Documento de Trabajo No. 01/00 Marzo Wage Differentials Between the Formal and the Informal Sector in Urban Bolivia.

Size: px
Start display at page:

Download "Documento de Trabajo No. 01/00 Marzo Wage Differentials Between the Formal and the Informal Sector in Urban Bolivia."

Transcription

1 Documento de Trabajo No. 01/00 Marzo 2000 Wage Dfferentals Between the Formal and the Informal Sector n Urban Bolva por Trne Monsted

2 Wage Dfferentals Between the Formal and the Informal Sector n Urban Bolva Trne Monsted 1 ILSECUCB La Paz, Bolva (February 2000) Summary: Ibs paper analyses wage dfferentals between the formal and nformal sectors n urban Bolva usng household survey data. As m other studes the wage dfferental between the formal and the nformal sector s found to be qute large. The wage gap s estmated by a human captal model correctng for selectvty bas usng Heckman s twostep procedure. A surprsng result s that mgrants tend not be employed n the formal sector. Another notceable result s that the probablty of enterng the formal sector was a hump shaped functon of alttude probably because the government ctes (La Paz and Sucre) are located at md alttudes. The wage generatng processes for the two sectors are qute dfferent, and the results ndcate that productvty s relatvely more mportant for wages n the nformal sector than n the formal sector. It s also found that returns to educaton are generally hgher for formal sector workers. In a nonlnear specfcaton of returns to schoolng the concluson s that basc educaton s not rewarded n the labour market, therefore serously questonng the qualty of the educatonal system. An Oaxaca decomposton of the wage dfferental between the two sectors shows that there are dfferences n characterstcs between formal and nformal sector workers, but these characterstcs are also rewarded dfferently n the two sectors. The analyss also confrms that educaton and experence are the man factors contrbutng to the wage dfferentals between sectors. 1 Ths paper was prepared durng my stay at IISECUCB as a vstng researcher m the fall of am very grateful to IISEC and Lykke Andersen for gvng me the possblty to stay m La Paz. 1 would also lke to thank Osvaldo Nna and Mguel Vera for provdng the data used n ths paper. Addtonally, comments from Lykke Andersen and Mette Yde Skaksen have been very useful. (The author s emal: eco941634@oecon.au.dk). 1

3 1. Introducton Tradtonal human captal theory predcts that educaton and tranng are the man determnants of earnngs and thereby of poverty, whch explans ther mportance as polcy varables. A hgher level of schoolng and experence mples a hgher level of ncome (e.g. Andersen, 1998 and Jensen et al, 1997). Bolva s the poorest country n South Amerca, thus poverty allevaton s a serous consderaton, whch s why earnngs are n focus. In addton the exstence of dual labour markets makes t mportant to determne the dstngushng features of these markets. The purpose of ths paper s to examne wage dfferentals n Bolva especally between the formal and nformal labour markets. Ths s mportant because dfferent mechansms may be workng at the two markets, whch mples that polces have to be constructed partcularly to the market n queston. The data comes from the Natonal Statstcal Offce (INE) and are collected n 8 rounds between 1989 and 1995 n the man ctes n Bolva. The paper s organzed as follows. Frst, some data ssues and defntons of the relevant varables wll be presented. Next, some descrptve statstcs wll provde an overvew of the labour markets n urban Bolva ncludng the wage dfferental between the formal and nformal labour markets, whch s the focus of ths paper. Thrd, a human captal model correctng for sample selecton bas wll be estmated assumng both lnear and nonlnear returns to educaton. Fourth, an Oaxaca decomposton of the wage gap wll be calculated and the last secton concludes. 2. The formal and nformal labour markets n Bolva The dvson of the labour market nto a formal and an nformal market s sgnfcant n most developng countres, where persons who cannot fnd employment n the formal labour market are forced to employ themselves n other ways. Ibs s also called hdden unemployment, as the people employed n the nformal market would be unemployed n an economy wth only a formal market. Normally t s observed that the productvty level at the nformal market s lower than that prevalng at the formal market, whch can be explaned by several factors. Frst, dfferences n technology s the man reason for varance n productvty. The formal market normally has access to more advanced equpment than the nformal market has, whch mples a hgher productvty for workers employed n the formal labour market. Second, people workng n the nformal market may have fewer sklls and lower abltes mplyng lower productvty for those workers. Snce t s expected that productvty wll be lower n the nformal sector, wages wll also be expected to be lower. Ibs analyss tres to dentfy the factors that determne partcpaton n the two sectors and the dstngushng features of the wage generatng processes. These processes are expected to dffer whch s why the dvson s mportant. 2.1 Data and defntons The data comes from 8 rounds of household surveys (Encuesta Integrada de Hogares) collected by the natonal statstcal offce (INE) n Bolva. Unavalablty makes t mpossble to use round 2 n ths paper. The questons have been posed to people lvng n the 2

