Measuring the impact of the reform of the education system on the return to higher education among young people in Poland

(1)

Bachelor thesis:

Measuring the impact of the reform of the

education system on the return to higher

education among young people in Poland.

Name: Magdalena Rola-Janicka

Student Number: 10090029

Program: Economics and Business BS

Supervisor: Prof. Sweder van Wijnbergen

(2)

1. Introduction

In the last years higher education became an important topic of public debate in Poland. Students and the youth fear that acquiring a university diploma will not help them find better jobs. At the same time employers present their dissatisfaction with insufficient skills of young graduates. All sides of the discussion agree that most of the problems can be attributed to flaws in the educational system itself.

Over the last 20 years the percentage of young people participating in tertiary education in Poland has been rising (MNISW, 2012). Nowadays most of the students face a much larger competition on the labor market then graduates in the past. Also programs offered by

universities and colleges do not match needs and expectations of employers, which results in a disequilibrium in the particular parts of the labor market. National statistics on unemployment among graduates summarize this situation, and so in 2012 over 25% of people who graduated from a college or a university were unemployed (Dziennik, 2012). The current state of higher education in Poland is also detrimental to employers and businesses. Number of low-skilled workers is shrinking as more young people decide to pursue higher education instead of vocational schooling.

The situation is not only the problem for individuals, it poses a serious threat to the economy as a whole. Loss of efficiency resulting from either too many people studying is equivalent to losing productivity of those who could do better without higher education and the costs of his or hers investment in university. An additional problem is a danger of losing the tool used by employees to screen their potential employers, which Arrow argued is an important function of higher education (Arrow, 1973). Hampering the solution of the asymmetric

information problem in the labor market might distort the relation between the productivity and the real wage.

Currently, reforms are proposed in order to improve the state of higher education and prospects of students. However, at this stage it is useful to evaluate the effect of the reform that took place in the previous decade as it might give a valuable insight for the policy making nowadays. This research takes a perspective of graduates and tries to evaluate whether educational reforms of 1999-2005 affected the rate of return to tertiary education.

The paper proceeds with providing background information on the reform of the

educational system of 1999-2005 and on major trends in higher education in Poland. Next, main arguments of major theories that link higher education with the labor market, are presented. Further the paper will give the methodology of the research and explain the econometric model at use. Finally, results of data analysis are given and they are being explored with respect to the research question. The last part of the paper concludes and provides an evaluation of the research, suggesting issues to be improved in the future.

(4)

2. Background information on education in Poland

Polish constitution guarantees free and equal access to the education. This is reflected in the schooling system, as an overwhelming majority of primary and secondary schools is public. However, there are more private then public tertiary education institutions. On top of that more than a half of students in tertiary education pay tuition fees.

Some research suggest that the majority of those who pay for their higher education are those from poorer and less-educated families (OECD, 2008). This is one of the reasons why OECD report, concluded that Polish system does not pay enough attention to providing equal chances to access higher education for all socio-economic groups . It is also pointed out that the main focus of the system and its development is placed on increasing the numbers of students, and not on improving equality indices among students (Rozmus & Pado,2013).

In order to understand the current state of affairs in Polish higher education it is crucial to acknowledge changes in the educational system that took place in 1999 and 2005. The first reform fundamentally changed primary and secondary education, the second reform was concerned with tertiary education.

2.1 Reforms of the education system

The reform of 1999 specified the system of financing and transformed the whole structure of schooling system in Poland. Changes were introduced because of growing belief that the old structure was not well crafted for functioning in the dynamic market economy and due to challenges of Polish integration with European Union. One of the main goals was to ensure equal chances in pursuing education for all and to increase the number of youngsters who follow secondary and tertiary education. It also aimed at improving vocational schooling and making it more closely connected to the labor market (Milejska, 2007).

In line with its main goals and motivations, the reform introduced a common syllabus for all schools in Poland. The main emphasis has been put on acquiring practical skills such as reading comprehension. The syllabus includes topics that need to be covered at a particular stage as well as recommendations regarding literature and educational materials. At the same time centralized examinations, which were supposed to test the extent to which the obligatory material was known and understood by students, have been introduced. Exams take place at the end of each stage of education and results serve as a recruitment criterion when applying to schools of next level. Most importantly examination at the end of secondary education,

traditionally called Matura was no longer assessed within the school by its teachers. Centralized examination have changed Matura into a more objective measure of students performance.

Instead of being composed of 8 years of primary education and 4 years of secondary education, the system was changed to 6 years of primary, 3 years of middle and 3 years of secondary schooling. In the first 9 years of studying all pupils follow a common track, the critical

(5)

moment comes when they enter secondary education. Due to the reform students at this level have become able to choose between:

 general high schools- which focus on preparing students to pass Matura (allowing them to follow tertiary education) ;

 profiled high schools-which prepare students to pass Matura and at the same time put much emphasis on one particular area, giving students a basic knowledge of a broad profession and thus allowing them to continue their education in the “post-high” schools that teach them a particular profession;

 technical secondary schools- which prepare students to pass Matura and at the same time teach them a profession so that they can obtain a professional eligibility;

 vocational schools-which teach a profession but do not give an opportunity to pass Matura and thus if one wants to pursue higher education he needs to continue with “post-high” school that provides chance to take Matura exam;

The distinction between these already existed, however profiled high schools were a new form that was supposed to give students more time to choose their future path. One of the main ideas while creating this system was to restrict vocational and technical schooling and promote general and profiled high schools so that more students would be able to continue with tertiary education. Policy makers set out a goal of increasing a participation rate of students in high schools to 80% and arriving at a 20% participation rate of students in vocational and technical schools. This was supposed to serve the purpose of increasing a participation rate in higher education to 65% by 2010 (MEN, 2000).

