Evaluating the Effects of FUNDEF on Wages and Test Scores in Brazil *

Transcription

1 Evaluatng the Effects of FUNDEF on Wages and Test Scores n Brazl * Naérco Menezes-Flho Elane Pazello Unversty of São Paulo Abstract In ths paper we nvestgate the effects of the 1998 reform n the fundng of fundamental educaton n Brazl (FUNDEF) on the relatve wages of publc school teachers and on the relatve profcency of publc school pupls. The evdence suggests that, on average, FUNDEF rased the publc school teachers relatve wages and mproved the relatve profcency of the publc school students. Some ndrect evdence was presented that showed that the effect of FUNDEF on profcency seems to be related to ts effect on wages and on school characterstcs. The effect on profcency seems to be concentrated n the muncpal schools n the Northeast of the country. * We would lke to thank the Mnstry of Educaton for provdng us the data used n ths analyss.

2 1) Introducton In terms of ncome dstrbuton, Brazl s one the most unequal countres n the world. 1 Educaton plays an mport role n explanng ths fact, as about 50% of the ncome dstrbuton n Brazl can be assocated wth educaton. Ths happens because returns to educaton are very hgh n Brazl and only a small proporton of the populaton has access to hgher levels of educaton. 2 Despte the fact that access to the frst schoolng year s almost unversal n Brazl, chldren from a poor background tend to drop out of the school system qute early on. 3 One the reasons behnd ths hgh drop-out rate may be the qualty of the educaton they receve n the publc system. In 1998, a reform n the fundng of the publc fundamental educaton system was ntroduced n Brazl, wth the creaton of FUNDEF (Fundo para Manutenção e Desenvolvmento do Ensno Fundamental e Valorzação do Magstéro Fund for Mantenance and Development of the Fundamental Educaton and Valorzaton of Teachng). FUNDEF man am s redstrbute resources from the rcher to the poorer regons and to ncrease publc teachers wages. The am of ths paper s to examne whether the ntroducton of FUNDEF has n fact ncreased the earnngs of the publc school teachers, relatve to ther prvate schools counterparts, and the relatve performance of publc school pupls n test scores. The Brazlan educaton system s dvded n cycles. The frst cycle (prmary educaton) conssts of four years, the second (secondary) also has four years, the thrd (hgh 1 The 10% n the top of the ncome dstrbuton approprate 50% of all ncome n Brazl. 2 See Menezes-Flho et al (2002). A college graduate earns about three tmes more than a hgh school graduate and only about 10% of the populaton has a college degree. 3 See Flmer, D and Prchett, L. (1998)

3 school) lasts three years and the fourth (college) usually lasts between four and fve years. The prmary and secondary cycles together form what s called the fundamental educaton, whch was affected by the ntroducton of FUNDEF. The system has both prvate (pad) and publc (free) schools. Fgure 1 presents the share of pupls studyng n prvate schools and the share of schools that are prvately owned n selected grades. One can notce that the share of students n prvate schools rses wth the level of educaton, whch can be explaned by the hgh drop-out rate among kds from a poor background, whch tend to study n publc schools. Moreover, the share of prvate schools s hgher than the share of students n prvate schools, especally from the 8 th grade onwards, whch means that prvate schools tend to have fewer students than the publc ones after that grade. In terms of college educaton, the stuaton s radcally dfferent, snce students from publc colleges perform much better on average n evaluaton tests than do the students from prvate nsttutons. As such, there s a very compettve exam to gan admsson nto each of the publc colleges, and students from prvate hgh schools generally do much better n these admssons exams than do pupls from the publc school system that managed to conclude hgh school. Therefore, most of the students from a poor background, whch went through the publc hgh school system, have to go to prvate colleges or try ther luck n the labor market and nequalty tends to self-perpetuate. For all these reasons, t s mportant to evaluate an educaton reform that amed at changng the fundng structure of the publc school system, n order to redstrbute resources to the poorest regons, such as FUNDEF. Barros et al (2001), usng household level data, found that the wages of publc school teachers rose by about 8% wth respect to those n the prvate sector n the southeast of Bra zl. Anuatt Neto et al (2003) also found

4 that the relatve wages of publc school teachers ncreased between 1997 and 1999, partcularly n the muncpal system n the Northeast of Brazl, whch they attrbute to FUNDEF. However, there s no study evaluatng the mpact of FUNDEF usng school level data and examnng ts effects on the relatve profcency of publc school pupls. We thnk that ths paper also relates to a broader lterature the tres to evaluate the mpact of resources spent on educaton and on teacher labor market (see Hanushek, 2003, for example). The structure of the paper s as follows. In secton 2 we descrbe the FUNDEF program and secton 3 descrbes the data. Secton 4 presents the econometrc methodology, whle the results are presented n secton 5 and the conclusons n secton 6. 2) The FUNDEF Program In each Brazlan muncpalty, the publc schools may belong to the State system or to the muncpalty system. The new Brazlan consttuton, whch took effect n 1988, stated that all States, Muncpaltes and the Federal Government had to spend a fxed share of ther tax and transfer revenues n ther publc educaton system. Ths share was equal to 25% n the cases of states and muncpaltes and 18% n the case of the federal government. Wth ths new legslaton, the amount of resources allocated to educaton ncreased, but the so dd heterogenety of the publc schools, snce rcher states wth a small share of students n ther system were spendng much more per pupl than were poor muncpaltes wth a large share of students. Moreover, there was no mechansm to enforce that the educaton

