Volume 23 Issue 1 Article 3 June 2021 An Empirical Analysis of the Effects of Student Work and An Empirical Analysis of the Effects of Student Work and Academic Performance on the Probability of Employment Academic Performance on the Probability of Employment Tjaš a Bartolj Institute for Economic Research, Ljubljana, Slovenia, bartoljt@ier.si Saš o Polanec University of Ljubljana, School of Economics and Business, Ljubljana, Slovenia, saso.polanec@ef.uni-lj.si Follow this and additional works at: https://www.ebrjournal.net/home Part of the Business Commons Recommended Citation Recommended Citation Bartolj, T., & Polanec, S. (2021). An Empirical Analysis of the Effects of Student Work and Academic Performance on the Probability of Employment. Economic and Business Review, 23(1), 26-39. https://doi.org/10.15458/2335-4216.1003 This Original Article is brought to you for free and open access by Economic and Business Review. It has been accepted for inclusion in Economic and Business Review by an authorized editor of Economic and Business Review. ORIGINAL ARTICLE An Empirical Analysis of the Effects of Student Work and Academic Performance on the Probability of Employment * Tjasa Bartolj a, *,Sa so Polanec b a Institute for Economic Research, Ljubljana, Slovenia b University of Ljubljana, School of Economics and Business, Ljubljana, Slovenia Abstract The theoretical relationship between student work and post-college probability of employment is ambiguous, due to theopposingdirectandindirecteffectsonhumancapitalaccumulation.Studentworkmayononehandloweracademic performance and thus harm the likelihood of getting a job, while on the other hand enabling students to acquire skills that increase their labour market odds. In this paper, we provide an answer to the question, whether the policy should encourage or limit student work, by using rich data which allow us to compare the effects of the two investments in human capital on the likelihood of employment. We use personal characteristics, socio-economic background, and ac- ademic performance in the propensity score matching to calculate the differences in the probability of employment for different amounts of student work. We find that only work experiences up to two years have a beneficial effect on employment prospects, while much larger effects are observed for improvements in educational attainment, like graduation and improvement in GPA. In the end, our results provide support for setting limits to the extent of student work during college, but certainly not for its prohibition. Keywords: Student work experience, Academic performance, Probability of employment, Propensity score matching JEL classification: I23, J21 Introduction T he debate on the advantages and disadvan- tages of student work is vibrant and has beenongoingfordecades,notonlybecauseofthe high percentage of students who work during college,butalsobecauseofitsopposingdirectand indirect effects. The latter work through academic performance on employment prospects (in the first years after leaving college), which according to the evidence, appears to have long-term effects on future employment and/or earnings (see, for example, Ellwood, 1982; Gregg & Tominey, 2005; Mroz & Savage, 2006; Nilsen & Reiso, 2011; Nordstrom Skans, 2011; Oreopoulos et al., 2012; Ryan, 2001; Schmillen& Umkehrer, 2017). In economics, two competing theoretical views explain the existence of positive effects of academic performance and work experience on the success of post-college entry on the labour market. The first is the human capital theory (Becker, 1962, 1964, 1993), according to which the skills accumulated in edu- cation or at work enhance an individual's produc- tivity, resulting in improved labour market outcomes. The second theoretical view is the screening/signalling theory (Arrow, 1973; Spence, 1973), which is based on the premise that education * We would like to thank the Slovenian Statistical Office for providing us with the data and allowing us to prepare the data in a secure room. We would also like to thank two anonymous referees for many valuable comments. Received 18 October 2019; accepted 14 May 2020. Available online 15 June 2021. * Corresponding author. E-mail addresses: bartoljt@ier.si (T. Bartolj), saso.polanec@ef.uni-lj.si (S. Polanec). https://doi.org/10.15458/85451.1003 2335-4216/© 2021 School of Economics and Business University of Ljubljana. This is an open access article under the CC-BY-NC-ND license (http://creativecommons. org/licenses/by-nc-nd/4.0/). andpreviousworkexperienceserveonly asasignal of an individual's productive characteristics. Althoughthecausesoftheeffectofeducationand student work experience on the post-college labour market outcomes are different in the theories listed above, they all predict a positive impact of the acquisition of both types of skills/credentials on employment probability. Concurrently, the theory of time allocation (Becker, 1965) provides founda- tions for a negative association between student work and academic achievementdstudents must allocate a limited time between the two activities, thus putting more hours into the accumulation of work experience likely harms academic perfor- mance. This is also supported by empirical evi- dence. Stinebrickner and Stinebrickner (2003), Auers et al. (2007), DeSimone (2008), Callender (2008), Kalenkoski and Wulff Pabilonia (2010),and Beerkens et al. (2011) found a negative effect of student work during college on GPA. In addition, Ehrenberg and Sherman (1987), Beerkens et al. (2011), Kosi-Antolic, Nastav and Sustersic found a negative effect of student work on ‘graduation-on- time’, while Darolia (2014) discovered a negative effect on the number of credits per term. The adverse effects of student work on GPA, number of exams passed and the probability of passing a year are also found in a paper analysing similar data as used in this study (Bartolj& Polanec, 2018). In sum, the indirect effect of student work on post-college labour market outcomes is negative. The relevant question for policymakers, in- stitutions,andstudentsthatweattempttoaddressin this article is the following: Should student work during college be limited or not? To answer the question,wepresentevidence-basedadvicefounded on a comparison of two effects: the effect of student work experience on the probability of employment, andtheeffectofacademicperformanceontheprob- ability of employment. For this purpose, we use administrativedataonasetofSlovenianundergrad- uate students, who were enrolled in 4-year under- graduate programs offered by the School of EconomicsandBusinessduringtheperiod1997e2008 andwereseekingworkbetween2002and2010.Toour knowledge, this study presents the first attempt to compare the relative impacts of student work and educationalperformanceontheprobabilityofgetting a job after college, using one data set and the same methodologytoestimatebotheffects.Thedifferences in the estimated effects thus cannot be attributed to the discrepancies in the institutional contexts, mea- surement and/or estimation method. The only anal- ysisthatiscomparabletooursexaminedonlymalesin the U.S. during the period 1972e1979. In addition, it concentrated on the effects on earnings and found evidence of a positive effect of higher grade-point averages,but failed tofind anyrelationship between studentworkandpost-collegeearnings. A distinguishing feature of our empirical analysis is also the applicationof propensity score matching, whichhasnotyetbeenusedinthisspecificcontext. 