The next stage of transformation of the educational system was the reform of 2005 which set out new law for higher education. Main goal of the reform was to clarify the system and to adjust it to requirements of EU-wide Bologna Process. Much of the value of the reform was in reorganizing administrative rules that apply to universities and linking all earlier resolutions with each other in one document.

As for changes that directly influenced students, the reform introduced a division into bachelor and master degrees as well as a system of assigning weight importance of courses by ECTS points. What’s more entrance exams to universities and colleges were eliminated and results of Matura exam became the only selection criterion for students applying to study at public universities. This selection process allows those with better results to participate in completely subsidized programs, while those with worse Matura results can follow the program if they pay their own tuition fee. The structure of these programs is generally the same or very similar, though in majority of cases, classes are scheduled in the afternoon and evening hours in order to give the students chance to work part time.

(6)

In academic year 2011/2012 new pro-quality reform of higher education was

introduced, however it did not impact any of the cohorts analyzed in this paper. The aim of the new reform was to support private institutions and provide more incentives to improve

performance by promoting competition between public and private universities. The new policy awards best organizations with a five year status of National Leading Scientific Centers . The best institutions are supposed to be chosen by an independent group of experts, and additional funds obtained can be distributed as grants for research or scholarships for best PhD students, they can also be invested in improving teaching skills of staff and enhancing career paths for students. The impact of this reform is yet to be seen.

2.2 Financing of higher education

Public universities are financed mostly through the governmental budget, however they also acquire their funds through European Union subsidies as well as tuition fees charged to students of unsubsidized programs.

The main contributions from national budget come to universities as two separate funds- the one for education and the one for research. Since 1993 the allocation of these funds is based on an algorithm, which was designed so that to allow objectiveness and promote

efficiency within universities. The most important component of the formula is the level of past year subsidy, however major weights are put on the number of students and PhD students and number of teachers. In practice the formula favors increasing the number of students in, and it does not promote improving the quality of studies (Puukka, Dubarle, Goddard, Hazelkorn, & Kuczera, 2013).

The trends in governmental funding of higher education are dependent on

macroeconomic situation and the trade-offs that the government is facing with respect to its budget. However since 2003 the subsidies from the national budget were growing significantly, from 7 billion PLN in 2003 to 10 billion PLN in 2007. The growth in terms of GDP was apparent from even earlier on- spending on higher education grew from 0.6% of GDP in 1995 to 1.6% of GDP in 2009.

The rest is covered mainly by tuition fees charged to “unsubsidized” students. Each institution sets its own fee levels and it often varies them according to the program, as a result some of the most popular programs, like bachelor studies in Law, are often priced higher than others. As pointed out in OECD report (OECD, 2008) most of the institutions offering

“unsubsidized” programs make use of economies of scale by structuring these programs so that to make best use of resources already involved in teaching the same program to “subsidized” students. This allows public universities to set their tuition fees lower than private universities. Lower fees put public institutions in an advantage while competing for students, who make their decision based on both costs and the reputation of the program and institution. The advantages

(7)

in financing and longer history tend to favor public universities and hamper the development of the private ones.

It is only in the last 20 years that the creation of private higher education institutions has become possible. In the first half of the 90’s this possibility was not widely used and so the growth of private universities and colleges was quite slow- in 1995/1996 there were only 80 of them. However the pace has increased quite substantially and already 5 years later the number of private academia’s was more than double, reaching 195. The trend continued and there were more of these schools set up, reaching over 300 in the year 2005/2006. Lately, due to both the saturation of the market and the economic slowdown, the growth was rather minor. In the academic year of 2010/2011 there were as much as 328 private universities (MNISW, 2012).

The growth of private sector of higher education has been an important factor causing increase in number of students enrolled in tertiary education. This has also made it easier for people living in smaller cities, towns and villages to pursue higher education without incurring high costs of moving to a major city, where the public universities are situated. As a result students of private universities are poorer and live in greater distance from major cities than those who can enjoy a free access to tertiary education (Siwinska, 2011).

It is important to note that these universities are financed mostly by student fees. Additionally many of them acquire extra funds through the system of EU subsidies, through European Funds Programme of Human Capital and Regional Funds (Perspektywy, 2012) . They are also partly subsidized by local governments and the Ministry of Science and Higher

Education.

2.3 Statistics on Polish students and graduates

One can notice a steady increase in the statistics that compare the numbers of students and graduates to the population of their age group.

0 0,1 0,2 0,3 0,4 2004 2005 2006 2007 2008 2009 2010

New entrants as percentage of population of 19

and 20 year olds

19 year old 20 year old

(8)

Over 37% of 19 year olds (the age when a typical general high school’s pupil finishes his secondary education)entered higher education in 2010, this is almost 10% more than in 2004 when 28% of people this age started their tertiary education. The proportion was increasing from 2004 until 2008 and since then it fluctuated around 37%. There is a more stable pattern when it comes to 20 year olds(the age when a typical technical school’s pupil finishes his

secondary education), proportion of new entrants of this age is fluctuating around 16%. It noted highest values in 2005, 2006 and 2010.