5 resources were effectvely beng spent on the educatonal system tself and not on other actvtes that could be remotely lnked to educaton. 4 The ntroducton of FUNDEF amed at changng the structure of fundng n fundamental educaton. Snce ts mplementaton (January 1 st 1998) and for a perod of 10 years, all muncpaltes and states had to spend 60% of ther educaton resources (that s, 15% of ther revenues) exclusvely wth the mantenance and development of ts fundamental educaton. However, nstead of beng drectly appled by the government unt, all resources were frstly drected to a common fund. In a second moment, the resources were redstrbuted to the states and muncpaltes, n drect proporton to the number of students enrolled n each state and muncpalty fundamental school system. Moreover, 60% of the resources receved through ths fund had to be spent wth teachers wages. Fnally, a mnmum amount of spendng per pupl was establshed, and n the cases where ths amount could not be acheved wth the fund resources alone, the federal government would complement t. Hence, FUNDEF affected the educaton system n several ways. Suppose that a muncpalty had revenues (from tax and transfers) that amounted to R$100. Wth the 1988 consttuton, t had to destne R$25 to educaton n any way t preferred. After FUNDEF, t had to donate R$15 to the fund, whereas the amount f receved back depended on the number of pupls enrolled n the fundamental educaton. If ts share of pupls was equal to ts share of resources to the fund, t would receve the same R$15 back. Moreover, at least R$9 had to be spent n teachers wages. 4 Rch muncpaltes wth a small number of publc schools, for example, spent the resources n actvtes remotely related to the educaton, lke street pavements near the school, sports gymnasum, etc..

6 Therefore, the mpact of FUNDEF on the schools and on teachers wages n a muncpalty or state depended on the amount of resources ntally allocated to the fundamental educaton system out of ts educaton budget; on the ntal share of wages out of ths amount and on ts share of enrollments as compared to ts share of revenues wthn the State. Table 1 reports the fnancal redstrbuton that took place between the each state and ts muncpaltes n 1998 for the dfferent regons. The transfers wthn a State would sum zero, were t not for the federal government transfers that complement the budget f the expendtures per student do not reach the mnmum amount. It s clear that n all regons, wth the excepton of the south-east (SE), the transfer favored the muncpaltes. Ths happened because ther proporton of enrolments was hgh relatve to the proporton of ther revenues. Fgure 2 shows the behavor the expendtures on educaton n each state as a proporton of the GDP over tme. Snce the proporton of the revenues spent on educaton n each unt should be constant over tme (determned by the Consttuton), the changes n the share of educaton expendtures should correspond to changes n the revenues/gdp raton. It s clear that there was a rse n the share of educaton expendtures n the country as a whole between 1997 and 1998, wth the man responsble beng the states and muncpaltes of the Northeast, where the rse actually startng n Therefore, a hgher share of resources was beng spent on educaton over the perod under analyss. FUNDEF establshed that 60% of all educaton resources should be spent on fundamental educaton. Fgure 3 shows that the states were, on average, already spendng more than 60% of ts educaton resources on fundamental educaton n Ths s also true for each state and muncpal system separately, except for the state system n Sao

7 Paulo, Ro de Janero and Paraná (fgure not shown). Between 1998 and 1999, one can notce a declne n share of resources accrung to fundamental educaton and a rse n the hgh school share of educaton expendtures. Fgure 4 shows that the Muncpal system as a whole was already spendng about 70% of ts educaton resources on the fundamental cycle. The São Paulo muncpaltes were the only ones spendng less than the mnmum requred, on average (fgure not shown). However, the share of resources spent on the fundamental educaton rose by about 8% between 1997 and 1998, wth a smlar declne n the amount destned to the pre-school system. Ths could be the result of the effort made by muncpaltes that were not prevously spendng the mnmum requred and had to substtute resources away from the pre-school to the fundamental educaton. Fgure 5 presents the evoluton of the total number of students n the fundamental educaton n each system (state, muncpalty and prvate schools) between 1997 and It s clear that the total number of students rose over tme, wth a rse n the number of students n the muncpal system more than compensatng for the declne n the number of students n the state system. It seems therefore that students are beng transferred fron the state to the muncpal system, whch could perhaps be assocated wth the shfts n the allocaton of the educaton resources from the states to the muncpaltes. It s nterestng to note however, that these movements to the muncpal system and away from the State system occurred even n the states whch experenced a shft n resources n the opposte drecton, lke São Paulo and Mnas Geras, whch means that t could actually reflect a trend that pre-dates the ntroducton of Fundef. In Fgure 6 we present the evoluton of the real expendtures per pupl n the fundamental educaton n the state and muncpal systems and n Brazl as a whole. It s