1 On the one hand, this method allows us to compare labour market outcomes of students with different academic performance but similar student work experience, personal characteristics, and socio-eco- nomic background. On the other hand, we can observe differences in the probability of employ- ment of students with similar academic perfor- mance,personalcharacteristics,andsocio-economic background, but diverse student work experience. We consider its property of putting emphasis on observationswithsimilarregressorsandthusgiving low or no weight to observations at a margin as an advantage over the methods that minimize squared errors and give such observations a high weight. Inlinewiththetheoreticalpredictions,ourresults, based on the data on the students of business and economics studies in Slovenia, confirm positive ef- fects for both types of skills/credentials (work experience and education) on the probability of being employed after college. However, the returns to student work experience seem to be diminishing. We find that 10e24 months of student work expe- rience(gainedwithinthefouryearsofstudies)were associated with an approximately 10 percentage point increase in the probability of employment, though increasing student work experience beyond 2 years was not beneficial. A similar increase in the probability of employment was also related with ranking among the top 25% of class, as opposed to the bottom quartile of the class, based on the grade point average (GPA). Nevertheless, writing and defending a thesis, which was supposed to take one year, is found to have been associated with more than a 20 percentage points increase in the proba- bility of employment in comparison to passing all examsbutnotgraduating.Wealsofindthatthetype ofstudentworkexperiencewashighlyrelevant.Our analysis shows that acquisition of student work experience in high-skilled jobs that were related to thefieldofstudyincreasedtheprobabilityofgetting 1 Quasi-experimental methods based on propensity score matching (PSM) are, however, still frequently used in evaluation of various treatments on the labour marketoutcomes.Infact,theseminal contributionbyDehejiaandWahba(1999)considered theeffects oftrainingdaformofinvestment inhuman capitaldon earnings. PSM has been used in more than 50 studies of active labour market policies alone (Vooren et al., 2018). ECONOMIC AND BUSINESS REVIEW 2021;23:26e39 27 a job after college by almost 16 percentage points in comparison to student work experience in low- skilledjobsthatwereunrelatedtothecollegemajor. The analysis provides support forsetting limits on theextentofstudentworkduringcollege,atleastfor the students of SEB UL. As the benefits of student work experience do not seem to exceed those of the analysed academic performance indicators per requiredunitoftime,itisnotworthwhiletosacrifice academic success in return for higher student work experience. Furthermore, results suggest that the optimal amount of student work coincides with the cumulative length of college school holidays, during whichstudentworkminimallyharmsacademic suc- cess. Thus, the policies, such as the one in Belgium thatlimitsstudentworktoupto475hperyear,seem tobewell-groundedandthemostefficient. 1 Review of empirical evidence The bulk of relevant literature on the association between academic performance and labour market outcomes is concentrated on earnings. The most substantial part of this literature deals with the estimation of returns to education (for a review of estimation approaches see Heckman et al., 2006; for the most recent review of results, see Psachar- opoulos & Patrinos, 2018; and for returns to educa- tioninSloveniadthecountryusedinthisstudydsee Bartoljetal.,2013).Theglobalaverageprivatereturn to a year of schooling based on these estimations is 9% a year (Psacharopoulos & Patrinos, 2018). The vastliteraturealsoprovidesevidenceonthepositive relationship between earnings and GPA, school qualityand/ordiploma.Seminalpapersinthisfields of research include Jones and Jackson (1990) on the impactof grades,Card andKrueger(1992)andBetts (1995) on the impact of school quality, and Hun- gerford and Solon (1987) and Belman and Heywood (1991) on sheepskin effects. However, the relation- ship between education and the probability of employment is less explored. Recent exceptions are experimental studies by Deming et al. (2016) and Piopiuniketal.(2018)whoconsidertheeffectsonthe likelihood ofa call back fortwodifferenttreatments: completion of the online educational program (compared to the on-campus ones) and improve- ment of GPA. Deming et al. (2016) find 22% lower probabilityofacallbackforbusinessbachelorsfrom for-profit online institutions than from non-selective public institutions, whereas Piopiunik et al. (2018) find that a one grade level increase in college GPA resultsina38percentagepointshigherlikelihoodof landing a job-interview. The literature that analyses the relationship be- tween student work experience and labour market outcomes can be divided on the basis of the level of education during which student experience is acquireddsecondary versus tertiary education. The empirical studies that analyse the effects of high- school work experience on post-study labour mar- ket outcomes (e.g. Light, 1999; Light, 2001; Ruhm, 1997) cannot be generalized to work performed by college students, as college students are more likely to find jobs that are related to their field of study, and thus enjoy higher returns to work experience. Thus, we concentrate solely on the effects found in the studies that use data from college students. Recentexperimentalstudiesyieldmixedresults.For example, while Baert et al. (2015) find no causal evidence of mentioning student work experience in a resume on the likelihood of a call back in their field experiment with fictitious job applications, Piopiunik et al. (2018), on the other hand, show that longer internship increases the probability of being invited to a job interview. However, in the system- atic literature review of the effectiveness of super- vised work placements in higher education (i.e. studentstaketimeoutofeducationtoworkfull-time in an organization), Inceoglu et al. (2018) conclude that such placements elicit overall small positive effects on career outcomes. Less structured student work experience seems to be positively related to post-college labour outcomes as well. For example, Hotz et al. (2002) find that returns to college employmentrangebetween4.6and5.4%forwhites. These results are confirmed by H€ akkinen (2006), whoalsoshowsthatoneadditionalyearofin-school work experience increases the probability to be employed one, two and three year(s) after gradua- tion by 5.6, 4.2 and 3.7 percentage points, respec- tively. Also, Scott-Clayton and Minaya (2016) do find a comparable relation between experience and the probability of employment, though the effects seem to be higher for the experience that is related to the field of study (Geel & Backes-Gellner, 2012). 2 Institutional context of the study Our empirical study relies on the data from the largest Slovenian university. Our sample consists of full-time undergraduate students, who were first enrolled in the four-year programs at the School of Economics and Business, 2 the University of 2 Until 2019, the name of the faculty was Faculty of Economics, University of Ljubljana. 28 ECONOMIC AND BUSINESS REVIEW 2021;23:26e39 Ljubljana (henceforth SEB UL), between 1997 and 2004, and were available in the labour market as regular employees between 2002 and 2010. Several studieshavealreadyusedtheSlovenianuniversities as their primary data source, which suggests that the data from the Slovenian tertiary institutions have been acknowledged as a suitable source for academic research (e.g. Bartolj & Polanec, 2012; Bartolj&Polanec,2018;Farcnik&Domadenik,2012; Cadezetal.,2017;Kosi-Antolicetal.,2013).Here,we provideashortdescriptionofthekeycharacteristics of the institutional context for this study, while an interested reader can find additional information in Bartolj and Polanec (2012), Cadez et al. (2017),and Bartolj and Polanec (2018), which also make use of the data from SEB UL. The sample of full-time students of SEB UL is nevertheless not representative of the respective cohorts of high-school graduates, as these could enrol in full-time four-year UG programs only, if they had completed any four-year high school and achieved sufficiently high weighted average grade 3 among the nationally ranked applicants. For such students with a Slovenian residence, studies were tuition-free, which implies that payment of tuition costs was not a motive for