The proportion of people aged 20-24 who are enrolled in tertiary education has been growing quite steadily for the last 10 years. The last years of 1990’s were much more dynamic in that respect. Through 1998 until 2010 there has been a growth of 19%. The fastest rates of growth in the last ten years were noted in 2005 and 2006.

0 0,05 0,1 0,15 0,2 0,25 0,3 0,35 0,4 0,45 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

Enrolled in tertiary education aged 20-24 as

percentage of corresponding population

(9)

As can be seen from the figure above the proportion of young people finishing higher education each year was rising from 2002 until 2010. The pace of the growth increased since 2007, which might be the effect of quite high growth of new entrants in 2003 through 2008.

3. Theoretical background

There are various theories that link education to the labor market. Two major ones are Human Capital Theory and Signaling Theory. The first one argues that schooling and training increase individuals skills and lead him to achieve higher productivity and through this allow him to earn higher wage. The second approach claims that educational achievement are an indication of underlying abilities and help the employers to select individuals with higher productivity. Following section describes these two theories and relates them to the research question.

3.1 Human Capital Theory

Among economists who write on Human Capital Theory, there are differences in approach and exact definition of the term “human capital”. One of the most widely used and standard

understandings is the one argued by Gary Becker, and this is also the one that will be followed in this paper.

Becker’s definition of an investment in human capital states that it comprises of “activities that influence future real income through the imbedding of resources in people” (Becker, 1964). The spectrum of such activities is quite broad and includes learning, both in school and during on-the-job training, as well as investing in ones’ health and in acquiring 0,00% 1,00% 2,00% 3,00% 4,00% 5,00% 6,00% 7,00% 2002 2003 2004 2005 2006 2007 2008 2009 2010

New graduates as proportion of population aged 20 to 29

(10)

information. Becker’s analysis of on-the-job training was quite extensive , however for the purpose of this paper attention is paid to his analysis of schooling.

First he defined school as an institution specializing in the production of training. He distinguished between different schools and how they focus on teaching more general set of skills or a more particular range. Later he noted that costs of the education are the sum of direct costs such as tuition and books, and indirect cost- the opportunity cost of working instead of studying. Using this definition of costs of learning net earnings of an individual are defined as difference between marginal productivity and the costs for given period (Becker 1964).

There is a difference between a stream of net earnings for a person who invests in human capital and for the person who does not. The relationship between age and earnings of a person who invests in human capital through activity lasting for certain period of time such as tertiary education (person A) and a person who does not (person B) is presented in the following graph. It can be seen that initially net earnings of a person B are higher than those of person A. However, after the investment is finished (at age G), and person A can enter the fulltime labor market, earnings of person A rise substantially. Person B’s net earnings are unchanged throughout (Becker 1964).

One can find that the cost of the investment in each period is a difference between net earnings of person A and person B, and the total investment cost is the present value of these differences.

∑

The return to such an investment can be expressed as difference between net earnings after the investment is finished and total return is the present value of these differences

∑

So in equilibrium the rate of return to education is going to equalize total costs and total revenues from this investment (Becker 1964).

(11)

The approach mentioned above can be applied if we use a simplifying assumption that person A does not gain marginal productivity before the investment is finished. So in case of higher education, a person who drops out of university after first or second year is treated as if he or she did not acquire any skills which would allow him or her to become more productive.

However, for the analysis of the investment in tertiary education that fully represents the human capital approach, the above mentioned assumption should be released. This is because Human Capital Theory argues that even incomplete studies can be defined as an activity that increases someone’s skills and knowledge. This could apply even to people who were forced to drop out from the study program because of failing too many courses- even they might have learned something.

If the assumption was released, one would have to take a closer look at how opportunity costs of a person change depending on the length of his studying at the university. One can expect that this would result in an increasing function of net earnings with respect to time. (That approach does not account for any obligations due to student loans and

scholarships, assuming that all costs of education are borne by an individual, and that once he quits studying he will not have to pay for his education anymore. This is quite applicable to Polish reality, where student loans are quite uncommon and one faces no extra charges if he decides to stop studying.) So the actual function we would expect looks more like this:

Human Capital Theory can explain the fluctuations in the proportion of high school graduates choosing to pursue higher education. It supports the intuitive prediction that lower return to education will decrease the number of students (Becker, 1964).

(12)

According to the framework presented above return to education could fall because during the same time spent at the university one learns less skills, or because the skills that he learns are becoming less useful in increasing his marginal productivity. Therefore Human Capital Theory predicts that wages of people who finished tertiary education compared to wages of those who did not will increase if the education system is improved. On the contrary if the rate of return to tertiary education is falling this can be a result of worse quality of education.

3.2 Signaling

The theory developed by Michael Spence argues that in conditions of imperfect information markets, signalling could be a successful strategy for achieving equilibrium. He uses the job market as an example, but argues that this strategy can work in other markets as well. The general idea is that because of the asymmetric information the recruiter is unable to assess the productivity of job candidates while each candidate knows his productivity. In order to promote a better equilibrium outcome candidates with higher productivity will choose to signal their potential, for example by achieving certain level of education. However, this kind of signalling can work only if the cost of educational achievement is negatively related with the productivity of an individual (Spence, 1973).