8 clear that there was a rse n real expendtures between 1997 and 1998, n both the state and the muncpal systems, despte the rse n the number of students, followed by a declne n the level of expendtures n the state system between 1998 and Fgures 7 and 8 present the equvalent numbers for the Northeast and Southeast regons separately. One can notce from fgure 7 that n the northeast the pattern s very smlar to the one observed for the country as a whole. The stablzaton of real expendtures between 1998 and 1999 despte fallng expendtures n both the muncpal and state systems can be explaned by the rse n the federal transfer to the unts that dd not reach the stpulated mnmum amount of expendtures per pupl. Fgure 8 shows that n the Southeast regon, where the state system was a net benefcary of the FUNDEF program, real expendtures n the muncpal system fell contnuously between 1997 and Fgure 9 shows that the number of schools offerng fundamental educaton fell between 1997 and 1999 especally due to the fall n the number of State schools, although there was a slght fall n the number of muncpal schools as well. Ths happened both n the Northeast ad n the Southeast (fgures not shown), wth the excepton of the number of muncpal schools n the southeast, whch rose between 1997 and 1998, despte the fall n real expendtures documented n the prevous fgure. Despte the fall n the number of schools, Fgure 10 shows that the total number of teachers actually rose between 1997 and 1999, manly due to the rse n the muncpal system, whch out-weghted the fall n the state system. It seems therefore that teachers also moved from the muncpal to the state system, followng ther students. Ths was true both n the Northeast and n the Southeast regons as well (fgures not shown). It s mportant to note that the number of prvate schools and of ther teachers has remaned constant over ths tme frame, snce they wll form our control group.

9 Fnally, Fgure 11 shows that average class szes remaned bascally constant between 1997 and 1999 n the system as a whole, but there was a rse n the average class sze n the muncpal system, whch was compensated by a fall n the prvate schools. Ths was especally true n the Northeast (fgures not shown). 3) Econometrc Methodology The emprcal strategy we wll follow to evaluate the mpact of the FUNDEF program s based on the dfferences-n-dfferences methodology, used n Card (1990) and descrbed n detals by Angrst and Kruege r (1999). In the frst step of hs methodology we evaluate whether the FUNDEF mpacted the publc schools teachers wages relatve to ther prvate schools counterparts. In the second step we nvestgate whether FUNDEF has mproved the profcency of the publc schools pupls wth respect to ther prvate schools counterparts. FUNDEF was ntroduced n Therefore, f FUNFED was effectve n rasng publc schools teachers wages, one should observe an ncrease n ther relatve wages n 1999 wth respect to ther relatve wages n More formally, suppose that the condtonal mean wages are defned by: E[ ] = β + γ (1) w o t s In the absence of FUNDEF, the teachers wages would be equal to the sum of a year effect that s common to all schools and a school effect (publc or prvate) that s fxed over tme. 5 Suppose also that the effect of FUNDEF was to rase wages by a constant, that s: E w ] = E[ w ] +δ (2) [ f o 5 As we do not observe the same schools n 1997 and 1999, we prefer to work wth the condtonal mean functon.

10 Ths means that the teachers wages n both prvate and publc schools n 1997 and 1999 can be wrtten as: w t = β + γ + δf + ε (3) t s t where F s a dummy varable equal to 1 f school was drectly affected by FUNDEF, that s, t was a publc school observed n Dfferentatng the wages across schools and years, we have: { E[ w / s = { E[ w / s = pub, t = 99] E[ w / s = prv, t = 99]} pub/ t = 97] E[ w / s = prv, t = 97]} = δ (4) As many school and students characterstcs may have changed between 1997 and 1999, and n publc schools dfferently from n the prvate ones, we wll stack the mcro data for all schools and years and estmate an equaton lke: w t = β + γ + δf + θx + λs + ε (5) t s t t t where X s a vector of school characterstcs and S s a vector of the teachers characterstcs. The man dentfcaton assumpton we need s that: { E[ ε { E[ ε / X, S, s = / X, S, s = pub, t = 99] E[ ε pub/ t = 97] E[ ε / X, S, s = prv, t = 99]} / X, S, s = prv, t = 97]} = 0 (6) that s, there could be no changes n the unobserved characterstcs of the publc schools or of ther teachers, relatve to the prvate ones, between 1997 and Snce we have no dea about the plausblty of ths assumpton, we wll nclude as many observable characterstcs as possble gven our data set, and compare ther means between 1997 and 1999.

