university choice, equality, and academic performance275129/fulltext01.pdf · university choice,...
TRANSCRIPT
University Choice, Equality, and Academic Performance
Acta Wexionensia No 188/2009 Economics
University Choice, Equality and Academic Performance
Susanna Holzer
Växjö University Press
University Choice, Equality, and Academic Performance. Thesis for the de-gree of Doctor of Philosophy, Växjö University, Sweden 2009. Series editor: Kerstin Brodén ISSN: 1404-4307 ISBN: 978-91-7636-681-3 Printed by: Intellecta Infolog, Göteborg 2009
Abstract Holzer, Susanna (2009). University Choice, Equality, and Academic Perform-ance. Acta Wexionensia No 188/2009. ISSN:1404-4307, ISBN: 978-91-7636-681-6. Written in English. This thesis consists of three essays that examine issues on university attendance behavior, factors of university completion, and the labor market value of a uni-versity diploma in Sweden.
Essay [I] analyzes how the rapid expansion of higher education that increased the geographical accessibility to higher education in the 1990s affected univer-sity enrollment decisions among various socioeconomic groups of young adults in Sweden. The empirical findings show that the probability of enrollment in uni-versity education increases with accessibility to university education. The results also indicate that accessibility adds to the likelihood of attending a university within the region of residence. Access to higher education more locally seems to have decreased the social distance to higher education, meaning that the option of attending higher education, as compared to entering the local labor market af-ter upper secondary school, has become a more common and a more natural al-ternative for more socioeconomic groups in society.
Essay [II] compares the performance of students in universities built before and after the large decentralization and expansion of the higher educational sys-tem in Sweden, starting in the late 1970s. Two outcome measures are used: (i) whether or not the student has obtained a degree within seven years after she ini-tiated her studies; and (ii) whether or not she obtained 120 credit points (the re-quirement for most undergraduate degrees) within seven years. Controlling for several background variables as well as GPA scores in a binomial probit model, we show that students at old universities are about 5 percentage points more likely to get a degree and about 9 percentage points more likely to obtain 120 credit points. However, in an extended bivariate model where we consider selec-tion on unobservables into university type, we cannot reject the possibility of no difference in performance between the two university types.
Essay [III] analyzes the labor market value of a university diploma (sheep-skin) in Sweden. In contrast to previous studies, this study only focuses on Swedish university students who have three years of full time university educa-tion or more − where some have obtained a university degree, others not. The re-sults show that for male students, the wage premium of possessing a degree, i.e. the sheepskin effect, is roughly 5-8 percent. For women, it is about 6-7 percent for those who have completed four years of fulltime or more. For students who attended a more prestigious university in the metropolitan areas in Sweden and majored in the natural sciences, a sheepskin effect of roughly 13 percent for men and 22 percent for women is traced. However, this result did not hold among stu-dents who attended a newer university outside the metropolitan reas and/or ma-jored in the social sciences.
Keywords: Higher education, university enrollment; university choice; accessi-bility; university completion; selection bias; propensity score matching, sheep-skin, human capital.
To my parents,
Irena & Franz Holzer
Contents
Acknowledgements
Introduction
Essay I: The Expansion of Higher Education in Sweden
and the Issue of Equality of Opportunity
Essay II: University Choice and Academic Success in
Sweden
Essay III: Are there Sheepskin Effects in the Return to
Higher Education in Sweden?
Svensk sammanfattning
Acknowledgments Less than two hours after I got my acceptance call from Karlstad University (I was on their reserve list), I sat on the train to Karlstad – little did I know that this was the start of one of the longest journeys in my life. With a short stop at Karl-stad University, followed by several years at Lund University, to finally end up at Växjö University with this dissertation … There are so many people to whom I am grateful who have crossed my path over the years that I hardly know where to start… First of all, there are two people that have my deepest gratitude – my supervi-sor Professor Mårten Palme and co-supervisor Dr. Håkan Locking. The German word for supervisor is Doktor Vater (doctor father). The parent metaphor em-braces, I think, so much more in terms of patience, support and humanity, than just a supervisor’s professional guidance in how to write a thesis. Just like ordi-nary parents you have shown a tremendous amount of patience and given me in-valuable support in both my thesis writing and in boosting my self-confidence when needed. For putting up with my crazy ideas, yet kindly guiding me back on track again… Mårten, your sharp insights and comments have been the key ele-ment when turning my thoughts on higher education in Sweden into the three es-says in this dissertation. Håkan, for ALWAYS being there for me and always finding the time to answer me whenever I needed to ask you for help – on almost anything and at almost any hour of the day. Thank you both for putting up with me and all the versions of this thesis, and for patiently guiding me to this final product. To Professor Inga Persson and Professor Curt Wells at Lund University, who both encouraged me to consider graduate school in the first place. A special thanks to Professor Harald Niklasson and Professor Jan Ekberg – who offered me a position at Växjö University and who encouraged me and al-lowed me to find my own way in research. To all my friends and colleagues at the School of Management and Econom-ics at Växjö University – especially at the Department of Economics and Statis-tics – for putting up with me for so many years, for sharing the joy and the ag-ony of economic sciences, and for reading and commenting on all versions of my papers in thesis − here it goes: Zangin Ahmad, Lars Andersson, Lina Andersson, Dominique Anxo, Abdullah Almasri, Lars Behrenz, Mårten Bjellerup, Lennart Delander, Jan Ekberg, Mats Hammarstedt, Hans Jonsson, Gösta Karlsson, Joel Karlsson, Yushu Li, Thomas Lindh, Håkan Locking, Monika Hjeds Löfmark, Andreas Mångs, Jonas Månsson, Maria Mikkonen, Maria Nilsson, Harald Nik-lasson, Mikael Ohlson, Osvaldo Salas, Klas Sandén, Ghazi Shukur, Carl-Erik Sjödahl, Jonas Söderberg, and Lars Tomsmark – THANK YOU ALL!
To my beloved fiend Karin Olsson, for whom I am so grateful. Through you I met my other dear friends in political sciences, Johanna Jormfeldt, Tobias Bro-mander, and Otto Petterson. Guys − what would these last few years have been without our breakfast club and lunch meetings? – Thanks for bringing out the laughter in me! Through our tight friendship over the years, I have gotten to know and have almost become a part of your department/school too – so thank you all at the School of Social Sciences! To my beloved friend Marie Eriksson, for our energizing power walks/talks around Växjösjön – I do not know how I would have survived without it! To Nick Barr, Peder Pedersen, Chris Taber and Jeff Smith – inspiring Profes-sors that I have stumbled across on conferences and seminars – thanks for taking the time to comment on early versions of my papers and inspiring me to keep on working! To the participants at the meetings of EEA in Budapest (2007), and Barcelona (2009), EALE in Prague (2006), SOLE in Boston (2009), the Arne Ryde Symposium in Lund (2006), Higher Education Systems, Decentralization and Educational Outcomes in Novara (2008), and seminars at Aarhus University – thank you all for your insightful discussions and comments on my work. To Professor Anders Björklund, for fruitful and helpful comments on my li-centiate thesis in 2006, and to Professor Mikael Lindahl, who was the appointed opponent at my final seminar of this thesis earlier this year, whose insightful comments have helped me improve this final version of the thesis. To Christina Lönnblad for improving my English. Financial support by Växjö University, Jan Wallander and Tom Hedelius’ Research Foundation, the Swedish Research Council, and the Sveriges Riksbank is gratefully acknowledged. Graduate studies can make the most social person asocial, which is why the presence of stubborn and beloved friends that refuse to give up on you becomes even more important. For this, Maria Wennerbo with family (my long lasting childhood friend), Annika and Per Lundfors, Paula Hallonsten and Ingvar Lind-holm, Marie and Magnus Eriksson, Mark Vooreveld, Monika and Martin Löf-mark – I am humble and grateful. An meine Familie in Polen und in Österreich – die Dorner, Glaner, Holzer, Leitner, Palme, Riebler, und die Trzcinski – Ich bin stolz ein Teil von euch zu sein! Last but not least – my beloved parents: Irena and Franz Holzer. –Hoppa inte i sjön bara för alla andra gör det, brukar du pappa säga. Exakt vad det betyder vet nog bara du, men min tolkning är att jag ska våga gå min egen väg och följa min egen övertygelse. En sak är säker, även om jag vågat och bitvis lyckats med di-verse galenskaper i mitt liv, är det för att ni båda funnit vid min sida med ert ovillkorliga stöd och kärlek – och för detta är jag er för evigt tacksam! For those I have forgotten to mention and for all of you that I have men-tioned: – Thank you all for crossing my path, for being such a source of inspiration, giv-ing me support and faith – and most of all for giving me joy. Susanna Holzer Stockholm, September 23, 2009.
Introduction When several reports showed that Sweden had been falling behind in relevant comparisons concerning national levels of higher education in the 1980s, strong criticism was raised towards the Swedish government for its centrally monitored and supply-side oriented higher education being under dimensioned.1 Sweden risked losing competence within several academic professions, if the sector of higher education did not expand its undergraduate education to compensate for large scale retirements in the 1990s.2 As a result, the sector of higher education became the target of a very substantial and state-funded expansion that started in the early 1990s.3 In less than 15 years, the student body grew from 150,000 stu-dents at the end of the 1980s, to more than 330,000 students in early 2000 – cor-responding to roughly 50 percent of all the younger birth cohorts in Sweden. The point of departure for this dissertation is this rapid expansion of the sector of higher education in Sweden in the 1990s. This thesis consists of three self-contained essays, each of which has a different approach to investigate the out-come of this expansion. The economic literature on higher education mainly examines the relationship between college (university) quality and labor market outcomes.4 The dominat-ing part of the literature is based on data for the United States; see e.g. Brewer and Ehrenberg (1996), Brewer et al. (1999), Berg Dale and Krueger (2002) and
––––––––– 1 See af Trolle (1990), UHÄ (1989), OECD (1993) and Hammarström (1996). 2 See The National Agency for Higher Education (HSV) (1998, p15f) for a brief discussion and over-
view. 3 The economic recession that Sweden and some other industrialized countries suffered in the early
1990s which made the labor market harsh, for especially for young adults, did have an impact on the speed of the expansion.
4 The theory that in basic all research on higher education rely on is the theory of human capital, see seminal work by Becker (1964), Schultz (1961) and Mincer (1974). The theory says that the deci-sion whether or not to participate in an education and for how many years is equivalent to making an investment, i.e. an investment in one’s own human capital. The rational individual has to decide if it is worth the forgone earnings he/she will suffer while studying, instead of participating in the labor market, in order to enhance his/her future chances of better labor market opportunities and la-bor market outcomes. In a computer age where individuals’ educational histories and wage devel-opments are well recorded, labor market economists have been using this information at an escalat-ing speed to provide overwhelming empirical evidence of the years invested in education being positively correlated with an increasing economic return (see Card (1999) and Harmon et al. (2003) for an overview of the literature.) However, a common feature of most of this research is that edu-cation is treated as a homogeneous good and the schools as “black boxes” where the educational production is taking place. Only in the last 15 years have empirical researchers started to pay more attention to institutional characteristics and other factors that might affect educational outcome – which, in turn, might affect later labor market outcomes.
Black and Smith (2004, 2006). Their overall conclusion is that not only does col-lege education have an impact on labor market outcomes, college quality (col-lege choice) is also of importance. Although they all have different ways of reaching their empirical findings, they all end up with similar results − that an investment in college studies has a positive effect on an increased labor market outcome up to about 20-50 percent, and attending high-quality colleges rather than low-quality colleges generally increases wages in the range of 5−15 percent. When classifying universities into quality groups in a Swedish context , the most common way of proceeding is to use the fact whether the university was estab-lished before or after the reform of the sector of higher education in 1977 – the universities are then commonly referred to as old and new universities, respec-tively. Sweden has had a similar development of the sector of higher education as the United States in terms of expansion and decentralization, in establishing new smaller (and for the majority of the population) more local regional univer-sities outside the metropolitan areas as a complement to the fewer old universi-ties (mostly situated in the metropolitan areas in Sweden).5 The newer establish-ments were initially offering shorter and a more limited choice of educational programs or courses. Over the years, however, there has been an increase in the amount of educational choices and their length. In contrast to the US higher edu-cational system, there are four important traits of the Swedish higher educational system that we must consider when interpreting outcomes of higher education in Sweden in the 1990s and later. Higher education is i) centrally monitored and quality controlled by the government; ii) state funded; iii) free of charge for the student; and iv) the new universities are, like the old universities, allowed to award students bachelor’s and master’s degrees on a regular basis. For Swedish evidence on estimating the labor market outcomes of a university education in general or a specific university choice, see e.g. Wadensjö (1991), Gartell and Regnér (2004), Lindahl and Regnér (2005), and Eliasson (2006). The studies that group universities into quality groups based on old versus new uni-versities find that graduates from old universities on average receive an earnings premium that is about 0-7 percent higher than that of their fellow peers at the new universities. When controlling for specific universities (and in some cases also the subject majors) the earning effects are in the range up to 20 percent or more. A common feature of most (if not all) of these empirical studies on Swedish data is that they only consider university graduates. In Sweden, roughly 40 percent of all students that enter higher education end up with a degree after seven years or more. That less than half of the students body ends up with a degree is not typi-cal for Sweden alone – on the contrary. Manski and Wise (1983) reported in-creasing drop out rates from US colleges already in the early 1980s, and OECD ––––––––– 5 Throughout this text and this thesis, I will refer to all Swedish establishments of higher education as
universities, based on the fact that all of them are offering educations up to the master’s level – which is not the case for some of the colleges in the US.
(2008) shows this to be quite common across most industrialized countries. What is special for Sweden, though, is that more than one-third of all students who ne-ver obtain a university degree are recoded as having three years or more worth of higher education, and most of it at intermediate or higher levels (see SCB (2007)). In the worst case scenario, this means that prior empirical research has underestimated the real labor market outcome of an investment in a university education, since these highly educated individuals with no formal degree are put in the control group together with upper secondary graduates. This raises the question of what is being measured in these empirical studies; labor market out-comes due to an educational investment − or the labor market value of a formal university degree? These issues are discussed and some of them are analyzed in Essay III of this thesis. The second drawback of only considering graduates, especially in terms of esti-mating the labor market outcomes associated with university qualities, is that the research made does not control for where the student has actually obtained his/her education – only which university that has issued the actual diploma is considered. About one third of all students in Sweden move between universi-ties.6 It is not uncommon that students that attend a regional (and in some opin-ions less prestigious) university move to an older (and more prestigious) univer-sity in their fourth year, in order to receive a master’s diploma from the latter university. What do any differences between labor market outcomes measure in this case? Part of this could, of course, be due to university quality, but any dif-ferences in wages could just as well be due to unobserved individual characteris-tics that made some students move – and others not. To analyze the institutional impact on educational performances – which labor market economists argue to be reflected in wage differences or other labor market outcomes later on – the in-stitutional impact on actual educational performances by university students is discussed and analyzed in Essay II. Besides the labor market motive for expanding higher education, egalitarian mo-tives are often used to motivate the state funding of the entire educational system in Sweden (from elementary education to tertiary education).7 The egalitarian idea of improving educational opportunities8 is often interpreted in terms of im-
––––––––– 6 See raw LINDA with LADOK information and Essay II in this thesis. Similar results are reported
on US students by Light and Strayer (2000). 7 In Sweden, the powerful Swedish labor movement in association with the Social Democratic Party
did already in the 1940s stress the importance of providing a nation-wide education. One of the lar-gest threats to a democratic society and the sustained wealth of a nation was, according to them, differences in educational standards among its population (see e.g. Erikson and Jonsson (1993:31ff)).
8 Equality of opportunity, or in this case equality of educational opportunity, relates to the extensive economic literature on intergenerational mobility, which looks at the relation between parents’ eco-nomic outcomes and the future economic outcome of their children. A high intergenerational mo-bility indicates a low relation between parents’ economic outcomes and the future economic out-come of their children. A weak relation between the child and parental characteristics indicates a high equality of opportunity in society, saying that children from different social origins have simi-
proving access for the population.9 A political action by Brint and Karabel (1989) and Rouse (1995) is referred to as democratization. That is, nearby easy-access would give all who want to participate in an educational program the op-portunity to do so or encourage those who might otherwise not have participated. This egalitarian motive for state funding of the expansion of the sector of higher education in Sweden in the 1990s and how increased access has affected differ-ent socioeconomics enrollment behavior before and after the expansion consti-tute the point of departure in Essay I. The empirical material used in all three essays in this thesis is based on regis-tered data provided by Statistic Sweden (SCB). The core database is the Longi-tudinal INdividual DAta for Sweden (LINDA), which is a representative sample of three percent of the population in Sweden and their household members (see Edin and Fredriksson (2000) for a description). LINDA goes back to 1968 and the years up until 2006 are used here. In addition, information from the Swedish upper secondary school register which contains information on final grades from upper secondary school, and from the Swedish higher education register, which reveals any activities within the higher educational system, has been attached to the main subjects in LINDA.
Essay I The empirical focus of this essay is to investigate three questions: (a) to what ex-tent did individuals living in the municipalities of the new universities of the late 1970s and later become more inclined to attend higher education in the 1990s; (b) to what extent did the choice of attending a university in general become less dependent on individual and family background characteristics in the 1990s (i.e. was there an increase in intergenerational educational mobility?); and (c) did the new universities divert students from lower socioeconomic backgrounds from at-tending older universities? With mostly shorter vocational-oriented programs be-ing offered at the new universities, did they divert potential students who might otherwise have attended an old university, i.e. who would otherwise have in-vested in longer educations? This effect is referred to as a diversion effect by Rouse (1995, 1998), and Leigh and Gill (2003, 2004).
lar chances in life; see seminal work such as Becker and Tomes (1979, 1986). Increased access more locally is considered to improve this intergenerational mobility.
9 In philosophical economics, there is an ongoing discussion on whether or not opportunity and ac-cess are two different things; see Roemer (2006). The main issue in modern egalitarian policy is, however, how to divide and obtain welfare, but the way of doing this differs. The difference in views was often on how much responsibility we can put on the individual alone, i.e. the cause of her actions that leads to a certain welfare outcome, and how much responsibility we can put on so-ciety as a whole, i.e. the circumstances/milieu in which she makes her decisions. This, in turn, can affect the welfare outcomes of her educational choices.
A sample of 299,944 individuals is extracted from LINDA (1968-2001), which is divided into 25 cohorts of 18-year-olds for the years 1977-2001. The cohorts of 1977-1989 represent the 1980s and the remaining cohorts represent the 1990s. All youths are conditioned on having at least one parent in LINDA, which pro-vides relations between parents and their youths. Two probit models are esti-mated, one on the democratization and one on the diversion effect. The results show that living and growing up in the area of a rapidly growing educational institution seems to have a strong positive effect on overall univer-sity attendance. With an average impact of roughly six percentage points, this indicates that the expansion of the higher educational sector in Sweden has had an overall democratization effect (the local effect at the new universities is roughly 41 percentage points). A more local access to higher education seems to have decreased the social distance to higher education, meaning that the option of attending higher education, as compared to entering the local labor market af-ter upper secondary school, has become a more common and possible alternative for more socioeconomic groups in society. The fact that more individuals chose to attend higher education further away in the 1990s among those living in the areas of new universities could be interpreted as the choice of moving being con-sidered less costly (a smaller risk) once the decision of attending higher educa-tion has been made in the first place and that the new universities do not have any diversion effect on overall university attendance. The relatively largest growth occurred among students whose parents had upper secondary school as their highest education, which can bee seen as some indications of political suc-cess in equalizing educational opportunities.
Essay II In this essay, a comparison is made between the academic performance of stu-dents that attended a university that was established before or after the large de-centralization and expansion of the higher educational system in Sweden which started in the late 1970s. Two outcome measures are used: (i) whether or not the student has obtained a degree within seven years after she initiated her studies; and (ii) whether or not she obtained 120 credit points (the requirement for most undergraduate degrees) within seven years. To model how university type might affect educational outcomes, two probit models are employed. First, we use a binomial probit model, where we control for possible heterogeneity by including observed characteristics as regressors. Second, the effect of selection on unobservables that might affect the educational outcome which is correlated with initial university choice is estimated with a bivariate probit model. The empirical material used in this study is a sample of 5,565 individuals that are extracted from LINDA. All those individuals entered a Swedish university for the very first time in the years 1996-1999, i.e. approxi-mately three percent of all new enrolled students at the time.
Controlling for several background variables as well as GPA scores in a bino-mial probit model, students who attended the old universities are about 5 per-centage points more likely to get a degree and about 9 percentage points more likely to obtain 120 credit points. However, when controlling for selection on unobservables with a bivariate probit model, we found that on the probability of completing 120 credit points or more, the selection parameter turned out to be significantly different from zero and the coefficient for an old university was not significantly different from zero. This means that we cannot rule out the possibil-ity that the higher probability of obtaining 120 credit points at older universities is attributed to selection on unobservables.
Essay III Although the positive relationship between educational investments and earnings is one of the most established relationships in the social sciences, we still argue about what exactly in the educational investment affects earnings – is it years of schooling, credentials, or perhaps a mixture of them both? Mincer (1974) argued that earnings are mainly affected by the number of years of education, while other researchers point at the importance of the acquisition of credentials by me-ans of formal degrees. In the latter case, an accredited worker earns more than its non-accredited counterparts, a phenomenon referred to as a sheepskin effect; see e.g. Hungerford and Solon (1987), Jaeger and Page (1996), and Flores-Lagunes and Light (2007). Two questions are raised in this study: First, is there a general difference in the economic outcome for former university students with a degree, as compared to those with no formal degree, i.e. is there a sheepskin effect? Second, does the sheepskin effect vary within groups of university types, subject majors, and edu-cational programs? Altonji (1993), Altonji et al. (2005), Arcidiacono (2004), Brewer et al. (1999), and Dale and Krueger (2002) all point out that information on school quality and educational programs, i.e. on university choice and univer-sity major, is of importance when the objective is to value the returns to univer-sity education. In contrast to most previous studies on sheepskin effects on the labor market, this study only focuses on university students in Sweden with about the same number of years of higher education (three years or more) and differences in labor mar-ket outcome due to whether or not they have a degree (bachelor’s or higher). In addition to university choice and within-university choices, individual and fam-ily characteristics and ability are also considered. A random sample of 2,363 in-dividuals was extracted from the cross section sample of LINDA 2006. Tradi-tional OLS-models are employed and complemented with models based on pro-pensity score matching. The results show that men face a wage-premium of possessing a degree, i.e. the sheepskin effect, of roughly 5-8 percent for those who have obtained 120 credit
points or more (i.e. three years or more). For women, this is about 6-7 percent for those who have obtained 160 credit points (i.e. four years or more) or more. For students who attended a more prestigious old university in the metropolitan areas in Sweden, and majored in the natural sciences, a sheepskin effect of roughly 13 percent for men and 20 percent for women is traced. However, this result did not hold among students who attended a newer university outside the metropolitan areas. Controlling for specific occupational programs for economists, engineers and teachers did not, regardless of gender, give any significant estimates of sheepskin effects.
References af Trolle, Ulf (1990), Mot en internationellt konkurrenskraftig AKADEMISK
UTBILDNING, Lund: Studentlitteratur. Altonji, Joseph G. (1993), The Demand for and Return to Education When Edu-
cation Outcomes are Uncertain, Journal of Labor Economics, 11(1):48-83.
Altonji, Joseph G., Todd E. Elder and Christopher R. Taber (2005), Selection on Observed and unobserved Variables: Assessing the Effectiveness of Catholic Schools, Journal of Political Economy, 113(1):151-184.
Arcidiacono, Peter (2004), Ability sorting and the returns to university major, Journal of Econometrics, 121(1-2):343-375.
Becker, Gary S. (1964[1993]), Human Capital, 3 ed. Chicago: University of Chicago Press.
Becker, Gary S. and Nigel Tomes (1979), An Equilibrium Theory of the Distri-bution of Income and Intergenerational Mobility, Journal of Political Economy, 87:1153–89.
Becker, Gary S. and Nigel Tomes (1986), Human capital and Rise and Fall of the Families, Journal of Labor Economics, 4(3):1-39.
Björklund, Anders, Mårten Palme and Ingemar Svensson (1995), Tax Reforms and Income Distribution: An Assessment Using Different Income Con-cepts, Swedish Economic Policy Review, 2:229–266.
Black, D. and Smith, J. (2004), How Robust is the Evidence on the Effects of College Quality? Evidence from Matching, Journal of Econometrics 121 (1−2):99−124.
Black, D. and Smith, J. (2006), Estimating the Returns to College Quality with Multiple Proxies for Quality, Journal of Labor Economics, 24(3):701-728.
Brewer, D. and Ehrenberg, R. (1996), Does it Pay to Attend an Elite Private Col-lege? Evidence from the Senior High School Class of 1980, Research in Labor Economics 15:239−271.
Brewer, D., Eide, E. and Ehrenberg, R. (1999), Does it Pay to Attend an Elite Private College? Cross-Cohort Evidence on the Effects of College Type on Earnings, Journal of Human Resources, 34 (1):104−123.
Brint, Steven and Jerome Karabel (1989), The Diverted Dream: Community Col-leges and the Promise of Educational Opportunity in America 1900-1985, New York: Oxford University Press.
Card, David (1999), The casual effect of education on earnings, in Handbook in Labor Economics, Vol. 3A, (red) Orley C. Ashenfelter and David Card, Amsterdam: North-Holland: Elsevier Science Publishers.
Dale, Stacy Berg and Allan Krueger (2002), Estimating the Payoff to Attending a More Selective University: An Application of Selection on Observables and Unobservables, Quarterly Journal of Economics, 117(4):1491-1527.
Edin, Per-Anders and Peter Fredriksson (2000), LINDA - Longitudinal INdivid-ual DAta for Sweden, Working paper 2000:19, Uppsala, Sweden: Depart-ment of Economics, Uppsala University.
Eliasson, Kent (2006), University choice and earnings among university gradu-ates in Sweden, Umeå Economic Studies No. 693.
Erikson, Robert and Jan O. Jonsson (1993), Ursprung och utbildning, in SOU:1993:85, Stockholm: Fritzes.
Gartell, Maria and Håkan Regnér (2004), Inkomstpremier av lärosäten för män och kvinnor som tog en examen under 1990-talet, Institutet för framtids-studier: 2004:1
Government-bill. (2000), Vuxnas lärande och utvecklingen av vuxenutbildning-en, Stockholm: Proposition 2000/01:72.
Hammarström, Margareta (1996), Varför inte Högskola?, Göteborg: Universita-tis Gothoburgensis.
Harmon, C., Oosterbeek, H. and Walker, I. (2003), The Returns to Education: Microeconomics, Journal of Economic Surveys,17(2):115−155.
HSV (1998), The Current Swedish Model of University Governamnce - Back-ground and Description, Raport : 1998:10S, Stockholm: National Agency for Higher Education.
Hungerford, Thomas and Garry Solon (1987), Sheepskin effects in the returns to education, Review of Economics and Statistics, 69:175–177.
Flores-Lagunes Alfonso and Audrey Light (2007), Interpreting Sheepskin Effects in the Returns to Education, Econ Working paper 0707, Department of Economics, University of Arizona
Jaeger, David. A. and Marianne E. Page (1996), Degrees Matter: New Evidence on Sheepskin Effects in the Returns to Education, Review of Economics and Statistics 78:733-740.
Kane, Thomas J. and Cecilia E. Rouse (1995), Labor Market Return to Two- and Four Year College, American Economic Review,. 85(3):600–14
--- --- (1999), The community college: educating students at the margin between college and work. Journal of Economic Perspectives 13(1):63–84.
Light, A., Audrey and Wayne Strayer (2000), The determinants of university completion: school quality or student ability?, Journal of Human Re-sources 35(2):299–332.
Lindahl, Lena and Regnér, Håkan, (2005), College Choice and Subsequent Earn-ings: Results Using Swedish Sibling Data, Scandinavian Journal of Eco-nomics, 107 (3), 437−457.
Leigh, Duane E. and Andrew M. Gill (2003), Do community college really di-vert the students from earning a bachelor’s degree?, Economic of Educa-tion Review, 22(1):23-30.
Leigh, Duane E. and Andrew M. Gill (2004), The effect of community colleges on changing students’ educational aspirations, Economic of Education Re-view, 23(1):95-102.
Manski, Charles F.and D. A. Wise (1983), University Choice in America. Cam-bridge, Massachusetts: Harvard University Press.
Mincer, Jacob. (1974), Schooling, Experience, and Earnings, NY Columbia University Press.
MIS (2000), Utbildningsklassificering, vol 1, Statistic Sweden. OECD (1993), Education at a Glance 1993, Paris: OECD. OECD (2008), Education at a Glance 2005, Paris: OECD. Rouse, Cecilia E. (1995), Democratization or Diversion? The effect of commu-
nity colleges on educational attainment, Journal of Business and Eco-nomic Statistics, 3(2):217–224.
--- (1998), Do Two-Year Colleges Increase Overall Educational Attainment? Evidence from the States, Journal of Policy Analysis and Management, 17(4): 595-620.
Roemer, John E. (2006), Democracy, Education, and Equality – Graz Schum-peter Lecture, Economic Society Monographs, Cambridge University Press.
SCB (2007), Universitet och högskolor: Genomströmning och resultat i högsko-lans grundutbildning rom 2005/06; Statistiska Meddelanden UF 20 SM 0702.
Schultz, Theodore W. (1961), Investment in Human Capital, The American Eco-nomic Review, 51(1):1-17
Solon, Gary (1999), Intergenerational Mobility in the Labor Market, in Orley Ashenfelter and David Card (eds.), Handbook of Labor Economics, Vol. 3A, Elsevier, Amsterdam.
SOU (1972), U68 – Högskolan, SOU1972:3, Stockholm: Fritzes. Light, A., Audrey and Wayne Strayer (2000), The determinants of university
completion: school quality or student ability?, Journal of Human Re-sources, 35(2):299–332.
UHÄ (1989) Högskoleutbidlningens framtida dimensionering, Stockholm: UHÄ1989:17.
Wadensjö, Eskil (1991), Högre utbildning och inkomster, in E. Wadensjö (ed.), Arbetskraft, arbetsmarknad och produktivitet, SOU1991:82 Expertrapport 4, Fritzes, Stockholm.
I
The Expansion of Higher Education in Sweden
and the Issue of Equality of Opportunity
Susanna Holzer
Abstract
This paper analyzes to what extent the political means of democratization have decreased educational inequalities (i.e. the choice of attending higher education has become less dependent on family background in the 1990s than before). Especially the new universities were heavily exposed to the expansion. The results show that living and growing up in the same area as rapidly growing educational institutions seems to have a strong overall positive effect on university attendance of roughly six percentage points, which indicates that the expansion of the higher educational sector in Sweden has had an overall democratization effect. Having more local access to higher education also seems to have decreased the social distance to higher education, meaning that the option of attending higher education, as compared to entering the local labor market after upper secondary school, has become a more common and possible alternative for more socioeconomic groups in society. The fact that more individuals chose to attend higher education further away in the 1990s among those living in the area of new universities, could be interpreted as the choice of moving being considered less costly (a smaller risk) once the decision to attend higher education has been made in the first place and the new universities do not have a clear diversion effect on overall university attendance. The relatively largest growth in attendance occurred among students whose parents had upper secondary school as their highest education, which can bee seen as an indication of a political success in equalizing educational opportunities in Sweden.
JEL Classification: I22, I28, J24
Keywords: Higher Education, Intergenerational Educational Mobility, Regionalization
Correspondence address: Susanna Holzer, School of Management and Economics, Växjö University, SE-351 95 Växjö, Sweden. Phone: +46 470 70 85 79. E-mail:[email protected]. I am grateful to Mårten Palme, Håkan Locking, Anders Björklund, Chris Taber, Nicholas Barr, Jan Ekberg, and Ghazi Shukur for helpful comments and suggestions. I have benefited from many useful comments from seminar and conference participants at Växjö University, Aahrhus University, SSGPE in Linköping, ECEO in Antwerpen, EALE in Prague and the Arne Ryde Symposium in Lund. I acknowledge financial support from Växjö University and Jan Wallander and Tom Hedelius’ Research Foundation.
2
1 Introduction
One central institution for implementing political views of social and distributive justice is
education. Swedish policy makers are by far no exception and by decreasing educational
inequalities in society they have, for the last century, used the egalitarian motive to reform the
entire education system. An important means of educational equalization has been to improve
access, both in terms of increasing the amount of slots at the existing institutions of education,
but also to geographically increase and spread the amount of institutions in the country that
could offer educational training. The egalitarian idea of improving educational opportunities
in terms of improving access for the population is referred to as a democratization by Brint
and Karabel (1989) and Rouse (1995). That is, nearby easy-access would give all who want to
attend an educational program the opportunity to do so or encourage those who might
otherwise not have attended.
