at the Education, Skills, and Jakubowski, M.: Poland: polish education reforms and evidence from interna- Labor Market Outcomes workshop in Oslo 2017, the meeting of the Hungar- tional assessments. In: Crato, N. (ed.) Improving a country’s education. ian labour economists at Szirák 2015, at the annual 2015 meeting of the Hun- Springer, Cham (2021). https:// doi. org/ 10. 1007/ 978-3- 030- 59031-4_7 garian Society of Economists, and seminar participants at the CERS-HAS for Jung-Miklaszewska, J.: The system of education in the Republic of Poland. their helping comments. The usual disclaimer applies. The authors gratefully Bureau for Academic Recognition and International Exchange, Warsaw acknowledge funding from the Hungarian Scientific Research Fund (Project (2003) no. 109338) and the Horizon 2020 Twinning grant EdEN (Project no. 691676). Kerr, P., Sari, T.P., Uusitalo, R.: School tracking and development of cognitive skills. J. Law. Econ. 31(3), 577–602 (2013) Kwiek, M.: Social perceptions versus economic returns of the higher Declarations education:the Bologna process in Poland. In: Studien zur international vergleichenden Erziehungswissenschaft. Schwerpunkt Europa - Studies Competing interests in International Comparative Educational Science. Focus: Europe. 28 The authors have no conflicts of interest to declare that are relevant to the November 2013. pp 147–182 (2014) content of this article. The authors gratefully acknowledge funding from the Liwiński, L.: The impact of compulsory schooling on Hourly Wage: evidence Hungarian Scientific Research Fund (Project no. 109338) as well as the Horizon from the 1999 education reform in Poland. Eval. Rev. 44(5–6), 437–470 2020 Twinning grant EdEN (Project no. 691676). All authors contributed (2020) equally to the paper. The paper uses the European Union Statistics on Income Malamud, O., Pop-Eleches, C.: General education versus vocational training: and Living Conditions (EU-SILC) database that can be accessed from Eurostat evidence from an economy in transition. Rev. Econ. Stat. 92(1), 43–60 at https:// ec. europa. eu/ euros tat/ web/ micro data/ europ ean- union- stati stics- (2010) on- income- and- living- condi tions. Meghir, C., Palme, M.: Educational reform, ability, and family background. Am. Econ. Rev. 95(1), 414–424 (2005) Author details Oosterbeek, H., Webbink, D.: Wage effects of an extra year of basic vocational 1 2 Central European University, Budapest, Hungary. Institute of Econom- education. Econ. Educ. Rev. 26(4), 408–419 (2007) ics, Centre for Economic and Regional Studies and Institute of Economics, Pekkarinen, T., Uusitalo, R., Kerr, S.: School tracking and intergenerational Corvinus University of Budapest, Budapest, Hungary. Faculty of Economic income mobility: evidence from the finnish comprehensive school Sciences, University of Warsaw, Warsaw, Poland. reform. J. Public. Econ. 93(7–8), 965–973 (2009) Pischke, J.-S.: The impact of length of the school year on student performance Received: 28 September 2021 Accepted: 29 July 2022 and earnings: evidence from the german short school years. Econ. J. 117(523), 1216–1242 (2007) Pischke, J.-S., von Wachter, T.: Zero returns to compulsory schooling in Ger- many: evidence and interpretation. Rev. Econ. Stat. 90(3), 592–598 (2008) Psacharopoulos, G., Patrinos, H.A.: Returns to education: a decennial review of References the global literature. Educ. Econ. 26(5), 445–458 (2018) Aakvik, A., Salvanes, K.G., Vaage, K.: Measuring heterogeneity in the returns Rivkin, S.G., Hanushek, E.A., Kain, J.F.: Teachers, schools, and academic achieve- to education using an education reform. Eur. Econ. Rev. 54(4), 483–500 ment. Econometrica 73(2), 417–458 (2005) (2010) Sacerdote, B.: Peer effects in education: how might they work, how big Angrist, J.D., Pischke, J.-S.: Mostly harmless econometrics: an empiricist’s com- are they and how much do we know thus far? Handb. Econ. Educ. 3, panion. Princeton University Press (2008) 249–277 (2011) Becker, GS.: “Human capital: a theoretical and empirical analysis, with special Strawiński, P., Broniatowska, P.: The impact of prolonging compulsory general reference to education”. National Bureau of Economic Research (1975) education on the labour market. Educ. Stud. (2021). https:// doi. org/ 10. Bialecki, I., Johnson, S., Thorpe, G.: Preparing for national monitoring in Poland. 1080/ 03055 698. 2021. 19576 70 Assess. Educ. Princ. Policy. Pract. 9(2), 221–236 (2002) Bukowski, P.: How history matters for student performance lessons from the Publisher’s Note partitions of Poland. J. Comp. Econ. 47(1), 136–175 (2019) Springer Nature remains neutral with regard to jurisdictional claims in pub- Chetty, R., Friedman, J.N., Rockoff, J.E.: Measuring the impacts of teachers I: lished maps and institutional affiliations. evaluating bias in teacher value-added estimates. Am. Econ. Rev. 104(9), 2593–2632 (2014) European Commission. Focus on the Structure of Higher Education in Europe 2004/05 - Teaching - EU Bookshop. European Commission, Directorate- General for Education, Youth, Sport and Culture (2005) Grenet, J.: Is extending compulsory schooling alone enough to raise earnings? Evidence from French and British compulsory schooling laws*. Scand. J. Econ. 115(1), 176–210 (2013) GUS: Oświata i wychowanie w roku szkolnym 2005/2006. Central Statistical Office, Poland (2006) Hall, C.: The effects of reducing tracking in upper secondary school evidence from a large-scale pilot scheme. J. Hum. Res. 47(1), 237–269 (2012) Hall, C.: Does more general education reduce the risk of future unemploy- ment? Evidence from an expansion of vocational upper secondary education. Econ. Educ. Rev. 52, 251–271 (2016) Hanushek, E.A., Woessmann, L.: Do better schools lead to more growth? Cognitive skills, economic outcomes, and causation. J. Econ. Growth. 17, 267–321 (2012) Heckman, J.J., Humphries, J.E., Veramendi, G.: The nonmarket benefits of education and ability. J. Hum. Cap. 12(2), 282–304 (2018). https:// doi. org/ 10. 1086/ 697535 Jakubowski, M.: "Opening up opportunities: education reforms in Poland". IBS Policy Papers (2015) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Journal for Labour Market Research Springer Journals

The labour market effects of the polish educational reform of 1999

Loading next page...