11 In the second step we wll use the same methodology, but usng the students performance n test scores nstead of the teachers wages as the dependent varable. Frstly, we wll estmate an equaton of profcency at the pupl level, as a functon of the dummy varables descrbed above and of the students characterstcs, n order to nvestgate whether FUNDEF has rased average test scores of the publc school students, as compared to prvate schools ones, uncondtonally: y = β + γ + δf + αz + ε (6) t s t t where Z s a vector of the students characterstcs. We then ntroduce the school characterstcs to examne ts effect onδ : y = β + γ + δf + αz + θx + ε (7) t s t t and fnally, we ntroduce the teachers characterstcs and ther wages: y = β + γ + δf + αz + θx + λs + κ W + ε (8) t s t t t t If the effect of FUNDEF on the students test scores was the result of mprovements of the school characterstcs, we should observe a declne n δ once we ntroduce the m n the regresson, and the same s vald for the teachers wages. Ths s the methodology we use to verfy how (f at all) has FUNDEF rased the profcency of the students n publc schools. 3) Data The data we use n ths part of the project come from SAEB (Sstema de Avalação do Ensno Básco) a survey carred out by the Mnstry of Educaton. Ths data set has nformaton on the test scores of a sample of students n both publc and prvate schools n

12 1995, 1997, 1999 and As FUNDEF was ntroduced n 1999 (see above) we wll only use the 1997 and 1999 waves. Each student n each school was tested for hs/her profcency n one out of three possble subjects: Portuguese, Mathematc s or Scences. The nformaton on the teacher responsble for ths subject and the school characterstcs were matched to each student to form the fnal data set. In ths verson of the paper, we wll use only the test scores of the students that were n the 8 th grade, the last grade of the fundamental educaton. The data set contans a very detaled set of characterstcs of each student, school, teacher and drector for all schools n the sample. Table 2 presents the summary statstcs of the students characterstcs. The percentage of boys s slghtly hgher n the prvate schools, although grls form the majorty of students n both systems. It s nterestng to note that the mean age n the prvate schools s much lower than n the publc ones, wth may reflect late start or hgher grade repetton. The dfferences n the famly background are qute strkng, as about 48% of the mothers of prvate school students have a college degree as compa red to 9% n the publc schools! Ths dfference remaned bascally the same n The percentage of pupls that have faled the grade exams n the past s very hgh, reachng 27% n the prvate and 59% n the publc system n 1997, declnng n both systems by about 5 percentage ponts between 1997 and Table 3 presents a summary of the teachers characterstcs. The frst thng to notce s that sample szes ncreased between the 97 and 99 sample. Ths may bas our estmaton results f t affected the composton of the publc and prvate school teachers dfferently n terms of unobservable characterstcs. 6 One can notce that about 93% of prvate school 6 Both the 1997 and the 1999 samples are representatve at the level of each State. It s not clear why the sample szes ncreased n the perod.

13 teachers were college educated n 1997, as compared to 80% n the publc schools, a dfference of about 13 percentage ponts. In 1999 the dfference n terms of college educaton was n the range of 10 percentage ponts. In terms of experence and age, there were no marked dfferences between the 1999 and the 1997 sample means. In terms of average wages however, we can see that the dfference between the prvate and publc schools that was R$512 n 1997, declned to about R$290 n 1999, a reducton of about 43%! 7 Table 4 presents the descrpton of some school characterstcs. Ths s the most problematc part of the data, snce there are not many school characterstcs n the 1999 survey, and there are dfferences n the way that the questons were formulated between the 1997 and the 1999 surveys. Therefore, a comparson between the 1997 and 1999 data s problematc, and we should concentrate on the comparson between publc and prvate schools n each year. 8 One can notce that n 1997 about 97% of the prvate school had computers, whereas only 37% of the publc schools had at least one. In 1999 the queston asked about the number of computers used by students, and so the proporton decreased to 66% n the prvate schools and 17% n the publc system. It s nterestng to note that the dfference n terms of the drector s wage between the prvate and publc schools has also declned between 1997 and 1999, from approxmately R$900 to about R$490, a change of about 45%, n lne wth the teachers wages. 4) Results 7 The orgnal nformaton on teachers wages was n the form of ntervals, so we used the mdponts of each nterval to construct the means, and converted nto real wages, usng the average nflaton rate n the perod. 8 We recently notced that the 1999 teacher survey does contan nformaton on several school characterstcs that could be used and compared to the ones present n the 1997 survey. The next verson of the paper wll nclude these varables.

14 Table 5 below presents the results of regresson that looks at the determnants of teachers wages for the pooled 1997 and 1999 sample. The frst column shows that real wages ncreased n 1999, that was n publc schools are lower on average then n the prvate ones and that wages n scence teachers are lower than those of the Mathematcs teachers. The second column however shows that there was an ncrease of about 32% n the average wages of publc school teachers n 1999, as compared to ther prvate schools counterparts. Column 3 then ncludes the teachers characterstcs and we can see that FUNDEF coeffcent declnes a lttle, but remans statstcally sgnfcant. In column (4) we control for the school characterstcs, ncludng the drector s wage and the coeffcent remans statstcally sgnfcant, meanng that Fundef rased the publc school teachers wages by about 20%. Interestngly enough there s a postve correlaton between the drector s and the teachers wages, whch may reflect matchng or school unobserved characterstcs. In Table 6 we present the results of the second state regresson, about the determnants of the students performance n test scores. In the frst column, one can note that there was a declne n students profcency n 1999 and that students fare worse n Portuguese and Scences than n Mathematcs (the omtted category). Moreover, boys tend to perform better than grls, young pupls than older one, and whte then non-whtes. Famly background, as measured by mother s schoolng does has a strong mpact on test scores. Moreover, pupls that faled n the past (a measure of unobserved ablty) tend to do worse. The results as a whole are n lne wth other studes n the school profcency lterature. In column (2) we nclude the publc school ndcator and the FUNDEF dummy, that s, the nteracton between publc school and It seems that publc school students tend

