Sustainable Education: Analyzing the Determinants of University Student Dropout by Nonlinear Panel Data Models

Transcription

1 susainabiliy Aricle Susainable Educaion: Analyzing he Deerminans of Universiy Suden Dropou by Nonlinear Panel Daa Models Donggeun Kim 1 and Seoyong Kim 2, * 1 Deparmen of Economics, Ajou Universiy, Worldcup-ro 206, Suwon 16499, Korea; kimdongg@ajou.ac.kr 2 Deparmen of Public Adminisraion, Ajou Universiy, Worldcup-ro 206, Suwon 16499, Korea * Correspondence: seoyongkim@ajou.ac.kr; Tel.: Received: 19 January 2018; Acceped: 20 March 2018; Published: 25 March 2018 Absrac: Universiy dropou is a serious problem. I affecs no only he individual who drops ou bu also he universiy and sociey. However, mos previous sudies have focused only on he subjecive/individual level. Universiy dropou is a very imporan issue in Souh Korea, bu i has no received much research aenion so far. This sudy examined he possible causes of universiy dropou in Souh Korea a he aggregae level, focusing on four fundamenal caegories: sudens, resources, faculy, and universiy characerisics. Three-year balanced panel daa from 2013 o 2015 were consruced and esimaed by using nonlinear panel daa models. The findings show ha cos and burden for sudens, financial resources, qualiaive and quaniaive feaures of faculy, and ype/size of he universiy have significan effecs on universiy dropou. Keywords: susainable educaion; dropou; universiy suden; deerminans of dropou; wihdrawal 1. Inroducion Educaion has played an essenial role in achieving rapid economic growh in Souh Korea. Higher educaion insiues such as colleges or universiies have been he main source of good-qualiy human capial and have boosed social mobiliy. I can easily be proven boh hisorically and saisically ha here exiss a very close posiive correlaion beween economic growh and higher educaion aainmen; Souh Korea is no excepion. According o he OECD [1], Korea is ranked fourh (45%) for year-olds and firs (69%) for year-olds in eriary educaion among 35 OECD counries. Moreover, i has seen a rapid increase in educaion aainmen, from 51% in 2010 o 69% in The high level and rapid growh of educaional aainmen seem o be necessary condiions for accumulaing human capial a he sociey level. However, he high level of paricipaion in eriary educaion does no ell us he whole sory. No all freshmen graduae from universiy and earn heir bachelor s degree. There are many reasons ouside of he inabiliy o cach up ha preven sudens from graduaing; hese can be dichoomized ino reasons from he sudens side and hose from he universiies side. The former includes a lack of full informaion on and saisfacion wih universiy educaion or a lack of hope for a brigh fuure. The laer includes poor-qualiy faculies, lower faciliy levels, and rickey financial saus, which usually induce dissaisfacion in sudens. No maer how a reason originaes, i can push sudens o drop ou. Table 1 shows he increasing dropou raes of he op 19 universiies among all 214 four-year universiies in Souh Korea. The average dropou rae increased from 1.17% in 2010 o 2.099% in I has almos doubled. This new phenomenon could be parly explained by a simple cos benefi analysis. During he same ime period, he GDP growh rae moved in exacly he opposie direcion; i was 6.5% in 2010 and 2.6% in The average GDP growh rae was 4.14% from 2006 o 2010 Susainabiliy 2018, 10, 954; doi: /su

2 Susainabiliy 2018, 10, of 18 and 2.96% from 2011 o If he cos of finishing one s universiy educaion exceeds he expeced benefi of earning a degree, here would be lile reason o say in school, even a a op-level universiy. Unil very recenly, enering a op-level universiy was regarded as walking on a royal road o social success, including geing a good job and gaining higher social saus in Souh Korea, which would be one reason why he counry mainains such a high educaion aainmen rae. However, his opimisic viewpoin was shaered as Souh Korea began o suffer from low economic growh and a high unemployed rae. As explored laer in Secion 3, he average employmen rae of universiy graduaes has consisenly decreased, from 62.72% in 2013 o 53.08% in As Table 1 shows, all universiies excep for UNIST show an upward rend in dropous. Table 1. Main universiy dropou raes No. Universiy Dropou Dropou Dropou Enrollmen Dropou Enrollmen Dropou Enrollmen Dropou Rae Rae Rae 1 Ewha 18, % 17, % 17, % 2 Sookmyung 13, % 13, % 12, % 3 Sogang 11, % 11, % 11, % 4 KonKuk 22, % 23, % 22, % 5 GIST % % 6 UNIST % % % 7 Sungkyungwan 26, % 28, % 27, % 8 Chung-ang 21, % 20, % 19, % 9 Hongik 16, % 16, % 16, % 10 Hanyang 21, % 22, % 22, % 11 Dongguk 19, % 20, % 19, % 12 Univ. of Seoul 13, % 12, % 12, % 13 HUFS 23, % 12, % 12, % 14 Kyunghee 34, % 17, % 17, % 15 KAIST % % % 16 Korea 27, % 28, % 28, % 17 S.N.U. 21, % 21, % 22, % 18 Yongsei 26, % 26, % 26, % 19 POSTECH % % % Toal or Mean 328, % 301, % 296, % Rapidly growing dropou raes can also be found in oher counries. Based on he Naional Educaional Longiudinal Sudy of 1988 (NELS: 88), Bound e al. [2] found ha among he classes of 1972 and 1992 high school cohors, eigh-year college compleion raes decreased naionally by 4.6 percenage poins, from 50.5% o 45.9%. The snapsho repor from he Naional Suden Clearinghouse issued on 27 June 2016, says ha for he academic years, 536,351 were earned by sudens wih no previous degrees or cerificaes. Wihin he nex six academic years, over 64% of hese sudens enrolled a a four-year insiuion and 41% earned a bachelor s degree. Based upon his repor, overall more han 30% were no enrolled and can be regarded as college dropous. So, why is he dropou rae of universiy sudens so imporan? Larsen e al. [3] describe he significan consequences of universiy dropou a he individual, universiy, and socieal levels. A he individual level, universiy dropou enails boh psychological and physical coss. Sudens may suffer from depression arising from feelings of inadequacy and self-doub, all of which are relaed o dropping ou. Moreover, hey will be aware of wasing personal resources, effor, ime, and money. A he universiy level, dropous have economic and academic consequences. From he academic poin of view, universiy dropous no only represen a failure o adap o he college life and sysem bu also signify a red ligh o he educaion sysem in regard o providing appropriae services for sudens. From he economic poin of view, he more dropous, he worse he financial sae of he universiy. A he socieal level, he socioeconomic impac of universiy dropou can never be ignored because he supply of universiy graduaes significanly influences boh he reurns o educaion and overall economic growh [2]. Gio e al. [4] commen ha weak universiy-level educaion may have negaive effecs no only on sudens bu also on economic growh and sociey as a whole. Therefore, i is imporan o analyze he causes of universiy dropou. Alhough here have been large, exensive sudies of universiy dropou raes, relaively few analyzed dropou a he aggregaed

