DISCUSSION PAPER SERIES IZA DP No. 4212 Mltary Conscrpton and Unversty Enrolment: Evdence from Italy Gorgo D Petro June 2009 Forschungsnsttut zur Zukunft der Arbet Insttute for the Study of Labor
Mltary Conscrpton and Unversty Enrolment: Evdence from Italy Gorgo D Petro Unversty of Westmnster and IZA Dscusson Paper No. 4212 June 2009 IZA P.O. Box 7240 53072 Bonn Germany Phone: +49-228-3894-0 Fax: +49-228-3894-180 E-mal: za@za.org Any opnons expressed here are those of the author(s) and not those of IZA. Research publshed n ths seres may nclude vews on polcy, but the nsttute tself takes no nsttutonal polcy postons. The Insttute for the Study of Labor (IZA) n Bonn s a local and vrtual nternatonal research center and a place of communcaton between scence, poltcs and busness. IZA s an ndependent nonproft organzaton supported by Deutsche Post Foundaton. The center s assocated wth the Unversty of Bonn and offers a stmulatng research envronment through ts nternatonal network, workshops and conferences, data servce, project support, research vsts and doctoral program. IZA engages n () orgnal and nternatonally compettve research n all felds of labor economcs, () development of polcy concepts, and () dssemnaton of research results and concepts to the nterested publc. IZA Dscusson Papers often represent prelmnary work and are crculated to encourage dscusson. Ctaton of such a paper should account for ts provsonal character. A revsed verson may be avalable drectly from the author.
IZA Dscusson Paper No. 4212 June 2009 ABSTRACT Mltary Conscrpton and Unversty Enrolment: Evdence from Italy Gven that a growng number of countres have abolshed or are consderng the abolton of mltary conscrpton, understandng the consequences of ths measure s of ncreased mportance. In ths paper we study the effect of the suppresson of compulsory mltary servce on unversty enrolment n Italy usng double and trple dfferences models. The emprcal results show that there s no compellng evdence suggestng that the abolton of mltary conscrpton has a causal effect on unversty enrolment. However, although there s no sgnfcant overall effect, we fnd some evdence of heterogeneous effects. Whle ths measure seems to ncrease unversty partcpaton among ndvduals from more advantaged backgrounds, t appears to have a detrmental effect on the enrolment of those from less advantaged backgrounds. JEL Classfcaton: I20, J24 Keywords: compulsory mltary servce, unversty enrolment Correspondng author: Gorgo D Petro Unversty of Westmnster Westmnster Busness School Department of Economcs and Quanttatve Methods 35 Marylebone Road NW1 5LS London Unted Kngdom E-mal: G.D.I.Petro@westmnster.ac.uk
1. Introducton A number of studes n the economcs lterature look at the effects of mltary conscrpton. Whle some of them (Angrst and Krueger, 1989; Angrst, 1990; Imbens and Wlbert van der Klaauw, 1995; Bauer et al., 2004) analyse the mpact of servng n the mltary on ndvduals subsequent earnngs, others focus on the consequences of the suppresson of compulsory mltary servce (CMS) on schoolng. These latter studes produce, however, mxed results. Maurn and Xenogan (2007) conclude that French young men who were affected by the 1997 mltary reform, whch abolshed CMS, have a lower educatonal attanment and a lower probablty of successfully completng hgh school relatve to ther peers who were not at rsk of servce. Ths reform s found to have a detrmental effect especally on the educatonal outcomes of ndvduals from less advantaged socoeconomc backgrounds. Results from studes by Angrst and Krueger (1992) and Card and Lemeux (2001) can also be nterpreted as supportng the hypothess that the abolton of CMS may have an adverse effect on schoolng. They fnd that n the US between 1965 and 1975 draft avodance behavours led to a rse n college enrolment and graduaton rates, mplctly suggestng that these would have been lower n the absence of conscrpton. On the other hand, fndngs from studes on the UK and Italy are at varance wth those reported above. Buonanno (2006) shows that n the UK the abolton of mltary conscrpton has led to an ncrease n school attendance among men aged 18 or above. Addtonally, hs estmates ndcate that ndvduals for whom the mltary servce was not compulsory have, on average, 0.25 years of educaton more than those who were subject to CMS. Cpollone and Rosola (2007) look at the consequences of exempton from CMS gven to a few cohorts of men lvng n areas of southern Italy that were ht by an earthquake n 1980. They fnd that ths exempton led to a 2 percentage ponts ncrease n male hgh school graduaton rates. In lght of these mxed results, n ths paper we attempt to provde new evdence on the relatonshp between educaton and the suppresson of CMS. We extend earler work of Cpollone and Rosola (2007) on Italy n three man respects. Frst, whle 3
ther analyss regards some parts of Italy, we look at the effects of the abolton of Natonal Servce n all the Italan terrtory. Second, our study uses very recent data as CMS was abolshed n Italy n 2005. Thrd, whlst Cpollone and Rosola (2007) examne how not beng at rsk of servce has affected the choce to fnsh hgh school, we analyse the effect of the abolton of CMS on the unversty enrolment decson. Understandng the factors nfluencng the transton from secondary to tertary educaton s very mportant gven that there are relevant benefts assocated wth nvestment n hgher educaton not only for ndvduals but also for socety as a whole (Baum and Payea, 2004). The abolton of mltary conscrpton may have conflctng effects on the decson to pursue educaton and n partcular on that of enrollng at unversty. On the one hand, t may dscourage male hgh school graduates from further nvestng n schoolng snce they no longer need to stay n educaton as a way to avod mmedate mltary servce. As outlned by Maurn and Xenogan (2007), many young men choose to defer the mltary servce as long as they can, snce ths experence s perceved to be assocated wth several costs (e.g. lvng n a barrack, sharng a room wth several other men and partcpatng n mltary tranng and drlls). Addtonally, holdng a unversty degree may sgnfcantly reduce these costs as unversty graduates are lkely to be chosen for mlder forms of natonal servce. On the other hand, however, the dsmssal of CMS may nduce more male hgh school graduates to nvest n unversty educaton as ths allows them to get returns from ths nvestment over a longer perod. Addtonally, although male unversty and hgh school graduates may both beneft from the abolton of CMS as they can both enter nto the labour market mmedately after completng ther studes, avodng career nterruptons s lkely to brng more advantages to the former group relatve to the latter 1. Ths s because the rate of deprecaton of educaton s lkely to ncrease wth years of educaton (Neuman and Wess, 1995) gven that more educated ndvduals typcally have jobs where techncal and economc changes happen at a fast rate. 1 The career nterrupton may last longer than the length of the mltary servce as typcally, due to bureaucratc reasons, there s a watng perod before men can actually start the servce. Imbens and Wlbert van der Klaauw (1995) suggest that n the Netherlands the cost of dong the mltary servce s approxmately equvalent to the cost of losng two years of potental work experence. 4