 
/lp/springer-journals/the-labour-market-effects-of-the-polish-educational-reform-of-1999-lb6yUXmTKD
Publisher
Springer Journals
Copyright
Copyright © The Author(s) 2022
ISSN
1614-3485
eISSN
2510-5027
DOI
10.1186/s12651-022-00317-4
Publisher site
See Article on Publisher Site

Abstract

We estimate the effect of the 1999 education reform in Poland on employment and earnings. The 1999 education reform in Poland replaced the previous 8 years of general and 3/4/5 years of tracked secondary education with 9 years of general and 3/3/4 years of tracked upper-secondary education. The reform also introduced new curricula, national examinations, teacher standards, and a transparent financing scheme. Our identification strategy relies on a differ - ence-in-differences approach using a quasi-panel of pooled year-of-survey and age-of-respondent observations from the Polish sample of the EU-SILC database. The results indicate that the reform has increased employment probability (by around 3 percentage points) and earnings (by around 4%). Keywords: Education reform, Returns to education, Poland, Detracking, Labour market, Difference-in-differences JEL Classification: I21, I24, I26, J24 time, the earnings of those with highly educated parents 1 Introduction decreased, but the average effect was positive. In this study, we look at the Polish education reform of The reform in Finland in the 1970s delayed the tracking 1999 that, among others, has increased general education of students from age 11 to 16 by extending comprehen- for all students, introduced a national core curriculum, sive education from 4 to 9  years. Pekkarinen, Uusitalo, standardised examinations at the end of each education and Kerr tested the effects of this reform on the income stage, introduced new teacher qualification require - elasticity (Pekkarinen et  al., 2009) and the average test ments, and adopted a transparent financing scheme. scores (Kerr et  al., 2013) and concluded that it had only Previous results on the effect of education reforms a small but overall positive effect on both dimensions. In on labour market outcomes are mixed. Comprehensive Norway, a school reform implemented between 1960 and education reforms in Scandinavia are usually hailed for 1972 increased compulsory education from 7 to 9 years. their positive effects on labour market outcomes. Swe - The 9  years of comprehensive education consisted of den carried out a major education reform in the 1950s, 6  years of primary school and a 3  year lower second- increasing compulsory education from 7/8 to 9  years, ary school. Lower secondary schools existed before the abolishing tracking (i.e. sorting students into general reform, too, but attending them was voluntary and they and vocational tracks) based on academic achievement were only available in some municipalities. This reform after the sixth grade, and introducing a national curricu- also improved the quality of education by standardising lum. Meghir and Palme (2005) showed that this reform the curriculum. Aakvik et al. (2010) find that the reform increased educational attainment and the future earn- increased the level of education in the country, and it also ings of children with low educated parents. At the same increased the returns to upper secondary and tertiary qualifications. However, some papers find no-returns to extending compulsory education in Germany (Pischke *Correspondence: drucker_luca-flora@phd.ceu.edu and von Wachter 2008) or France (Grenet 2013), and Central European University, Budapest, Hungary the story is similar if we look at education reforms that Full list of author information is available at the end of the article © The Author(s) 2022. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/. 13 Page 2 of 21 L. F. Drucker et al. increased the time of education particularly for voca- shortened by one year so that the overall length of pri- tional students: no (substantial) effects were found for mary and secondary education remained 12 years, except the Netherlands (Oosterbeek and Webbink 2007). in the basic vocational track, which was 3  years before These differences in the results might stem from the the reform and remained 3 years. differences in the type of education extended. While in Besides these structural changes, the reform changed Germany and France, the increase in education was after other major parts of the education system: Poland intro- tracking, in the Anglo-Saxon (Grenet 2013) and Scan- duced a national core curriculum, standardised examina- dinavian countries, where a positive effect of additional tions at the end of each education stage, introduced new years of schooling was found, it was before tracking. teacher qualification requirements, and adopted a trans - We argue that the reason for this is that “before track- parent financing scheme. Using PISA data, Jakubowski ing reforms” not only extended education but also (2015) showed that the between-school variance of test improved the quality of inputs, such as the curriculum, scores decreased after the reform. Jakubowski et  al. teachers, or peers, as these are crucial in the education (2016) also showed, using difference-in-differences production function (e.g. Chetty et al. 2014; Rivkin et al. approach adjusted with propensity score matching, that 2005; Sacerdote 2011). When additional years of edu- the fast increase in Polish students’ PISA test scores was cation are "inserted" before students are selected into driven by potential vocational students, who after the tracks, the effect is different from when they are already 1999 reform were in the comprehensive lower second- selected into academic or vocational tracks. ary school instead of already having been tracked into the Offering increased general training to vocational stu - low-tier vocational schools. They argue that the increased dents does not affect teachers and peers, as these inputs resources of an additional year of better teachers, peers, are unlikely to change due to the increased length of and general curriculum have improved these students’ vocational education or increased general content. test scores. They also show that the improvements in However, if the system is ’de-tracked’, i.e. the age of selec- student achievement at the age of 15 are still substantial tion is increased, the composition of teachers and peer at the age of 16 or 17 when students are already tracked groups will remain as they were before selection in that to different secondary school programs. These results “inserted” year, and thus–at least for the later vocational confirm our expectation that improved inputs result not track students–a higher peer and teacher effect might only in better achievement but provide a stronger basis improve their long term outcomes. for further development of foundational skills among stu- The 1999 reform of Poland, which we analyse in this dents who continue education in vocational tracks. study, was a comprehensive one like the Scandinavian Additional years of education translate into higher reforms. However, it was unique in the sense that besides wages (Psacharopoulos, Patrinos, 2018). This effect is extending comprehensive education, it altered all of the partly explained by the human capital theory, which above-mentioned essential inputs in the education pro- assumes that better education translates into skills that duction function. This reform replaced the former 8 year increase our productivity (Becker, 1975). An alternative general primary school with a 6 year general primary and explanation is that by passing through the education sys- a 3  year general lower secondary school. Therefore, the tem, one signals its innate ability and employers reward age of first selection was postponed from age 15 to 16. it with higher wages (see Page, 2010, for a review). I our The length of tracked upper-secondary education was case, both theories imply that the Polish reform should increase wages as it improved skills but also educational attainment. Research also shows that at the country level the improvement of cognitive skills translates into higher A counterexample is presented by Malamud and Pop-Eleches (2010), who economic growth (Hanushek, Woessmann, 2012). More- find no effect before tracking. over, better skills bring non-economic benefits (Heckman Improving only the curriculum, however, might not be sufficient. In Swe- et al., 2018). den in the 1990’s, academic content in all vocational tracks was increased. This change reduced curriculum differences between the academic and The empirical research documents well that the Pol- vocational tracks at the upper secondary level, also allowing vocational ish reform was beneficial for students’ school perfor- graduates to apply to university. The reform was preceded by a pilot mance, but we are aware of only a couple of attempts scheme, introducing the new system in selected schools throughout the country. Using this time and spatial variance Hall tested the effect of this to assess its long-term impacts on labour market out- reform on tertiary enrolment and wages (Hall 2012) and on unemployment comes. Liwiński (2020) applies a regression discon- chances (Hall 2016). The results of these analyses have shown no difference tinuity design (RDD) focusing on the differences in between pre- and post-treatment cohorts in outcomes. Hall argues that a potential reason for the non-effect is the increased dropout rate of the voca- employment probability