supplying student work. 4 SEB UL offerededuring the period of analysisethree economics majors (banking and finance, international economics, and national eco- nomics) and five business majors (accounting and auditing, business informatics, finance, marketing, and management and organization). Business ma- jors were chosen by a large majority of students, among which dominated finance, management and organization, and marketing majors. Students were expected to complete the four-year study program within five years (including the year for writing the final thesis), although study duration varied be- tween four and six years and could extend to beyond ten years. The grading scheme in Slovenian universities operates on a ten-point scale, with 1 as the lowest and 10 as the highest grade, 5 and 6 as the minimum passing grade. Students can attempt to pass exams three times per academic year in each course. The Slovenian regulation at the time limited stu- dent work to full-time students between 15 and 26 years of age, who were enrolled in any state- approved primary, vocational, high school, or undergraduate program. The student work con- tracts, called referrals, were issued by licensed intermediariesestudent employment agencies. The incentives for hiring student workers were strong during the analysed period, as regular employment contracts were subject to high social contributions, which amounted to 38.2% of gross wages, whereas student-employment contracts were not subject to any such tax. Regular employees were also entitled to a bonus for working night shifts, Sundays, holi- days, higher wages for overtime, seniority and job performance bonuses, while no such rights were given to students. Further, regular employees were also paid tax-free costs for meals during working hours and daily commuting costs (SSC Act, 2001), and their incomes were subject to a progressive payroll tax. Despite the advantageous tax regime, student workwassubjecttotaxation.Specifically,employers had to pay concession fees (on top of student gross earnings), value-added tax on the concession fee, while students were subject to personal income tax. Duringtheperiodoftheanalysis,theconcessionfee increased from 10 percent of students' gross earn- ings (1997e2003) to 12 percent (2004e2006), and then even further to 14 percent (2006e2008). As the VAT rate was 20% on the concession fee, the total cost for student work for employers in 2008 was EUR 1.168 for every EUR of gross earnings. More- over, students' personal incomes were subject to positive tax rates above significantly higher in- comes, due to additional (student-specific) personal tax deduction. Hence, net and gross earnings were the same for a vast majority of students. The comparatively low cost of student work was the major driver of demand for this type of labour. In 2008, 114,391 Slovenian students in all types of tertiary education, as many as 927,809 student employment contracts were issued and 54,363,336 h of work performed. This amount of work is equiv- alent to work performed by 26,000 full-time em- ployees or2.9% ofaggregateemployment(inclusive of student work). 3 Data sources We aim to estimate the causal effects of student employment and academic performance on the probability of employment (emphasized arrows in Fig. 1). For this purpose, we constructed an 3 This average was calculated from the grade percentage averages achieved in the third and fourth year of high school study and a national exam- dmaturadthe Slovene equivalent of the SAT in the US. 4 Bartolj and Polanec (2020) study the determinants of labour supply decisions by UL students enrolled in four- to six-year undergraduate programs during this period. 5 However, grades below 5 were rarely used in practice, which led us to set such grades to equal 5. ECONOMIC AND BUSINESS REVIEW 2021;23:26e39 29 individual-level panel of employment histories during and after studies, academic performance, earnings and personal characteristics of full-time undergraduate students at SEB UL, by merging in- dividual-level data from the following sources in a secure room at the Slovenian Statistical Office (SORS): Slovenian Tax Authority (TARS): The data on student and regular-employee earnings are re- ported to TARS by student employment agencies and employers. 6 TARS provided us with the information on the labour incomes earned by persons during and after the completion of studies. While students with suf- ficientlylowearningsaretypicallynotobligedto report personal incomes, student employment agencies have a legal obligation to report earn- ings received by each working high school or college student. In addition, TARS is also the source of data for incomes of students' families and post-college earnings of students. Personal incometaxfilingsalsoincludelabourandcapital incomes, which are an essential piece of infor- mationfortheconstructionofthevariablefamily income per member, i.e. one of the variables used in the propensity score matching proced- ure. Furthermore, labour incomes of families include not only wages and salaries, but also bonuses, perks, wages earned from short-term labour contracts, and royalties, while capital in- comes include interest, dividends, rents, and incomes of sole proprietors. SEB UL: From this source of data, we use appli- cation-sheet data and the data on all attempts to pass exams, grades achieved for all students enrolled in the four-year programs, and year of graduation. We extract the information on age, gender, the location of permanent residence, chosen major, study year, grades and year of graduation. Based on the enrolment history of each student, we also at this point construct variables that indicate if students passed a year, repeatedayear,ordroppedoutofaprogram.In addition, exam results are used to construct variables on study performance of students. National Examination Center: We extract infor- mation on the third- and fourth-year average grades and the grades from the final (external) examination called matura from this source. These grades are used to construct the high school GPA. SORS: We obtain the data from the Central Registry of Population, which allow us to estab- lish the identity of the students' parents, using a unique person identifier for each student, and also attribute family incomes and transfers to each student. Knowing the identity of parents enables us to determine their educational attainmentandfamilyincome.From thissource, we also obtain information on all scholarships receivedbystudents,suchassocialscholarships, targeting students with low-income families, scholarships for talented individuals (Zois scholarships) and scholarships granted by pro- spective employers. For a subsample of students, we also use data from a student employment agency, e- Studentski servis. This is an agency with a market share exceeding 50 percent in student work intermedia- tion.Asthisemploymentagencyhasmoreoutletsin central Slovenia, its market share in total student employment is likely even higher for SEB UL stu- dents, who typically select a student employment agency that intermediates the student work trans- actions. Their data contain not only information on theincomespayedforeachstudentworker,butalso information on the number of working hours, the identityoftheemployer,andthetypesofjobsforall studentswhousedtheirservices.Unfortunately,the information from this source of data is limited to referrals issued between January 2006 and December 2010, which reduces our sample to roughly half of all observations. 4 Description of variables and summary statistics As already mentioned on pages 4 and 5, we restricted our sample to a set of full-time Fig. 1. Representation of causal chain. 6 Astandardprocedurefordatacollectionbytaxauthoritiesisreportingincomesbyindividuals,whichwasalsothecaseinSloveniaduringtheperiodof our analysis. However, the data we use were reported by employers for regular employment or student employment agencies for student work, and were initially used for inspection purposes. 