In a simple model there are two groups of potential employees, Group A with productivity equal to , and Group B with productivity equal to , where < . It is also the case that the cost of achieving a certain level of education is defined as 𝑐_𝐴 𝑦 for Group A and 𝑐_𝐵 𝑦/2 for Group B (𝑦 being the level of education).

In separating equilibrium the employer believes that individuals who achieve education level 𝑦∗_{are certainly members of Group B, while those who did not are certainly members of}

Group A. In this situation 𝑦∗_{is high enough to make the cost of achieving it too high for Group A}

members and low enough to make members of Group B invest in theirs education, this leads to a unique

equilibrium in which members of Group A do not achieve any education 𝑦 , and members of Group B achieve exactly the threshold level 𝑦∗_{. And the members}

of Group A will end up earning , while Group B members will obtain 𝑦∗_/2

(Spence, 1973).

(13)

Other equilibria are also possible, in which both groups chose to invest in education at the same level. This is called a pooling equilibrium and in this case employers cannot be certain about the productivity of an individual they

hire if they are basing their opinion solely on the education level. The wage will be set at the expected productivity of the whole population that is 𝐴

𝑁

⁄ 𝐵

𝑁 ⁄ , where 𝐴 is the number of individuals in

Group A, 𝐵 is number of individuals in

Group B and 𝑁 is the population size (Spence, 1973).

With negative relationship between productivity and cost of acquiring education being a

necessary condition for signalling to work, it becomes clear that if the educational reform allows people with lower productivity to face the same cost of education, it will make the signalling impossible. This is as saying that when reform allows less able people to finish tertiary education the signalling is hampered, and employers are forced to set a single wage level for both groups. In this scenario all people would benefit if they all stopped pursuing higher so the equilibrium is not efficient.

3.3 Signalling with Human Capital

The signalling model does not rule out the possibility for education to enhance human capital. Spence proposed a following model that takes into account the productivity gain resulting from investing in education. He assumes that the productivity of a worker from group A depends on the level of education and is described by function 𝑠_𝐴 𝐸 , and that of a worker from group B is 𝑠_𝐵 𝐸 . Both of the functions are increasing, but 𝑠_𝐵 𝐸 > 𝑠𝐴 𝐸 and 𝑠 𝐵 𝐸 > 𝑠 𝐴 𝐸 .

As earlier, cost of investing in the education is higher for individuals in group A, it is described by function 𝑐_𝐴 𝐸 , while the cost function for people from group B is 𝑐_𝐵 𝐸 . It is assumed that 𝑐_𝐴 𝐸 > 𝑐𝐵 𝐸 and 𝑐 𝐴 𝐸 > 𝑐 𝐵 𝐸 . Functions 𝑠𝐴 𝐸 and 𝑠𝐵 𝐸 are concave, while

functions 𝑐_𝐴 𝐸 and 𝑐_𝐵 𝐸 are convex, as a result net income functions 𝑁_𝐴 𝐸 𝑠𝐴 𝐸 𝑐𝐴 𝐸

and 𝑁_𝐵 𝐸 𝑠𝐵 𝐸 𝑐𝐵 𝐸 are concave.

It is also important to specify the function 𝑉_𝐴 𝐸 𝑠𝐵 𝐸 𝑐𝐴 𝐸 which describes the

net income of a member of group A if he is mistakenly recognized to be from group B (Spence, 2001).

(14)

The separating equilibrium for this model is described in the picture above. Equilibrium level of Group B is equal to 𝐸_𝐵. In this case, members of Group A are better off if they obtain level 𝐸_𝐴, then if they try to imitate Group B by choosing 𝐸_𝐵 (Spence, 2001).

The above figure represents the other scenario, in which the equilibrium value of the investment for Group B, 𝐸𝐵, is low enough to allow individuals from group A to invest in education and

achieve net income higher than if they invested only 𝐸𝐴. There are two possible results of this

situation- one is the emergence of a pooling equilibrium, the other is a separating equilibrium characterized by overinvestment in education by Group B to a point 𝐸∗

𝐵> 𝐸̅.

FIGURE 8

(15)

Which outcome will emerge depends on the size of Group A compared to size of Group B. If there are much less people in Group A, the outcome will be a pooling equilibrium- because the smaller the proportion of Group A individuals, the higher will be the expected productivity within the whole population, and it will not be beneficial for the Group B members to overinvest in education (Spence, 2001).

According to this model the difference in the cost of achieving education is crucial. If both groups find it as costly to achieve education, 𝑐_𝐴 𝐸 𝑐𝐵 𝐸 , then 𝑉𝐴 𝐸 𝑁𝐵 𝐸 and there is no

separating equilibrium, as members of group A are strictly better by imitating Group B level of education. Also in case there is no chance for efficient separating equilibrium, it might be the case that pooling equilibrium is more Pareto efficient then separating equilibrium with overinvestment (Spence, 2001).

When looking at the model in the context of this research it is important to notice that, just as with regular Signalling theory, when the reform of education decreases the cost of acquiring education, the separating equilibrium might be lost. Also in a similar manner to basic Human Capital Theory, this model predicts that the better the quality of the education the greater the benefit for those who pursue it.

In order to draw some predictions from this model one needs to know whether changes in costs or benefits occur equally for all groups or if they are stronger for one of them. If the reform makes it easier to get into university it might decrease the costs for a Group A members more than for Group B members. In this case pooling equilibrium might emerge.