and wages among adults with tional track students induced by the academic content. basic vocational degrees. He found that the reform At least in the short run, and thus, the typically utilised regression-discon- improved hourly wages by 13%. However, as the tinuity approach shows no effects. The labour market effects of the polish educational reform of 1999 Page 3 of 21 13 reform has also changed patterns of selection into dif- 1.1 The educational reform of 1999 in Poland ferent upper secondary tracks, with a decreasing num- The 1999 educational reform was one of the four ber of students attracted to basic vocational schools, reforms—social security, healthcare, public administra- the results comparing just adults with basic vocational tion, and education—implemented by the government training might provide a biased picture of the over- elected in 1997. The three main goals of the education all effect of the reform. Strawinski and Broniatowska reform were to increase the level of education in soci- (2021) applied a similar RDD approach to the Polish ety, provide equal educational opportunities to every- LFS data, but limiting their sample to people with sec- one, and improve the quality of education (Bialecki et al. ondary education degree only. Their result suggests a 2002). While the other three reforms were implemented rather small but positive overall impact of the reform. simultaneously, they similarly affected all students and They provide additional results showing larger ben- all adults in the population. The education reform was efits for vocational school graduates in rural areas. rolled out so that the older cohorts followed the old sys- However, these results might be biased for the same tem, while only the younger cohorts, born from 1986, fol- reasons as the analysis of Liwiński (2020) as a larger lowed the new system with the new structure, curricula, number of students in the post-reform cohorts contin- and standardised national exams. ued education into tertiary level. The reform consisted of many distinct parts, of which These papers use RDD and rely on the assumption we concentrate mainly on the structural changes but that cohorts before and after the reform are as if they briefly discuss all parts here (see Jakubowski, 2021, for a were randomly allocated. This is a strong assumption. more detailed discussion of the reform components). In Typically, RDD studies use high-density data around sum, the 1999 reform (1) extended comprehensive gen- the cut-off, for these studies, this is clearly not the case eral education by one year, and (2) changed the struc- (but see Grenet 2013 for a similar approach). Thus, ture of general education by dividing the previous 8 year using a quasi-DID identification strategy could offer a primary school into 6  years of primary and 3  years of strong robustness check to the RDD results. This study general lower secondary school. It also (3) shortened aca- aims at filling this gap by adding more evidence on demic and technical upper secondary education by one the labour market outcomes of the reform. We show year, but the basic vocational school remained 3-years- that the reform had a significant positive effect on the long. Besides these structural changes, the reform (4) earnings and employment chances of the post-reform introduced a core curriculum, which guided teachers in cohorts. We argue that these results come from the determining their syllabi, giving them more autonomy, comprehensive nature of the reform, i.e. general educa- and it (5) introduced the teacher professional attainment tion was extended, and its quality was improved before ladder with four levels and strong incentives for profes- students were tracked to upper secondary schools. sional development and obtaining master degrees for all Unlike the Scandinavian reforms, the Polish reform teachers. The third line of changes concerned the testing did not have a pilot stage; instead, it was introduced and admission system: (6) new, standardised tests were simultaneously in the whole country. Thus, our esti- introduced at the end of each education stage (primary, mations cannot exploit any time or spatial variation. lower secondary, and upper secondary), and parallel to Due to the lack of high-frequency micro-data before that, (7) the admission system to each stage shifted from and after the reform, it is impossible to use a regres- entrance examinations to using the results of these end- sion discontinuity design. However, by pooling sev- of-stage standardised exams. Finally, (8) the reform also eral years of cross-sectional surveys, we can generate transferred ownership of nearly all schools to local gov- a quasi-panel of year-of-survey and age brackets, ernments and introduced a new per pupil formula-based which we will use to estimate difference-in-differences financing system. models. Unlike in the Scandinavian studies, the vari- Figure 1 shows the structural changes in the Polish edu- ance does not come from the time of implementation cation system. Most importantly, the newly established but from the time (year) of observation. This way, we comprehensive lower secondary schools (gimnazjum will directly compare the employment chances and in Polish) had to follow the same general curriculum real earnings of pre-reform (control) and post-reform and admit all students from their catchment area with- (treatment) cohorts. In Sect.  2, we discuss the 1999 out any additional requirements. They could accept reform in more detail. Section  3 presents the data and additional applicants from other areas to the remaining descriptive analyses. In Sect.  4, we present our esti- places. Lower secondary schools were larger, and there mation methodology and baseline results. Section  6 were fewer of them than primary schools. They opened shows robustness checks, and Sect. 7 concludes. 13 Page 4 of 21 L. F. Drucker et al. Fig. 1 Pre- and post-reform structure of the Polish education system. The figure shows the structure of the Polish education system before and after the reform. The width of the specific upper-secondary level tracks approximates the ratio of students attending these tracks in 2000 (pre-reform) and 2004 (post-reform). Source of data: Polish Statistical Office in larger settlements, which was especially important in comprehensive education, the end of compulsory educa- rural areas, where one gimnazjum collected the children tion increased to age 18. from neighbouring villages who previously attended dif- Finally, Poland introduced a new curriculum and stand- ferent local primary schools. So, students spent their last ardised final exams to measure whether students achieve 3  years of general education in a comprehensive institu- the curricular goals at the end of the 6th and 9th grades. tion intended to give the same high-quality education for The 6th-grade test was low-stakes and provided informa - students all over the country. tion on student and school performance. The 9th-grade Before the reform, there were three upper secondary tests were high-stakes and were used as an entrance exam tracks: a 4  year academic secondary school (liceum), a to the upper secondary stage. These exams were first 5  year technical secondary school (technikum), and a launched in 2002 for the first cohorts that followed the new 3  year basic vocational school. Only liceum and techni- curriculum in the new comprehensive lower secondary kum ended with a maturity exam, which did not change schools. For the same cohort, the new standardised matu- with the reform. After the reform, academic and techni- rity exams were introduced in 2005 and replaced higher cal secondary schools became one year shorter, but the education admission exams. The standardised examina - basic vocational schools remained 3  years long. There - tions strengthened the impact of the extended general fore, lower secondary graduates and those who chose the education as they were compulsory for all students. vocational track received one additional year of compre- Overall, the reform revolutionised all parts of the hensive education. A new institution was operating for school system. However, one can reasonably expect a short time, the so-called profiled academic secondary that these changes differently affected various groups of school, but it was abolished after a few years. In general, students and had different impacts over time. One can the upper secondary education curricula and programs expect a delayed effect on student outcomes of changes were not changed substantially. In academic and techni- in curriculum, professional teacher standards, or system cal upper secondary schools the curricula were adjusted governance structure. Even the reformers assumed that where necessary as parts of the material taught in first several years were needed to benefit from these changes. grades were now moved to general lower secondary However, one year’s general education extension schools. Before 1999, education was compulsory until age 17 Compulsory age of schooling was increased from 17 to 18 in 1997 in arti- cle 70 of the Polish Constitution. This change affected all cohorts already in with the possibility of part-time education after finish - school, i.e. both pre- and post-treatment cohorts (see Jung-Miklaszewska ing the 8  year primary school. With the extension of 2003). The labour market effects of the polish educational reform of 1999 Page 5 of 21 13 immediately provided