30 ECONOMIC AND BUSINESS REVIEW 2021;23:26e39 undergraduate SEB UL students, who started studies during the period from 1997 to 2004, were 18e20yearsoldwhenenrollinginthefirstyear,and passed all mandatory exams during the entire four- year study. These students were expected to finish their study with a defence of the thesis within one yearafterpassingtheexams.Werefertothisperiod the final year of study. As we have no information ontheactualstartof thesearchforaregular job,we assume thefirst and the second years on the labour market to be one and two calendar years after the expected year of studies (inclusive of thefinal year). We determine the amount of student work experi- ence, the post-college employment status and aca- demic performance for these two periods, which refer to the time period from 2002 to 2010 for our sample of students (see also Fig. 2). Thesampleconsistsof2616and2347observations in the first and second year on the labour market, amongwhich around 60% are females (see Table 1). Note that in the vast majority of cases, the employment status for the same person is observed in the first and the second calendar year after the final year of study. The post-college employment status is measured with an indicator variable that assumes value 1, if a person worked at least 1 h and earnedpositiveearningsinregularemploymentina calendar year, and value 0 otherwise. 7 Table 2, which contains the summary statistics, reveals that 65.0% and 86.3% of persons were employed for at least 1 h in the first and second year, respectively. One of the main questions that we aim to address is whether student work experience has any effect on the likelihood of post-college employment. We construct a measure of student work from earnings reported by student employment agencies. As we donothaveinformationontheactualhoursofwork performed, we use information on the average hourly gross wage rates, reported annually by the largest student employment agency for regular Fig. 2. Representation of the sample. Table 1. Sample size by gender. 1st Year 2 nd Year Number of observations 2616 2347 Males 1068 952 Females 1548 1395 Table 2. Summary statistics. 1st Year 2nd Year Mean Sd Mean Sd Employed after college 0.650 0.477 0.863 0.344 Student work experience in years 1.833 1.147 1.868 1.156 Graduated 0.661 0.473 0.809 0.393 Time to final year 4.522 0.749 4.521 0.741 No. of exam attempts 54.718 12.620 54.454 12.451 Avg. grade 6.801 0.750 6.808 0.744 Age (at enrolment to faculty) 18.895 0.407 18.885 0.414 High school GPA 0.511 0.155 0.518 0.153 University or higherdmother 0.208 0.406 0.206 0.404 University or higherdfather 0.235 0.424 0.235 0.424 Family business 0.162 0.369 0.156 0.363 Step parent 0.235 0.424 0.239 0.426 No. of siblings 0.790 0.750 0.804 0.746 Student parent 0.006 0.085 0.006 0.085 Non-labour income 8005 5794 7920 5682 Conditional income share 0.147 0.232 0.153 0.237 Capital income share 0.042 0.088 0.041 0.088 Expected net wage 15.852 2.484 15.749 2.452 Year 2006.7 2.2 2007.3 1.9 Note: All income-related variables are in constant (2004) Euros. The exchange rate in 2004 was 1 EUR ¼ 1.24 USD. Variables describing family characteristics and the economic conditions of students during studies are measured in the final year of study. 7 We also considered alternative (stricter) definitions of employment status, like employment with indefinite contracts, and found qualitatively similar results. ECONOMIC AND BUSINESS REVIEW 2021;23:26e39 31 college students, to calculate the total number of workinghours. 8 Table2showsthattheaveragetotal student work experienceecalculated by dividing hours of student work by the total number of hours of a full-time employee per yearewas roughly 1.8 years.Assumingtheaveragedurationofstudies(5.5 years), the average amount of work was around 3.3 months or 11 h per week. In addition to the effects of student work experi- ence, we are also interested in the treatment effects of study outcomes on the probability of employ- ment, as employers seeking regular workers may select them based on their study results, as shown by Pinto and Ramalheira (2017) for business stu- dents. The first such indicator pertains to gradua- tion, which assumes value 1, if the student graduated (passed all exams and successfully defended their thesis), and 0 otherwise. This vari- able aims to capture the well-known ‘sheepskin’ effect (Belman & Heywood, 1991; Hungerford & Solon,1987).Table2revealsthatthemajorityofSEB UL students defended their theses on time, as the share of graduates in thefirst year was 66.1%, while by the second year this share increased to 80.9%. The other measures of study performance are time- needed to reach the final year, the total number of attempts to pass all exams, and the average grade achieved in all exams. Table 2 shows that the average time to the final year was around 4.5 years, which implies that students needed roughly half a year more than expected. While the total number of examswas38,theaveragenumberofexamattempts was significantly highere54. The average grade for all exams was around 6.8. Next, we describe a set of variables, based on which we perform propensity score matching. This set covers the students' own and their families' characteristics, values, and the structure of non-la- bour incomes, major-specific expected net wages, indicator variables for year of entry to the labour market, and the region of permanent residence of students. Starting with personal characteristics, Table 2 reports the average age when enrolled at university, and high school GPA, a measure of general ability. As aresult of theconstructionof our sample,whichincludesonlyfull-timestudents,who chose programs offered by the SEB ULimmediately after completing high-school studies, the average age was only 19 years. Our measure of general ability, namely high school GPA, is calculated from gradesachievedinthethirdandfourthyearofstudy and thefinal exam. We then normalize the measure to range between 0 and 1 (by subtracting 2, the minimum passing grade, and further dividing this difference by 3). The average normalized GPA is around 0.5 in both periods. Turning to family characteristics, we use the data ontheeducationalattainmentofparents,ownership of family business by either of the two parents, having a step parent, number of siblings below the age of 27, and parental status before entry into the labour market. These variables are included in the estimation of propensity scores, as these may affect student work and academic success, as well as la- bourmarketoutcomes.Theeducationalattainments ofparentsaremeasuredwiththeindicatorvariables that assume value 1, if they completed at least a four-year undergraduate college degree, and value 0 otherwise. Table 2 shows that around 20% of mothers and fathers had a college degree, while roughly 16% of students had parents who owned a family business. On average, 24% of students had step parents, and less than 1% had a child during their studies. When entering the labour market, persons had on average less than one sibling under the age of 27. The economic situation of a student during the last year of study is captured by three variables: non-labour income, conditional income share and capital income share in non-labour income. Non- labour income is calculated as the sum of all in- comes that are unrelated to student work: (i) net family income per family member, which is con- structed as the sum of parental net income, divided by the number of family members 9 , and serves as a proxy for parental transfers; (ii) scholarships; and (iii) pension received after deceased parents. The share of income that depends on academic success (conditionalincomeshare)iscalculatedasashareof scholarships and pension benefit payments from a deceased parent 10 in the student's non-labour in- come. Capital income share, on the other hand, is the share of capital incomes in non-labour income, where we use information on capital gains, divi- dends, copyright income, etc. from personal income tax statements. The average non-labour income in the final year of study for persons who had entered the labour market was EUR 8,005, 14.7% of which 8 The observed differences in hours could in principle also reflect the differences in hourly wage rates. However, this limitation of the data could not be avoided. 