University of Amsterdam
Faculty of Business and Economics
Master of Science
Selective schools and the gender gap in test scores
Thesis supervisor:
Student:
Erik Plug
Fabian Szekeres
11392207
Economics MSc, Public Policy track
Course: 60 ECTS; Thesis: 15 ECTS
1
Statement of Originality
This document is written by Fabian Szekeres who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
2
Table of contents
1. Introduction ... 3
2. Literature review ... 5
3. Data and Methodology ... 8
3.1 Data ... 8
3.2 Methodology ... 12
4. Results ... 15
4.1 Main results ... 15
4.2 Heterogeneity ... 19
4.3 Tenth grade mathematics and reading scores ... 21
5. Discussion and conclusion ... 25
3 Abstract
Hungarian students have the option to leave the general track of schooling at age 10 or 12 and enrol in high schools that offer 8-year or 6-year tracks, respectively. Approximately 90 percent of them stay in the general track. These children do not leave for high school until at least finishing eighth grade at age fourteen. There is an annual country-wide Competency Survey in mathematics and reading mandatory for all sixth and eighth graders. Between grades 6 and 8 the students in 6-year and 8-year tracks have been proven to improve their scores more than those staying in the general track, even after controlling for previous test scores. This thesis offers evidence that suggests males benefit even more from the selective environment than girls. The ordinary least squares (OLS) estimates and the instrumental variable (IV) approach indicate that between grades 6 and 8 girls in 6-year tracks gain between 0 and 75 percent of what boys in 6-year tracks do.
1. Introduction1
When discussing the main drivers of the economic success of individuals and countries, education is one of the main topics of interest. Most nations invest a significant share of their gross domestic product (GDP) in their education systems in the hope of creating an environment which allows future generations to thrive. The goal is the same, though the institutional setting differs from country to country. The ages at which children start attending school or at which they are selected into different tracks vary significantly, as does the degree of the systems’ centralization. A great indicator of the success of different approaches is the standardized PISA test. The test compares the mathematics, comprehension and science skills of 15 year olds in the partaking countries.
The PISA tests are of great value when making country-to-country comparisons. Policy implications, however, can be derived not only from contrasting the systems of different
1 I would like to take this opportunity to express my gratitude to Daniel Horn, a leading researcher at the
Economics Department of the Hungarian Academy of Sciences for his guidance in tackling the research question. I am also thankful for all the input provided in the early stages of my research by Hessel Oosterbeek, my lecturer at the University of Amsterdam. Last, but not least, I am extremely grateful to Erik Plug, my thesis supervisor and a professor at the University of Amsterdam for all his advice and ideas that helped shape this thesis.
4
nations. In most countries there are multiple paths students can take. Examining the results produced by each of these can help in determining which might be the most effective within each country’s institutional setting. The age at which parents choose to, or can afford to, enrol their child in high school is one aspect to consider when examining the academic success of students within a country. The goal of education systems should be to maximize the potential of all youngsters, granting equal opportunity to all, regardless of social status and gender. However, the method and age of selection can greatly influence how children with different backgrounds are affected. (Ammermüller, 2005; Arnett, 2007).
Before the change of regime in the early 1990s Hungary would not have been of as much interest in this research topic. The path of students only diverged at one point: they all took part in general education until reaching the end of eighth grade at age fourteen. At this point they were assigned to high schools and vocational schools based on their academic achievements. The end of communism brought with it a change in institutional setting. High schools offering 8-year and 6-year tracks, selecting their students at age 10 and 12, respectively, were established (Horn, 2010). Due to the constantly shrinking student population of the country, these schools had to offer certain advantages to stay viable. One of these edges is the earlier selection of the best performing students, which ensures a more competitive environment with peers of higher ability. The success of these institutions has been substantial. Nowadays approximately 10 percent of eighth graders attend either an 8-year or a 6-year track. The positive effect they have on their pupils’ academic achievements has been substantiated by Horn (2013). He uses the results of an annual, nationally administered test in mathematics and comprehension mandatory for all sixth, eighth and tenth graders and a portion of those in fourth grade. The author’s findings add to the extensive literature on the effect of selective schooling on academic achievement (Abdulkadiroğlu et al., 2014; Dobbie and Fryer, 2014; Clark and Bono, 2016). The international results are mixed, some find the effects of selective schools to be significantly positive, while others arrive at a contrasting conclusion.
The added value provided by the highly selective schools offering 8-year and 6-year tracks compared to the general track in Hungary in and of itself is already grounds for discussion. The potentially different effects they have on boys and girls is a further area of interest. This paper will investigate whether boys choosing the more competitive 6-year track instead of continuing their general education until age 14 gain more than girls making the same decision. If they do, the question becomes why boys might benefit more, why the gender gap might open up. A potential explanation in this case would be a selection bias at the cut-off point.
5
Also, the behaviour of girls and boys in competitive settings might differ. This avenue could be worthwhile to explore, as it has been an explanation provided by a number of economists for other observed gender gaps in academic achievement and track choices (Niederle and Vesterlund, 2011; Landaud et al., 2016). Another fascinating explanation involves the different effect the gender of teachers might have on female and male performance (Carrell et al., 2010). Carrell shows that the gender of the teacher does not impact male achievement, while female scores increase by 10 percent of a standard deviation in introductory math and science courses if there is a female faculty, closing two thirds of the measured gender gap.
The main results will underscore the notion that boys benefit more from going to a 6-year track than girls. For males attending 6-6-year tracks there is an approximately 0.20 standard deviations bump in eighth grade mathematics and reading test scores compared to their counterparts in general education, holding the sixth grade results even. For girls the effect is between 0.04 and 0.10 standard deviations smaller than for boys. These results do not appear to be driven by a particular subsample. Urban citizens are just as affected as students in rural schools. The gender gap shifts in favour of the males irrespective of socioeconomic status and the effect does not differ whether the students are growing up in a household with both of their parents or not. The instrumental variable model estimates the shift in gender gap to be even larger. Interestingly, one methodological approach indicates that this apparent discrepancy disappears by tenth grade, while the instrumental variable estimation provides evidence to the contrary. The former assessment would be consistent with some earlier findings that discrepancies in test scores tend to fade with time (Fletcher and Kim, 2016).
The following pages of the paper will be divided into four distinct sections. The first section will discuss the relevant literature, providing a basis for my research and explaining why certain topics were further investigated. The second part will detail the dataset and the methodology used to answer the research question. The third part will contain the results and analysis. The concluding chapter will summarize the paper, discuss its shortcomings and suggest further avenues to explore.
2. Literature review
The topic of gender differences in payoffs attributed to attending 6-year high schools in Hungary has not yet been extensively researched. The discovery that high schools selecting their students at an earlier age are more beneficial to students’ academic development than the
6
general tracks in Hungary has been made by Horn (2013) but the gender gap is not included in that particular paper. However, there are some studies on the effect selective schooling has on male and female academic achievements in different countries.