15 to do worse than prvate school ones, but that ths dfference has declned between 1997 and 1999, after the ntroducton of FUNDEF. Column (3) ncludes the teachers characterstcs (other than ther wages) and there s hardly any change n the FUNDEF dummy, mplyng that the change n teacher characterstcs s not responsble for the mprovement n publc school students test scores. In column (4) we nclude the teachers wages as an addtonal regressor. It attracts a postve and statstcally sgnfcant coeffcent and t reduces the FUNDEF dummy by almost a half. Ths n ndrect evdence that the mprovements n the performance of the publc school pupls may be partly explaned by the hgher wages of the publc school teachers. Column (5) then ncludes the school characterstcs and the magntude of the FUDNEF coeffcent declnes further. Fnally, column (6) ncludes the drector s wages and ths leads to a further declne n the FUNDEF coeffcent, whch s now not statstcally dfferent from zero. Ths evdence suggests that the mprovements n the publc school performance may be explaned by the mprovements n teachers and drectors wages and other school characterstcs. Gven the changes n the process of fundng educaton ntroduced by FUNDEF, whereby most of the muncpaltes were net benefcares, wth the excepton of the southeast, we expect the effects of FUNDEF to dffer substantally between states and muncpaltes and across dfferent regons. We therefore repeated the exercses above separately for the state and muncpal systems n two regons, Southeast and Northeast. The results are presented n tables 7 to 14 In terms of teachers wages, tables 7 to 10 show that they ncreased for publc school teachers n both the state and muncpal system, n the Northeast as well as n the Southeast, as ndcated by column (2) n each table. It s nterestng to note however, that they ncreased by more n the Southeast, where, as can be noted from column (1), the

16 dfferences n wages between the publc and prvate schools were hgher to start wth. Moreover, t seems that the ncreases n the southeast are robust to the ncluson of other teacher and school characterstcs, as shown by columns (3) and (4), whle n the state system n the Northeast, the ncluson of school characterstcs and drectors wages wpes out the FUNDEF effect. In the muncpal level however, the FUNDEF effect s robust to the ncluson on other characterstcs. Tables 11 to 14 repeat the exercses of table 6 for both systems n the Northeast and Southeast regons and the results look qute nterestng. Whle the dfferences between the prvate and publc schools n terms of profcency were broadly smlar n all systems and regons (about 30 ponts), one can only observe a rse n the profcency of publc schools n the muncpal system n the Northeast. In the other system/regons, the FUNDEF dummy attracted a coeffcent that s not sgnfcantly dfferent from zero. Moreover, n the Northeast muncpal level, the magntude of the coeffcent also declnes when other possble FUNDEF outcomes are ncluded n the regresson, lke teachers wages (column 5), and the drectors wage and other school characterstcs (column 6). 5) Conclusons In ths paper we nvestgate the effects of a 1998 reform n the fundng scheme of fundamental educaton n Brazl (FUNDEF) on the publc teachers wages and on the profcency of publc school pupls. The evdence suggests that, on average, FUNDEF rased the teachers relatve wages and mproved the relatve profcency of the publc school students. Some ndrect evdence was presented that showed that the effect of FUNDEF on profcency seems to be related to ts effect on wages and on school

17 characterstcs. The effect on profcency however, seems to be concentrated n the muncpal schools n the Northeast of the country. Tabela 1: FUNDEF Fnancal Impact by Brazlan Regons-(1998) Regon State Government Contrbuton Revenues from FUNDEF to FUNDEF Prncpal Federal Total (A) Compl. (B) B A Muncpal Government Contrbuton Revenues from FUNDEF to Prncpal Federal Total FUNDEF (A) Compl. (B) B - A N 731,7 655, ,6 (10) 262,5 338,5 46,6 385,1 122,6 NE 1810,6 1203,2 157,9 1361,3 (449,5) 966,1 1573,8 216,1 1789,9 823,8 SE 4327,7 4500,2-4500,2 173,2 1973,3 1799,9-1799,9 (173,4) S 1283,4 1152,5-1152,5 (130,9) 717,2 848,1 848,1 130,9 CO 452,0 446,3-446,3 (5,7) 247,0 252,5-252,5 5,5 Brazl 8604,7 7957,8 223,9 8181,7 (423,0) 4166,1 4818,8 262,7 5075,5 909,4 Source Educaton Secretary MEC R$ Mllons