If the reform brings higher quality of education it simultaneously increases benefits for both groups, though probably more productive Group B will benefit more than Group A. However most of the times if the quality is increased, it becomes more difficult to finish ones’ studies, therefore the costs of the investment in education increase, and in this case Group A might experience a higher increase in costs than Group B, as they are less able and for them studying requires more time and energy. Therefore, increasing the quality might have two different results on the labour market- it might create a separating equilibrium, or if the increase in costs faced by Group A is not big enough it might lead to a pooling equilibrium.

As can be seen from the discussion above, the implications of this model are ambiguous and one has to remember that it all depends on the magnitude of impacts that it has on each of the groups, which are often very hard to assess.

(16)

4. Methodology

In order to answer the research question following hypothesis is stated:

The return to higher education for people between 21 and 29 years old was lower in 2011 than in 2009 and 2007.

4.1 Data description

The hypothesis is tested by performing econometric analysis on the panel data from Diagnoza Społeczna. It is a project introduced in 2000 by The Council for Social Monitoring. Second survey took place after 3 years and since then they are taken regularly every 2 years. Surveys comprise questions about both economic and social issues and are performed at household and individual level. For this research data from 2007, 2009 and 2011 surveys of individuals are relevant. The analyzed variables are Age, Gender, Income, Master, FatherEdu and Fulltime Wave0907. The analysis is focused on comparing the effect of finishing a Master program rather than of finishing Bachelor because before the reform most programs offered at higher education institutions were uniform Master programs, therefore finishing university with Bachelor diploma was not common. Also in order to compare Master graduates only with those who finished secondary education, Bachelor and PhD graduates are excluded from the sample.

On top of that, the group is restricted to those who finished their education (either secondary or tertiary) in the 2 years prior to the survey. In this way the influence of years of professional experience is minimized, and the focus is placed at analyzing the impact of tertiary education on entering wage. The restriction also allows to distinguish between different cohorts of graduates, as precise data on the year of graduation is not given, it can only be inferred in this way.

For the purpose of further analysis the range of Age is restricted so that only people between 21 and 29 years old are taken into account. In order to perform econometric tests on the data Master and FatherEdu are transformed into a dummy variables. FatherEdu is a dummy

indicating whether individuals father finished higher education or not. Monthly income is adjusted for inflation so all the values can be compared to 2007 levels, and natural logarithm of this variable is calculated.

Distribution of natural logarithm of monthly income (further called Income)

0 .2 .4 .6 .8 1 D e n si ty 4 6 8 10 income FIGURE 10

(17)

in the sample resembles normal distribution. There are 6077 observations and the mean is 7.12 and the standard deviation is 0. 549.

The gender proportions in the analyzed sample were quite even, 46,6% of respondents were woman and 53,4% were man. While the majority- 90.5% had a fulltime job. Variable Wave0907 is a dummy variable that takes a value of 1 when the person was recorded in year2009 or 2007 and takes value of 0 if the person answered questions in 2011.

4.2 Description of models

In order to test the hypothesis two models were built and two separate tests were conducted.

4.2.1 Model A

𝑐 𝑠 Income- natural logarithm of real monthly income

Master- binary variable; 1=obtained Masters diploma, 0=did not finish higher education Gender-binary variable; 1=male, 0=female

Fulltime-binary variable; 1=works fulltime, 0=does not work fulltime

It is a trend in Poland that more women than men pursue higher education which makes variable Gender correlated with Master, however many studies proof that there is still some degree of discrimination so Gender is relevant in explaining Income, therefore it is worth controlling for gender effects in the sample. Fulltime employment has an obvious effect on income through influencing the number of hours worked per month.

For the estimators of the model to be consistent the problem of endogenity of Master needs to be solved. This is done by using Two Stage Least Squares regression. Following the practice mentioned in the literature (Blundell, Dearden, Goodman, & Reed, 2000) FatherEdu is used as an instrumental variable. Even though the instrument is often criticized as weak (Colm Harmon, 2003) it is the only variable available within data set which can explain the variation Master and be assumed to be uncorrelated with the error term.

The first stage of TSLS regression requires Master to be predicted using estimators of the function where instrumental variable and exogenous variables serve as independent variables and Master is a dependent variable. Since Master is a binary variable the equation of the first stage of TSLS takes a following probit form:

(18)

Next the predicted value of Master is found and applied in the second stage of regression in order to estimate the initial model (Wooldridge, 2002).

4.2.2 Model B

𝑐 𝑠 𝑠 Incomedif- difference between natural logarithm of real monthly income and natural logarithm of average real monthly income in the economy in a given year

Master- binary variable; 1=obtained Masters diploma, 0=did not finish higher education Wave0907-binary variable; 1=participant of 2007 or 2009 survey, 0=participant of 2011 survey

Master x Wave0907- interaction term between Master and Wave0907 Gender-binary variable; 1=male, 0=female

Fulltime-binary variable; 1=works fulltime, 0=does not work fulltime

In the same way as in the previous model, the problem of endogenity of variable Master has to be solved by using TSLS. In this case the probit model that will be used to predict Master takes a following form:

𝐸

The predicted value of Master is found and applied in the second stage of regression and the interaction term is created as a cross product of the predicted Master and Wave0907

4.3 Testing hypothesis

4.3.1 Model A

The above mentioned procedure is first applied to the full data set, then each of the steps is performed on data constrained to the year 2011 and to a group created by joining year 2007 and 2009. These regressions will take this form:

𝑐 𝑠

Coefficients that are obtained in these regressions are compared by performing Chow test for equality of coefficients. The null hypothesis of this test asserts that:

(19)

The test statistic is calculated according to following formula: _⁄

𝑁 𝑁 2 ⁄

Where:

is the sum of squared residuals, is the number of observations,

is the number of estimated parameters.