students with one more year of and health status data at the personal and household the general curriculum. The new accountability system level. The data consists of private households with all established by new standardised examinations provided household members surveyed, but only above age 16 are additional incentives to students and teachers to cover interviewed personally for the income data. In this paper, the general curriculum. However, that made a difference we use the cross-sectional database of EU-SILC. only to students who would otherwise have gone to the To generate a balanced “quasi-panel”, we pool the cross- basic vocational schools. Before the reform, the other sectional datasets between 2005 and 2013 and restrict groups had to follow the academic curriculum and pass the sample to participants between ages 20 and 27. This secondary school entrance examinations. allows us to compare the pre-reform and post-reform The evaluation of changes in student achievement con - groups at the same age: in 2005, the youngest control firms that the main group that the reform affected imme - group members were 20 years old, and in 2013, the oldest diately was the students of basic vocational schools. treatment group members were 27 (see Table  1 below). Jakubowski et  al. (2016) demonstrate that the immediate This means we have 16 cohorts in the sample, eight in the impact of the reform on 15-year-olds’ achievement was close treatment (T86 to T93–where 86, 87, etc., signal the birth to one standard deviation for students whose socioeconomic year of the cohort) and eight in the control group (C78 background is identical to former students of basic voca- to C85). These people were born between 1978 and 1993 tional schools. Moreover, additional comparisons for 16- and (see Tables 8 and 9 in the Appendix). There are around 48 17-year-old students suggest that this effect is long-lasting. 500 observations in the sample, with 23 500 in the con- After 1–2  years of basic vocational education, students still trol group and 25 000 in the treatment group. In Poland, show improved reading and mathematics outcomes close to the school starting age is 7, and the threshold is January an equivalent of one year of instruction. A similar achieve- 1. In the sample, everyone born after January 1 1986 is ment effect for students in the academic track was negligible considered to be treated—to have studied in the new sys- even 6 years after the reform was implemented. tem—and everyone born until December 31 1985, is con- As the new comprehensive lower secondary schools sidered to be in the control group. There is a scope for started to operate already in 1999/2000, the first cohort that misclassification around the threshold because of grade started their 7th grade in the new system is the 1986 cohort. retention or skipping a grade. However, grade repetition So, we categorise everyone born after 1986 as a member of was always rare in Polish schools and was mainly limited the treated group. We argue that structural changes imme- to students with special needs. The earliest data on grade diately affected every cohort from 1986 onwards, while the repetition are available for 2005, and they show that in other changes were beneficial in the long run, affecting both primary schools, only 0.6% of students were repeating a pre- and post-treatment cohorts, although to a slightly dif- grade (GUS, 2006). ferent extent. So, if we compare a few cohorts before and For educational attainment, we rely on the ISCED clas- after the reform, the potential differences in their labour sification : those with ISCED 2 (lower secondary) qualifi - market outcomes are most likely caused by the additional cation or below are considered low-educated, those with year of general education before the delayed tracking. ISCED 3 (upper secondary) or 4 (post-secondary non- tertiary) are at the medium level, and those with ISCED 5 1.2 Da ta and descriptive statistics (tertiary) are highly educated. We see the highest educa- We use the EU Statistics on Income and Living Condi- tion level attained in the data and the year when it was tions (EU-SILC) data between 2005 and 2013. The EU- achieved, but unfortunately, we do not observe the exact SILC contains detailed income and labour, education, type of school the person graduated from. So, in the case of upper secondary education, we see how old the per- son was when she finished the upper secondary level, but School enrollment cut-off date in Poland is January 1. we do not see whether it was a liceum, a technikum, or a 6 th For example, the first 6 grade standardised exams took place in 2002, so basic vocational school. Figure  2 shows the distribution the 1986 and 1987 cohorts did not have to take them yet. of the ages when each level of education was achieved, Previous studies looking at the effect of the Polish education reform of 1999 have used the Polish LFS data (Liwiński 2020, Strawinski and Bronia- separately for the treatment and control group. The dis - towska 2021). We have also planned our initial analysis on the Polish LFS, tribution of finishing ages is different for all three educa - however when we asked the Polish Statistical Office for the official data- tion levels for the two groups. The most visible change is base, their offer has vastly exceeded our budget. So we decided to use the EU-SILC instead, which was freely available through the Eurostat. Admit- tedly, the national LFS is better in its level of education measure as the SILC. However—as we point out below—education is a ‘bad control’ in this reform. We also admit that the number of observations is higher in the LFS 8 The summary of the variables can be found in the Appendix (Table 7). than in the SILC, however, earnings are much better recorded in the EU- Note that each figure shows only those people who have that particular SILC. us we b Th elieve that the SILC is not an inferior option compared to education level as the highest finished level at the date of the survey. LFS for this analysis. 13 Page 6 of 21 L. F. Drucker et al. Table 1 Distribution of treatment and control group cohorts by age and year of survey Year of survey 2005 2006 2007 2008 2009 2010 2011 2012 2013 Age 20 C85 T86 T87 T88 T89 T90 T91 T92 T93 21 C84 C85 T86 T87 T88 T89 T90 T91 T92 22 C83 C84 C85 T86 T87 T88 T89 T90 T91 23 C82 C83 C84 C85 T86 T87 T88 T89 T90 24 C81 C82 C83 C84 C85 T86 T87 T88 T89 25 C80 C81 C82 C83 C84 C85 T86 T87 T88 26 C79 C80 C81 C82 C83 C84 C85 T86 T87 27 C78 C79 C80 C81 C82 C83 C84 C85 T86 The table shows the distribution of the treatment and control cohorts in our sample. The numbers in the cells indicate the birth year of the cohort that falls into that year-of-survey-age cell. The letters C and T show if the cohort belongs to the control group or the treatment group For the number of observations, see the Appendix, Table 8 in the median finishing age of the lowest educated: the cohorts, but this average zero effect masks a significant median age increased from 15 to 16, mainly because of composition effect: the low educated stay about 0.9 years the one additional year of comprehensive education. longer in school, while the average upper secondary The difference for upper secondary graduates (ISCED graduate stays a little over 1  month (0.09  years) longer 3) is not so salient, but we also see the distribution shift in school. This average effect for the medium educated to the right, driven by the one extra year for vocational is probably due to the ca. 15% of students in basic voca- students. The distribution of tertiary (ISCED 5) gradu - tional tracks, who stay about one year longer in school. ation age became bimodal, as, in the Bologna system, a On the other hand, higher educated people finish educa - BA degree also counts as a tertiary qualification, while tion about 0.7 years earlier, which is probably due to the earlier only long-cycle programmes existed. Note that previously non-existent BA degree. When looking only studying the differences between treated and non-treated at those who are currently not pursuing any education– cohorts in this dimension is not straightforward. Poland which means that they have finished their educational signed the Bologna Declaration in 1999 and 28 other career, at least for a while–the pattern is similar. How- European countries, which introduced the typical three- ever, effect sizes are slightly different: low educated stay level system of tertiary education—bachelor, master, and in school 0.7  years longer, medium level educated over doctorate. The three-cycle structure of higher education 2 months longer and higher educated about 0.8 years less. was implemented gradually. In 2003 Poland was already The two main outcome variables we are interested in implementing the Bologna structure (Kwiek 2014). By are employment status and earnings. EU-SILC classifies the school year 2004/2005 10% of state higher educa- activity status into four categories: at work, unemployed, tion institutions had already adopted the 2-cycle model in retirement or early retirement, and other inactive. The (Bachelor and Master) in all fields of study and 50% of the first category covers those who work either full-time or institutions in at least 50% of the fields of study (see Euro - part-time or are self-employed full-time or part-time. pean Commission 2005). In 2008 all tertiary students When looking at employment chances, we compare were enrolled in the Bologna system (see Kwiek 2014). employed people to unemployed, as we find that the Consequently, the first bachelor-level graduates of share of the active population compared to the inactive the new system entered the labour market in around did not change during the sample period. We drop those 2006. Therefore, the 1984 and 1985 cohorts also had the in retirement or in early retirement as there are only 55 of opportunity to study in the Bologna system, depending these people in our sample. on their institution and field of study–however, the first Income data is collected as gross current monthly full "Bologna cohort" was the 1989 cohort (who finished earnings before the deduction of taxes and social insur- in 2011). This coincidence makes it hard for us to study ance contributions. Income is given in Euros and cur- the effect of the reform on the upper end of the education rent prices, so we converted this data to Polish Złoty in distribution. 