The topic of the paper by Jackson (2010) is the one that most closely resembles that of this thesis. Jackson sets out to check if better quality schools with higher ability peers provide a payoff compared to less competitive surroundings. The country of interest is Trinidad and Tobago, where after fifth grade all students are to enrol in a high school. School assignment is based on matching the preferences of the students with the results they achieve on a country-wide test administered after fifth grade. A great difference between Trinidad and Tobago and Hungary is the fact that Hungarian children are not required to write the standardized assignment tests after fourth, sixth or even eighth grade. Partaking is an individual choice that is influenced by a multitude of factors.
The importance of the differences in the two systems comes into play when examining Jackson’s methodology. The regression discontinuity model used is more appropriate in Trinidad and Tobago’s setting than would be in the case of Hungary due to the fact that the cut-off points of the high schools create ability groups. The tail end of a higher ability group will have very similar test scores to the top end of the group just below it. This allows Jackson to see whether those who just make it into a better high school are better off than those with very similar scores falling just short of the requirements. In contrast, in Hungary only a fraction of the students even take part in the application process to the early selective high schools and there are no clear thresholds therefore using the scores of these tests would be questionable at best. However, the country-wide assessment administered every year, compulsory for all sixth, eighth and tenth graders and a large portion of fourth graders provides an extensive dataset, but more on this later.
Jackson uses the test scores achieved after fifth grade and the results of the tests at the end of high school to assess the effect a better high school with higher ability students might have. He finds that the more selective institutions have a significant added value, improving test scores and chances for admission to tertiary education. This could be due to multiple influencers: peer effect; the better schools attracting more qualified teachers; the higher scores inviting more funding which in turn creates even more added value, generating a cycle. All these benefit girls approximately twice as much as boys in Trinidad and Tobago – the exact opposite of what this paper will find in Hungary. Students at the lower end of the ability
7
distribution gain less from being surrounded by marginally higher achieving peers than those at the top.
In spirit, Landaud, Ly and Maurin’s (2016) recent work is quite similar to Jackson’s insofar as that the methodology is also a regression discontinuity design. The two main differences stem from the contrasting approaches of the researched countries. The early selection observed in Trinidad and Tobago is not present in France. French students are not assigned to a high school until after they have finished grade nine. This in itself is fascinating, since there is some evidence supporting the idea that earlier selection leads to higher social inequality (Meghir and Palme, 2005; Pekkarinen et al.. 2009; Jakubowski, 2008). However, the main focus of the Landaud, Ly and Marin’s work, as well as mine, is not the effect selective schooling takes on social inequalities but rather how it affects the academic achievements and choices of males and females.
The authors investigate the education region of Paris, where the school assignment is quite similar to that in Trinidad and Tobago: it is based on the preferences listed by the students and their results on the test administered after grade nine. After the first round of assignments there are some schools that are oversubscribed and some that are not as sought after. Landaud, Ly and Maurin identify the cut-off point and check to see if the oversubscribed, generally more reputable high schools provide a benefit to those just making it compared to those who miss out by a small margin and therefore are forced to enrol in an undersubscribed institution. They also set out to see whether the more competitive environment inspires girls to specialize in science instead of humanities or social sciences.
Contrary to Jackson’s results the authors find that in France being surrounded by higher ability peers in a more highly regarded school does not manifest itself in better test scores or graduation rates. The claim that girls do not benefit as much from the more selective and competitive environment is substantiated. In fact, there is a negative effect associated with being admitted into a more selective high school. Female students are less likely to specialize in science in the more competitive environment in their last two years of secondary education and are more prone to specialize in humanities. Male choice on the other hand does not seem to be influenced by surroundings. This finding is also underscored by the fact that before the assignment system became so centralized and ability-based in 2008 the gender gap was narrower. These results substantiate the claim of behavioural scientists that females shy away from competition more than males (Niederle and Vesterlund, 2011).
8
Whereas there are multiple studies on the effect of educational settings on academic achievements on the international scene, they are scarce in Hungary. One of the reasons for this is the quite recent change of regime in the early 1990s and the accompanying lack of trustworthy data for previous years. Horn (2013), however, did publish a study using the first few nationally administered Competency Surveys in which he sets out to determine the added value selective schools provide their students with and the social inequalities they cause.
The author provides evidence substantiating the notion that students attending 8-year and 6-year tracks gain a competitive advantage between sixth and tenth grade. Moreover, between sixth and eighth grade those selected at age 10 gain even more than their peers enrolling in high school at age twelve. The earlier the selection, the higher the added value. This holds true even after controlling for test scores and socioeconomic status, indicating that the difference in added value does not stem from ability but rather from the more selective environment. The author makes another key observation: the mathematics and reading scores in tenth grade are more contingent on socioeconomic status in smaller settlements offering 6-year track than in those that do not. This means that the opportunity for earlier enrolment contributes to an increase in a town’s social inequalities.
Also, those enrolling in high school after grade 6 do not only have higher test scores in sixth grade but also come from more prominent families. Taking this and multiple other factors into account Horn arrives at a similar conclusion to one that will also be drawn in this thesis: high school tracks selecting their students at a younger age contribute to an increase in social inequalities. Part of the reason for this is that gifted children in smaller and poorer communities are too far from these 8-year and 6-year tracks that can generally be found in larger and richer settlements.
3. Data and Methodology
3.1 Data
Before the change of regime around 1990 and for more than a decade after it there was no reliable source on younger children’s academic achievements in Hungary. In 2001 this changed with the introduction of a standardized test that is similar in its goals and construction to the PISA test: assessing the comprehension and mathematics tools of the students rather than their
9
so called book knowledge. At first they were only administered every two years and in a less comprehensive manner. The first major change came in 2006 when the Survey became annual, and the second shift in 2008 brought with it the comprehensiveness that makes the last almost ten years’ data extremely valuable. Starting with 2008 all students were assigned an identifier which made it possible to track their scores through the years. The assessment also became mandatory for all sixth, eighth and tenth graders, while examining fourth graders in 200 nationally representative schools.
Because tracking individual students has only been a possibility since 2008 this thesis works with a database which starts in that year and ends in 2015. The Competency Survey provides invaluable data not only on test results but thanks to an extremely thorough follow-up questionnaire also on the track type, family background, grades achieved, gender, age and a whole host of other individual attributes. In the interest of a clearer picture the test scores that were originally standardized to a mean of 500 with a standard deviation of 100 are standardized to a mean of 0 and a standard deviation of 1.
This dataset is amended by two additional ones that serve different purposes. The socio-economic status (SES) index is included to have a good control on social status. This is a variable that is standardized to a mean of 0 and a standard deviation of 1. It is comprised of three separate functions. The first of these is the amount of years the parents spent in the education system. The second is a standardized index of the households’ valuables: number of bedrooms, cellular phones, cars, bathrooms, personal book-collection of the student and the book-collection of the household, and whether the student has a room, a desk and a personal computer to themselves. The third is a standardized index of the parents’ status on the labour-market (employed or unemployed). The Competency Suvey is also amended by a database which contains the students’ distance from the closest high school offering a 6-year track.