Chow test statistic follows F distribution with and degrees of freedom

(Gould, 2011).

4.3.3 Model B

The results of t-tests generated by Stata are analyzed, with special attention given to the result for the coefficient by the interaction term of variables Master and Wave0907.

The coefficient will show how did the income of those who graduate with Masters change between 2009 and 2011 and it should be able to account for the effects of the reform. It is expected this test will show that the coefficient by the interaction term is positive, as it would indicate that the return to tertiary education is lower after introduction of the reform than it was before it.

5. Data analysis

5.1 Model A

First the model for the whole set of data is calculated. Below are the results of the probit

regression which is used to predict the value of variable Master (the predicted variable is named Masterall). The variable FatherEdu is not a weak instrument, because the coefficients in the first stage regression are significant, as indicated by a Chi-squared test.

Following is the linear regression in which all the coefficients are different from zero at 1% significance level. The model estimates that having a Master degree results in 78% increase in

(20)

monthly income, man’s income is on average 23% higher and fulltime workers earn 41% more than part-time workers.

Next the model for the data gathered in 2011 survey is calculated. Below are the results of the probit regression which is used to predict the value of variable Master (the predicted variable is named Master11). The variable FatherEdu is not a weak instrument, because the coefficients in the first stage regression are significant, as indicated by a Chi-squared test.

Following is the second stage linear regression. All the coefficients are different from zero at 1% significance level. The model estimates that in 2011 having a Master degree resulted in 70% increase in monthly income, man’s income was on average 15% higher and fulltime workers earned 51% more than part-time workers.

. _cons 6.398891 .0337297 189.71 0.000 6.332766 6.465016 fulltime .4126219 .0272373 15.15 0.000 .3592247 .466019 gender .2343336 .0188022 12.46 0.000 .197473 .2711942 masterall .7813586 .0902463 8.66 0.000 .6044362 .9582811 income Coef. Std. Err. t P>|t| [95% Conf. Interval] Total 1503.33309 4987 .30145039 Root MSE = .51941 Adj R-squared = 0.1050 Residual 1344.63156 4984 .269789638 R-squared = 0.1056 Model 158.701538 3 52.9005127 Prob > F = 0.0000 F( 3, 4984) = 196.08 Source SS df MS Number of obs = 4988 . regress income masterall gender fulltime

(option pr assumed; Pr(master)) . predict masterall _cons -.8086861 .1193461 -6.78 0.000 -1.0426 -.574772 fulltime .4107718 .121701 3.38 0.001 .1722423 .6493013 gender -.3803429 .0789432 -4.82 0.000 -.5350686 -.2256171 fatheredu .8858322 .1204925 7.35 0.000 .6496714 1.121993 master Coef. Std. Err. z P>|z| [95% Conf. Interval] Log likelihood = -691.77402 Pseudo R2 = 0.0558 Prob > chi2 = 0.0000 LR chi2(3) = 81.79 Probit regression Number of obs = 1192 Iteration 3: log likelihood = -691.77402

Iteration 2: log likelihood = -691.77403 Iteration 1: log likelihood = -691.89292 Iteration 0: log likelihood = -732.66758 . probit master fatheredu gender fulltime

(21)

Finally the model for the data gathered in 2009 and 2007 survey is calculated. Below are the results of the probit regression which is used to predict the value of variable Master (the predicted variable is named Master0907). The variable FatherEdu is not a weak instrument, because the coefficients in the first stage regression are significant, as indicated by a Chi-squared test.

Following is the second stage linear regression. All the coefficients are different from zero at 1% significance level. The model estimates that in 2009 and 2007 having a Master degree resulted in 95% increase in monthly income, man’s income was on average 28% higher and fulltime

workers earned 36% more than part-time workers.

_cons 6.509536 .0484246 134.43 0.000 6.414563 6.604508 fulltime .5089723 .0437123 11.64 0.000 .4232418 .5947028 gender .1585585 .0289061 5.49 0.000 .1018665 .2152505 master11 .5046472 .1278198 3.95 0.000 .2539611 .7553333 income Coef. Std. Err. t P>|t| [95% Conf. Interval] Total 511.946043 1854 .276130552 Root MSE = .49486 Adj R-squared = 0.1131 Residual 453.292828 1851 .244890777 R-squared = 0.1146 Model 58.6532144 3 19.5510715 Prob > F = 0.0000 F( 3, 1851) = 79.84 Source SS df MS Number of obs = 1855 . regress income master11 gender fulltime if year==2011

(3133 missing values generated) (option pr assumed; Pr(master)) . predict master11 if year==2011

_cons -1.001327 .21146 -4.74 0.000 -1.415781 -.5868724 fulltime .521622 .2130115 2.45 0.014 .1041272 .9391169 gender -.4159867 .1327312 -3.13 0.002 -.6761351 -.1558382 fatheredu .9951292 .1883125 5.28 0.000 .6260434 1.364215 master Coef. Std. Err. z P>|z| [95% Conf. Interval] Log likelihood = -242.6727 Pseudo R2 = 0.0760 Prob > chi2 = 0.0000 LR chi2(3) = 39.94 Probit regression Number of obs = 443 Iteration 4: log likelihood = -242.6727