2005 prices. The database contains data on experience, Appendix Table 11 expl or es the age of finishing educa - expressed as the number of years spent as an employee or tion in a regression framework. Treated cohorts, on aver- self-employed since the respondent first started a regular age, tend to stay just as long in education as the control job. As low educated people studied for one more year The labour market effects of the polish educational reform of 1999 Page 7 of 21 13 Fig. 2 Distribution of the age when the highest educational level was attained. The figures show the distributions of the age when the respondents finished their highest education level. Each figure corresponds to a subsample with a specific education level: those with at most lower secondary education (ISCED 2), upper secondary (ISCED 3), and tertiary education (ISCED 5) 13 Page 8 of 21 L. F. Drucker et al. after the reform, and this change was immediate, there Table 2 Ages when control and treatment group members started their first job was a year (2001) when, in theory, no one graduated from the lowest education level. There is a similar gap year Age Control Treated Difference (t-stat.) (2004) for vocational graduates because their studies also a. Full sample became longer by one year. This, and other differences 20 18.14 18.52 0.38 in labour market characteristics make it very important (2.79) to control for labour market entry in our estimations. 21 18.87 19.33 0.46 There is data on the year of starting the first job, which, of (4.85) course, is only available for those who have ever worked. 22 19.52 19.71 0.19 To correct for this, we generated a variable called ’labour (2.27) market entry year’ that equals the year of starting the 23 20.00 20.03 0.03 first job, or, when it is not available, the year when the (0.35) respondent finished their highest education level (pro - 24 20.33 20.30 −0.03 vided in the EU-SILC database). We corrected this vari- (−0.36) able so that the labour market entry age of respondents is 25 20.81 20.88 0.07 at least 18, as this is the legal age when young people can (0.70) start to work full time in Poland. 26 21.12 21.21 0.09 Table 2a–c show when each age group of pre- and post- (0.75) reform cohorts started their first job. Since post-reform 27 21.22 21.38 0.16 cohorts finished their education later, they started work - (0.95) ing later (at 20.47 vs 20.35  years, p < 0.01 ). The differ - b. ISCED 2 or below ence is significant for the younger cohorts and disappears 20 16.90 18.14 1.24 in the older cohorts. Surprisingly, later job starting does (3.27) not go together with less experience: treated people of 21 18.26 18.81 0.55 the same age tend to have higher years of experience than (1.83) the control group (2.87 vs 2.58 years on average, p < 0.01, 22 18.75 19.22 0.47 see Table  3a–c for more detail). This could be due to (1.53) different employment chances. If treated cohorts have 23 19.02 19.34 0.33 higher employment chances than the control cohorts, (0.90) then–on average–even if they start working later, they might secure more stable jobs and gain more experience 24 19.00 19.75 0.75 over a shorter period. Table  4a–c seem to underline this (1.90) assumption, as we see a higher share of employed peo- 25 19.28 20.25 0.96 ple in younger cohorts among the treatment group than (2.47) in the control group, and the differences are even higher 26 19.75 20.19 0.44 within the lowest educated people. Across all age groups (0.79) in the full sample the means are 79 percent vs 77 percent 27 19.91 19.28 −0.62 (p < 0.01). (−0.93) Figure  3 shows the real earnings distribution of the c. ISCED 3 pre- and post-reform cohorts. Treated people tend to 20 18.34 18.62 0.28 earn more on average (PLN 1664 vs PLN 1388), mainly (2.03) because fewer people are at the bottom of the earn- 21 18.92 19.38 0.46 ings distribution in the treatment group. That is, the (4.49) earnings distribution is shifted to the right, moving 22 19.55 19.72 0.18 (2.03) 23 19.94 19.95 0.01 10 th The last primary school graduates finished 8 grade in 2000, and the first (0.07) th lower secondary graduates finished 9 grade in 2002. 24 20.16 20.07 −0.09 The results are essentially the same if we change the minimum working (−0.88) age to 15 for the control and 16 for the treatment cohorts. The pre-reform 25 20.26 20.33 0.07 cohorts could start working in part-time jobs at the age of 15. Along with the extension of comprehensive education until the age of 16, the legal job- (0.57) starting age for part-time jobs was also extended to 16. 26 20.30 20.26 −0.04 Means are weighted by the inverse of the size of each age group in the (−0.24) sample. The labour market effects of the polish educational reform of 1999 Page 9 of 21 13 Table 2 (continued) Table 3 Mean years of experience by age Age Control Treated Difference (t-stat.) Age Control Treated Difference (t-stat.) a. Full sample 27 20.24 20.32 0.08 20 0.79 0.85 0.06 (0.44) (0.69) Tables 2a–c show the mean age when treated and control group members 21 0.99 1.11 0.12 started their first regular job, separately by age groups. The third column shows the differences between the two groups at each age with t-statistics in the (1.78) parentheses. 2a presents the data of the whole sample, Table 2b for those who 22 1.36 1.64 0.28 have at most a lower secondary qualification, and 2c for those with an upper (4.55) secondary qualification 23 1.83 2.22 0.38 (5.70) 24 2.36 2.84 0.48 those at the bottom of the distribution to the middle. (6.21) This shift is apparent in the full sample and for the low 25 2.79 3.23 0.45 educated. (4.90) From the raw data and from the research before us, 26 3.44 3.81 0.36 we assume that Poland’s 1999 comprehensive educa- (3.08) tion reform had a non-negligible and positive effect 27 4.23 4.40 0.17 on the Polish labour market. We believe that it was (1.04) especially young people (where the level of educa- b. ISCED 2 or below tion and skills gained matter most) and those at the 20 1.02 1.01 −0.01 bottom of the education distribution, who stayed one (−0.05) more year in school, who benefited the most from the 21 0.83 1.32 0.50 reform. (2.75) 22 1.79 1.70 −0.08 1.3 M ethodology and baseline results (−0.36) We now turn to our causal estimates. Year of birth deter- 23 2.33 2.47 0.14 mined the assignment into treatment, which means that (0.50) self-selection into the treatment or control group was 24 2.97 2.95 −0.03 impossible. However, there are cohort-specific differ - (−0.08) ences, as the treatment and control group members were 25 3.09 3.18 0.10 born in different years. To handle this issue, we control (0.26) for age fixed-effects. There are also differences between 26 3.54 3.83 0.28 the years each survey was taken, so we control for sur- (0.41) vey year fixed effects. We assume that in the absence 27 4.40 5.78 1.38 of the treatment, in both the treatment and the control (1.34) group, changes in outcomes between two consecutive c. ISCED 3 survey years would have been the same for all ages, and 20 0.75 0.81 0.06 vice versa: changes between two consecutive ages would (0.57) have been the same for all survey years (parallel trends 21 1.02 1.09 0.07 assumption). For this reason, as a baseline, we opted for (1.00) a difference-in-differences method, where age and year- 22 1.36 1.63 0.27 of-survey act as the two dimensions of the estimation (4.16) 23 1.86 2.27 0.41 (5.36) We also included labour-market entry fixed effects in some regressions, 24 2.48 3.06 0.58 which further diminish the problem of cohort-specific differences. See below (6.03) for more detail. 25 3.14 3.74 0.60 For instance, suppose the only available survey years are 2005-2006. Then the average change between 2005 and 2006 for age groups ranging from 21 (5.05) to 27 is a counterfactual change for age 20 (the only age group from which 26 3.96 4.65 0.69 we have both treated and control observations in the sample). Now, suppose (4.31) we have only 2006 and 2007, the assumption is that average change between 27 4.85 5.13 0.27 2006 and 2007 for cohorts 22-27 and 20 is a counterfactual for change at age 21 etc. And similar logic applies for the other dimension. See Appendix B (1.16) for further discussion. 13 Page 10 of 21 L. F. Drucker et al. Table 3 (continued) Table 4 Share of employed people among those who are active (employed or unemployed) Tables 3a–c show the mean years of experience in the treatment and the control group, separately by age groups. The third column shows the differences Age Control Treated Difference (t-stat.) between the two groups at each age with t-statistics in the parentheses. 