3 Susainabiliy 2018, 10, of 18 level. The main novely of he presen research is ha we esed a more inegraed model ha included four facors from no only he demand-side bu also he supply-side deerminans of dropou. Few previous sudies have compared four facors (suden, faculy, resource, and srucural facors) in one model. Such an approach would allow us o pursue a sysemaic and horough undersanding of universiy dropou. In paricular, here have been few sysemaic sudies on universiy dropou in Souh Korea. Our sudy, mainly focusing on objecive/supply side (more han demand side) variables, aims o implemen an empirical sudy o analyze possible facors relaed o universiy dropou wih aggregae-level hree-year panel daa on universiies in Souh Korea. The remainder of he paper is organized as follows: in Secion 2, previous sudies are summarized and he heoreical background for universiy dropou is provided. In Secion 3, an economeric model for examining imporan variables affecing universiy dropou is inroduced. In Secion 4, he daa are described and he empirical findings are repored. Secion 5 summarizes he empirical findings. Finally, Secion 6 conains concluding remarks and policy implicaions. 2. Theoreical Background 2.1. Research Trends Relaed o Dropou a he Universiy Level The erm universiy dropou is commonly used o describe siuaions where a suden leaves he universiy in which (s)he has enrolled before having obained a formal degree. The erms used o describe universiy dropou from a suden s perspecive are many: dropou, deparure, wihdrawal, academic failure, nonconinuance, and noncompleion. Their posiive counerpars include persisence, coninuance, compleion, ec. [3] (p. 5). Dropou akes place for various reasons. Spady [5] and Tino [6] argue ha a suden s persisence/wihdrawal decision is he resul of a long series of ineracions beween he suden and he academic and social sysems of he insiuion. Lam [7] poins ou he various reasons for dropou suden saus, residence, financial sources, disance of homeown from he universiy, goal fulfilmen, and saisfacion wih he overall universiy amosphere which are useful in making predicions. Also, according o an exensive lieraure review by Larsen e al. [3], various facors explain dropou from universiy: he suden s sociodemographic background, academic compeencies, moivaions for sudying, social and academic inegraion a universiy, and living condiions. There have been exensive effors o analyze he deerminans of universiy dropou. Those sudies can be divided ino four rends. Firs, as a convenional approach, a lo of sudies on universiy dropou ended o focus more on he demand (suden) side raher han he supply (universiy) side. According o Gio e al. [4], dropou sudies have heavily focused on demand-side characerisics such as sudens prior educaional achievemens (before enering universiy), gender, age, moivaion, financial consrains, sociodemographic background, and individual circumsances. Using pah analysis, Bean [8] finds ha college grades, insiuional fi, and insiuional commimen are imporan inervening variables in dropou syndrome. He also poins ou ha he supply-side deerminans of dropou (e.g., organizaional condiions wihin universiies) have only recenly sared receiving aenion. Moreover, he shows ha a suden s peers are more imporan agens of socializaion han informal faculy conacs and ha a suden may play a more acive role in socializing wih his peers han previously perceived. Heublein e al. [9,10] include pre-universiy facors in explaining universiy dropou. Those facors comprise he suden s financial siuaion/living condiions, advice/suppor from friends/family and oher opporuniies for counseling, as well as he suden s own fuure plans. Second, previous sudies have focused more on subjecive or psychological facors a he individual level han on objecive variables a he aggregae or universiy level. For example, Bean [8] focuses on individual-level variables such as college grades, insiuional fi, and insiuional commimen. He shows ha college grades seem more a produc of selecion han of socializaion. Moreover, according o Spady [5], psychological facors such as insiuional commimen and

4 Susainabiliy 2018, 10, of 18 perceived social inegraion (i.e., a subjecive sense of belonging and fiing in) have a sronger influence on he decision o drop ou in one s firs year. Third, compared o he numerous sudies focusing on he demand/subjecive side a he individual level, relaively recen and few sudies have been examining he supply/objecive side a he aggregae level. However, insiuional supply-side facors also have an influence on dropou decisions. Larsen e al. [3] menion causal facors of dropou a he insiuional level, which concern he course/subjec of sudy, deparmen, faculy, universiy, or he enire sysem of higher educaion. Gio e al. [4] demonsrae ha some supply-side facors such as he srucure of universiy courses and reorganizaion of remoe branches have a significan impac on he likelihood of dropou occurring. Bound e al. [2] argue ha even hough he supply-side facors of higher educaion play an imporan role in explaining changes in suden oucomes, hey have ofen been disregarded. Lasly, a more recen rend for analyzing universiy dropou is o adop a more balanced view o reflec he supply/objecive side as well as he demand/subjecive one. There exis several previous sudies ha adop a more balanced model. Based on he Suden Inegraion Model, Tino [6] emphasizes he ineracions beween individual sudens aribues and insiuional srucures wihin universiies. He argues ha no only suden facors bu also universiy facors influence dropou, as follows: Firs, sudens aribues such as family background, personal characerisics, and prior schooling influence his/her abiliies and prerequisies for sudy a a universiy. They provide he resources for adapaion o he suden lifesyle. Second, universiy facors consis of academic and social sysems. The firs is concerned wih academic performance and ineracions wih professors/saff. The laer is abou exracurricular aciviies and social relaions. Gio e al. [4] commen ha even hough universiy dropou may be explained by he supply-side characerisics of he universiy as well as by sudens individual characerisics, many sudies have focused only on he laer facors [2,7,8]. Based on survey daa, Pierrakeas e al. [11] show ha no only inrinsic (suden-relaed) facors such as sickness, work/school conflics, ec., bu also exrinsic (insiuion-relaed) facors such as sudy mehods and maerials, educaional approach, and uors influence dropou a he universiy level. The research rend seems o move from a unilaeral view considering demand-side/individual level or supply-side/aggregae level o comprehensive or bilaeral view such as a balance model. When analyzing universiy dropou, we will mainain he balanced view and focus on no only demand-side suden facors bu also supply-side insiuional universiy facors. Furhermore, for more deph sudy, we divide he supply-side ino hree fundamenal caegories (i.e., faculy, resource, and srucural facors) and sugges a modified balanced model. Our main research quesion concerns which facors criically conribue o explaining dropou a he universiy level. Nex, we will review previous findings relaed o hese facors and presen he research hypoheses Suden Facors Every year, many sudens wih differen backgrounds, characerisics, and culures go o universiy. There are many facors ha affec dropou decisions. Various ypes of suden characerisic have some influence on hese decisions. Among hese various suden facors, we will concenrae on wo: suden qualiies and coss/benefis. The former concerns he abiliy, compeence, moivaion, and performance of a suden. The laer relaes o he financial cos and he expeced or realized benefis ha a suden carries on or receives afer enering universiy. Firs, he qualiy of sudens influences how hey will be inegraed ino or drop ou of a universiy s social and academic sysems. Tius [12] demonsraes wheher or no precollege academic performance has an impac on suden persisence in universiy. He shows ha hose who have a high SAT score demonsrae more persisence. Also, Asin [13] regards high-achieving sudens as a useful resource because large numbers of such sudens enhance he qualiy of he learning environmen for all sudens. Simpson e al. [14] show ha good performance, such as a higher GPA, decreases failing dropous and ransfers, and ha higher SAT scores also reduce failing ransfers. Gio e al. [4] corroborae he inuiive proposiion ha sudens who do well in high school are more