4 largest ctes n Bolva, and the subsequent analyss therefore apples only to the urban areas. Panel data models would have used the nformaton n the sample more effcently snce tme seres nformaton s ncluded n these models, but the present dataset does not allow for such an analyss. Ibs s because the sample s not dentcal between rounds. Therefore, the model presented wll only use cross sectonal nformaton. The workng populaton s defned as all respondents aged between 12 and 66 havng a strctly postve ncome. Ths populaton s dvded nto two groups, those employed n the formal market and those employed n the nformal market. The dataset does not contan any explct varable descrbng whether a person belongs to the formal or the nformal sector, but t s possble to construct such a varable from a combnaton of some of the other varables. When asked about ther occupatonal status seven dfferent categores were possble: worker, employed, employer, professonal ndependent, selfemployed, domestc worker and famly worker. A rough dvson of the two last categores as the nformal market and the fve frst as the formal may be approprate, but the data allow for some further adjustments. These lead to the followng defnton of the nformal sector. A person employed as a domestc worker or famly worker havng a postve salary wll always be ncluded n the sample of nformal market workers. In addton the type of frm n combnaton wth the form of payment, avalablty of socal securty, membershp of a unon 2 and the status of the job as permanent or temporary have been used. If a person not employed as a domestc worker or famly worker, s employed outsde the publc sector and gets hs or her salary as nknd compensaton or per job accomplshed, s not a member of a unon or has access to a socal securty scheme and has a temporary employment status, ths person wll be ncluded n the nformal sector. Ibs dvson leads to the followng dstrbuton of formal and nformal sector workers n the dfferent rounds. Table 1: Partcpaton Rates Round Partcpaton Rate Formal Sector % 1(1989) 46, (1990) 43,40 (45,47%) (1991) 45,77 (62,5 1%) (1992) 45,26 (59,81%) (1993) 47,52 (61,37%) (1994) 48,37 56,27% (1995) 49,33 (76,35%) 6799 Source: Author s calculatons (77,78%) Informal Sector 3031 (54,53%) 3217 (37,49%) 3429 (40,19%) 3077 (38,63%) 2840 (43,73%) 2161 (23,65%) 1942 (22,22%) As can be seen. from table 1, the partcpaton rate averages 46%, that s, 46% of the populaton between 12 and 66 years are actually workng. Of these, some are employed at the formal market and others at the nformal market. There s an ncreasng share of employed at the formal labour market. Ibs pattern could be completely random though. Frst, the sample 2 It not possble to determne whether an ndvdual s a member of a unon or not n ah rounds. Ths s because n some rounds the queston has not been posed. The varable s ncluded n rounds 1 and 5, not n the other rounds. 3

5 s not dentcal over tme; every year dfferent people have been asked dfferent questons. Second, n some rounds t has not been possble to dstngush between people beng members of a unon or not whch agan may have led to a downward bas n the share of employed at the nformal markets. Fnally, although the defnton of the nformal market has been approxmately the same across rounds, a lower share of formal sector workers mght occur f one used other dvson crtera. Ibs paper wll contnue wth ths dstncton though, snce there s no obvous way to mprove the dvson crtera from the data at hand. 2.2 Earnngs Earnngs are measured n the surveys as the ncome a partcular person s recevng. In table 2 the wage gap between the two sectors s obvous. Table 2: Wage Gap Between the Formal and the Informal Sectors n Bolva Round Formal Sector Informal Sector Tstatstc a Mean Std. Dev. Mean Std. Dcv. 1 (1989) 2,25 0,06 1,76 0,07 277,11 3 (1990) 2,94 0,14 1,70 0,06 476,78 4 (1991) 3,20 0,08 2,08 0,07 666,16 5 (1992) 3,60 0,08 2,40 0,08 651,82 6 (1993) 4,33 0,11 2,65 0,07 709,83 7 (1994) 4,77 0,08 2,12 0, ,0 8 (1995) 5,39 0,11 2,45 0, ,1 a) H 0 : Formal wage = Informal wage Source: Author s calculatons In all years, the wage for formal sector employees has been hgher than for those employed n the nformal sector, and the gap has ncreased substantally durng the perod under nvestgaton. Ibs gap between wages n the two sectors reflects the lower productvty n the nformal labour market, but also other effects lke generally lower educaton levels n the nformal sector, a hgher proporton of people not havng Spansh as ther mother tongue etc. The rest of ths paper wll be concerned wth an analyss of ths wage gap n order to dentfy the dstngushng features of the wage generatng processes n the two markets. 3. The Wage Gap Ibs secton wll try to dentfy some of the factors nfluencng the dfference n wages between the formal and nformal labour markets n Bolva. Several ndvdual characterstcs may turn out to be sgnfcant explanatons of the wage gap. Frst, human captal theory predcts that educaton and experence are the man determnants of earnngs, as also presented by Mncer (1974). Other characterstcs lke ethncty, regon of resdence, and gender are also ncluded here as determnants of earnngs. In secton 4 a human captal model correctng for selectvty bas wll be estmated, takng ah these aspects nto account smultaneously. 3.1 Educaton and experence As mentoned above educaton and experence are mportant determnants of earnngs. A hgh level of educaton generally leads to hgher earnngs, as does experence. Educaton can 4

6 be measured ether by completed level or by years of educaton. Experence s measured by on the job experence and general potental experence at the labour market. 3 Table 3: Educaton Levels Across Sectors Round Formal Sector Informal Sector Tstatstc a N Mean Std. Dev. N Mean Std. Dev. 1 (1989) 3 (1990) 4 (1991) 5 (1992) 6 (1993) 7 (1994) 8 (1995) ,24 9,38 12,16 14,09 13,90 10,46 13,48 0,085 0,058 0,083 0,13 0,15 0,071 0, ,61 8,41 9,40 9,70 9,85 7,82 8,73 0,074 0,064 0,073 0,092 0,091 0,09 0, ,6 709,3 1537,9 1600,2 1238,9 1412,5 1804,5 a) H 0 : Mean educaton n formal sector = Mean educaton n nformal sector Source: Author s calculatons The educatonal dfferences between the two sectors are llustrated n table 3, measured as years of educaton. Educaton s thus unevenly dstrbuted between the two sectors. In all rounds, years of educaton for nformal sector workers are sgnfcantly lower than that of formal sector workers. Earnngs are postvely related to educaton (e.g. Wood and Patrnos, 1994). And snce educaton s lower n the nformal sector, earnngs n ths sector are also lower. 3.2 Ethncty In most countres wth dfferent ethnc groups; ndgenous people usually earn less than no ndgenous. Usng language as a proxy for ethncty, 4 the followng results emerge. Table 4 Ethncty and Wages 4.1 Round 1, 1989 Informal Formal 4.2 Round 3, 1990 Informal Formal Indgenous 1,57 1,69 Indgenous 1,55 2,09 NonIndgenous 1,96 2,50 NonIndgenous 1,77 3, Round 6, 1993 Informal Formal Indgenous NonIndgenous 2,24 2, Round 7, 1994 Indgenous NonIndgenous Informal Formal 1,97 4,17 2,22 5,20 Mean wage for each sector, condtonal on ethncty Source: Author s calculatons 3 4 Experence s calculated as age years of educaton 6, where 6 s the age where schoolng normally starts. Unemployment s the reason why ths s only potental experence. See Oaxaca (1973) for further problems wth usng ths measure. People of ethnc orgn are defned as those speakng an ndgenous language. 5