Of the hundreds of variables listed in the Competency Survey database this thesis only uses a certain set. In the main part of the research test results from sixth and eighth grade are included, in a later check for long-term effects tenth grade achievement is also taken into account. Because those attending 8-year tracks are already in high school by sixth grade, for all intents and purposes they are not included. This is simply because the question in mind is whether 6-year tracks offer more of an advantage to boys than they do to girls when compared to the general track.
10
Besides the test scores there are multiple variables in the set that are included in some shape or form. Sex of the students is of course of high importance. The track type is also taken into account in all models. Then there are dummy variables created using the following methods, used as controls in the main regressions and later as instruments for heterogeneity checks. A student is considered urban if according to the questionnaire attached to the Survey they attended a school in a city, a county centre or the capital, Budapest. They are grouped into the rural category if the institution is located in a village. Students are considered well-off if compared to others they are living well, or above average and assigned to the other group if they live on the average or beneath it. The last dummy variable created takes into account whether the children grew up in a broken home. They are considered to have lived in a household with both parents if they lived with both their father and their mother at the time of the survey. If either parent was not living under the same roof as the child, the student is assigned to the group of kids with broken homes.
Table 1 Descriptive statistics
General track 6-year track
Female Male Female Male
Control variables Urban 70.92% 70.57% 99.38% 99.31% Well-off 30.44% 35.65% 41.11% 49.46% Both parents 70.28% 72.66% 77.38% 79.81% Socioec. status -.113944 (.9859879) -.1326955 (.9845952) .8219211 (.7511705) .9245962 (.739984) Reading in 6th .0827993 (.9371976) -.1957022 (.9694714) .8869096 (.7860893) .7841644 (.8236861) Math in 6th -.1011085 (.9466201) -.0211172 (.9772141) .7076169 (.8640608) .9955255 (.8986764) Dependent variables Reading in 8th .0538028 (.966288) -.222792 (.9813079) .8023176 (.7938637) .6930263 (.8137373) Math in 8th -.1110937 (.9649687) -.0478686 (.9992354) .500528 (.8520975) .8586078 (.901215) Reading in 10th .0565691 (.9378442) -.1676833 (.9939686) .8308526 (.7568154) .7711588 (.7873937) Math in 10th -.1460521 (.9183637) .0239328 (.999219) .6248187 (.8308628) 1.035533 (.8661449) Instrumental variable Minutes by bus to nearest 6-year track in 6th 42.60006 (31.66228) 42.47008 (31.68851) 20.98314 (22.62972) 20.4414 (21.88505) Observations* 206 009 (48.64%) 217 562 (51.36%) 13 382 (51.91%) 12 399 (48.09%) 423 571 (94.26%) 25 781 (5.74%)
11
When assessing descriptive statistics (Table 1) grouping the students into their respective tracks allows us to instantly spot some significant differences. While approximately 30 percent of both female and male students staying in general education until after grade 8 attend a school in a rural setting, almost all children in 6-year tracks are enrolled in an urban institution. Smaller communities do not have enough children and resources to maintain both a 6-year track and a general track, therefore they only offer the latter in almost all cases. As expected this also translates into a higher percentage of students in the 6-year tracks being better off than the average. Another observation can be made regarding the social status of the observed children: male pupils appear to be coming from better-off families regardless of the track. Similarly, around 80 percent students attending high school for six years come from homes where both parents are present, while this holds true for only a little over 70 percent of children in the general track. Males are a little better off in this department as well but the difference is not as substantial as in the case of social status.
Looking at the standardized test scores in grade 6 and 8 as well as the socio-economic status there is a substantial separation between the two groups. Both male and female students who opt for a 6-year track achieved results that are almost a whole standard deviation better than their general track counterparts’ in both mathematics and reading in sixth grade. The same holds true for the socio-economic status, which indicates that not only are the students attending high school for six years better off, but they also come from more prominent families and have a more stable background. Also, the descriptive statistics show us that the students attending 6-year tracks live closer to these institutions than those staying in the general track. The latter group would have to travel over 40 minutes on a bus to get to a high school that offers the more selective environment. The former group lives half as far away. We can conclude therefore that distance from the nearest institution that offers a 6-year track also factors into the decision about early enrolment.
As can be seen in figures 1 through 4, students who are in a 6-year track are already at the higher end of the ability distribution in grade 6, before they go off to high school. This is true for both males and females in both reading and writing. However, there already appears to be a gender gap in the selection: among males the system is even more selective, an even higher part of the grade 6 score distribution opts to leave the general track. This might indicate that any potential widening of the gender gap might stem from the fact that the system is even more selective when it comes to boys. Unfortunately, there is no measure for motivation in this database, but if we assume that higher ability is accompanied by a higher level of motivation
12
this can be part of the explanation for the gender gap potentially widening between grades six and eight.
Figure 1: Distribution of grade 6 math scores, general population Figure 2: Distribution of grade 6 math scores, 6-year track
Figure 1: Distribution of grade 6 reading scores, general population Figure 2: Distribution of grade 6 reading scores, 6-year track
3.2 Methodology
The methodology used in this thesis does not resemble the regression discontinuity (RD) design applied by Landaud, Ly and Maurin (2016) in France or by Jackson (2010) in Trinidad and Tobago. The dataset does not allow for an RD design because a threshold above which students are accepted to 6-year tracks cannot be assessed. However, the data provided by the Competency Survey can be used to check on the added value these institutions provide by examining the results achieved in grades 6 and 8. In addition, using the above described dummy variables there are heterogeneity checks included. Also, the main models are applied to see whether any potential effects last until tenth grade.