5 Susainabiliy 2018, 10, of 18 likely o succeed a universiy and, as a resul, are less likely o drop ou. Bound e al. [2] repor ha preparaions for enering college (as measured by mah es scores) by a suden explain one-hird of he observed decline in compleion raes. Moreover, Spady [5] shows ha hose who have academic poenial show higher graduaion raes. According o a lieraure review by Larsen e al. [3], 14 ou of 22 sudies found ha higher (upper) secondary school marks significanly lowered he risk of dropou. Pierrakeas e al. [11] repor ha hose who are no qualified enough (i.e., do no have he required knowledge for a specific course) o pursue universiy-level sudies are fel o be dropous. Second, he coss and benefis ha sudens face a universiy influence heir dropou decisions. In erms of coss, financial burdens such as uiion, scholarships, and suden loans criically influence dropou decisions. Li and Killian [15] examine paerns of ariion a a midwesern research universiy and repor ha among he mos ofen proffered reasons for leaving, financial facors are he mos imporan ones in relaion o persisence in higher educaion. If he uiion is high, more sudens will feel psychological and economic hardships, hereby negaively affecing heir dropou decision and paricipaion in higher educaion. Financial aid from universiies is closely relaed o dropou. According o Ishiani e al. [16], generally hose who receive financial aid show lower dropou raes han nonaided sudens. Moreover, afer esimaing he effec of financial aid on dropou from and compleion of a five-year universiy degree in Denmark, Arend [17] shows ha an increase in grans decreases he dropou rae bu has no overall impac on he compleion rae. In erms of benefis, expeced or realized benefis will reduce he dropou rae. Sudens will no leave universiy if hey can envision he prospecive fuure ha awais afer hey graduae: for example, geing a good job, enering a beer graduae school, and having oher chances in heir personal life. Simpson e al. [14] show ha among sudens of good sanding, hose who have high vocaional orienaion have a lower probabiliy of wihdrawing Faculy Facors A faculy is a buffering resource ha makes conribuions o reducing suden dropou a a universiy. Spady [5] and Tino [6] sress ha social and academic inegraion is criical influencing facors on suden persisence. Pascarella e al. [18] argue ha he faculy suden ineracion is he key facor in he academic and social inegraion of sudens. Such conacs have a srong impac on suden ariion. Based on daa colleced from sudens a a privae universiy, hey also argue ha academic and social conacs wih faculy a some insiuions, paricularly privae ones, influence ariion-relaed conen more han hose a oher insiuions. Pascarella and Parick [19] show ha a faculy has wo inegraion funcions: academic and social ones. The firs relaes o faculies concern for eaching and suden developmen and sudens informal conac wih faculy for he sake of inellecual discussions. The second is abou informal relaions wih a faculy; for example, helping o resolve a personal problem. Such ineracions have a significanly negaive impac on college ariion. If he quaniy and qualiy of professors are enhanced, here will be beer roles for hem. The former is abou he volume of faculy (e.g., he number of professors, he suden faculy raio, and how many classes are augh by professors) whereas he laer is abou he faculy s research compeences (i.e., he number of publicaions and research funding). Relaed o he quaniy of faculies, Asin [13] explains ha he lower he suden faculy raio becomes, he greaer he learning and personal developmen, because reduced eaching load encourages professors o spend more ime and energy supporing sudens. Gio e al. [4] find ha he raio of he number of lecurers o newly enrolled sudens is negaively correlaed o dropou. This implies ha he qualiy of eaching has some obvious influences on dropou decisions. According o Bound e al. [2], an increase in he suden faculy raio explains a large porion of he oal observed compleion rae reducion. Educaion is affeced no only by quaniaive aspecs bu also by qualiaive ones: Educaional environmens can be improved by bringing in beer or higher-qualiy professors. Our research assumes ha higher-qualiy professors wih many good publicaions and significan research funds

6 Susainabiliy 2018, 10, of 18 help o boos he qualiy of educaion o a more desirable level, leading o a decrease in he dropou rae. However, previous sudies have no explicily examined such qualiaive aspecs of faculies Resource Facors Pleniful resources ha universiies can uilize provide he fundamenal srucure for beer educaion. Gio e al. [4] and Bound e al. [2] argue ha even hough he supply side of higher educaion plays an imporan role in explaining changes in sudens oucomes, is imporance has been overlooked. Bound e al. [2] show ha resources ha are available wihin hese insiuions are key facors in influencing college compleion in he Unied Saes. They demonsrae ha a decrease in insiuional resources explains abou one quarer of he observed compleion rae decline. According o Asin [13] (p. 520), hese resources include a wide range of ingrediens believed o enhance suden learning, such as fiscal resources (financial aid, endowmens, and exramural research funds) and physical faciliies (laboraories, libraries, and audiovisual aids). Our sudy focuses on wo resource facors: financial resources and faciliy level. The former includes he amoun of revenue and expendiure mainly based on uiion, funding from donaions, and grans from he governmen, whereas he laer concerns building a space for educaion. If universiies have enough resources from revenues, higher educaion will afford more qualiaive services. Tius [12] analyzes he effec of revenue and expendiure paerns. He shows ha an increased percenage of resources from uiion, no from sae appropriaions or grans, will increase suden persisence. On he conrary, a higher percenage spen on adminisraion (no on insrucion, suden services, grans and scholarships, or research) decreases i. Moreover, if here is a higher percenage of revenues based on governmen subsidies, hen paricipaion in higher educaion will be higher. Governmen grans or subsidies play some role in expanding access hrough lowering uiion levels. Gross e al. [20] examine he effecs of insiuional financial aid on he persisence of universiy undergraduae and graduae sudens. They find ha insiuional financial aid from public universiies has a posiive bu modes effec on persisence. Ineresingly, he effecs of aid are greaer for men han for women, all else being equal. The physical condiions a a universiy, such as he age and size of faciliies and he qualiy of he classrooms, have an influence on dropou. According o Sewasew [21], a lack of faciliies is one of he major challenging facors in he universiy environmen and has been idenified as one reason for dropou. Bowers and Burke [22] compare schools in erms of age and find ha sudens a a modern school have higher aendance raes han hose a an older faciliy. However, McGowen [23] shows ha school faciliies influence discipline and behavior bu no achievemen, aendance, or compleion rae Srucural Characerisics of Universiies: Type, Size, and Locaion Much like how sociodemographic variables a he individual level such as gender, residence, age, and social saus affec universiy dropou, characerisics a he universiy level such as ype, size, and locaion also have some effecs on i. Firs, Tius [13] shows ha he ype (privae versus public), locaion (rural versus urban), and size (number of enrollmens) of a universiy do no influence sudens persisence. Bound e al. [3] menion he criical role of insiuion ype: collegiae characerisics have more of an influence on dropou. Bound e al. [3] demonsrae ha he kind of college ha sudens aend is an imporan facor in he changes in college compleion in he Unied Saes. Also, Gio e al. [4] show ha he dropou raes a public universiies appear higher han privae ones. Second, he size of he universiy influences he qualiy of he educaion as well. In he educaion field, he principle small is beauiful is generally and empirically validaed. The smaller he school, he more efficien is educaional oucomes. Piman and Haughwou [24] invesigae how school size affecs he dropou rae. They assume ha size no only indirecly influences he dropou rae bu also direcly affecs he diversiy of academic offerings and he school s social climae. They find ha he social climae creaes poenial links beween school size and he dropou rae.