The remander of ths paper s as follows. In Secton 2 we develop a smple model to nvestgate the effect of the abolton of CMS on the unversty enrolment decson. Secton 3 brefly descrbes the nsttutonal features of CMS n Italy before t was abolshed. In Secton 4 we outlne the dentfcaton strategy used to determne the causal effect of the suppresson of CMS on unversty enrolment. Secton 5 presents the data employed n ths study. In Secton 6 emprcal results are reported and dscussed. Secton 7 concludes. 2. A smple theoretcal framework Unlke prevous studes on the effect of the abolton of CMS on educaton, n ths paper we set up a smple theoretcal framework n order to understand how not havng to do the mltary servce may nfluence the schoolng decson of young male ndvduals. We adopt a human captal approach (Becker, 1964) and, gven the purpose of ths study, we focus our attenton on the decson whether or not to enrol at unversty. Let s frst analyse ths choce n the presence of CMS. In such a case, the present value of the earnngs stream for a male hgh school graduate who, at the age of 18, chooses not to enrol at unversty s 64 HS W (1) PV = 19 HS n 18 where the superscrpt HS refers to hgh school educaton, r s the dscount rate and W s the yearly wage. Assumng that the mltary servce lasts for one year, ths ndvdual s expected to earn W HS from age 19 untl the age of retrement (.e. 65). If nstead, the male hgh school graduate chooses to enrol at unversty, then the present value of the stream of hs future ncomes would be 64 U U W (2) PV = 24 n 18 where the superscrpt U refers to unversty educaton. Assumng that ths ndvdual s able to graduate n four years and that the mltary servce stll lasts for one year, he s expected to earn W U from age 24 untl retrement. 5
Now we look at what happens when CMS s abolshed. In ths crcumstance, the present value of the earnngs stream for a male hgh school graduate who, at the age of 18, chooses not to enrol at unversty becomes 64 HS W (3) PV = 18 HS n 18 If nstead, after the suppresson of CMS, the male hgh school graduate chooses to enrol at unversty, then the present value of the stream of hs future ncomes would become 64 U U W (4) PV = 23 n 18 As one can see from Equatons (3) and (4), the abolton of CMS brngs fnancal benefts for the ndvdual f he decdes not to attend unversty as well as f he chooses to enrol at unversty. In the presence of CMS, a male hgh school graduate wll choose to enrol at unversty f: 64 U 64 HS W W > n 18 n 18 1+ r) 24( 19 23 HS 64 U HS W W W (5) + > 0 n 18 n 18 19 where 23 19 W 64 U W W and n 24 HS HS 18 n 18 24 represents the opportunty cost of gong to unversty 2 s the gan from havng a hgher wage afterwards. In the absence of CMS, a male hgh school graduate wll choose to enrol at unversty f: 64 U 64 HS W W > n 18 n 18 1+ r) 23( 18 22 HS 64 U HS W W W (6) + > 0 n 18 n 18 18 where 22 18 W HS n 18 23 represents the opportunty cost of gong to unversty 2 It s assumed that the drect cost of unversty educaton s zero. 6
64 U W W and n 23 HS 18 s the gan from havng a hgher wage afterwards. Ths smple model shows that the of abolton of CMS encourages a male hgh school graduate to enrol at unversty f the left hand sde of Equaton (6) s bgger than the left hand sde of Equaton (5): 22 18 HS W + n 18 ( 1+ r) n 64 = 23 U W W n HS 18 > 23 W HS + 64 n 18 19 ( 1+ r) 24 U W W HS n 18 W HS HS W + U W W + HS > 5 5 U HS HS 1 W W (7) W ( 1) + > 0 5 5 The condton (7) can be re-wrtten as: (8) W U > ( 1+ r) 5 W HS 0 From Equaton (8) t emerges that the suppresson of CMS wll have a postve mpact on the unversty enrolment decson f 1) the ndvdual tme dscount rate s small enough 2) the unversty wage s large enough relatve to the hgh school wage 3. Compulsory mltary servce n Italy In Italy CMS was abolshed n 2005 followng a mltary reform (Legge Martno) approved by the Parlament n July 2004. Accordng to ths reform, men born after 1 st January 1986 are no longer subject to CMS 3. The ratonale for ths was to move toward a professonal mltary n lne wth several other countres ncludng the US and the UK. In Italy the procedures for the selecton of conscrpts were smlar to those mplemented n other countres such as Germany and the Netherlands. All young men were called for medcal and psychologcal examnatons shortly after or shortly before they turned 18. These examnatons typcally lasted between 2 and 3 days, at the end 3 Whlst n the early 1980s the length of servce was 12 months, n the late 1990s ths was reduced to 10 months. 7
of whch perod potental recruts were splt nto three man groups: a) those who had a good health status and no psychologcal problems were declared to be sutable to do the mltary servce; b) those who had serous health or physologcal problems were mmedately exempt from the mltary servce; c) those who needed to undertake further medcal and psychologcal examnatons. Based on the results of these addtonal examnatons, they were declared to be ft to serve or not. Fnally, even f on the grounds of medcal and psychologcal examnatons a young man was found to be sutable to do the mltary servce, he could stll have been exempt for specal reasons 4. As outlned by Cpollone and Rosola (2007), n Italy the reasons for exempton from servce were strctly coded and restrctve. Men who at the age of 18 were n school could defer mltary servce, provded that they could successfully complete hgh school by the age of 22. Male hgh school graduates who enrolled at unversty mmedately after completng ther studes could also obtan one-year deferment. Those students performng well at unversty could obtan successve one-year deferments, provded that they could graduate by the age of 27. The 2004 mltary reform also put an end to conscentous objectors. Snce 1972, men, who were declared to be ft to serve n the mltary, had the possblty to become conscentous objectors. Conscentous objectors were freed from mltary servce but they had to perform publc servce for a perod of tme typcally longer than that assocated wth the mltary servce. On the other hand, they enjoyed consderable freedom n selectng the nsttuton where to the perform publc servce. 4. Identfcaton strategy In ths Secton we sketch the methodologcal approach used to estmate the effect of the abolton of CMS on unversty enrolment. Gender and the date of brth jontly determne whether a hgh school graduate s affected by the abolton of CMS. Women are exempt from mltary servce n Italy. Only those male hgh school graduates born after 1 st January 1986 beneft from the suppresson of CMS. Hence, to estmate the effect of the abolton of CMS on the decson to attend unversty, our 4 For nstance, f he was orphan from both parents, son or brother of a solder ded n war or wdower wth dependant chld 8