13
The main results stem from two ordinary least squares (OLS) estimates. One of these includes additional control variables.. The initial estimations are as follows:
𝑚𝑎𝑡ℎ8𝑖 = 𝛼 + 𝛽1∗ 𝑚𝑎𝑡ℎ6𝑖+ 𝛽2∗ 𝑟𝑒𝑎𝑑𝑖𝑛𝑔6𝑖 + 𝛽3∗ 𝑓𝑒𝑚𝑎𝑙𝑒8𝑖+ 𝛽4∗ 6𝑦𝑡𝑟𝑎𝑐𝑘𝑖+ 𝛽5∗ 𝑚𝑎𝑡ℎ6𝑖𝑋𝑓𝑒𝑚𝑎𝑙𝑒8𝑖 + 𝛽6∗ 𝑟𝑒𝑎𝑑𝑖𝑛𝑔6𝑖𝑋𝑓𝑒𝑚𝑎𝑙𝑒8𝑖 + 𝛽7∗ 6𝑦𝑡𝑟𝑎𝑐𝑘𝑖𝑋𝑓𝑒𝑚𝑎𝑙𝑒8𝑖+ 𝑢𝑖 (1) 𝑟𝑒𝑎𝑑𝑖𝑛𝑔8𝑖 = 𝛼 + 𝛽1∗ 𝑚𝑎𝑡ℎ6𝑖 + 𝛽2∗ 𝑟𝑒𝑎𝑑𝑖𝑛𝑔6𝑖 + 𝛽3∗ 𝑓𝑒𝑚𝑎𝑙𝑒8𝑖 + 𝛽4∗ 6𝑦𝑡𝑟𝑎𝑐𝑘𝑖+ 𝛽5∗ 𝑚𝑎𝑡ℎ6𝑖𝑋𝑓𝑒𝑚𝑎𝑙𝑒8𝑖 + 𝛽6∗ 𝑟𝑒𝑎𝑑𝑖𝑛𝑔6𝑖𝑋𝑓𝑒𝑚𝑎𝑙𝑒8𝑖+ 𝛽7∗ 6𝑦𝑡𝑟𝑎𝑐𝑘𝑖𝑋𝑓𝑒𝑚𝑎𝑙𝑒8𝑖+ 𝑢𝑖 (2)
Where the dependent variables are the mathematics and comprehension results in grade 8 achieved by student ‘i’. These are estimated by a constant; the individual’s scores from grade 6; the track attended in grade 8 (6𝑦𝑡𝑟𝑎𝑐𝑘𝑖), which in this case is either the general or the 6-year track; the sex of the child in eighth grade (𝑓𝑒𝑚𝑎𝑙𝑒8𝑖); all independent variables interacted with the student’s sex; and an error term (u). The parameter of interest is the interaction of the track attended in grade 8 and the sex of the student in both equations. The coefficient of this parameter will show if females benefit less from attending 6-year tracks than males. If 𝛽7 is significant and negative the above statement is corroborated. I use clustered standard errors based on the school attended in sixth grade. The interacted terms are necessary to control for gender differences already present in grade six. The assumption is that sixth grade scores control for ability. A similar assumption has been made by Arcidiacono et al. (2010) in their paper. This estimation might potentially not be accurate enough since there is no control introduced for the socioeconomic status, and including the type of settlement, the wellbeing of the family and the status of the family could also help in reducing the omitted variable bias. Therefore there is a second round of estimations which includes all these and their interactions with the sex of the student. However, if we assume that past performance is an appropriate ability control including these additional control variables should not change the results significantly. The results section will show that the results do in fact remain the same after the inclusion of said controls.
𝑚𝑎𝑡ℎ8𝑖 = 𝛼 + 𝛽1∗ 𝑚𝑎𝑡ℎ6𝑖+ 𝛽2∗ 𝑟𝑒𝑎𝑑𝑖𝑛𝑔6𝑖 + 𝛽3∗ 𝑓𝑒𝑚𝑎𝑙𝑒8𝑖+ 𝛽4∗ 6𝑦𝑡𝑟𝑎𝑐𝑘𝑖+ 𝛽5∗ 𝑚𝑎𝑡ℎ6𝑖𝑋𝑓𝑒𝑚𝑎𝑙𝑒8𝑖 + 𝛽6∗ 𝑟𝑒𝑎𝑑𝑖𝑛𝑔6𝑖𝑋𝑓𝑒𝑚𝑎𝑙𝑒8𝑖 + 𝛽7∗ 6𝑦𝑡𝑟𝑎𝑐𝑘𝑖𝑋𝑓𝑒𝑚𝑎𝑙𝑒8𝑖 + 𝛽8∗ 𝑢𝑟𝑏𝑎𝑛8𝑖 + 𝛽9∗ 𝑤𝑒𝑙𝑙𝑜𝑓𝑓8𝑖+ 𝛽10∗ 𝑝𝑎𝑟𝑒𝑛𝑡𝑠8𝑖 + 𝛽11∗ 𝑢𝑟𝑏𝑎𝑛8𝑖𝑋𝑓𝑒𝑚𝑎𝑙𝑒8𝑖 + 𝛽12∗ 𝑤𝑒𝑙𝑙𝑜𝑓𝑓8𝑖𝑋𝑓𝑒𝑚𝑎𝑙𝑒8𝑖 + 𝛽13∗ 𝑝𝑎𝑟𝑒𝑛𝑡𝑠8𝑖𝑋𝑓𝑒𝑚𝑎𝑙𝑒8𝑖 + 𝑢𝑖 (3) 𝑟𝑒𝑎𝑑𝑖𝑛𝑔8𝑖 = 𝛼 + 𝛽1∗ 𝑚𝑎𝑡ℎ6𝑖 + 𝛽2∗ 𝑟𝑒𝑎𝑑𝑖𝑛𝑔6𝑖 + 𝛽3∗ 𝑓𝑒𝑚𝑎𝑙𝑒8𝑖+ 𝛽4∗ 6𝑦𝑡𝑟𝑎𝑐𝑘𝑖 + 𝛽5∗ 𝑚𝑎𝑡ℎ6𝑖𝑋𝑓𝑒𝑚𝑎𝑙𝑒8𝑖 + 𝛽6∗ 𝑟𝑒𝑎𝑑𝑖𝑛𝑔6𝑖𝑋𝑓𝑒𝑚𝑎𝑙𝑒8𝑖+ 𝛽7∗ 6𝑦𝑡𝑟𝑎𝑐𝑘𝑖𝑋𝑓𝑒𝑚𝑎𝑙𝑒8𝑖+ 𝛽8∗
14
𝑢𝑟𝑏𝑎𝑛8𝑖 + 𝛽9∗ 𝑤𝑒𝑙𝑙𝑜𝑓𝑓8𝑖+ 𝛽10∗ 𝑝𝑎𝑟𝑒𝑛𝑡𝑠8𝑖 + 𝛽11∗ 𝑢𝑟𝑏𝑎𝑛8𝑖𝑋𝑓𝑒𝑚𝑎𝑙𝑒8𝑖 + 𝛽12∗
𝑤𝑒𝑙𝑙𝑜𝑓𝑓8𝑖𝑋𝑓𝑒𝑚𝑎𝑙𝑒8𝑖 + 𝛽13∗ 𝑝𝑎𝑟𝑒𝑛𝑡𝑠8𝑖𝑋𝑓𝑒𝑚𝑎𝑙𝑒8𝑖 + 𝑢𝑖 (4)
Another approach that might estimate the difference in effect 6-year tracks have on girls and boys is an instrumental variable model. This is useful because in the above described estimations the omitted variable bias and the endogeneity bias can heavily influence the results. The latter in this case might pose an issue because track choice is influenced by ability and motivation, two things that also greatly predict test scores. Test score affects track choice. An instrumental variable is only viable if it checks out two boxes. Firstly, it has to significantly correlate with the variable deemed endogenous, in this case track choice. Secondly, it shall not correlate with the error term.