9 We count parents and children under the age of 27 as family members, following the income tax act that defines a dependent family member as a person up to the age of 26 (in addition to other requirements). 10 Childrenhavetherighttoreceiveapensionaftertheirdeceasedparentuntiltheendoftheirschoolingoruntiltheyare26yearsold.Therefore,students who are not enrolled in a program lose pension. 32 ECONOMIC AND BUSINESS REVIEW 2021;23:26e39 depended on academic success, and 4.2% of which the income pertained to capital. Finally, Table 2 also contains summary statistics for the expected net hourly wage, which is calcu- lated separately for each year, major, and gender. We presume that students formed expectations based on the most recent net wage of persons who graduated in their major. 11 To capture the differences in the specific labour market conditions, our empirical model also in- cludes indicator variables for year of observation (Table 2), major (Table 3), and region of permanent residence of persons (Table 4). Table 3 reveals a significant variation in the popularity of different majors. The most popular were business majors, such as finance, marketing, and management and organization, while among the economics majors, banking and finance dominated. 12 The regional structure reveals that the students were most likely to originate from the Osrednjeslovenska region, where SEB UL is also located (see Table 4). 5 Estimation method In order to estimate the effect of student work on the probability of employment after college, as well as the effect of academic performance on post-col- lege probability of employment, we employ the propensity score matching technique (see e.g. Angrist & Pischke, 2009; Lee, 2005). This type of estimation enables us to match students with different work histories during their studies, but with similarly predicted probabilities or propensity scores of student employment level, and compare their post-college labour market outcome. Analo- gously, we can compare the post-college employ- ment status of ‘similar’ 13 students, who differed ‘only’ in their academic performance. The advan- tages of propensity score matching are two-fold: (i) it avoids the dimensionality problem of finding matched subjects, if there are many control vari- ables, and (ii) it imposes minimal structure on esti- mation. Another advantageous property of matching is the fact that it emphasizes observations with similar regressors, namely, observations at margin are likely to get lower weights. In contrast, OLStriestominimizesquarederrorsandthusgives observations at margin relatively high weights. The estimation of treatment effects is done in two steps. In the first step, we estimate propensity scores. Since we measure academic success and student work experience at the end of studies, extensive student work could have harmed aca- demic success in a current year, but at the same time, poor academic performance could have low- ered student work in the subsequent year (this is presented with a double headed arrow in Fig. 1). Thus, the logit regression for the probability of working k hours during study (denoted SW k ), in- cludes not only the students' personal and family characteristics,amountsandstructureofnon-labour incomes, major-specific expected net wages, indi- cator variables for year of entry to labour market, region of permanent residence of students (denoted x), but also academic performance (A) as explana- tory variables. Similarly, we use the variables in x and student work as explanatory variables in the estimation of the propensity scores for academic performance: Pr[SW ki ¼ 1]¼ a 0 þ a 1 x i þ a 2 A i þ u i (1) Table 3. Structure of sample by major. Major 1st Year 2nd Year National Economy 1.22 1.24 International Economics 6.27 6.99 Banking and Finance 9.59 10.31 Marketing 19.07 18.07 Finance 31.65 31.23 Accounting 9.29 9.08 Management and Organization 13.38 13.98 Business Informatics 9.52 9.12 Note: Table presents shares in percent of the respective column total. Table 4. Structure of sample by region. Region 1st Year 2nd Year Pomurska 1.45 1.62 Podravska 1.26 1.24 Koroska 1.80 1.62 Savinjska 7.11 6.90 Zasavska 2.14 1.96 Spodnjeposavska 2.33 2.39 Jugovzhodna 9.25 9.76 Osrednjeslovenska 45.95 45.97 Gorenjska 13.38 12.82 Notranjskoe Kraska 2.48 2.39 Goriska 7.19 7.37 Obalnoe Kraska 5.66 5.97 Note: Table presents shares in percent of the respective column total. 11 BartoljandPolanec(2012)demonstratedthatSEBULundergraduatestudentsmakecollegemajorchoicesbasedonpastnetwages,whichsuggeststhat the assumed formation of expectations is reasonable. 12 Note that Slovenian employers often require specific fields of specialization in job advertisements. 13 Similar in values of the observed variables. ECONOMIC AND BUSINESS REVIEW 2021;23:26e39 33 Pr[A ji ¼ 1]¼ b 0 þ b 1 x i þ b 2 SW i þ e i. (2) This conditional probability of working k hours during the study (or achieving level j of academic performance), given the explanatory variables, is used to match the students who worked different numbers of hours during the study (or performed differently in terms of study outcomes), but have similar propensity scores values. The matching al- gorithm used in our analysis is radius matching with replacement and imposed common support. As suggested by Austin (2011), we use a calliper equal to 0.2 of the standard deviation of the logit of the propensity score. 14 Weexpectdifferentlevelsofstudentworktohave different impacts on labour market outcomes. Therefore, we create six different indicator vari- ables, which allows an estimation of six different treatment effects. We estimate these weighted average conditional differences (WACD), using the same mathematical formulas that are used in esti- mating the average treatment effects on the treated (ATETs). However, as we cannot explain the full variation of the selected treatments with the observable variables, our estimated effects reflect both the differences in treatment levels and the differences in unobserved characteristics. Conse- quently, the estimated effects do not have causal interpretation, and we hence abstain from using the term average treatment effect on the treated. As shown in Table 5, we compare students who had less than 10 months of student work experi- ence 15 with those with 10e24 months of work experience (WACD 11 ), students who gained 2e3 years of work experience (WACD 12 ), and students with more than three years of work experience (WACD 13 ). Similarly, we compare students who gained10e24monthsofworkexperiencewiththose with 2e3 years of student work experience (WACD 22 ), and so on. The benefits of such estima- tionare describedin BartoljandPolanec (2018).The effects of academic performance on labour market outcomes are estimated by comparing students with, on one hand, similar propensity scores (described with Equation (2)) and, on the other hand, differences in two measures of academic successecumulative grade point average (GPA) of all exams taken in college and graduation. All effects are estimated for the first and second year on the labour market. It should be noted that we do not estimate the ‘dynamic model’, which would include the lagged employment in the esti- mation of propensity scores for the second year, 16 because we are interested in the impacts that reflect the total effects, including the indirect effect of for example student work on labour market outcomes in the second year through the outcomes in thefirst year on the labour market. Nevertheless, we recognize the fact that propensity score matching could only reduce the part of the endogeneity bias that is captured by observable determinants of stu- dent work and academic performance. While these might be correlated to unobservable characteristics, suchasmotivation,abilityorpreferences,wecannot be certain that conditioning on observable charac- teristics fully eliminates the effect of selection on unobservable characteristics. In this case, our esti- mated associations reflect both the effects of the variables of interest and unobserved heterogeneity. Thatiswhywedonotinterpretthemastheaverage treatmenteffectonthetreated(ATET),asisusualin the propensity score matching literature. 