3 a presents the data of the whole sample, 3b for those who have at most a lower a. Full sample secondary qualification, and 3c for those with an upper secondary qualification 20 0.52 0.64 0.12 (the first differences) and the treatment variable as the (2.82) diff-in-diff (second difference) estimator. 21 0.60 0.70 0.10 The baseline specification of the multivariate model is (3.90) the following: 22 0.70 0.76 0.06 (3.04) Y = α + βTreat + ρX + γ + δ +μ + ε asri as asri a s asri, 23 0.77 0.79 0.02 (1.34) where Y is the outcome variable (current educational sta- 24 0.80 0.81 0.01 tus, employment status, or earnings) for each individual (0.76) (i). Treat is the treatment dummy, which can vary across 25 0.82 0.82 −0.01 ages (a) and year-of-survey (s). X are individual-level (−0.34) variables (gender and highest level of education, in some 26 0.83 0.87 0.03 specifications), and γ, δ, and μ are age, year of survey, (1.97) and region fixed effects. ε is the idiosyncratic error term, 27 0.85 0.82 −0.03 while α, β, and ρ are parameters to be estimated. For a similar estimation framework, see, for instance, Pischke (−1.41) (2007). b. ISCED 2 or below Table 5 shows the results of our baseline regression on 20 0.41 0.59 0.19 labour market outcomes. In columns 1 and 2 we estimate (1.89) linear probability models of the employment probability 21 0.44 0.59 0.15 on the sample of the active population. Columns 3 and (2.02) 4 show the results of regressions of log real earnings. We 22 0.58 0.71 0.13 estimate all models with age and year-of-survey fixed (1.84) effects as well as region fixed effects. While these should 23 0.54 0.68 0.14 take out much of the unobserved heterogeneity across (1.79) cohorts in educational outcomes, pre- and post-treat- 24 0.62 0.65 0.03 ment cohorts differ in their year of labour market entry, (0.33) too, which means they might have faced very different 25 0.60 0.58 −0.02 labour market conditions. This difference can easily bias (−0.19) the effect of the treatment on longer-run labour market 26 0.65 0.67 0.03 outcomes. Consequently, in columns 2 and 4 we also (0.26) include year-of labour-market-entry fixed effects. On 27 0.60 0.41 −0.18 the one hand, controlling for the year when one enters (−1.36) the labour market seems to be essential as different c. ISCED 3 demand-side factors can alter the entrants’ employment 20 0.54 0.65 0.11 probabilities and initial earnings. On the other hand, the (2.24) year of labour market entry might be considered a ’bad 21 0.62 0.71 0.09 control’ (see Angrist and Pischke 2008) as it correlates (3.25) well with years spent in schooling, which depend on the 22 0.71 0.76 0.05 reform. Moreover, year of labour market entry correlates (2.48) strongly with age and year-of-survey, which inflates the 23 0.78 0.80 0.02 (0.89) 24 0.80 0.81 0.01 The treatment variable is basically a simplified interaction term between (0.29) the age and the year-of survey, as it is shown in table 1. 25 0.82 0.85 0.03 We have tested for potential differences in composition across regions (see Bukowski 2019). Including regional fixed-effects in the regressions do (1.30) not change any of the results. 26 0.82 0.85 0.04 For people with no experience, we imputed their year of labour market (1.38) entry with their year of finishing highest education (see above). The labour market effects of the polish educational reform of 1999 Page 11 of 21 13 Table 4 (continued) we cannot compare the average finishing age for the full pre- and post-treatment sample. However, we can Age Control Treated Difference (t-stat.) compare the lowest educated sub-population, as there 27 0.84 0.82 −0.02 was no compositional change regarding this education (−0.59) group. Appendix Fig.  5 shows the first stage of our IV: Tables 4a–c show the share of employed respondents in the active population the age distribution for those with only ISCED 2 or less in the treatment and the control group, separately by age groups. The third for each pair of cohorts. The median finishing age for this column shows the differences between the two groups at each age with level was 15 until the 1985 cohort and became 16 with t-statistics in the parentheses. 4a presents the data of the whole sample, 4b for those who have at most a lower secondary qualification, and Table 4c for those the 1986 cohort. Columns 1 and 2 in Appendix Table  12 with an upper secondary qualification show that the age when the highest degree was obtained for this population has zero effect in itself on log of earn - ings and employment probability. 1st stage estimates in variance of the model. Nevertheless, substantial results of columns 3 and 5 show that the reform can act as a strong the models do not differ much with or without the year of instrument: treated cohorts are 0.4 or 0.6  years older labour market entry fixed effect. than the pre-treatment cohorts when they finish their The results in Table  5 show that the treated group is highest degree of schooling. In columns 4 and 6 we see about 3 percentage points more likely to be employed and that the 2SLS coefficients of finishing age on earnings and earn 4 to 5% higher earnings. Additional results pre- employment are high and positive, 13.6% and 12 percent- sented in Fig. 4a and b show how these average treatment age points, respectively. Unfortunately, they are insignifi - effects vary across age cohorts. The estimates are too cant, due most likely to the small power of our analysis, imprecise for employment probability to detect any sta- but they are of a very similar magnitude to the signifi - tistically significant and consistent age-related patterns. cant estimates of Table  6 in earnings. Nevertheless, However, people closer to their twenties benefited from both estimations show that the lowest educated popula- a 10 percentage points increase in their earnings after the tion would have been more likely to earn more had they reform compared to the pre-treatment people of similar attended comprehensive lower secondary schooling for age. This effect declines with age and disappears around an additional year. age 24–26. These results can be explained by arguing that the reform improved labour market entrance, but the 2 Conclusion effects disappear with age as experience becomes more We analysed the effects of the 1999 education reform on important than general skills learned at school. labour market outcomes, in particular on employment probability and earnings. This reform was comprehen - 1.4 Robustness checks sive, as it extended general education by one year before In a robustness check, we simplify our models and com- students were tracked into upper secondary education. It pare only two cohorts–right before and right after the also improved the quality of education for the low-track reform–with each other. This might decrease the power students before they were tracked and increased the gen- of our analysis substantially but also highlight the impor- eral skills of potential vocational students and those who tance of the “inserted” year of education before tracking, later chose the vocational track. We show that the reform as all other reform elements have impacted these two resulted in a 3 percentage point increase in employment cohorts similarly. Table  6 presents the same models as probability. The reform also increased earnings by 4–5% in Table  5 regressed only on the 1985 and 1986 sample, on average. The results suggest that the positive effects on so the cohorts born right before and right after the 1986 earnings decrease with age, so that the reform treatment January 1 cut-off. The effects on the employment prob - could have been beneficial in the labour market entrance, abilities are not significant; the 1986 cohort earns 9–11% but its long-term impact is relatively small. more than the 1985 cohort on average. Our research tries to answer the question of why there In another robustness check, we test the mechanisms are mixed results in the literature about the returns of by using the reform as an instrument for the finishing age education. Increasing the length of compulsory education of schooling. Unfortunately, the database does not con- or an additional year of education offered to vocational tain data on the years spent in education, but we know students usually do not help their employment chances the age when the highest educational level was attained. or increase their earnings. We show that a comprehensive As shown above, due mainly to the Bologna Process, 18 19 The effects on earnings are not different by gender. The effects on employ - As treated are only 0.4-0.6 years older than the pre-treatment cohorts, the ment probabilities are, on the other hand, greater for women than for men. causal estimate from the IV should be around 13.6*0.4 = 5.44% for earnings See Appendix Table 13 for the results by gender. and 12*0.58 = 6.96 percentage points for employment. 