7 Indgenous people tend to earn less than nonndgenous regardless of sector choce and the dfference s sgnfcant. 5 Ibs could ndcate that ndgenous are ether less productve or possess fewer sklls than nonndgenous, or dscrmnaton aganst ndgenous people. 3.3 Alttude An earler study of the Bolvan labour market (Andersen, 1999) shows that alttude has sgnfcant explanatory power for wages. Ths paper wll examne ths result n order to see whether t stll holds after takng the dual labour markets nto account. Fgure 1: Alttude and Wages Ctes: 1=Cobja (0.221), 2=Trndad (0.236), 3=Santa Cruz (0.416), 4=Tarja (1.866), 5=Cochabamba (2.558), 6=Sucre (2.790), 7=La Paz (3.640), 8=Oruro (3.709), 9=E1 Alto (3.848), 1O=Potos (4.070). Heght above sea level (n km) s mentoned n parentheses. Source: Author s calculatons As s vsble from fgure 1, there are qute large dfferences between the formal and nformal sector wages, also n the dfferent regons. Wages are generally fallng wth alttude, whch could be a reflecton of the lower productvty n the ctes hgher aboye sea level than others Gender Normally women earn less than men, whch s also llustrated n fgure 2. It s seen that the dfference between genders n the nformal sector s even bgger than the dfference between genders n the formal sector. After havng dentfed some explanatons of the wage gap, a model ncorporatng sector choce and the possble dfferences n the determnaton of wages n the two sectors, wll be presented n the next secton. 5 6 Even though results are only presented usng rounds 1 and 8, the general pcture carres over to the other rounds Usng a ttest shows sgnfcant dfferences n ah cases. Results are not reported here, but are avalable upon request. 6 Even though results are only presented usng rounds 1 and 8, the general pcture carres over to the other rounds. 6

8 Fgure 2: Gender and Wages Source: Author s calculatons. 4. Wage Estmaton Ths secton wll look more closely at the earnngs dfferental between the formal and the nformal sectors n urban Bolva. Estmaton of a human captal model wth correcton for sample selecton bas wll be carred out. Then returns to educaton under dfferent assumptons about the functonal relatonshp wll be tested and, lastly, a decomposton of the wage dfferental wll be presented. 4. The Model The man pont of nterest s estmaton of a wage equaton of the type proposed by Mncer (1974). However, n a sample lke the one at hand, there are a dvson of ndvduals nto two dfferent sectors, and ths dvson may not be random. Ths possble nonrandomness of the sample mples that OLS may not be consstent, and the soluton to ths problem s to apply a 2step procedure proposed by Heckman (1976, 1979). The frst step s to estmate a probt selecton equaton, whch determnes the probablty of enterng the formal sector 7 (one mnus ths probablty s thus the probablty of enterng the nformal sector) gven some ndvdual characterstcs. The equaton wll look lke F = β `Y + v, where F s a bnary choce varable, whch takes the value 1 f the worker s employed n the formal sector, and O f not. Y s a vector of ndvdual characterstcs expected to nfluence the choce of sector. The method of estmaton s maxmum lkelhood. From ths equaton t s possble to calculate the probablty of beng n the formal sector gven that you are a member of the urban labour force. Ths probablty s called lambda, and when t s ncluded n the earnngs equatons n the next step, t corrects for a possble sample selecton bas. 7 It could have been nterestng to nvestgate whether selfselecton has an effect on partcpaton n the labour force n general. However, ths s outsde the scope of ths paper, and for now t s assumed that labour force partcpaton s random. 7

9 An earnngs equaton wll be estmated separately for each market because t s not the same process that generates wages n the two sectors,.e. the determnants may dffer sgnfcantly between sectors. Addtonally, for the sake of dentfcaton of the emprcal model t s mportant not to nclude all varables from the selecton equatons n the wage equatons. Ibs problem s easly solved, more varables are relevant for earnngs than for sector choce and some of the proxes n the selecton equatons are exchanged wth ther true varables,.e. experence. The dependent varable s log wages n each sector, and the estmated equatons wll have the followng form. Ln( W Ln( W f ) ) = = α γ + + f β β X X + + w u f, where the superscrpt f denotes the formal sector and the nformal sector. X s a vector of ndvdual characterstcs expected to nfluence the ncome generatng process n each sector, and u and w are resduals. The reason for takng the logarthm to wages s that the dstrbuton of earnngs s skew, and better results emerge when the logarthm s used. The method of estmaton s thus OLS, whch produces unbased and consstent estmates of the coeffcents. (See Heckman 1976, 1979 or Greene, 1997 for more techncal detals). 4.2 The selecton equaton The results of estmaton are presented n the table below. The dependent varable s a bnary choce varable wth the value 1 when an ndvdual belongs to the formal sector and zero f the ndvdual belongs to the nformal sector. The pont of departure was to nclude the followng explanatory varables n the selecton equaton: years of educaton, age, age squared (both as a proxy for experence), alttude, alttude squared (both as a proxy for regon /productvty), whether the ndvdual s a household head, female, marred and mgrant. In the table only the sgnfcant coeffcents are reported (tstatstcs n parentheses). As s vsble from the table, educaton ncreases the probablty of employment n the formal sector sgnfcantly. Ibs s expected snce the tasks performed by the formal sector are n general more complex and therefore requre more sklled labour. Experence as approxmated by age seems to have a hump shaped relaton wth the estmated probablty. Untl the age of 49 8 the probablty of beng employed formally s ncreasng wth age, thereafter the probablty s decreasng. One possble nterpretaton s that the formal sector manly employs ndvduals n ther most productve years and the nformal sector s stuck wth the young and nexperenced or the old, both groups havng a lower productvty than those employed n the other sector. One very sgnfcant fndng s that the relaton between the probablty of beng employed n the formal sector and alttude of resdence s hump shaped. Ibs means that the probablty of formal employment s ncreasng for resdents n the lowlands and valleys, but 8 Ths s an outweghed average of the optmum from the seven estmated equatons. he lowest age s 44 years n round 1 and the hghest s 52 years n round 6. 8

10 s decreasng n the hghlands. 9 Ths s probably because the two government ctes (Sucre and La Paz, ctes 6 and 7) contan most of the publc sector jobs, whch account for a large share of formal jobs. It s also consstent wth the result that ndgenous people have a hgher probablty of nformal employment than nonndgenous, because the densty of the ndgenous populaton s ncreasng wth alttude (Wood and Patrnos, 1994). Table 5: Probt Estmaton of the Probablty of Beng n the Formal Sector Intercept 4,31 (91,40) 5,38 (176,68) 5,42 (171,87) 5,43 (163,32) 5,50 (140,73) 5,95 (280,66) 5,92 (277,04) Years of Educaton 0,027 (20,99) 0,0048 (6,76) 0,0030 (4,15) 0,00328 (6,03) 0,0030 (4,38) 0,00323 (8,83) 0,0028 (7,52) Age 0,008 (3,06) 0,021 (12,64) 0,0205 (11,49) 0,021 (11,34) 0,024 (11,46) 0,011 (9,05) 0,001 (8,11) Age Squared 0,00009 (2,70) 0,00021 (10,26) 0,00021 (9,33) 0,00021 (9,22) 0,00023 (9,06) 0,00011 (7,38) 0,0001 (6,41) Alttude 0,23 (14,62) 0,179 (20,87) 0,189 (19,31) 0,188 (19,12) 0,16 (12,80) 0,135 (18,31) 0,13 (17,63) Alttude Squared 0,043 (11,82) 0,029 (14,48) 0,03 1 (13,97) 0,03 1 (13,77) 0,022 (7,92) 0,019 (11,40) 0,018 (10,82) Head 0,0152 0,010 (1,68) (1,27) Marred 0,034 (4,66) 0,043 (5,59) 0,049 (6,12) 0,053 (5,68) 0,024 (4,55) 0,018 (3,35) Female 0,09 (8,12) 0,06 (6,87) 0,052 (6,64) 0,06 (8,01) 0,028 (6,29) 0,016 (3,47) Mgrant 0,041 (2,77) 0,025 (2,90) 0,046 (2,24) 0,010 (1,11) 0,017 (1,33) 0,018 (1,33) Ethncty 0,056 0,008 0,0 10 0,017 (4,64) (1,14) (1,26) (3,27) N Source: Author s calculatons; tc sgnfcance level s 10%. Beng a household head s sgnfcantly ncreasng the estmated probablty n two cases out of seven, thus ths s not a very robust result. The effect of beng female s sgnfcantly postve, but vanes over tme n magntude. One mght argue that the probablty of beng employed s lower for women than for men, but once they get nto the labour force, they have a hgher probablty of enterng the formal sector than ther male counterparts. The reason why women are generally less lkely to work than men s that women have other consderatons before enterng the labour market. They may have chldren requrng maternal care etc., whch may make women more reluctant to enter the labour market as frequently as men do. Beng marred ncreases the estmated probablty as s often the case n such estmatons. Ibs may be due to marrage beng correlated wth some other unobserved characterstcs lke e.g. culture. Mgraton s a sgnfcant factor explanng employment n the two sectors. Mgraton s defned as a person who has moved n the last 5 years 10 and the relaton s qute robust. A 9 10 The probablty of employment n the formal sector s ncreasng for an ndvdual lvng n areas below 3230 meters on average. After ths pont the probablty s decreasng. Ths means that ndvduals lvng n La Paz, El Alto, Potos and Oruro have a lower probablty of enterng tc formal than ndvduals lvng n ctes stuated lower. In some rounds t was not possble to use ths defnton because only questons regardng changes of locaton durng the past year were asked. Ths was tc case n round 4 and 8. However, tc results do not change n these cases. 9

11 recently mgrated person has a hgher probablty of beng employed n the nformal sector, whereas a nonmgrant s normally employed n the formal sector. There are several plausble reasons for ths. One factor could be that mgrants may come from rural areas, where schoolng s generally lower and qualfcatons n general are worse. Another reason s the fact that mgrants may have moved locaton because ther earnngs n the prevous locaton were nsuffcent. Snce earnngs and qualfcatons n general are related postvely (Wood and Patrnos, 1994), the reason for low earnngs stays wth the ndvdual even after movng. Therefore the probablty of employment n the formal sector decreases wth mgraton. A thrd reason may be that when a person mgrates, he or she does not have any connectons n the new locaton whch could mean that t gets harder to obtan employment n the formal sector. 4.2 The earnngs equatons As a startng pont the followng explanatory varables are tested, n order to see how well they explan wages. Educaton, general experence and experence n current job, both squared to allow for a hump shaped profle of experence. The varable controllng for selecton bas, lambda, s also ncluded, n order to se whether there s postve selfselecton n to the formal sector and negatve selfselecton nto the nformal sector. The sgn of the coeffcent to lambda measures the extent to whch mean ncomes n the sector n queston are lower or hgher than the populaton mean. Therefore a negatve coeffcent n the formal sector ndcates that on average persons employed n ths sector earn more than the total populaton. On the other hand, for the nformal sector a postve coeffcent to lambda would ndcate the same. In ths case there s postve selfselecton nto each sector, ah ndvduals choose the sector where they do best compared to the average. If the coeffcent to lambda turns out to be nsgnfcant, t ndcates that no selfselecton takes place. Alttude wll be ncluded because t has been shown to have sgnfcant explanatory power for wages (Andersen, 1999). The effect of beng the man provder to a famly, measured as the household head, gender, and economc sector wll also be ncluded. In addton language measures, degree of llteracy and mgraton wll be ncluded such as to llustrate dscrmnaton of ethnc mnortes, llterates and new resdents. The table shows the sgnfcant results from the estmaton of wages for the formal and the nformal sector respectvely. For the formal sector, t can be seen that years of educaton s a very sgnfcant factor explanng the wages n the sector. There are large returns to educaton n ths sector of the economy. The return to general experence s hump shaped as expected n rnost cases. On the other hand returns to experence n current job seem to be almost lnearly related to earnngs. The coeffcent to lambda, the varable controllng for selfselecton, s sgnfcantly negatve n ah rounds, ndcatng postve selfselecton nto the formal sector. Ths s an mportant fndng because t ndcates that the formal sector s better at attractng ndvduals wth the best unobserved characterstcs such as ablty, motvaton etc. compared to the nformal sector. Alttude does not only have explanatory power for the probablty of employment n the formal sector, t also explans wages very well. The hgher an ndvdual hves the less he/she earns. The relaton s lnear as opposed to the probt equaton estmated above. Ths means that people employed n the formal sector n the hghlands earn sgnfcantly less than ther colleagues n the valleys and the lowlands. 10