The supply-side oriented and centrally monitored higher education in Sweden was
concentrated to six universities in the early 1970 (mainly located in the metropolitan areas). In
the late 1970s, these universities were complemented with an additional 12 smaller and
geographically more dispersed regional universities.1 However, improved geographical access
did not initially have any major impact on the size of the student body. A change toward the
end of the 1980s, though, led to a rapid expansion in the institutions of higher education. In
less than ten years, there was an increase from approximately 150,000 students during the
entire 1980s to about 330,000 students in 1999. The new universities, which in some cases
grew by 400 percent in terms of the number of students, were a main contribution to this
expansion.2
The empirical focus in this paper is to investigate three questions: (a) to what extent did
individuals living in the municipalities of the new universities of the late 1970s and later
become more inclined to attend higher education in the 1990s; (b) to what extent did the
choice of attending a university in general become less dependent on individual and family
background characteristics in the 1990s (i.e. did intergenerational educational mobility
1 Through the reform of the higher educational system in Sweden in 1977, there was a change in the entire definition of higher education. This makes any before and after comparisons of the higher educational system almost pointless, since it would be comparing apples to pears. Universities established in 1977 or later are throughout the paper referred to as new universities, universities established before 1977 are referred to as old universities. Since both universities and university colleges in Sweden are allowed to provide educations on master level, both institutions are throughout the text referred to as universities. 2 See SOU (1972:3) for the political motives for the educational expansion.
3
increase?); and (c) did the new universities divert students from lower socioeconomic
backgrounds from attending older universities? With mostly shorter vocational-oriented
programs being offered at the new universities, did they divert potential students who might
otherwise have attended an old university, i.e. who would otherwise have invested in longer
educations? An effect referred to as a diversion effect by Rouse (1995), and Leigh and Gill
(2003, 2004).
Increased access and shorter traveling distances decrease the investment cost, which should
make the investment more appealing to more groups with limited economic resources. This
supports the democratization effect. Access to higher education nearby might decrease the
physical traveling distance and the social distance. If higher education becomes a natural
alternative for more groups in society, the decision to enroll in alternative higher education
programs than what is offered by the local university might become more interesting. The
decision to move in order to attend a higher education elsewhere could be interpreted as less
costly, since the actual decision to attend higher education in the first place has been less
dramatized by having local access to some higher education.
The empirical data used in this study is a sample of roughly 300,000 individuals extracted
from the Swedish Longitudinal INdividual DAta panel (LINDA). The sample is divided into
25 cohorts of 18-year-olds for the years 1977-2001, where the cohorts of 1977-1989 represent
the 1980s and the remaining cohorts the 1990s. All youths are conditioned as having at least
one parent in LINDA, which provides associations between parents and their youths. Two
models, one on the democratization and one on the diversion effect, are estimated.
The results here show that living and growing up in the area of a rapidly growing educational
institution seems to have a strong positive effect on overall university attendance. With an
impact of roughly six percentage points, this indicates that the expansion of the higher
educational sector in Sweden has had an overall democratization effect (the local effect at the
new universities is roughly 41 percentage points). Access to higher education more locally
seems to have decreased the social distance to higher education, meaning that the option of
attending higher education, as compared to entering the local labor market after upper
secondary school, has become a more common and possible alternative for more
socioeconomic groups in society. The fact that more individuals chose to attend higher
education further away in the 1990s among those living in the areas of new universities could
4
be interpreted as the choice of moving being considered less costly (a smaller risk) once the
decision of attending higher education has been made in the first place and that the new
universities do not have any diversion effect on overall university attendance. The relatively
largest growth occurred among students whose parents had upper secondary school as their
highest education, which can bee seen as some indications of political success in equalizing
educational opportunities.
The paper is organized as follows: The next section gives an overview of Sweden’s higher
educational policy, followed by a brief overview of the literature in Section 3. Section 4
presents the empirical specifications and Section 5 presents the data management. Section 6
reports the results and Section 7 concludes the paper.
2 Brief History of Sweden’s Higher Educational System
A fast growing industry and a rapidly increasing demand for a larger and more highly
educated labor force obliged Sweden, like most industrialized countries after World War II, to
put political and economical resources into improving the entire educational system. The
education offered was, and still is, almost entirely financed and monitored by the government
and offered free of charge (i.e. from compulsory education to the university level). This
increasing demand for education, together with an egalitarian vision of equality of opportunity
by the leading Social Democratic Party in Sweden, brought about a substantial reform of the
entire education system in the 1950s and 1960s. By both improving and extending the years
spent in compulsory school and by rapidly increasing the access to upper secondary school,
more people than ever before qualified and chose to attend more schooling beyond the
mandatory level.3 This, together with the introduction of a student aid program that provided
university students with grants and generous student loans (that were independent of the
financial status of the parents), triggered the development of the few Swedish universities
becoming literally overcrowded. The chaotic situation suffered by the institutions of higher
education in the late 1960s called for drastic measures for solving the future increasing
demand for higher education.4
3 See Meghir and Palme (2005) on effects of an earlier reform in the Swedish educational system on educational attendance, and Erikson and Jonsson (1993) for an overview of the Swedish higher educational system and issues concerning educational opportunities in Sweden. 4 During the 1960s alone, the amount of newly enrolled students rose from 7,800 during the academic year 1960/61, to 27,100 in 1969/70 – see SOU 1972:3, p 84.
5
Whereas the Swedish population was offered compulsory and upper secondary schools in
their local region, higher education in the early 1970s was mainly concentrated to six
university (mostly metropolitan) areas in Sweden.5 In terms of regional policy, local access
was sought to favor those who would like to stay, live and work locally. A successful regional
policy in offering higher education nearby would not only increase the overall educational
level, it would most likely also increase the number of job opportunities in the region and
promote local economical growth (see SOU (1972:3)). Hence, the future goal of Swedish
higher education could be interpreted both as egalitarian and as a form of regional
redistribution.
There was a dramatic change in the Swedish university structure as a result of the 1977 Act of
Higher Education. Besides the six old universities, 12 new regional universities were
established.6 The new institutions of higher education were in most cases former schools for
teacher training, military training, and nursing schools that were granted an upgraded status as
tertiary educations due to the reform. Therefore, the majority of the programs initially offered
at those institutions had a vocational character and were shorter (less than three years). Most
of the new institutions of higher education were only to conduct undergraduate education
without any research connections. The geographical locations of all institutions of higher
education from 1977 and onwards are presented in Figure 1.1.
Initially, there were no limitations in the admissions to higher education in the 1960s, which
was also a contributing factor for the institutions becoming overcrowded. As a result, overall
restrictions and limitations in admissions to higher education were implemented in 1977 to
1979. However, to encourage new student groups to attend higher education, the limited
admission was softened, with alternative ways of qualifying for higher education. Besides the
traditional and most common way of qualifying for higher education (i.e. by having a degree
from an upper secondary school with Grade Point Average scores (GPA)), degrees from adult
5 Göteborg, Linköping, Lund-Malmö, Stockholm, Umeå and Uppsala. In the same geographical areas as the universities were three large institutes in Stockholm: the Royal Institute of Technology, the Karolinska Institute of Medicine, and Stockholm School of Economics; and two others: the Chalmers Institute of Technology in Gothenburg, and the Institute of Agriculture in Uppsala. All institutions of higher education in these six university municipalities are included in the definition of old universities in this paper. 6 The new universities of 1977 were: Borås, Eskilstuna/Västerås, Falun/Borlänge, Gävle/Sandviken, Jönköping, Kalmar, Karlstad, Kristianstad, Sundsvall/Härnösand, Växjö, Örebro, Östersund, Luleå. Later university establishments were: 1983 – Halmstad and Skövde, 1988 – Ronneby/Karlskrona, 1990 - Uddevalla/Trollhättan, 1995 - Södertörn, and 1998 – Gotland and Malmö. In this paper, Malmö is never separated from Lund.
6
schooling (folkhögskola), four years of labor market experience, or good results from the
Swedish Scholastic Aptitude Test (högskoleprovet) became new means of qualifying for
higher education.7 Note that Swedish universities cannot choose freely among eligible
students. The qualifications of the presumptive student are the only means for the individual
of competing for a student slot in a certain education. The centrally monitored admission
system from 1977 remains more or less the same today.
Figure 1.1 The geographical location of the institutions of higher education in Sweden. Note: The old universities (established prior to 1977) are in capitals, the rest are new universities that received the status of independent institutions of higher education in 1977 or later. Source: Statistic Sweden.
Despite the increased amount of institutions of higher education and the overall increased
geographical access, there was a very modest development in the sector of higher education in
the first ten years after the reform. In fact, the number of students at the old universities and
the new universities was roughly the same in 1987 as it had been in 1977.
The reforms of the lower educational levels resulted in Sweden having one of the highest
population rates with upper secondary qualifications in an international comparison in the
7 See Kim (1998) and Öckert (2001) for a description and discussion about the admission rules of 1977.
7
1980s. Yet only a smaller proportion of the Swedish population made the transition to higher
education. In the late 1980s, several reports stated that Sweden had fallen behind in relevant
comparisons concerning national levels of higher education.8 Sweden received strong
criticism for its higher education, centrally monitored by the government, being under-
dimensioned with respect to the demand. Reports stated that Sweden risked losing
competence within several academic professions if the sector of higher education did not
expand its undergraduate education to compensate for large scale retirements in the 1990s.9
To meet the present and future demand of higher education, the sector of higher education
became the target of a very massive expansion in the 1990s. Especially the new universities
became heavily exposed, where in some cases the enrollment grew by 400 percent in less than
ten years. The growth in enrollment into the universities in Sweden is presented in Figure1.2.
Figure 1. Students enrolled in Swedish higher education 1977‐2001 Note: The figure illustrates the enrollments at old and new universities for the years 1977‐2001. Source: Statistic Sweden.
The strict diversion between establishments mainly conducting undergraduate education and
establishments conducting both undergraduate education and research was softened in the
1990s. This made it possible for the new universities to conduct research on a larger scale
than before. The increased research activities at the new universities, together with overall
structural changes of educational programs in the early 1990s, made it possible for more of
8 See af Trolle (1990), UHÄ (1989), OECD (1993) and Hammarström (1996). 9 See HSV (1998, p 15f) for a brief discussion and overview.
8
the new universities to offer longer educational programs than before. However, most of the
prestigious educational programs in law, medicine and art, for which the competition among
students is the highest, are until the present day restricted to the old universities.
3 Previous Literature
Equality of educational opportunity relates to the extensive economic literature on
intergenerational mobility, which looks at the association between parents’ economic
outcomes and the future economic outcome of their children. A high intergenerational
mobility indicates a low association between parents’ economic outcomes and the future
economic outcome of their children. A weak association between the child and parental
characteristics indicates a high equality of opportunity in society, saying that children from
different origins have similar chances in life (see e.g. Becker and Tomes (1979, 1986) and
Solon (1999) for a survey of the literature on intergenerational mobility).10
On the issue of expanding and increasing access to higher education in Sweden, only modest
economic research has been performed on how the expansion as a political means has affected
intergenerational mobility. One of the few examples is Holm and Häggström (1972) who
conducted cost-benefit analysis on early pilot projects to relocate higher education into new
areas in Sweden in the late 1960s. They argued that they could detect some positive effects on
recruiting youth from the new regions with less traditional family backgrounds. This was seen
as early indications of increased local access increasing educational mobility and promoting
social mobility in the region. However, Fasth (1980), who did research similar to that of Holm
and Häggström, found no support for the hypothesis that having more local access encouraged
more people from different socioeconomic groups in these new geographical areas to invest in
higher education than in other areas in Sweden.
The effect of traveling distance as part of a student’s investment cost was investigated by
Kjellström and Regnér (1999). They assumed that the distance to an institution of higher
education was to be positively correlated with the cost of attending higher education – the
10 See Björklund and Jäntti (1997) and Björklund, Lindahl and Sund (2003) for some Swedish examples. An alternative track in the economic literature of intergenerational mobility is also found in the economics of philosophy; see e.g. Roemer (1996, 2006) on issues of democracy, equality and distributive justice when discussing the impact of educational policies.
9
longer the distance, the higher the cost. When examining the probability of attending higher
education among 10,000 individuals in Sweden born in 1967, the authors found that traveling
distances had a significant negative effect on the probability of attending higher education.
However, the effect was so small that their conclusion was that this was most likely not the
strongest determinant for the individual in deciding whether to attend higher education.
More extensive research on the effects of the expansion of higher education on educational
inequalities in Sweden has been conducted by sociologists like Erikson and Jonsson (1993,
1994, 1996, 2006) and Dryler (1994, 1998).11 Dryler (1994, 1998) used aggregated data from
the population census and data from the Higher Education Register in order to follow 17-24
year olds from 1968-1990 in three geographical areas that received new establishments of
higher education in the 1970s. She found no support for the hypothesis that new
establishments had increased the probability of enrollments of individuals from lower
socioeconomic backgrounds.
Erikson and Jonsson (2006) present an individual-educational-choice model, which they use
to analyze to what extent expanding higher education has reduced inequalities in education in
Sweden. They address three angles on the issue of higher education: how GPA from upper
secondary school and social origin affect higher education attainment; how increased access
has affected the association between social origin and the educational outcome of an
individual over the last century; and how these intergenerational associations were affected by
the higher educational reform in the late 1970s. The last question was divided into two parts.
First, how did the increased access affect individuals with different social origins? Second,
did the increased access to shorter, tertiary programs divert individuals from lower
socioeconomic groups to attend the shorter educational programs, instead of attending the
traditional, longer university educations?
To answer their questions, they used data from several registers, the population census for
several years and interviews with several students from several years, covering cohorts from
the period 1892-1970. Their key finding is that the association between parents and their
children has weakened during the second half of the twentieth century. Nonetheless, they state
that there may have been an increase in social mobility, but the political means of expanding
11 See also Broady, Börjesson and Palme (2002).
10
and increasing access to higher education had a very modest equalizing affect on educational
opportunities in Sweden.
Comparisons between the political outcomes of how expansion has affected inequality in
Swedish educational opportunities and the outcome of similar expansion in other countries
should be made with caution, since educational systems differ among countries. Differences
in educational and financial aid systems give individuals different conditions in each country.
Therefore, the effect of the educational expansion and increased access may vary from
country to country. One of few examples is Rouse (1995) which could be taken as a good
comparison to the Swedish development. She analyzes the possible effect on higher
educational attainment of the implementation of several regional two-year colleges, as a
complement to four-year colleges, in the United States. Two-year colleges were established to
increase local access to higher education and were considered to influence more people from
lower social origins to consider investing in higher education, i.e. the democratization effect.
She also discusses how increased access to shorter tertiary educational tracks at the two-year
colleges nearby might divert able individuals from lower socioeconomic backgrounds to settle
for shorter educations instead of investing in a longer educational program at four-year
colleges. Even though she found some tendencies of a diversion effect of the two-year
colleges on higher educational investment, the overall democratization effect was so much
stronger, giving the overall expansion and increased educational access a positive effect on
increasing educational attainment in the United States.12
Although some examples from the United States indicate that an increased access raises the
democratization effect, most of the literature on the effect of expanding and increasing the
access to higher education in Sweden provides no or very modest support for its being an
effective means of educational equalization.
12 See Kane and Rouse (1999) for an overview of the higher educational system in the United States; see also Kane and Rouse (1995) and Leigh and Gill (2003, 2004) for more examples of research.
11
4 Empirical Specification
This study applies a differences-in-differences methodology to investigate how higher
educational attendance choice behavior and school choice behavior for individuals living in
the geographical areas of the new universities differ from the Swedish population in general.13
This kind of methodology allows us to estimate differences in choice behaviors before and
after the expansion of higher education in Sweden during the 1990s. It also allows us to
estimate how individuals living in the areas of the new universities may differ in their choice
behavior from the rest of the population. The areas of the new universities will henceforth be
referred to as the areas of NEW.
The following latent variable specification is used both for modeling the individual propensity
for attending higher education and the propensity for new university colleges of diverting
students from attending old traditional universities:
(1) NEWuZA jijij 3210*
90*90* 210 DuDZYear jij
,210 90**90**90* ijjij DNEWuDNEWZDNEW
where *ijA is a latent variable measuring attendance to higher education, defined as:
(2)
,0
1 *
otherwise
cAifA ij
ij
where ijA is the binary outcome variable for student i living in county j that reveals if the
student attends higher education or not. c is a threshold, Zij is a vector of personal and family
background characteristics, Year is a vector of year dummies, NEW is a dummy variable for
living in a municipality where a new university started in the late 1970s or later, uj is the
county youth unemployment rate in the county labor market in which the individual lived at
the time of entrance, D90 is a dummy variable taking the value of zero for the years 1977-
1989, and unity for the years 1990-2001. ij is a random error term representing all omitted
variables that might affect individual choice behavior and it is assumed to be approximated by
a normal distribution.
13 See Angrist and Krueger (1999) for an overview of differences-in-differences methodology.
12
The key policy parameters in Equation (1) are α3, γ0, γ1, and γ2. They all measure how
individual choice behavior changes in the 1990s and how this behavior differs for individuals
living in the areas of NEW compared to the top rest of the Swedish population.
4.1 The democratization effect of the expansion
The effects of increased access to higher education, the democratization effects, are modeled
in a latent variable specification described in Equation (2). Here, the binary response variable
takes the value of 1 if an individual attends higher education, and zero otherwise. The control
group here is individuals who have never attended higher education.
According to standard human capital theory on intergenerational mobility, both educational
level and income level are to some extent to be transmitted across generations within families,
i.e. children from homes of more highly educated parents are more likely to attend higher
education themselves.14 The impact of individual and family characteristics (such as gender,
level of parental education, and parental income) on higher educational attendance in the
1980s is measured by α1.
More local access to higher education should make it possible for more people from different
social origins to attend universities without moving or making use of long distance
communications. This is a motivation for regionalization and educational equalization that to
a high extent motivated the rapid expansion, especially at the new universities in the 1990s.
The overall effects of increased access to universities on attendance behavior in the 1990s,
based on the same set of individual and family characteristics as before, are measured by β1.
The differences-in-differences effect of the individual and family characteristics on
attendance in the 1990s for those individuals who lived and grew up in the area of NEW, as
compared to the population in the rest of Sweden, is measured by γ1.
The county youth unemployment rate, uj, is included as a factor of a macro economic
externality. Besides parental influence, local youth unemployment is assumed to have a
positive influence on the educational investment choice by the youth. Poor job opportunities
should decrease the alternative cost of educations; see e.g. Freeman (1980) and Rouse (1995).
Yet it is not easy to make any assumptions about the effect of local youth unemployment in
14 See Becker and Tomes (1986) and Becker (1993) for an example, or Solon (1999) for an overview of the literature.
13
the 1980s, measured in α2.15 Sweden enjoyed a huge economic boom in the 1980s, where
unemployment was to a very large extent due to individual characteristics, rather than a lack
of job opportunities. As for the 1990s, unemployment was very much caused by the sustained
loss of job opportunities due to the dramatic crisis in the early 1990s, an impact on higher
educational attendance that is represented by β2, and is assumed to be positive. A majority of
the areas of the new universities were the parts of Sweden most affected by the economic
crisis in the 1990s. Parameter γ2 measures how youths living and growing up in the area of
NEW were differently affected by unemployment in their region, as compared to Sweden as a
whole in the 1990s.
The difference in the propensity to attend higher education for individuals living in the areas
of NEW during the 1980s, as compared to the Swedish population in general, is measured by
α3. How this propensity changes due to the overall [är det vad du menar?] population in the
1990s is measured by γ0. Yearly changes in attendance from the years 1977-2001 are
measured by β0, using 1989 as a base year. Since the supply of student slots increases over
time and the impact of family background probably weakens, we should expect that trend of
attendance to stand out as positive. Since the dependent variable is binary, a probit model is
used to estimate the model of higher educational attendance.
4.2 The diversion effect of new university colleges
To study how the new universities divert students from attending old universities, Equation
(1) is once more used. However, the binary response variable A now takes the value of 1 if an
individual attends an old university and zero otherwise. The control group here contains both
individuals who have never attended higher education and those attending a new university.
The possible diversion effect of the new universities in Sweden on higher educational
attendance from old universities in the 1980s, based on individual and family characteristics,
is measured by α1. How the increase of student slots at the new university colleges may affect
a diversion in the 1990s is measured by β1. However, the increased and improved educations
offered at the new universities in the 1990s should have attracted more youths from higher 15 Sweden was badly hurt by an economic crisis at the beginning of the 1980s. However, after the devaluation of the Swedish krona in 1982, the economy experienced a rapid recovery and the unemployment rate became historically small. If a person was unemployed in the late 1980s, the reason for this was most likely not a lack of job opportunities. Rather, to a considerable extent, it had to be looked for in individual characteristics. The labor market of the 1990s did not display such a corresponding characteristic. Rather, the 1990s became marked by a dramatic loss of job opportunities.
14
socioeconomic groups to the new universities than before. This, in turn, would give a
contractive effect on the diversion in attendance between some socioeconomic groups.
Living in an expansive NEW area in the 1990s should have encouraged more individuals from
these areas to attend their local institutions of higher education (i.e. this should have an
expected negative effect on attendance at old universities). How the propensity to attend an
old university differs for youths living in the area of NEW from youths overall in Sweden,
based on individual and family characteristics, is measured in the differences-in-differences
coefficient γ1. The impact of just living in NEW in the 1980s, measured by α3, is expected to
have a negative effect on attendance at an old university. The expansion of higher education
in the 1990s, in which were the most exposed in the areas of NEW, is expected to manifest
itself in an even higher negative effect on attendance at old universities in the 1990s.
Finally, β0 measures if and how the trend in attendance changes over the years and whether
increased access to the new university colleges in the 1990s had a negative effect on
attendance at the old universities.
5 Data and Measurements
The empirical analysis is based on data from Longitudinal INdividual DAta for Sweden
(LINDA). LINDA is a random sample of approximately three percent of the population of
Sweden, where the information is based on income-tax registers, population censuses and
other register based data (see Edin and Fredriksson (2000) for a description). In addition to
the main subjects in LINDA, family members and cohabits that belong to the same household
as the main subjects are also included in the data. In total, the dataset contains information
regarding nearly one million individuals.
Information from the Swedish Higher Education Register has been added, which reveals if
and where any of the individuals in LINDA have attended an institution of higher education in
Sweden. This information is only available from 1977 and onwards, which gives this analysis
the natural starting point of 1977. The last year considered is 2001.
15
5.1 The Sample
At the age of 18, most youths in Sweden still live at home and attend the last year of upper
secondary school. Conditioning the 18-year-old to have at least one parent in LINDA allows
for intergenerational connections in the dataset.16 All in all, 299,944 18-year-olds divided into
25 cohorts were extracted from LINDA between the years 1977 and 2001, i.e. roughly 10
percent of the entire cohorts of 18-year-olds in Sweden for the same time period. Table 5.5
shows descriptive statistics of all cohorts. The sample of 18-year-olds extracted from LINDA
is either main subjects in LINDA (i.e. one of the randomly sampled three percent of the
Swedish population) or in the LINDA-data as a family member of a main subject. Due to this,
the amount of 18year-olds can exceed three percent of the year cohort.
Extracted information about the 18-year-olds includes gender and the geographical location of
the household in which the youths were registered at the age of 18, both at the county and the
municipality level. The only information available about the youths after the age of 18 is if
and where they attend higher education (at which university) before the age of 26.17
Family background information is based on information about the parents registered in
LINDA. The financial status of the household is represented by parental income, measured by
disposable income after taxes and received benefits. The nominal income of the household
has been transformed into a relative income of the household as compared to all households in
the year their youth turns 18.18 To have some indicator of whether the parents have economic
problems, two financial aid forms, social welfare and unemployment benefits, are included.
Both social welfare and unemployment benefits are transformed into dummy variables,
indicating if any of the parents received one or both financial aid forms during the year their
youth turned 18.
16 An intergenerational connection based on the multi-generation-register of Sweden Statistic over the population. 17 There has been an outspoken wish by policymakers to encourage young adults to attend higher education at an early age. In fact, the political goal is to encourage 50 percent of an age-cohort to enter higher education before the age of 26; see Government Bill (2000/01:72). In line with this goal, this study will be restricted to studying youth up to the age of 26. 18 In the two-parent household case, household income has been divided by 1.7 in order to compare the economical support capacity of a two-parent household to that of a one-parent household (see Björklund, Palme and Svensson (1995)).
16
Table 5.1. The sample of 18‐year‐olds in LINDA 1977‐2001. (The proportion of women in the sample is presented in parentheses.)
Cohort
�1977 11,594 (0.47) 1,996 (0.52) 696 (0.69) 8,902 (0.44)
1978 11,118 (0.47) 1,929 (0.52) 686 (0.67) 8,503 (0.45)
1979 11,541 (0.47) 2,001 (0.53) 718 (0.65) 8,822 (0.45)
1980 11,554 (0.47) 1,952 (0.53) 741 (0.61) 8,861 (0.44)
1981 12,157 (0.48) 2,204 (0.54) 769 (0.64) 9,184 (0.45)
1982 12,804 (0.48) 2,281 (0.54) 883 (0.63) 9,640 (0.46)
1983 13,217 (0.48) 2,262 (0.54) 935 (0.62) 10,020 (0.45)
1984 13,254 (0.48) 2,284 (0.54) 1,006 (0.62) 9,964 (0.45)
1985 13,110 (0.48) 2,299 (0.55) 996 (0.60) 9,815 (0.46)
1986 12,285 (0.47) 2,211 (0.53) 1,027 (0.58) 9,047 (0.45)
1987 11,601 (0.48) 2,201 (0.55) 1,164 (0.55) 8,236 (0.45)
1988 11,978 (0.47) 2,388 (0.54) 1,283 (0.51) 8,307 (0.45)
1989 12,253 (0.48) 2,650 (0.54) 1,481 (0.53) 8,122 (0.45)
1990 11,906 (0.48) 2,754 (0.54) 1,683 (0.55) 7,469 (0.45)
1991 12,240 (0.47) 2,982 (0.54) 1,922 (0.54) 7,336 (0.42)
1992 12,550 (0.48) 3,138 (0.55) 2,064 (0.54) 7,348 (0.43)
1993 11,943 (0.48) 3,021 (0.55) 2,067 (0.54) 6,855 (0.43)
1994 11,691 (0.47) 3,057 (0.54) 2,113 (0.55) 6,521 (0.41)
1995 11,508 (0.48) 2,740 (0.55) 2,012 (0.55) 6,756 (0.43)
1996 11,167 (0.48) 2,479 (0.55) 1,899 (0.55) 6,789 (0.43)
1997 11,505 (0.48) 2,393 (0.56) 1,745 (0.56) 7,367 (0.43)
1998 11,647 (0.48) 2,160 (0.57) 1,573 (0.58) 7,914 (0.44)
1999 11,493 (0.48) 1,641 (0.55) 1,015 (0.61) 8,837 (0.45)
2000 11,567 (0.48) 707 (0.54) 477 (0.65) 10,383 (0.479
2001 12,261 (0.48) 22 (0.59) 4 (0.75) 12,232 (0.48)
N 299,944 (0.47) 55,752 (0.54) 30,959 (0.57) 213,230 (0.45)
Number of
observations
Attends
OLD*
Attends
NEW*
Control
group
Note: *Attendance is divided into attendance at an old university and attendance at a new university. The control group is the 18‐year‐olds who never attended a Swedish university before the age of 26. The majority of the attending youths in the sample attend higher education between ages 20‐22. It may appear that attendance decreases for the cohorts after 1994. However, this is more due to the fact that the study period of this paper ends in 2001, so the cohorts after 1994 are not followed up to the age of 26.
Moreover, the parents’ highest educational level is also included, divided into four
educational levels:19 compulsory school of a maximum of nine years, upper secondary school,
19 The educational history is based on the SUN-code and it is a standard used in classifying individual educational programs (see MIS (2000)). In this study, all forms of compulsory schooling have been merged into one level of elementary education, the same goes for upper secondary school. In case the parents have a university degree as their highest education, they are separated into two categories; one in which the degree is taken after passing a shorter higher educational program (shorter than three years), and one in which the degree is worth three years or more of higher education. I have kept the two university levels separated due to the fact
17
a shorter university education of less than three years and a longer university education of
three years or more.
County youth unemployment rates have been attached to this data in order to incorporate the
possibility of external influences in this analysis. The youth unemployment rate corresponds
to the year and county in which the individual lived at the age of 18. All explanatory variables
are presented in the Appendix.
5.2 Descriptive Statistics
In the descriptive statistics regarding the explanatory variables presented in Table 5.2, we can
see that the sample of 18-year-olds is presented in three panels, showing the mean value of all
explanatory variables of the cohorts summarized into 1980s and 1990s, respectively.
The first panel in Table 5.2 shows the entire sample of 18-year-olds used in this analysis. We
can see that the data contains slightly more men than women. Comparing the mean values of
educational levels of the parents in the 1980s with the values of the 1990s, the share of
parents only having compulsory schooling declined by roughly 12-16 percentage points. The
share with upper secondary schooling increases, as does the share of parents with one of the
two university levels.
In terms of family finances, relative income has increased over time. Sweden went through a
turbulent macroeconomic period in the early 1990s with high unemployment and low
economic growth. That households were affected by harsh financial times in the 1990s can be
seen by there being an increase in the amount of households receiving social welfare or
unemployment benefits or both.
In the second panel of Table 5.2, the youths attending higher education before the age of 26
are compared to youths never attending higher education, for each decade respectively. In
section three, we can compare youths who attend an old university with youths who attend a
new university, for each decade respectively.
that they give a relatively good signal of what sort of university education the parents have. As briefly mentioned in the policy section, numerous vocational educations (with less than three years of duration) received university status in the late 1970s. Longer university educations (i.e. three years or more) are dominated by traditional university degrees in law, medicine and art.
18
As can be seen in both Tables 5.1 and 5.2, women are in the majority among the youths who
attended higher education. This result is totally in line with the overall trend in Sweden for the
same time period.
Furthermore, we can see that the parents of attending youths are, on average, more highly
educated than the parents of non-attending youths. Those who attend an old university on
average have more highly educated parents than those who attend a new university. The
overall educational level of parents rose in the 1990s. An intergenerational pattern can be
traced from the 1980s into the 1990s: youths who never attend higher education are those
who, on average, have the lowest educated parents, followed by the parents of youths that
attend a new university.
Patterns can also be traced in the variables of the household economy. The households of
youths who never attend higher education on average have the lowest incomes. Family
finances are, on average, lower in households whose 18-year-olds attend a new university, as
compared to households whose youths attend an old university.
Table 5.2
Descriptive statistics: the sam
ple of 18‐year‐olds
The table is divided
into three sections: the first describes the m
ean values of all characteristics of the entire sam
ple, also divided
into 1980s and 1990s; the second section describes the
characteristics of the youths who Attend higher education in
the 1980s or 1990s, compared to youths who never atten
d higher education (the Control group); and the third section describes
the characteristics of the youth who atten
d an Old university and those atten
ding a New
university, also divided
into the 1980s and 1990s.