6 Results of the empirical analysis 6.1 Student work and post-college probability of employment Let us start the discussion of our results by focusing on the relationship between student work and the probability of employment after college (represented by an emphasized arrow running be- tween student work and Pr[Employment] in Fig. 1). Before commenting on the results, note that the Table 5. Estimated WACDs based on the amount of student work experience. Student work experience 10e24 months 2e3 years more than 3 years (32%) (23%) (22%) less than 10 months (23%) WACD 11 WACD 12 WACD 13 10e24 months (32%) WACD 22 WACD 23 2e3 years (23%) WACD 33 Note: Values in parentheses present the share of students in the sample that belongs to each group. 14 We also tried other matching algorithms and other calliper values, but obtained qualitatively similar results. This method is selected based on the recommendation by Lee (2005) in order to make a comparison group as localized as possible and the baseline differences between treated and controls as small as possible. 15 The upper bound of this interval is an equivalent of five-years of 2-month summer jobs. 16 Hotz et al. (2002) show that estimated returns from working while in high school or college dramatically diminish, when a dynamic selection model is used. 34 ECONOMIC AND BUSINESS REVIEW 2021;23:26e39 propensity score matching procedure balances all the included variables in the logit regression. In Table 6, we report the estimated WACDs, corre- sponding to the theoretical effects of student work on the probability of employment described in Table 5. The estimated effects are mostly positive and seem to exhibit diminishing returns to work experience, both one and two years after the entry on the labour market. Focusing on the estimated WACD 11 , WACD 12 and WACD 13 ,wefind that stu- dents in thefirst (second) year after entering labour market, who worked 10e24 months, 2e3 years, and more than three years during their study had, on average,9.3(5.6),5.9(9.7),and10.3(12.2)percentage points higher probability of being employed than comparable students (in terms of observed socio- economiccharacteristicsandacademicsuccess)who obtained less than 10 months of work experience, respectively. With one exception, gaining work experience beyond two years did not yield a statis- tically significant effect on the likelihood of employment.Thesizeoftheeffectsiscomparableto those reported in the existing studies (e.g. H€ akkinen, 2006; Hotz et al., 2002; Scott-Clayton & Minaya, 2016). 6.2 Related versus unrelated student work experience Next, we examine whether different types of stu- dent work affect student labour market outcomes differently. The effects are estimated on a subsample of students who used referrals issued by one of the student employment agencies (e- Studentskiservis)intheperiod2006e2010.Thetotal number of observations is 1186 and 983 in the first and the second year on the labour market, with approximately 70% of females. This data set con- tains information on the actual type of work per- formed by students, as e- Studentski servis distinguishes between more than 100 occupations and reclassifies them according to the International Standard Classification of Occupations (ISCO 1988). We sort these occupations into three groups: i) related high-skilled occupations (e.g. business ana- lysts, accountants, programmers), 17 ii) related but less-skilled occupations (e.g. office work, data preparation), 18 and iii) unrelated low-skilled occu- pations (e.g. serving tables). 19 Since we do not observetheentireemploymenthistoriesofstudents who used the services of this agency, we cannot estimatethe relations in thesame manneras shown in Table 6. Instead, we compare students who per- formed at least some hours of related less-skilled work and related high-skilled work, with those who performed only unrelated, low-skilled work. Then we compare those with related less-skilled student work experience with those that had related high- skilled work experience. In the estimation of pro- pensity scores, we add the amount of student work experience as an additional control variable. The summary statistics of the subsample in Table 7 reveals that, compared to the full sample, the average student work experience is lower. This can Table 6. Estimated WACDs between student work and probability of employment. Student work experience 10e24 months 2e3 years More than 3 years 1st Year Student work experience less than 10 months 0.093** 0.059 0.103* (0.027) (0.034) (0.041) 10e24 months 0.031 0.006 (0.026) (0.030) 2e3 years 0.038 (0.032) 2nd Year Student work experience less than 10 months 0.056** 0.097** 0.122** (0.022) (0.029) (0.033) 10e24 months 0.023 0.044* (0.019) (0.021) 2e3 year 0.032 (0.023) Note: *p < 0.05;**p < 0.01. Standard errors are reported in parentheses. 17 ISCO broad categories 1 and 2. 18 ISCO broad categories 3, 4, 5, and 6. 19 ISCO broad categories 7, 8, and 9. ECONOMIC AND BUSINESS REVIEW 2021;23:26e39 35 be attributed to the observation period, which put moreweightontheyearsduringthefinancialcrisis, which decreased student work hours (see Bartolj et al., 2015). Furthermore, persons in the subsample had poorer academic results (lower graduation rates, a higher number of exam attempts) and highernon-labourincome.Weattributethelatterto the concentration of the e- Studentski servis branch network in the wealthier parts of the country. Complementing the results of Geel and Backes- Gellner (2012),wefind (see Table 8) that related high-skilledstudentworkwasassociatedwitha15.8 (10.7) percentage points higher probability of employment in the first year on the labour than unrelated low-skilled (related less-skilled) student work. On the other hand, we find no statistically significant effects in the second year on the labour market, which suggests that the effects of the different types of student jobs are temporary. 6.3 Academic performance and probability of employment The last set of results provides evidence on the effect of academic performance on labour market outcomes. In this study, we concentrate on two measuresegraduation and GPAewhich we believe are most likely to be observed by employers in the selection process of regular workers, and may thus have the highest impact on the post-college labour market outcomes. More specifically, we compare students with a similar socio-economic background and student work experience, but different GPA rank and graduation status. We define two types of treatment: (i) GPA in the first quarter of the distri- bution and (ii) student has graduated. In our data, the average GPA in the top quartile was 7.86, whereas the average GPA below the 75 percentile was 6.45. Before turning to results, we must point out that propensity score matching does not balance all the conditioning variables in the logit regression. Nevertheless, after matching, the balancing prop- erty is significantly improved. Namely, in the esti- mation of the average graduation effect on graduates, there are 15 unbalanced variables in the unmatchedsampleandonly2unbalanced variables in the matched one (out of 43 variables). Further- more, the differences in the mean values of vari- ables between compared groups are significantly lower after matching. The results in Table 9 show that graduation was associated with 28.6 and 21.7 percentage points higher probability of employment in the first and second year on the labour market, respectively. The differences in employment likelihood of persons with different GPA were smaller in size. We find that persons with an above average GPA had a 9.0 percentagepointshigherprobabilityofemployment Table 8. Estimated relation between different types of student work and probability of employment. 