Iteration 3: log likelihood = -242.6727 Iteration 2: log likelihood = -242.67273 Iteration 1: log likelihood = -242.77643 Iteration 0: log likelihood = -262.64053

(22)

Based on the above regressions it is possible to test whether the coefficients in the regression based on 2011 data are significantly different from those based on data from 2007 and 2009. Chow test statistic is found to be:

2 2 2 _⁄

2 2 2

⁄

Critical value for α=0.01 is F(4,∞)= 3.319<13.671, therefore it can be concluded that the coefficients of the two regression are unequal at 1% significance level. It can be inferred that having Master degree increased the monthly income by higher percentage in 2007 and 2009 than in 2011 and therefore that the return to education fell between 2009 and 2011. Also the income gap between man and women is smaller, while working fulltime has even greater impact on income in 2011 than it had in 2009 and 2007.

_cons 6.324436 .0460907 137.22 0.000 6.234065 6.414807 fulltime .357149 .0344309 10.37 0.000 .2896395 .4246584 gender .2751431 .0242968 11.32 0.000 .2275038 .3227824 master0907 .9529273 .1221514 7.80 0.000 .7134223 1.192432 income Coef. Std. Err. t P>|t| [95% Conf. Interval] Total 978.004042 3132 .312261827 Root MSE = .52934 Adj R-squared = 0.1027 Residual 876.734031 3129 .280196239 R-squared = 0.1035 Model 101.27001 3 33.7566701 Prob > F = 0.0000 F( 3, 3129) = 120.48 Source SS df MS Number of obs = 3133 . regress income master0907 gender fulltime if year!=2011

(1855 missing values generated) (option pr assumed; Pr(master)) . predict master0907 if year!=2011

_cons -.7157446 .1453866 -4.92 0.000 -1.000697 -.4307922 fulltime .3619263 .1492174 2.43 0.015 .0694656 .6543871 gender -.3604506 .098446 -3.66 0.000 -.5534013 -.1674999 fatheredu .8343078 .1578074 5.29 0.000 .5250109 1.143605 master Coef. Std. Err. z P>|z| [95% Conf. Interval] Log likelihood = -447.07262 Pseudo R2 = 0.0468 Prob > chi2 = 0.0000 LR chi2(3) = 43.88 Probit regression Number of obs = 749 Iteration 3: log likelihood = -447.07262

Iteration 2: log likelihood = -447.07262 Iteration 1: log likelihood = -447.12188 Iteration 0: log likelihood = -469.01124

(23)

While drawing conclusions it is important to note, that the test used is only able to compare full models and is not testing the exact change in the impact of obtaining master degree on income. All other variables that help to explain the income are also taken into account and might

influence the results. So while it can be said that the structural break was observed, it is not possible to assess the change in the impact that our variable of interest- Master has on income.

5.2 Model B

Below are the results of the probit regression which is used to predict the value of variable Master (the predicted variable is named MASTER). The variable FatherEdu is not a weak

instrument, because the coefficients in the first stage regression are significant, as indicated by a Chi-squared test.

Following is the linear regression in which all the coefficients are different from zero at 1% significance level. The model estimates that having a Master degree results in 87% increase in monthly income, man’s income is on average 23% higher and fulltime workers earn 41% more than part-time workers. Furthermore income of a Master graduates was approximately 16% higher in 2011 than in previous years.

Since the change in income that results from the economic growth is already accounted for by using a percentage difference from the mean income as the dependent variable, therefore a possible explanation is that an increase in return to education occurred after introduction of the reform. This would mean that the hypothesis of this paper is rejected.

Of course there are some imperfections in the way in which economic growth and resulting increase in incomes was accounted for. The estimates of average real income that were used were not age specific and therefore if the dynamism of incomes in different age groups varied in these years, that could not be accounted for by this model.

Even after realizing the possible flaws in the test, its results are very important. They indicate that the return to higher education increased in the time period that is analyzed and could therefore suggest that the reform of education improved the quality of education and did not significantly hamper its usefulness in signaling.

(24)

6. Evaluation and conclusions

The paper described the education system in Poland and the changes introduced in the reform while paying attention to factors that may be crucial for explaining the job market situation of the graduates. It can be seen that, the reform of education which was guided by the aim of increasing tertiary education participation rates have achieved both positive and negative outcomes.