All
1980s
1990s
1980s
1990s
1980s
1990s
Attent
Control
Attent
Control
Old
New
Old
New
Mean
Mean
Mean
Mean
Mean
Mean
Mean
Mean
Mean
Mean
Mean
(Std.dev)
(Std.dev)
(Std.dev)
(Std.dev)
(Std.dev)
(Std.dev)
(Std.dev)
(Std.dev)
(Std.dev)
(Std.dev)
(Std.dev)
The 18‐year‐old:
Female
0.48
0.47
0.48
0.57
0.45
0.54
0.44
0.54
0.63
0.54
0.55
(0.50)
(0.50)
(0.50)
(0.50)
(0.50)
(0.50)
(0.50)
(0.50)
(0.48)
(0.50)
(0.50)
Living in NEW
0.22
0.21
0.22
0.23
0.21
0.24
0.20
0.18
0.38
0.15
0.37
(0.41)
(0.41)
(0.41)
(0.42)
(0.41)
(0.43)
(0.40)
(0.38)
(0.49)
(0.36)
(0.48)
Father’s education:
Compulsory
0.41
0.47
0.35
0.28
0.52
0.21
0.45
0.24
0.37
0.17
0.26
(0.49)
(0.50)
(0.48)
(0.45)
(0.50)
(0.41)
(0.50)
(0.43)
(0.48)
(0.38)
(0.44)
Upper secondary
0.39
0.38
0.40
0.40
0.38
0.39
0.41
0.40
0.40
0.36
0.43
(0.49)
(0.49)
(0.49)
(0.49)
(0.48)
(0.49)
(0.49)
(0.49)
(0.49)
(0.48)
(0.50)
University < 3 years
0.08
0.06
0.09
0.10
0.05
0.13
0.07
0.10
0.09
0.13
0.13
(0.27)
(0.24)
(0.29)
(0.30)
(0.22)
(0.33)
(0.25)
(0.30)
(0.29)
(0.33)
(0.33)
University ≥ 3 years
0.12
0.09
0.15
0.22
0.05
0.27
0.07
0.26
0.14
0.33
0.18
(0.32)
(0.28)
(0.36)
(0.42)
(0.21)
(0.44)
(0.25)
(0.44)
(0.34)
(0.47)
(0.38)
Mother’s education:
Compulsory
0.40
0.48
0.32
0.31
0.52
0.17
0.41
0.29
0.38
0.15
0.20
(0.49)
(0.50)
(0.47)
(0.46)
(0.50)
(0.38)
(0.49)
(0.45)
(0.48)
(0.36)
(0.40)
Upper secondary
0.40
0.38
0.42
0.36
0.38
0.40
0.44
0.35
0.39
0.36
0.45
(0.49)
(0.48)
(0.49)
(0.48)
(0.49)
(0.49)
(0.50)
(0.48)
(0.49)
(0.48)
(0.50)
University < 3 years
0.10
0.08
0.13
0.14
0.06
0.19
0.09
0.15
0.12
0.20
0.18
(0.30)
(0.27)
(0.34)
(0.35)
(0.24)
(0.39)
(0.29)
(0.36)
(0.32)
(0.40)
(0.39)
University ≥ 3 years
0.09
0.07
0.12
0.18
0.04
0.23
0.06
0.20
0.11
0.28
0.16
(0.29)
(0.25)
(0.33)
(0.38)
(0.19)
(0.42)
(0.23)
(0.40)
(0.32)
(0.45)
(0.37)
Economy of
the household:
Income A
1.04
1.03
1.05
1.20
0.99
1.15
0.99
1.24
1.10
1.20
1.06
(0.99)
(0.49)
(1.31)
(0.54)
(0.46)
(0.86)
(1.51)
(0.60)
(0.35)
(0.99)
(0.62)
Social welfare
0.05
0.03
0.06
0.01
0.04
0.03
0.09
0.01
0.01
0.03
0.03
(0.21)
(0.17)
(0.24)
(0.08)
(0.19)
(0.16)
(0.28)
(0.07)
(0.09)
(0.16)
(0.16)
Unemployment benefit
0.10
0.06
0.14
0.04
0.07
0.11
0.17
0.03
0.06
0.09
0.13
(0.30)
(0.24)
(0.35)
(0.19)
(0.25)
(0.31)
(0.37)
(0.17)
(0.23)
(0.29)
(0.33)
Single parent
0.24
0.22
0.26
0.14
0.24
0.19
0.30
0.15
0.13
0.20
0.18
(0.42)
(0.42)
(0.44)
(0.35)
(0.42)
(0.39)
(0.46)
(0.35)
(0.33)
(0.39)
(0.38)
Macro variable:
Youth unemployment
4.10
2.54
5.60
2.60
2.52
5.39
5.72
2.48
2.92
5.11
5.77
(2.55)
(1.10)
(2.63)
(1.09)
(1.11)
(2.88)
(2.47)
(1.05)
(1.13)
(2.85)
(2.88)
Number of
Observations:
299,944
147,001
153,943
29,578
117,423
57,133
95,810
21,142
8,436
34,634
22,549
Note: A) Household income is defined
as relative net‐income the year the youth turns 18. See the Appendix for a description. The case of a one‐paren
t‐household is considered
in the
estimation. This is presented in
the Appendix.
6 Results
The main results are presented in Tables 6.1 and 6.2.20 The first panel of each table tells us
how the covariates affect the probabilities of attending higher education in the 1980s; the
second panel tells us how the marginal effect of the explanatory variables changed in its
influence on this probability in the 1990s. To see how the impact of the covariates in the
model may differ depending on where in Sweden the youths grew up, the model is estimated
with and without interaction with a dummy variable NEW, in the tables referred to as the Base
model and the Model with interactions. The results in the third panel in both tables tell us how
the differences-in-differences in the marginal effect of the explanatory variables of individuals
who grew up in the area of NEW differ from those of youths who grew up elsewhere in
Sweden in the 1990s. All results are transformed into marginal effects, and the marginal
effects are to be read additively from panel to panel.21
6.1 The democratization effect of the expansion
The democratization effect is here measured by estimating how the probability of attending
higher education in general has changed due to the large expansion in the 1990s. As described
in the brief history of the higher educational sector in Sweden, no major changes occurred in
the 1980s – which could explain why the effect of growing up in the area of NEW in the
1980s on the probability of attending higher education, as shown in Table 6.1, had no
significant impact. In the 1990s, however, the effect of NEW on the probability of university
attendance tells us a somewhat different story. In the Base model, where we do not consider
any interaction between individual and family covariates with the dummy variable NEW, the
impact of living in the same municipally as a rapidly expanding university shows an average
positive effect of roughly six percentage points in Sweden. The difference-in difference 20 The outcome of the estimated models rests upon the assumption that all 18-year-olds who desire to attend higher education before the age of 26 are qualified to do so. Through the Act of Higher Education of 1977, alternative ways of qualifying for higher education (besides having a degree from upper secondary school – which is the most frequent way of qualifying for higher education) were introduced; a scholastic aptitude test was introduced, offering a chance of qualifying for higher education for applicants with poor degrees, or for applicants who were 25 years or older, with more than four years of work experience, work experience could be accounted for as a qualifications (see Kim (1998)). Moreover, according to Statistic Sweden, roughly 80-85% of the cohorts before 1990 had upper secondary school qualifications at the age of 26 and, correspondingly, 90% of the cohorts from the 1990s had upper secondary school qualifications. For adults who did not attend or perhaps did not finish upper secondary school in their early youth, Sweden had and still has very generous possibilities of attending local adult education (”komvux”) to help them receive the qualifications of upper secondary school. Those attending local adult education were also given financial support by the government. All in all, this strengthens the assumption about the university qualifications of 18-year-olds made above. 21 Complete regressions from both models are presented in the Appendix.
21
results of living in the area of a new university, showed in the Model with interaction, show
that living in the area of NEW has a local average effect of about 41 percentage points in the
1990s. Compared to the impacts of any other observables accounted for in the model, this is
the largest by far. Thus, increased local access has a greater impact on encouraging youths
(especially men from low educated parents) to enter higher education overall than what is the
impact of their socioeconomic background or local labor market conditions.
In the 1980s, it only made a small difference whether the parents had elementary or upper
secondary schooling as their highest education for their youth’s probability of enrolling in
higher education. However, if the parents had lower university degrees, it had a considerable
larger positive impact on the probability of attending higher education. The impact rose even
more if the parents had a higher university degree. This indicates that intergenerational
mobility (i.e. that students attend more and higher education as compared to their parents) in
Sweden seems to have been very low in the 1980s. This is in line with previous findings by
Erikson and Jonsson (1993, 1996) and Dryler (1998). Notably, the impact of the mother’s
education on the probability of her child attending higher education is significantly different
and larger as compared to the impact of the father’s education. These findings are comparable
to those of Currie and Moretti (2003) when analyzing the intergenerational transmission
power of education and health between a mother and her child.
The impact of parental education changed somewhat in the 1990s. All levels above
compulsory schooling as the highest level of schooling for the parents became more
important. The probability of attending higher education tripled for youths whose parents had
upper secondary schooling as their highest education, as compared to those whose parents
only had compulsory schooling as their highest education. This might indicate that the
increased number of student slots in the 1990s has improved the intergenerational educational
mobility for at least one of two groups of youths from less educated parents. The impact of
high-educated parents on their youths’ educational choices grew even stronger in the 1990s,
thus to a smaller extent than for parents with a somewhat lower education.
The differences-in-differences effect of living in the area of NEW, presented in the third panel
in Table 6.1 under the Model with interaction, all interacted covariates show a negative effect.
But remember that the marginal effects are to be read additively through panels, meaning that
the negative values tell us that the impact of individual and parental covariates and local youth
22
unemployment has a smaller, yet overall positive, impact on the probability of university
attendance among youths who grew up in the area of NEW, as compared to youths who grew
up elsewhere in Sweden in the 1990s.22
In the 1980s, most income factors seem to affect higher educational attendance, as could be
expected. If any of the parents in the household received social welfare, the impact on the
attendance probabilities were strongly negative. If any or both of the parents received
unemployment benefits, this had a slightly negative effect on attendance. The 1990s was a
turbulent macro economic decade, however. There was a dramatic increase in the rate of
unemployment and more people were declared as poor than ever before in modern time. This
may account for why social welfare and unemployment benefits became more or less
naturalized in the 1990s (i.e. circled around a zero effect). Parental income, on the other hand,
rose from a small negative income effect on the probability of attending higher education in
the 1980s to a small positive effect in the 1990s. The differences-in-differences effect of
household income for youth living in the areas of NEW is almost zero.
According to the descriptive data in Table 5.2, more women than men attended higher
education, both in the 1980s and the 1990s. The marginal effects on higher educational
attendance indicate that the probability of attending higher education is larger for women than
men. In fact, the probability of attending higher education for women tripled in the 1990s as
compared to the 1980s. For individuals in the areas of NEW, the increase . doubled from the
1980s. This means that the expansion of the new universities affected males who grew up in
the area of NEW to a larger extent than the rest of the country.
The marginal effect of a macro externality, here represented by county youth unemployment
rates, on the probability of attending higher education is negative in the 1980s. A possible
explanation for this somewhat odd result could be that the existing unemployment during the
booming economic cycle Sweden in the 1980s was, to a relatively large extent, probably
caused by individual characteristics rather than a lack of job opportunities. In the 1990s, on
the contrary, much of the observed high unemployment was caused by the sustained loss of
job opportunities due to the dramatic crisis in the early 1990s. Now, the local youth 22 For instance, the probability of a youth whose mother only had upper secondary school attending higher education in the 1980s was 2.1 percentage points. In the 1990s, the corresponding probability was (2.1) + (14.9) = 17 percentage points and (2.1) + (14.9) + (– 7.6) = 9.4 percentage points if the youth lived in NEW in the 1990s.
23
unemployment rate has a positive marginal effect of roughly three percentage points on the
probability of attending higher education. In the area of NEW, this local effect is roughly two
percentage points lower, i.e. about one percentage point.
To illustrate how the chances of attending higher education changed in the 1980s to the 1990s
for various socioeconomic groups in Sweden, four groups of youths are illustrated in Figure
6.1a for Sweden and Figure 6.1b for youth living in the municipalities of NEW. The figures
illustrate the changes in the probability of youths attending higher education based on the
mean characteristics of the parents, given that they have the same highest educational level.23
It is clearly shown in both figures that all groups, except one, have considerably increased
their probabilities of attending higher education in the 1990s. The exception is youths whose
parents have elementary education as their highest education. According to the presented
results, their situation has become worse. Yet is notable that this lower socioeconomic group
has become a more selective group during the 1990s than what was previously the case. This
can be seen in the descriptive statistics presented in earlier sections of this paper.
23 Alternative combinations of household incomes and educational levels by the parents show similar developments as what is shown in Figures 6.1a and 6.1b.
24
Table 6.1 Democratization Effects The table shows the probability of attending higher education in Sweden in the 1980s, changes in the 1990s, and the differences‐in‐differences in change for those living in NEW in the 1990s, as compared to Sweden overall. The effects are presented as marginal effects.
Variables Model with interaction
Std. dev. Std. dev.
Effects on democratization: From living in NEW 0.008 0.014 0.014 0.013
Female 0.050 *** 0.004 0.053 *** 0.005
Upper secondary~father 0.041 *** 0.004 0.044 *** 0.004
University < 3 years~father 0.145 *** 0.008 0.147 *** 0.008
University ≥ 3 years~father 0.295 *** 0.009 0.295 *** 0.008
Upper secondary~mother 0.018 *** 0.004 0.021 *** 0.004
University < 3 years~mother 0.167 *** 0.007 0.169 *** 0.007
University ≥ 3 years~mother 0.301 *** 0.010 0.301 *** 0.009
Household income ‐0.044 *** 0.009 ‐0.033 *** 0.008
Social welfare ‐0.163 *** 0.007 ‐0.160 *** 0.007
Unemployment benefit ‐0.059 *** 0.006 ‐0.058 *** 0.006
Single parent ‐0.090 *** 0.005 ‐0.086 *** 0.005
Youth unemployment ‐0.055 *** 0.007 ‐0.050 *** 0.006
Changes in the 1990s:From living in NEW 0.056 *** 0.020 0.412 *** 0.026
Female 0.094 *** 0.005 0.103 *** 0.005
Upper secondary~father 0.107 *** 0.007 0.118 *** 0.007
University < 3 years~father 0.101 *** 0.010 0.118 *** 0.009
University ≥ 3 years~father 0.047 *** 0.009 0.066 *** 0.010
Upper secondary~mother 0.133 *** 0.007 0.149 *** 0.007
University < 3 years~mother 0.111 *** 0.008 0.128 *** 0.008
University ≥ 3 years~mother 0.079 *** 0.009 0.096 *** 0.010
Household income 0.070 *** 0.009 0.064 *** 0.008
Social welfare 0.132 *** 0.016 0.129 *** 0.017
Unemployment benefit 0.052 *** 0.008 0.051 *** 0.008
Single parent 0.154 *** 0.010 0.165 *** 0.011
Youth unemployment 0.088 *** 0.009 0.088 *** 0.008
Interacted effects; Changes in the 1990s and of living in NEW: Female ‐0.050 *** 0.007
Upper secondary~father ‐0.064 *** 0.008
University < 3 years~father ‐0.068 *** 0.010
University ≥ 3 years~father ‐0.070 *** 0.010
Upper secondary~mother ‐0.076 *** 0.008
University < 3 years~mother ‐0.071 *** 0.009
University ≥ 3 years~mother ‐0.069 *** 0.009
Household income ‐0.025 *** 0.005
Social welfare ‐0.021 0.015
Unemployment benefit ‐0.004 0.009
Single parent ‐0.076 *** 0.010
Youth unemployment ‐0.022 *** 0.002
Base model
Marg. eff. Marg. eff.
Note: The standard errors are robust and are adjusted by 283 clusters – equivalent to the municipalities accounted for in the paper. ***, **, * denote significance at 1, 5, and 10 percent, respectively. The cases of one‐parent and two‐parent households are considered in the estimations. Not presented in this paper, some other variables have been considered and analyzed as well; e.g. the municipalities adjoining the municipalities of the new universities have been included in the dummy NEW, resulting in somewhat weaker results than those presented here; variables of family and individual wealth have also been estimated – showing non‐significant results; kilometer distances to the nearest old university and the nearest new university gave significant estimations – but the marginal effect was extremely small.
25
Figure 6.1a The probability of attending higher education in the 1980s and 1990s.
0
10
20
30
40
50
60
70
80
90
100
1980s 1990s
Perc
ent
ED1 ED2 ED3 ED4 Figure 6.1b The probability of attending higher education in municipalities of NEW
0
10
20
30
40
50
60
70
80
90
100
1980s 1990s
Perc
ent
ED1 ED2 ED3 ED4
Note: The probabilities shown in Figure 6.1a and Figure 6.1b. are based on the average value of the covariates given the highest educational level of the parents. ED1 = elementary, ED3= upper secondary, ED3= shorter tertiary education, ED4 = longer tertiary education.
26
6.2 The diversion effect of new universities
The diversion effect is here measured by estimating how the probability of attending an old
university changed due to the huge expansion of the higher educational sector in Sweden in
the 1990s.
Let us initially consider the Base model, where the covariates are not interacted with the
variable NEW. In the 1980s, the chances of attending an old university for those youths that
grew up in the area of the new universities were about 4 percentage points lower as compared
to youths who lived elsewhere in Sweden. A possible explanation for this is that the
regionalization effect of higher education in the late 1970s did succeed in encouraging youths
in the areas of NEW to consider attending higher education in their region instead of moving.
This could, of course, be interpreted as the new universities diverting the youths in their areas
from attending an old university, i.e. a longer university education. Considering the massive
expansion of the universities in the area of NEW in the 1990s (of which some grew by 400
percent), the probability of attending an old university in the 1990s only fell by 0.4 percentage
points. Yet, living in the area of NEW has an average negative effect for attending an old
university, indicating that we can see a small diversion effect among individuals living in the
area of NEW.
Looking at the impact of parental education on the probability of attending an old university,
the effects are pretty much what could be expected. The higher is parental education in the
1980s, the higher is the probability of their children attending an old university – all compared
to youths whose parents only have elementary education as their highest education. Even
though the parental impact on the probability of attending an old university grew stronger in
the 1990s, the relatively largest increase is among students whose parents have upper
secondary education as their highest education. In some sense, this result contradicts the
diversion effect, since at least one of two low socioeconomic groups did respond positively to
the overall expansion of higher education in Sweden.
Turning to the Model with interaction, the story becomes slightly different, but only for the
effects in the 1990s. The marginal effects of the covariates in the 1980s are similar to those in
the Base model. Allowing the covariates to interact with the dummy variable NEW, however,
shows that the impact of living in the area of NEW goes from a small negative effect to a
small positive effect in the 1990s. This does not necessarily mean that the local institutions
27
failed to absorb the youths in their region. In fact, it could just be the other way around. Local
access to higher education has most likely decreased the social distance to higher education
overall, meaning that the option of attending higher education has become a more common
and possible alternative for more socioeconomic groups in society. The fact that more
individuals chose to attend higher education further away in the 1990s among those living in
the area of NEW, could be interpreted as the choice to move being considered less costly (a
smaller risk) once the decision to attend higher education at all has been made.
Compared to the youths whose parents only have compulsory education as their highest
education, the trend in the 1990s seems to be that there is an overall small positive significant
increase in the probability of going to an old university among all other groups. Once more,
the overall largest relative increase is among students whose parents only have upper
secondary education as their highest education.
For youths growing up in the area of NEW, however, the differences-in-differences effect
presented in the third panel does not indicate that the expansion of the new universities has
caused (or increased) a diversion among socioeconomic groups in Sweden. In fact, the
increased local access in the area of NEW seems to have weakened the importance of family
background across all socioeconomic groups in these regions for attending an old university.
28
Table 6.2 The Diversion Effects
The table shows the probability of attending an old university in Sweden in the 1980s, changes in the 1990s, and the differences‐in‐differences in the change for those living in NEW in the 1990s, as compared to Sweden overall. The effects are presented as marginal effects.
Variables Model with interaction
Std. dev. Std. dev.
Effects on diversion: From living in NEW ‐0.043 ** 0.018 ‐0.039 ** 0.017
Female 0.020 *** 0.003 0.023 *** 0.003
Upper secondary~father 0.041 *** 0.003 0.044 *** 0.003
University < 3 years~father 0.116 *** 0.008 0.119 *** 0.008
University ≥ 3 years~father 0.247 *** 0.010 0.249 *** 0.010
Upper secondary~mother 0.015 *** 0.003 0.017 *** 0.003
University < 3 years~mother 0.126 *** 0.006 0.128 *** 0.006
University ≥ 3 years~mother 0.232 *** 0.010 0.233 *** 0.010
Household income ‐0.002 0.003 0.002 0.003
Social welfare ‐0.100 *** 0.005 ‐0.098 *** 0.005
Unemployment benefit ‐0.046 *** 0.005 ‐0.045 *** 0.005
Single parent ‐0.038 *** 0.004 ‐0.035 *** 0.005
Youth unemployment ‐0.048 *** 0.005 ‐0.044 *** 0.005
Changes in the 1990s:From living in NEW ‐0.004 *** 0.029 0.058 *** 0.004
Female 0.052 *** 0.004 0.063 *** 0.006
Upper secondary~father 0.057 *** 0.006 0.060 *** 0.009
University < 3 years~father 0.051 *** 0.009 0.037 *** 0.007
University ≥ 3 years~father 0.025 *** 0.007 0.080 *** 0.007
Upper secondary~mother 0.075 *** 0.006 0.060 *** 0.009
University < 3 years~mother 0.052 *** 0.008 0.048 *** 0.008
University ≥ 3 years~mother 0.041 *** 0.007 0.013 *** 0.003
Household income 0.014 *** 0.003 0.108 *** 0.014
Social welfare 0.121 *** 0.014 0.038 *** 0.007
Unemployment benefit 0.037 *** 0.007 0.119 *** 0.010
Single parent 0.116 *** 0.009 0.047 *** 0.005
Youth unemployment 0.049 *** 0.006 0.197 *** 0.053
Interacted effects; Changes in the 1990s and of living in NEW: Female ‐0.033 *** 0.005
Upper secondary~father ‐0.048 *** 0.008
University < 3 years~father ‐0.049 *** 0.007
University ≥ 3 years~father ‐0.052 *** 0.007
Upper secondary~mother ‐0.041 *** 0.006
University < 3 years~mother ‐0.045 *** 0.007
University ≥ 3 years~mother ‐0.043 *** 0.008
Household income ‐0.010 *** 0.003
Social welfare 0.014 0.011
Unemployment benefit ‐0.008 0.008
Single parent ‐0.045 *** 0.014
Youth unemployment ‐0.009 *** 0.002
Base model
Marg. eff. Marg. eff.
Note: The standard errors are robust and adjusted by 283 clusters – equivalent to the municipalities accounted for in the paper. ***, **, * denote significance at 1, 5, and 10 percent, respectively. The cases of one‐parent and two‐parent households are considered in the estimations. Not presented in this paper, some other variables have been considered and analyzed as well; e.g. the municipalities adjoining the municipalities of the new universities have been included in the dummy NEW, resulting in somewhat weaker results than those presented here; variables of family and individual wealth have also been estimated – showing non‐significant results.
29
7 Conclusions and Discussions
Is the expansion of higher education a successful political means of decreasing educational
inequalities? The results here show that living and growing up in the same area as a rapidly
growing educational institution seems to have a strong positive effect on overall university
attendance. With a positive average marginal effect of six percentage points in Sweden, this
result supports the hypothesis that nearby-easy access encourages more people to attend
higher education (the local average effect is even higher, 41 percentage points). Local access
to higher education has most likely decreased the social distance to higher education,
meaning that the option of attending higher education rather than entering the local labor
market after upper secondary school has become a more common and possible alternative for
more socioeconomic groups in society. The fact that more individuals chose to attend higher
education further away in the 1990s among those living in the area of NEW, could be
interpreted as the choice of moving being considered less costly (a smaller risk) once the
decision to attend higher education at all has been made.
The relatively largest growth occurred among students whose parents had upper secondary
school as their highest education, which can be seen as some indications of political success
in equalizing educational opportunities. On the other hand, for youths whose parents only had
elementary education as their highest education and where the family overall had a low family
income – their probabilities of entering higher education seem to have become even lower in
the 1990s than before. This results can, however, be attributed to the fact that over the 25
years accounted for in this study, this group has become increasingly selective, as parental
generations have become increasingly educated over time.
References af Trolle, Ulf (1990), Mot en internationellt konkurrenskraftig AKADEMISK UTBILDNING,
Lund: Studentlitteratur. Angrist, Joshua and Allan B. Krueger (1999), Empirical Strategies in Labor Economic, in
Handbook in Labor Economics, ed. Orley C. Ashenfelter and David Card. Vol. 3A Amsterdam: North-Holland: Elsevier Science Publishers.
Becker, Gary S. (1964[1993]), Human Capital, 3 ed. Chicago: University of Chicago Press. Becker, Gary S. and Nigel Tomes (1979), An Equilibrium Theory of the Distribution of
Income and Intergenerational Mobility, Journal of Political Economy, 87:1153–89.
30
--- (1986), Human Capital and Rise and Fall of the Families, Journal of Labor Economics, 4(3):1-39.
Björklund, Anders, Mårten Palme and Ingemar Svensson (1995), Tax Reforms and Income Distribution: An Assessment Using Different Income Concepts, Swedish Economic Policy Review, 2:229–266.
Björklund, Anders and Mikael Jäntti (1997), Intergenerational Income Mobility in Sweden Compared to the United States, American Economic Review, 87:1009–1018.
Björklund, Anders, Mikael Lindahl and Krister Sund (2003), Family background and school performance during a turbulent era of school reforms, Swedish Economic Policy Review 10(2):111–136.
Brint, Steven and Jerome Karabel (1989), The Diverted Dream: Community Colleges and the Promise of Educational Opportunity in America 1900-1985, New York: Oxford University Press.
Broady, Donald, Mikael Börjesson and Mikael Palme (2002), Det svenska högskolefältet under 1990-talet: Den sociala snedrekryteringen och konkurrensen mellan lärosäten, in Perspektiv på högskolan - i ett förändrat Sverige, Stockholm: Högskoleverket.
Currie, Janet and Enrico Moretti (2003), Mother’s education and the intergenerational transmission of human capital: Evidence from college openings, The Quarterly Journal of Economics, 118(4): 1495-1532.
Dryler, Helen (1998). Educational Choice in Sweden: Studies on the Importance of Gender and Social Context, Stockholm University, Stockholm: Swedish Institute for Social Research No 31.
Dryler, Helene (1994), Etablering av nya högskolor - ett medel för minskad snedrekrytering, in Skola och Sortering - Studier av snedrekrytering och utbildningens konsekvenser, ed. Robert Eriksson and Jan O. Jonsson. Stockholm: Carlssons Förlag.
Currie, Janet and Enrico Moretti (2003), Mother’s education and the intergenerational transmission of human capital, The Quarterly Journal of Economics, 1998(4):1495-1523.
Edin, Per-Anders and Peter Fredriksson (2000), LINDA - Longitudinal INdividual DAta for Sweden, Working paper 2000:19, Uppsala, Sweden: Department of Economics, Uppsala University.
Erikson, Robert and Jan O. Jonsson (1993), Ursprung och utbildning, in SOU:1993:85, Stockholm: Fritzes.
Erikson, Robert and Jan O. Jonsson (1994), Sortering i skolan, Stockholm: Carlssons Bokförlag.
Erikson, Robert and Jan O. Jonsson (1996), Can Education be Equalized: The Swedish Case in Comparative Perspective, Boulder and Oxford: Westview Press, Social Inequality Series.
Erikson, Robert and Jan O. Jonsson (2007), Why educational expansion is not such a great strategy for equality: Theory and evidence for Sweden, in Stratification in Higher Education, ed. Adam Gamoran Yossi Shavit, Tichard T. Aurum and Gila Menahem. Stanford, CA: Stanford University Press.
Fasth, Eva (1980), Aspects on Relocalization of Higher Education. Göteborg: UHÄ National Board of Universities and Colleges.
Freeman, Richard B. (1980), The Facts about the Declining Economic Value of College, Journal of Human Resources, 15:124–142.
Government-bill (2000), Vuxnas lärande och utvecklingen av vuxenutbildningen. Stockholm: Proposition 2000/01:72.
31
Hammarström, Margareta (1996), Varför inte Högskola?, Göteborg: Universitatis Gothoburgensis.
Holm, Einar and Nils Häggström (1972), Högre utbildning - regional rekrytering och samhällsekonomiska kalkyler, SOU:1972:23, Stockholm: Fritzes.
HSV (1998), The Current Swedish Model of University Governamnce - Background and Description, Raport: 1998:10S, Stockholm: National Agency for Higher Education.
Kane, Thomas J. and Cecilia E. Rouse (1995), Labor Market Return to Two- and Four Year College, American Economic Review, 85(3):600–14.
--- (1999), The community college: educating students at the margin between college and work, Journal of Economic Perspectives, 13(1):63–84.
Kim, Lillemor (1998), Val och urval till högre utbildning: en studie erfarenheterna av 1977 års tillträdesreform, Uppsala: Uppsala Universitet.
Kjellström, Christian and Håkan Regnér (1999), The Effect of Geographical Distance on the Decision to Enroll in University Education, Scandinavian Journal of Education Research. 43(4): 335-348.
Leigh, Duane E. and Andrew M. Gill (2003), Do community college really divert the students from earning a bachelor’s degree?, Economic of Education Review, 22(1):23-30.
Leigh, Duane E. and Andrew M. Gill (2004), The effect of community colleges on changing students’ educational aspirations, Economic of Education Review, 23 (1):95-102.
Meghir, Costas and Mårten Palme (2005), Educational Reform, Ability and Parental Background, American Economic Review, 95(1):414–424.
MIS (2000), Utbildningsklassificering, vol 1, Statistic Sweden. OECD (1993), Education at a Glance, Paris: OECD. Moulton, Brent R (1986), Random group effects and the precision of regression estimates,
Journal of Econometrics, 23(3): 385-397. Rouse, Cecilia E. (1995), Democratization or Diversion? The effect of community colleges on
educational attainment, Journal of Business and Economic Statistics, 3(2):217–224. Roemer, John E. (2006), Democracy, Education, and Equality – Graz Schumpeter Lecture,
Economic Society Monographs, Cambridge University Press. Roemer, John E. (1996), Theories of Distributive Justice, Harvard University Press. Solon, Gary (1999), Intergenerational Mobility in the Labor Market, in Handbook in Labor
Economics, ed. Orley C. Ashenfelter and David Card. Vol. 3A Amsterdam: Elsevier Science Publishers.
SOU (1972), U68 – Högskolan, SOU1972:3, Stockholm: Fritzes. UHÄ (1989), Högskoleutbildningens framtida dimensionering, Stockholm: UHÄ1989:17. Öckert, Björn (2001), Effects of Higher Education and the Role of Admission Selection,
Stockholm University, Stockholm: Swedish Institute for Social Research No 52.
32
APPENDIX Table A.1 Description of the explanatory variables
Variable name Description
The 18‐year‐old:
Female 1 if female, 0 otherwise
NEW* 1 if the youth lives in the municipality that received
local access to a higher education through a new university
in 1977 or later, 0 otherwise.
Father’s highest education**:
Compulsory School ≤ 9 years 1 if it is compulsory schooling, 0 otherwise.
Upper Secondary ≤ 3 years 1 if it is upper secondary school, 0 otherwise.
University < 3 years 1 if it is less than 3 years of university, 0 otherwise.
University ≥ 3 years 1 if it is 3 years or more of university, 0 otherwise.
Mother’s highest education**:
Compulsory School ≤ 9 years 1 if it is compulsory schooling, 0 otherwise.
Upper Secondary ≤ 3 years 1 if it is upper secondary school, 0 otherwise.
University < 3 years 1 if it is less than 3 years of university, 0 otherwise.
University ≥ 3 years 1 if it is 3 years or more of university, 0 otherwise.
Economy of the Household:
Income*** Relative net‐income of the household.
Social Welfare 1 if the household receives social welfare, 0 otherwise.
Unemployment Benefit 1 if the household receives unemployment benefits, 0 otherwise.
Single parent 1 if the household consists of one adult (one parent), 0 otherwise.
Macro variable:
Youth Unemployment County unemployment rate for the age group 16‐24 years.
Year/Cohort dummies:
L1977 1 if the 18‐year old is 18 years in 1977, 0 otherwise.
L1978 1 if the 18‐year old is 18 years in 1978, 0 otherwise.
....
L2001 1 1 if the 18 year old is 18 years in 2001, 0 otherwise.
D90 1 for the years 1990‐2001, 0 for the years 1977‐1989. Note1: (*) Municipalities represented in NEW: Boden, Borlänge, Borås, Eskilstuna, Falun, Gotland, Gävle, Halmstad, Helsingborg, Härnösand, Jönköping, Kalmar, Karlskrona, Karlstad, Kristianstad, Luleå, Ronneby, Skövde, Sundsvall, Uddevalla, Vänersborg, Västerås, Växjö, Örebro, Örnsköldsvik and Östersund. For municipalities that obtained a university after 1977, the dummy variable shifts from zero to one the year the university was officially established there. An example: the University College of Blekinge (in the municipalities Karlskrona and Ronneby) was established in 1990, and the variable NEW has the value of zero before 1990, and one in 1990 and after for the municipalities Karlskrona and Ronneby. Note 2: (**) Information about individuals’ educational history in LINDA starts in 1990. Due to this lack of information for the cohorts prior to 1990, all information on parental educational level is based on the educational level that is registered in the census of 1990. Note3: (***) Family income is presented as relative net‐income (after tax reduction and received benefits) for the household to which the student belonged at the age of 18.