18 Table 2: Descrptve Analyze - Student's Characterstcs Prvate Publc Prvate Publc Boy ,0% ,3% ,7% ,8% Whte ,8% ,0% ,8% ,1% (Age 7) 7,4 9,0 7,3 8,6 (1,2) (2,2) (1,1) (1,8) Mother s educaton = ,7% ,7% ,3% ,3% Secondary Mother s educaton = ,7% ,7% ,2% ,1% hgh school Mother s educaton = ,2% ,0% ,5% ,0% college Faled Before ,4% ,4% ,8% ,4% Scences ,0% ,8% ,0% ,2% Score Portuguese ,4% ,0% ,3% ,0% Score Mathematcs ,6% ,2% ,8% ,8% Score Number of Observatons

19 Table 3: Descrptve Analyze - Teacher's Characterstcs Teacher's Characterstcs Prvate Publc Prvate Publc Hgh School 68 6,8% ,0% ,2% ,4% College ,2% ,5% ,7% ,4% Experence >= 6 and <= ,8% ,1% ,3% ,8% Experence >= 11 and <= ,9% ,6% ,8% ,1% Experence >= 16 and <= ,4% ,9% ,7% ,0% Experence >= ,0% ,5% ,2% ,9% Age >= 21 & <= ,1% ,2% ,4% ,5% Age >= 26 & <= ,0% ,0% ,5% ,0% Age >= 31 & <= ,7% ,5% ,3% ,9% Age >= 36 & <= ,8% ,3% ,4% ,7% Age >=41 & <= ,1% ,5% ,4% ,9% Age >= 46 & <= ,3% ,1% 190 8,1% ,1% Age >= ,1% 200 5,7% 113 4,8% 270 5,6% Men ,2% ,8% ,0% ,4% Wage Number of observatons 1.104,08 592, ,21 711,16 (625,28) (396,28) (629,46) (456,48)

20 School's Table 4: Descrptve Analyze - School's Characterstcs Characterstcs Prvate Publc Prvate Publc Computer ,2% ,5% ,9% ,0% Clean classroom 97 83,6% ,3% ,7% ,5% Clean bathroom 93 80,2% ,4% ,2% ,8% Drector s Wage Number of Observaton 1.981, , , ,64 (845,29) (644,92) (872,47) (553,26)

21 Table 5: Frst Stage Dependent varable Ln Wage Model 1 Model 2 Model 3 Model 4 Year 99 0,165-0,073-0,017 0,072 (0,013) (0,025) (0,024) (0,022) Publc -0,432-0,646-0,577-0,375 (0,014) (0,023) (0,022) (0,024) Year 99 * publc 0,319 0,291 0,209 (0,029) (0,026) (0,025) Scences -0,106-0,106-0,051-0,055 (0,015) (0,015) (0,013) (0,013) Portuguese 0,002 0,002-0,021-0,019 (0,014) (0,014) (0,013) (0,012) Dummes of states Yes yes yes yes Teacher s characterstcs Hgh School 0,200 0,260 (0,145) (0,140) College 0,702 0,686 (0,145) (0,140) Years of experence 0,211 0,207 >= 6 and <=10 (0,016) (0,016) Years of experence 0,327 0,315 >= 11 and <=15 (0,019) (0,018) Years of experence 0,412 0,395 >= 16 and <=20 (0,022) (0,021) Years of experence 0,560 0,543 >= 21 (0,024) (0,024) Men 0,067 0,062 (0,012) (0,011) Age >= 21 & <= 25 0,212 0,193 (0,054) (0,052) Age >= 26 & <= 30 0,347 0,323 (0,053) (0,051) Age >= 31 & <= 35 0,337 0,306 (0,054) (0,052) Age >= 36 & <= 40 0,313 0,294 (0,054) (0,052) Age >= 41 & <= 45 0,365 0,336 (0,055) (0,053) Age >= 46 & <= 50 0,363 0,328 (0,057) (0,055) Age >= 51 0,333 (0,059) 0,311 (0,057)

22 (School s characterstcs) Clean classroom 0,061 (0,013) Clea bathroom -0,017 (0,011) Computer 0,060 (0,012) Drector s Ln wage 0,229 (0,010) Constant 6,886 7,044 5,753 4,047 (0,037) (0,041) (0,157) (0,169) Nº of Observaton F(k, n-k ) 139,88 142,13 210,34 214,60 Prob > F 0,000 0,000 0,000 0,000 R-squared 0,245 0,254 0,436 0,477 *Robust standard-errors n brackets.