13 Page 12 of 21 L. F. Drucker et al. Fig. 3 Distribution of real earnings of treatment and control group in 2005 PLN. The figures show the distribution of real earnings in the treatment and the control group in 2005 Polish zloty. The first figure presents data for the whole sample, the second for those who have at most lower secondary qualification (ISCED 2), and the third for those who have upper secondary qualification (ISCED 3) The labour market effects of the polish educational reform of 1999 Page 13 of 21 13 Table 5 The effect of the reform on labour market outcomes, linear models Variables (1) (2) (3) (4) Employed vs unemployed Employed vs unemployed Log earnings Log earnings Treated 0.0314*** 0.0325*** 0.0361** 0.0514*** (0.0108) (0.0110) (0.0157) (0.0168) Female −0.0438*** −0.0517*** −0.183*** −0.183*** (0.00607) (0.00618) (0.00888) (0.00893) Constant 0.565*** 0.453*** 6.764*** 6.716*** (0.0218) (0.0457) (0.0323) (0.0672) Observations 23,274 23,274 15,499 15,499 R-squared 0.051 0.061 0.187 0.191 Age fixed-effect y y y y Year-of-survey fixed-effect y y y y Region fixed-effect y y y y Year of LM entry fixed-effect n y n y Robust standard errors in parentheses All models are linear regressions on the sample of the active population. Columns 1–2 show the results of an LPM on employment probability, while columns 3–4 show the results of the linear regression on earnings. In columns 2 and 4, labour market entry fixed effects are included, too ***p < 0.01, **p < 0.05, *p < 0.1 type of reform can successfully improve the labour mar- Finally, despite being successful in improving the edu- ket outcomes, and the effects are driven by the young and cation outcomes (as shown by the PISA study) and the the low educated (and likely by vocational students). This labour market outcomes of post-reform students (as study has its drawbacks: due to the lack of a proper edu- shown by this study, but also by Liwiński, (2020) and cation measure in the EU-SILC we can only estimate an Strawinski and Broniatowska (2021), using different average effect for all education levels. This tells us little data, methods and subgroups), after 18  years, the Pol- about the potential mechanisms driving the results, thus ish government reversed the reform in January 2017 we can only speculate that the increased general educa- and re-introduced the old system. So, from September tion and decreased tracking (better peers and teachers) 2017, students study again in the old 8 + 4/5 system. caused the effects. After 1999 students in Poland were Along with this reform, they try to improve the qual- forced to sit an additional year in less selected classes than ity of vocational education. Based on the current paper before and were taught by teachers who were less selected. and research before us, we would warn against a re- This change was likely beneficial for the low-track chil - tracking reform like this as it might not be the best way dren, as their composition of peers and teachers improved to improve the labour market conditions of vocational substantially. Our results suggest that the reform has students. reached its initial goal of decreasing inequalities. 13 Page 14 of 21 L. F. Drucker et al. Fig. 4 a The effect of the reform on employment probability and earnings by age groups. The figure shows the effects of the reform on each age cohort estimated by models 2 and 4 in Table 5, extended with treatment*age interactions. The left figure shows the treatment effect on the probability of being employed (interpreted as percentage points), while the right figure on the logarithm of earnings (interpreted as percentage change in the earnings). The controls include gender and age, year-of-survey, region, and year-of-labour-market-entry fixed effects. The bars show the average effects with 95% confidence intervals. b The effect of the reform on employment probability and earnings by age groups, without labour market entry fixed effects. The figure shows the effects of the reform on each age cohort estimated by models 1 and 3 in Table 5, extended with treatment*age interactions. The left figure shows the treatment effect on the probability of being employed (interpreted as percentage points), while the right figure on the logarithm of earnings (interpreted as percentage change in the earnings). The controls include gender and age, year-of-survey, and region fixed effects. The bars show the average effects with 95% confidence intervals The labour market effects of the polish educational reform of 1999 Page 15 of 21 13 Table 6 The effect of the reform on employment probability and earnings—cohorts 1985/1986 Variables (1) (2) (3) (4) Employed vs unemployed Employed vs unemployed Log earnings Log earnings Treated 0.0184 −0.00361 0.0876*** 0.107*** (0.0127) (0.0141) (0.0204) (0.0232) Female −0.0474*** −0.0493*** −0.171*** −0.167*** (0.0113) (0.0118) (0.0180) (0.0182) Constant 0.569*** 0.571*** 6.725*** 6.645*** (0.0400) (0.0530) (0.0961) (0.116) Observations 5082 5082 3439 3439 R-squared 0.049 0.059 0.223 0.245 Age fixed-effect y y y y Year-of-survey fixed-effect y y y y Region fixed-effect y y y y Year of LM entry fixed-effect n y n y Robust standard errors in parentheses All models are linear regressions on the sample of the active population born in 1985–1986. Columns 1–2 show the results of an LPM on employment probability, while columns 3–4 show the results of the linear regression on earnings. In columns 2 and 4, labour market entry fixed effects are included, too ***p < 0.01, **p < 0.05, *p < 0.1 Appendix A See Tables 7, 8, 9, 10, 11, 12, 13 Table 7 Summary of independent variables Variable Obs Mean Std. Dev Min Max Year of birth 48,557 1985.7 3.60 1978 1994 Female 48,557 0.48 0.50 0 1 Experience 26,697 2.27 2.26 0 12 Age when the first job began 23,186 20.28 2.47 8 27 Age 48,557 22.87 2.57 19 27 Level of educ.: low 43,057 0.18 0.39 0 1 Level of educ.: medium 43,057 0.65 0.48 0 1 Level of educ.: high 43,057 0.17 0.37 0 1 Treated 48,557 0.52 0.50 0 1 Gross real earning (PLZ) 16,762 1473 795 15 16,939 Labour market status At work 48,557 0.44 0.50 0 1 Unemployed 48,557 0.12 0.33 0 1 Retired 48,557 0.00 0.03 0 1 Inactive 48,557 0.43 0.50 0 1 13 Page 16 of 21 L. F. Drucker et al. Table 8 Number of observations in each age/year-of-survey cell Age Year of survey Total 2005 2006 2007 2008 2009 2010 2011 2012 2013 19 929 814 679 657 626 586 532 497 529 5849 20 843 821 732 623 572 569 538 499 485 5682 21 860 730 732 655 523 528 498 511 464 5501 22 926 773 687 655 546 539 518 496 486 5626 23 812 801 665 658 573 491 503 469 424 5396 24 746 708 732 618 582 523 495 489 420 5313 25 790 644 618 640 558 515 494 479 467 5205 26 708 676 587 555 567 534 499 495 465 5086 27 675 599 603 531 488 527 515 501 460 4899 Total 7289 6566 6035 5592 5035 4812 4592 4436 4200 48,557 Table 9 Number of observations in each year-of-birth/year-of-survey cell Year of birth Year of survey 2005 2006 2007 2008 2009 2010 2011 2012 2013 Total 1978 675 0 0 0 0 0 0 0 0 675 1979 708 599 0 0 0 0 0 0 0 1307 1980 790 676 603 0 0 0 0 0 0 2069 1981 746 644 587 531 0 0 0 0 0 2508 1982 812 708 618 555 488 0 0 0 0 3181 1983 926 801 732 640 567 527 0 0 0 4193 1984 860 773 665 618 558 534 515 0 0 4523 1985 843 730 687 658 582 515 499 501 0 5015 1986 929 821 732 655 573 523 494 495 460 5682 1987 0 814 732 655 546 491 495 479 465 4677 1988 0 0 679 623 523 539 503 489 467 3823 1989 0 0 0 657 572 528 518 469 420 3164 1990 0 0 0 0 626 569 498 496 424 2613 1991 0 0 0 0 0 586 538 511 486 2121 1992 0 0 0 0 0 0 532 499 464 1495 1993 0 0 0 0 0 0 0 497 485 982 1994 0 0 0 0 0 0 0 0 529 529 Total 7289 6566 6035 5592 5035 4812 4592 4436 4200 48,557 The labour market effects of the polish educational reform of 1999 Page 17 of 21 13 Table 10 Number of observations and row percentage of activity status and current education activity Basic activity status Current education activity Total Not in education In education Missing At work 13,958 4450 3047 21,455 % 65 21 14 100 Unemployed 5027 424 549 6000 % 84 7 9 100 In retirement or early retirement 18 36 1 55 % 32.73 65.45 1.82 100 Other inactive 3219 15,996 1832 21,047 15.29 76 8.7 100 Total 22,222 20,906 5429 48,557 % 45.76 43.05 11.18 100 Table 11 Linear regressions on the age of finishing education Variables (1) (2) (3) (4) Full sample Currently not in education Age of finishing ed Age of finishing ed Age of finishing ed Age of finishing ed Treated 0.0164 0.0900*** −0.0208 0.184*** (0.0480) (0.0285) (0.0674) (0.0426) Low ed −3.824*** −3.847*** (0.0490) (0.0518) High ed 3.946*** 4.297*** (0.0377) (0.0369) Treat* low ed 0.806*** 0.494*** (0.0590) (0.0757) Treat* high ed −0.785*** −0.956*** (0.0654) (0.0817) Female 0.548*** −0.0444*** 0.749*** −0.0142 (0.0247) (0.0148) (0.0377) (0.0232) Constant 17.62*** 18.95*** 17.68*** 18.77*** (0.0502) (0.0280) (0.0817) (0.0533) Observations 37,569 37,569 21,510 21,510 R-squared 0.188 0.724 0.138 0.731 Age fixed-effect y y y y Year-of-survey fixed-effect y y y y Robust standard errors in parentheses *** p < 0.01, **p < 0.05, *p < 0.1 13 Page 18 of 21 L. F. Drucker et al. Table 12 The effect of the reform on employment probability and earnings—IV estimation for people with ISCED2 or less Variables (1) (2) (3) (4) (5) (6) OLS 1st stage 2sls 1st stage 2sls log earning Employed log real earning Employed Age when highest education −0.00307 0.00633 0.136 0.120 was attained (0.0150) (0.00923) (0.194) (0.0768) Treated 0.401** 0.583*** (0.188) (0.123) Female −0.171*** −0.178*** 0.186 −0.202*** 0.166** −0.201*** (0.0615) (0.0298) (0.143) (0.0649) (0.0805) (0.0362) Constant 6.422*** 0.361** 15.46*** 4.829* 15.22*** −1.247 (0.247) (0.158) (0.295) (2.865) (0.189) (1.115) Observations 914 1917 914 914 1917 1917 R-squared 0.174 0.101 0.151 0.079 0.115 0.009 Region fixed-effect y y y y y y Age fixed-effect y y y y y y Year-of-survey fixed-effect y y y y y y Robust clustered standard errors in parentheses Columns 1–2 show the association of the age when the highest degree was obtained with log earning and with employment probability, respectively for those who finished at most ISCED 2 level. Columns 3–4 and 5–6 show the 2SLS estimations of log real earning and employment probability with the treatment used as an instrument for the age when they finished the ISCED 2 level *** p < 0.01, **p < 0.05, *p < 0.1 Table 13 The effect of the reform on employment probability and earnings separately by gender Variables (1) (2) (3) (4) Employed vs Employed vs log earnings log earnings unemployed unemployed Treated 0.0230* 0.0210* 0.0377** 0.0514*** (0.0121) (0.0121) (0.0165) (0.0175) Female −0.0506*** −0.0607*** −0.182*** −0.183*** (0.00710) (0.00715) (0.0116) (0.0116) Treated* Female 0.0184 0.0244* −0.00325 0.000110 (0.0136) (0.0136) (0.0187) (0.0187) Constant 0.569*** 0.457*** 6.763*** 6.716*** (0.0219) (0.0460) (0.0322) (0.0671) Observations 23,274 23,274 15,499 15,499 R-squared 0.051 0.061 0.187 0.191 Age fixed-effect y y y y Year-of-survey fixed-effect y y y y Region fixed-effect y y y y Year of LM entry fixed-effect n y n y Robust standard errors in parentheses ***p < 0.01, **p < 0.05, *p < 0.1 The labour market effects of the polish educational reform of 1999 Page 19 of 21 13 Figure 5 Fig. 5 Distribution of ages when highest educational level was attained by pairs of cohorts for students with ISCED 2 or below. The coloured bars show the distribution of the ages of the second birth cohort, while the empty bars show that of the first birth cohort in each figure. The median age of schooling was 15 for all birth cohorts until 1985 and became 16 for all birth cohorts starting from 1986 Figure  7 compares people of different ages in the same Appendix B survey year, so as we move to the right in the figure, the Unfortunately we cannot do a proper common trend cohorts presented become younger. Thus, none of these analysis with our data. As this is a quasi-panel—of age- graphs is informative enough alone. The only way to of-respondent and year-of-observation—, we could compare pre-and post-reform cohorts is by accounting either compare people of similar age in different survey for both dimensions in a regression framework—includ- years or people of different ages in the same survey year. ing both age and year-of-observation as fixed-effects; and Figure  6 compares age cohorts in different survey years. this is what we do in the paper. The increase in the log wages as we move to the right in Figures 6, 7 the graph is therefore partly caused by economic trends. 13 Page 20 of 21 L. F. Drucker et al. Fig. 6 Log wage by age groups before and after the reform Fig. 7 Log wage by survey-year groups before and after the reform The labour market effects of the polish educational reform of 1999 Page 21 of 21 13 Acknowledgements Jakubowski, M., Harry, P., Emilio, P., Jerzy, W.: The effects of delaying tracking in We would like to thank Pawel Bukowski, Adam Booij, Eva Holb, Karolina secondary school: evidence from the 1999 education reform in Poland. Kurpaska, Heiko Rachinger, Balázs Reizer and Ágnes Szabó-Morvai, and two Educ. Econ. 24(6), 557–572 (2016) anonymous referees as well as the participants at the Education, Skills, and Jakubowski, M.: Poland: polish education reforms and evidence from interna- Labor Market Outcomes workshop in Oslo 2017, the meeting of the Hungar- tional assessments. In: Crato, N. (ed.) Improving a country’s education. ian labour economists at Szirák 2015, at the annual 2015 meeting of the Hun- Springer, Cham (2021). https:// doi. org/ 10. 1007/ 978-3- 030- 59031-4_7 garian Society of Economists, and seminar participants at the CERS-HAS for Jung-Miklaszewska, J.: The system of education in the Republic of Poland. their helping comments. The usual disclaimer applies. The authors gratefully Bureau for Academic Recognition and International Exchange, Warsaw acknowledge funding from the Hungarian Scientific Research Fund (Project (2003) no. 109338) and the Horizon 2020 Twinning grant EdEN (Project no. 691676). Kerr, P., Sari, T.P., Uusitalo, R.: School tracking and development of cognitive skills. J. Law. Econ. 31(3), 577–602 (2013) Kwiek, M.: Social perceptions versus economic returns of the higher Declarations education:the Bologna process in Poland. In: Studien zur international vergleichenden Erziehungswissenschaft. Schwerpunkt Europa - Studies Competing interests in International Comparative Educational Science. Focus: Europe. 28 The authors have no conflicts of interest to declare that are relevant to the November 2013. pp 147–182 (2014) content of this article. The authors gratefully acknowledge funding from the Liwiński, L.: The impact of compulsory schooling on Hourly Wage: evidence Hungarian Scientific Research Fund (Project no. 109338) as well as the Horizon from the 1999 education reform in Poland. Eval. Rev. 44(5–6), 437–470 2020 Twinning grant EdEN (Project no. 691676). All authors contributed (2020) equally to the paper. The paper uses the European Union Statistics on Income Malamud, O., Pop-Eleches, C.: General education versus vocational training: and Living Conditions (EU-SILC) database that can be accessed from Eurostat evidence from an economy in transition. Rev. Econ. Stat. 92(1), 43–60 at https:// ec. europa. eu/ euros tat/ web/ micro data/ europ ean- union- stati stics- (2010) on- income- and- living- condi tions. Meghir, C., Palme, M.: Educational reform, ability, and family background. Am. Econ. Rev. 95(1), 414–424 (2005) Author details Oosterbeek, H., Webbink, D.: Wage effects of an extra year of basic vocational 1 2 Central European University, Budapest, Hungary. Institute of Econom- education. Econ. Educ. Rev. 26(4), 408–419 (2007) ics, Centre for Economic and Regional Studies and Institute of Economics, Pekkarinen, T., Uusitalo, R., Kerr, S.: School tracking and intergenerational Corvinus University of Budapest, Budapest, Hungary. Faculty of Economic income mobility: evidence from the finnish comprehensive school Sciences, University of Warsaw, Warsaw, Poland. reform. J. Public. Econ. 93(7–8), 965–973 (2009) Pischke, J.-S.: The impact of length of the school year on student performance Received: 28 September 2021 Accepted: 29 July 2022 and earnings: evidence from the german short school years. Econ. J. 117(523), 1216–1242 (2007) Pischke, J.-S., von Wachter, T.: Zero returns to compulsory schooling in Ger- many: evidence and interpretation. Rev. Econ. Stat. 90(3), 592–598 (2008) Psacharopoulos, G., Patrinos, H.A.: Returns to education: a decennial review of References the global literature. Educ. Econ. 26(5), 445–458 (2018) Aakvik, A., Salvanes, K.G., Vaage, K.: Measuring heterogeneity in the returns Rivkin, S.G., Hanushek, E.A., Kain, J.F.: Teachers, schools, and academic achieve- to education using an education reform. Eur. Econ. Rev. 54(4), 483–500 ment. Econometrica 73(2), 417–458 (2005) (2010) Sacerdote, B.: Peer effects in education: how might they work, how big Angrist, J.D., Pischke, J.-S.: Mostly harmless econometrics: an empiricist’s com- are they and how much do we know thus far? Handb. Econ. Educ. 3, panion. Princeton University Press (2008) 249–277 (2011) Becker, GS.: “Human capital: a theoretical and empirical analysis, with special Strawiński, P., Broniatowska, P.: The impact of prolonging compulsory general reference to education”. National Bureau of Economic Research (1975) education on the labour market. Educ. Stud. (2021). https:// doi. org/ 10. Bialecki, I., Johnson, S., Thorpe, G.: Preparing for national monitoring in Poland. 1080/ 03055 698. 2021. 19576 70 Assess. Educ. Princ. Policy. Pract. 9(2), 221–236 (2002) Bukowski, P.: How history matters for student performance lessons from the Publisher’s Note partitions of Poland. J. Comp. Econ. 47(1), 136–175 (2019) Springer Nature remains neutral with regard to jurisdictional claims in pub- Chetty, R., Friedman, J.N., Rockoff, J.E.: Measuring the impacts of teachers I: lished maps and institutional affiliations. evaluating bias in teacher value-added estimates. Am. Econ. Rev. 104(9), 2593–2632 (2014) European Commission. Focus on the Structure of Higher Education in Europe 2004/05 - Teaching - EU Bookshop. European Commission, Directorate- General for Education, Youth, Sport and Culture (2005) Grenet, J.: Is extending compulsory schooling alone enough to raise earnings? Evidence from French and British compulsory schooling laws*. Scand. J. Econ. 115(1), 176–210 (2013) GUS: Oświata i wychowanie w roku szkolnym 2005/2006. Central Statistical Office, Poland (2006) Hall, C.: The effects of reducing tracking in upper secondary school evidence from a large-scale pilot scheme. J. Hum. Res. 47(1), 237–269 (2012) Hall, C.: Does more general education reduce the risk of future unemploy- ment? Evidence from an expansion of vocational upper secondary education. Econ. Educ. Rev. 52, 251–271 (2016) Hanushek, E.A., Woessmann, L.: Do better schools lead to more growth? Cognitive skills, economic outcomes, and causation. J. Econ. Growth. 17, 267–321 (2012) Heckman, J.J., Humphries, J.E., Veramendi, G.: The nonmarket benefits of education and ability. J. Hum. Cap. 12(2), 282–304 (2018). https:// doi. org/ 10. 1086/ 697535 Jakubowski, M.: "Opening up opportunities: education reforms in Poland". IBS Policy Papers (2015)

Journal

Journal for Labour Market ResearchSpringer Journals

Published: Dec 1, 2022

Keywords: Education reform; Returns to education; Poland; Detracking; Labour market; Difference-in-differences; I21; I24; I26; J24

References