1st year 2nd year Related less-skilled Related high-skilled Related less-skilled Related high-skilled Unrelated, low-skilled 0.038 0.158* 0.001 0190 (0.034) (0.057) (0.029) (0.052) Related, less-skilled 0.107* 0.006 (0.049) (0.039) Note: Standard errors are reported in parentheses. *p < 0.05. Table 7. Summary statistics for the subsample with information on the type of student work. 1st Year 2nd Year Mean Sd Mean Sd Employed after college 0.659 0.474 0.859 0.349 Hourly gross wage after college 4.391 6.684 6.852 11.063 Student work experience in years 1.373 0.898 1.452 0.920 Graduated 0.593 0.492 0.743 0.437 Time to final year 4.582 0.774 4.601 0.789 No. of exam attempts 55.899 13.071 55.713 13.011 Avg. grade 6.761 0.736 6.768 0.737 Age at enrolment 18.908 0.393 18.894 0.402 High school GPA 0.486 0.156 0.496 0.155 University or higherdmum 0.221 0.415 0.221 0.415 University or higherddad 0.232 0.422 0.229 0.420 Family business 0.145 0.352 0.130 0.337 Step parent 0.225 0.418 0.233 0.423 No. of siblings 0.770 0.741 0.777 0.733 Student parent 0.004 0.077 0.002 0.045 Non-labour income 8487 6623 8381 6514 Conditional income share 0.135 0.229 0.143 0.238 Capital income share 0.047 0.095 0.046 0.095 Expected net wage 15.944 2.522 15.722 2.463 Year 2008.4 1.2 2008.8 1.1 Note: All income-related variables are in constant (2004) Euros. The exchange rate in 2004 was 1 EUR ¼ 1.24 USD. Variables describingfamily characteristics and economicconditions during studies are measured in the final year of study. 36 ECONOMIC AND BUSINESS REVIEW 2021;23:26e39 in the first year than those who had a GPA below the 75th percentileea much weaker effect than the one found by Piopiunik et al. (2018). While these results are informative for the SEB UL, it is impor- tanttonotethattheseeffectsclearlydonottranslate to other study programs and other countries, due to the differences in the curricula and preferences of employers. 6.4 Comparison of the two effects Directcomparisonoftheeffectofstudentworkon the probability of employment and the effect of ac- ademic performance on the probability of employ- ment is only meaningful, if both input and output variables are measured in the same units, rewards have the same temporal structure and risks are similar. Clearly, this is not the case for these two typesofinvestmentsinhuman capitalethey require different amounts of time and returns are likely to exhibit different risks. These differences prevent us from making any statements regarding an optimal allocation of time between the two types of investments. Nevertheless, some guidelines can be given even onthebasisoftheseresults.Theinvestigationofthe effect of student work on the probability of employment shows that it is worth working 10 monthsormore.Itdoesnot,however,payoffforthe student to acquire more than 2 years of student work experience. In other words, for 10 months to 2 years of effort put into student work, persons could gain an approximately 9 percentage point increase intheprobabilityofemployment.Asimilareffecton the chances of getting a job is associated with ranking in the top 25% of the class, according to the GPA. However, this kind of academic success re- quires 4 school years invested in studying. There- fore, it seems that the highest payoff could be obtained by graduating, after students pass all the exams.Thewritingandthedefenceofthethesisare supposed to take one year and are associated with a more than 20 percentage points increase in the probability of employment. 7 Conclusions In this paper, we estimate the effects of two types of investment in human capitalestudent work and academic performanceeon the probability of employment, by comparing the labour market outcome of persons with similar socio-economic characteristics and human capital that differed only intheinvestigateddependentvariable.Wefindthat both types of investments positively and signifi- cantly affect the likelihood of regular employment. Our results show consistent benefits of increasing work experience from less than 10 months to 10e24 months during undergraduate study, especially when it is high-skilled work in occupations related to college major. Additional student work experi- encehaspositive effects,butthesizeofthese effects is small and typically statistically insignificant, which gives support to the policies that limit the amount of student work. Although direct compari- son of the estimated relations is not possible due to the variation of inputs, time structure of rewards and riskiness, our rule of thumb comparison sug- gests that the preparation of the final thesis was associated with higher gains than any amount of student work. Although we base our results on the data from a single educational organization that is located in a small economy, we believe that our work bears relevance for other countries. Slovenia is an open economywithintheEuropeanUnion,andweexpect its employers to be influenced by similar economic forces as their counterparts in other EU economies. While the two types of investments in human cap- ital might have different returns in different coun- tries, employers should exhibit broadly similar relative preferences to those reported in our anal- ysis.However,employersmayattachdifferentvalue tostudentworkexperienceinjobsfilledbystudents of other study programs (e.g. jobs in IT) that we do not analyse in this research, and thus, the relative sizes of the impacts of student work and academic performance might be different for those students. Furtherresearchshouldtrytoovercometwomain limitations of this study. Firstly, while we were able to control for a rich set of variables that are corre- lated with the preferences for working after college (such as family income, parental education, high- school performance), our estimations might be biased, if there were significant differences in the inclination to work after college among for example those who graduated and those who did not. In addition, our data do not allow us to observe the actual hours of student work. And even though Bartolj and Polanec (2018) show that hours Table 9. Estimated relation between academic performance and proba- bility of employment. 1st Year 2nd Year Graduated 0.286** ,2 0.217** ,6 (0.022) (0.025) GPAinthe75thpercentile or higher 0.090** ,1 0.028 (0.025) (0.019) Note 1 6 : Number indicates how many of the 48 explanatory vari- ablesarenotbalancedbetweencomparedgroupsatp<0.01.**p< 0.01.Standarderrorsarereportedinparentheses. ECONOMIC AND BUSINESS REVIEW 2021;23:26e39 37 calculated in the manner that is employed in this research is a good enough proxy, further analysis should try toestimatethese effects by applyingdata on the actual hours of student work. Funding The authors disclosed receipt of the following financial support for the research, authorship, and/ or publication of this article: This work was sup- ported by the Slovenian Research Agency [grant number P5-0096]. Data availability statement Thedatathatsupportthefindingsofthisstudyare available in the secure room of the Statistical Office oftheRepublicofSlovenia.Restrictionsapplytothe availability of these data. For more details see https://www.stat.si/StatWeb/en/StaticPages/Index/ For-Researchers. References Angrist,J., & Pischke,J.