Z
i
Z
i itit
itit
HousholdFAMincome
FAMincomeincomeFamily
1 1/
_
where itincomeFamily _ stands for the nominal income of the household of student i at time t. t = 1968, ...,
2001) indicates the year the student turned 18. The sum of all nominal incomes in year t is divided by all households in the same year. In the two‐parent household case, the nominal family income has been divided by 1.7 to compare income levels with one‐parent households (see Björklund, Palme, and Svensson (1995)). Note 4: All individual data is based on the database LINDA. Youth unemployment is taken from the open database over unemployment in Sweden. The intergenerational connection between parents and their youths is based on the multigenerational register of Statistics Sweden over the population in Sweden. All information on activities within the higher educational system in Sweden is based on the Swedish Higher Education Register. All data has been provided by Statistics Sweden.
Table A2
Probit estim
ates of ‘The democratization effect of the expan
sion’
Estimating the probability of attending higher education in
Swed
en in
the 1980s, changes in the probabilities in
the 1990s, and the differences‐in‐differences effect of living
in a m
unicipality where a new
university college was established
in the late 1970s or later.
OLS
PROBIT
Coeff.
Std. D
ev.
Coeff.
Std. D
ev.
Coeff.
Std. D
ev.
Marg. Eff.
Std. D
ev.
Coeff.
Std. D
ev.
Marg. Eff.
Std. D
ev.
Effects on dem
ocratization:
From living in NEW
0.006
0.012
0.011
0.011
0.027
0.044
0.008
0.014
0.046
0.040
0.014
0.013
Female
0.036
***
0.004
0.039
***
0.004
0.159
***
0.015
0.050
***
0.004
0.171
***
0.015
0.053
***
0.005
Upper Secondary~father
0.029
***
0.004
0.031
***
0.004
0.129
***
0.012
0.041
***
0.004
0.139
***
0.013
0.044
***
0.004
University < 3 years~father
0.124
***
0.007
0.126
***
0.007
0.418
***
0.021
0.145
***
0.008
0.425
***
0.021
0.147
***
0.008
University ≥ 3 years~father
0.278
***
0.009
0.278
***
0.008
0.813
***
0.022
0.295
***
0.009
0.815
***
0.021
0.295
***
0.008
Upper Secondary~mother
0.008
**0.004
0.011
***
0.004
0.056
***
0.013
0.018
***
0.004
0.067
***
0.013
0.021
***
0.004
University < 3 years~m
other
0.143
***
0.006
0.145
***
0.006
0.482
***
0.018
0.167
***
0.007
0.488
***
0.017
0.169
***
0.007
University ≥ 3 years~m
other
0.272
***
0.009
0.273
***
0.009
0.827
***
0.025
0.301
***
0.010
0.828
***
0.024
0.301
***
0.009
Household income
‐0.038
***
0.010
‐0.031
***
0.009
‐0.141
***
0.031
‐0.044
***
0.009
‐0.106
***
0.026
‐0.033
***
0.008
Social W
elfare
‐0.097
***
0.004
‐0.094
***
0.004
‐0.684
***
0.036
‐0.163
***
0.007
‐0.672
***
0.036
‐0.160
***
0.007
Unem
ploym
ent Ben
efit
‐0.045
***
0.005
‐0.044
***
0.004
‐0.201
***
0.020
‐0.059
***
0.006
‐0.197
***
0.020
‐0.058
***
0.006
Single Paren
t‐0.079
***
0.004
‐0.075
***
0.004
‐0.310
***
0.017
‐0.090
***
0.005
‐0.292
***
0.017
‐0.086
***
0.005
Youth Unem
ploym
ent
‐0.051
***
0.006
‐0.047
***
0.006
‐0.177
***
0.021
‐0.055
***
0.007
‐0.161
***
0.020
‐0.050
***
0.006
Changes in
the 1990s:
From living in NEW
0.051
***
0.016
0.270
***
0.019
0.172
***
0.058
0.056
***
0.020
1.133
***
0.071
0.412
***
0.026
Female
0.093
***
0.004
0.099
***
0.005
0.289
***
0.013
0.094
***
0.005
0.314
***
0.015
0.103
***
0.005
Upper Secondary~father
0.103
***
0.006
0.108
***
0.006
0.322
***
0.020
0.107
***
0.007
0.353
***
0.019
0.118
***
0.007
University < 3 years~father
0.109
***
0.010
0.119
***
0.009
0.299
***
0.028
0.101
***
0.010
0.346
***
0.026
0.118
***
0.009
University ≥ 3 years~father
0.037
***
0.011
0.050
***
0.012
0.144
***
0.028
0.047
***
0.009
0.201
***
0.030
0.066
***
0.010
Upper Secondary~mother
0.117
***
0.005
0.122
***
0.005
0.398
***
0.019
0.133
***
0.007
0.444
***
0.019
0.149
***
0.007
University < 3 years~m
other
0.115
***
0.008
0.119
***
0.008
0.327
***
0.023
0.111
***
0.008
0.375
***
0.023
0.128
***
0.008
University ≥ 3 years~m
other
0.070
***
0.009
0.076
***
0.010
0.239
***
0.026
0.079
***
0.009
0.287
***
0.027
0.096
***
0.010
Household income
0.053
***
0.013
0.047
***
0.013
0.227
***
0.030
0.070
***
0.009
0.206
***
0.026
0.064
***
0.008
Social W
elfare
0.031
***
0.008
0.034
***
0.008
0.382
***
0.042
0.132
***
0.016
0.374
***
0.044
0.129
***
0.017
Unem
ploym
ent Ben
efit
0.035
***
0.005
0.036
***
0.005
0.161
***
0.022
0.052
***
0.008
0.157
***
0.024
0.051
***
0.008
Single Paren
t0.118
***
0.008
0.125
***
0.008
0.450
***
0.028
0.154
***
0.010
0.479
***
0.030
0.165
***
0.011
Youth Unem
ploym
ent
0.083
***
0.007
0.083
***
0.007
0.282
***
0.027
0.088
***
0.009
0.283
***
0.024
0.088
***
0.008
(Model with inteaction)
(Base model)
(Base m
odel)
(Model with inteaction)
Table A2 cont.
Changes in the 1990s and from living in NEW:
Female
‐0.046
***
0.008
‐0.172
***
0.026
‐0.050
***
0.007
Upper Secondary~father
Upper Secondary~father***
0.010
‐0.224
***
0.030
‐0.064
***
0.008
University < 3 years~father
‐0.058
***
0.013
‐0.239
***
0.041
‐0.068
***
0.010
University ≥ 3 years~father
‐0.065
***
0.012
‐0.246
***
0.040
‐0.070
***
0.010
Upper Secondary~mother
‐0.050
***
0.008
‐0.270
***
0.030
‐0.076
***
0.008
University < 3 years~mother
‐0.042
***
0.011
‐0.250
***
0.036
‐0.071
***
0.009
University ≥ 3 years~mother
‐0.045
***
0.010
‐0.242
***
0.036
‐0.069
***
0.009
Household income
‐0.011
0.012
‐0.079
***
0.015
‐0.025
***
0.005
Social W
elfare
‐0.033
***
0.012
‐0.071
0.052
‐0.021
0.015
Unemploym
ent Benefit
‐0.013
0.008
‐0.013
0.029
‐0.004
0.009
Single Parent
‐0.071
***
0.011
‐0.271
***
0.039
‐0.076
***
0.010
Youth Unemploym
ent
‐0.020
***
0.002
‐0.072
***
0.007
‐0.022
***
0.002
Year dummies :
L1977
0.071
***
0.007
0.080
***
0.006
0.367
***
0.023
0.126
***
0.008
0.414
***
0.023
0.143
***
0.009
L1978
0.097
***
0.008
0.104
***
0.008
0.448
***
0.027
0.157
***
0.011
0.488
***
0.026
0.172
***
0.010
L1979
0.092
***
0.007
0.099
***
0.007
0.431
***
0.026
0.150
***
0.010
0.473
***
0.025
0.166
***
0.010
L1980
0.077
***
0.007
0.085
***
0.007
0.381
***
0.025
0.131
***
0.009
0.424
***
0.026
0.147
***
0.010
L1981
0.118
***
0.009
0.123
***
0.008
0.518
***
0.030
0.183
***
0.012
0.552
***
0.029
0.196
***
0.012
L1982
0.155
***
0.013
0.158
***
0.012
0.645
***
0.047
0.233
***
0.019
0.668
***
0.044
0.242
***
0.018
L1983
0.163
***
0.014
0.164
***
0.014
0.677
***
0.051
0.245
***
0.020
0.694
***
0.048
0.252
***
0.019
L1984
0.137
***
0.012
0.140
***
0.011
0.584
***
0.043
0.209
***
0.017
0.606
***
0.041
0.217
***
0.016
L1985
0.108
***
0.010
0.111
***
0.009
0.471
***
0.035
0.165
***
0.013
0.495
***
0.033
0.174
***
0.013
L1986
0.087
***
0.009
0.089
***
0.009
0.380
***
0.035
0.131
***
0.013
0.402
***
0.035
0.139
***
0.013
L1987
0.052
***
0.006
0.055
***
0.006
0.245
***
0.023
0.082
***
0.008
0.267
***
0.024
0.089
***
0.009
L1988
0.026
***
0.004
0.028
***
0.005
0.135
***
0.019
0.044
***
0.006
0.152
***
0.021
0.049
***
0.007
L1990
‐0.228
***
0.007
‐0.245
***
0.008
‐0.738
***
0.022
‐0.170
***
0.004
‐0.828
***
0.028
‐0.183
***
0.004
L1991
‐0.251
***
0.009
‐0.267
***
0.008
‐0.823
***
0.029
‐0.183
***
0.004
‐0.905
***
0.028
‐0.193
***
0.004
L1992
‐0.320
***
0.015
‐0.333
***
0.013
‐1.056
***
0.054
‐0.210
***
0.006
‐1.128
***
0.048
‐0.217
***
0.005
L1993
‐0.408
***
0.024
‐0.418
***
0.022
‐1.352
***
0.092
‐0.234
***
0.006
‐1.415
***
0.082
‐0.237
***
0.006
L1994
‐0.385
***
0.024
‐0.397
***
0.022
‐1.279
***
0.090
‐0.229
***
0.007
‐1.346
***
0.081
‐0.233
***
0.006
L1995
‐0.409
***
0.024
‐0.420
***
0.022
‐1.350
***
0.090
‐0.233
***
0.006
‐1.417
***
0.080
‐0.237
***
0.006
L1996
‐0.443
***
0.025
‐0.453
***
0.024
‐1.461
***
0.095
‐0.239
***
0.006
‐1.523
***
0.087
‐0.242
***
0.005
L1997
‐0.479
***
0.024
‐0.489
***
0.022
‐1.572
***
0.092
‐0.245
***
0.005
‐1.634
***
0.083
‐0.247
***
0.005
L1998
‐0.474
***
0.019
‐0.485
***
0.017
‐1.549
***
0.072
‐0.244
***
0.004
‐1.616
***
0.065
‐0.247
***
0.004
L1999
‐0.537
***
0.015
‐0.548
***
0.015
‐1.776
***
0.059
‐0.253
***
0.004
‐1.843
***
0.054
‐0.255
***
0.004
L2000
‐0.452
***
0.015
‐0.469
***
0.014
‐1.601
***
0.057
‐0.247
***
0.003
‐1.696
***
0.053
‐0.250
***
0.003
L2001
‐0.520
***
0.011
‐0.538
***
0.011
‐3.318
***
0.089
‐0.283
***
0.004
‐3.429
***
0.091
‐0.284
***
0.004
_cons
0.233
***
0.009
0.206
***
0.011
‐0.890
***
0.023
‐1.022
***
0.035
Observations:
2,999,944
2,999,944
2,999,944
2,999,944
2,999,944
2,999,944
R‐squared
0.235
0.238
Pseuda R2
0.209
0.213
Log pseudolikelihood
‐142,563.63
‐141,888.29
Note: R
obust standard errors (shown in paren
thesis) are adjusted
by 283 clusters – eq
uivalen
t to the municipalities accounted for in the paper. ***, **, * den
ote
significance at 1, 5, and 10 percent, respectively.
Table A3
Probit estim
ates of ’The diversion effect of new university colleges’
Estimating the probability of attending an
old university in the 1980s, changes in probabilities in
the 1990s, and the differences‐in‐differences effect of living in a m
unicipally
where a new
university was established
in the late 1970s or later.
OLS
PROBIT
Coeff.
Std. D
ev.
Coeff.
Std. D
ev.
Coeff.
Std. D
ev.
Marg. Eff.
Std. D
ev.
Coeff.
Std. D
ev.
Marg. Eff.
Std. D
ev.
Effects on dem
ocratization:
From living in NEW
‐0.044**
0.020
‐0.041**
0.018
‐0.205**
0.092
‐0.043**
0.018
‐0.188**
0.084
‐0.039**
0.017
Female
0.013***0.003
0.015***0.003
0.092***
0.012
0.020***0.003
0.102***0.013
0.023***0.003
Upper Secondary~father
0.031***0.003
0.033***0.003
0.180***
0.012
0.041***0.003
0.192***0.013
0.044***0.003
University < 3 years~father
0.102***0.007
0.103***0.007
0.434***
0.024
0.116***0.008
0.443***0.024
0.119***0.008
University ≥ 3 years~father
0.254***0.009
0.254***0.009
0.829***
0.023
0.247***0.010
0.834***0.023
0.249***0.010
Upper Secondary~mother
0.007***0.003
0.009***0.003
0.067***
0.014
0.015***0.003
0.078***0.015
0.017***0.003
University < 3 years~m
other
0.115***0.006
0.117***0.006
0.471***
0.018
0.126***0.006
0.478***0.018
0.128***0.006
University ≥ 3 years~m
other
0.233***0.008
0.234***0.008
0.786***
0.024
0.232***0.010
0.790***0.024
0.233***0.010
Household income
‐0.011***0.004
‐0.006
0.004
‐0.010***
0.016
‐0.002
0.003
0.009
0.013
0.002
0.003
Social W
elfare
‐0.062***0.005
‐0.059***0.005
‐0.629***
0.039
‐0.100***0.005
‐0.617***0.039
‐0.098***0.005
Unem
ploym
ent Ben
efit
‐0.036***0.004
‐0.035***0.004
‐0.228***
0.025
‐0.046***0.005
‐0.224***0.025
‐0.045***0.005
Single Paren
t‐0.041***0.005
‐0.037***0.005
‐0.180***
0.021
‐0.038***0.004
‐0.164***0.022
‐0.035***0.005
Youth Unemploym
ent
‐0.051***0.005
‐0.048***0.005
‐0.215***
0.021
‐0.048***0.005
‐0.201***0.021
‐0.044***0.005
Changes in
the 1990s:
From living in NEW
‐0.005***0.030
0.177***0.032
‐0.018***
0.133
‐0.004
0.029
0.700***0.156
0.058***0.004
Female
0.064***0.004
0.072***0.004
0.222***
0.015
0.052***0.004
0.243***0.016
0.063***0.006
Upper Secondary~father
0.060***0.006
0.069***0.005
0.235***
0.023
0.057***0.006
0.260***0.021
0.060***0.009
University < 3 year s~ fa the r
0.062***0.009
0.079***0.010
0. 206***
0.031
0.051***0.009
0.241***0.032
0.037***0. 007
University ≥ 3 ye ars~father
0.031***0.009
0.054***0.010
0.108***
0.028
0. 025***0.007
0.156***0.029
0.080***0.007
Upper Se condary~mothe r
0.071***0.006
0.077***0.006
0. 309***
0.023
0.075***0.006
0.326***0.025
0.060***0. 009
University < 3 years~m
other
0.062***0.008
0.076***0.010
0.211***
0.028
0.052***0.008
0.241***0.031
0.048***0.008
University ≥ 3 years~m
other
0.050***0.009
0.064***0.010
0.172***
0.028
0.041***0.007
0.199***0.030
0.013***0.003
Household income
0.023***0.008
0.018**
0.009
0.063***
0.016
0.014***0.003
0.057***0.013
0.108***0.014
Social W
elfare
0.038***0.005
0.030***0.006
0.446***
0.042
0.121***0.014
0.406***0.044
0.038***0.007
Unem
ploym
ent Ben
efit
0.024***0.005
0.024***0.005
0.156***
0.028
0.037***0.007
0.160***0.028
0.119***0.010
Single Paren
t0.096***0.008
0.100***0.008
0.445***
0.030
0.116***0.009
0.453***0.031
0.047***0.005
Youth Unemploym
ent
0.055***0.007
0.055***0.006
0.219***
0.025
0.049***0.006
0.210***0.023
0.197***0.053
(Base model)
(Model with inteaction)
(Base model)
(Model with inteaction)
Table A3 cont.
Changes in the 1990s and from living in
NEW:
Female
‐0.046***0.008
‐0.160***0.028
‐0.033***0.005
Upper Secondary~father
‐0.065***0.009
‐0.248***0.043
‐0.048***0.008
University < 3 years~father
‐0.081***0.015
‐0.253***0.042
‐0.049***0.007
University ≥ 3 years~father
‐0.095***0.018
‐0.275***0.040
‐0.052***0.007
Upper Secondary~mother
‐0.049***0.008
‐0.205***0.032
‐0.041***0.006
University < 3 years~mother
‐0.073***0.018
‐0.232***0.039
‐0.045***0.007
University ≥ 3 years~mother
‐0.071***0.024
‐0.216***0.047
‐0.043***0.008
Household income
‐0.006
0.010
‐0.045***0.014
‐0.010***0.003
Social W
elfare
0.012
0.008
0.060
0.047
0.014
0.011
Unemploym
ent Benefit
‐0.005
0.006
‐0.039
0.036
‐0.008
0.008
Single Parent
‐0.053***0.016
‐0.229***0.079
‐0.045***0.014
Youth Unemploym
ent
‐0.012***0.003
‐0.041***0.010
‐0.009***0.002
Year dummies :
L1977
0.078***0.006
0.085***0.006
0.457***
0.027
0.124***0.009
0.495***0.026
0.137***0.009
L1978
0.102***0.008
0.108***0.008
0.552***
0.032
0.156***0.011
0.584***0.031
0.167***0.011
L1979
0.095***0.007
0.101***0.007
0.527***
0.027
0.147***0.009
0.561***0.026
0.158***0.009
L1980
0.080***0.007
0.086***0.006
0.458***
0.028
0.125***0.009
0.493***0.028
0.136***0.009
L1981
0.120***0.008
0.124***0.007
0.621***
0.030
0.179***0.011
0.648***0.029
0.189***0.011
L1982
0.153***0.012
0.154***0.011
0.754***
0.044
0.227***0.016
0.771***0.042
0.233***0.016
L1983
0.159***0.013
0.160***0.012
0.781***
0.048
0.236***0.018
0.792***0.046
0.240***0.017
L1984
0.132***0.011
0.133***0.010
0.654***
0.040
0.190***0.014
0.670***0.039
0.196***0.014
L1985
0.109***0.009
0.111***0.009
0.544***
0.034
0.152***0.011
0.560***0.032
0.158***0.011
L1986
0.089***0.009
0.090***0.008
0.430***
0.034
0.116***0.011
0.444***0.034
0.120***0.011
L1987
0.049***0.006
0.051***0.006
0.244***
0.024
0.061***0.007
0.258***0.025
0.065***0.007
L1988
0.024***0.005
0.026***0.005
0.132***
0.023
0.031***0.006
0.141***0.024
0.034***0.006
L1990
‐0.145***0.008
‐0.160***0.007
‐0.429***
0.025
‐0.076***0.005
‐0.462***0.023
‐0.080***0.004
L1991
‐0.146***0.012
‐0.160***0.011
‐0.417***
0.042
‐0.074***0.007
‐0.441***0.039
‐0.077***0.007
L1992
‐0.159***0.019
‐0.170***0.018
‐0.434***
0.078
‐0.076***0.011
‐0.447***0.075
‐0.078***0.011
L1993
‐0.172***0.030
‐0.180***0.029
‐0.438***
0.134
‐0.077***0.018
‐0.439***0.129
‐0.077***0.018
L1994
‐0.166***0.031
‐0.176***0.029
‐0.422***
0.134
‐0.075***0.019
‐0.425***0.130
‐0.075***0.018
L1995
‐0.189***0.030
‐0.199***0.029
‐0.508***
0.130
‐0.086***0.017
‐0.512***0.126
‐0.086***0.016
L1996
‐0.207***0.030
‐0.216***0.029
‐0.570***
0.130
‐0.093***0.015
‐0.571***0.127
‐0.093***0.015
L1997
‐0.225***0.030
‐0.234***0.028
‐0.641***
0.129
‐0.100***0.014
‐0.642***0.125
‐0.100***0.014
L1998
‐0.244***0.024
‐0.254***0.023
‐0.745***
0.099
‐0.110***0.010
‐0.751***0.096
‐0.110***0.010
L1999
‐0.283***0.021
‐0.293***0.020
‐0.931***
0.082
‐0.123***0.007
‐0.942***0.079
‐0.124***0.007
L2000
‐0.219***0.017
‐0.234***0.015
‐0.830***
0.063
‐0.117***0.006
‐0.858***0.058
‐0.119***0.006
L2001
‐0.273***0.018
‐0.289***0.017
‐2.312***
0.096
‐0.161***0.006
‐2.354***0.097
‐0.161***0.006
_cons
0.159***0.010
0.137***0.012
‐1.246***
0.041
‐1.351***0.051
Observations:
2,999,944
2,999,944
2,999,944
2,999,944
2,999,944
2,999,944
R‐squared
0.173
0.177
Pseuda R2
0.1744
0.1769
Log pseudolikelihood
‐118,938.56
‐118,583.54
Note: R
obust standard errors (shown in paren
thesis) are adjusted
by 283 clusters – eq
uivalen
t to the municipalities accounted for in the paper. ***
,**, * den
ote
significance at 1, 5, and 10 percent, respectively.
II
University choice and academic success in Sweden
By Susanna Holzer
Abstract We compare the performance of students in universities built before and after the large decentralization and expansion of the higher educational system in Sweden that started in the late 1970s. Two outcome measures are used: (i) whether or not the student has obtained a degree within seven years after initiating her studies; and (ii) whether or not she obtained 120 credit points (the requirement for most undergraduate degrees) within seven years. Controlling for several background variables as well as GPA scores in a binomial probit model, we show that students in old universities are about 5 percentage points more likely to get a degree and about 9 percentage points more likely to obtain 120 credit points. However, in an extended bivariate model, where we consider selection on unobservables into university type, we cannot reject the possibility of no difference in performance between the two university types.
JEL Classification: I23, I28, J24
Keywords: Higher Education, Government Policy, Human Capital
Correspondence address: Susanna Holzer, School of Management and Economics, Växjö University, SE-351 95 Växjö, Sweden. Phone: +46 470 70 85 79. E-mail: [email protected]. I am grateful to Mårten Palme, Håkan Locking, Thomas Lindh, Lennart Delander, Harald Niklasson, Abdullah Almasri, Ghazi Shukur and seminars participants at Växjö University, and conference participants at the conference Higher Education Systems, Decentralization and Educational Outcomes, Novara, Italy, November 2008, for helpful comments and suggestions. I acknowledge funding from Växjö University, Jan Wallander and Tom Hedelius’ Research Foundation, and the Swedish Research Council.
2
1 Introduction
Providing and ensuring access to education is one of the strongest political concepts of
democratization in most Western democracies.1 Increased access is considered to encourage
more socioeconomic groups in society to invest in education and thus, enhance (especially
young) people’s chances of equality of opportunity in life. Focusing only on the publicly
financed and centrally monitored higher education in Sweden in the following, it is shown
that policies of democratization and regionalization have lead to a dramatic change in the
geographical and physical access to higher education during the last fifty years. For instance,
in the 1960s, only a very select group in society, amounting to roughly 25,000 students,
attended higher education at the six (mostly metropolitan located) universities. Fifty years
later, more than 360,000 students attend higher education in more than 40 university areas
that are widely spread across Sweden.
Although the student body has grown to cover roughly half of all younger birth cohorts in
Sweden, the composition of the student body is still highly skewed. Among students born in
the 1970s and the 1980s, about 22 percent of the children with parents with compulsory
education as their highest education attended higher education, whereas the corresponding
attendance rate among children from parents with a post-graduate education was 86 percent.2
Once this skewed (but more heterogeneous as compared to the composition of the student
body in the 1960s) student body has entered the higher educational system, only modest
attention has so far been given to educational outcome. This issue of equality of opportunity
becomes even more essential knowing that less than half of all students who attended higher
education in the 1990s and the early 2000s managed (or chose) to complete their education
with a university degree, which should be compared to a completion rate of more than 80
percent in the 1960s.3
Just like in Sweden, there has been an increase in the number of institutions providing higher
education in the United States, which has resulted in a rapid growth of the student body. Also
in the US has there been a sharp fall in the relative completion rates. When Manski and Wise
1 By enhancing the access to postsecondary education the hypotheses is that community colleges may increase the years of schooling completed - an effect known as a democratization effect; see Karabel and Brint (1989), and Rouse (1995) for a longer discussion of the concept. 2 SCB (2008). 3 See SCB (1975, 2007).
3
(1983) first brought attention to this development in the US, they explained the increasing
drop-out rates as simply a matter of ability and taste for education. However, later studies
argue that especially high tuition costs at the older, more prestigious, and fewer four-year
colleges, as compared to the newer local community (two-year) colleges, inhibit the
enrollment of students from lower socioeconomic backgrounds; see Empty Promises (2002).
This should be kept in mind when considering studies by Dougherty (1994), Rouse (1995),
Kane and Rouse (1995, 1999) and Leigh and Gill (2003, 2004) who all find college choice to
be of importance for an individual’s chances for academic success. They argue that
community colleges have a diversion effect on overall educational attainment, in that they
show that students who first enter the local community colleges on average complete fewer
years of schooling than for students starting at four-year colleges.4 In terms of university
degrees, students who first attend a four-year college have a higher likelihood of ending up
with a bachelor’s degree, as compared to students who first attend a two-year college.
Turning back to the case of Sweden, smaller regional universities were introduced at the end
of the 1970s in order to complement the six older universities that are often regarded as more
prestigious. (Henceforth, the universities established prior to 1977 are referred to as old
universities and the rest as new universities.5) In contrast to the US higher educational system,
there are four important traits of the Swedish higher educational system that we must
consider: higher education in Sweden is i) centrally monitored and quality controlled by the
government; ii) publicly funded; iii) free of charge for the student (no fees!); iv) and the new
universities, like the old universities, are allowed to award students bachelor’s and (often)
master’s degrees.6
Keeping these traits of the Swedish higher educational system in mind, the following
questions are investigated; a) does academic success depend on university type in Sweden, i.e.
if the student attended an old or a new university? Since all institutions of higher education
are equally funded per student by the government, this action is interpreted as the assumption
4 A student who first enters a four-year college has a greater chance of ending up with at least a bachelor’s degree than a student that first enters a two-year college – even if they initially have similar abilities, and even if the student at the two-year college were given the chance of transferring to a four-year college. 5 In the following, both universities and university colleges are refereed to as universities, since both are allowed to proved educations at the master’s level in Sweden. 6 There are some examples of colleges in Sweden that do not have the Swedish government as their superior, e.g. Stockholm School of Economics and University College of Jönköping for example. However, they all rely on public funding for their undergraduate education and, like other colleges, they are subordinated Swedish law and regulations.
4
by the government being that educational outcome across institutions is homogeneous7; b)
does university choice affect a student’s educational prospects differently if we control for
his/her socioeconomic background? Performance accountability at the two university types is
explored as a mechanism for examining detailed student outcomes in order to analyze equity
issues.
Two probit models are employed in order to model how university type might affect
educational outcomes. First, we use a binomial probit model, where we control for possible
heterogeneity by including observed characteristics as regressors and second, as mentioned
above, the composition of the student body is skewed, so it is fair to assume that we face a
self-selection into universities.8 The effect of selection on unobservables that might affect
educational outcome that is correlated with initial university choice is estimated with a
bivariate probit model. The empirical material used in this study is a sample of 5,565
individuals that are extracted from the Swedish Longitudinal INdividual DAta (LINDA).
They all entered a Swedish university for the very first time during the years 1996-1999, i.e.
they constitute approximately three percent of all new enrolled students in that period of time.
When we do not control for selection on unobservables into university types, the results from
the binomial probit model show that we have an average university type effect of 5 and 9
percentage points on completing university with a degree or 120 credit points, respectively,
where attending an old university is more favorable for any of these. In this first case, family
background has no impact on educational outcome. When controlling for selection on
unobservables with a bivariate probit model, however, we found that on the probability of
completing 120 credit points or more, the selection parameter turned out to be significantly
different from zero and the coefficient for an old university was not significantly different
from zero. This means that we cannot rule out the possibility that the higher probability of
obtaining 120 credit points at older universities is attributed to selection on unobservables.
Once more, family background seems to have no impact on educational outcomes.
The paper is organized as follows: In Section 2, the institutional settings of higher education
in Sweden are described in brief. Section 3 provides an overview of the literature, Section 4 7 Student slots in the same subject major are equally funded across universities; however, the funding differs across the subject majors. More on this issue in the next section of this paper. 8 Also studied and discussed by Manski (1989), Altonji (1993), Light and Strayer (2000), and Arcidiacono (2004).
5
shows the empirical specification of the paper and Section 5 views the data and measurements
used. The empirical findings are presented in Section 6 and Section 7 concludes the paper.
2 Brief facts about Swedish Higher Education In 1977, there was a dramatic change in Swedish higher education, which is supply-side
oriented and centrally monitored by the Swedish government. The six older universities,
which constituted the sector of higher education and most of which were located in the
metropolitan areas in Sweden, were now complemented with (initially) 12 new and more
vocationally-oriented regional universities located all over Sweden. The main difference
between an old and a new university, i.e. a university established prior to and after 1977
is that only the old universities are entitled to award postgraduate degrees on a regular
basis and all prestigious educations in medicine, law and art are concentrated to the old
universities.9 In contrast to the US university system, all new universities in Sweden have
the right to provide educations up to the bachelor’s level and, in most cases, up to the
master’s level.
All student slots are restricted by the government which, in most cases, also sets the eligibility
requirements for a certain educational program or course. The students may, of course, choose
the education and the university they desire, but they have to compete for the limited amounts
of students slots, most often with their Grade Point Average (GPA) from upper secondary
school as the only means of competing. The selection of students is mostly made centrally by
the National Admissions Office to Higher Education (VHS), meaning that the universities
cannot choose their students themselves.10
9 The author is aware of the upgraded status that Luleå Technical University received in 1998, and Karlstad University, Växjö University, and Örebro University in 1999, where they all were granted the right to conduct research and offer graduate educations on a more permanent basis than what was previously the case. 10 The Swedish Agency for Higher Education (VHS). Since the outlined rules of admission are the only tools that must be considered by VHS in its selection process, the admission system can be said to be fairly transparent in the sense that the institutions of higher education cannot freely choose among eligible students. The admission rules currently in force in Sweden have been subjected to an intense debate since they were introduced in 1977/1979, but apart from some minor modifications, they still remain the same today. To be admitted to undergraduate education, the applicant must fulfill the basic eligibility requirements, which are the same for all courses and programs of education. Basic eligibility is obtained by having a degree from upper secondary school or if the applicant is 25 years of age and has at least four years of work experience and possesses a knowledge of English and Swedish comparable to that obtained at upper secondary school. In addition to the basic requirements of a degree from upper secondary school, the applicant can improve his/her chances by adding good scores from the Swedish Scholastic Aptitude Test (see Öckert (2001) for a thorough overview).
6
Basically all student performances are registered in the national system for documentation
of academic performances (Ladok).11 The system was jointly started by a majority of all
universities in Sweden in 1993. Although Ladok has been available since 1993/1994, the
administrative routines did not become stable until 1996/1997, which is the reason why
all aggregated input factors of Swedish higher educational production, Table 2.1, are only
presented for the years 1997 to 2005. Besides total production for all universities in
Sweden, educational production is also divided into the two university types of interest here.12
University is free of charge for the student and alternative funding is rare, so all universities
depend on public funding for their undergraduate education.13 A student slot in a specific
subject major is funded equally regardless of university, but educations in the natural
sciences, medicine, and technology (a group henceforth referred to as natural sciences)
receive more than twice the funding of educations in the humanities, law, and the social
sciences (a group henceforth referred to as social sciences). The proportion of educations in
the natural sciences is larger at the older universities as compared to the newer universities,
which could explain why the older universities in Table 2.1 on average receive slightly more
funding per student.