The instrumental variable of choice in this case was the time it would have taken by bus from each child’s home to the nearest school offering a six-year track. The validity of this IV will be tested in the results section but it is also worthwhile to examine it intuitively. The distance should not affect the abilities or motivation levels of a child. It, however, should have an impact on the track choice, given that with the time spent with traveling growing larger the inclination to attend a 6-year track diminishes. At first glance it seems that the eighth grade reading and mathematics scores are affected by distance to the nearest 6-year track only through the track choice made in sixth grade. In a similar setting Card (1993) also uses the same IV, this was also taken into consideration when making the decision. Thus, the IV estimations consist of the first-stage (5) and the second-stage estimates (6,7):
6𝑦𝑡𝑟𝑎𝑐𝑘𝑖 = 𝛼 + 𝛽1∗ 𝑚𝑎𝑡ℎ6𝑖 + 𝛽2∗ 𝑟𝑒𝑎𝑑𝑖𝑛𝑔6𝑖 + 𝛽3∗ 𝑓𝑒𝑚𝑎𝑙𝑒8𝑖+ 𝛽4∗ 𝑏𝑢𝑠𝑡𝑖𝑚𝑒6𝑖+ 𝛽5∗ 𝑚𝑎𝑡ℎ6𝑖𝑋𝑓𝑒𝑚𝑎𝑙𝑒8𝑖 + 𝛽5∗ 𝑟𝑒𝑎𝑑𝑖𝑛𝑔6𝑖𝑋𝑓𝑒𝑚𝑎𝑙𝑒8𝑖+ 𝛽6∗ 𝑏𝑢𝑠𝑡𝑖𝑚𝑒6𝑖𝑋𝑓𝑒𝑚𝑎𝑙𝑒8𝑖 + 𝑢𝑖 (5) 𝑚𝑎𝑡ℎ8𝑖 = 𝛼 + 𝛽1∗ 6𝑦𝑡𝑟𝑎𝑐𝑘𝑖+ 𝛽2 ∗ 6𝑦𝑡𝑟𝑎𝑐𝑘𝑖𝑋𝑓𝑒𝑚𝑎𝑙𝑒8𝑖 + 𝛽3 𝑚𝑎𝑡ℎ6𝑖 + 𝛽4 ∗ 𝑟𝑒𝑎𝑑𝑖𝑛𝑔8𝑖 + 𝛽5∗ 𝑓𝑒𝑚𝑎𝑙𝑒8𝑖+ 𝛽6∗ 𝑚𝑎𝑡ℎ6𝑖𝑋𝑓𝑒𝑚𝑎𝑙𝑒8𝑖 + 𝛽7∗ 𝑟𝑒𝑎𝑑𝑖𝑛𝑔6𝑖𝑋𝑓𝑒𝑚𝑎𝑙𝑒8𝑖+ 𝑢𝑖 (6) 𝑟𝑒𝑎𝑑𝑖𝑛𝑔8𝑖 = 𝛼 + 𝛽1∗ 6𝑦𝑡𝑟𝑎𝑐𝑘𝑖 + 𝛽2∗ 6𝑦𝑡𝑟𝑎𝑐𝑘𝑖𝑋𝑓𝑒𝑚𝑎𝑙𝑒8𝑖+ 𝛽3 𝑚𝑎𝑡ℎ6𝑖 + 𝛽4∗ 𝑟𝑒𝑎𝑑𝑖𝑛𝑔8𝑖 + 𝛽5∗ 𝑓𝑒𝑚𝑎𝑙𝑒8𝑖+ 𝛽6∗ 𝑚𝑎𝑡ℎ6𝑖𝑋𝑓𝑒𝑚𝑎𝑙𝑒8𝑖 + 𝛽7∗ 𝑟𝑒𝑎𝑑𝑖𝑛𝑔6𝑖𝑋𝑓𝑒𝑚𝑎𝑙𝑒8𝑖+ 𝑢𝑖 (7)
Besides these estimation there are a few other noteworthy calculations that do not differ that much in their method. First of all there will be three heterogeneity checks conducted using the
15
three dummy variables introduced earlier (urban or rural; well-off or not well-off; intact home with both parents or broken home). Then the estimations described in equations 1-7 are used to see if anything changes when the outcome and control variables (except for test score in grade 6, naturally) are shifted to grade ten.
4. Results
4.1 Main results
Using the methodology and data described in the previous section one can assess the effect selective schooling has on the performance of males and females. The difference in added value between the two sexes can also be measured. The interaction of the sex and the track of the student helps identify the difference in added value. The results will show that both in reading and in mathematics the added value of 6-year tracks between grades 6 and 8 is greater for boys than it is for girls. This holds true for all methods used: the OLS estimation with only the main controls; the OLS estimation including a set of additional control variables; and the IV approach.
In Table 2 all three approaches are visible for both the effects on eighth grade mathematics and reading scores. Columns 1-4 contain the OLS estimations. In the first two the eighth grade mathematics score, while in the latter two the eighth grade reading score is the dependent variable. Columns 1 and 3 provide the estimations with only the main set of control variables included: reading and mathematics scores in grade 6; the track attended by the student; the interaction of the aforementioned three with the sex of the child; and the sex of the pupil. Columns 2 and 4 provide the estimations with the additional control variables included (urban or rural; well-off or not well-off; living with both parents or in a broken home). Columns 5 and 6 include the IV estimations, with the top half of the table portraying the first stage estimates.
16
Table 2
Estimations of eighth grade mathematics and reading scores
(1) (2) (3) (4) (5) First stage (6) First stage
VARIABLES Main math Controls math Main reading Controls reading IV math IV reading
Math in 6th 0.0316*** 0.0316*** (0.00120) (0.00120) Reading in 6th 0.0261*** 0.0261*** (.00137) (.00137) Female 0.00787*** 0.00787*** (.00199) (.00198) Math in 6th_female -0.00306** -0.00306** (0.00127) (0.00127) Reading in 6th_female 0.00118 0.00119 (.00125) (.00125) Bus time -0.000931*** -0.000930*** (0.0000324) (0.0000324) Bus time_female -0.000141*** -0.000141*** (0.0000259) (0.0000259) Constant 0.0961*** 0.00286 0.0961*** 0.00286
(1) (2) (3) (4) (5) Sec. stage (6) Sec. stage
VARIABLES Main math Controls math Main reading Controls reading IV math IV reading
Math in 6th 0.581*** 0.572*** 0.246*** 0.227*** 0.600*** 0.227*** (0.00445) (0.00479) (0.00281) (0.00315) (0.00581) (0.00518) Reading in 6th 0.214*** 0.186*** 0.557*** 0.539*** 0.217*** 0.565*** (0.00306) (0.00330) (0.00296) (0.00330) (0.00463) (0.00445) 6-year track 0.206*** 0.162*** 0.182*** 0.157*** 0.435*** 0.898*** (0.0367) (0.0376) (0.0307) (0.0254) (0.121) (0.128) Female -0.0841*** -0.0607*** 0.134*** 0.151*** -0.0641*** 0.141*** (0.00268) (0.00676) (0.00232) (0.00612) (0.00540) (0.00499) Math in 6th_female -0.0375*** -0.0395*** -0.00945*** -0.0124*** -0.0302*** -0.00781*
17 (0.00361) (0.00409) (0.00317) (0.00355) (0.00470) (0.00433) Reading in 6th_female 0.0205*** 0.0164*** 0.0469*** 0.0285*** 0.0305*** 0.0612*** (0.00346) (0.00387) (0.00325) (0.00374) (0.00443) (0.00416) 6-year track_female -0.102*** -0.0888*** -0.0428*** -0.0427*** -0.408*** -0.299*** (0.00867) (0.00960) (0.00815) (0.00897) (0.0860) (0.0812) Socioec status 0.106*** 0.106*** (0.00272) (0.00250) Socioec_female 0.00216 0.0259*** (0.00315) (0.00295) Urban -0.0288 -0.0995** (0.0533) (0.0445) Urban_female -0.0294*** -0.0201*** (0.00630) (0.00555) Well-off -0.00398 -0.0150*** (0.00448) (0.00414) Well-off_female -0.0133** -0.0191*** (0.00627) (0.00556) Parents 0.0214*** -0.00132 (0.00377) (0.00361) Parents_female 0.00468 0.00455 (0.00525) (0.00475) Constant -0.00601** 0.0333 -0.118*** -0.0144 -0.0197** -0.156*** (0.00258) (0.0387) (0.00217) (0.0323) (0.00768) (0.00700) Observations 397,461 291,536 397,577 291,606 385,740 385,853 R-squared 0.627 0.639 0.672 0.683 0.588 0.622
Notes: The F-statistic for the first-stage of the IV estimations are 169.75 and 169.76 (they only differ in the first stage of the IV estimation for reading and math because there is a minor difference in the number of observations – this is also why there are extremely small differences in the first stage estimations and why I reported both separately.) The first stage estimates substantiate the notion that the distance to the nearest 6-year track does statistically significantly
correlate with the attended track and can be used if it does not correlate with the error term. Here, it is assumed that the instrumental variable of choice does not correlate with the error term so it is used as the instrumental variable. Robust standard errors in parentheses
18