basc strategy s to track the dfference n unversty enrolment between male hgh school graduates born after and before 1 st January 1986, and then compare ths wth the correspondng dfference between smlar female hgh school graduates. Ths comparson, whch generates the so-called dfference-n-dfferences (DD) or doubledfference estmator, allows us to control for unobserved age-group specfc effects as well as for common macro-effects. Usng a sample of hgh school graduates who successfully completed ther studes n 2004, one may observe that those born after 1 st January 1986 were 18 years old or less at that tme. Thus our DD estmator s: 2 18 19 18 19 (9) = ( E E ) ( E E ) 2004 M M F F where subscrpts M and F denote males and females, respectvely. Whlst E 18 represents the unversty enrolment choce of ndvduals aged 18 or less when they fnshed hgh school, 19 E ndcates the unversty enrolment choce of ndvduals aged 19 or more when they fnshed hgh school. The subscrpt 2004 ndcates that we use a sample of people who fnshed hgh school n 2004. Adjustng the DD estmator for observables requres estmatng the followng equaton: (10) E = α + β 0 X + β1m + β 2T + β 3T M + ε where E s the unversty enrolment choce observed for hgh school graduate. It s a dchotomous varable that s equal to 1 f the hgh school graduate enrols at unversty, and 0 otherwse. X s a vector of ndvdual and famly characterstcs that are thought to nfluence the unversty enrolment decson. M s a dummy takng the value 1 f the hgh school graduate s male, and 0 otherwse. T s a dummy varable takng the value 1 f the ndvdual fnshed hgh school when he/she was 18 years old or younger, and 0 otherwse; and ε s the usual error term. The DD estmator s the coeffcent β 3. It tests whether, followng the abolton of CMS, the dfference n unversty enrolment between hgh school graduates aged 18 or less and ther peers aged 19 or above vares systematcally across gender. β 3 dentfes the true effect of the suppresson of CMS assumng there are no shocks 9
takng place concdentally at the same as the 2004 mltary reform and affectng only the unversty enrolment decson of male hgh school graduates aged 18 or less. Our DD estmator would be based f the change n unversty enrolment between hgh school graduates aged 18 or less and ther peers aged 19 or above systematcally evolves dfferently across males and females. One way of testng ths assumpton s to carry out a pre-program test usng a sample of ndvduals who fnshed hgh school n the years precedng the abolton of CMS. More precsely, usng a sample of hgh school graduates who successfully completed ther studes n 2001, the followng DD estmator can be computed: 2 18 19 18 19 (11) = ( E E ) ( E E ) 2001 M M F F An adjusted double-dfference for the 2001 sample of hgh school graduates can be computed by estmatng an equaton analogous to Equaton (10). If n ths equaton the estmated DD coeffcent turns out not to be statstcally sgnfcant, such a result would support the hypothess that β 3 n Equaton (10) dentfes the causal effect of the abolton of CMS on unversty enrolment. If, however, n the 2001 sample the dfference n unversty enrolment between hgh school graduates aged 18 or less and ther peers age 19 or above s found to be hgher among males than among females, then the DD estmator n Equaton (10) s lkely to overstate the true effect of the abolton of CMS on unversty enrolment. Fnally, a negatve and statstcally sgnfcant value of the adjusted double-dfference estmator for the 2001 sample mples that the DD coeffcent n Equaton (10) s lkely to understate the true effect of the abolton of CMS on unversty enrolment. By poolng the 2001 and 2004 samples of hgh school graduates t s possble to obtan a dfference-n-dfference-n-dfferences (DDD) or trple-dfference estmator followng the approach of Hamermesh and Trejo (2000) and Bansak and Rapahel (2001). We subtract the DD estmator for the 2004 sample from the comparable DD estmator for the 2001 sample. (12) 3 = 2 2004 2 2001 10
To compute the DDD estmator wthn a regresson framework, the followng regresson s estmated: (13) E = + β 0 X + β1m + β 2T + β 3D + β 4TM + β 5DM + β 6DT + β 7 α D M T + ε where D s dummy varable takng the value 1 f the hgh school graduate successfully completed hs/her studes n 2004, and 0 otherwse. The coeffcent β 7 represents the trple-dfference estmator. It measures the extent to whch n the 2004 sample the unversty enrolment gap between our two age-groups of male hgh school graduates, relatve to the unversty enrolment gap between the two smlar female groups, s statstcally dfferent from the comparable dfference-n-dfferences n the 2001 sample. If the adjusted double-dfference for the 2001 sample s equal to zero, then the trple dfference estmator wll reduce to the double-dfference n Equaton (10). Our estmaton strategy on the effect of the abolton of CMS on unversty enrolment s accompaned by one drawback whch s common to the large majorty of studes on the costs/benefts of mltary servce. We assume that all male hgh school graduates born before 1 st January 1986 are subject to CMS. Ths s obvously not true because one would expect some of them to have been exempt for the reasons outlned n footnote 4. Such an ssue, however, s unlkely to bas our estmates as long as the reasons behnd the exempton, though possbly correlated wth the unversty enrolment decson, are randomly dstrbuted across males and females aged 18 or less when they fnshed hgh school. Equatons (10) and (13) can be estmated by OLS as ths allows us to drectly estmate the parameter of nterest 5. As ponted out by Angrst (2001), the problem of causal nference does not sgnfcantly dffer between lmted dependent varables and contnuous outcomes. Ths means that f there are no covarates or the covarates are sparse and dscrete, then lnear models can be used to estmate models wth lmted 5 As shown by Chunrong and Norton (2003), whle n lnear models the nterpretaton of the coeffcent of the nteracton between two varables s straghtforward, ths does not hold n non-lnear models. 11
dependent varables as well as models wth other types of dependent varables. Berlnsk and Galan (2007) argue that nsttutonal reform and polcy experments clearly fall wthn ths framework snce control varables are prmarly added n an attempt to mprove the effcency of the estmates, but ther omsson s unlkely to sgnfcantly bas the estmates of the parameter of nterest. Borjas (2003, 2004) uses OLS to estmate double and trple dfferences models where the dependent varable s dchotomous. 5. Data The data for ths study are taken from two waves (.e. 2004 and 2007) of a natonal cross-sectonal survey (Percors d Studo e d Lavoro de Dplomat) carred out by the Italan Natonal Statstcal Insttute (ISTAT). Each wave conssts of a representatve sample of hgh school graduates who are surveyed three years after completng ther studes. Thus these data enable us to examne the post-hgh school decsons made by two cohorts of ndvduals who completed ther studes n 2001 6 and 2004. Although one of the possble destnatons of these ndvduals s unversty enrolment, there are other possble trajectores. Some people may start to work mmedately whle others may become job seekers, and they can all change ther mnd after the ntal choce. The survey contans ndvdual nformaton on prevous educatonal attanment, parents soco-economc status as well as a range of personal attrbutes. In Italy all hgh school graduates gan the automatc rght of entry to unversty, provded that they have successfully completed fve years n hgh school 7. The underlyng dea behnd ths provson s to gve to each hgh school graduate the opportunty of expermentng to see whether he/she s unversty materal (Mansk, 1989). Our emprcal analyss consders four dfferent sets of characterstcs that are lkely to affect the decson to enrol at unversty. The frst group conssts of three proxes for the ndvdual s academc ablty. These ndcators are: lower secondary school 6 Earler cohorts of hgh school graduates cannot be consdered for the purpose of ths study gven that n 2001 the Italan unversty system embarked on a process of reform whch led to the ntroducton of the 3+2 model (for more detals about ths reform see Bratt et al., 2006) 7 There are no admsson standards, except for specalzed dscplnes such as medcne and archtecture. 12