7 Susainabiliy 2018, 10, of 18 Third, he locaion of he universiy for example, rural versus urban changes he dropou rae. Alhough Jordan e al. [25] show ha high school graduaion raes were very similar across he rural urban coninuum in he early 2000s, Pallas [26] finds ha urban sudens drop ou more frequenly. He repors ha a rural educaion is beer han ha obained in a congesed urban area because a rural environmen leads o fewer problems. Gio e al. [4] demonsrae ha universiies locaed in Norhern Ialy show lower dropou raes. The previous sudies show ha universiy dropou is closely relaed o he four facors which were described above. However, few sudies have considered he four facors in a model. The aim of his sudy is o inquire ino a sysemaic and horough undersanding of universiy dropou wih a more inegraed model. For his purpose, we adop a modified balanced model including he four facors, examine i wih he nonlinear panel daa regression models and es for he hypoheses as follows: H1: Dropou decreases when sudens are more qualified and more benefis are expeced from he universiy; i increases when sudens have o bear higher coss. H2: Increased qualiy and quaniy among faculy decreases universiy dropou. H3: As resource facors, beer financial and faciliy condiions reduce universiy dropou. H4: Dropou raes a public universiies are higher han a privae ones. The smaller he size of he universiy, he smaller he dropou rae. Universiies locaed in rural areas have less dropou han hose in urban areas. 3. Model Le {(X i, y i ) : i = 1, 2,..., N; = 1, 2,..., T} denoe he panel daa observaions where X i is a 1 P vecor of he explanaory variables and y i is he dependen variable. Le E(y i X i, a i ) be he condiional expecaion of y i given X i and a i where a i = e µ i is he unobserved effec. As he dependen variable is a coun variable ha akes on nonnegaive ineger values,y i 0, he mos popular parameric funcion is an exponenial funcion of X i and he P 1 vecor of parameers β 0 ; exp(x i β 0 ). If he unobserved and ime-consan effec a i is muliplicaive and independen of explanaory variables, E(y i x i, a i ) = a i exp(x i β 0 ). Assume ha y i X i, a i Poisson[a i exp(x i β 0 )] (1) y i and y is f or = s are independen condiional on X i and a i (2) a i = e µ i Gamma(δ 0, δ 0 ) and a i is independen of X i (3) The Poisson probabiliy specificaion is derived under he assumpions (1) (3): pr(y i X i, µ i ) = e λ ie µi (λ i e µ i) y i. (4) y i! The join densiy of (y i1, y i2,..., y it ) and µ i akes he form pr(y i1, y i2,..., y it, µ i X i1, X i2,..., X it ) = pr(y i1, y i2,..., y it X i1, X i2,..., X it, µ i )g(µ i ) = λ y i i y i! e eµi λ i µ i y i e g(µi ) (5)

8 Susainabiliy 2018, 10, of 18 where g(µ i ) is he probabiliy densiy funcion of µ i. The condiional densiy of µ i given X i is he same as he uncondiional densiy of µ i in Equaion (5). Since µ i is an unobservable random variable, i can be inegraed ou from Equaion (5) as follows: pr(y i1, y i2,..., y it X i1, X i2,..., X it ) = [ ] 0 λ y i i y i! e α i λ iα y i i f (α i )dα i = [ λ y i i y i! ] [ δ ] δ( λ i +δ λ i + δ ) ( y i Γ Γ(δ) ) y i +δ (6) where Γ( ) is he gamma funcion and Γ(z) = z 1 e for z > 0. Equaion (6) is he random effecs (RE) Poisson model suggesed by Hausman, Hall, and Griliches [27] (hereafer, HHG), and he parameers of he model in Equaion (6) can be consisenly esimaed via quasi-maximum likelihood esimaion. Le u i = Equaion (7): δ λ i +δ and λ i = exp(x i β). Then, Equaion (6) is rewrien as he likelihood funcion in pr(y i1, y i2,..., y it X i1, X i2,..., X it ) = ( λ y i i Γ ( y i! Γ(δ) ) y i +δ ) y i ui δ(1 u i) λ i y i (7) The esimaor of he RE Poisson model can be obained by maximizing he log likelihood funcion in Equaion (8). L(β) = N { ( log Γ i=1 δ + T ) y i T log Γ(1 + y i ) log Γ(δ) + δ log u i =1 =1 + log(1 u i ) T y i + T ( ) T T y i (x i β) y i log( exp(x i β) =1 =1 =1 =1 )} (8) where Γ(1 + y i ) = y i!. If assumpion (3) is violaed, hen an arbirary dependence beween X i and a i exiss, and he Poisson RE model can no longer be a consisen esimaor. HHG [27] conduced seminal work on his issue and sugges he fixed effecs (FE) Poisson esimae. Using n i T y i for each cross secion i, HHG [27] shows ha y i does no depend on a i and ha he condiional disribuion of y i given n i and x i follows mulinomial disribuion: =1 y i n i, X i, a i Mulinomial{n i, p i1 (X i, β 0 ), p i2 (X i, β 0 ),..., p it (X i, β 0 )} where p i (X i, β 0 ) = a i exp(x i β 0 ) T a i exp(x i β 0 ) =1 From Equaion (9), he join densiy funcion of (y i1, y i2,..., y it ) is derived as ( ) pr y i1, y i2,..., y it X i, y i ( = = y i )! ( y i )! y i! exp(x i β) y i { } yi y i! exp(x ik β) k p y i i (9) (10) where p i = exp(x iβ) y i { } yi. exp(x ik β) k