Table 2.1 also reports the input factor teachers. Here, all sorts of teacher resources are
included, i.e. from all levels of professors to graduate students acting as teaching
assistants – all transformed into full year equivalents. The older universities supply
roughly two-thirds of the higher educational production in Sweden and employ more
teachers than the newer universities. If we break down the teacher-resource to a teacher-
student ratio, we can see that students at the older universities on average have almost
twice the teacher resources per student than students at the new universities. The
additional funding received by the older universities could, of course, be a possible 11 ADB-baserat studiedokumentationssystem (Ladok). At the time of the introduction of Ladok, 21 of 40 universities were connected to the system, but by the end of the 1990s, 36 out of 40 universities were connected. Ladok is the largest source of information for the government on higher educational production. Universities that are not connected to Ladok are mainly universities specializing in art, drama, and music, but basically all universities that offer more general educations in the social sciences, the natural sciences, technology, the humanities, medicine etc. are connected to Ladok (see www.ladok.se). 12 Note that the sum of the two university types is slightly lower that the total for Sweden, since only facts based on universities that are included in the present study are reported, i.e 24 of 39 universities. More on this in the section on sample selection. 13 Even though not all universities in Sweden are officially subordinated the Swedish government, they are all dependent on public funding to survive and all educations must be in accordance with the Higher Education Act and the Higher Education Ordinance and thus, basically all universities are monitored by the government – directly or indirectly.
7
explanation for why they have more staff. Another explanation is that the older
universities have more access to cheaper teacher resources, e.g. graduate students and
research assistants. Since the new universities (given some exceptions) are not allowed to
conduct research on the same regular basis as old universities, they do not have any
automatic access to these possible additional teacher resources. Table 2.1 Average input factors of Swedish higher education for the years 1997-2005
All UniversitiesA Old UniversitiesB New UniversitiesB
Input factors
FundingC 17,083 9,235 4,689
Students 272,191 162,157 91,524
Funding per studentD 63 57 51
Teachers 24,861 17,526 5,766 Teacher Student Ratio 0.08 0.11 0.06
Note: A) All universities during 1997 to 2005. B) Only universities that are accounted for in this study. C) Funding is presented in billion SEK. D) Funding per student is presented in thousand SEK. Source: Statistics Sweden and the Agency for Higher Education (2008).
Educational performances by the students are measured in credit points, where one week
of successful full time study corresponds to one credit point. A student who studies full
time is assumed to obtain 40 credit points in one year. Most academic degrees require at
least 120 credit points (equivalent to three years of full-time study) – which is the lowest
credit point level requirement for receiving a bachelor’s degree. Table 2.2 shows student-
specific performances for students who entered higher education in Sweden during the
academic years 1993/94–2005/06. These are then followed for up to seven years or up to
the academic year 2005/06. Only about 16 percent have obtained at least 120 credit
points within three years, 54 percent within five years, and 61 percent within seven years.
After seven years, only marginal changes in educational performances occur. Therefore,
the follow up is restricted to seven years in the following.14
Given all sorts of academic degrees, roughly 6 percent had an academic degree after
three years, 29 percent after five years, and 45 percent after seven years. This leaves
about 20 percent of all students who have obtained 120 credit points or more with no
formal degree. A possible explanation for this is that the students themselves must apply
for a degree from the university they attended, and will obtain it if they fulfill the 14 A full-time student is granted student grants and loans for up to twelve semesters (= six years). This is most likely the reason why there is a rapid decrease in university activities after six years.
8
requirements set by the government and, to some extent, the university, i.e. they have passed
the required combination of courses. Among these 20 percent who lack a formal degree,
some of the students might lack one or more of the required courses to fulfill the degree
requirements, or may simply not value a formal degree so much as to consider it worth
applying for. A side effect of this is that it may make the Swedish official registers of the
educational level of the population in Sweden somewhat skewed and thus underestimate the
actual education level, since these registers often only consider the highest completed
degree.15
Table 2.2 Student performances for the academic years 1993/34 to 2005/06
Academic University Ladok 160 ≤ creditC
Year Entrants (in %) (in %)
3 years 5 years 7 years 3 years 5 years 7 years 7 years
Sweden ‐ All UniversitiesA
1993/94 59,490 79 10 32 46 17 52 59 35
1995/96 61,920 88 8 28 43 16 51 58 33
1997/98 58,930 93 6 28 44 16 54 61 35
1999/00 63,032 97 6 29 47 19 58 65 37
2001/02 69,514 98 5 29 20 57
2003/04 71,308 98 6 17
Old Universities B
1993/94 30,853 96 6 25 44 16 54 61 40
1995/96 32,022 97 4 20 38 16 53 60 39
1997/98 29,770 97 4 22 42 15 54 62 41
1999/00 30,654 97 5 26 47 18 60 60 38
2001/02 36,583 96 4 26 19 59
2003/04 37,508 96 6 17
New UniversitiesB
1993/94 20,496 78 10 35 45 19 49 55 26
1995/96 21,434 97 7 34 46 17 50 56 25
1997/98 22,892 95 6 30 44 18 52 59 28
1999/00 27,093 97 5 32 46 19 56 72 37
2001/02 29,007 94 5 33 21 64
2003/04 29,652 94 6 18
Awarded DegreeC
Passed 120 ≤credit pointsC
(in %) (in %)
Note: A) All universities in Sweden. B) Only universities that are accounted for in this study. This will be described in the next section. C) The credit point rate is only based on performances by students at universities affiliated with Ladok. An Academic Year starts at the end of August in one year and ends in early June the following year. Entrants refer to all first time enrolled students in higher education in Sweden. Here, only university entrants who are permanent residents in Sweden are accounted for. Ladok refers to how many of the students, expressed in percent, that entered a university that was affiliated with the Ladok‐system. All educational performances are presented as awarded degrees and credit points in percent of all university entrants. Observe that all sorts of degrees are accounted for here, i.e. they have a theoretical production time of two to five and a half years. The degree rate and the credit point rate are measured after 3, 5 and 7 years. Source: Statistic Sweden (2007).
Table 2.2 shows that new universities are better at producing degrees within three years,
as compared to older universities. However, it should be kept in mind that the table
shows all sorts of degrees, i.e. even degrees that require less than 120 credit points
15 Only in some cases are 120 credit points or more accounted for; see Statistic Sweden (2007).
9
(although they are few). Compared to the new universities, the older universities have a
larger share of educational programs that, in theory, require more time, i.e. three years or
more, which can most likely explain why most of the differences between the old and
new universities’ educational production have disappeared after seven years. In terms of
credit points, the older universities show a slightly higher rate of production than the new
universities and their share of students with at least 160 credit points (equivalent to four
years of full-time study) is also higher, when measured after seven years.
3 Previous literature
As in most fields in economics, the existing evidence on what determinates higher educational
completion is very much dominated by US research and the empirical research can be
considered from three angles.
From the first angle, we have the issue of the democratization effect of new establishments in
the US, i.e. the institutional impact on college completion of first attending a two-year college
instead of a four-year college. The empirical research is entirely based on National
Longitudinal Youths Surveys (NLYS) and the Longitudinal Study of High School Class of
1972 (NLS72). The advocates of geographically spreading higher education nationwide argue
that having access to a college nearby decreases the overall cost of attending higher education
– both in terms of lower tuitions fees , which is usually the case for the community (two-year)
colleges, and in terms of lower commuting costs. This does in particular encourage students
from non-academic backgrounds to invest in higher education to a larger extent than what was
previously the case (see e.g. Dougherty (1994), Rouse (1995)) and Leigh and Gill (2003,
2004)). The critics, on the other hand, argue that two-year colleges “channel” students into
vocational-oriented educations, away from studies for a bachelor’s degree – i.e., they divert
potential students towards settling for less education than what they have the capacity for (see
e.g. Karabel and Brint (1989), Rouse (1998) and Kane and Rouse (1995,1999)). In the US
system, students who initially entered a two year-college and who desire a bachelor’s degree
or an even higher degree must transfer to a four-year college. According to the critics, this
transfer decision is associated with high costs, which is the reason why students at the margin
between pursuing their studies at a four-year college or entering the local labor market (often
10
students from less fortunate backgrounds), may be lost to the latter. Rouse (1998, p. 602)
expresses the benefits for college completion of attending a four-year college already at the
start where “the four-year college environment helps keeping students ‘on-track’ and focused
on the bachelor’s degree” which, according to her, is not the case for students who first attend
a two-year college.
From the second angle, Light and Strayer (2000) try to answer the question of whether college
completion depends on college quality or student ability, when only controlling for quality
levels among four-year colleges.16 They use the NLYS, who all graduated from high school in
1978. In total, 780 four-year colleges were categorized into four quality groups. Their main
findings are that the more able (measured by pre-college educational performances) is an
individual, the more likely he/she is to attend higher education – regardless of the rank level
of the educational institution. However, in terms of attendance and completion decisions,
students tend to sort themselves by ability, i.e. the matching between student ability and
college quality is of importance for the likelihood of a student completing college. A low-
ability student has a higher likelihood of graduating if he/she attends a less select college –
compared to if he she attends a more highly ranked college; whereas a high-ability student is
more likely to graduate if he/she attends a high-quality college – compared to if he/she attends
a weaker school.
From the third angle, Altonji (1993) (followed by Arcidiacono (2004)) showed evidence of
within-university choices being far more important determinants of college completion than
college choice and family background. Based on NSL72 data, he found that within university
choices, like university majors (especially in the natural sciences), had a far greater impact on
university completion and on later economic outcomes as compared to the impact of
individual and family background characteristics and, in some cases, college choice.
A common criticism of most studies on college choice and college performances is that they
often fail to control for initial differences in student characteristics that may cause self-
selection to be inherent in the educational outcomes. As in the Light and Strayer (2000) case
above, better performing students "self-select" into better quality schools, and students either
16 See also Dale and Krueger (1998) and Brewer, Eide and Ehrenberg (1999) for more US examples.
11
choose to pursue their education or choose to drop out (see also e.g. Keane and Wolpin (1997)
and Cameron and Heckman (1998, 2001)).
4 Empirical specifications
4.1 Model 1: A binomial probit model on university completion
Two different models are used in the empirical analysis. The first is a binomial probit model:
(1) *iD = 0 + 1 OLDi + iX2 + iZ3 + iCohort4
iX5 * OLDi + iZ6 * OLDi + iCohort7 * OLDi + ,i
where the *iD is a latent variable measuring student performances and is defined as:
(2)
,0
1 *
otherwise
cDifD i
i ,
where Di is the binary educational outcome for student i, c is a threshold defining a degree or
a certain credit point amount, OLD is a dummy variable for attending an old university, Xi is a
vector of personal-specific characteristics and family background information, Zi is a vector
of within university choices (e.g. course majors, program participation, transfer decisions) and
Cohort is a vector of year dummies indicating to which of four ‘university-entry-cohorts’ the
student belongs. Xi, Zi, and Cohort are interacted with the dummy variable OLD in order to
study how the impact of the independent variables differs between students who attended an
old university and those who attended a new university.
1 is the key parameter, measuring what is the impact of attending an old university on the
probability of university completion, as compared to attending a new university. The politics
concerning the higher educational sector in Sweden assumes it to be homogenous and that the
university type is of no importance for the likelihood of educational success. For this to hold,
1 , 5 , 6 , and 7 should not differ from zero. i is a random error term representing all
omitted variables that might affect individual completion behavior and it is approximated by a
normal distribution.
12
4.2 Model 2: A bivariate probit model on university choice and university completion
University choice can be seen as a search process made by people in their discovery of what
they like and dislike in terms of education.17 By modeling this process into sequences, we
capture how possible effects of self selection into one university type endogenously affect the
probability of a positive educational outcome. Endogeneity is also assumed to be correlated
with unobservables that might affect the educational outcome of the students, e.g. teacher
quality, educational preferences, work preferences, peer group quality etc.
To model choice of university type and how possible endogeneity affects educational
outcomes, a recursive bivariate probit model is used (see e.g. Greene (2008)).18 The latent
variable specification for the bivariate probit is:
(3a) *iD = 0 + 1 OLDi + iX2 + iZ3 + iCohort4
iX5 *OLDi+ iZ6 * OLDi + Cohort7 * OLDi + ,i
(3b) *iOLD = 0 + iX1 + iCOM2 + Cohort3 + ,i
where both *iD and *
iOLD are latent variables measuring educational performances and
university choice, respectively, and they both follow the assumption made about *iD in
equation (2). Vectors Xi, Zi, and Cohort are the same as in the binomial probit model case. 17 This is inspired by Manski (1989), Altonji (1993), Light and Strayer (2000) and Arcidiacono (2004) who model the educational choices in sequences and the final outcome of an investment is assumed to be uncertain, but it is endogenously affected by earlier educational choices. 18 Some simplifications of the model are necessary. First, the sequential process of student application, university admission, and student acceptance decisions is collapsed into a single choice. Second, only attendance and graduation decisions at the first university are considered. Although one third of the students in the sample used in this study have recorded activities at other universities, besides the one to which they were first admitted, university-transfer or parallel studying decisions are beyond the scope of this study. Third, according to Altonji (1993) and Arcidiacono (2004), university majors affect the probability of graduating from university. They categorized university majors into two categories; mathematics (including technology and natural sciences) and humanities (including social sciences) and find that students in the mathematics category had a higher probability of completing university, as compared to the other category. This broad definition of university majors will be considered in this study, but only as an individual-specific factor that might influence the outcome in the second choice. Fourth, students tend to enroll and reenroll quite frequently at all universities, which is why continuous time-spells for the educational production are hard to come by. For Cameron and Heckman (1993) and Light (1995), the timing of the educational decision was in focus, in terms of education completion. Official statistics of educational performances at Swedish universities reveals that the activities at the universities after seven years are small, which is why the second decision – whether to complete university or not – will be collapsed into one single (person-specific) time period. A possible explanation for why the activity decreases so rapidly after six years is the financial support system for students which (given some exceptions), stops after twelve semesters (= six years).
13
According to Kjellström and Regnér (1999), the distance to an institution of higher education
is positively correlated with the cost of attending higher education – the longer the distance,
the higher the cost.19 So, in the present case, we can assume that if a student lives close to an
old university, it is more likely that this student will attend this institution instead of a new
university further away. The other way around is assumed to hold for students living close to
a new university. To account for possible effects of having one university type in the students’
direct neighborhood where they lived/grew up prior to university entry, the vector iCOM is
included in the model. iCOM contains two dummy variables which indicate if a student’s
residence one year prior to university entry was in the same or in a neighboring municipally to
a new and old university, respectively. and are parameters and i , and i are random
error terms that are assumed to be approximated by a normal bivariate distribution with
E( i )=E( i )=0, and var( i )=var( i )=1.
The identification of this model relies on the assumption that Zi is unrelated to the
student’s ability to complete a university education. Simultaneously estimating equation
(3a) and equation (3b), we allow the respective outcomes to be dependent on each other.
There are two core features of this model; one is that OLD in equation (3a) is
endogenous (i.e. directly an outcome of equation (3b)), the second is that the
simultaneous estimations allow the error terms to be correlated, i.e. cov( i , i ) = . The
basic assumption is: if there is an endogenous effect of choice of university type on
educational outcome that is not accounted for by the covariates in the model, i.e. if there
is a selection into university that indirectly affects educational outcome, this will be
caught by the covariance, , and it will differ significantly from zero.
5 Data and measurement The primary data source is the Longitudinal INdividual DAta for Sweden (LINDA), a random
sample of approximately three percent of the population in Sweden. The core data is based on
income-tax registers of 1994 and it contains a sample of roughly 300,000 individuals. In
19 Higher education is free of charge for students in Sweden. The costs referred to here are mainly those associated with room, board, and books. I assume that the overall cost for a student is lower if he/she can attend higher education within a commuting distance, while living with his/her parents, as compared to if the student has to move in order to attend higher education and explicitly pay for room and board.
14
addition, population censuses and other register based data are added (see Edin and
Fredriksson (2000) for a description). The data is traced back to 1968 and up to 2006, and
individuals who for some reason leave the registers (die or leave the country) are replaced,
leaving the data to be used as longitudinal and as a representative annual cross-section sample
of the population in Sweden. In case the LINDA subject is a part of a family, the family
members are also registered, but only as long as they share the same household as the main
subject. Intergenerational relations among the individuals in LINDA are controlled for
through the intergenerational register.
Information about the highest level of completed education, based on highest degree, is to be
found in LINDA. However, more detailed information about education prior to university and
any activities made by a LINDA subject within the higher educational system in Sweden is
added: grade point average (GPA) of the final grades from upper secondary school and its
track character are added through the Swedish upper secondary school register (that dates
back to 1973); the Swedish Higher Education Register reveals activities within the higher
educational system such as first time entrance, reenrollments, choice of university, choice of
subjects, choice of courses, length of courses, courses passed, if the course was part of an
educational program etc. The main information in the Swedish Higher Education Register is
based on Ladok. Only students who attended a Ladok-connected university are considered,
meaning that the education offered was of a more ‘general’ kind (i.e. the remaining
universities have no outspoken specialization in music, drama, art, agriculture or sports).
Observe that in 1996, the 24 Ladok-connected universities (as compared to 36 today) hosted
roughly 90 percent of all students.20
5.1 The Sample
The basic LINDA sample used here is a cross-section sample of 1996, conditioned on having
GPA from upper secondary school, at least one parent in the LINDA-data base, and being a
20 Old universities: Chalmers University of Technology, Göteborg University, Karolinska Institutet, Linköping University, Lund University, Stockholm University, The Royal Institute of Technology, Umeå University and Uppsala University. New universities: Blekinge Institute of Technology, Halmstad University, Högskolan Dalarna, Jönköping University, Kalmar University, Karlstad University, Kristianstad University, Luleå University of Technology, Mid Sweden University, Mälardalen University, University of Borås, Skövde University, Växjö University, University West and Örebro University.
15
first time university entrant between the years 1996 and 1999, which leaves a sample of 5,565
students.21 The final sample is presented in Table 5.1.
All individuals in the sample are aged 18 to 37, with an average age of 22 years, containing
slightly more women than men and with a proportion of 95 percent that were born in Sweden.
The socioeconomic background is here entirely based on the information on the subjects’
parents. A large number of the parents in LINDA are in the data as a family member and once
their child moves away from home, information about the parents is not followed any further.
Data on family background is therefore collected at the age of 18, an age at which the vast
majority of youths attend their last year of upper secondary school, at which preferences for
university are taking form, and at which most youths still live at home with their parents. Data
on parents is collected between the years 1977 and 1996. Table 5.1 shows that more than half of the parents have elementary or upper secondary school
as their highest education. Only about one percent of all students only have one parent.
Family income is measured as a relative income, i.e. the income of one family is set relative
to all households that had an 18-year old in the LINDA-data base the same year, i.e. roughly
11,000 households per year (about 10 percent of all 18 year-olds in Sweden) since 1977.22
The household income, by which the students was affected at the age of 18, shows that, on
average, students at the older universities come from households with a higher income as
compared to students at the new universities. The GPA is notably lower for students at the new universities as compared to students at the
old universities. As for socioeconomic background, there are also obvious differences among
the subsamples. Students at older universities have more highly educated parents and their
family income is higher on average. Both in terms of educational level and income, some of
the differences can be explained by the fact that the average population in Sweden is more
educated and have a higher income in the metropolitan areas, where most of the old
universities are located.
21 In order to avoid the risk that a student is recorded twice as a new student in the higher educational system in Sweden, the oldest students considered here are aged 37. They were 18 years in 1977, date of the creation of the Register of Higher Education and the likelihood of they having been enrolled prior to 1977 is very small. 22 See the Appendix for an explanation.
16
Table 5.1 Sample description
Variables Mean (Std. Dev). Mean (Std. Dev). Mean (Std. Dev).
Individual characteristicsAge 22.35 (4.12) 22.07 (4.67) 22.75 (4.35)
Woman* 0.56 (0.50) 0.55 (0.50) 0.58 (0.49)
Swedish* 0.95 (0.21) 0.95 (0.22) 0.96 (0.20)
Child* 0.05 (0.23) 0.04 (0.20) 0.07 (0.26)
Married* 0.06 (0.23) 0.04 (0.20) 0.08 (0.27)
GPA 14.69 (2.82) 15.13 (2.94) 14.10 (2.53)
‐ Natural Science track 0.33 (0.47) 0.39 (0.49) 0.26 (0.44)
‐ Social Sciences track 0.41 (0.49) 0.41 (0.49) 0.42 (0.49)
Lived in an area of NEW 0.41 (0.49) 0.26 (0.43) 0.62 (0.49)
Lived in an area of OLD 0.32 (0.46) 0.45 (0.49) 0.13 (0.33)
Family background at age 18
Father’s highest education:Elementary* 0.18 (0.38) 0.14 (0.35) 0.22 (0.41)
Upper Secondary* 0.32 (0.47) 0.29 (0.45) 0.36 (0.42)
University* 0.31 (0.46) 0.37 (0.48) 0.23 (0.48)
Missing* 0.18 (0.38) 0.19 (0.39) 0.18 (0.41)
Mother's highest education:Elementary* 0.18 (0.38) 0.15 (0.36) 0.21 (0.41)
Upper Secondary* 0.40 (0.40) 0.35 (0.50) 0.43 (0.50)
University* 0.38 (0.49) 0.44 (0.50) 0.29 (0.46)
Missing* 0.04 (0.20) 0.04 (0.20) 0.04 (0.19)
Income quota 1.02 (0.57) 1.06 (0.60) 0.96 (0.52)Single parent* 0.01 (0.10) 0.01 (0.09) 0.01 (0.10)
SAMPLE
New UniversitiesAll Universities
A
Old Universities
5,565 3,211 2,354
Note: A) All 24 universities accounted for in this paper. *Dummy variables. All variables are explained in the Appendix.
As mentioned in earlier sections, all student slots at all universities are restricted and subject o
competition. The most common way of competing is with GPA. GPA takes a value between 0
and 20, where 10 equals pass and 15 pass with distinction. The average GPA of the sample
presented in Table 5.1 of (14.69) is well above pass. That the older universities have a higher
GPA can partly be explained by the fact that many of students with high GPA apply to more
prestigious educational programs in medicine and law that are only offered at the old
universities. But only about 6-8 percent of all student slots at the older universities are
occupied by medical students and students in law, meaning that these prestigious educations
17
just constitute a small proportion of the total educational production at these universities and
can only partly explain why their GPA is higher.23
5.2 Educational performances
Turning to educational performances, Table 5.2 shows student performances after seven years
and they are broken down into subgroups based on individual and family characteristics and
by old and new universities. In the traditional case, only completed degrees are measured as a
performance. Since so many students obtain a considerable amount of credit points but lack a
formal degree, educational outcomes are measured both in terms of degree rates (Degree), and
in terms of obtained credit points (120 credits), i.e having obtained 120 credit points or more.
In the degree rate case, the rates of the old universities are slightly higher compared to those
of the new universities. The difference, however, increases more if we compare the rates of
students that have obtained 120 credit points or more. On average, a student at an old
university obtains 113 credit points and 57 percent of all its students manage to obtain at least
120 credit points. Comparable performances at the new universities are 95 credit points on
average, and 50 percent have obtained at least 120 credit points.
From the descriptive statistics in Table 5.2, there are no obvious differences as concerns
university types among any of the personal characteristics displayed. Students at the old
universities follow the descriptive statistics in Table 2.2 over Sweden, where they on average
take more credit points compared to students at the new universities. Once more, it must be
kept in mind that the proportion of longer educations is larger at the old universities as
compared to the new ones.
Men prefer studies in the natural sciences (Nat), including mathematics, technology etc., over
the social sciences (Soc), which include humanities, political science, law, economics, nursing
etc. Women seem to have preferences toward the social sciences. A higher proportion of
women attends educations that lead to a profession where the degree equals a license ( i.e a
professional degree, (License), nursing school for example. More than two thirds of all
courses for which both genders register are included in an educational program (Program)24 –
the proportion is slightly larger at the new universities as compared to the old universities.
23 Phone interview with Olof Nelsson, director at the Office of Evaluation, Lund University, December 1, 2008. Lund University is the largest university in Sweden. 24 A preset course package.
18
According to Light and Strayer (2000), roughly one third of all students in the United States
transferred to other universities. The Swedish transfer rate is about the same (Transfers).
However, although some of the students presented in the transfer rate do change universities,
the majority of the students stay at the university of first choice and study in parallel or get a
second degree at other universities while or after studying at the university of first choice.
This behavior is prevalent for both university types.
Table 5.2
Educational perform
ances within seven years after first entering a university
Stud
ent
Cred
it point cha
racter (>
50%) :
Tran
sfers
Entrants
Soc
Nat
Program
License
Degree
120 ≤
Cred
it points
Variables
Total
Mean
(Std.D.)
(in %)
(in %)
(in %)
(in %)
(in %)
(in %)
(in %)
All (24
) Universities
5,56
510
5.50
(69.11)
6333
728
4354
33
Old Universities
3,21
111
2.92
(72.96)
6135
698
4457
33Individu
al cha
racteristics:
Man
1,456
114.34
(75.13)
4650
734
4257
32Wom
an1,752
111.73
(71.09)
7323
6711
4557
34Sw
edish bo
rn3,041
111.23
(73.07)
6135
708
4457
34Not Swed
ish bo
rn170
107.37
(70.80)
6232
6811
3456
26Family backgroun
d at age 18
Parents' highe
st edu
catio
n:Elem
entary*
603
106.56
(72.03)
6530
6811
3754
32Upp
er Secon
dary*
1,104
105.50
(69.56)
6530
676
4454
29Co
llege*
1,504
120.91
(74.95)
5640
729
4760
37
New
Universities
2,35
495
.37
(62.07)
6630
749
4150
33Individu
al cha
racteristics:
Man
994
95.34
(63.44)
4848
802
3547
34Wom
an1,360
94.67
(61.06)
7816
7113
4553
33Sw
edish bo
rn2,251
95.74
(61.92)
6629
748
4151
33Not Swed
ish bo
rn103
87.29
(65.01)
5837
7711
3741
43Family backgroun
d at age 18
Parents' highe
st edu
catio
n:Elem
entary*
472
91.46
(62.21)
6729
748
3745
32Upp
er Secon
dary*
1,097
94.59
(60.91)
6728
748
4252
28Co
llege*
785
98.83
(63.48)
6332
759
4151
41
Educationa
l outcome:
Passed
Cred
it points
Note: *Here highest ed
ucation among both paren
ts is steering the categorization. Source: LINDA and Svenska Högskoleregistret .
20
6 Results
Table 6.1 presents the results of the binomial probit model on the probability of university
completion transformed into marginal effects, for both the educational outcome “degree”
(Degree) and obtaining 120 credit points or more (P120). The model is estimated with and
without interaction with a dummy variable OLD (1 equals attending an old university, zero
equals attending a new university), and is referred to the Base Model and the Interacted
Model in the table. The first model allows us to measure the average effect of the various
covariates in the model and through the interacted model, we can see if the effects of the
covariates differ in impact depending on choice of university type.
The results of key interest here are if university choice is of importance for educational
outcome. In the Base Model, we can see that attending an old university increases the
students’ chances of university completion with on average 5 and 9 percentage points, for the
educational outcomes Degree and P120, respectively. In the Interacted Model, we control for
the interacted effect of the covariates and university choice and besides some differences in
the interacted covariates, there is no additional university type effect on university
completion.
If we look at how individual and family background characteristics affect university
completion, almost none of the covariates of family background show any impact whatsoever.
The impact of age is overall significantly negative on university completion, which sounds
reasonable. The older the students are, the less likely they will be to finish, due to factors of
family formation and the outside labor market that affect a student’s chances and willingness
to complete a university education. However, the age impact is so small that it can hardly be
considered a huge determinant of university completion. Other factors like having a child or if
the student is married seem to have a far greater impact on P120 than age. Having a child
increases the probability by roughly 17 percentage points, and being married decreases the
probability by about 10 percentage points. That especially children have a positive impact
could be interpreted as parents on maternity leave combining child care with taking self-
contained courses at a university. This may also explain why the child effect (and the
marriage effect) is not to be found for the Degree-outcome, i.e. their incentive when studying
may not be to take a degree, just the courses themselves.
21
Heckman and Cameron (1993) argued that the value of the GPA is one of the greatest means
of forecasting whether a student will succeed at university or not. The higher the GPA, the
higher the likelihood of success. According to the Swedish results in Table 6.1, we can see
that this also seems to hold for Sweden, where the impact of GPA on university completion is
on average 2 percent, i.e. the higher the GPA, the greater the chances of university
completion. If we also take into account what track the student took in upper secondary
school, having a background in theoretical tracks like Social Sciences or Natural Sciences
increases the impact by roughly 5-10 percentage points on the outcome P120, as compared to
a more vocational-oriented track in upper secondary school.
Vocationally oriented programs at the universities, like the nurse-training program, are among
the largest educational programs. They are dominated by women and it is a profession in
which a degree is equivalent to a professional degree. This is most likely the reason why the
probability of obtaining a degree is higher if the student is a Woman, the university education
is oriented towards the Social Sciences (which here include the caring professions), if the
degree is equivalent to a License, and if the education is part of a preset course program,
Program. Here, upper secondary educational background seems to be of less importance.
Other huge educational programs at the universities are in engineering and teaching, for none
of which a license is required to practice in a corresponding profession. This could explain
why the pattern is not quite the same if we look at the educational outcome for P120. The
gender impact is gone, and the impact of License is roughly one-third in size compared to its
impact on a Degree. But, once more, participating in a set course program, Program,
increases the student’s chances of university completion by roughly 50 percentage points.
22
Majoring in the social sciences enhances the chances of university completion by roughly 9
and 14 percentage points, for educational outcome Degree and P120, as compared to
majoring in the natural sciences among those who study at a new university. Only in the
Degree-case does the impact of university subject major differ between the two university
types; the impact of studying in the social sciences is roughly 16 percentage points for
students attending a new university. At the old universities, on the other hand, we cannot see
that a subject major in the natural sciences or in the social sciences differs in impact on
university completion.25
The results from the bivariate probit model are shown in Table 6.2. Under Y1 in Table 6.2, the
outline is similar to that presented for the binomial probit estimations in Table 6.1, where the
probabilities of university completion are expressed as marginal effects. OLD is now
endogenous, however. If there are other factors associated with university choice that are not
accounted for in the model, e.g. selection on observables, this will be absorbed by the
covariance ρ. The probabilities of initial university choice, expressed as marginal effects, are
presented under Y2 in Table 6.2.
First of all, there are two outcomes that we want to highlight. By allowing the impact of
university type, OLD, on university completion to be endogenous under Y1 in Table 6.2, we
can see that this weakens its precision and its impact on university completion for both
educational outcomes, Degree and P120. The effect of university type is no longer significant.
Furthermore, in the P120-case, we can see that the covariance ρ is significant. This significant
and positive covariance means that we have a selection effect on observables of attending an
old university that is positively correlated with the probability for a student of completing
university with at least 120 credit points.
In the degree-outcome case, this effect of the selection on observables is not significant, nor
can we see any effect of choice on university type. A possible explanation is that the outcome
Degree only reflects the more vocational-oriented programs at the universities – and in some
cases, the students are more homogeneous across university types, whereas the educational
outcome P120 absorbs and reflects the entire educational production at both university types. 25 The university effect of studying in the social sciences instead of the natural sciences at the old universities in Sweden is roughly zero. This is traced by looking at the interacted models. In the Degree case, the marginal effect of attending an old university and subject majoring in the social sciences is 0.165-0.160=0.005 – i.e. almost no difference in the effect .
23
For all other covariates under Y1, the results are pretty much the same as in the binomial
probit models presented in Table 6.1.
In Table 6.2 under Y2, we can see that the covariates explaining the probability of attending
an old university act pretty much as expected. Having highly educated parents increases the
probability of attending an old university by roughly 8 percentage points. GPA has an overall
positive impact of on average 3 percentage points on attending an old university. Moreover,
knowing that the proportion of educations in the natural sciences is larger at the older
universities explains why having a background in the natural sciences from upper secondary
school increases the possibility of attending an old university by roughly 16 percentage
points, as compared to those choosing a more vocational-oriented education. A theoretical
background in the social sciences has a positive impact of about six percentage points on
attending an old university. This is a somewhat weaker impact, which can most likely be
explained by that fact that the younger universities to a larger extent constitute a possible
alternative for more students in the social sciences.
Controlling for several individual and family background variables as well as GPA scores in
the bivariate probit model when estimating the probability of attending an old university
allows us to exclude them in the interpretation of the covariance ρ. However, although we can
see that individual characteristics and family background are obviously of importance for
university choice, there are other unobserved factors, correlated with university choice that
seem to be of importance for university completion.
24
Table 6.1 Binomial probit model on university completion
Two educational outcomes are accounted for: Degree and P120. The results are transformed into marginal effects
Y1( Degree) Y1( P120)
Marg. Std. Err. Marg. Std. Err. Marg. Std. Err. Marg. Std. Err.