23 Table 6: Second Stage Dependent Varable Student s profcency Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Year 99-6,272 (0,345) -9,352 (0,631) -9,086 (0,631) -9,250 (0,626) -7,824 (0,635) -7,028 (0,638) Publc -28,597 (0,577) -28,443 (0,580) -25,588 (0,599) -21,470 (0,659) -20,361 (0,664) Year 99 * publc 2,893 2,966 1,664 0,759 0,287 (0,734) (0,733) (0,734) (0,744) (0,745) Scences -1,039-1,235-0,868-0,560-0,612-0,723 (0,406) (0,392) (0,397) (0,397) (0,396) (0,396) Portuguese -7,221-7,347-7,481-7,310-7,357-7,402 (0,403) (0,391) (0,415) (0,414) (0,413) (0,413) Dummes of states yes yes yes yes yes yes Students Characterstcs Boy 6,218 5,595 5,654 5,625 5,598 5,579 (0,340) (0,330) (0,329) (0,329) (0,328) (0,328) Age 7-12,454-8,892-8,721-8,567-8,439-8,343 (0,666) (0,643) (0,644) (0,643) (0,642) (0,642) (age 7) 2 0,384 0,258 0,252 0,245 0,242 0,238 (0,032) (0,031) (0,031) (0,031) (0,031) (0,031) Whte 5,370 3,648 3,611 3,518 3,405 3,415 (0,351) (0,341) (0,341) (0,340) (0,339) (0,339) Mother s educaton = 5,624 5,108 4,883 4,783 4,738 4,678 secondary (0,605) (0,597) (0,595) (0,594) (0,594) (0,594) Mother s educaton = 20,834 13,360 12,806 12,585 12,246 12,101 hgh school (0,700) (0,695) (0,695) (0,694) (0,693) (0,693) Mother s educaton = 33,677 19,556 18,838 18,042 17,416 17,045 college (0,748) (0,762) (0,763) (0,762) (0,761) (0,761) Faled Before -13,455-12,084-12,170-12,112-12,057-12,061 (0,446) (0,432) (0,431) (0,430) (0,430) (0,429) Teachers Characterstcs Ln Wage 5,183 4,819 3,970 (0,296) (0,296) (0,304) Hgh School -0,301 2,799 3,118 2,533 (3,239) (3,255) (3,230) (3,223) College -4,907-2,397-1,911-1,523 (0,482) (0,500) (0,499) (0,499) Experence 2,131 1,027 1,159 1,217 >= 6 and <=10 (0,513) (0,515) (0,515) (0,514) Experence 2,032 0,297 0,370 0,466 >= 11 and <=15 (0,587) (0,593) (0,592) (0,592)

24 Experence 2,914 0,951 0,960 1,120 >= 16 and <=20 (0,697) (0,705) (0,704) (0,704) Experence >= 21 3,876 1,280 1,486 1,807 (0,774) (0,787) (0,786) (0,785) Men 0,374 0,105-0,032-0,109 (0,365) (0,364) (0,364) (0,363) Age >= 21 & <= 25 3,714 2,796 2,042 2,035 (1,450) (1,462) (1,457) (1,454) Age >= 26 & <= 30 2,941 1,474 0,945 1,025 (1,431) (1,444) (1,439) (1,437) Age >= 31 & <= 35 1,470-0,008-0,561-0,688 (1,452) (1,465) (1,460) (1,457) Age >= 36 & <= 40 3,005 1,682 1,257 1,322 (1,483) (1,496) (1,490) (1,488) Age >= 41 & <= 45 4,890 3,149 2,651 2,611 (1,509) (1,522) (1,517) (1,514) Age >= 46 & <= 50 2,786 1,121 0,625 0,463 (1,569) (1,581) (1,576) (1,574) Age >= 51 1,304-0,308-0,984-0,867 (1,662) (1,674) (1,669) (1,667) School Characterstcs Drector s Ln wage 3,202 (0,278) Clean classroom 1,098 1,139 (0,361) (0,360) Clean bathroom 2,990 2,858 (0,396) (0,396) Computer 5,110 4,638 (0,404) (0,405) Constant 314, , , , , ,038 (3,424) (3,297) (3,572) (4,075) (4,081) (4,334) Nº of Observaton F(k, n-k ) 655,63 758,33 566, , , ,78 Prob > F 0,000 0,000 0,000 0,000 0,000 0,000 R-squared 0,267 0,308 0,311 0,315 0,317 0,318 *Robust standard-error between parentheses.

25 Table 7: Frst Stage Dependent varable Ln Wage Northeast Brazl STATE SCHOOLS Model 1 Model 2 Model 3 Model 4 Year 99 0,064-0,095-0,031 0,215 (0,031) (0,052) (0,049) (0,042) Publc -0,343-0,532-0,511-0,032 (0,030) (0,053) (0,050) (0,053) Year 99 * publc 0,282 0,228-0,001 (0,063) (0,058) (0,052) *Robust standard-error between parentheses. Table 8: Frst Stage Dependent varable Ln Wage Northeast Brazl MUNICIPAL SCHOOLS Model 1 Model 2 Model 3 Model 4 Year 99 0,065-0,112-0,037 0,197 (0,034) (0,051) (0,049) (0,042) Publc -0,413-0,636-0,593-0,092 (0,030) (0,057) (0,052) (0,053) Year 99 * publc 0,317 0,365 0,131 (0,068) (0,060) (0,053) *Robust standard-error between parentheses. Table 9: Frst Stage Dependent varable Ln Wage Southeast Brazl STATE SCHOOLS Model 1 Model 2 Model 3 Model 4 Year 99 0,198-0,140-0,073-0,062 (0,033) (0,050) (0,046) (0,045) Publc -0,713-1,101-0,869-0,835 (0,032) (0,053) (0,050) (0,059) Year 99 * publc 0,584 0,426 0,425 (0,066) (0,059) (0,062) *Robust standard-error between parentheses. Table 10: Frst Stage Dependent varable Ln Wage Southeast Brazl MUNICIPAL SCHOOLS Model 1 Model 2 Model 3 Model 4 Year 99 0,175-0,116-0,062-0,017 (0,036) (0,049) (0,045) (0,043) Publc -0,391-0,737-0,623-0,575 (0,033) (0,055) (0,052) (0,061) Year 99 * publc 0,512 0,419 0,392 (0,068) (0,063) (0,064) *Robust standard-error between parentheses.