9 Susainabiliy 2018, 10, of 18 Equaion (10) is he likelihood funcion of he FE Poisson model. The condiional log likelihood funcion is given by aking logarihms of he model in Equaion (10) and summing up over individual observaion i: L(β) = { ( N T ) log Γ y i + 1 i=1 =1 T =1 log Γ(y i + 1) + The FE Poisson model is esimaed by maximizing Equaion (11). 4. Empirical Findings 4.1. Daa Descripion T =1 y i log p i } According o he Ac on Informaion Disclosure of Educaional Insiuions, Secion 6, all educaional insiuions in Souh Korea mus have horough informaion disclosure and regulaed relevan deails. The purpose of his Ac is o ensure he righ o be informed, promoe academic and policy research, encourage paricipaion in school educaion, and improve efficiency and ransparency in educaional adminisraion. The educaional websie Higher Educaion in Korea ( designed and operaed by he Korean Council for Universiy Educaion, collecs deailed informaion for all universiies in Korea and makes i available o he public. The websie provides 63 iems in 14 caegories (e.g., sudens enrolled, enure-rack faculy, annual uiion and fees, scholarship recipiens, graduae employmen, research funding, journal publicaions, suden accommodaions, library budge and holdings, financial resources, universiy budge and selemen, donaion and reserves). These iems are updaed once or wice a year, usually in Augus, and are available o he public for hree years. The raw daa were obained from his websie. The original daase conained academic informaion on a oal of 214 four-year universiies. Among hem, 22 universiies repored imprecise and erroneous informaion and were removed from he daase. Afer removing and adjusing he missing values, a hree-year balanced panel daase of 192 universiies from 2013 o 2015 was consruced for he sudy. We seleced and arranged he variables for he sudy under four comprehensive caegories, which are discussed in Secion 2. They were suden, faculy, resource, and universiy characerisics. Table 2 explains he variables in hese four caegories. Dropou is he oal number of universiy dropous, and i is considered he dependen variable for regression analysis. The suden caegory conained seven variables in hree subcaegories. Special HS is he rae of sudens from special high schools o all newly enrolled sudens. Grad school is he rae of sudens who go on o graduae school. Job is he rae of sudens who find jobs afer hey graduae. Tuiion is annual uiion measured in unis of 1000 Korean won. Scholarship is he average amoun of financial scholarship funding received per suden and is also expressed in unis of 1000 Korean won. Loan is he rae of sudens who receive loans. The faculy caegory conained five variables: sudens/prof, lecure, enure, sci, and exernal fund. Sudens/prof is he number of sudens per enure-rack professor. Lecure is he rae of lecures augh by enure-rack professors. Tenure is he number of enure-rack professors. Sci is he average number of aricles per professor published in SCI (Science Ciaion Index) or SSCI (Social Science Ciaion Index) journals. Exernal fund is he average amoun of research funds received from oher insiuions or organizaions. The resource caegory conained four variables in wo subcaegories: bookkeep, donaion, projec, and faciliy. Bookkeep is he annual selemen amoun measured in unis of 1000 Korean won. Donaion is he annual increase in donaions measured in unis of 1000 Korean won. Projec is he oal amoun of financial subsidy graned by he Minisry of Educaion, Science, and Technology (MEST) and is measured in unis of 1000 Korean won. Faciliy is he professor faciliies provision rae. (11)

10 Susainabiliy 2018, 10, of 18 Table 2. Explanaion of variables. Facor Concep Variable Explanaion Suden Facor Faculy Facor Resource Facor Srucural Characerisics of Universiy Dependen Variable Dropou Number of Dropous Qualiy of Suden special HS Rae of freshmen from special high schools Benefi Cos Quaniy of Faculy Qualiy of Faculy Finance grad school job uiion scholarship loan sudens/prof lecure enure sci exernal fund bookkeep donaion projec Rae of graduaes o grad school Rae of graduaes who find jobs Annual uiion Average scholarship Rae of sudens who receive loans Number of sudens per professor on enure rack Rae of lecures augh by professors on enure rack Number of professors on enure rack Number of SCI/SSCI aricles published Research funds from ouside he universiy Selemen amoun Annual amoun of donaions Toal amoun of projec funds from governmen Faciliy faciliy Faciliy provision rae Size Type: Public/Privae Universiy sudens saff sae_d Toal number of enrolled sudens Number of saff Sae dummy: 1 if sae universiy, 0 oherwise Locaion ciy_d Locaion dummy: 1 if in ciy, 0 oherwise Age age Toal number of years since a four-year insiuion Lasly, he universiy characerisics caegory conained five variables in four subcaegories: sudens, saff, sae_d, ciy_d, and age. Sudens is he oal number of currenly enrolled sudens. Saff is he number of saff members. Sae_d is a saus dummy variable. I was assigned 1 if a universiy was a naional, sae, or ciy universiy and 0 if oherwise. Ciy_d is a locaion dummy variable. I was assigned 1 if a universiy was locaed in a ciy area and 0 if oherwise. Age is he number of years since he universiy became a four-year universiy. Table 3 repors he means and sandard deviaions as summary saisics for he variables from 2013 o 2015 lised in Table 2. Mos variables show a very seady paern. For example, he average number of dropous per universiy is jus under 400 and mainains sabiliy: in 2013, in 2014, and in Alhough he average uiion is almos fixed, he amoun of scholarship funds per suden keeps increasing from 2,773,000 won in 2013 o 3,407,000 won in This reflecs he compulsory policy by MEST in Korea. However, lavishing scholarship funds on sudens while holding back uiion has been criicized as one of he leading reasons for universiies aggravaed budge siuaion. Table 4 repors he oal number of enrolled and dropou sudens. The oal number of dropou sudens comes from eigh differen caegories in he original daa, hree of which were direcly relaed o he sudens sudies. To analyze possible facors ha promp sudens o leave universiy, he oher five non-sudy-relaed caegories were considered in his sudy. These five caegories are unregisered, unenrolled, drop off, exceeding due periods, ec. The oal number of sudens a he 192 universiies slighly increased from 2,002,163 o 2,043,902 over hree years, while he oal number of dropous was almos sagnan. The dropou rae varied slighly, from 3.71% o 3.77%.

11 Susainabiliy 2018, 10, of 18 Table 3. Means and sandard deviaions for he variables, Variable dropou (316.51) (304.57) (298.49) special HS (7.928) (7.242) (5.902) grad school (15.23) (10.21) (15.23) job (17.14) (16.51) (17.92) uiion ( ) ( ) ( ) scholarship (961.99) (945.12) (926.98) loan (6.64) (6.43) (5.99) sudens/prof (7.74) (6.84) (6.79) lecure (12.64) (11.44) (11.16) enure (346.18) (347.01) (350.27) sci 0.20 (0.23) 0.21 (0.24) 0.24 (0.25) exernal fund 49, (130,315.2) 40, (60,893.39) 89, (636,924.7) bookkeep ( ) ( ) ( ) donaion 4,056,260 (8,081,197) 6,962,536 ( ) 4,273,196 (9,532,227) projec ( ) ( ) ( ) faciliy (85.22) (209.00) (198.73) sudens 10, ( ) 10, ( ) 10, ( ) saff (212.06) (214.38) (217.70) age (18.28) (18.28) (18.28) Table 4. Toal number of dropou sudens, Year Toal Sudens Toal Dropous ,002,163 75, ,028,892 76, ,043,902 75, Regression Resuls Iniially, 28 variables were considered for seing up a model. Among hem, dropou, a coun variable, was used as he dependen variable, five dummy variables for ime and universiy characerisics, and 22 variables from he four caegories were considered as a se of independen variables. Based on he preliminary regression analysis, some redundan variables were removed. The finalized nonlinear panel regression model for he sudy is given in Equaion (12): dropou i = a i exp β 0 + δ 1 sae_d i + δ 2 ciy_d i + δ 3 d14 i + δ 4 d15 i + β 1 special HS +β 2 grad school i + β 3 job i + β 4 log (uiion) i + β 5 log (scholarship) i +β 6 loan i + β 7 sudens/pro f i + β 8 lecuure i + β 9 log (enure) i +β 10 sci i + β 11 log (exernal f und) i + β 12 log(bookkeep) +β 13 log (donaion) i + β 14 log (projec) i + β 15 f aciliy i +β 16 log (sudens) i + β 17 sa f f i v i (12) where v i is idiosyncraic errors, d14 is he year dummy for 2014, and d15 is he dummy for The independen variable age was removed in he model due o he perfec collineariy problem wih he year dummy variables. Table 5 repors he regression resuls. Columns 4 6 lis he regression resuls from he linear models ordinary leas squares, RE model, and FE model, respecively. These models presume ha here exiss a linear relaionship beween he dependen and independen variables, which was no appropriae in he case of he coun dependen variable. Columns 7 and 8 lis he regression resuls from he nonlinear panel daa models, he Poisson RE and Poisson FE models, respecively.