The chosen track has a far greater effect in the IV estimation than it does in the OLS estimations. Columns 1-4 show that attending a 6-year track boosts eighth grade mathematics and reading scores by 0.15-0.21 standard deviations even after controlling for grade 6 results and a host of other characteristics. In Columns 5 and 6 the IV approach estimates this effect to be almost 0.50 standard deviations in mathematics and 90 percent of a standard deviation in the case of reading and comprehension. Because the number of observations is so large and the data is reliable, measurement errors could not have caused the large differences between the OLS and the IV estimations. One explanation could be that the IV identifies the impact on the compliers. However, according to all results boys attending 6-year tracks gain an extreme competitive advantage compared to their counterparts who had the similar abilities in grade 6 but stayed in the general education for two more years.
The impact is smaller for girls in all three estimations. The OLS approach estimates that the added value of 6-year tracks is statistically significantly smaller for females than for males. The difference ranges from 0.09-0.10 standard deviations in mathematics and is approximately 4 percent of a standard deviation in reading and comprehension. This means that according to the OLS estimates 6-year tracks provide a payoff to all students between grades 6 and 8 but girls gain only half and 75 percent of what boys do in mathematics and reading, respectively. While these are already statistically significant impacts on the gender gap, they are substantially smaller than the effect estimated by the IV approach. Columns 5 and 6 show us that girls achieve 40 percent and 30 percent of a standard deviation less of a payoff in mathematics and reading, respectively, than do boys. This means that girls attending 6-year tracks only gain about 0.017 standard deviations, effectively nothing, in mathematics when compared to their counterparts in the general tracks but are still provided with an added value of about 60 percent of a standard deviation in reading.
Looking at the OLS estimations including the additional controls the main conclusion that can be drawn is that they do not account for the widening of the gender gap caused by 6-year tracks. The main explanatory variables take on very similar estimations with and without the added controls. There are some interesting observations, however, that can be made when assessing the OLS estimations that include the controls. Unsurprisingly, being one standard deviation higher in the distribution of socioeconomic status improves both mathematics and reading scores by 10 percent of a standard deviation. Living with both parents does not affect reading scores and has only an unsubstantial impact on mathematics results. This is counterintuitive, coming from broken homes would be considered a disadvantage by most.
19
Attending an urban school has a significant negative impact on reading scores and a statistically insignificant but also negative effect on mathematics results all else being equal. The main results are all in Table 2, however, interest in which subsample might be driving the effect 6-year tracks have on the gender gap prompted tests to check if the above described impacts are heterogeneous.
4.2 Heterogeneity
The below table (Table 3) contains the OLS estimates described in Equations 1 and 2 for only specific parts of society. In Columns 1 and 4 only students attending urban schools are taken into account. They represent a little over 70 percent of all pupils. Looking at the estimates it is clear that it is not specifically the rural or the urban population that drives the widening of the gender gap in 6-year tracks. Female students in these tracks yet again appear to reap approximately half and 75 percent of the benefits boys enjoy in mathematics and reading, respectively.
Similarly, when excluding all those who grow up in a household leading an average or below average lifestyle, we find that the results do not change substantially. Female children of well-off families tend to do better in reading than their male counterparts by over 10 percent of a standard deviation and worse in mathematics by a little less than 0.10 standard deviation. These estimations are extremely similar in fact almost identical to the ones seen in Table 1. The gender gap effect of the 6-year tracks also stays the same, substantiating the notion that it is not a certain segment of society that is driving the effects.
20
Table 3
Heterogeneity checks
(1) (2) (3) (4) (5) (6)
VARIABLES Urban M Socio M Parents M Urban R Socio R Parents R
Math in 6th 0.603*** 0.596*** 0.588*** 0.249*** 0.251*** 0.246*** (0.00519) (0.00555) (0.00483) (0.00320) (0.00428) (0.00336) Reading in 6th 0.205*** 0.206*** 0.216*** 0.558*** 0.549*** 0.565*** (0.00356) (0.00449) (0.00361) (0.00340) (0.00452) (0.00356) 6-year track 0.191*** 0.193*** 0.180*** 0.178*** 0.177*** 0.162*** (0.0363) (0.0422) (0.0374) (0.0306) (0.0451) (0.0321) Female -0.0893*** -0.0829*** -0.0837*** 0.131*** 0.139*** 0.132*** (0.00314) (0.00464) (0.00320) (0.00270) (0.00415) (0.00285) Math in 6th_female -0.0399*** -0.0315*** -0.0378*** -0.00659* -0.0109* -0.00919** (0.00409) (0.00639) (0.00450) (0.00367) (0.00583) (0.00403) Reading in 6th_female 0.0246*** 0.0160** 0.0185*** 0.0450*** 0.0526*** 0.0414*** (0.00402) (0.00623) (0.00432) (0.00379) (0.00585) (0.00408) 6-year track_female -0.0916*** -0.108*** -0.103*** -0.0396*** -0.0474*** -0.0418*** (0.00881) (0.0123) (0.00999) (0.00833) (0.0124) (0.00944) Constant 0.00120 -0.0101** 0.0248*** -0.101*** -0.120*** -0.101*** (0.00335) (0.00432) (0.00304) (0.00283) (0.00448) (0.00261) Observations 289,554 112,529 230,973 289,632 112,564 231,019 R-squared 0.640 0.659 0.645 0.674 0.692 0.687
Notes: Dependent variable: eighth grade mathematics and reading scores. Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
When only looking at children who live in a household with both of their parents present, the same conclusions can be drawn. All explanatory variables have a by and large similar coefficient as they do in Table 1. Female students in 6-year tracks yet again prove to benefit less from selective schooling than males. The difference is stagnant around the 0.1 and 0.04 standard deviations mark for mathematics and reading, respectively. This means that it is neither the children coming from broken homes, nor the ones living with both their parents that affect the gender gap increase but rather both groups. The three heterogeneity checks and the results they provide allow one to conclude that regardless of which subsample is examined, 6-year tracks provide added value to all enrolled students between grades 6 and 8, but are statistically significantly and substantially more beneficial for boys than they are for girls.
4.3 Tenth grade mathematics and reading scores
A further area of interest that raises its head when examining this particular topic is whether the gender gap continues to grow in later years due to selective schooling. To examine this, the mathematics and reading scores in tenth grade can be used. Of course because the dataset is limited to the years between 2008 and 2015 and there is a four year gap between these two grades the number of observations shrinks significantly. Allowing for one year in which a student might fail only pupils attending grade 6 in the years 2008, 2009 and 2010 can be included. However, because of the extensiveness of the Competency Survey this still means a large number of observations that should be enough to draw some conclusions. In Table 4 (below) Columns 1 and 2 are the OLS estimates that are based on Equations 1 and 3, while the results Columns 3 and 4 are based on the Equations 2 and 4. Columns 5 and 6 show the results of the IV estimation.
The competitive advantage provided by 6-year tracks diminishes by grade 10 according to the OLS estimations. The added value shrinks to about 6 percent and 12-14 percent of a standard deviation in reading and mathematics, respectively. The IV model, however offers starkly contrasting results: the added value of 6-year tracks substantially increases. Both in tenth grade reading and mathematics scores the payoff is over one standard deviation for males in 6-year tracks. This is more consistent with intuition: the 6-6-year tracks offer better quality education and are filled with higher ability peers not only between grades six and eight but also between grades eight and ten. It stands to reason that their positive effect on test scores only increases with time. The extreme differences in the results provided by the two approaches
22
cannot be accounted for by measurement errors. Therefore the conclusion can be made that the IV model simply identifies the impact on the compliers.
The main concern of this thesis is the estimated effect gender has on the benefits of attending a more selective 6-year track. The different approaches, again, provide very different estimations. According to the OLS estimations, by grade 10 girls in selective schools make up for the earlier shift and by that time selective schools do not appear to have a significant effect on the gender gap. Girls attending 6-year tracks are just as much better off in both mathematics and reading compared to their counterparts who stayed in general schooling until after grade 8 as boys enrolled in the same type of programme. This also suggests that between eighth and tenth grade females in the more selective 6-year tracks gain more of a competitive advantage than males. The finding that the size of the effects decline or in fact vanish is consistent with other literature (Fletcher and Kim, 2016).
The IV model shows us a very different picture. Females in 6-year tracks gain about 30 and 40 percent of a standard deviation less than males do in mathematics and comprehension, respectively. This suggests that the gender gap shift stays approximately the same, it does not change further between eighth and tenth grade. This does not mean, however, that girls do not benefit from the selective environment. This only suggests that while males in 6-year tracks gain over a standard deviation in test scores, females benefit from a 0.6-1 standard deviations bump compared to their similarly talented general track counterparts. This is still an extremely substantial advantage over those who do not attend 6-year tracks. Because the IV model is more sophisticated in identifying the impact on compliers, one can assume that in reality the shift in gender gap caused by 6-year tracks does not disappear by tenth grade. This conclusion, again, is more consistent with intuition: girls get acclimated to the new, competitive environment by grade eight and do not fall further behind the boys in their tracks but cannot make up for the earlier developments. The differences in benefits reaped stay stagnant after eighth grade.
Table 4
Estimations for tenth grade mathematics and reading scores
(1) (2) (3) (4) (5) First stage (6) First stage
VARIABLES 10th math 10math controls 10th reading 10read controls IV math IV reading
Math in 6th 0.0315*** 0.0314*** (0.00230) (0.00230) Reading in 6th 0.0233*** 0.0233*** (0.00173) (0.00173) Female 0.00704** 0.00701** (0.00333) (0.00333) Math in 6th_female -0.00345* -0.00343* (0.00179) (0.00179) Reading in 6th_female 0.00116 0.00117 (0.00176) (0.00176) Bus time -0.000857*** -0.000856*** (0.0000532) (0.0000531) Bus time_female -0.000120*** -0.000120*** (0.0000412) (0.0000411) Constant 0.0839 0.0839 (0.00452) (0.00452)
(1) (2) (3) (4) (8) Sec. stage (5) Sec. stage
VARIABLES 10th math 10math controls 10th reading 10read controls IV math IV reading
Math in 6th 0.511*** 0.515*** 0.215*** 0.210*** 0.581*** 0.261*** (0.00526) (0.00596) (0.00315) (0.00368) (0.0101) (0.00869) Reading in 6th 0.126*** 0.110*** 0.453*** 0.446*** 0.209*** 0.527*** (0.00389) (0.00445) (0.00354) (0.00411) (0.00752) (0.00736) 6-year track 0.141*** 0.122*** 0.0664*** 0.0606*** 1.065*** 1.352*** (0.0190) (0.0204) (0.0153) (0.0167) (0.238) (0.240) Female -0.225*** -0.131*** 0.0297*** 0.0607* -0.182*** 0.0974*** (0.00443) (0.0442) (0.00406) (0.0347) (0.0106) (0.00999) Math in 6th_female -0.0692*** -0.0689*** -0.0240*** -0.0263*** -0.0768*** -0.0258***
24 (0.00479) (0.00586) (0.00395) (0.00479) (0.00772) (0.00692) Reading in 6th_female 0.0362*** 0.0393*** 0.0405*** 0.0351*** 0.0614*** 0.0680*** (0.00500) (0.00606) (0.00428) (0.00508) (0.00712) (0.00687) 6-year track_female -0.0148 -0.0114 0.0192 -0.000344 -0.298* -0.471*** (0.0127) (0.0140) (0.0120) (0.0139) (0.166) (0.166) Socieoec. status 0.0862*** 0.0888*** (0.00379) (0.00375) Socioec._female -0.000753 0.00783* (0.00480) (0.00455) Urban -0.0800 0.00915 (0.193) (0.177) Urban_female -0.0840* -0.0231 (0.0441) (0.0349) Well-off -0.0316*** -0.0282*** (0.00574) (0.00570) Well-off_female -0.00301 0.000283 (0.00785) (0.00780) Parents -9.32e-05 -0.0128** (0.00564) (0.00519) Parents_female -0.0100 -0.00774 (0.00694) (0.00658) Constant 0.0429*** 0.171 -0.0854*** -0.0437 -0.0413*** -0.193*** (0.00245) (0.191) (0.00221) (0.175) (0.0152) (0.0150) Observations 219,309 141,819 219,405 141,861 211,518 211,607 R-squared 0.649 0.657 0.673 0.676 0.551 0.566
Notes: The F-statistic for the first-stage of the IV estimations is 53.41. The first stage estimates substantiate the notion that the distance to the nearest 6-year track does statistically significantly correlate with the attended track and can be used if it does not correlate with the error term. Here, it is assumed that the instrumental variable of choice does not correlate with the error term so it is used as the instrumental variable. There are extremely small differences in the first stage estimations for the two first stage estimations only because the number of observations slightly differs. This is why I reported both separately.