12 Beng llterate affects earnngs sgnfcantly n only one year. Ths may be due to the fact that educaton measures approxmately the same effect causng multcollnearty n the model. Beng a woman sgnfcantly ncreases the probablty of gettng employment n the formal sector, but once a female s employed, she earns sgnfcantly less than her male counterpart. Household heads generally earn more than people who are not the man provders n a famly. Table 6: Formal Sector Estmatons Intercept 1,017 0,6 1,102 1,041 (2,8) (1,677) (3,375) (1,917) Years of educaton 0,05618 (17,735) 0,08814 (27,973) 0,0963 (16,976) 0,07736 (19,536) 0,09551 (23,969) 0,09973 (27,726) 0,106 (41,182) Experence 0,03777 (8,955) 0,02827 (8,445) 0,01123 (4,66) 0,01887 (3,397) 0, (3,931) 0,02174 (5,213) 0,03 (8,168) Experence squared 0,00057 (6,523) 0,00039 (5,841) 0,00026 (2,339) 0,00021 (2,577) 0,0003 (3,818) Job experence 0, (3,846) 0, (2,622) 0,00021 (3,821) 0, (2,545) 0, (5,123) 0, (4,222) Job exp. Squared 0,00000 (1,9) Lambda 1,073 (11,901) 1,099 (10,833) 2,712 (4,236) 1,420 (2,269) 1,988 (3,239) (2,7) 0,451 (6,221) Alttude 0,111 (10,222) 0,17 (19,084) 0,162 (7,858) 0,156 (8,148) 0,151 (7,628) 0,124 (4,945) 0,123 (14,002) Ethncty 11 0,202 0, ,129 0,0709 (6,134) (2,557) (4,073) (2,72) Illteracy 0,293 (2,712) Woman 0,241 (6,552) 0,163 (4,118) 0,401 (12,667) 0,224 (6,811) 0,136 (4,198) Head 0,210 (6,436) 0,282 (11,020) 0,143 (3,772) 0,115 (3,587) 0,08958 (2,623) Publc 0,138 0,174 Admn. (3,736) (3,903) R 2 0,293 0,405 0,252 0,349 0,355 0,449 0,808 Fstatstc 153, ,388 97, ,922 25, , ,501 H 0 : All coef = 0 (0,000) (0,000) (0,000) (0,000) (0,000) (0,000) (0,000) N Source: Author s calculatons; tc sgnfcance level s 10%. Employment n the publc sector ncreases earnngs n some cases and when the varable s sgnfcant t s postve. However the result does not seem partcularly robust, therefore no frm concluson can be reached here. The choce of sector seems not to be of mportance for earnngs n the formal sector whch s opposte of the fndngs for the nformal sector as wll be llustrated next. Another result that emerges s that mgraton s not sgnfcantly nfluencng earnngs n the formal sector. The reason may be that mgraton nfluences the sector choce, but once employment n a sector s obtaned, whether an ndvdual has lved n the area for a long 11 Ths varables cannot be constructed 1991, 1992 and 1995 whch s why tc varable s not ncluded n these years. 11

13 tme or not s no longer of mportance when explanng the wage settng process between the two sectors. Table 7: Informal Sector Estmatons Intercept 2,41 1,117 (3,289) (1,41) Years of educaton 0, (4,243) 0,08759 (23,594) 0,09425 (25,352) 0,08117 (25,872) 0,105 (31,208) 0,0574 (9,792) 0,05737 (8,534) Experence 0,02899 (4,064) 0,02529 (6,578) 0,03692 (9,260) 0,02865 (7,401) 0,0408 (11,028) 0,01539 (2,686) 0,0049 (1,74) Experence 0, , , ,0004 0, ,00028 squared (2,399) (4,518) (6,408) (4,402) (7,211) (2,808) Job experence 0,001 (3,059) 0, (6,285) 0, (4,814) 0, (5,449) 0,00435 (4,647) 0,0006 (3,429) 0,00049 (4,576) Job exp. Squared 0,00000 (2,13) 0,00000 (4,545) 0,00000 (2,997) 0,00001 (3,091) 0,00000 (2,047) Lambda 0,467 (2,903) 1,883 (22,418) 1,052 (11,188) 0,491 (5,243) 1,093 (11,029) 3,432 (3,310) 1,09 (0,97) Alttude 0,144 (7,734) 0,16 (18,22) 0,153 (17,092) 0,176 (19,378) 0,127 (12,095) 0,232 (7,05) 0,166 (4,566) Ethncty 0,103 (3,189) Woman 0,237 0,154 0,172 0,147 0,101 0,24 (2,599) (4,285) (4,289) (3,987) (2,734) (4,825) Head 0,149 (4,174) 0,08737 (2,273) 0,231 (6,782) 0,08525 (2,24) 0,08205 (1,992) 0,157 (3,424) Mgrant 0,165 0,132 (2,110) (3,024) Trade 0,279 (2,219) 0,224 (5,399) 0,211 (5,070) 0,133 (3,322) 0,112 (2,554) 0,129 (2,298) Manufacturng 0,153 (4,052) 0,154 (4,4) 0,166 (4,736) 0,116 (3,224) 0,181 (2,843) Utlty 0,474 0,255 (2,746) (1,958) Constructon 0,267 0,138 0,125 0,146 (6,082) (3,269) (2,853) (3,941) Hotel 0,33 0,329 (4,8 14) (2,755) Agrculture 0,472 0,176 0,271 (5,041) (1,923) (2,772) Transport 0,177 (3,948) 0,316 (3,236) 0,781 (4,096) Mnng 0,197 0,335 (2,051) (3,783) R 2 0,194 0,309 0,437 0,545 0,355 0,264 0,202 Fstatstc H 0 : All coef = 0 13,310 (0,000) 115,972 (0,000) 131,591 (0,000) 230,974 (0,000) 235,302 (0,000) 48,731 (0,000) 35,111 (0,000) N Source: Author s calculatons; the sgnfcance level s 10%. In the nformal sector, the pcture s more vared. The returns to educaton are stll very sgnfcantly postve, but n most cases less than for ndvduals employed n the formal sector. 12 Ths means that for people employed n the nformal sector, one addtonal year of schoolng would contrbute sgnfcantly to earnngs but less so than n the formal sector. 12 The average return to schoolng n tc formal sector s 0,08846 whereas t s only 0,0735 n tc nformal sector. 12

14 Returns to general experence are hump shaped as n the formal sector, whch means that prevous experence ncreases earnngs to a certan pont beyond whch experence decreases earnngs. Ibs s probably related to the fact the elderly people are less productve than younger ones who have been workng for some years. The returns to experence n current job are also of a hump shaped character, even more so than n the formal sector. Ibs means that earnngs n the formal sector may be generated by tenure rather than qualfcatons, whereas tenure s less mportant n the nformal sector and productvty declnes after a certan tme perod n the same job. The coeffcent to the selecton varable, lambda, s negatve, whch ponts to negatve selfselecton nto the nformal sector, because the formal sector was the reference category when estmatng the probablty of gettng employment n each sector. Whereas the formal sector s best at attractng ndvduals wth favourable unobserved characterstcs, the nformal sector has to employ whoever s left. Ths fndng suggests that formal sector workers would earn more regardless of sector choce, mplyng that the same son of abltes are rewarded n both sectors. Alttude s mportant for earnngs; the hgher an ndvdual lves, the less he/she earns. On average the coeffcents to alttude s hgher for the nformal sector, than for ts formal counterpart. 13 Ibs means that earnngs are more senstve to alttude n the nformal sector than n the formal sector and could be an ndcaton of productvty beng relatvely more mportant for wages n the nformal sector. In the formal sector on the other hand, there are ndcatons that productvty s not the man determnant of earnngs snce returns to experence n current job s not hump shaped and margnal productvty s usually fallng. Ethncty does not nfluence earnngs as much as n the formal sector, another ndcaton of earnngs beng determned by productvty n ths sector to a hgher degree than n the formal sector, assumng that ndgenous people are as productve as nonndgenous. Thus, dscrmnaton does not seerm to be a problem n the nformal sector maybe because productvty s more mportant than other ndvdual characterstcs. Beng female stll decreases earnngs sgnfcantly, ndcatng dscrmnaton n ths case f t s assumed that women have the same productvty as men. Ths s not the case, however, f t s taken nto account that nformally employed women often fulfll other dutes such as chldcare whle workng. Poor famles tend to get more chldren than nonpoor (Wood and Patrnos, 1994), and snce nformal workers earn less than ther formal sector counterparts they are lkely to get more chldren too. Therefore ths fndng also suggests payment accordng to productvty. Household heads earn more than other people employed n the nformal sector. Beng a mgrant does n some cases nfluence earnngs, and n ths case postvely. Ths means that beng a mgrant s not only mportant for sector choce; when a mgrant s employed n the nformal sector he/she s rewarded wth a hgher wage than the average employee. The result should be drawn wth cauton however snce the varable only turns out sgnfcant n two cases. Nevertheless ths s more than n the case of the formal sector, and agan productvty may be the explanaton. Informal earnngs generally seem to be more senstve to choce of lne of busness than formal sector earnngs. Employment n trade, manufacturng, the hotel sector or agrculture sgnfcantly decreases earnngs n some cases. On the other hand employment n 13 The smple average s 0,142 for the formal sector and 0,165 for the nformal sector. 13

15 utlty, constructon, mnng and transport to a certan degree, ncreases nformal sector earnngs. It seems that the equaton has more explanatory power for the formal sector than for the nformal sector judged by the hgher R 2 n the former. 14 So far the analyss has assumed that returns to schoolng were related to earnngs n a loglnear way. Ths assumpton may show up to be too restrctve, however, snce dfferent levels of schoolng may have dfferent returns. Ths s a result of some levels of schoolng beng more effcent n rasng an ndvdual s margnal product than other levels. The next secton elaborates ths possblty n depth by allowng for dfferent levels of schoolng to have dfferent returns. 4.3 Earnngs Equatons wth NonLnear Returns to Schoolng The nonlnear returns to schoolng are modeled wth dummy varables ndcatng the hghest level attaned. Hence there s a dummy varable for each level of educaton rather than one varable measurng years of educaton. The models from the prevous secton are appled agan, adjusted for these new educatonal varables. The reason for keepng the two sectors separated n ths secton as well s that dfferent sectors may gve dfferent returns due to several factors: Imperfect markets, barrers to entry n a sector, or a specfc level of educaton may be rrelevant n a partcular sector. As s demonstrated n the prevous secton there are rather large dfferences between the two sectors when t comes to the wagesettng process and the factors nfluencng wages. The regressons have no schoolng as the reference category, and the man results are shown n table 8. The rest of the regresson results do not dffer much from the results presented above and are not reported here. Table 8: NonLnear Returns to Educaton, Formal Sector None Prmary 0,605 (7,042) 0,07694 (0,43) 0,677 (6,716)) 0,65 (7,425) 0,951 (8,947) 0,492 (8,027) 0,791 (11,50) Intermedate 0,475 (5,757) 0,313 (1,766) 0,629 (6,687) 0,555 (6,586) 0,723 (6,863) 0,42 (7,126) 0,655 (10,41) Medate 0,231 (3,316) 0,572 (3,36) 0,379 (4,42) 0,329 (4,58) 0,498 (4,965) 0,238 (4,687) 0,345 (6,485) Normal 0,125 (1,697) 0,855 (5,092) 0,103 (1,134) 0,09507 (1,224) 0,09461 (0,866) 0,29 (4,84) 0,368 (6,284) Unversty 0,47 (7,05) 1,362 (8,154) 0,407 (4,627) 0,449 (5,998) 0,41 (4,037) 0,605 (11,239) 0,8 (15,031) Techncal 0,101 0, ,121 0,115 (1,292) (0,129) (1,368) (1,032) Techncal Medate 0,797 (4,474) 0,18 (2,592) 0,361 (4,672) Techncal 1,045 0,338 0,512 Superor (5,864) (4,30) (5,922) Source: Authors calculatons, sample szes as n table 6 14 There are sgnfcant dfferences across gender and ethncty n the earnngs regressons. Further analyss of these dfferences shows that only devatons n the numencal value of the coeffcents dffer across groups, not sgns, and/or relatve mportance. 14

16 It s seen that returns to educaton are not lnear n years of schoolng. 15 The lberalzaton largely gnores the negatve returns for the low levels of educaton. In the formal sector the returns to prmary, ntermedate and medate levels of educaton are largely negatve. Ths means that persons possessng relatvely low levels of educaton earn less than those wth no schoolng at all. Indvduals who have obtaned a hgher level of educaton on the other hand are secured postve returns accordng to these estmates, especally those havng a unversty degree. In the nformal sector prmary and ntermedate educaton gves the same return (or less) than no schoolng. Thus, there s no ncentve to obtan low levels of educaton n ether sector. Hgher levels of educaton tend to yeld (slghtly) postve returns n the nformal sector as well. Basc school sklls are n general not rewarded n ether labour market. Ths can be ether because the demand for sklls s low or because the accumulated sklls are not useful n the labour market. The relatvely large dfferences n returns across levels of educaton ndcate that the lnear specfcaton s not approprate for measurng returns to educaton. Addtonally t s worth notcng the low/negatve returns to basc educaton. Ibs serously questons the qualty of schoolng n Bolva; the results ndcate that the value of addtonal educaton s very low, n many cases even negatve. For a country wth relatvely large problems wth poverty, ths s an area worth further nvestgaton and nvestment. It s known that educaton decreases poverty, but f t s not worth gong to school, people may choose workng nstead, thus not reducng poverty n the long run. Table 9: NonLnear Returns to Educaton, Informal Sector None Prmary 0,0315 (0,217) 0,222 (1,246) 0,102 (0,832) 0,101 (0,755) 0,769 (6,231) 0,0075 (0,102) 0,214 (2,24) Intermedate 0, (0,6) 0,01 (0,58) 0,267 (2,192) 0,153 (1,156) 0,699 (5,833) 0,02935 (0,394) 0,071 (0,74) Medate 0,15 (1,151) 0,273 (1,643) 0, 478 (4,121) 0,414 (3,286) 0,531 (4,617) 0,124 (1,687) 0,041 (0,42) Normal 0,439 (0,971) 0,778 (4,265) 1,13 (7,436) 0,821 (4,819) 0,0268 (0,16) 0,319 (1,269) 0, (0 334) Unversty 0,71 (4,18) 0,936 (5,688) 1,232 (10,557) 1,232 (9,941) 0,279 (2,359) 0,745 (7,991) 0,553 (4,683) Techncal 0,233 1,042 1,017 0,156 (1,215) (8,45) (7,719) (1,274) Techncal Medate 0,505 (2,863) 0,253 (1,741) 0,07214 (0,416) Techncal 0,519 0,798 0,579 Superor (2,825) (4,121) (2,575) Source: Authors calculatons, sample szes as n table Oaxaca Decomposton The wage gap between the two sectors can be dvded between dfferences n characterstcs and dfferences n rewards to these characterstcs. An Oaxaca decomposton makes ths dstncton possble (Oaxaca, 1973). Assumng that all workers would be pad lke nformal 15 One could also have searched for degree effect,.e. f there are any dfferences between ndvduals who have completed a certan degree and others who have not. However ths s outsde tc scope of ths paper. 15

17 sector workers on average, t s possble to construct a decomposton between how much of the wage dfferental s due to dfferences n characterstcs and how much s due to dfferences n rewards on the two markets. Lettng W be the nformal sector wage and W f be the formal sector wage, the followng decomposton makes t possble to dfferentate between the factors mentoned before: Ln ( W ) Ln ( W ) = ΔZ` ˆ β Z f `Δ ˆ, β f where Z s a vector of average characterstcs from each sample, A denotes dfferences, and βˆ s a vector of estmated coeffcents from the earnngs equatons n secton 4.2. The frst pan of the rght hand sde of the equaton s thus the share of the wage gap that can be ascrbed to dfferences n characterstcs, whereas the second pan s due to dfferences n rewards. Therefore the nterestng pan of ths analyss s the dfferences n rewards; f the two sectors reward the same characterstcs dfferently, t ndcates that there are dfferent wage settng mechansms at the two markets. The results are as follows for round 1 and Table 10: Oaxaca Decomposton, Round 1, 1989 Intercept Years of educaton Experence Experence Squared Job Experence Job Ex. Squared Lambda Alttude Woman Head Ethncty Publc Admn. Analphabetc Trade Ln (monthly wage) Dfference Note: Job experence s measured n weeks. Source: Author s calculatons Average Characterstcs Decomposton Formal Informal Dfferences n characterstcs Dfferences n rewards ,24 6,61 0,115 0, ,69 22,28 0,104 0, ,32 706,02 0,081 0, ,69 115,13 0,173 0, ,7 0,107 0,146 0,46 0,51 0,023 0,278 2,46 2,02 0,063 0,08 1 0,31 0,52 0,05 0,072 0,61 0,46 0 0,127 0,38 0,52 0 0,077 0, ,023 0,02 0,11 0 0,006 0,06 0,34 0,079 0,016 5,52 5,19 0,34 0,0898 0,1634 It s seen that for round 1, approxmately 27% of the wage dfferental can be accounted for by dfferences n characterstcs, whereas dfferences n rewards account for approxmately 49% of the wage gap. 17 The dfferences n rewards and characterstcs are prmarly explaned by returns to educaton and experence. Ths means that n the nformal sector workers tend to have worse characterstcs than workers employed n the formal sector. In addton formal sector workers are rewarded dfferently than nformal sector workers after havng controlled for dfferences n characterstcs (the fourth column n table These two rounds were chosen because they represent dfferent tme perods and nclude approxmately tc same varables. Especally t was judged that tc varable ethncty, whch s ncluded n both samples, would make the comparson more approprate between rounds 1 and 7 nstead of rounds 1 and 8. There s also an unexplaned part of the wage dfferental due to the fact that the earnngs equatons do not explan earnngs perfectly,.e. R 2 <1 n te equatons. Ths unexplaned part amounts to 24% n round 1 and 17% n round 7. 16

18 9), they are rewarded more for educaton and general experence and less for experence n current job, whch s another concluson renforcng the results from the analyss above. Round 7 shows slghtly dfferent results. In 1994, 54% of the wage dfferental can be accounted for by dfferences n characterstcs, whereas dfferences n rewards only account for 29% of the wage gap. Dfferences n characterstcs manly consst of formal sector workers havng more educaton and experence, both n general and job specfc. Rewards n the two sectors depend prmarly on educaton and general experence, selfselecton and alttude. However, t s not possble to conclude that the declne n the pan of the wage dfferental, whch s due to dfferences n rewards, s a phenomenon that wll contnue over tme due to the large dfferences across rounds, whch was llustrated above. The man concluson n ths secton s that the results from the decomposton confrm the prevous analyss, whch contrbutes to the robustness of the results. Table 11: Oaxaca Decomposton, Round 7, 1994 Intercept Years of educaton Experence Experence Squared Job Experence Job Ex. Squared Lambda Alttude Woman Head Ethncty Publc Admn. Analphabetc Trade Ln (monthly wage) Dfference Note: Job experence s measured n weeks. Source: Author s calculatons 5. Concluson Average Characterstcs Decomposton Formal Informal Dfferences n characterstcs Dfferences n rewards 0 1,369 10,46 7,82 0,152 0,443 20,49 14,96 0,085 0, ,64 375,35 0,062 0, ,24 91,85 0,100 0, ,03 5 0,047 0,52 0,55 0,105 0,678 2,57 2,39 0,042 0,278 0,43 0,34 0,02 1 0,007 0,56 0,43 0,010 0,018 0,42 0,40 0 0,030 0,05 0,27 0,03 2 0,007 0,11 0,14 0,004 0,012 0,02 0,03 0,003 0,007 6,22 5,73 0,485 0,260 0,142 As n other studes the wage dfferental between the formal and the nformal sector has been found to be qute large n urban Bolva. Ths wage gap has been estmated by a human captal model correctng for selectvty bas usng Heckman s twostep procedure. The probablty of enterng the formal sector was estmated, and the most surprsng result was that mgrants tend not be employed n ths sector. Another notceable result was that the probablty of enterng the formal sector was a hump shaped functon of alttude. Wages were then estmated usng a Mnceran specfcaton, and there were some dfferences between the formal and nformal labour markets. Returns to educaton are generally hgher for formal sector worker, whereas job specfc experence dd not exhbt the usual hump shape for these workers. Instead wages depend lnearly on experence n current job, ndcatng that tenure s more mportant than productvty. Beng female or ndgenous decreases formal sector earnngs, and alttude turned out to be a sgnfcant explanaton of 17

19 wages as well. People lvng n the valleys and lowlands earn more n general than ther counterparts n the hghlands. Sector choce turned out to be less sgnfcant n the formal sector than n the nformal sector. In the nonlnear specfcaton of returns to educaton, t turns out that returns to schoolng dffer sgnfcantly between levels. Basc educaton s ether not rewarded n the labour market at ah, or the returns are negatve. Only returns for relatvely hgh degrees are sgnfcantly postve, ndcatng that the qualty of basc educaton n Bolva s very low. A decomposton of the wage dfferental nto dfferences n characterstcs and dfferences n rewards between the two sectors only confrm the frst analyss, because educaton and experence are the man factors contrbutng to the dfferences between sectors. 18

20 References Andersen, Lykke E. and Zafrs Tzannatos (1998), Econometrc Analyss of Formal and Informal Labor Markets n Thaland , Workng Paper, World Bank. Andersen, Lykke E. (1999), Wage Dfferentals Between Bolvan Ctes, Workng Paper 02/99, HSECUCB, La Paz, Bolva. Greene, Wllam H. (1997), Econometrc Analyss, Prentce Hall, Inc., New Jersey, USA. Heckman, James J. (1976), The Common Structure of Statstcal Models of Truncaton, Sample Selecton and Lmted Dependent Varables and a Smple Estmator for Such Models, Annals of Economc and Socal Measurement 5/4. Heckman, James J. (1979), Sample Selecton Bas as a Specfcaton Error, Econometrca vol. 47, no. 1. 1NE (Insttuto Naconal de Estadístca) ( ), Encuesta Integrada des Hogares, La Paz, Bolva. Jensen et al (1997), Employment and Earnngs n Zamba n 1993, Workng Paper, Centre for Labour Market and Socal Research, Unversty of Aarhus, Denmark. Mncer, J. (1974), Schoolng, Experence and Earnngs, NBER. Oaxaca, Ronald (1973), MaleFemale Wage Dfferentals n Urban Labor Markets, Internatonal Economc Revew, vol. 14 no. 3. Wood and Patrnos (1994), Urban Bolva, n Indgenous People and Poverty n Latn Amerca, an Emprcal Analyss, (1994), edted by Psacharopoulos and Patrnos, World Bank Regonal and Sectoral Studes. 19