(scuola meda) fnal grade, hgh school fnal grade and a dummy varable recordng whether the ndvdual has faled and had to repeat at least one year at hgh school. The second set of factors ncludes school-related varables. We nclude ndcators for the type of hgh school attended by the ndvdual and a dummy varable recordng whether ths hgh school was publc or prvate. The thrd group of varables comprses ndvdual personal attrbutes such as gender, age and area of resdence 8. Fnally, we nclude famly background characterstcs such as parents hghest educatonal attanment and father s occupaton 9. Followng the approach of Cappellar (2004) and D Petro and Cutllo (2008), we elmnate from the fnal sample those ndvduals who graduated from those hgh schools manly desgned to tran prmary school teachers (scuole magstral). These schools have a dfferent structure relatve to the other types of hgh schools. Addtonally, we exclude from the fnal sample hgh school graduates lvng abroad and those who, though they enrolled at unversty, dd not do so mmedately after completng ther studes. After deletng observatons wth mssng varables of nterest, we are left wth 20,846 and 16,959 hgh school graduates n the 2004 and 2001 samples, respectvely. Table 1 provdes some descrptve statstcs for all the varables used n ths study. Although n Italy the large majorty of people successfully complete hgh school at the age of 19 or above, some manage to do t at the age of 18 (or less n very exceptonal cases). These are ndvduals who, at the age of 5, attended a specal course (prmna), upon successful completon of whch they were allowed to skp the frst year of prmary school. Ths means that at the age of 6, when most chldren begn prmary school, they were admtted to the second year of prmary school 10. Insert Table 1 about here 8 Four geographcal areas are consdered: North-East, North-West, South and Centre. Unfortunately, we are unable to defne areas of resdence at a more dsaggregated level (.e. regon) gven that ths nformaton s unavalable. The ncluson of dummes for area of resdence allows us to control for labour market condtons across the geographcal areas. 9 Gven the low drect cost of unversty educaton n Italy (compared to other countres such as, for nstance, the US), tuton fees are unlkely to exert an nfluence on the enrolment decson even for ndvduals from less advantaged backgrounds. Prevous studes (see for, nstance, D Petro and Cutllo, 2006) on the determnants of unversty enrolment n Italy have not consdered the mpact of ths factor. 10 Typcally parents can freely decde to make ther chld attend prmna. The chld has the rght to enrol on the second year of prmary school as long as he/she passes a fnal exam. 13
From Table 1 t emerges that n both the 2001 and 2004 samples unversty enrolment rate s sgnfcantly hgher among ndvduals who fnshed hgh school at the age of 18 or less relatve to ther older peers. For nstance, n the 2004 sample the proporton of people who enrolled at unversty mmedately after successfully completng hgh school s 80% among the former group, compared to 57% among latter group. Ths s also due to the fact that people who fnshed hgh school at the age of 18 or less typcally show hgher academc ablty and are more lkely to come from a more prvleged famly background than ther older peers. Ths fndng s n lne wth our expectatons gven that well-educated parents have hgh educatonal expectatons for ther chldren and hence they may want them to start school earler. 6. Emprcal results The results reported n the frst half of Table 2 present the double-dfference estmates for the 2004 sample, whch emerge from a very basc unversty enrolment specfcaton. The DD coeffcent s statstcally sgnfcant at all conventonal levels and has a postve sgn. Our emprcal results show that after the abolton of CMS the change n the probablty of enrollng at unversty between our two age-groups of hgh school graduates was 15 percentage ponts hgher across males relatve to females. Insert Table 2 about here A number of personal, famly background and school-related characterstcs are added to our basc unversty enrolment equaton. The result for ths new specfcaton s depcted n the second half of Table 2. The emprcal fndngs are generally consstent wth prevous research. Attendng a general hgh school (lceo), performng well at hgh school, not repeatng a hgh school year, beng female, studyng at a publc hgh school 11, havng well-educated parents all result n a hgher probablty of enrollng at unversty. Movng to the varable of prmary nterest n ths study, one may note that the DD coeffcent s stll statstcally sgnfcant at the 1 percent level. Nevertheless, ts value s smaller than the correspondng one n the frst half of Table 2. More precsely, our results mply that controllng for observable ndvdual characterstcs 11 Several studes (see, for nstance, Bertola and Checch, 2002) suggest that n Italy prvate schools play a remedal role. They tend to attract less talented students from wealthy backgrounds. 14
removes approxmately one-thrd of the estmated dfference-n-dfferences shown n the frst half of Table 2. To test whether the results depcted n Table 2 are robust, we check the senstvty of our estmates to our defnton of unversty enrolment. Hence we nclude n the fnal sample those hgh school graduates who enrolled at unversty, but nstead of dong t mmedately after fnshng ther studes, they dd t two or three years later. The ncrease n sample sze s, however, qute modest gven that n Italy the majorty of hgh school graduates enrol at unversty straght after completng ther studes. Estmates based on ths larger sample are depcted n Table 3. Not only n both model specfcatons the DD coeffcents are stll postve and statstcally sgnfcant at the 1 percent level, but also ther values are relatvely close to the correspondng values of the DD coeffcents depcted n Table 2. In lght of these results, t s possble to conclude that our estmates are not senstve to changes n the defnton of unversty enrolment Insert Table 3 about here Our DD estmates seem to suggest that the abolton of CMS s actually drvng the hgher unversty enrolment growth experenced by male hgh school graduates aged 18 or less compared to ther peers aged 19 or above, relatve to the growth n unversty enrolment between smlar female groups. Ths fndng, however, reles on the dentfcaton assumpton accordng to whch n the absence of the abolton of CMS the ncrease n unversty enrolment of the hgh school graduates aged 18 or less relatve to ther older peers would have been the same across males and females. In order to test ths hypothess, two dfferent tests are carred out. Frst, followng the approach of Duflo (2001), we perform a control experment. An mplcaton of the dentfcaton assumpton s that, snce n 2004 males who successfully completed hgh school at the age of 19 or above were not affected by the mltary reform, the ncrease n unversty enrolment between two cohorts n ths age-group should not be systematcally dfferent across males and females. Hence we restrct our 2004 sample to those ndvduals who successfully completed hgh school at the age of 19 or above and then we splt them n two cohorts: a cohort of ndvduals who fnshed hgh school at the age of 19 and a cohort of ndvduals who were older than 19 years when 15