-S. (2009). Mostly harmless econometrics: An empiricist's companion. Princeton (NJ): Princeton University Press. Arrow, K. J. (1973). Higher education as a filter. Journal of Public Economics, 2(3), 193e216. Auers, D., Rostoks, T., & Smith, K. (2007). Flipping burgers or flipping pages? Student employment and academic attain- ment in post-Soviet Latvia. Communist and Post-Communist Studies, 40(4), 477e491. Austin, P. C. (2011). Optimal caliper widths for propensity-score matching when estimating differences in means and differ- ences in proportions in observational studies. Pharmaceutical Statistics, 10(2), 150e161. Baert,S., Rotsaert, O., Verhaest, D., & Omey, E. (2015). A signal of diligence? Student work experience and later employment chances. Technical Report. IZA DP No. 9170. Bartolj, T., Ahcan, A., Feldin, A., & Polanec, S. (2013). Evolution of private returns to tertiary education during transition: Evidence from Slovenia. Post-Communist Economies, 25(3), 407e424. Bartolj, T., Feldin, A., & Polanec, S. (2015). Does low family income deter college students from supplying labour? Mimeo. Bartolj, T., & Polanec, S. (2012). College major choice and ability: Why is general ability not enough? Economics of Education Review, 31(6), 996e1016. Bartolj, T., & Polanec, S. (2018). Does work harm academic per- formance of students? Evidence using propensity score matching. Research in Higher Education, 59(4), 401e429. Bartolj, T., & Polanec, S. (2020). Does low family income deter college students from supplying labor? Mimeo. Becker, G. S. (1962). Investment in human capital: A theoretical analysis. Journal of Political Economy, 70(5, Part 2), 9e49. Becker,G.S.(1964). Human capital.NewYork:NationalBureauof Economic Research. Becker, G. S. (1965). A theory of the allocation of time. Economic Journal, 493e517. Becker, G. S. (1993). Human capital: A theoretical and empirical analysis, with special reference to education. Chicago: University of Chicago Press. Beerkens, M., M€ agi, E., & Lill, L. (2011). University studies as a side job: Causes and consequences of massive student employment in Estonia. Higher Education, 61(6), 679e692. Belman, D., & Heywood, J. S. (1991). Sheepskin effects in the returns to education: An examination of women and minor- ities. The Review of Economics and Statistics, 720e724. Betts, J. R. (1995). Does school quality matter? Evidence from the national longitudinal survey of youth. The Review of Economics and Statistics, 231e250. Cadez, S., Dimovski, V., & Zaman Groff, M. (2017). Research, teaching and performance evaluation in academia: The salience of quality. Studies in Higher Education, 42(8), 1455e1473. Callender, C. (2008). “The impact of term-time employment on higher education students' academic attainment and achievement. Journal of Education Policy, 23(4), 359e377. Card, D., & Krueger, A. B. (1992). School quality and black-white relative earnings: A direct assessment. Quarterly Journal of Economics, 107(1), 151e200. Darolia, R. (2014). Working (and studying) day and night: Het- erogeneous effects of working on the academic performance of full-time and part-time students. Economics of Education Review, 38,38e50. Dehejia, R. H., & Wahba, S. (1999). Causal effects in nonexperi- mental studies: Reevaluating the evaluation of training pro- grams. Journal of the American Statistical Association, 94(448), 1053e1062. Deming,D.J.,Yuchtman,N.,Abulafi,A.,Goldin,C.,&Katz,L.F. (2016). The value of postsecondary credentials in the labor market:Anexperimentalstudy.TheAmericanEconomicReview, 106(3), 778e806. DeSimone, J. S. (2008). The impact of employment during school on college student academic performance. National Bureau of Eco- nomic Research. NBER Working Paper No. 14006. Ehrenberg, R. G., & Sherman, D. R. (1987). Employment while in college, academic achievement, and postcollege outcomes: A summary of results. Journal of Human Resources, 22(1), 1e23. Ellwood, D. T. (1982). Teenage unemployment: Permanent scars ortemporaryblemishes?.InInTheyouthlabourmarketproblem: Its nature, causes, and consequences (pp. 349e390). University of Chicago Press. Farcnik, D., & Domadenik, P. (2012). Has the Bologna reform enhancedtheemployabilityofgraduates?Earlyevidencefrom Slovenia. International Journal of Manpower, 33(1), 51e75. Geel, R., & Backes-Gellner, U. (2012). Earning while learning: When and how student employment is beneficial. Labour, 26(3), 313e340. Gregg, P., & Tominey, E. (2005). The wage scar from male youth unemployment. Labour Economics, 12(4), 487e509. H€ akkinen, I. (2006). Working while enrolled in a university: Does it pay? Labour Economics, 13(2), 167e189. Heckman, J. J., Lochner, L. J., & Todd, P. E. (2006). Earnings functions, rates of return and treatment effects: The Mincer equation and beyond. In E. Hanushek, & F. Welch (Eds.), Handbook of the economics of education (pp. 307e458). Amster- dam: North-Holland. Hotz, V. J., Xu, L. C., Tienda, M., & Ahituv, A. (2002). Are there returns to the wages of young men from working while in school? The Review of Economics and Statistics, 84(2), 221e236. Hungerford, T., & Solon, G. (1987). Sheepskin effects in the returns to education. The Review of Economics and Statistics, 175e177. Inceoglu, I., Selenko, E., McDowall, A., & Schlachter, S. (2019). (How) do work placements work? Scrutinizing the quantita- tive evidence for a theory-driven future research agenda. Journal of Vocational Behavior, 110(B), 317e337. ISCO (1988) International standard classification of occupations. Url: https://www.ilo.org/public/english/bureau/stat/isco/ Jones, E. B., & Jackson, J. D. (1990). College grades and labour market rewards. Journal of Human Resources, 25(2), 253e266. Kalenkoski,C.M.,&WulffPabilonia,S.(2010).Parentaltransfers, student achievement, and the labour supply of college stu- dents. Journal of Population Economics, 23(2), 469e496. Kosi-Antolic, T., Nastav, B., & Sustersic, J. (2013). Does student employment deteriorate academic performance? The case of Slovenia. Revija za Socijalnu Politiku, 20(3), 253e274. 38 ECONOMIC AND BUSINESS REVIEW 2021;23:26e39 Lee, M.-J. (2005). Micro-econometrics for policy, program, and treat- ment effects. New York: Oxford University Press. Light, A. (1999). High school employment, high school curricu- lum, and post-school wages. Economics of Education Review, 18(3), 291e309. Light, A. (2001). In-school work experience and the returns to schooling. Journal of Labor Economics, 19(1), 65e93. Mroz,T.A.,&Savage,T.H.(2006).Thelong-termeffectsofyouth unemployment. Journal of Human Resources, 41(2), 259e293. Nilsen, Ø., & Reiso, K. (2011). Scarring effects of unemployment. NHH Dept. of Economics Discussion Paper No. 26/2011. NordstromSkans,O.(2011).Scarringeffectsofthefirstlabourmarket experience. IZA Discussion Paper No. 5565. Oreopoulos, P., Von Wachter, T., & Heisz, A. (2012). The short- and long-term career effects of graduating in a recession. American Economic Journal: Applied Economics, 4(1), 1e29. Pinto, L. H., & Ramalheira, D. C. (2017). Perceived employability of business graduates: The effect of academic performance and extracurricular activities. Journal of Vocational Behavior, 99, 165e178. Piopiunik, M., Schwerdt, G., Simon, L., & Woessmann, L. (2018). Skills, signals, and employability: An experimental investigation. CESifo working paper series No. 6858. Psacharopoulos, G., & Patrinos, H. A. (2018). Returns to invest- mentineducation:Adecennialreviewofthegloballiterature. Education Economics, 26(5), 445e458. Ruhm, C. J. (1997). Is high school employment consumption or investment? Journal of Labor Economics, 15(4), 735e776. Ryan, P. (2001). The school-to-work transition: A cross-national perspective. Journal of Economic Literature, 39(1), 34e92. Available at: http://www.aeaweb.org/articles?id¼10.1257/jel. 39.1.34. Schmillen, A., & Umkehrer, M. (2017). The scars of youth: Effects of early-career unemployment on future unemployment experience. International Labour Review, 156(34), 465e494. Scott-Clayton, J., & Minaya, V. (2016). Should student employ- ment be subsidized? Conditional counterfactuals and the outcomes of work-study participation. Economics of Education Review, 52,1e18. Available at: http://www.sciencedirect.com/ science/article/pii/S0272775715000825. Spence, M. (1973). Job market signaling. Quarterly Journal of Eco- nomics, 87(3), 355e374. SSC Act. (2001). “Social security contributions act”,of ficial Gazette of theRS[Zakonoprispevkihzasocialnovarnost.UradnilistRS]no. 5/1996, 97/2001, 34/1996. Stinebrickner, R., & Stinebrickner, T. R. (2003). Working during school and academic performance. Journal of Labor Economics, 21(2), 473e491. Vooren, M., Haelermans, C., Groot, W., & Maassen van den Brink, H. (2018). The effectiveness of active labor market policies: A meta-analysis. Journal of Economic Surveys, 33(1), 125e149. ECONOMIC AND BUSINESS REVIEW 2021;23:26e39 39