26 Table 11: Second Stage Dependent Varable Student s profcency - Northeast Brazl STATE SCHOOLS Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Year 99-7,017-9,260-9,038-9,628-6,487-4,163 (0,749) (1,070) (1,080) (1,059) (1,087) (1,110) Publc -26,202-25,826-22,119-13,813-11,695 (1,154) (1,179) (1,189) (1,320) (1,329) Year 99 * publc 1,624 1,729 0,312-2,252-3,856 (1,450) (1,458) (1,446) (1,457) (1,467) *Robust standard-error between parentheses. Table 12: Second Stage Dependent Varable Student s profcency - Northeast Brazl MUNICIPAL SCHOOLS Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Year 99-7,014-9,470-9,261-9,751-6,669-4,718 (0,762) (1,070) (1,079) (1,059) (1,086) (1,109) Publc -30,118-30,118-25,710-18,767-16,466 (1,245) (1,258) (1,267) (1,377) (1,400) Year 99 * publc 6,040 7,081 4,436 1,703 0,811 (1,459) (1,467) (1,463) (1,475) (1,478) * Robust standard-error between parentheses. Table 13: Second Stage Dependent Varable Student s profcency - Southeast Brazl STATE SCHOOLS Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Year 99-6,105-8,088-7,339-7,339-6,369-6,347 (1,041) (1,465) (1,479) (1,474) (1,497) (1,504) Publc -34,423-33,315-29,324-23,757-23,717 (1,596) (1,642) (1,869) (2,188) (2,201) Year 99 * publc 0,247-1,140-2,948-4,554-4,571 (2,008) (2,046) (2,080) (2,183) (2,185) *Robust standard-error between parentheses. Table 14: Second Stage Dependent Varable Student s profcency - Southeast Brazl MUNICIPAL SCHOOLS Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Year 99-7,131-8,175-7,484-7,447-5,760-5,678 (1,045) (1,467) (1,476) (1,471) (1,510) (1,515) Publc -27,484-26,625-23,736-16,953-16,926 (1,628) (1,641) (1,720) (2,052) (2,052) Year 99 * publc 0,797-0,342-2,526-6,632-6,659 (2,005) (2,029) (2,063) (2,179) (2,179) *Robust standard-error between parentheses.

27 Fgure 1 - Share of Prvate Schools % st grade 4th grade 8th grade 11th grade college students schools Fgure 2 - Expendtures on Educaton per GDP by Regon 8,00 7,00 6,00 5,00 % 4,00 3,00 2,00 1,00 0, BRAZIL NORTHEAST SOUTHEAST

28 Fgure 3 - Expendture Share n Each Cycle - State System 70% 60% 50% 40% 30% 20% 10% 0% pre-school fundamental hgh school college Fgure 4- Expendture Share n Each Cycle - Muncpal System 80% 70% 60% 50% 40% 30% 20% 10% 0% pre-school fundamental hgh school college

29 Fgure 5- Number of Students n Fundamental Educaton - Brasl total state muncpo prvate Fgure 6- Real Expendtures per Pupl - Fundamental Educaton - BRAZIL total state muncpo

30 Fgure 7- Real Expendtures per Pupl n Fundamental Educaton - NE total state muncpo Fgure 8 - Real Expendtures per Pupl - Fundamental Educaton -SE total state muncpo

31 Fgure 9 - Number of Schools - BR Total State Muncpo Prvate Fgure 10 - Number of Teachers - BR Total State Muncpo Prvate

32 Fgure 10 - Class Szes - BR Total State Muncpo Prvate 7 -References Anuatt Neto, F., Fernandes, R. and Pazello, E. (2003) Avalação dos Saláros dos Professores da Rede Públca de Ensno Fundamental em Tempos de FUNDEF, Unversdade de São Paulo- mmeo. Barros, R., Mendonça, R. and Blanco, F. (2001). O Mercado de Trabalho para Professores no Brasl, Anas do XXIX Encontro Naconal de Economa ANPEC, Salvador-BA. Hanushek, E.(2003) The Falure of Input-Based Schoolng Polces, The Economc Journal, vol. 113, pp. F64-F98.

33 Menezes-Flho, N., Fernandes, R. and Pcchett, P (2002). Rsng Human Captal but Constant Inequalty: The Educaton Composton Effect n Brazl, Unversty of Sao Paulo, mmeo.