Robust standard errors in parentheses Robust standard errors in parentheses
5. Discussion and conclusion
This thesis aimed to examine whether boys benefit more from selective schooling than girls between grades 6 and 8 in Hungary. The OLS estimations with and without controls both conclude that males have significantly more to gain in these two years compared to their counterparts not leaving for high school after sixth grade than females. Both approaches estimate the payoff being around 15-21 percent of a standard deviation for boys in both mathematics and reading compared to the general track. This same payoff is approximately half as large in mathematics and about 25 percent smaller in reading and comprehension for girls in the more competitive environment. The IV estimates for the same effects are substantially larger. In reading, the added value nears 0.9 standard deviations for males and 0.6 for females. In mathematics boys reap a benefit of about 50 percent of a standard deviation, whereas girls appear not to gain anything from attending a selective 6-year track.
The heterogeneity checks confirmed that there is not a certain segment of society driving these effects. Irrespective of the students being enrolled in urban or rural schools the above described impacts stay the same on average. Similarly, the estimates do not change when only examining students whose living situation is better than average, implying that the same holds true for those who live on the average or beneath it. Boys living in a household with both of their parents seem to benefit just as much more from selective schooling as those growing up living in broken homes. The discrepancy between male and female competitive advantage gained in 6-year tracks between grades 6 and 8 disappears by tenth grade according to the OLS estimations. The IV model on the other hand estimates that the additional added value that boys benefit from between grades 6 and 8 in 6-year tracks stays the same. The latter implies that girls gain the same between eighth and tenth grade as do boys but cannot make up for the differences caused in the first two years of high school.
The estimates show that boys benefit more from 6-year tracks than girls. The reason for this might be understood when examining the Kernel density estimates (graphs 1-4). These graphs illustrate that even though in the general population male and female ability does not differ that substantially in sixth grade, among those enrolling in 6-year tracks boys come from a higher end of the ability distribution than girls. This could also coincide with the fact that males are proven to grow up in better off and more prominent families and that those attending 6-year tracks also have a higher socioeconomic status. Because better off and higher ability
26
boys enrol into these selective schools than girls it might also be concluded that family support and higher motivation will them into working even harder than their female counterparts at the early stages thereby benefitting more from the system.
Taking into account all these factors there can be some policy implications assessed: these selective schools that offer significant competitive advantages should be made more accessible to children growing up in less prominent families. Another avenue that could be taken to level the playing field is not allowing for this early selection, but rather following the example of Trinidad and Tobago and France and assessing a fixed cut-off point at which all children would have to choose between high schools and vocational schools. Pushing this cut-off point back to where it originally was, namely after grade 8 would not result in statistically significant losses in efficiency but would help eliminate the inequalities inherent to the current system. This has been proven in Sweden by Meghir and Palme (2005), in Finland by Pekkarinen et al. (2009) and in Poland by Jakubowski (2008). All these countries pushed selection to an older age at one point and made gains in social equality, while not losing efficiency. There is no reason to believe Hungary would be any different.
There are multiple ways this research could be improved and expanded. It would be advisable to check if there are similar effects associated with 8-year tracks in grades 5 and 6 when most students are still in the general track. Because selection in those schools is even earlier, at age 10, they potentially invite even higher ability students creating an even more competitive environment providing higher benefits. Another extension of this study can be made when the first few groups will reach an age at which their labour market success could be observed. Of course, tracking these children using their individual identification numbers could pose a serious issue, there is no such initiative in Hungary as of yet. This latter extension would be extremely useful, however, as it would provide information on how selective schools affect real life success of males and females as opposed to the purely academic success measured in this thesis. Test scores have already been shown to heavily impact future earnings, but even so, because of the uniqueness of the Hungarian education system examining the aforementioned effects would help policy makers immensely in their efforts to calibrate the country’s educational approach to be efficient and fair.
27 6.References
Abdulkadiroğlu, A., Angrist, J., & Pathak, P. (2014). The elite illusion: Achievement effects at Boston and New York exam schools. Econometrica, 82(1), 137-196.
Ammermüller, A. (2005). Educational opportunities and the role of institutions.
Arcidiacono, P., Bayer, P., & Hizmo, A. (2010). Beyond Signaling and Human Capital: Education and the Revelation of Abilit. American Economic Journal: Applied Economics, 2(4), 76-104.
Arnett, S. M. (2007). Influences of National Education Policies on the Academic
Achievement of Highand Low-Social Status Students. In European Forum-Assessing the Quality of Education and Its Relationship with Inequality in European and the Modern Societies, Florence.
Card, D. (1993). Using geographic variation in college proximity to estimate the return to schooling (No. w4483). National Bureau of Economic Research.
Carrell, S. E., Page, M. E., & West, J. E. (2010). Sex and science: How professor gender perpetuates the gender gap. The Quarterly Journal of Economics, 125(3), 1101-1144.
Clark, D., & Bono, E. D. (2016). The long-run effects of attending an elite school: evidence from the United Kingdom. American Economic Journal: Applied Economics, 8(1), 150-176. Dobbie, W., & Fryer Jr, R. G. (2014). The impact of attending a school with high-achieving peers: Evidence from the New York City exam schools. American Economic Journal: Applied Economics, 6(3), 58-75.
Fletcher, J., & Kim, T. (2016). The effects of changes in kindergarten entry age policies on educational achievement. Economics of Education Review, 50, 45-62.
Horn, D. (2010). The Political Background of Structural Changes in the Educational System of Hungary, 1985-1994. Central European University: Institute for Policy Studies Working Paper. Horn, D. (2013). Diverging performances: the detrimental effects of early educational selection on equality of opportunity in Hungary. Research in Social Stratification and Mobility, 32, 25-43.
Jakubowski, M., Patrinos, H. A., Porta, E. E., & Wisniewski, J. (2010). The impact of the 1999 education reform in Poland. Washington, DC: World Bank.
Kirabo Jackson, C. (2010). Do Students Benefit from Attending Better Schools? Evidence from Rule‐based Student Assignments in Trinidad and Tobago. The Economic
Journal, 120(549), 1399-1429.
Landaud, F., Ly, S. T., & Maurin, E. (2016). Competitive Schools and the Gender Gap in the Choice of Field of Study.
Meghir, C., & Palme, M. (2005). Educational reform, ability, and family background. The American Economic Review, 95(1), 414-424.
Niederle, M., & Vesterlund, L. (2011). Gender and competition. Annual Review of Economics, 3(1), 601-630.
Pekkarinen, T., Uusitalo, R., & Kerr, S. (2009). School tracking and intergenerational income mobility: Evidence from the Finnish comprehensive school reform. Journal of Public Economics, 93(7), 965-973.