12 Susainabiliy 2018, 10, of 18 Table 5. Regression Resuls. Facor Concep Variable OLS RE FE Poisson RE Poisson FE Suden Facor Faculy Facor Resource Facor Srucural Characerisics of Universiy Qualiy of Suden Benefi Cos Quaniy of Faculy Qualiy of Faculy Finance Faciliy Type: Public/privae Univ. Size Locaion Year special HS grad school job log (uiion) log (scholarship) loan sudens/prof lecure log (enure) sci log (exernal fund) log (bookkeep) log (donaion) log (projec) faciliy sae_d log (sudens) saff ciy_d d14 d15 consan Hausman es *** (2.557) (1.723) (0.908) * ( ) (59.641) (2.393) ** (3.055) *** (1.163) *** (49.969) *** (86.193) *** (10.035) *** (18.633) *** (8.997) (6.471) *** (0.215) *** (71.121) *** (43.927) (0.099) * (23.318) (21.383) ** (28.946) ( ) (3.122) (1.351) (0.536) * ( ) (51.096) (3.130) *** (2.811) *** (1.019) *** (52.478) ** (90.338) (9.271) ** (22.401) * (7.193) (8.012) (0.140) *** ( ) ** (35.279) (0.113) ** (32.122) (11.241) (19.172) ( ) (1.482) (1.543) (0.333) ( ) (62.754) (5.190) (5.592) (1.242) ( ) ( ) (13.138) (11.301) (6.712) (13.275) * (0.068) *** (23.279) (0.095) (11.438) (19.316) *** (0.004) (0.003) *** (0.001) (0.277) *** (0.051) (0.005) (0.004) *** (0.001) *** (0.084) (0.1115) (0.010) ** (0.018) *** (0.007) *** (0.008) (0.0002) *** (0.184) *** (0.062) (0.0001) *** (0.064) (0.010) (0.020) (2.425) (0.005) *** (0.004) * (0.001) (0.981) *** (0.055) (0.006) (0.007) *** (0.001) *** (0.193) *** (0.135) (0.011) (0.019) ** (0.007) *** (0.008) ** (0.0003) *** (0.081) (0.0001) (0.011) (0.023) R 2 = R 2 = R 2 = LL = LL = p-value = Noes: 1. *, **, and *** denoe 10%, 5%, and 1% levels of significance, respecively; 2. LL in columns 7 and 8 sands for log likelihood; 3. Sandard errors in he linear models are heeroscedasiciy robus sandard errors. - As shown in columns 7 and 8 in Table 5, mos of he regression resuls from boh he Poisson RE and FE regression models look very similar, which is no an uncommon incidence, hough. Among hem, seven coefficiens from boh models were saisically significan: he coefficiens of job and log (scholarship) in he suden caegory, he coefficiens of lecure and log (enure) in he faculy caegory, he coefficiens of log (donaion) and log (projec) in he resource caegory; and he coefficien of log (suden) in he characerisics-of-universiy caegory. Meanwhile, few coefficiens showed opposie signs or differen magniudes. The coefficien of grad school in he suden caegory had opposie

13 Susainabiliy 2018, 10, of 18 signs, bu only he coefficien from he FE Poisson was significan a he one-percen significance level. The coefficiens of sci in he faculy caegory and faciliy in he resource caegory in he FE Poisson were also significan a he one- and five-percen significance levels, respecively. Alhough boh he RE and FE Poisson models yielded similar regression resuls, i was worh discerning he correcly specified model beween hem. For his purpose, he Hausman es was applied; i is repored in he boom row of Table 5. The Hausman es saisic was wih a p-value of 0.00, which indicaed ha he FE Poisson model was preferable o he RE model. The main empirical findings based on he FE Poisson regression resuls in column 8 in Table 5 are lised as follows: 1. Grad school, job, and log (scholarship) in he suden caegory are significan. A. The coefficien of grad school is and significan a he one-percen significance level. i. Inerpreaion: If he rae of going o graduae school goes up one percen, hen he dropou rae is also increased by one percen. B. The coefficien of job is and significan a he en-percen significance level. i. Inerpreaion: Dropou rae decreases by wo percen if he rae of employmen increases by en percen. ii. This coincides wih Simpson e al. [14]. C. Going o graduae school and finding jobs are regarded as imporan facors for evaluaing good academic insiuions. However, his empirical resul showed ha hese wo have opposie effecs, indicaing ha sudens seem o use he universiy as a means of finding a job. D. Special HS as a proxy for he qualiy of sudens had a coefficien of bu was no significan, and he effec of special high schools on he dropou rae is no clear. E. The coefficien of log (scholarship) he dropou elasiciy of scholarships was and significan a he 1% significance level. i. Inerpreaion: If scholarships increase by one percen, dropou decreases by 0.248%. ii. Implicaion: Financial siuaion is a very imporan facor in deermining wheher one says in school. iii. This resul is concurren wih Li and Killian [15], Ishiani e al. [16], and Arend [17]. 2. Lecure, log (enure), and sci in he faculy caegory had significan coefficiens. A. The coefficien of lecures was and also significan a he one-percen significance level. i. Inerpreaion: A one-percen increase in professors lecures causes dropou o increase by 0.4%. ii. iii. iv. This is an ineresing finding because i suggess ha sudens drop ou when more lecures are given by enure-rack professors. This empirical finding appears o conradic he fac ha a beer qualiy of educaion booss sudens persisence a universiy. However, as shown by Gio e al. [4], sudens may see beer-qualiy educaion as a heavier learning burden. B. The coefficien of log (enure) was and significan a he one-percen significance level. i. Inerpreaion: If he number of enure-rack professors increases by one-percen, dropou decreases by 0.55%.

14 Susainabiliy 2018, 10, of 18 ii. The number of enure-rack professors plays a very imporan role in reducing college dropou. C. The coefficien of sci was and significan a he one-percen significance level, which is a very large effec. i. Inerpreaion: Publishing one more SCI/SSCI journal aricle increases dropou by 40.1%. ii. iii. This migh seem odd, bu i provides srong empirical evidence ha research and educaion migh no go in he same direcion. Professors research abiliies, such as publishing academic journal aricles or obaining research funds, did no seem o encourage sudens o say in school. D. The rae of sudens o professors was no significan in his empirical sudy, which is conrary o Asin [13] and Bound e al. [2]. 3. Log (donaion), log (projec), and faciliy in he resource caegory were significan. A. The coefficien of donaions was and significan a he five-percen significance level. B. The coefficien of log (projec) was and significan a he one-percen significance level. i. Inerpreaion: A 10% increase in governmen funding raises he dropou rae by 0.4%. ii. iii. This finding implies ha governmen subsidies graned by MEST were no very helpful for keeping sudens in school. This finding also indicaes ha while governmen subsidies migh be a very imporan resource for easing financial burdens on universiies, hey do no reduce universiy dropou. iv. This is he opposie of he empirical findings in Bound e al. [2], Tius [12], and Gross e al. [20]. C. The coefficien of faciliy was and significan a he five-percen significance level. i. This finding implies ha a beer educaional environmen plays a negaive role in reducing dropou, which is unusual and meris furher research. ii. This empirical finding is he opposie of McGowen [23]. 4. Only log (suden) in he srucural characerisics of he universiy caegory affeced dropou. A. The coefficien of log (sudens) was and significan a he one-percen significance level. i. This finding implies ha he old principle in he educaional field ha small is beauiful holds rue for universiies in Souh Korea. ii. This resul aligns wih Piman and Haughwou [24]. 5. Due o he naure of he FE regression model, all of he ime-consan variables were eliminaed and could no be esimaed. However, hey were esimaed in he Poisson RE model. Boh he saus and locaion dummy variables in he universiy characerisics caegory were significan. A. The coefficien of sae_d was and significan a he one-percen significance level. i. Implicaion: Public universiies, such as naional, ciy, or sae universiies, have 46.4% fewer dropous han privae ones. B. The coefficien of ciy_d was and significan a he one-percen significance level. i. Implicaion: Universiies in urban areas have 14.5% fewer dropous han hose in rural areas.