OLD 0.054 *** 0.015 0.040 0.188 0.091 *** 0.016 0.141 0.188
Age ‐0.005 ** 0.002 ‐0.006 0.004 ‐0.011 *** 0.003 ‐0.009 ** 0.004
Woman 0.059 *** 0.016 0.051 ** 0.025 0.017 0.016 0.031 0.025
Swedish 0.070 ** 0.033 ‐0.001 0.057 0.044 0.036 0.060 0.057
Child 0.049 0.047 0.019 0.065 0.175 *** 0.042 0.154 ** 0.061
Married ‐0.001 0.044 0.026 0.063 ‐0.101 ** 0.044 ‐0.060 0.063
GPA 0.022 *** 0.003 0.019 *** 0.005 0.024 *** 0.003 0.018 *** 0.005
‐Nature Science 0.012 0.022 0.046 0.034 0.097 *** 0.023 0.103 *** 0.034
‐Social Science ‐0.040 ** 0.020 ‐0.036 0.028 0.054 *** 0.020 0.042 0.029
Father Upper Sec. 0.034 ** 0.018 0.029 0.026 0.002 0.018 0.030 0.027
Father UNI 0.029 0.019 0.041 0.031 0.018 0.020 0.055 0.032
Mother Upper Sec. ‐0.002 0.019 ‐0.025 0.028 ‐0.003 0.020 ‐0.002 0.029
Mother UNI ‐0.017 0.021 ‐0.047 0.032 0.000 0.022 ‐0.043 0.034
Fam_inc 0.000 0.014 0.020 0.023 ‐0.004 0.014 0.015 0.024
Single ‐0.121 0.074 ‐0.025 0.122 ‐0.046 0.081 ‐0.002 0.124
Social Science 0.088 *** 0.017 0.165 *** 0.022 0.142 *** 0.019 0.142 *** 0.024
Program 0.437 *** 0.013 0.468 *** 0.020 0.503 *** 0.014 0.542 *** 0.021
Transfers ‐0.172 *** 0.015 ‐0.199 *** 0.023 ‐0.263 *** 0.015 ‐0.244 *** 0.024
License 0.225 *** 0.028 0.273 *** 0.043 0.073 *** 0.028 0.049 0.042
Cohort 1997 0.031 0.020 0.039 0.033 0.027 0.021 0.037 0.033
Cohort 1998 0.050 ** 0.021 0.018 0.032 0.065 *** 0.021 0.046 0.033
Cohort 1999 0.039 * 0.021 0.007 0.032 0.067 *** 0.021 0.076 ** 0.032
Interaction with OLD:
Age 0.000 0.005 ‐0.003 0.005
Woman ‐0.007 0.032 0.025 0.033
Swedish 0.110 0.072 ‐0.024 0.074
Child 0.072 0.095 0.040 0.094
Married ‐0.059 0.083 ‐0.086 0.089
GPA 0.004 0.006 0.010 0.006
‐Nature Science 0.004 0.040 ‐0.013 0.046
‐Social Science ‐0.061 0.043 0.018 0.041
Father Upper Sec. 0.008 0.036 ‐0.048 0.037
Father UNI ‐0.018 0.040 ‐0.059 0.041
Mother Upper Sec. 0.046 0.040 0.001 0.040
Mother UNI 0.057 0.044 0.070 0.044
Fam_inc ‐0.036 0.030 ‐0.034 0.031
Single ‐0.203 0.124 ‐0.089 0.163
Social Science ‐0.160 *** 0.029 0.001 0.033
Program ‐0.097 ** 0.041 ‐0.086 * 0.041
Transfers 0.039 0.034 ‐0.035 0.034
License ‐0.119 ** 0.050 0.043 0.057
Cohort 1997 ‐0.010 0.041 ‐0.016 0.043
Cohort 1998 0.058 0.043 0.035 0.043
Cohort 1999 0.058 0.042 ‐0.013 0.043
Log likelihood ‐3046.0197 ‐3020.1037 ‐2975.8287 ‐2965.2944
(Base Model) ( Interacted Model) (Base Model) ( Interacted Model)
Note: *, **, *** indicates a significance level at 10, 5, and 1 percent, respectively. Complete regressions are presented in the APPENDIX.
Table 6.2
Recursive bivariate probit m
odel on university choice and university completion
Two educational outcomes are accounted for: Deg
ree and P12
0. The results are transform
ed into m
arginal effects.
Y1( Degree)
Y1( P120)
Y2(OLD)
Y2(OLD)
when Y1(Degree)
when Y1(P120)
M
arg.
Std. Err. Marg.
Std. Err. M
arg.
Std. Err. M
arg.
Std. Err.
M
arg.
Std. Err. M
arg.
Std. Err.
OLD
0.026
0.038
0.012
0.192
0.022
0.039
0.064
0.193
Age
0.009***
0.002
0.009***
0.002
Age
‐0.005*
0.003
‐0.006
0.004
‐0.010***0.003
‐0.008**
0.004
Woman
0.001
0.015
0.001
0.015
Woman
0.059***0.016
0.051**
0.025
0.017
0.016
0.031
0.025
Swedish
‐0.041
0.034
‐0.040
0.034
Swedish
0.068**
0.033
‐0.005
0.057
0.039
0.036
0.049
0.057
Child
‐0.079*
0.046
‐0.078*
0.046
Child
0.046
0.047
0.017
0.065
0.169***0.043
0.148**
0.061
Married
‐0.078*
0.043
‐0.079*
0.043
Married
‐0.004
0.044
0.022
0.063
‐0.107**
0.044
‐0.069
0.063
GPA
0.032***
0.003
0.032***
0.003
GPA
0.023***0.003
0.020***0.005
0.026***0.003
0.020***0.005
‐Nature Science
0.158***
0.020
0.158***
0.020
‐Nature Science
0.017
0.023
0.052
0.034
0.109***0.023
0.118***0.034
‐Social Science
0.059***
0.019
0.059***
0.019
‐Social Science
‐0.037*
0.020
‐0.034
0.028
0.060***0.021
0.048*
0.029
Father Upper Sec.
‐0.009
0.017
‐0.009
0.017
Father Upper Sec
0.034*
0.018
0.029
0.026
0.001
0.018
0.029
0.026
Father UNI
0.079***
0.019
0.080***
0.019
Father UNI
0.031
0.020
0.043
0.032
0.023
0.020
0.061*
0.032
Mother Upper Sec.
0.008
0.019
0.008
0.019
Mother Upper Se
‐0.002
0.019
‐0.024
0.028
‐0.003
0.020
‐0.001
0.029
Mother UNI
0.080***
0.020
0.080***
0.020
Mother UNI
‐0.014
0.021
‐0.044
0.033
0.006
0.022
‐0.035
0.034
Fam_inc
0.012
0.013
0.012
0.013
Fam_inc
0.000
0.014
0.021
0.023
‐0.002
0.014
0.018
0.024
Single
0.016
0.077
0.015
0.077
Single
‐0.121
0.074
‐0.026
0.122
‐0.047
0.081
‐0.003
0.123
Cohort 1997
0.005
0.020
0.005
0.020
Social Science
0.089***0.017
0.165***0.022
0.142***0.018
0.142***0.024
Cohort 1998
‐0.022
0.020
‐0.022
0.020
Program
0.436***0.013
0.468***0.020
0.501***0.014
0.543***0.022
Cohort 1999
‐0.041**
0.020
‐0.041**
0.020
Transfers
0.172***0.015
‐0.198***0.023
‐0.262***0.015
‐0.239***0.024
Live near Old
0.240***
0.017
0.239***
0.017
License
0.225***0.028
0.272***0.043
0.074***0.028
0.048
0.042
Live near New
‐0.255***
0.016
‐0.256***
0.016
Cohort 1997
0.031
0.020
0.040
0.033
0.027
0.021
0.039
0.033
Cohort 1998
0.049**
0.021
0.018
0.032
0.063***0.021
0.046
0.033
Cohort 1999
0.038*
0.021
0.006
0.032
0.064***0.021
0.073**
0.032
Interaction with OLD:
Age
0.001
0.005
‐0.002
0.005
Woman
‐0.007
0.032
0.026
0.033
Swedish
0.112
0.072
‐0.017
0.073
Child
0.073
0.095
0.039
0.094
Married
‐0.058
0.083
‐0.083
0.089
GPA
0.004
0.006
0.009
0.006
‐Nature Science
‐0.063
0.043
‐0.017
0.046
‐Social Science
0.004
0.040
0.019
0.041
Father Upper Sec.
0.008
0.036
‐0.049
0.037
Father UNI
‐0.018
0.039
‐0.060
0.041
Mother Upper Sec.
0.046
0.040
0.001
0.040
Mother UNI
0.057
0.044
0.069
0.044
Fam_inc
‐0.037
0.030
‐0.036
0.031
Single
‐0.203
0.124
‐0.088
0.161
Social Science
‐0.160***0.029
0.000
0.032
Program
‐0.099**
0.041
‐0.092**
0.041
Transfers
0.036
0.034
‐0.043
0.034
License
‐0.118**
0.050
0.046
0.057
Cohort 1997
‐0.011
0.041
‐0.019
0.043
Cohort 1998
0.057
0.043
0.032
0.043
Cohort 1999
0.058
0.042
‐0.013
0.043
Roh (coefficient)
0.051
0.064
0.055
0.065
0.122**
0.064
0.141**
0.065
Log likelihood
‐6160.964
‐6134.9958
‐6089.2398
‐6078.1525
(Base M
odel)
(Interacted M
odel)
(Base M
odel)
(Interacted M
odel)
Note: *, **, ***
indicate a significance level at 10, 5, and 1 percent, respectively. Y1=1
if the studen
t has obtained
120 credit points or more or has taken
a degree, and zero
otherwise. Y2=1
if an OLD
university, zero if a new
university. The outcome for Y2
is the same, given
both the base and the interacted
model in
Y1, so only one is presented
here from the respective educational outcome. Complete regressions are presented in the APPEN
DIX.
26
7 Conclusions
Is university type of any importance for university completion and is university success
affected by family background in Sweden? In a binomial probit model, where we do not
control for selection on observables that might affect the students’ choice of university type
which, in turn, might affect educational outcomes, the empirical results show that we have an
average university type effect of 5 and 9 percentage points, respectively, on a student’s
chances for educational success, where attending an old university increases the chances. In
this first case, family background has no impact on educational outcome.
In addition, we also estimate a bivariate model, where selection on unobservables into the
two types of universities is considered. For one of the outcomes, i.e. completing more than
120 credit points, the selection parameter turned out to be significantly different from zero
and the coefficient for an old university was not significantly different from zero. This means
that we cannot rule out the possibility that the higher probability of obtaining 120 credit
points in older universities is attributed to selection on unobservables.
References
Altonji, Joseph G. (1993), The Demand for and Return to Education When Education Outcomes are Uncertain, Journal of Labor Economics, 11(1): 48-83.
Arcidiacono, Peter (2004), Ability sorting and the returns to university major, Journal of Econometrics, 121(1-2): 343-375.
Björklund, Anders, Mårten Palme and Ingemar Svensson (1995), Tax Reforms and Income Distribution: An Assessment Using Different Income Concepts, Swedish Economic Policy Review, 22(2):267–269.
Brint, Steven and Jerome Karabel (1989), The Diverted Dream: Community Colleges and the Promise of Educational Opportunity in America 1900-1985, New York: Oxford University Press.
Cameron, Stephen, and James Heckman (1993), The Nonequivalence of High School Equivalents, Journal of Labor Economics, 11(1): 1-47.
--- (1998), Life Cycle Schooling and Dynamic Selection Bias: Models and Evidence for Five Cohorts of American Males, Journal of Political Economy, 106(2): 262-333.
--- (2001), The Dynamics of Educational Attainment for Black, Hispanic, and White Males, Journal of Political Economy, 109(3): 455-499.
Dale, Stacy Berg and Alan B. Krueger (2002), Estimating the Payoff to Attending A More Selective University: An Application of Selection on Observables and Unobservables, The Quarterly Journal of Economics, 117(4):1491-1527.
Dougherty, Kevin James (1994), The contradictory community university: the conflicting origins, impacts and futures of the community university, Albany, NY: State University of New York Press.
27
Edin, Per-Anders and Peter Fredriksson (2000), LINDA - Longitudinal INdividual DAta for Sweden.Working Paper 2000:19, Uppsala, Sweden: Department of Economics, Uppsala University.
Eliasson, Kent (2006), College Choice And Earnings Among University Graduates In Sweden, Umeå Economic Studies 693, Umeå University, Department of Economics.
Empty promises: The myth of university access in America (2002), Washington, DC: Advisory Committee on Student Financial Assistance.
Eherenbergh, Ronald. G. (2004), Econometric studies of higher education, Journal of Econometrics, 121(1-2):19-37.
Greene, William H. (2008), Econometric Analysis, sixth edition, Prentice Hall. Kane, Thomas J. and Cecilia E. Rouse (1995), Labor Market Return to Two- and Four Year
College, American Economic Review, 85(3):600–614. --- (1999), The community university: educating students at the margin between university
and work, Journal of Economic Perspectives, 13(1):63–84. Kjellström, Christian and Håkan Regnér (1999), The Effect of Geographical Distance on the
Decision to Enroll in University Education, Scandinavian Journal of Education Research, 43(4): 335-348.
Keane, Michael P. and Kenneth I. Wolpin (1997), The Career Decisions of Young Men, Journal of Political Economy, 105(3):473-522.
Leigh, Duane E. and Andrew M. Gill (2003), Do community universities really divert students from earning bachelor's degrees?, Economics of Education Review, 22: 23–30.
Leigh, Duane E. and Andrew M. Gill (2004), The effect of community universities on changing educational aspirations, Economics of Education Review, 23:95–102.
Light, Audrey A., and Wayne Strayer (2000), The determinants of university completion: school quality or student ability?, Journal of Human Resources, 35(2):299–332.
Light, Audrey (1995), Hazard Model Estimates of Decision to Reenroll in School, Labour Economics, 2(4):381-406.
Manski, Charles F., (1989), Schooling as experimentation: a reappraisal of the postsecondary dropout phenomenon, Economics of Education Review, 8(4):305-312.
Rouse, Cecilia E. (1995), Democratization or diversion? - The effect of community universities on educational attainment, Journal of Business and Economic Statistics 13(2):217–224.
--- (1998), Do Two-Year Colleges Increase Overall Educational Attainment? Evidence from the States, Journal of Policy Analysis and Management, 17(4): 595-620.
Statistic Sweden (SCB) (1975), Högskolestatistik I. Nyinskrivna, närvarande och examinerade vid universitet och högskolor 1962/63 – 1971/72. (Promemoria från SCB 1975:2) Stockholm: Statistiska centralbyrån.
--- (2007): Universitet och Högskolor. Genomströmning och resultat i högskolans grundutbildning t.o.m.2005/06. Statistiska meddelanden UF 20 SM 0702.
--- (2008), Universitet och högskolor - Högskolenybörjare 2007/08 och doktorandnybörjare 2006/07 efter föräldrarnas utbildningsnivå [Higher education. Level of parental education], Statistiska meddelanden UF 20 SM 0802.
--- (2009): Befolkningens utbildning 2008[Educational attainment of the population 2008]. Statistiska meddelanden UF 37 SM 0901.
Öckert, Björn (2001), Effcets of Higher Education and the Role of Admission Selection, Stockholm University, Stockholm: Swedish Institute for Social Research No 52.
Alternative sources: Internet: www.ladok.se (December 1, 2008) Phone interview: Olof Nelsson,, Director of the Office of Evaluation, Lund University, (December 1, 2008), Lund University is the largest university in Sweden.
28
APPENDIX
A. Variable descriptions Table A1 1. Variable description
Variables Description
120 credits 1 if the students obtains has least 120 credit points, 0 otherwise
Age Numerical, 18‐46 years
Child 1 if the student has at least 1 child, 0 othertwise
Cohort 1996 1 if the student was enroled for the first time in 1996, 0 otherwise
Cohort 1997 1 if the student was enroled for the first time in 1997, 0 otherwise
Cohort 1998 1 if the student was enroled for the first time in 1998,, 0 otherwise
Cohort 1999 1 if the student was enroled for the first time in 1998, 0 otherwise
Degree 1 if the students obtains a degree, 0 otherwise
Lived in an area of NEW 1 if the student lived in the same or in the direct neighbor municipally of a new
university the year before university entrance otherwiseLived in an area of OLD 1 if the student lived in the same or in the direct neighbor municipally of an old
old university the year before university entrance, 0 otherwise
Fam_inc A Family income, measured as relative income
Father Elementary 1 if the father has elementary education as highest education, 0 otherwise
Father Missing 1 if information about father's education is missing, 0 otherwise
Father UNI 1 if the father has a university education, 0 otherwise
Father Upper 1 if the father has upper secondary school as highest education, 0 otherwise
GPA Grade Point Average from upper secondary School with a value between 0 and 20.
‐ Natural Sciences track 1 if a theoretical track in natural or technology track in upper secondary school, 0 otherwise
‐ Social Sciences track 1 if a theoretical track in social studies, economics or humanities in upper secondary school, 0 othe
License 1 if the education results in an occupation where the degree equals a license to practice it, 0 other
Male 1 if the students is a man, 0 otherwise
Married 1 if the student is married/cohabiting, 0 otherwise
Mother Elementary 1 if the mother has elementary education as highest education, 0 otherwise
Mother Missing 1 if information about mother's education is missing, 0 otherwise
Mother UNI 1 if the mother has a university education, 0 otherwise
Mother Upper 1 if themother has upper secondary school as highest education, 0 otherwise
Nature 1 if university the major is in mathematics, technology, chemistry, biology, etc. , 0 otherwise
Old 1 if an old college, zero equals a new college
Program 1 if the student is enrolled in an educational program, 0 otherwise
Single 1 if the family is a single‐parent household
Swedish 1 if the student was born in Sweden
Transfers 1 if Ladok has records of activities at other colleges, 0 otherwise Note: A) Family income is presented as relative net income (after tax reduction and received benefits) for the household to which the student belonged at the age of 18.
Z
i
Z
i itit
itit
HousholdFAMincome
FAMincomeincomeFamily
1 1/
_
where itincomeFamily _ stands for the nominal income of the household of student i at time t. t =
(1968,...,1996) indicates the year in which the student turned 18. The sum of all nominal incomes year t is divided by all households the same year. In the two‐parent household case, the nominal income has been divided by 1.7 to able to compare one‐parent households with two‐parent households incomes of the 18‐year old (see Björklund, Palme, and Svensson (1995).
29
B. Regression results Table: B1: Binomial probit model – Educational outcome: Degree
Y1 (Degree) Coef. Std. Err. Coef. Std. Err.
OLD 0.141 *** 0.040 0.105 0.492
Age ‐0.012 ** 0.006 ‐0.015 0.010
Woman 0.154 *** 0.041 0.133 ** 0.065
Swedish 0.187 ** 0.092 ‐0.003 0.148
Child 0.125 0.119 0.050 0.167
Married ‐0.003 0.113 0.066 0.160
GPA 0.057 *** 0.007 0.049 *** 0.012
‐Nature Science 0.032 0.058 0.119 0.087
‐Social Science ‐0.104 ** 0.051 ‐0.094 0.074
Father Upper Sec. 0.089 ** 0.046 0.075 0.068
Father UNI 0.075 0.050 0.105 0.081
Mother Upper Sec. ‐0.004 0.050 ‐0.064 0.074
Mother UNI ‐0.043 0.054 ‐0.123 0.085
Fam_inc ‐0.001 0.036 0.052 0.060
Single ‐0.335 0.221 ‐0.067 0.325
Social Science 0.232 *** 0.046 0.441 *** 0.062
Program 1.341 *** 0.052 1.474 *** 0.088
Transfers ‐0.462 *** 0.042 ‐0.539 *** 0.066
License 0.572 *** 0.071 0.697 *** 0.114
Cohort 1997 0.081 0.053 0.101 0.084
Cohort 1998 0.129 ** 0.053 0.046 0.083
Cohort 1999 0.101 * 0.053 0.018 0.082
Interaction with OLD:
Age 0.001 0.013
Woman ‐0.019 0.083
Swedish 0.288 0.189
Child 0.184 0.240
Married ‐0.158 0.229
GPA 0.011 0.015
‐Nature Science 0.010 0.104
‐Social Science ‐0.162 0.116
Father Upper Sec. 0.021 0.092
Father UNI ‐0.047 0.104
Mother Upper Sec. 0.119 0.101
Mother UNI 0.147 0.112
Fam_inc ‐0.095 0.078
Single ‐0.605 0.454
Social Science ‐0.429 *** 0.080
Program ‐0.255 ** 0.110
Transfers 0.101 0.086
License ‐0.326 ** 0.147
Cohort 1997 ‐0.026 0.108
Cohort 1998 0.148 0.109
Cohort 1999 0.150 0.108
Constant ‐2.244 *** 0.240 ‐2.098 *** 0.374
Log likelihood ‐3046.0197 ‐3020.1037 Note: *, **, *** indicates a significance level at 10, 5, and 1 percent, respectively.
30
Table: B2: Binomial probit model – Educational outcome: P120
Y1(P120) Coef. Std. Err. Coef. Std. Err.
OLD 0.228 *** 0.040 0.355 0.479
Age ‐0.027 *** 0.006 ‐0.023 ** 0.009
Woman 0.043 0.041 0.078 0.064
Swedish 0.111 0.090 0.150 0.144
Child 0.455 ****0.118 0.399 ** 0.166
Married ‐0.254 ** 0.113 ‐0.150 0.159
GPA 0.060 *** 0.007 0.045 *** 0.012
‐Nature Science 0.246 *** 0.058 0.259 *** 0.086
‐Social Science 0.137 *** 0.051 0.105 0.073
Father Upper Sec. 0.005 0.046 0.074 0.067
Father UNI 0.046 0.050 0.137 * 0.080
Mother Upper Sec ‐0.008 0.050 ‐0.005 0.073
Mother UNI ‐0.001 0.055 ‐0.108 0.084
Fam_inc ‐0.010 0.036 0.037 0.060
Single ‐0.116 0.204 ‐0.005 0.311
Social Science 0.358 *** 0.047 0.357 *** 0.061
Program 1.399 *** 0.049 1.545 *** 0.081
Transfers ‐0.673 *** 0.041 ‐0.623 *** 0.065
License 0.186 *** 0.072 0.123 0.107
Cohort 1997 0.067 0.053 0.093 0.083
Cohort 1998 0.164 *** 0.054 0.117 0.082
Cohort 1999 0.168 *** 0.053 0.191 ** 0.081
Interaction with OLD:
Age ‐0.007 0.013
Woman 0.062 0.083
Swedish ‐0.060 0.185
Child 0.101 0.239
Married ‐0.217 0.227
GPA 0.024 0.015
‐Nature Science ‐0.032 0.116
‐Social Science 0.046 0.104
Father Upper Sec. ‐0.121 0.092
Father UNI ‐0.147 0.104
Mother Upper Sec. 0.002 0.101
Mother UNI 0.176 0.112
Fam_inc ‐0.087 0.079
Single ‐0.225 0.414
Social Science 0.003 0.082
Program ‐0.215 ** 0.103
Transfers ‐0.089 0.084
License 0.109 0.144
Cohort 1997 ‐0.041 0.108
Cohort 1998 0.088 0.109
Cohort 1999 ‐0.032 0.108
Constant ‐1.760 *** 0.234 ‐1.885 *** 0.362
Log likelihood ‐2975.8287 ‐2965.2944 Note: *, **, *** indicates a significance level at 10, 5, and 1 percent, respectively.
31
Table: B3: Recursive bivariate probit model – Educational outcome: Degree
Y1(Degree) Coef. Std. Err. Coef. Std. Err.Y2(OLD) Coef. Std. Err. Coef. Std. Err.
OLD 0.069 0.099 0.031 0.499 Age 0.023 *** 0.006 0.023 *** 0.006
Age ‐0.012 * 0.007 ‐0.015 0.010 Woman 0.002 0.040 0.002 0.040
Woman 0.154 *** 0.041 0.133 ** 0.065 Swedish ‐0.106 0.089 ‐0.106 0.089
Swedish 0.181 ** 0.092 ‐0.012 0.149 Child ‐0.201 * 0.117 ‐0.201 * 0.117
Child 0.119 0.119 0.043 0.168 Married ‐0.199 * 0.109 ‐0.199 * 0.109
Married ‐0.011 0.114 0.058 0.161 GPA 0.083 *** 0.007 0.083 *** 0.007
GPA 0.059 *** 0.008 0.051 *** 0.013 ‐Nature Science 0.420 *** 0.054 0.420 *** 0.054
‐Nature Science 0.044 0.060 0.135 0.088 ‐Social Science 0.154 *** 0.049 0.154 *** 0.049
‐Social Science ‐0.098 * 0.052 ‐0.087 0.074 Father Upper Sec ‐0.024 0.044 ‐0.024 0.044
Father Upper Se 0.089 * 0.046 0.075 0.068 Father UNI 0.208 *** 0.049 0.208 *** 0.049
Father UNI 0.080 0.050 0.112 0.081 Mother Upper Se 0.022 0.049 0.021 0.049
Mother Upper S ‐0.004 0.050 ‐0.064 0.074 Mother UNI 0.208 *** 0.053 0.208 *** 0.053
Mother UNI ‐0.037 0.055 ‐0.115 0.086 Fam_inc 0.032 0.034 0.032 0.034
Fam_inc 0.001 0.036 0.054 0.060 Single 0.040 0.201 0.040 0.201
Single ‐0.334 0.221 ‐0.068 0.324 Cohort 1997 0.014 0.052 0.014 0.052
Social Science 0.233 *** 0.046 0.441 *** 0.062 Cohort 1998 ‐0.058 0.053 ‐0.058 0.053
Program 1.340 *** 0.052 1.476 *** 0.088 Cohort 1999 ‐0.105 ** 0.052 ‐0.105 ** 0.052
Transfers ‐0.462 *** 0.042 ‐0.535 *** 0.067 Live near Old 0.654 *** 0.050 0.654 *** 0.050
License 0.572 *** 0.071 0.696 *** 0.114 Live near New ‐0.661 *** 0.043 ‐0.661 *** 0.043
Cohort 1997 0.081 0.053 0.103 0.084 Constant ‐1.651 *** 0.222 ‐1.651 *** 0.222
Cohort 1998 0.127 ** 0.053 0.046 0.083
Cohort 1999 0.098 * 0.053 0.016 0.082
Interaction with OLD: Age 0.001 0.013
Woman ‐0.018 0.083
Swedish 0.294 0189
Child 0.186 0.239
Married ‐0.156 0.228
GPA 0.011 0.015
‐Nature Science 0.010 0.104
‐Social Science ‐0.167 0.116
Father Upper Sec. 0.020 0.092
Father UNI ‐0.048 0.104
Mother Upper Sec. 0.119 0.101
Mother UNI 0.147 0.111
Fam_inc ‐0.096 0.078
Single ‐0.604 0.453
Social Science ‐0.429 *** 0.080
Program ‐0.262 ** 0.110
Transfers 0.093 0.087
License ‐0.324 ** 0.147
Cohort 1997 ‐0.029 0.108
Cohort 1998 0.145 0.109
Cohort 1999 0.149 0.108
Constant ‐2.251 *** 0.240 ‐2.111 *** 0.374
Roh 0.051 0.064 0.055 0.065
Log likelihood
(Base Model) ( Interacted Model)
‐6160.964 ‐6134.9958
(Base Model) ( Interacted Model)
Note: *, **, *** indicates a significance level at 10, 5, and 1 percent, respectively.
32
Table: B4: Recursive bivariate probit model – Educational outcome: P120
Y1(P120) Coef. Std. Err. Coef. Std. Err. Y2(OLD) Coef. Std. Err. Coef. Std. Err.
OLD 0.055 0.098 0.161 0.485 Age 0.023 *** 0.006 0.023 *** 0.006
Age ‐0.025 *** 0.006 ‐0.021 ** 0.009 Woman 0.002 0.040 0.002 0.040
Woman 0.042 0.041 0.078 0.064 Swedish ‐0.105 0.089 ‐0.104 0.089
Swedish 0.097 0.090 0.124 0.144 Child ‐0.199 * 0.116 ‐0.199 * 0.117
Child 0.439 *** 0.119 0.382 ** 0.166 Married ‐0.200 * 0.109 ‐0.200 * 0.109
Married ‐0.270 ** 0.113 ‐0.173 0159 GPA 0.083 *** 0.007 0.083 *** 0.007
GPA 0.065 *** 0.007 0.051 *** 0.012 ‐Nature Science 0.419 *** 0.054 0.419 *** 0.054
‐Nature Science 0.274 *** 0.060 0.298 *** 0.087 ‐Social Science 0.154 *** 0.049 0.154 *** 0.049
‐Social Science 0.151 *** 0.052 0.121 * 0.073 Father Upper Sec. ‐0.024 0.044 ‐0.023 0.044
Father Upper Sec 0.004 0.046 0.073 0.067 Father UNI 0.208 *** 0.049 0.209 *** 0.049
Father UNI 0.059 0.051 0.154 * 0.080 Mother Upper Sec. 0.020 0.049 0.020 0.049
Mother Upper Se ‐0.007 0.050 ‐0.003 0.073 Mother UNI 0.208 *** 0.053 0.207 *** 0.053
Mother UNI 0.015 0.055 ‐0.089 0.084 Fam_inc 0.031 0.034 0.031 0.034
Fam_inc ‐0.006 0.036 0.044 0.060 Single 0.040 0.201 0.040 0.201
Single ‐0.117 0.204 ‐0.008 0.309 Cohort 1997 0.013 0.052 0.013 0.052
Social Science 0.358 *** 0.047 0.357 *** 0.061 Cohort 1998 ‐0.057 0.053 ‐0.058 0.053
Program 1.392 *** 0.049 1.546 *** 0.081 Cohort 1999 ‐0.106 ** 0.052 ‐0.106 ** 0.052
Transfers ‐0.671 *** 0.041 ‐0.610 *** 0.065 Live near Old 0.650 *** 0.050 0.650 *** 0.050
License 0.188 *** 0.072 0.120 0.107 Live near New ‐0.665 *** 0.043 ‐0.665 *** 0..043
Cohort 1997 0.068 0.053 0.099 0.083 Constant ‐1.649 *** 0.222 ‐1.651 *** 0.222
Cohort 1998 0.160 *** 0.054 0.116 0.082
Cohort 1999 0.161 *** 0.053 0.185 * 0.081
Interaction with OLD: Age ‐0.006 0.013
Woman 0.064 0.083
Swedish ‐0.043 0.184
Child 0.099 0.237
Married ‐0.209 0.225
GPA 0.024 0.015
‐Nature Science ‐0.042 0.115
‐Social Science 0.047 0.103
Father Upper Sec. ‐0.123 0.092
Father UNI ‐0.149 0.103
Mother Upper Sec. 0.001 0.101
Mother UNI 0.174 0.111
Fam_inc ‐0.090 0.078
Single ‐0.221 0.411
Social Science 0.000 0.081
Program ‐0.232 ** 0.103
Transfers ‐0.109 0.084
License 0.117 0.144
Cohort 1997 ‐0.048 0.107
Cohort 1998 0.080 0.108
Cohort 1999 ‐0.033 0.107
Constant ‐1.775 *** 0.233 ‐1.911 *** 0.361
Roh 0.122 ** 0.064 0.141 ** 0.065
Log likelihood ‐6089.2398 ‐6078.1525
(Base Model) ( Interacted Model) (Base Model) ( Interacted Model)
Note: *, **, *** indicates a significance level at 10, 5, and 1 percent, respectively.
III
Are there sheepskin effects in the return
to higher education in Sweden?
By Susanna Holzer
Abstract In contrast to previous studies on sheepskin effects (diploma effects), this study only focuses on Swedish university students with about the same number of years of education. A sample of 2,363 individuals is extracted from the Swedish Longitudinal INdividual DAta (LINDA). Individual characteristics, family background, ability, university type, and university major are controlled for, and the students are conditioned to have obtained at least 120 credit points (corresponding to three years of full-time study). Traditional OLS-models (log-wage models) are complemented with models based on propensity score matching to minimize possible bias due to self selection. Forcing the empirical material to become as homogeneous as possible, the idea here is to isolate possible sheepskin effects from other impacts that might be caused by pure heterogeneity in data. The results show that for male students, the wage-premium of possessing a degree, i.e. the sheepskin effect, is roughly 5-8 percent. For women, it is about 6-7 percent for those who have obtained 160 credit points or more. For students who attended a more prestigious university in the metropolitan areas in Sweden and majored in the natural sciences, a sheepskin effect of roughly 13 percent for men and 22 percent for women is traced. However, this result did not hold among students who attended a newer university outside the metropolitan areas and/or majored in the social sciences. JEL Classification: I23, J24, J31 Keywords: Returns to education, Sheepskin effects, Propensity Score Matching, Higher Education, Human Capital
Correspondence address: Susanna Holzer, School of Management and Economics, Växjö University, SE-351 95 Växjö, Sweden. Phone: +46 470 70 85 79. E-mail: [email protected]. I am grateful to Mårten Palme, Håkan Locking, Thomas Lindh, Lennart Delander, Harald Niklasson, Mikael Lindahl and seminars participants at Växjö University and the SOLE-meeting in Boston, for helpful comments and suggestions. Financial support by Växjö University, Jan Wallander and Tom Hedelius’ Research Foundation, and the Swedish Research Council is acknowledged . All remaining errors are my own.
2
1 Introduction
Although the positive relationship between educational investments and earnings is one of the
most established relationships in the social sciences, we still argue about what exactly in the
educational investment affects earnings—is it years of schooling, credentials, or perhaps a
mixture of them both? Human capital theorists like Mincer (1974) argued that earnings are
mainly affected by the number of years of education, while other researchers point at the
importance of the acquisition of credentials by means of formal degrees.1 In the latter case, an
accredited worker earns more than its non-accredited counterparts, a phenomenon referred to as a
sheepskin effect; see e.g. Hungerford and Solon (1987), Jaeger and Page (1996), and Flores-
Laguunes and Light (2007).2 Understanding the impact and the importance of credentials as well
as of years of education is important for shaping educational policies. Empirical studies based on
competing views on the role of educational investments for labor market outcomes may help us
better characterize the relationship between such investments and their economic returns.
Once a university student in Sweden has passed and fulfilled a set of course-requirements for a
certain university degree, he or she can apply for a formal degree issued by the university he/she
attended. For the vast majority of degrees, the examination requirements are stipulated by
Swedish law and the formal certificate (the diploma) itself is free of charge for the student, except
for the costs in terms of time and effort spent on preparing the formal application. So, the basic
dilemma faced by all students is to value if the possible revenue of a formal degree on the labor
market exceeds the cost of applying for it. In 2005, less than 50 percent of all students who
entered university in the 1990s had a formal university degree (e.g. a bachelor’s, master’s or a
higher degree), another 20 percent were recorded to have passed university courses
corresponding to at least three years of full time study (the theoretical time length of study for a
bachelor’s degree), yet they had no formal degree. Some of them might not value a degree, and
therefore not worth applying for. A more common explanation, though, is the absence of one or
more of the courses that are required to fulfill the degree requirements. However, in terms of
years of education (or the amount of university credits points), they can be compared to other
students who have obtained their degree.
1 On human capital theory, see seminal work by Becker (1964 [1993]) and Schultz (1961). 2 Also referred to as a diploma effect or a credential effect.
3
Two questions are raised in this study: First, is there a general difference in the economic
outcome for former university students with a degree, as compared to those with no formal
degree, i.e. is there a sheepskin effect? Second, does the sheepskin effect vary within groups of
university types, subject majors, and educational programs? Altonji (1993), Altonji et al. (2005),
Arcidiacono (2004), Brewer et al. (1999), and Dale and Krueger (2002) all point out that
information on school quality and educational programs, i.e. on university choice and university
majors, is of importance when the objective is to value the returns to university education.
Adding more information should make us divide students into various groups that, in turn, will
make the students within each group more homogeneous, which should make it possible to
isolate sheepskin effects on the labor market from other heterogeneity in the data. We also
control for whether a student attended an old (and often regarded as more prestigious) university,
if he/ she majored in mathematics, humanities, social sciences etc., and if the student was
enrolled in a vocational educational program such as a teacher training program, a program in
engineering, or a program in economics – in order to see how possible sheepskin effects may
vary depending on student group.
In contrast to most previous studies on sheepskin effects on the labor market, this study only
focuses on university students with about the same number of years of education. In addition to
university choice and within-university choices, individual and family characteristics and ability
are also considered. A random sample of 2,363 individuals was extracted from the Swedish
Longitudinal INdividual DAta (LINDA). The issue of whether the labor market values students’
educational performances differently if they have a formal degree is evaluated by controlling for
individual characteristics, family background, prior educational background, and by conditioning
the students in the sample to have obtained at least 120 credit points (corresponding to three years
of full-time study). Traditional OLS-models are employed and are complemented with models
based on propensity score matching.
The results show that men face a wage-premium of possessing a degree, i.e. the sheepskin effect,
of roughly 5-8 percent. For women, it is about 6-7 percent for those who have obtained 160 credit
points or more. For students who attended a more prestigious old university in the metropolitan
areas in Sweden, and majored in the natural sciences, a sheepskin effect of roughly 13 percent for
4
men and 20 percent for women is traced. However, this result did not hold among students who
attended a newer university outside the metropolitan areas. Controlling for specific occupational
programs for economists, engineers and teachers did not, regardless of gender, give any
significant estimates of sheepskin effects.
The paper is organized as follows: Section 2 gives a description of how higher educational
performances are measured in Sweden. Section 3 briefly outlines previous research. Section 4
describes the data and measurements. Section 5 presents the econometric strategy that is
employed. Section 6 presents the estimated results and Section 7 concludes the paper.
2 Measurement of higher educational performances in Sweden The most common forms of higher education offered to the students at Swedish universities are
program based and normally result in students being eligible for a formal university degree (e.g. a
bachelor’s or master’s degree, a nursing diploma etc.). Attending the university is free of charge
for the student. The students may freely choose university and education, but they must compete
for the student slots available in their desired education, since these are limited by the
government.3 The universities may set certain requirements, but they are not allowed to choose
among eligible students. For the vast majority of students, the Grade Point Average (GPA) from
upper secondary school is the only characteristic with which they compete.
As a consequence of a structural reform of the sector of higher education in Sweden in 1993, a
majority of all universities jointly introduced a system for documentation of the academic
performances by all their students – called Ladok.4 Educational performances are measured in
3 The sector of higher education in Sweden is to a very large extent centrally monitored by the Swedish government and it is supply-side oriented. Since no universities are allowed to charge any students a university fee, and alternative income sources are rare, all universities are dependent on public funding to finance their educational production. The government, in turn, sets stipulations of performance funding, performance budgeting, and outcome assessment on each university, and thereby steers the entire production of higher education in Sweden. 4 An ADB-based system for documentation of university studies. At the time of the introduction of Ladok, 21 of 40 universities were connected to Ladok. By the end of the 1990s, that number had grown to 36. This means that it has become the government’s largest source of information on higher educational production. Colleges that are not connected to Ladok are, mostly, universities that specialize in art, drama, and music, but basically all universities that offer more general educations in social sciences, natural sciences, technology, humanities, medicine, etc. are connected to Ladok (see www.ladok.se).
5
credit points, where one week of successful full-time study by a student corresponds to one
credit point (given that the student has passed the required examinations and fulfilled all
other requirements). A student who studies full time is assumed to obtain 40 credit points in
one year.
At least 120 credit points (three years of full-time study) are usually required to qualify for
an academic degree - which is the lowest requirement to qualify for a bachelor’s degree.5
Table 2.1 shows that a good 15 percent have obtained at least 120 credit points within three
years, about 55 percent within five years, and about 60 percent within seven years. After
seven years, only marginal changes in educational performances occur.6
Table 2.1 Student performances for the academic years 1993/34 to 2003/04
Academic College Earned degrees in percent 120 ≤ credit points in precentB
160 ≤ credit pointsB
YearA
Entrats3 years 5 years 7 years Ladok 3 years 5 years 7 years 7 years
Sweden
1993/94 59,490 10 32 46 47,070 17 52 59 35
1995/96 61,920 8 28 43 54,475 16 51 58 33
1997/98 58,930 6 28 44 54,955 16 54 61 35
1999/00 63,032 6 29 47 61,384 19 58 65 37
2001/02 69,514 5 29 68,054 20 57
2003/04 71,308 6 70,183 17
Note: A) An academic year starts at the end of August and ends in early June the following year. B) New university entrants only include students who were enrolled at a university that was connected to the Ladok‐system at the time of entrance. Source: Statistics Sweden (2007).
Once the student has obtained enough credit points and has fulfilled the requirements for a certain
degree, he/she can apply for a formal degree at the university he/she attended. Table 2.1 shows a
lower frequency of degrees as compared to credit point production; about 6 percent had an
academic degree after three years, 29 percent after five years, and 45 percent after seven
years.7 This leaves roughly 20 percent of the students without a formal degree; they have,
5 There are more requirements, but they will not be further discussed here. See www.hsv.se for more information. 6 A possible explanation for the lower activities after seven years is that students are only allowed student grants and student loans for twelve semesters at most, i.e. six years. 7 In Sweden, a student must apply for a degree from the university that he or she attends, and will obtain it if he or she fulfills the requirements set by the government and, to some extent, the university, i.e. he/she has obtained the required combination of courses. This procedure in itself seems to undermine the incentives by the students to apply for a degree. As long as the degree is not equivalent to a license, e.g. a degree from a nursing school which is required for a nurse to be allowed to practice in the desired profession, employers seem to accept university transcripts of passed courses as proof of competence. This makes the Swedish official registers over the educational
6
however, obtained 120 credit points or more.8 Possible explanations for this could, of course,
be that some of these students lack one or more of the required courses to fulfill the degree
requirements, or they may simply not value a formal degree sufficiently high to consider it
worthwhile applying for.
3 Previous research in brief As mentioned above, the most commonly used application of human capital theory is the one
associated with the work of Mincer (1974). He argues that accreditations have no impact on labor
market earnings. In brief, human capital theories argue that an investment in education adds to an
individual’s productivity and therefore increases the labor market value of his labor, reflected in
labor market earnings. Screening theories, like the theory of the sheepskin effect, argue that the
returns of an educational investment are due to specific educational credentials rather than
accumulated years of education. In accordance with the strict screening theory, years of education
only work as a filter/signal to sort more able students from less gifted ones. A degree works as an
additional signal of a student’s potential and endowments with characteristics that are desired by
the labor market, which is why we should observe higher labor earnings associated with degree
completion.9
Empirical studies by Hungerford and Solon (1987) found that the wage premium rose for
individuals in the year when they were assumingly awarded a university diploma. They used a
population survey from 1978 of roughly 16,500 individuals and found that the wage increase was
significantly larger between the 15th and 16th year of education (the year of a university diploma
from a four-year college), as compared to the difference between the 14th and 15th year of
education. Their conclusion was that about 10 percent of the additional wage increase in the year
of a diploma was caused by a screening effect by the labor market due to the individual holding a
formal diploma. These findings, in particular the discontinuity at 16 years of education, have
level of the population in Sweden somewhat askew since they only consider highest completed degree or, in some cases, whether the student has obtained 120 credit points. 8 Statistics Sweden does not provide this information divided into genders. 9 See seminal work on screening theories by Arrow (1973), Spence (1973) and Stiglitz (1975). Arrow (1973) argued that going to university could bee seen as a filter. The fact that some students are accepted for an education and others not provide a signal of a person’s traits that might be relevant on the labor market, and a university degree is only assumed to make this signal stronger.
7
been confirmed in studies by Belman and Haywood (1991), Card and Kreuger (1992) and
Heckman, Layane-Farrar and Todd (1996).
Kane and Rouse (1995) analyze the returns to two-year community colleges and four-year
colleges using data that provides information on completed course credits for graduates and
dropouts. They found that course credits at two-year and four-year colleges have similar payoffs
on the labor market. The strongest evidence for potential sheepskin effects was found for
associate’s degrees10 for women (approximately 12 percent) and bachelor’s degrees for men
(approximately 24 percent).
Jaeger and Page (1996) used a population survey from 1992 of roughly 18,700 individuals in
which they controlled for both actual years of education and whether a person had received a
formal degree or not. In addition, Jaeger and Page tried to see if there were any differences in
possible sheepskin effects on the labor market besides the number of years of education if they
controlled for individual characteristics such as gender and race. They could confirm that there
was a sheepskin effect of receiving a bachelor’s degree of roughly 25 percent for white men, 30
percent for black men, 22 percent for white women, and 39 percent for black women. The authors
could only statistically verify that there was a sheepskin effect, the differences between races and
sexes were not statistically significant.
After having obtained employment, we expect the value of a diploma to decrease since the
employer now has more direct information about an individual’s labor market qualities and
abilities as a basis for assessing his/her productivity (mirrored in the wage premium). This is
confirmed by Belman and Heywood (1997) on basis of a comparison between wage incomes for
five different age-cohorts, which showed that there was a decrease in the wage premium of a
formal degree as the individual became older and gained work experience. They concluded that
the sheepskin effect on the labor market is to be seen as a short-term effect.
One of few Swedish attempts at detecting possible sheepskin effects on the Swedish labor market
was made by Antelius and Björklund (2001). For the purpose of analyzing the quality value of
10 An associate’s degree can be obtained in the US after two years of college studies, and is most frequently rewarded at the two-year colleges.
8
using Swedish educational register data when estimating possible economic returns to an
educational investment, they compared the quality of the data in this register data with more
precise information based on the Swedish Level of Living Surveys (SLLS). Whereas the register
data mainly reveals highest completed education, based on highest completed degree, the SLLS
also gave information on partial educational performances. Partial educational performance was
defined as having some years of upper secondary school or university. The authors found that
partial education has a significant impact on wage income and thus, they concluded that
sheepskin effects are a minor/rare occurrence in Sweden.
4 Data and measurement The primary data source is Longitudinal INdividual Data for Sweden (LINDA), which is a
random sample of approximately three percent of the Swedish population. The core data is based
on income tax registers of 1994, and it contains a sample of roughly 300,000 individuals which in
addition was merged with population censuses and other register based data (see Edin and
Fredriksson (2000) for a description). The data is traced back to 1968 and up to 2006. In case the
LINDA subject is a part of a family and as long as he or she belongs to the same household, the
family members are registered as well — which makes it possible to trace intergenerational
relationships in the data. Such relationships in LINDA are controlled for by matching them
against the Swedish intergenerational register.
The vast majority of empirical research on education is based on data on self-reported
educational levels, of which most data only has one measure of educational attainment; years of
completed education or the highest degree obtained. The problem with self-reported data, as
pointed out by Card (1999) and Kane et.al. (1999), is that people with a university degree tend to
report this to a higher extent than individuals who only have a partial university education and, in
some cases, people tend to lie. This might cause traditional OLS estimates to understate the
returns per year of education.
To minimize the problem of self-reported data in this study, detailed information on subjects’
educational history, both prior to university and during a university education, is based on official
public records: the Swedish upper secondary school register (dates back to 1973) provides the
9
GPA of the final grades from upper secondary school,11 and the Swedish Higher Education
Register reveals activities within the higher educational system, such as first-time entrance,
reenrollments, choice of university, choice of subjects, choice of courses, length of courses,
passed courses, if the course was part of an educational program, etc.
The main information in the Swedish Higher Education Register is based on Ladok, which dates
back to the school year 1993/1994. However, the information on where and when a student was
first enrolled goes back to 1977, which makes it possible for older individuals with university
experiences prior to 1977 to appear as ”new” students if they reenrolled in 1977 or later. To avoid
this, the oldest students that will be considered were born in 1959, i.e. those who were 18 years of
age in 1977, which is the age at which students on enter university at the earliest.
Moreover, only universities that became connected to the Ladok system no later than in 1993
are considered, meaning that 21 of today’s 40 universities are accounted for. However, they
roughly hosted 80 percent of the entire student population in 1993.12 These are all
universities that offer general and similar educations in social sciences, natural sciences,
technology, medical care, teacher’s training, etc.
4.1 Sample selection and descriptive statistics The basic sample consists of individuals who were first-time students at a Swedish Ladok
connected university during the years 1994-1996. If there are any sheepskin effects on the
Swedish labor market, the effect is expected to be strongest during the first years after the student
has left university. But to give as many students as possible the chance to fulfill their educational
aspirations and the chance to settle down on the labor market, the cross-section sample is taken
from the year 2006, and consists of 4,025 individuals. 11 Due to a reform of the entire upper secondary school system in Sweden in the mid 1990s, the system and rules for all grades changed as well. However, the Agency for Higher Education and the National Admissions Office to Higher Education (VHS) use a translation key in order to compare students with older and students with newer grades from upper secondary school; a method which is also employed in this study. 12 By restricting the universities to have become connected to Ladok no later than 1993, new establishments are excluded from the sample of universities, and the students’ study time is considered to be equal at all universities. The 21 universities accounted for are: Blekinge Institute of Technology, Chalmers University of Technology, Göteborg University, Halmstad University, Dalarna University, Kalmar University, Karlstad University, Karolinska Institutet, Kristianstad University, Linköping University, Luleå University of Technology, Lund University, Mid Sweden University, Mälardalen University, Stockholm University, The Royal Institute of Technology, Umeå University, Uppsala University, University West, Växjö University and Örebro University. See Holzer (2007) for a longer discussion. Observe that the vast majorities of schools in healthcare and nursing were not incorporated into Ladok until the late 1990s, which is why most of them are not accounted for in this study.
10
Using annual wage instead of hourly wage could, according to Card (1999), cause an
overestimation of the educational impact on wage income. However, according to Antelius and
Björklund (2000), a wage-income level of at least 100,000 SEK in 1991 gave fair and comparable
results to using hourly wage income. The 100,000 SEK were based on three base amounts in
1991, and the comparable value of three base amounts in 2006 is 130,000.13 Furthermore,
regional labor market effects on wage-incomes are deflated, i.e. differences, if any, in average
wage structure caused by local labor markets are removed.14 Conditioning on university entry
years and an annual income of 130,000 SEK in 2006, this leaves a sample of 3,536 individuals.
Due to the fact that men and women tend to have different wage structures on the labor market,
which is reflected in labor market earnings, the individuals in the present study are grouped
according to gender.
Moreover, the sample is restricted in that only students that have obtained at least 120 credit
points, i.e. equivalent to three years of full-time study and the minimum number of credit points
required to qualify for a bachelors degree. Part of the obtained credit points are, furthermore,
controlled for to be at an intermediate and master level. In addition, all students in the sample are
conditioned by having at least one parent in LINDA and having GPA from upper secondary
school, which leaves a final sample of 2,363 individuals. The final sample is presented in Table
4.1, where the data is grouped into gender and whether the individual possessed a degree in 2006.
The sample of students contains slightly more women than men, and 97 percent of the individuals
were born in Sweden. In 2006, the individuals in the sample were aged between 28 and 44 years,
with an average age of 32-33 years. More than one third had a parent with a university education.
The average income of the household in which the students lived at the age of 18 was slightly
higher than that of comparable households in Sweden at that time.
13 Various income cut-off points of 100,000 SEK, 120,000 SEK,140,000 SEK and 160,000 SEK have been controlled for, but 130,000 SEK are used in this paper. If I go any lower, I tend to include part time workers, PhD students and other less representative labor market participants which just add more noise into the data. 14 In accordance with definitions by the EU and Statistics Sweden, I have divided Sweden into eight labor market regions. The average wage in each of these regions is used to calculate a ‘regional-wage-deflator’ that is used in order to clear the wage-data from regional differences. The regions are all described in the Appendix.
11
GPA from upper secondary school takes a value between 0 and 20, where 10 equals pass and
qualification for university. According to Table 4.1, the average GPA of the sample is well above
a pass. There is a higher rate of female students with an educational background in the social
sciences from upper secondary school, whereas there is a higher rate of male students with an
educational background in the natural sciences. Students that are awarded a degree on average
have a slightly higher GPA as compared to those who are not awarded a degree.
Of all students in the final sample, 60 percent of the women and 73 percent of the men had 160
credit points or more in 2006, and 81 percent of the women and 72 percent of the men had a
university degree. The educational patterns from upper secondary school are followed by similar
choices of university majors: more women are majoring in the social sciences as compared to
men who to a higher degree major in the natural sciences. A higher share of women studied to
become teachers as compared to men, and more men than women chose to study to become
engineers.
12
Table 4.1 Descriptive data Women Men
Degree No Degree Degree No Degree
Variables Mean Std. Dev. Mean Std. Dev. Mean Std. Dev. Mean Std. Dev.
Individual and family characteristics:
Age 32.248 3.121 32.372 2.985 32.509 3.097 32.785 3.131
Swedish 0.968 0.175 0.957 0.202 0.972 0.166 0.979 0.143
College fatherA 0.307 0.462 0.306 0.462 0.361 0.481 0.336 0.473
College motherA 0.373 0.484 0.438 0.497 0.448 0.498 0.356 0.480
Single parent A 0.243 0.429 0.240 0.428 0.187 0.390 0.221 0.416
Relative income of the household A 1.023 0.391 1.074 1.199 1.065 0.552 1.050 0.373
Upper Secondary School:
Grade Point Average 15.923 2.224 15.280 2.227 15.389 2.481 14.517 2.488
Natural SciencesB 0.242 0.428 0.225 0.418 0.568 0.496 0.457 0.499
Social SciencesB 0.599 0.490 0.671 0.471 0.278 0.448 0.370 0.484
University:
Metropolitan university 0.478 0.500 0.477 0.500 0.529 0.500 0.498 0.501
Educational achievements:
Average credit points 179.983 47.195 173.493 39.697 189.697 52.298 166.602 34.887
120 ≤ credit points 1.000 0.000 1.000 0.000 1.000 0.000 1.000 0.000
160 ≤ credit points 0.604 0.489 0.566 0.497 0.734 0.442 0.547 0.499
Majors and programs: Natural SciencesC 0.228 0.426 0.217 0.413 0.585 0.493 0.574 0.495
Social Sciences C 0.695 0.491 0.767 0.423 0.379 0.485 0.351 0.478 Teacher's program 0.293 0.455 0.140 0.347 0.101 0.301 0.073 0.260 Economist program 0.100 0.300 0.105 0.307 0.089 0.284 0.104 0.306 Engineer program 0.076 0.265 0.054 0.227 0.286 0.452 0.228 0.421
Economic outcome:
Annual Wage 2006 D 277,994 93,811 267,796 93,543 360,839 12,2997 341,022 131,863
Number of observations: 1071 258 745 289
Note: All students have obtained at least 120 credit points, i.e. three years of full time study. A) The family background variables relate to conditions when the students were 18 years of age. B) Omitted are the vocational oriented tracks that are not to be classified into the two presented tracks in natural sciences and social sciences. C) Omitted are university majors that are vocational oriented and not to be classified as natural sciences and social sciences. D) Wage in Swedish kronor (SEK), where 8 SEK ≈ 1 US Dollar (March 25, 2009). All variables are described in the Appendix.
13
5 Econometric strategy Two probit models will be used in the following, first a log-wage model in the form of a
traditional Mincer equation and then a binary model where propensity score matching methods
will be employed.
According to the Mincer equation:
(1) uDXEXPEXPSLnW 542
3210 ,
where W stands for wages, S for years of schooling, EXP for years of labor market experience,
and X is a vector of additional individual and other covariates that may influence earnings, such
as age, gender, nationality, place of residence, etc. Hungerford and Solon (1987) and most of
their followers changed the numerical variable S into a vector of year dummies in order to isolate
possible sheepskin effects between years of schooling. Jaeger and Page (1996) followed this
strategy, but added an extra dummy variable, D, into the equation that indicated whether the
individual had formal educational credentials. is a parameter vector that measures the marginal
effect of each variable on the logarithm of a wage. The models were regressed with a traditional
OLS.
A drawback in comparing a treatment group with a non-experimental comparison group
could be that the results are biased due to self-selection (see Dehejia and Wahba (1999,
2002)). For instance, is the group of individuals with a degree self-selected so that these students
possess different traits as compared to their non-accredited counterparts? A possible solution to
minimize this possible bias is, according to Rosenbaum and Rubin (1983), to use a propensity
score matching method in which the first step is to estimate each individual’s probability of being
awarded a degree, i.e. each individual’s propensity score (p(X)). This is preferably done with a
binary model, e.g. a binomial probit model. The effect on the labor market of possessing a degree
can then be estimated by comparing the economic outcomes between individuals who have a
degree with those lacking one, given that they have similar propensity scores. (In this section,
individuals with a degree will henceforth be referred to as treated, and their counterparts with no
degree as untreated.)
14
The propensity score matching method relies on the assumption of selection on observables, i.e.
that treatment participation and treatment outcome are independent conditional on observable
characteristics of students (see Heckman and Robb (1985). But for the matching to hold, it is
necessary to condition on the support common to both treated and untreated (see e.g., Heckman
et al. (1997, 1998) and Dehejia and Wahba (1999, 2002)). The common support assumption rests
on matches being made when individuals are similar on a certain amount of comparison attributes
(e.g. age, gender, GPA from upper secondary school), which constitutes the core in minimizing
the possible impact of the problem of self-selection on potential sheepskin effects. This
assumption forces the characteristics and the propensity scores of the treated and the untreated to
be as equal as possible. Furthermore, when using propensity score matching, only matches are
used, i.e. observations that lack common support are excluded.
It is important to note that when using matching methods, all empirical information must either
be time invariant or conditioned to hold prior to the treatment. Unlike the Mincer equation above,
no information after the fact that the students were awarded a degree or left school is added, e.g.
labor market experience is excluded. For the OLS estimates to be comparable with the results
based on matching, the covariate conditions that hold when using the matching methods will be
used when employing the OLS. Since all subjects are university students with a similar number
of years, years of education are excluded from the log-wage model as well. However, each
individual credit point obtained will be included.
In the ideal case, we would find a one-to-one match, but that often requires a very large (infinite)
sample size. Instead, we could match each person in the treatment group with the person among
the untreated whose propensity score is closest to that of the treatment group observation—
nearest neighbor matching. Considering that in the current case we have a small sample and that
the untreated are fewer than the treated, it would be necessary to implement a method that allows
the same comparison observation to be repeatedly used. Using replacement could, however,
result in large standard errors (since few of the observations of the untreated may be heavily
used). According to Monte Carlo simulation studies by Frölich (2004) and a recommendation by
Black and Smith (2004), density matching, using Epanechnikov kernel matching, performs much
better and gives more robust matching estimates than nearest neighbor matching, which is why
the former will be used in this study. Kernel density matching also allows for conditioning on
15
covariates, yet we set bandwidths on how far apart a match is allowed to be in terms of its
propensity score, and still be considered a match.
Standard errors are obtained by bootstrapping. Here I allow 1000 iterations.
6 Results The results from employing OLS and propensity score matching are presented in Table 6.1 for
men and in Table 6.2 for women.15 The results are all conditioned on individual characteristics,
family background, GPA from upper secondary school, if the student has an upper secondary
education in the natural sciences or the social sciences, and if the student attended a metropolitan
university or not, studied in the field of natural sciences or social sciences and how many credit
points the student has obtained.
In the first panel of Table 6.1, the male students are conditioned to have obtained at least 120 or
160 credit points. In the 120 credit-point case, all students have just enough credit points to
qualify for a bachelor’s degree, and the estimations do show some indications that may be
interpreted as a sheepskin effect on the labor market. Here, an average diploma effect according
to the OLS results is roughly five percent, and according to the results obtained by propensity
score matching, the average treatment effect (ATE) of possessing a degree is roughly eight
percent. Increasing the restrictions to having obtained at least 160 credit points increases the
average diploma effect to eight percent, an effect which is not found in the results obtained by
propensity score matching.
In the second panel of Table 6.1, all male students have obtained at least 120 credit points and are
now sampled into groups based on university choice, university majors and a combination of
these two aspects. Among students that attended a metropolitan university, the estimates indicate
that having a degree increases the labor market earnings by almost 11 percent according to the
OLS, and by 15 percent according to the matched result. Among students who have attended less
prestigious universities outside the metropolitan areas in Sweden, no sheepskin effects are traced.
15 All matching estimates are obtained by using PSMATCH2 for STATA, by Leuven and Sianesi (2003).
16
Only among students who majored in the natural sciences does the OLS results indicate a
diploma effect of six percent. These are results that cannot be traced in the social sciences or in
any of the results obtained by propensity score matching for both majors.
Conditional on both university choice and university major, the only combination in which any
sheepskin effects are found is among male students who attended a metropolitan university and
majored in the natural sciences, where the labor market value of a formal degree manifests itself
in an increase in labor market earnings by 14 percent according to the OLS, and by 15 percent
according to the results obtained by propensity score matching.16 No degree effects are found for
students studying at universities outside the metropolitan areas and/or majoring in the social
sciences or the natural sciences.
According to the results for women shown in Table 6.2, the only significant sheepskin effects are
found among students who have obtained 160 credit points or more. The OLS results indicate an
average diploma effect of six percent and the matched result of roughly seven percent. When
controlling for choice of university type and university major in the second panel of the table, no
conclusive and significant indications of sheepskin effects are obtained. One exception, however,
is that majoring in and having a degree in the natural sciences seems to be more highly valued on
the labor market, due to the fact that a degree gives a reward of roughly eight percent according
to the OLS and 13 percent additional wages according to the results obtained by propensity score
matching.
Sampling female students on both university choice and university major gives some additional
indications of sheepskin effects. Among students that majored in the natural sciences and
attended a metropolitan university, the possession of a degree gives an average value added on
the labor market of 19-23 percent.17 No such effect can be traced among any other combinations
of university type and university majors.
16 Similar results are to be traced if we narrow down the sample to only include students who have obtained 160 credit points or more. The results also hold if we exclude students in medicine. 17 The results also hold if we exclude students in medicine.
17
Comparing the results for men and women, it seems most likely that students studying the natural
sciences at metropolitan universities are those driving the positive results obtained when only
controlling for the amount of credit points in the first panels of both tables.
For both men and women, three specific educational programs that are associated with a certain
occupation group are singled out; economists, engineers, and teachers. The matched average
effect of possessing a degree is all negative for men. Excluding engineers, this also holds for
women. However, note that these results are based on extremely small samples, which is why any
interpretations should be made with some caution and since no results are significant, no
conclusions on possible sheepskin effects can be drawn from these results.
The results obtained by propensity score matching are overall somewhat larger than the OLS
results, but at the price of weaker precisions.
Table 6.1
Propensity score estim
ates of the effect of possessing a university degree on the Swedish labor market – the case of men
Credit Points
Conditioning on college choice, college m
ajor, and a combination of these
BOccupationsB
120 ≤
160 ≤
Metro.
Country
Nature
CSocial
CMet+N
atC
Met+SocC
Cou+N
atC
Cou+SocC
Engineer
Teacher
Economist
ATE
0.081**
0.088
0.148**
0.021
0.064
0.056
0.154**
0.120
‐0.049
‐0.005
‐0.013
‐0.0547
‐0.064
St. D
ev.
0.042
0.056
0.061
0.042
0.048
0.054
0.080
0.093
0.092
0.049
0.075
0.053
0.109
Bandwidth
0.0006
0.001
0.002
0.002
0.001
0.002
0.001
0.004
0.0015
0.02
0.0025
0.1
0.007
OLS
0.047**
0.080***
0.107***
‐0.002
0.064**
0.052
0.135***
0.118***
0.014
‐0.011
‐0.020
‐0.0464
‐0.105
Std. D
ev
0.022
0.029
0.034
0.029
0.031
0.033
0.048
0.051
0.040
0.044
0.039
0.0406
0.0674
Pseudo R2 before
0.079
0.095
0.132
0.048
0.116
0.062
0.153
0.128
0.081
0.030
0.113
0.071
0.105
Pseudo R2 after
0.026
0.055
0.054
0.041
0.082
0.026
0.053
0.035
0.055
0.037
0.131
0.015
0.080
Treated on support
321
238
206
223
320
183
215
57
65
142
83
72
58
Untreated on support
223
119
106
132
115
137
58
67
38
79
40
19
30
Treated off support
424
309
188
128
116
99
28
69
127
4130
18
Untreated off support
66
39
38
13
829
129
27
126
20
Sample
1,034
705
538
496
559
448
302
222
257
226
279
94
96
Note: *, *, and ***
indicate a significance level o
f 10, 5, and 1 percent. Bandwidths are selected
after using a minim
um root mean squared criterion, w
hich im
plies
that the differences in covariates between treated and control group used in the matching procedure are non‐significantly different from zero. Bootstrap
ped
estimates for matching estimates are based
on 1,000
replications, and standard errors are presented ben
eath the ATE results. Robust standard errors from the OLS
estimates are presented ben
eath. A) Covariates: age, born Swed
ish, university father, university m
other, household relative income, upper secondary school: GPA,
mathem
atics or social scien
ce directions, atten
ding a metr opol ita n
uni versit y. B) All studen
ts have obtained at least 120 credit point s. Excluded by the covariate s
are the metr opolita n
uni ver sity indi cator and university m
a jor s. C) Excluded here ar e studen
ts that are m
a joring in fields that could not be class ifie d
int o either the
social or the na tural scie n
ces: 14 studen
ts at the metropol itan
universities (Metro) and 13 student s at the universities outside the metr opolitan
(Country) areas in
Swed
en.
Table 6.2
Propensity score estim
ates of the effect of possessing a university degree on the Swedish labor market – the case of women
Credit Points
OccupationsB
120 ≤
160
≤Metro.
Country
Nature
CSocial
CMet+N
atC
Met+SocC
Cou+N
atC
Cou+SocC
Engineer
Teacher
Economist
ATE
0.029
0.072**
0.029
0.037
0.127*
0.022
0.225**
‐0.008
0.040
0.034
0.058
‐0.038
‐0.014
St. D
ev.
0.027
0.035
0.049
0.034
0.058
0.030
0.113
0.045
0.078
0.033
0.206
0.056
0.114
Bandwidth
0.001
0.0008
0.001
0.003
0.005
0.0025
0.01
0.02
0.04
0.2
0.1
0.0025
0.01
OLS
0.030
0.057**
0.031
0.025
0.086*
0.012
0.190***
‐0.014
‐0.008
0.029
0.011
‐0.007
0.1011
Std. D
ev
0.021
0.029
0.032
0.029
0.050
0.024
0.072
0.036
0.076
0.033
0.102
0.041
0.0771
Pseudo R2 before
0.028
0.051
0.042
0.038
0.163
0.022
0.220
0.028
0.298
0.030
0.072
0.056
0.212
Pseudo R2 after
0.014
0.035
0.030
0.030
0.093
0.019
0.104
0.025
0.110
0.010
0.062
0.068
0.137
Treated on support
801
510
265
501
174
683
95
332
79
409
81171
46
Untreated on support
242
141
114
126
46
195
23
91
20
106
934
20
Treated off support
270
153
247
58
70
61
46
318
0143
18
56
Untreated off support
16
59
910
312
07
12
09
Sample
1,329
809
635
694
300
942
176
426
124
516
95350
131
Conditioning on college choice, college m
ajor, and a combination of these
B
Note: *, *, and ***
indicate a significance level o
f 10, 5, and 1 percent. Bandwidths are selected
after using a minim
um root mean squared criterion, w
hich im
plies
that the differences in covariates between treated and control group used in the matching procedure are non‐significantly different from zero. Bootstrap
ped
estimates for matching estimates are based
on 1,000
replications, and standard errors are presented ben
eath the ATE results. Robust standard errors from the OLS
estimates are presented ben
eath. A) Covariates: age, born Swed
ish, university father, university m
other, household relative income, upper secondary school: GPA,
mathem
atics or social scien
ce directions, atten
ding a metropolitan
university. B) All studen
ts have obtained at least 120 credit points. Excluded by the covariates
are the metropolitan
university indicator and university m
ajors. C) Excluded here are studen
ts m
ajoring in fields that could not be classified
in the social or the
natural scien
ces: 14 students at metropolitan
universities and 13 students at universities outside the metropolitan
areas in Swed
en.
7 Conclusions
In contrast to most studies on sheepskin effects, the focus of this study is only on university
students that have invested in a similar number of years and fields of education. The idea is to
make the empirical sample as homogenous as possible in order to isolate possible sheepskin
effects of having obtained a formal university degree from other heterogeneities in the data. For
male students with 120 credit points or more (corresponding to three years of full-time study), the
wage-premium of possessing a degree, i.e. the sheepskin effect, is roughly 5-8 percent. For
women who have obtained 160 credit points or more, it is about 6-7 percent. These results on
Swedish students on the Swedish labor market are comparable to the US findings on US data
made by Hungerford and Solon (1987), Belman and Haywood (1991), Card and Kreuger (1992),
Heckman, Layane-Farrar and Todd (1996) and Kane and Rouse (1995).
When controlling for university type and university majors, we found that students (both genders)
who attended a more prestigious university in the metropolitan areas in Sweden and majored in
the natural sciences gained a sheepskin effect of roughly 13-22 percent. These results are most
likely driving the overall results found when only controlling for obtained credit points. No
diploma effects were traced for students who attended a newer university outside the
metropolitan areas, or who majored in the social sciences or for a combination of them both.
Controlling for specific occupational programs for economists, engineers and teachers did not,
regardless of gender, give any significant estimates of sheepskin effects. However, these results
are based on extremely small data samples and thus, alternative outcomes in future studies on
sheepskin effects cannot be ruled out.
21
References
Altonji, Joseph G. (1993), The Demand for and Return to Education When Education Outcomes are Uncertain, Journal of Labor Economics, 11(1):48-83.
Altonji, Joseph G., Todd E. Elder and Christopher R. Taber (2005), Selection on Observed and unobserved Variables: Assessing the Effectiveness of Catholic Schools, Journal of Political Economy, 113(1):151-184.
Antlius, Jesper and Anders Björklund (2000), How Reliable are Register Data for Studies of the Return on Schooling? An examination of Swedish data, Scandinavian Journal of Educational Research, 44(4):341-355.
Arcidiacono, Peter (2004), Ability sorting and the returns to university major, Journal of Econometrics, 121(1-2):343-375.
Arrow, Kenneth (1973), Higher education as a filter, Journal of Public Economics, 2:193-216. Becker, Garry. S. (1964 [1993]), Human Capital, 3 ed, Chicago: University of Chicago Press. Belman, Dale and John S. Heywood (1991), Sheepskin effect in the Return to Education: An
Examination of Women and Minorities, The Review of Economic and Statistics, 73:720-724.
Belman, Dale, and John S. Heywood (1997), Sheepskin Effects by Cohort: Implications of Job Matching in a Signaling Model, Oxford Economic Papers, New Series, 49(4):623-637.
Björklund, Anders, Mårten Palme and Ingemar Svensson (1995), Tax Reforms and Income Distribution: An Assessment Using Different Income Concepts, Swedish Economic Policy Review, 2: 229–266.
Black, Dan and Jeffery Smith (2004), How Robust is the Evidence on the effect of University Quality? Evidence from Matching, Journal of Econometrics, 121(1-2):99-124.
Brewer, Dominic J and Ronald G. Ehrenberg (1996), Does it pay to attend an élite private university? Evidence from the senior class of 1980, Research in Labor Economics, 15:239-72.
Brewer, Dominic J., Eric Eide, and Ronald G. Ehrenberg (1999), Does It Pay To Attend An Elite Private University? Cross Cohort Evidence on the Effects of University Quality on Earnings, Journal of Human Resources, 34(1):104-123.
Dale, Stacy Berg and Allan Krueger (2002), Estimating the Payoff to Attending a More Selective University: An Application of Selection on Observables and Unobservables, Quarterly Journal of Economics, 117(4):1491-1527.
Card, David. (1999), The casual effect of education on earnings, in Handbook in Labor Economics Vol. 3A, (red) Orley C. Ashenfelter and David Card, Amsterdam: North-Holland: Elsevier Science Publishers.
Dehejia, Rajeev, and Sadek Wahba, (2002), Propensity score matching methods for nonexperimental causal studies, Review of Economics and Statistics, 84:1, pp 151–161.
Dehejia, Rajeev, and Sadek Wahba (1999), Causal effects in nonexperimental studies: reevaluating the evaluation of training programs, Journal of the American Statistical Association 94(448):1053–1062.
Edin, Per-Anders and Peter Fredriksson (2000), LINDA - Longitudinal INdividual DAta for Sweden.Working Paper 2000:19, Uppsala, Sweden: Department of Economics, Uppsala University.
Flores-Lagunes Alfonso and Audrey Light (2007), Interpreting Sheepskin Effects in the Returns to Education, Econ Working paper 0707, Department of Economics, University of Arizona
Frölich, Markus (2004), Finite sample properties of propensity score matching and weighting estimators, Review of Economics and Statistics, 86:(1): 77–90.
22
Heckman, James and Richard Robb (1985), Alternative Methods for Evaluating the Impact of Interventions, in J. Heckman and B.Singer (Eds.), Longitudinal Analysis of Labor Market Data, Econometric Society Monograph, No. 10 (Cambridge, UK: Cambridge University Press, 1985: 63–113.
Heckman, James Heckman, Anne Layne-Farrar and Petra Todd (1996), Human Capital Pricing Equations with an Application to Estimating the Effect of Schooling Quality on Earnings, The Review of Economics and Statistics, 78(4), 562-610.
Heckman, James, Hidehiko Ichimura, and Petra Todd (1997), Matching as an Econometric Evaluation Estimator: Evidence from Evaluating a Job Training Program, Review of Economic
Studies, 64(4): 605–654. Heckman, James, Hidehiko Ichimura and Petra Todd (1998), Matching as an Econometric
Evaluation Estimator, Review of Economic Studies, 65: 261-294. Holzer, Susanna (2007), The Expansion of Higher Education in Sweden and the Issue of Equality
of Opportunity, CAFO working paper series No 5, Växjö University. Hungerford, Thomas and Garry Solon (1987), Sheepskin effects in the returns to education”,
Review of Economics and Statistics, 69: 175–177. Jaeger, David. A. and Marianne E. Page (1996), Degrees Matter: New Evidence on Sheepskin
Effects in the Returns to Education, Review of Economics and Statistics, 78:733-740. Kane, Thomas J. and Cecilia Elena Rouse (1995), Labor Market Return to Two- and Four Year
College, American Economic Review, 85(3):600–14. Kane, Thomas J., Cecilia Elena Rouse and Douglas Staiger (1999), Estimating Returns to
Schooling when Schooling is Misreported, NBER Workin Paper, No. 7235. Mincer, Jacob (1974), Schooling, Experience, and Earnings, NY Columbia University Press. Rosenbaum, P. and D. Rubin, 1983, The central role of the propensity score in observational
studies for causal effects, Biometrica, 70:41-55. Schultz, Theodore W. (1961), Investment in Human Capital, The American Economic Review,
51(1):1-17. Spence, Michael, (1973), Job market signaling, Quarterly Journal of Economics, 87:355-374. Stiglitz, Joseph E (1975), The Theory of ‘Screening,’ Education, and the Distribution of Income,
American Economic Review, 65: 283-300.
23
APPENDIX
A1. Variable description
Table A1.1 Variable description
Variable name Description
Age Age in 2006Age Squared Age in 2006, squaredSwedish 1 if born in Sweden, 0 otherwise College parents 1 if the father has a college degree, 0 otherwise
Relative income A relative income for the households when the student was 18 years old
Upper Secondary School: Grade Point Average clusters ‐11.99; 12‐13.99: 14‐15.99: 13‐17.99; 18‐20‐ Mathematics 1 if the direction was towards the natural sciences, 0 otherwise‐ Social Sciences 1 if the direction was towards the social sciences, 0 otherwise
University type:
Metropolitan college B 1 if Chalmers University of Technology, Göteborg University, Karolinska institutet, Lund University, Uppsala University, Stockholm University, The Royal Institute of Technology, 0 otherwise
Natural Sceinces C 1 if the direction was towards the natural sciences, 0 otherwise
Social Sciences C 1 if the direction was towards the social sciences, 0 otherwiseCredit points cluster Clustered as: 120‐139; 140‐159; 160‐179; 180‐199; 200‐229; 230‐160 ≤credit points 1 if the student have achieved 160 credit points or more, 0 otherwise120 ≤credit points 1 if the student have achieved 120 credit points or more, 0 otherwiseTeacher program 1 if the student has 120 credit points or more at the program, 0 otherwiseEconomist program 1 if the student has 120 credit points or more at the program, 0 otherwiseEngineer program 1 if the student has 120 credit points or more at the program, 0 otherwiseDegree 1 if the student has a college degree, 0 otherwise
Economic outcome:
Wage D Labor market wage in 2006
Individual and family characteristics:
Note: A) Family income is presented as the relative net‐income (after tax reduction and received benefits) of the household to which the student belonged at the age of 18.
Z
i
Z
i itit
itit
HousholdFAMincomeFAMincomeincomeFamily
1 1/
_ ,
where itincomeFamily _ is the nominal income of the household of student i at time t. t = (1968,...,1996) indicates the
year in which the student turned 18. The sum of all nominal incomes in year t is divided by all households the same year. In the two‐parent household case, the nominal income has been divided by 1.7 in order to compare the household income by a one‐parent with the ones by a two‐parent (see Björklund, Palme, and Svensson (1995)). B) These are all universities situated in the metropolitan areas in Sweden, they are somewhat older and normally looked upon as more prestigious universities. C) The social sciences include: Humanities, Social Sciences, Economics, History and Law. The natural sciences include Technology and Medicine. Observe that the caring profession is excluded in both groups, yet it is part of the total student sample. D) Annual wage is “regional deflated”, meaning that the direct impact on regional labor market impact is deflated away. According to Statistic Sweden and the EU, Sweden is divided into eight labor market regions (NUTS‐2 regions): Stockholm, östra Mellansverige, Småland med öarna, Sydsverige, Västsverige, norra Mellansverige, mellersta Norrland, and övre Norrland.
1
Svensk sammanfattning Föreliggande sammanläggningsavhandling består av tre fristående artiklar röran-de svensk högskolepolitik, huvudsakligen under 1990-talet och början av 2000-talet. Avhandlingens empiriska delar bygger på statistiskt material från Statistis-ka Centralbyrån. De representativa urval av individer som används i samtliga studier har hämtats ur databasen LINDA. I avhandlingens två första artiklar delas de svenska högskolorna in i gamla respektive nya högskolor beroende på om de var etablerade som självständiga högskolor före respektive efter högskolerefor-men 1977. Artikel [I] analyserar hur den kraftiga och snabba expansionen av den svenska högskolan under 1990-talet påverkade ungdomars antagningsbeteenden, med be-toning på deras socioekonomiska bakgrund. Expansionen skedde vid samtliga högskolor i Sverige, men var särskilt markant vid de yngre lärosätena utanför storstadsregionerna – där några av dem ökade sitt studentintag med upp till 400 procent på tio år. De empiriska resultaten antyder att den ökade tillgången till högre studier i den lokala och regionala närmiljön har ökat sannolikheten för unga individer att studera - och då inte bara i närheten av individens bostadsort. Den ökade geografiska tillgängligheten verkar ha minskat det ”sociala avståndet” till högre utbildning. Mer precist betyder det att det har blivit mer allmänt accep-terat bland fler socioekonomiska grupper (även icke-akademiska) att betrakta högre utbildning som ett minst lika självklart val efter gymnasiala studier, som att inträda på arbetsmarknaden. Den relativt största ökningen av tillströmmande studenter har skett bland ungdomar från icke-akademiska hem, mer precist; gruppen ungdomar vars föräldrar har högst gymnasial utbildning. Artikel [II] i avhandlingen analyserar högskolevalets betydelse för individers chanser att klara sina studier. I undersökningen används två utfallsvariabler: (i) huruvida studenten har tagit en examen (motsvarande kandidatexamen eller hög-re) inom sju år efter högskoleinträdet eller ej; (ii) huruvida studenten uppnått minst 120 högskolepoäng eller mer (minsta poängkrav för att erhålla kandidatex-amen) inom sju år efter högskoleinträdet eller ej. Genom en binom probit-modell – där individ- och familjebakgrundsspecifika variabler kontrollerats, samt ge-nomsnittligt avgångsbetyg från gymnasiet inkluderats – visar resultaten att studi-er vid en äldre högskola ökar individens chanser att klara sina stunder med 5 procentenheter om utfallet är examen, och med 9 procentenheter om utfallet är 120 poäng eller mer. Med en utökad bivariat probit-modell – där vi tar hänsyn till att det kan förekomma en selektion in till respektive högskolekategori på grund av individers ickeobserverbara egenskaper – visar resultaten dock att det
2
inte kan uteslutas att skillnader i prestation inte kan bindas till respektive hög-skolekategori. Artikel [III] i avhandlingen analyseras värdet av en examen (motsvarande minst kandidatexamen) på den svenska arbetsmarkanden. Till skillnad från tidigare studier fokuserar föreliggande studie endast på de studenter som har motsvaran-de minst tre års (heltids) högskolestudier – varav delar på avancerad nivå – där en del tagit en formell examen och andra inte. Resultaten visar att lönepremien av att ha en formell examen är cirka 5-7 procent för män. För kvinnor som har motsvarande minst fyra års heltidsstudier visar resultaten att examenspremien är 6-7 procent. För studenter som studerat vid de äldre lärosätena (Lund, Göteborg, Stockholm och Uppsala) och inom naturvetenskapliga ämnen, visar resultaten att lönepremien av att ha en formell examen är nästan 13 procent för män, respekti-ve 22 procent för kvinnor, jämfört med studenter från samma lärosäten som sak-nar en examen. Motsvarande lönepremie av en formell examen kunde inte esti-meras bland studenter vid andra lärosäten i Sverige, eller bland studenter som läst i huvudsak samhällsvetenskapliga ämnen.
Acta Wexionensia Below please find a listing of publications in the Acta Wexionensia series. For more in-formation, please see www.vxu.se Series III (ISSN 1404-4307). From 2007 and onward. 106. Ann-Charlotte Larsson 2007, Study of Catalyst Deactivation in Three Different In-
dustrial Processes (doktorsavhandling), ISBN: 978-91-7636-533-5. 107. Karl Loxbo, 2007,Bakom socialdemokraternas beslut. En studie av den politiska
förändringens dilemman - från 1950-talets ATP-strid till 1990-talets pensionsuppgörel-se (doktorsavhandling), ISBN: 978-91-7636-535-9.
108. Åsa Nilsson-Skåve, 2007, Den befriade sången. Stina Aronsons berättarkonst (dok-torsavhandling), ISBN: 978-91-7636-536-6.
109. Anne Haglund Morrissey, Daniel Silander (eds.), 2007, The EU and the Outside World - Global Themes in a European Setting, ISBN: 978-91-7636-537-3.
110. Robert Nyqvist, 2007, Algebraic Dynamical Systems, Analytical Results and Nume-rical Simulations (doktorsavhandling), ISBN: 978-91-7636-547-2.
111. Christer Fritzell, Lena Fritzén, 2007, Integrativ didaktik i olika ämnesperspektiv. ISBN: 978-91-7636-548-9.
112. Torgny Klasson, Daniel Silander, 2007. Hot och hotbilder i globaliseringens tid – en studie av den svenska säkerhetspolitiska debatten. ISBN: 978-91-7636-550-2
113. Olof Eriksson (red.), 2007. Översättning och Kultur. Föredrag från ett symposium vid Växjö universitet 17-18 november 2006, ISBN: 978-91-7636-552-6
114. Henrik Tryggeson, 2007. Analytical Vortex Solutions to the Navier-Stokes Equation (doktorsavhandling), ISBN: 978-91-7636-555-7.
115. Sofia Ask, 2007. Vägar till ett akademiskt skriftspråk (doktorsavhandling), ISBN: 978-91-7636-557-1.
116. Cesar Villanueva Rivas, 2007 Representing Cultural Diplomacy: Soft Power, Cos-mopolitan Constructivism and Nation Branding in Mexico and Sweden. (doktorsav-handling), ISBN: 978-91-7636-560-1.
117. Elisabet Frithiof, 2007. Mening, makt och utbildning. Delaktighetens villkor för per-soner med utvecklingsstörning (doktorsavhandling). ISBN: 978-91-7636-554-0.
118. Mats Johansson, 2007. Product Costing for Sawmill Business Management (dok-torsavhandling). ISBN: 978-91-7636-564-9.
119. Rune Svanström, 2007. När väven blir skör och brister – erfarenheter av att leva med demenssjukdom (doktorsavhandling). ISBN: 978-91-7636-565-6
120. Sofia Almerud, 2007. Vigilance & Invisibility. Care in technologically intense envi-ronments (doktorsavhandling). ISBN: 978-91-7636-569-4.
121. Urban Ljungquist, 2007. Core Competence Matters: Preparing for a New Agenda (doktorsavhandling) . ISBN: 978-91-7636-567-0.
122. Jimmy Engren, 2007. Railroading and Labor Migration. Class and Ethnicity in Ex-panding Capitalism in Northern Minnesota, the 1880s to the mid 1920s (doktorsav-handling). ISBN: 978-91-7636-566-3.
123. Susanne Källerwald, 2007. I skuggan av en hotad existens – om den onödiga striden mellan biologi och existens i vården av patienter med malignt lymfom (doktorsavhand-ling). ISBN: 978-91-7636-568-7.
124. Gunilla Härnsten, Britta Wingård, 2007. Högskoleutbildning – Javisst, men med vem och för vad? ISBN: 978-91-7636-570-0.
125. Thérèse Eng, 2007. Traduire l´oral en une ou deux lignes – Étude traductologique du sous-titrage français de films suédois contemporains (doktorsavhandling). ISBN : 978-91-7636-570-0.
126. Andreas Jansson, 2007. Collective Action Among Shareholder Activists (doktorsav-handling). ISBN: 978-91-7636-573-1.
127. Karl-Olof Lindahl, 2007. On the linearization of non-Archimedean holomorphic functions near an indifferent fixed point (doktorsavhandling) ISBN : 978-91-7636-574-8.
128. Annette Årheim, 2007. När realismen blir orealistisk. Litteraturens ”sanna historier” och unga läsares tolkningsstrategier (doktorsavhandling). ISBN: 978-91-7636-571-7.
129. Marcela Ramírez-Pasillas, 2007. Global spaces for local entrepreneurship: Stret-ching clusters through networks and international trade fairs (doktorsavhandling). ISBN: 978-91-7636-577-9.
130. Daniel Ericsson, Pernilla Nilsson, Marja Soila-Wadman (red.), 2007. Tankelyft och bärkraft: Strategisk utveckling inom Polisen. ISBN: 978-91-7636-580-9.
131. Jan Ekberg (red.), Sveriges mottagning av flyktingar – några exempel. Årsbok 2007 från forskningsprofilen Arbetsmarknad, Migration och Etniska relationer (AMER) vid Växjö universitet. ISBN: 978-91-7636-581-6.
132. Birgitta E. Gustafsson, 2008. Att sätta sig själv på spel. Om språk och motspråk i pe-dagogisk praktik (doktorsavhandling). ISBN: 978-91-7636-589-2.
133. Ulrica Hörberg, 2008. Att vårdas eller fostras. Det rättspsykiatriska vårdandet och traditionens grepp (doktorsavhandling). ISBN: 978-91-7636-590-8.
134. Mats Johansson, 2008. Klassformering och klasskonflikt i Södra och Norra Möre hä-rader 1929 – 1931 (licentiatavhandling). ISBN: 978-91-7636-591-5.
135. Djoko Setijono, 2008. The Development of Quality Management toward Customer Value Creation (doktorsavhandling). ISBN : 978-91-7636-592-2.
136. Elisabeth Björk Brämberg, 2008. Att vara invandrare och patient i Sverige. Ett indi-vidorienterat perspektiv (doktorsavhandling). ISBN: 978-91-7636-594-6.
137. Anne Harju, 2008. Barns vardag med knapp ekonomi. En studie om barns erfarenhe-ter och strategier (doktorsavhandling). ISBN: 978-91-7636-595-3.
138. Johan Sjödin, 2008. Strength and Moisture Aspects of Steel-Timber Dowel Joints in Glulam Structures. An Experimental and Numerical Study (doktorsavhandling). ISBN: 978-91-7636-596-0.
139. Inger von Schantz Lundgren, 2008. Det är enklare i teorin… Om skolutveckling i praktiken. En fallstudie av ett skolutvecklingsprojekt i en gymnasieskola (doktorsav-handling). ISBN: 978-91-7636-600-4.
140. Lena Nordgren, 2008. När kroppen sätter gränser – en studie om att leva med hjärt-svikt i medelåldern (doktorsavhandling). ISBN: 978-91-7636-593-9.
141. Mirka Kans, 2008. On the utilisation of information technology for the management of profitable maintenance (doktorsavhandling). ISBN : 978-91-7636-601-1.
143. Christer Fritzell (red.), 2008. Att tolka pedagogikens språk – perspektiv och diskur-ser. ISBN: 978-91-7636-603-5.
144. Ernesto Abalo, Martin Danielsson, 2008. Digitalisering och social exklusion. Om medborgares användning av och attityder till Arbetsförmedlingens digitala tjänster. ISBN: 978-91-7636-608-0.
145. Patrik Wahlberg, 2008. On time-frequency analysis and pseudo-differential opera-tors for vector-valued functions (doktorsavhandling). ISBN: 978-91-7636-612-7.
146. Morgan Ericsson, 2008. Composition and Optimization (doktorsavhandling). ISBN: 978-91-7636-613-4.
147. Jesper Johansson, 2008. ”Så gör vi inte här i Sverige. Vi brukar göra så här.” Retorik och praktik i LO:s invandrarpolitik 1945-1981 (doktorsavhandling). ISBN: 978-91-7636-614-1.
148. Monika Hjeds Löfmark, 2008. Essays on transition (doktorsavhandling). ISBN: 978-91-7636-617-2.
149. Bengt Johannisson, Ewa Gunnarsson, Torbjörn Stjernberg (red.), 2008. Gemensamt kunskapande – den interaktiva forskningens praktik. ISBN: 978-91-7636-621-9.
150. Sara Hultqvist, 2008. Om brukardelaktighet i välfärdssystemen – en kunskapsöver-sikt. ISBN: 978-91-7636-623-3.
151. Jaime Campos Jeria, ICT tools for e-maintenance (doktorsavhandling). ISBN: 978-91-7636-624-0.
152. Johan Hall, Transition-Based Natural Language Parsing with Dependency and Con-stituency Representations (doktorsavhandling). ISBN: 978-91-7636-625-7.
153. Maria Fohlin, L’adverbe dérivé modifieur de l’adjectif. Étude comparée du français et du suédois (doktorsavhandling). ISBN: 978-91-7636-626-4.
154. Tapio Salonen, Ernesto Abalo, Martin Danielsson, 2008. Myndighet frågar medbor-gare. Brukarundersökningar I offentlig verksamhet. ISBN: 978-91-7636-628-8.
155. Ann-Christin Torpsten, 2008. Erbjudet och upplevt lärande i mötet med svenska som andraspråk och svensk skola (doktorsavhandling). ISBN: 978-91-7636-629-5.
156. Guillaume Adenier, 2008. Local Realist Approach and Numerical simulations of Nonclassical Experiments in Quantum Mechanics (doktorsavhandling). ISBN: 978-91-7636-630-8.
157. Jimmy Johansson, 2008. Mechanical processing for improved products made from Swedish hardwood (doktorsavhandling). ISBN: 978-91-7636-631-8.
158. Annelie Johansson Sundler, 2008. Mitt hjärta, mitt liv: Kvinnors osäkra resa mot häl-sa efter en hjärtinfarkt (doktorsavhandling). ISBN: 978-91-7636-633-2.
159. Attila Lajos, 2008. På rätt sida om järnridån? Ungerska lantarbetare i Sverige 1947-1949. ISBN: 978-91-7636-634-9.
160. Mikael Ohlson, 2008. Essays on Immigrants and Institutional Change in Sweden (doktorsavhandling). ISBN: 978-91-7636-635-6
161. Karin Jonnergård, Elin K. Funck, Maria Wolmesjö (red.), 2008. När den professio-nella autonomin blir ett problem. ISBN: 978-91-7636-636-3
162. Christine Tidåsen, 2008. Att ta över pappas bolag. En studie av affärsförbindelser som triadtransformationer under generationsskiften i familjeföretag (doktorsavhand-ling). ISBN: 878-91-7636-637-0
163. Jonas Söderberg, 2009. Essays on the Scandinavian Stock Market (doktorsavhand-ling). ISBN: 978-91-7636-638-7
164. Svante Lundberg, Ellinor Platzer (red.), 2008. Efterfrågad arbetskraft? Årsbok 2007 från forskningsprofilen Arbetsmarknad, Migration och Etniska relationer (AMER) vid Växjö universitet. ISBN: 978-91-7636-639-4
165. Katarina H. Thorén, 2008 “Activation Policy in Action”: A Street-Level Study of Social Assistance in the Swedish Welfare State. ISBN: 978-91-7636-641-7
166. Lennart Karlsson, 2009. Arbetarrörelsen, Folkets Hus och offentligheten i Bromölla 1905-1960 (doktorsavhandling). ISBN: 978-91-7636-645-5.
167. Anders Ingwald, 2009. Technologies for better utilisation of production process re-sources (doktorsavhandling) ISBN: 978-91-7636-646-2.
168. Martin Estvall, 2009. Sjöfart på stormigt hav – Sjömannen och Svensk Sjöfarts Tid-ning inför den nazistiska utmaningen 1932-1945 (doktorsavhandling). ISBN: 978-91-7636-647-9.
169. Cecilia Axelsson, 2009. En Meningsfull Historia? Didaktiska perspektiv på historie-förmedlande museiutställningar om migration och kulturmöten (doktorsavhandling). ISBN: 978-91-7636-648-6.
170. Raisa Khamitova, 2009. Symmetries and conservation laws (doktorsavhandling). ISBN: 978-91-7636-650-9.
171. Claudia Gillberg, 2009. Transformativa kunskapsprocesser för verksamhetsutveck-ling – en feministisk aktionsforskningsstudie i förskolan (doktorsavhandling). ISBN: 978-91-7636-652-3.
172. Kina Hammarlund, 2009. Riskfyllda möten. Unga människors upplevelser av sexu-ellt överförbara infektioner och sexuellt risktagande (doktorsavhandling). ISBN: 978-91-7636-653-0.
173. Elin K. Funck, 2009. Ordination Balanced Scorecard – översättning av ett styrin-strument inom hälso- och sjukvården (doktorsavhandling). ISBN: 978-91-7636-656-1.
174. Ann-Kari Sundberg, 2009. Le poids de la tradition. La gestion professorale de l’altérité linguistique et culturelle en classe de FLE (doktorsavhandling). ISBN : 978-91-7636-657-8.
175. Peter Bengtsson, 2009. Development towards an efficient and sustainable biofuel drying (doktorsavhandling). ISBN: 978-91-7636-659-2.
176. Linda Reneland-Forsman, 2009. A changing experience – communication and meaning making in web-based teacher training (doktorsavhandling). ISBN: 978-91-7636-660-8.
177. Anders Andersson, 2009. Numerical conformal mappings for waveguides (doktorsavhan-dling). ISBN: 978-91-7636-661-5.
178. Rune Svanström, 2009. När livsvärldens mönster brister – erfarenheter av att leva med demenssjukdom (doktorsavhandling). ISBN: 978-91-7636-662-2.
179. Mats Anderberg och Mikael Dahlberg, 2009. Strukturerade intervjuer inom missbruks-vården – en grund för kunskapsutveckling (doktorsavhandling). ISBN: 978-91-7636-663-9.
180. Arianit Kurti, 2009. Exploring the multiple dimensions of context: Implications for the design and development of innovative technology-enhanced learning environments (doktorsavhandling). ISBN: 978-91-7636-665-3.
181. Joakim Krantz, 2009. Styrning och mening – anspråk på professionellt handlande i lärarutbildning och skola (doktorsavhandling). ISBN: 978-91-7636-671-4.
182. Hans Lundberg, 2009. Kommunikativt entreprenörskap: Underhållningsidrott som totalupplevelse före, under och efter formeringen av den svenska upplevelseindustrin 1999-2008 (doktorsavhandling). ISBN: 978-91-7636-673-8.
183. Jens Nilsson, 2009. Transformation and Combination in Data-Driven Dependency Parsing (doktorsavhandling). ISBN: 978-91-7636-674-5.
184. Uffe Enokson, 2009. Livspusslet: Tid som välfärdsfaktor (doktorsavhandling). ISBN: 978-91-7636-676-9.
185. Karin Olsson, 2009. Den (över)levande demokratin. En idékritisk analys av demo-kratins reproducerbarhet i Robert Dahls tänkta värld (doktorsavhandling). ISBN: 978-91-7636-677-6.
186. Rüdiger Lincke, 2009. Validation of a Standard- and Metric-Based Software Quality Model (doktorsavhandling). ISBN: 978-91-7636-679-0.
187. Lina Andersson, 2009. Essays on economic outcomes of immigrants and homosexu-als (doktorsavhandling). ISBN: 978-91-7636-680-6.
188. Susanna Holzer, 2009. University Choice, Equality, and Academic Performance (doktorsavhandling). ISBN: 978-91-7636-681-3.
Växjö University Press S-351 95 Växjö www.vxu.se, [email protected]