they fnshed hgh school. To compute the new dfference-n-dfferences, we estmate the followng equaton: (14) E = α + β 0 X + β1m + β8k + β 9K M + ε where K s a dummy varable takng the value 1 f the ndvdual fnshed hgh school when he/she was 19 years old, and 0 otherwse The estmates from the less parsmonous specfcaton, whch are depcted n the second half of Table 4, suggest that assgnng a causal nterpretaton to the DD coeffcent shown n the second half of Table 2 s suspect: also the change n the probablty of enrollng at unversty enrolment between hgh school graduates aged 19 and ther peers older than 19 years s hgher across males than across females after the abolton of CMS. Insert Table 4 about here A more convncng test for the valdty of the dentfcaton assumpton s represented by a pre-program test. The dea s to check whether n the years precedng the mplementaton of the mltary reform the unversty enrolment gap between our two age-groups of hgh school graduates already exhbted a dfferent trend across males and females. Hence we perform a pre-program test usng a sample of people who fnshed hgh school n 2001. The estmates reported n Table 5 suggest that n the perod before the abolton of CMS the trend n relatve hgh school graduates aged 18 or less/ hgh school graduates aged 19 or above unversty enrolment was already dfferent across males and females. Durng such a perod ths trend was flatter across females relatve to males. More specfcally, n the less parsmonous specfcaton the dfference n the probablty of enrollng at unversty between our two age-groups of hgh school graduates was 5.4 percentage ponts hgher across males than across females. Ths fndng confrms the napproprateness of our DD dentfcaton assumpton. In lght of the evdence reported n Table 5, the DD estmates depcted n Table 2 are lkely to overestmate the effect of the suppresson of CMS on unversty enrolment. Insert Table 5 about here 16
Next we need to adjust the DD estmates reported n Table 2 n order to consder the dfferental unversty enrolment trend between our two age-groups across males and females n the years precedng the mplementaton of the mltary reform. As outlned n Secton 4, one way to account for ths dfferental trend s to employ the trpledfference model. Ths model attempts to subtract the bas revealed through the preprogram test from the effect of the abolton of CMS estmated usng the DD model. Specfcally, t tests whether n the 2004 sample the growth n unversty enrolment shown by male hgh school graduates aged 18 or less compared to ther peers aged 19 or above, relatve to the comparable growth between smlar female groups, s statstcally dfferent from the correspondng dfference-n-dfferences n the 2001 sample. These trple-dfference estmates are reported n Table 6. Whlst n our very basc specfcaton the DDD coeffcent has a postve sgn and s statstcally sgnfcant at the 5 percent level, n the full control specfcaton t retans the postve sgn but loses ts statstcal sgnfcance. Consequently, there appears to be no compellng evdence ndcatng that n the 2004 sample the gender dfference n the probablty of enrollng at unversty between hgh school graduates aged 18 or less and ther older peers has actually been drven by the abolton of CMS. Our results suggest that, not only ths gender dfference already exsted before the abolton of CMS, but also that the dfference between these two dfferences s not statstcally sgnfcant. Insert Table 6 about here Although the DDD estmates reported n Table 6 show that the abolton of CMS has no sgnfcant overall effect on unversty enrolment, t s stll possble that the effect dffers across ndvduals wth dfferent characterstcs. As outlned n the theoretcal framework set up n Secton 2, holdng wages for unversty and hgh school graduates constant, the suppresson of CMS may encourage unversty partcpaton among ndvduals wth lower dscount rates, but at the same tme t may dscourage unversty partcpaton among those wth hgher dscount rates. In the returns to educaton lterature t s generally assumed (see, for nstance, Card, 1994) that people from more advantaged famly backgrounds (.e. hgher-ncome people) tend to have lower dscount rates than those from less advantaged famly backgrounds. Hence, to test whether the 2004 mltary reform s assocated wth any heterogeneous effects, we 17
estmate our trple-dfference model for ndvduals from more and less advantaged famly backgrounds, separately. Father s occupaton s used as a proxy for famly background. Whle the frst half of Table 7 reports the DDD estmates for ndvduals whose father has ether a hgh-level occupaton or a medum-level occupaton, the second half of Table 7 presents the correspondng estmates for those whose father has ether a low-level occupaton or other/unknown occupatons 12. Insert Table 7 about here Table 7 shows some evdence of heterogeneous effects. Whle n the frst half of Table 7 the DDD coeffcent s statstcally sgnfcant at the 10 percent level and has a postve sgn, n the second half of Table 7 the correspondng coeffcent s statstcally sgnfcant at the 5 percent level but has a negatve sgn. Thus these results would seem to support the hypothess that, whle the abolton of CMS ncreased unversty partcpaton among ndvduals from more prvleged famly backgrounds, t had a negatve mpact on the enrolment of those from less prvleged famly backgrounds. Although male ndvduals wth hgher dscount rates choose lower levels of educaton, the presence of CMS, because of ts assocated costs, could have dverted them towards a longer nvestment n schoolng. On the other hand, the longer payoff perod assocated wth nvestment n unversty educaton, whch s trggered by the abolton of CMS, could encourage male hgh school graduates wth a stronger taste for schoolng (or lower dscount rates) to contnue studyng. 7. Conclusons Gven that several countres throughout the world have already abolshed or are consderng the abolton of CMS, understandng the consequences of ths measure s of ncreased concern. In ths paper we have attempted to analyse the effect of the suppresson of CMS on the decson to enrol at unversty among Italan male hgh school graduates. Studyng the factors nfluencng the transton from secondary to tertary educaton s very mportant as nvestng n hgher educaton s expected to brng sgnfcant benefts not only to ndvduals but also to socety as a whole. 12 The ncluson of geographcal area dummes n the specfcaton allows us to control for dfferences n the wage dfferental between unversty and hgh school graduates lvng n the same area. 18
The abolton of CMS may have conflctng effects on the unversty enrolment decson of male hgh school graduates. On the one hand, t may dscourage them from attendng unversty as they no longer need to contnue studyng n order to avod mltary servce. Ths argument hnges on the dea that young men typcally perceve the mltary servce to be assocated wth several costs and hence they want to postpone t as long as they can. On the other hand, however, hgh school graduates may be more wllng to nvest n unversty educaton as the dsmssal of CMS mples a longer payoff perod for ths nvestment. In Italy n 2004 those male ndvduals who successfully completed hgh school at the age of 18 or less were not at rsk of servce. Our dentfcaton strategy s to compare ther unversty enrolment decson wth that of ther older peers, relatve to the correspondng dfference between smlar female groups. Ths approach allows us to account for unobserved age-group specfc effects and common macro-effects n addton to observable characterstcs. Double-dfference estmates show that the dfference n unversty enrolment between these two age-groups of hgh school graduates s hgher across males relatve to females. However, a pre-program test reveals that n the years precedng the abolton of CMS the unversty enrolment gap between hgh school graduates aged 18 or less and ther older peers was already hgher across males than across females. In lght of ths result, we compare the gender dfference n unversty enrolment between these two groups durng the perod after the abolton of CMS wth the correspondng one durng the perod before the abolton of CMS. As the dfference between these two dfferences s not statstcally dfferent from zero, our estmates suggest that the abolton of CMS had no statstcally sgnfcant effect on unversty partcpaton rate. Although the trple-dfference estmates suggest that the abolton of CMS has no sgnfcant overall effect on unversty enrolment, we fnd some evdence of heterogeneous effects. Whle ths measure seems to ncrease unversty partcpaton among ndvduals from more advantaged backgrounds, t appears to have a detrmental effect on the enrolment of those from less advantaged backgrounds. These fndngs may have consderable polcy mplcatons. Gven that Italy s charactersed by great nequalty n access to unversty educaton (D Petro, 2008), there s the rsk that the abolton of CMS may further exacerbate ths stuaton. 19
References Angrst, J.D. (1990) Lfetme earnngs and the Vetnam era draft lottery: evdence from socal securty admnstratve records, Amercan Economc Revew, 80(3), 313-336. Angrst, J.D. (2001) Estmaton of lmted-dependent varable models wth bnary endogenous regressors: smple strateges for emprcal practce, Journal of Busness Economcs and Statstcs, 19 (1), 2 16. Angrst, J.D. and Krueger, A.B. (1989) Why do World War II veterans earn more than nonveterans?, Workng Paper 2991, Natonal Bureau of Economc Research. Angrst, J.D. and Krueger, A.B. (1992) Estmatng the payoff to schoolng usng the Vetnam-era draft lottery, Workng Paper 4067, Natonal Bureau of Economc Research Bansak, C. and Raphael, S. (2001) Immgraton reform and the earnngs of Latno workers: Do employers sanctons cause dscrmnaton?, Industral and Labor Relatons Revew, 54(2), 275-295. Bauer, T.K., Bender, S. and Schmdt, C.M. (2004) Evaluatng the labor market effects of compulsory mltary servce: A regresson-dscontnuty approach, unpublshed manuscrpt, IZA, Bonn. Baum, B.S., and Payea, K. (2004) Educaton pays 2004: The benefts of hgher educaton for ndvduals and socety, The College Board. Becker, G.S. (1964) Human Captal. New York: Columba Unversty Press. Berlnsk S. and Galan, S. (2007) The effect of a large expanson of pre-prmary school facltes on preschool attendance and maternal employment, Labour Economcs, 14 (3), 665-680. Bertola, G. and Checch, D. (2002). Sortng and prvate educaton n Italy, Center for Economc Polcy Research, Dscusson Paper, No. 3198. Borjas, G.L. (2004) Food nsecurty and publc assstance, Journal of Publc Economcs, 88 (7-8), 1421-1443. Borjas, G.L. (2003) Welfare reform, labour supply, and health nsurance n mmgrant populaton, Journal of Health Economcs, 22 (6), 933-958. Bratt, M., Broccoln, C. and Staffolan, S. (2006) Is '3+2' Equal to 4? Unversty Reform and Student Academc Performance n Italy, Unverstà Poltecnca delle Marche, Dpartmento d Economa, Workng Paper, n. 251. Buonanno, P. (2006) Long-term effects of conscrpton: Lessons from the UK, Workng Paper 2006/4, Dpartmento d Scenze Economche Hyman P. Mnsky, Unverstà degl stud d Bergamo 20
Cappellar, L. (2004) Hgh school types, academc performance and early labour market outcomes, IZA Dscusson paper no. 1048. Card, D. and Lemeux, T. (2001) Gong to college to avod the draft: the unntended legacy of the Vetnam war, Amercan Economc Revew, 91(2), 97-102. Card, D. (1994) Earnngs, schoolng and ablty revsted, Workng Paper No. 4832, Natonal Bureau of Economc Research. Chunrong, A. and Norton, E. (2003) Interacton terms n logt and probt models, Economcs Letters, 80 (1), 123-129. Cpollone, P. and Rosola, A. (2007) Socal nteractons n hgh school: Lessons from an earthquake, Amercan Economc Revew, 97(3), 948-965. D Petro, G. (2008) Changes n the extent of nequalty n access to unversty educaton n Italy, Rvsta d Poltca Economca, 1, 175-213. D Petro, G. and Cutllo, A. (2008) Degree flexblty and unversty drop-out: the Italan experence, Economcs of Educaton Revew, 27(5), 546-555. D Petro, G. and Cutllo, A (2006) Does attendng a catholc school make a dfference? Evdence from Italy, Bulletn of Economc Research, 58 (3), 193-234. Duflo, E. (2001) Schoolng and labor market consequences of school constructon n Indonesa: Evdence from am unusual polcy experment, Amercan Economc Revew, 91(4), 795-813 Hamermesh, D.S. and Trejo S.J. (2000) The demand for hours of labor: Drect evdence from Calforna, Revew of Economcs and Statstcs, 82(1), 38-47. Imbens, G.and van der Klaauw, W. (1995) Evaluatng the cost of conscrpton n the Netherlands, Journal of Busness and Economc Statstcs, 13(2), 207-215 Mansk, C.F. (1989) Schoolng as expermentaton: A reapprasal of the postsecondary dropout phenomenon, Economcs of Educaton Revew, 8(4), 305-312. Maurn, E. and Xenogan, T. (2007) "Demand for educaton and labor market outcomes: Lessons from the abolton of compulsory conscrpton n France", Journal of Human Resources, 42(4), 795-819 Neuman, S. and Wess, A. (1995) "On the effects of schoolng vntage on experenceearnngs profles: Theory and evdence", European Economc Revew, 39(5), 943-955. 21
Table 1. Mean and standard devaton (n brackets) of varables 2004 hgh school graduates 2001 hgh school graduates Aged 18 or less Aged 19 or more Aged 18 or less Aged 19 or more Enrolled at unversty=1, otherwse= 0 0.80 (0.40) 0.57 (0.49) 0.82 (0.38) 0.56 (0.50) Male=1, otherwse 0 0.47 (0.50) 0.52 (0.50) 0.41 (0.49) 0.53 (0.50) Area of resdence North-West=1, otherwse= 0 0.08 (0.27) 0.22 (0.41) 0.05 (0.22) 0.23 (0.42) North-East=1, otherwse= 0 0.04 (0.20) 0.17 (0.37) 0.03 (0.16) 0.17 (0.38) Centre=1, otherwse= 0 0.11 (0.31) 0.20 (0.40) 0.08 (0.27) 0.21 (0.41) South=1, otherwse= 0 0.77 (0.42) 0.41 (0.49) 0.84 (0.36) 0.39 (0.49) Academc ablty Faled hgh school year=1, otherwse= 0 0.01 (0.10) 0.20 (0.40) 0.01 (0.11) 0.25 (0.43) Performance at lower secondary school Pass =1, otherwse= 0 0.12 (0.33) 0.27 (0.44) 0.08 (0.28) 0.26 (0.44) Good=1, otherwse= 0 0.25 (0.43) 0.30 (0.46) 0.20 (0.40) 0.30 (0.46) Very good =1, otherwse= 0 0.27 (0.44) 0.22 (0.42) 0.29 (0.45) 0.21 (0.41) Excellent =1, otherwse= 0 0.36 (0.48) 0.21 (0.40) 0.43 (0.50) 0.23 (0.42) Hgh school degree classfcaton 60-69=1, otherwse= 0 0.17 (0.38) 0.32 (0.46) 0.21 (0.41) 0.35 (0.48) 70-79=1, otherwse= 0 0.22 (0.42) 0.26 (0.440 0.18 (0.38) 0.27 (0.44) 80-89=1, otherwse= 0 0.18 (0.38) 0.19 (0.39) 0.24 (0.43) 0.18 (0.38) 90-100=1, otherwse= 0 0.43 (0.49) 0.24 (0.43) 0.37 (0.48) 0.20 (0.40) School-related characterstcs Prvate hgh school=1, otherwse= 0 0.08 (0.27) 0.07 (0.25) 0.04 (0.20) 0.04 (0.20) Type of hgh school General =1, otherwse= 0 0.58 (0.49) 0.32 (0.47) 0.66 (0.47) 0.33 (0.47) Professonal=1, otherwse=0 0.07 (0.25) 0.20 (0.40) 0.04 (0.20) 0.18 (0.38) Techncal=1, otherwse=0 0.29 (0.45) 0.45 (0.50) 0.22 (0.42) 0.46 (0.50) Artstc==1,, otherwse=0 0.07 (0.25) 0.04 (0.19) 0.07 (0.26) 0.03 (0.18) Famly background Father s educaton Prmary school or less=1, otherwse=0 0.06 (0.23) 0.13 (0.34) 0.06 (0.23) 0.14 (0.35) Lower secondary school=1, otherwse=0 0.21 (0.41) 0.38 (0.49) 0.21 (0.41) 0.39 (0.49) Hgh school=1, otherwse=0 0.43 (0.49) 0.39 (0.49) 0.41 (0.49) 0.36 (0.48) Unversty=1, otherwse=0 0.31 (0.46) 0.10 (0.30) 0.32 (0.47) 0.11 (0.31) Mother s educaton Prmary school or less=1, otherwse=0 0.05 (0.23) 0.14 (0.34) 0.08 (0.27) 0.17 (0.37) Lower secondary school=1, otherwse=0 0.22 (0.41) 0.39 (0.49) 0.22 (0.41) 0.40 (0.49) Hgh school=1, otherwse=0 0.43 (0.49) 0.39 (0.49) 0.41 (0.49) 0.35 (0.48) Unversty=1, otherwse=0 0.30 (0.46) 0.09 (0.28) 0.30 (0.46) 0.09 (0.28) Father s occupaton Hgh-level=1, otherwse=0 0.36 (0.48) 0.17 (0.38) 0.39 (0.49) 0.20 (0.40) Medum-level=1, otherwse=0 0.43 (0.49) 0.39 (0.49) 0.39 (0.49) 0.38 (0.49) Low-level=1, otherwse=0 0.18 (0.38) 0.42 (0.49) 0.20 (0.40) 0.39 (0.49) Other-unknow1, otherwse=0 0.03 (0.18) 0.02 (0.15) 0.02 (0.15) 0.03 (0.16) Number of observatons 783 20,063 591 16,368 Survey weghts have been used 22
Table 2. Effect of the abolton of mltary conscrpton on unversty enrolment: double-dfference estmates Coeffcent Standard Error Coeffcent Standard Error Constant 0.640* 0.005 1.141* 0.012 Aged 18 or less when fnshed hgh school=1, otherwse 0 0.143* 0.020-0.047** 0.016 Male=1, otherwse 0-0.126* 0.007-0.019* 0.006 Aged 18 or less when fnshed hgh school=1, otherwse 0 * Male=1, otherwse 0 0.150* 0.029 0.095* 0.023 Area of resdence Reference category s South North-West=1, otherwse= 0 0.027* 0.007 North-East=1, otherwse= 0-0.008 0.008 Centre=1, otherwse= 0 0.003 0.007 Faled hgh school year=1, otherwse= 0-0.069* 0.007 Hgh school degree classfcaton- Reference category s 90-100 60-69=1, otherwse= 0-0.250* 0.008 70-79=1, otherwse= 0-0.146* 0.008 80-89=1, otherwse= 0-0.075* 0.008 Performance at lower secondary school- Reference category s excellent Pass =1, otherwse= 0-0.126* 0.010 Good=1, otherwse= 0-0.039* 0.009 Very good =1, otherwse= 0 0.015*** 0.009 Prvate hgh school=1, otherwse= 0-0.082* 0.011 Type of hgh school- Reference category s General Professonal=1, otherwse=0-0.495* 0.010 Techncal=1, otherwse=0-0.305* 0.007 Artstc==1,, otherwse=0-0.428 0.015 Father s educaton- Reference category s Unversty Prmary school or less=1, otherwse=0-0.108* 0.015 Lower secondary school=1, otherwse=0-0.065* 0.012 Hgh school=1, otherwse=0-0.021*** 0.011 Mother s educaton- Reference category s Unversty Prmary school or less=1, otherwse=0-0.120* 0.014 Lower secondary school=1, otherwse=0-0.071* 0.012 Hgh school=1, otherwse=0-0.003 0.010 Father s occupaton- Reference category s Hgh-level Medum-level=1, otherwse=0-0.010 0.009 Low-level=1, otherwse=0-0.054* 0.009 Other-unknow1, otherwse=0-0.096* 0.019 R-squared 0.027 0.408 Number of observatons 20,846 20,846 * denotes statstcally sgnfcant at the 1% level ** denotes statstcally sgnfcant at the 5% level *** denotes statstcally sgnfcant at the 10% level Survey weghts are used n the regresson analyss 23
Table 3. Effect of the abolton of mltary conscrpton on unversty enrolment: double-dfference estmates dfferent unversty enrolment defnton Coeffcent Standard Error Coeffcent Standard Error Constant 0.671* 0.005 1.132* 0.011 Aged 18 or less when fnshed hgh school=1, otherwse 0 0.143* 0.018-0.032** 0.015 Male=1, otherwse 0-0.113* 0.007-0.018* 0.006 Aged 18 or less when fnshed hgh school=1, otherwse 0 * Male=1, otherwse 0 0.131* 0.027 0.077* 0.022 Area of resdence Reference category s South North-West=1, otherwse= 0 0.024* 0.007 North-East=1, otherwse= 0-0.010 0.008 Centre=1, otherwse= 0 0.005 0.007 Faled hgh school year=1, otherwse= 0-0.045* 0.007 Hgh school degree classfcaton- Reference category s 90-100 60-69=1, otherwse= 0-0.237* 0.008 70-79=1, otherwse= 0-0.132* 0.008 80-89=1, otherwse= 0-0.068* 0.008 Performance at lower secondary school- Reference category s excellent Pass =1, otherwse= 0-0.128* 0.010 Good=1, otherwse= 0-0.027* 0.009 Very good =1, otherwse= 0 0.019** 0.008 Prvate hgh school=1, otherwse= 0-0.081* 0.010 Type of hgh school- Reference category s General Professonal=1, otherwse=0-0.463* 0.009 Techncal=1, otherwse=0-0.277* 0.007 Artstc==1,, otherwse=0-0.381* 0.014 Father s educaton- Reference category s Unversty Prmary school or less=1, otherwse=0-0.100* 0.014 Lower secondary school=1, otherwse=0-0.065* 0.012 Hgh school=1, otherwse=0-0.023** 0.011 Mother s educaton- Reference category s Unversty Prmary school or less=1, otherwse=0-0.109* 0.013 Lower secondary school=1, otherwse=0-0.068* 0.011 Hgh school=1, otherwse=0 0.0002 0.010 Father s occupaton- Reference category s Hgh-level Medum-level=1, otherwse=0-0.006 0.008 Low-level=1, otherwse=0-0.049* 0.009 Other-unknow1, otherwse=0-0.073* 0.018 R-squared 0.023 0.360 Number of observatons 22,859 22,859 * denotes statstcally sgnfcant at the 1% level ** denotes statstcally sgnfcant at the 5% level *** denotes statstcally sgnfcant at the 10% level Survey weghts are used n the regresson analyss 24
Table 4 Effect of the abolton of mltary conscrpton on unversty enrolment: double-dfference estmates control experment Coeffcent Standard Error Coeffcent Standard Error Constant 0.358* 0.104 1.026* 0.016 Aged 19 when fnshed hgh school=1, otherwse 0 0.356* 0.012 0.115* 0.011 Male=1, otherwse 0-0.078* 0.013-0.065* 0.011 Aged 19 when fnshed hgh school=1, otherwse 0 * Male=1, otherwse 0 0.009 0.015 0.069* 0.012 Area of resdence Reference category s South North-West=1, otherwse= 0 0.028* 0.007 North-East=1, otherwse= 0-0.007 0.008 Centre=1, otherwse= 0-0.002 0.007 Faled hgh school year=1, otherwse= 0 0.048* 0.010 Hgh school degree classfcaton- Reference category s 90-100 60-69=1, otherwse= 0-0.249* 0.009 70-79=1, otherwse= 0-0.145* 0.008 80-89=1, otherwse= 0-0.081* 0.009 Performance at lower secondary school- Reference category s excellent Pass =1, otherwse= 0-0.110* 0.011 Good=1, otherwse= 0-0.041* 0.009 Very good =1, otherwse= 0 0.010 0.009 Prvate hgh school=1, otherwse= 0-0.062* 0.011 Type of hgh school- Reference category s General Professonal=1, otherwse=0-0.487* 0.010 Techncal=1, otherwse=0-0.300* 0.007 Artstc==1,, otherwse=0-0.411* 0.016 Father s educaton- Reference category s Unversty Prmary school or less=1, otherwse=0-0.101* 0.015 Lower secondary school=1, otherwse=0-0.069* 0.013 Hgh school=1, otherwse=0-0.016 0.012 Mother s educaton- Reference category s Unversty Prmary school or less=1, otherwse=0-0.104* 0.014 Lower secondary school=1, otherwse=0-0.070* 0.012 Hgh school=1, otherwse=0-0.011 0.011 Father s occupaton- Reference category s Hgh-level Medum-level=1, otherwse=0-0.010 0.009 Low-level=1, otherwse=0-0.051* 0.009 Other-unknow1, otherwse=0-0.084* 0.019 R-squared 0.122 0.413 Number of observatons 20,063 20,063 * denotes statstcally sgnfcant at the 1% level ** denotes statstcally sgnfcant at the 5% level *** denotes statstcally sgnfcant at the 10% level Survey weghts are used n the regresson analyss 25
Table 5. Effect of the abolton of mltary conscrpton on unversty enrolment: double-dfference estmates pre-program test Coeffcent Standard Error Coeffcent Standard Error Constant 0.607* 0.006 1.144* 0.013 Aged 18 or less when fnshed hgh school=1, otherwse 0 0.225* 0.021-0.036** 0.016 Male=1, otherwse 0-0.087* 0.008 0.012** 0.006 Aged 18 or less when fnshed hgh school=1, otherwse 0 * Male=1, otherwse 0 0.062*** 0.006 0.054** 0.024 Area of resdence Reference category s South North-West=1, otherwse= 0-0.011** 0.008 North-East=1, otherwse= 0-0.001 0.009 Centre=1, otherwse= 0-0.008 0.008 Faled hgh school year=1, otherwse= 0-0.100* 0.008 Hgh school degree classfcaton- Reference category s 90-100 60-69=1, otherwse= 0-0.248* 0.009 70-79=1, otherwse= 0-0.139* 0.009 80-89=1, otherwse= 0-0.051* 0.010 Performance at lower secondary school- Reference category s excellent Pass =1, otherwse= 0-0.089* 0.011 Good=1, otherwse= 0-0.042* 0.010 Very good =1, otherwse= 0 0.001 0.009 Prvate hgh school=1, otherwse= 0-0.052* 0.014 Type of hgh school- Reference category s General Professonal=1, otherwse=0-0.557* 0.011 Techncal=1, otherwse=0-0.353* 0.008 Artstc==1,, otherwse=0-0.513* 0.017 Father s educaton- Reference category s Unversty Prmary school or less=1, otherwse=0-0.061* 0.015 Lower secondary school=1, otherwse=0-0.038* 0.013 Hgh school=1, otherwse=0 0.008 0.012 Mother s educaton- Reference category s Unversty Prmary school or less=1, otherwse=0-0.115* 0.014 Lower secondary school=1, otherwse=0-0.075* 0.012 Hgh school=1, otherwse=0-0.010 0.011 Father s occupaton- Reference category s Hgh-level Medum-level=1, otherwse=0-0.026* 0.009 Low-level=1, otherwse=0-0.076* 0.010 Other-unknow1, otherwse=0-0.030 0.019 R-squared 0.023 0.438 Number of observatons 16,959 16,959 * denotes statstcally sgnfcant at the 1% level ** denotes statstcally sgnfcant at the 5% level *** denotes statstcally sgnfcant at the 10% level Survey weghts are used n the regresson analyss 26
Table 6 Effect of the abolton of mltary conscrpton on unversty enrolment: trple-dfference estmates Coeffcent Standard Error Coeffcent Standard Error Constant 0.607* 0.005 1.132* 0.009 Aged 18 or less when fnshed hgh school=1, otherwse 0 0.225* 0.019-0.028*** 0.015 Male=1, otherwse 0-0.087* 0.007 0.009 0.006 Aged 18 or less when fnshed hgh school=1, otherwse 0 * Male=1, otherwse 0 0.062** 0.030 0.059** 0.023 Fnshed hgh school n 2004=1, otherwse 0 0.033* 0.007 0.023* 0.006 Fnshed hgh school n 2004=1, otherwse 0 * Male=1, otherwse 0-0.038* 0.010-0.025* 0.008 Fnshed hgh school n 2004=1, otherwse 0 *Aged 18 or less when fnshed hgh school=1, otherwse 0-0.078* 0.029-0.025 0.022 Aged 18 or less when fnshed hgh school=1, otherwse 0 * Male=1, otherwse 0 * Fnshed hgh school n 2004=1, otherwse 0 0.089** 0.043 0.033 0.033 Area of resdence Reference category s South North-West=1, otherwse= 0 0.009 0.005 North-East=1, otherwse= 0-0.003 0.006 Centre=1, otherwse= 0-0.002 0.005 Faled hgh school year=1, otherwse= 0-0.085* 0.005 Hgh school degree classfcaton- Reference category s 90-100 60-69=1, otherwse= 0-0.249* 0.006 70-79=1, otherwse= 0-0.142* 0.006 80-89=1, otherwse= 0-0.063* 0.006 Performance at lower secondary school- Reference category s excellent Pass =1, otherwse= 0-0.110* 0.008 Good=1, otherwse= 0-0.042* 0.007 Very good =1, otherwse= 0 0.007 0.006 Prvate hgh school=1, otherwse= 0-0.071* 0.009 Type of hgh school- Reference category s General Professonal=1, otherwse=0-0.524* 0.007 Techncal=1, otherwse=0-0.329* 0.005 Artstc==1,, otherwse=0-0.469* 0.011 Father s educaton- Reference category s Unversty Prmary school or less=1, otherwse=0-0.085* 0.010 Lower secondary school=1, otherwse=0-0.052* 0.009 Hgh school=1, otherwse=0-0.006 0.008 Mother s educaton- Reference category s Unversty Prmary school or less=1, otherwse=0-0.118* 0.010 Lower secondary school=1, otherwse=0-0.073* 0.008 Hgh school=1, otherwse=0-0.007 0.008 Father s occupaton- Reference category s Hgh-level Medum-level=1, otherwse=0-0.019* 0.006 Low-level=1, otherwse=0-0.065* 0.007 Other-unknow1, otherwse=0-0.063* 0.013 R-squared 0.025 0.422 Number of observatons 37,805 37,805 * denotes statstcally sgnfcant at the 1% level ** denotes statstcally sgnfcant at the 5% level *** denotes statstcally sgnfcant at the 10% level Survey weghts are used n the regresson analyss 27
Hgh school graduates whose father s occupaton s Low-level or Other/unknown Table 7 Effect of the abolton of mltary conscrpton on unversty enrolment: trple-dfference estmates- heterogeneous effects Hgh school graduates whose father s occupaton s Hghlevel or Medum-level Coeffcent Standard Error Coeffcent Standard Error Constant 1.096* 0.010 1.267* 0.038 Aged 18 or less when fnshed hgh school=1, otherwse 0-0.004 0.018-0.081* 0.031 Male=1, otherwse 0 0.012 0.008 0.006 0.009 Aged 18 or less when fnshed hgh school=1, otherwse 0 * Male=1, otherwse 0-0.0004 0.027 0.250* 0.049 Fnshed hgh school n 2004=1, otherwse 0 0.012 0.008 0.037* 0.009 Fnshed hgh school n 2004=1, otherwse 0 * Male=1, otherwse 0-0.010 0.011-0.044* 0.012 Fnshed hgh school n 2004=1, otherwse 0 *Aged 18 or less when fnshed hgh school=1, otherwse 0-0.038 0.025 0.056 0.046 Aged 18 or less when fnshed hgh school=1, otherwse 0 * Male=1, otherwse 0 * Fnshed hgh school n 2004=1, otherwse 0 0.070*** 0.038-0.144** 0.070 Area of resdence Reference category s South North-West=1, otherwse= 0 0.013*** 0.007 0.004 0.008 North-East=1, otherwse= 0 0.009 0.008-0.018** 0.009 Centre=1, otherwse= 0-0.005 0.007 0.005 0.008 Faled hgh school year=1, otherwse= 0-0.092* 0.007-0.077* 0.008 Hgh school degree classfcaton- Reference category s 90-100 60-69=1, otherwse= 0-0.223* 0.008-0.291* 0.010 70-79=1, otherwse= 0-0.112* 0.008-0.190* 0.010 80-89=1, otherwse= 0-0.044* 0.008-0.098* 0.010 Performance at lower secondary school- Reference category s excellent Pass =1, otherwse= 0-0.107* 0.010-0.123* 0.012 Good=1, otherwse= 0-0.033* 0.008-0.065* 0.011 Very good =1, otherwse= 0 0.019** 0.008-0.024** 0.010 Prvate hgh school=1, otherwse= -0.058* 0.010 0 Type of hgh school- Reference category s General Professonal=1, otherwse=0-0.512* 0.010-0.561* 0.011 Techncal=1, otherwse=0-0.296* 0.007-0.391* 0.009 Artstc==1,, otherwse=0-0.446* 0.015-0.521* 0.017 Father s educaton- Reference category s Unversty Prmary school or less=1, -0.109* 0.014-0.148* 0.038 otherwse=0 Lower secondary school=1, -0.076* 0.009-0.116* 0.037 otherwse=0 Hgh school=1, otherwse=0-0.024* 0.008-0.080** 0.037 Mother s educaton- Reference category s Unversty Prmary school or less=1, -0.122* 0.013-0.154* 0.023 otherwse=0 Lower secondary school=1, -0.072* 0.010-0.116* 0.022 otherwse=0 Hgh school=1, otherwse=0-0.009 0.008-0.047** 0.022 R-squared 0.403 0.370 Number of observatons 20,207 17,598 * denotes statstcally sgnfcant at the 1% level ** denotes statstcally sgnfcant at the 5% level *** denotes statstcally sgnfcant at the 10% level Survey weghts are used n the regresson analyss 28