15 Susainabiliy 2018, 10, of Summary The aim of he paper was o examine possible causes for universiy dropou wih a focus on objecive facors a he aggregae level. For his purpose, four fundamenal facors of universiy educaion were seleced and caegorized: suden, faculy, resource, and srucural facors. We assumed ha hese four caegories would have consisen effecs on universiy dropou. Using nonlinear panel daa models, we carried ou an empirical analysis wih hree-year balanced panel daa from 2013 o 2015, which were obained from he Higher Educaion in Korea websie. The empirical findings of he sudy can be summarized as follows. Firs, geing a job and receiving a scholarship in he suden caegory were wo major facors in lowering he universiy dropou rae. Sending sudens o graduae schools is one of he mos imporan asks of a universiy, bu i migh pu sress or exra burdens on sudens and discourage hem from saying a universiy. Second, lecure, enure, and sci in he faculy caegory affeced universiy dropou. A larger number of professors was a very supporive facor for sudens persisence, bu a beer qualiy of educaion, represened as lecures from professors on he enure rack, was regarded as an exra burden for sudens and made hem consider quiing school. Publishing many aricles is a very imporan virue of professors and also creaes a good repuaion for he universiy. However, he dark side of a research-oriened universiy, especially for undergraduae, is ha sudens may feel malreaed. Third, donaions, projecs, and faciliies in he resource caegory affeced universiy dropou. Even hough donaions and projecs are wo good resources for improving a universiy s financial siuaion, hese wo facors had opposie effecs on universiy dropou. The former can be direcly used o improve he educaional environmen and make sudens more saisfied wih heir universiy. The laer, however, requires limied universiy resources (e.g., professors ime, effor, and energy) o apply for and obain funds; as such, i helps o improve he universiy s financial siuaion a he cos of aking less care of sudens. Finally, he size facor in he universiy characerisics caegory was very closely relaed o universiy dropou. I seems ha small is beauiful is also valid for universiies in Souh Korea. 6. Conclusion and Policy Implicaions In Souh Korea, universiy dropou has become a nagging problem for universiies as well as sociey. The counry also faces a serious populaion cliff problem, which is anoher negaive facor ha will only worsen universiy dropou. The oal number of sudens enering high school in 2018 will be abou 500,126 while he oal number of universiy freshmen will be 530,655. To examine his issue and evaluae he siuaion universiies currenly face, we invesigaed he possible causes of universiy dropou wih a focus on objecive facors a he aggregae level Theoreical Implicaions Previous research has exensively analyzed he deerminans of universiy dropou. However, hose sudies ended o focus on one of hree main rends: demand-side (suden) facors, individual-level subjecive or psychological facors, and insiuional supply-side facors. Gio e al. [4] noed ha many sudies have focused only on demand-side facors while few (e.g., Gio e al. [4] and Larsen e al. [3]) have examined he supply/objecive side a he aggregae level. Recenly, here has been a focus on adoping a more balanced view ha reflecs boh he demand/subjecive side and he supply/objecive side, wih only a few previous sudies (e.g., Tino [6] and Pierrakeas e al. [11]) having considered boh. Our sudy has some disincive feaures compared o previous research. Firs, we adoped a modified balanced model and focused on supply-side universiy facors, including faculy, resource, and srucural facors, as well as demand-side suden facors. Second, unlike mos previous sudies using cross-secional daa, our sudy used hree-year balanced panel daa. Alhough panel daa

16 Susainabiliy 2018, 10, of 18 are usually more difficul o collec han cross-secional daa, hey allowed us o conrol for cerain unobserved effecs of individual unis, analyze he significance of lags in behavior, and make causal inferences, all of which are very difficul wih cross-secional daa Pracical Implicaions The empirical findings in Secion 4 indicae ha 10 of he 19 variables were saisically significan. Going o graduae school, finding a job, and receiving scholarships were significan in he suden caegory. I is worh noing ha going o graduae school has a negaive impac on lowering dropou. Geing a job and receiving scholarships boh help o reduce dropou, which aligns wih Li and Killian [15] and Ishioni e al. [16] for scholarships and Simpson e al. [14] for high vocaional orienaion. Lecures augh by professors, he number of professors, and he number of published SCI/SSCI aricles were significan in he faculy caegory. The finding ha he more faculy, he lower he dropou is congruen wih Pascarella and Parick [19] and Gio e al. [4]. Asin [13] and Bound e al. [2] found ha a lower suden faculy raio reduced he dropou rae. Our empirical finding for his effec was economically significan bu saisically insignifican. I is remarkable ha publishing more SCI/SSCI journal aricle increases dropou. The annual amoun of donaions, oal amoun of projec funds from he governmen, and faciliy provision raes were significan in he resource caegory. Donaions help o reduce dropou, and his resul agrees wih Gross e al. [20]. The more projec funding from governmen increases dropou. Sewasew [21] found ha a lack of faciliies was one reason for dropou, while McGowen [23] found no effec on compleion rae. However, we found a negaive role in reducing dropou, which is unusual and meris furher research. Only he oal number of enrolled sudens was significan in he characerisics-of-universiy caegory. Our resul implies ha small is beauiful, similar o Piman and Haughwou [24]. Based on our empirical findings, i would seem ha universiy dropou could be lowered by changing some exising policies. Firs, privae universiies in Souh Korea comprised 79.2% of all universiies in 2016 (179 ou of 226). Unlike hose in he US, he majoriy are in poor financial shape and canno coninue wihou suppor from MEST in he form of projec-based funding. However, his ype of suppor aggravaes he qualiy of universiy educaion and leads o a higher dropou rae. Thus, more direc and insananeous ypes of subsidies should be adoped o srenghen universiies financial condiions. Second, even hough geing a job is a very imporan facor for reducing universiy dropou, universiies should fulfill heir original purpose and devoe hemselves o raising he qualiy of heir resources. If his is done, only hose who need higher educaion will come and aemp o complee heir sudies. Lasly, if educaion and research canno go in he same direcion, running wo racks, nururing eaching-oriened professors on he one hand while encouraging research-oriened professors on he oher hrough selecion and concenraion processes, could also help o lower he universiy dropou rae Limiaions This sudy aemped o conribue o he lieraure on universiy dropou by examining hree supply-side facors as well as demand-side facors. Due o a lack of deailed daa, however, we could no divide dropous ino volunary and involunary ypes. For a richer empirical analysis and inerpreaion, considering universiy heerogeneiy may play an essenial role. Unlike schools in he US and some oher developed counries, here is no clear division beween research-oriened and educaion-oriened schools in Souh Korea. In general, wo- or hree-year colleges are regarded as vocaional schools and were excluded in our sudy. Alhough we used panel daa regression analysis o avoid unnecessary and unobserved effecs on he model, we could no successfully conrol for all he relevan variables. Fuure research can address such limiaions. Moreover, alhough universiy heerogeneiy and he effec of educaional facors such as educaional mismach on universiy dropou are very imporan, we did no include hem in he model because here was no daase o suppor hem. Moreover, Di Piero [28] effecively showed he negaive relaionship beween regional unemploymen

17 Susainabiliy 2018, 10, of 18 raes and universiy dropou raes. Aina [29] showed ha, in Ialy, parenal background influences dropou: sudens wih fahers or boh parens who only had compulsory schooling were more likely o drop ou. Our sudy did no consider hose opics. Such limiaions can be addressed in fuure sudies. Acknowledgmens: This research was conduced wihou any research grans. Auhor Conribuions: The wo auhors conribued equally o his work. Conflics of Ineres: The auhors declare no conflics of ineres. References 1. OECD. Educaion a a Glance; OECD: Paris, France, Bound, J.; Lovenheim, M.F.; Turner, S. Why have college compleion raes declined? An analysis of changing suden preparaion and collegiae resources. Am. Econ. J. Appl. Econ. 2010, 2, [CrossRef] [PubMed] 3. Larsen, M.R.; Sommersel, H.B.; Larsen, M.S. Evidence on Dropou Phenomena a Universiies; Danish Clearinghouse for Educaional Research: Copenhagen, NV, USA, Gio, L.; Minervini, L.F.; Monaco, L. Universiy dropous in Ialy: Are supply side characerisics par of he problem. Econ. Anal. Policy 2016, 49, [CrossRef] 5. Spady, W.G. Dropous from higher educaion: Toward an empirical model. Inerchange 1971, 2, [CrossRef] 6. Tino, V. Dropou from higher educaion: A heoreical synhesis of recen research. Rev. Educ. Res. 1975, 45, [CrossRef] 7. Lam, Y.L.J. Predicing dropous of universiy freshmen: A logi regression analysis. J. Educ. Adm. 1984, 22, Bean, J.P. Ineracion effecs based on class level in an explanaory model of college suden dropou syndrome. Am. J. Educ. Res. 1985, 22, [CrossRef] 9. Heublein, U.; Huzsch, C.; Schreiber, J.; Sommer, D.; Besuch, G. Ursachen des Sudienabbruchs in bachelor- und in herkömmlichen sudiengängen: Ergebnisse einer bundesweien befragung von exmarikulieren des sudienjahres 2007/2008; HIS: Hannover, Germany, Heublein, U.; Spangenberg, H.; Sommer, D. Ursachen des sudienabbruchs:analyse 2002; HIS: Hannover, Germany, Pierrakeas, C.; Xenos, M.; Panagioakopoulos, C.; Vergidis, D. A comparaive sudy of dropou raes and causes for wo differen disance educaion courses. The Inernaional Review of Research in Open and Disribued Learning. Available online: hp:// (accessed on 30 December 2017). 12. Tius, M.A. Undersanding he influence of he financial conex of insiuions on suden persisence a four-year colleges and universiies. J. High. Educ. 2006, 77, [CrossRef] 13. Asin, A.W. Suden involvemen: A developmenal heory for higher educaion. J. Coll. Sud. Dev. 1999, 40, Simpson, C.; Baker, K.; Mellinger, G. Convenional failures and unconvenional dropous: Comparing differen ypes of universiy wihdrawals. Sociol. Educ. 1980, 53, [CrossRef] 15. Li, G.; Killian, T. Sudens Who Lef College: An Examinaion of Their Characerisics and Reasons for Leaving. In Proceedings of he AIR 1999 Annual Forum Paper presened a he 39h Annual Forum of he Associaion for Insiuional Research, Seale, WA, USA, May 30 June Ishiani, T.T.; DesJardins, S. A longiudinal invesigaion of dropou from college in he Unied Saes. J. Coll. Sud. Re. 2002, 4, [CrossRef] 17. Arend, J.N. The effec of public financial aid on dropou from and compleion of universiy educaion: Evidence from a suden gran reform. Empir. Econ. 2013, 44, [CrossRef] 18. Pascarella, E.T.; Duby, P.B.; Iverson, B.K. A ex and reconcepualizaion of a heoreical model of college wihdrawal in a commuer insiuion seing. Sociol. Educ. 1983, 56, [CrossRef] 19. Pascarella, E.T.; Terenzini, P.T. Ineracion effecs in Spady and Tino s concepual models of college ariion. Sociol. Educ. 1979, 52, [CrossRef] 20. Gross, J.P.K.; Hossler, D.; Ziskin, M. Insiuional aid and suden persisence: An analysis of he effecs of insiuional financial aid a public four-year insiuions. NASFAA J. Sud. Financ. Aid 2007, 37,

18 Susainabiliy 2018, 10, of Sewasew, D.T. Ariion causes among universiy sudens: The case of Gondar Universiy, Gondar, norh wes Ehiopia. Innov. J. Soc. Sci. 2014, 2, Bowers, J.H.; Burke, C.W. Effecs of physical and school environmen on sudens and faculy. CEFPI s Educ. Facil. Plan. 1989, 27, McGowen, R.S. The Impac of School Faciliies on Suden Achievemen, Aendance, Behavior, Compleion Rae and Teacher Turnover Rae in Seleced Texas High Schools. Ph.D. Thesis, Texas A&M Universiy, College Saion, TX, USA, Piman, R.B.; Haughwou, P. Influence of high school size on dropou rae. Educ. Eval. Policy Anal. 1987, 9, [CrossRef] 25. Jordan, J.L.; Kosandini, G.; Mykerezi, E. Rural and urban high school dropou raes: Are hey differen? J. Res. Rural Educ. 2012, 27, Pallas, A.M. School dropous in he Unied Saes. In Dropous, Pushous and Oher Casualies; Denon, W.T., Ed.; Phi Dela Kappa: Bloomingon, IN, USA, 1987; pp Hausman, J.A.; Hall, B.H.; Griliches, Z. Economeric models for coun daa wih an applicaion o he paens-r&d relaionship. Economerica 1984, 52, Di Piero, G. Regional labour marke condiions and universiy dropou raes: Evidence from Ialy. Reg. Sud. 2006, 40, [CrossRef] 29. Aina, C. Parenal background and universiy dropou in Ialy. High. Educ. 2013, 65, [CrossRef] 2018 by he auhors. Licensee MDPI, Basel, Swizerland. This aricle is an open access aricle disribued under he erms and condiions of he Creaive Commons Aribuion (CC BY) license (hp://creaivecommons.org/licenses/by/4.0/).