As mentioned in OECD reports and by many Polish critics of the education system, the reform increased the participation rates, but did not solve the problem of unequal chances faced by those with lower incomes and from lower social groups. Also, the system of financing

encourages public universities to increase the number of students rather than improve the

_cons 6.342407 .0333777 190.02 0.000 6.276972 6.407842 fulltime .4115233 .0269487 15.27 0.000 .3586919 .4643547 gender .2313968 .0186053 12.44 0.000 .1949223 .2678713 MasterW~0907 -.1622965 .0491492 -3.30 0.001 -.2586507 -.0659423 Master .8776553 .0941794 9.32 0.000 .6930223 1.062288 incomediff Coef. Std. Err. t P>|t| [95% Conf. Interval] Total 1476.48269 4987 .29606631 Root MSE = .51389 Adj R-squared = 0.1080 Residual 1315.92387 4983 .264082654 R-squared = 0.1087 Model 160.558823 4 40.1397059 Prob > F = 0.0000 F( 4, 4983) = 152.00 Source SS df MS Number of obs = 4988 . regress incomediff Master MasterWave0907 gender fulltime

. generate MasterWave0907=Master*wave0907 (option pr assumed; Pr(master))

. predict Master _cons -.8086861 .1193461 -6.78 0.000 -1.0426 -.574772 fulltime .4107718 .121701 3.38 0.001 .1722423 .6493013 gender -.3803429 .0789432 -4.82 0.000 -.5350686 -.2256171 fatheredu .8858322 .1204925 7.35 0.000 .6496714 1.121993 master Coef. Std. Err. z P>|z| [95% Conf. Interval] Log likelihood = -691.77402 Pseudo R2 = 0.0558 Prob > chi2 = 0.0000 LR chi2(3) = 81.79 Probit regression Number of obs = 1192 Iteration 3: log likelihood = -691.77402

Iteration 2: log likelihood = -691.77403 Iteration 1: log likelihood = -691.89292 Iteration 0: log likelihood = -732.66758 . probit master fathered gender fulltime

(25)

quality of studies. On top of that, comes the problem of the cost advantage of public universities, which hampers the competition between public and private colleges, and reduces the extent to which students can “vote by their feet” while choosing their program of studies.

Data analysis provided ambiguous outcomes, with model A implying that the return to education was lower in 2011 than in previous years and model B suggesting a contradicting conclusion. The Chow test used in model A has indicated that there was a structural break in the 2009-2011 period and the regressions used to construct this test reveal that the return to education was decreased in that period. The second model shows that the change that occurred in this time period was positive- that is in 2011 the income of master graduates was higher than in previous years..

Having these results in mind it is important to realize the imperfections in the econometric study. To begin with, Chows test compares more than just the impact of higher education, it also accounts for the change in importance of other variables that are used in these regressions. Therefore the test is not narrow enough to conclude with certainty that the return to education was decreased in the period analyzed. On top of that comes the problem with the data set used. Data on the exact date of graduation from university was not available from these surveys. Furthermore, details on school performance and the years spent on education could be very helpful in shaping a more accurate model.

Summing up, the above research suggests that the education reform in Poland increased the return to education of the university graduates. However, since results of Model A indicate otherwise, definite answer to the research question cannot be given. There is a lot of room for improvement, and more accurate results could be obtained if a survey focused on this particular issue was held.

(26)

Bibliography

Dziennik. (2012). Retrieved from http://gospodarka.dziennik.pl/praca/artykuly/414967,bezrobocie-wsrod-absolwentow-moze-przekroczyc-30-proc-w-2013-roku.html

MNISW. (2012, 11 23). Retrieved from http://www.nauka.gov.pl/szkolnictwo-wyzsze/dane-statystyczne-o-szkolnictwie-wyzszym/

Perspektywy. (2012, 12 20). Retrieved from

http://www.perspektywy.pl/index.php?option=com_content&task=view&id=4059&Itemid=8 75

Arrow, K. J. (1973). Higher Education as a filter. Journal of Public Economics, 193-216.

Becker, G. (1964). Human capital: A theoretical and empirical analysis, with special reference to education. New York: National Bureau of Economic Research.

Blundell, R., Dearden, L., Goodman, A., & Reed, H. (2000). The Returns to Higher Education in Britain: Evidence From a British Cohort. The Economic Journal, 82-99.

Carneiro, P., Heckman, J., & Vytlacil, E. (2011). Estimating Marginal Returns to Education. American Economic Review, 2754-2781.

Cline, H. M. (1982). The measurement of change in the rate of return to education: 1967-75. Economics of Education Review, 275-293.

Colm Harmon, H. O. (2003). The Returns to Education: Microeconomics. Journal of Economic Surveys, 115-156.

Gould, W. (2011, 07). www.stata.com. Retrieved 04 26, 2013, from

http://www.stata.com/support/faqs/statistics/computing-chow-statistic/ MEN. (2000). Reforma systemu edukacji, szkolnictwo ponadgimnazjalne. Warszawa.

Milejska, M. K. (2007). Zreformowany system edukacji i jego wpływ na kształcenie i wychowanie dzieci w szkołach publicznych. Pedagogical study, Katowice.

OECD. (2008). OECD Reviews of Tertiary Education – POLAND.

Puukka, J., Dubarle, P., Goddard, J., Hazelkorn, E., & Kuczera, M. (2013). Higher education in Regional and City Development: Wrocław, Poland 2012. OECD.

Rozmus, A., & Pado, K. (n.d.). www.efinanse.com. Retrieved 03 20, 2013, from Finansowanie szkolnictwa wyższego w Polsce – wybrane dylematy i sugerowane rozwiązania: http://www.e-finanse.com/artykuly_eng/114.pdf

Siwinska, B. (2011, 10 16). University World News. Retrieved 05 02, 2013, from POLAND: Private Higher Education under Threat:

http://www.universityworldnews.com/article.php?story=20111015213651212

(27)

Spence, M. (2001). Signaling in retrospect and the Informational Structure of Markets. Prize Lecture. Stockholm.

Measuring the impact of the reform of the education system on the return to higher education among young people in Poland

Bachelor thesis: