Unemployment in Mexico : its characteristics and determinants

_ _ _ __ _ _5 lZ'O POLICY RESEARCH WORKING PAPER 1 230 Unemployment in Mexico AJthoughMexico's unemployment rates, Its Characteristics measured over a week, are low (3 to 6 percent), i 5 to 20 and Determinants percent of the population experiences at least one spell Ana Revenga of unemployment over a Michelle Ribozud year Unemployment is concentrated among the young; Half the workers under 20 experience a spelf of unemployment over a year, but only a tenth of workers over 30. The World Bank latin America and the Caribbean Country Departnent II Huirnan Resources Operations Division Country Operations 1 and the Environment Division December 1993  [i,iH R FSEAR( H W0RKKNOI 1'F I2,3(0 Sutmmary findings The restI iruturtrii I of M e\iL. )s ck i wn III has, h I d perCICiti 1( t. I ; reL enit. %~ rh[ th( lrte it, rcises ill sil '~~risi n~ivlittle t'ttc'tt oII Nl.\Ie i,A ll Hlif IIFF i1CIflliiv1`, :ii FA JfMn pLu7uCO'11 Imdill 20F oi %itlil WhidlC I O slW C-VeFI ill tIlt' WorSt \CLIIS. (e y 11 FK idl t FiI : - i',~i 1 tirUt. rer it UM0 lIpl VF,IIII( R~iii udi tisk: ILs th lie it:Iic di dlet iii U II if I 111plit i\FieO-lf 1ii' mm l i ii 'li' *i (),, Ii 'I cd il wh r i'I l. wvho eit' it huiidi'Fi I, FiFe wclidr c iisF FlllctiFlo i,1u1Iellt lFe priiirThil1Ito oI ullF1('FpI()VFFlilt-fi IIIFeFF1~lOrFtF.'Flt widiesplIr )I A Ci ti riFF1,11p 0ipu11Ft hIl iCe's\ .h O.l dc',dii.Cit 11 ItIIfir 1)111 d1d10111' 1112 k r1f)IIf wlIl giritips espe~.aaliv '% u eI hi to it, freperteIdlv hlit IK~I
F.nU \1Frrvlg IL Issi&ioItcd ot with1 It t~ oI' risk iiit Fuld tIetetifOre is'Flco of AiI IFteiFoFll Is Fllisi 1 fi-IellIi 11ielt I flilt - IF I tikii bi ) IllU te( k-i'di,ii' 5SIFFFeFI Un1eflIphIr'VlFe t ass ciated ithi Ti rufnl ttirilover hit Illiire ptb ()I), Iitv o fIf uiFenplm Flicut for wt ilotei with tuIovc1uueuits trol)n uone job to F1"thUer) mr Avith eertiin less cd Ip,~l iiiFh. induividluals Incing not of work ai imy. FIIl III lFe IIX iL'Il spell I Ii )I rlFp(IvI~ Ittltt is, lot hiF: R{eC%Cga, on,d R(it-ou.d lddress these Ii~Ul Uslilg mOelo duitii iti 0 rtiiu ith I uI, t itr IIT tind ~.2)Tl tFFthls nanel data trouti the, nII,IrtCrlN Urh;ur lalhor torec survev, . f( r tincrF (whIfli ch \plaiinIf thei h ichcr averaige household-based surve%y of 1 t6 urban Arc%as, aInd daimtF ff"lii llinifllpliivnleF,"t rare fi)r w5otllet). rthe NitiotFFil Frnplovmet ~UrvcN clure d out every two * He duratXIW it 1(lnpli en oger fir older to thre ears. 1Kv titi tliFat: ,%orkcrs, b,tt does, Fiit va j uhstor'oiivace iding to ihe sti ticttire- of tFnleiiFplovnicntl iii Niexici IK edH;,cat 1 ooIIti ATlni IICet." I leads,l ill household s andi br,ildlo, ' rbiiLir to thar in other Li.u trices. t eFiciplito FFFOFU inkihldI tills with hiii sehiiid respo1n,Lihtics Ftu dloto exit is IigigKst tiot thlise I t) ti 2.i secib wolle~ i. trio( Il IIF'telplsl~) floU't Lisier. Stirprisitlgiv hiev it is t'i' 11(Fully, MOld i11lX lilmwii~h U it,~ '.'i sp lF fI-IbiteFi .iiVrIIettt IS schlool glrIU~Iutes thal all II)Iig the le edJLatedL. re,jIttiveX HFt FF1 OFj)s 71 per-cent of all iutrinphiyvun -It U lelFiplo!, Fitlit as ffc11o cl; Icd y 'Fe .~t 5%. In :I I () s) I itiioihi F sp is lIdstiIiit dt1 least siN Frd ii is rentilUld illidcrxct dMltrFlig ldJfUStF: -lb bt. ..\1sF j1. ibd () piOIocFt .repile FI( spells List-ili.r at .ljFIlIFFiccLIt rrd throug1~2h the. ro('il ~atge rIi1 Ii1Ir0 1l~,Ist I edo%: tru hunetoplovotent. Ih r li iv lsi, ihezi reinlIo.c Iv lik * A thnwh unemlpl I\ menr rues. ds lo,easFrtec OVet a tmIFFrcI:t'tiFFll. )If t If c I iils.. 1ec :~i ii IF t gliFe o. l rII j iMt '11 ;\m : FeIFFti'o. F Ut' rlip! 'FF0 U , hil- Iug spIeis oIit i,t lit: Itl"iti t(iLer' mr.FF'.tFiifn Ill lid iou it tibi I,' (';O !It \N iF'bK* I' () I% Ws()n tn xrk-cr5 ovt r raises tWIt fIte iii Illalet 1F11IFrI1l,lltli\ lletl tFIo 1 )8N Fr I I 4 i his paper - .1 jiiitt proddi't il llcFF IiIi e o e )irn lns aoJ ildlltry ~ eF t' tldlo tII I Fbvirii'llolt D)ivisions, Latin Anwr-i ito1 the ( an'ihell' o11 ( oulIlt!" DI)epartIIetF It --,- is ." p; tF a liger 01t11 Ml Itii delmrtmcnt it ilihor Flarket iss,ues. ([opies ot ilth p.tir irt iiI iI Froe tr' ml tile \ )rid h,mnk1 1 I I I `st'cc NW, \\sI" IIgton 1)( 20433, Pease ciort-at I{FlktFlFblr Si ~Ipicnt. roiO FU f I 6's o-\rte tJ p i-hi (20 pier 'a,) I k't !1'L FlI~3 1' nilailced O FIt I ',s Rcsat,ii l)sscrFUiO',1. (ta:Fc  UNEMPLOYMENT IN MEXICO: AN ANALYSIS OF ITS CHARACTERISIICS AND DETERMINANTS Ana Revenga and Michelle Riboud  i EXECUTIVE SUMMARY Over the past six years, Mexico has successfully implemented a program of sweeping economic reforms and made major strides towards a complete restructuring of its economy. Surprisingly, the reform process has occurred with little impact on unemployment. Official statistics drawn from the census and from employment surveys report an unemployment rate of 2.8% in 1991. An analysis of trends indicates that, even in the worst years of economic crisis, average unemployment rates did not increase beyond 6%. These figures seem very low when compared to those for other countries, especially given the magnitude of reforms and structural changes recently experienced by the Mexican economy. They raise a number of questions about the nature and relative importance of unemployment in Mexico. A first question is whether the official definition of unemployment adequately reflects the importance of the phenomenon. In other words, is unemployment properly measured? A second issue relates to who bears the burden of unemployment. From a welfare perspective it matters greatly whether the cost of unemploy ment is widely spread or whether it falls primarily on a few. Even if only a small fraction of the labor force is unemployed at any point in time, these individuals may have specific characteristics that would make them particularly and repeatedly vulnerable, and therefore deserving of special attention. What are, in fact, the characteristics of the unemployed? Can one identify population groups that are more vulnerable to unemployment? Within each population group, is the risk of unemployment concentrated on a small number of individuals who are repeatedly hit? Another issue revolves around the relative importance of long-term unemployment. Specifically, one would want to know whether most unemployment is associated with normal turnover (movements from one job to the next), or rather comprised primarily of individuals who are out of work for a long period of time. All these questions have important implications for the design of policies and programs aimed at the unemployed. In this report, we attempt to answer these questions using data drawn from two surveys. Our first source of data is the quarterly urban labor force survey (ENEU), a household-based survey of sixteen main urban areas. A key feature of the ENEU is its panel structure, which allows us to analyze certain aspects of unemployment --such as unemployment duration, persistence and turnover- which could not be analyzed with purely cross-sectional surveys. The ENEU's main drawback, however, is its limited coverage; it provides no information on the population from the rural areas or from smaller urban centers. Our second source of data is the National Employment Survey (ENE), which expands beyond the ENEU sample to cover in addition all other main urban areas and a sample of the rural population. The ENE is carried out every 2-3 years. We work with both surveys, drawing on the larger ENE sample principally for the analysis of the characteristics and determinants of unemployment and on the iNEU, and its panel structure, for the analysis of the dynamics of unemployment. The main findings of the report are the following:  ii (1) The overall structure of unemployment is broadly similar to that observed for other countries. Unemployment rates are highest for the young -particularly for those 16 to 25 years of age- and have been consistently higher among women than among men. With regard te education, the highest unemployment rates for males correspond to those with either incomplete or complete lower secondary schooling (7 to 9 years of school). For females, the highest rates are found among those with either complete low secondary or higher secondary (9 to 12 years of school). This pattern of higher unemployment rates for secondary school graduates differs somewhat from that observed in other countriec, where unemployment appears to be more prevalent among the less-educated. (2) Unemployment rates as measured by official statistics are quite low, and have remained moderate throughout the adjustment process. This reflects the fact that most of the adjustment occurred through the real wage --which showed substantial downward flexibility-- rather than through employment. One consequence of this form of adjustment is that there has been relatively little productivity-enhancing employment restructuring. (3) The definition of unemployment used in official statistics tends to underestimate the true number of people who are jobless, because it fails to consider transitions in and out of the labor force, which are extremely frequent. The data show that 25% of all unemployment spells for men and 53% for women end in withdrawal from the labor forr, and that a large fraction of those who withdraw reenter the labor force within 3 months. These short spells out of the labor force should, in most cases, be considered as unemployment. (4) Using a more extensive alternative definition of unemployment!', the rate of male unemployment in 1988 is shown to increase from 3.4% to 6.4%. The largest increases in unemployment are observed for individuals under 20 years of age, ! :or those with little education. The use of this alternative definition thus yields a structurc jf unemployment by education more similar to that observed in most countries. (5) Multivariate analysis confirms that age, sex and education are key determinants of unemployment. Estimates obta' i' - from a probit model reveal that the probability of unemployment decreases with age and education for both men and women. The probit analysis also finds a strong effect of marital status, although this effect tends to go in opposite .lirections for men and women. Marriage is associated with lower risk of unemployment for men and foL more educated women, but increases the probability of unemployment for women at low levels of education. (6) The analysis of the distribution of completed unemployment spells suggests that the typical unemployment spell is not long. About 40% of all unemployment spells among men end within three months, and one-half of all spells are completed within four to five months.w The mean duration of a completed spell for males is about 5.7 months. Mean duration for 1/ Which includes individuals who are not working and not actively looking for a job, but who are not studying, nor taking care of the household, nor retired nor physicaUy disabled. 2' These figures refer to the standard definition of unemployment. Corresponding figures for the alte.r-a,n.e defunition of unemployment are presented in Table 11. Throughout the report, results regarding composition, duration and persistence of unemployment are presented separately for both definitions o' the unemployed.  iii women is 7.2 months. Thus higher average unemployment rates for women are partly explained by longer duration. (7) Although the typical spell of unemployment is relatively short, there is a sizeable proportion of unemployed ineividuals (129o for all males) who suffer spells of over a year. As a result, almost 70% of all unemployment in 1990-91 was attributable to spells lasting at least six months, and 30% corresponded to spells lasting at least a year. (8) Duration of unemploymen is longer for older workers, but does not seem to vary substantially by educational attainment. The report also tinds that household heads and individuals with houFehold responsibilities tend to exit from unemployment faster. (9) Finally, the report examines the degree of persistence of unemployment over time for the whole population and within age and education categories. It finds that although unemployment rates, as measured over a one-week period, are low (on the or,.!er of 3-6%), a significant fraction of the population (15-20%) experiences at least one spell of unemptoyment over a year. Sharp differences exist between young and adult workers. About one-half of teernagers experience at least one spell of unemployment in the course of a year, as compared to 10% of workers over the age of 30. This suggests that while the incidence of unemployment is widely shared among youth, it is concentrated on a much smaller group among older workers.  L IN1 ODUCTION Over the past six years, Mexico has successfully implemented A program of sweeping economic reforms and made major strides towards a complete restructuring of its economy. Until 1985, the Mexican economy was highly protected, dominated by state-run industries, and heavily burdened by debt. As a result of the policy changes and reforms that have taken place since then, the economy's prospects today are vastly different. A fast and far-reaching trade reform has turned Mexico into one of the more open economies in the world. Domestically, the reforms have been equally comprehensive. The financial sector has been restructured and liberalized, fiscal expenditures have been cut dramatically, and the tax system has been overhauled. Key sectors, such as transport and telecommunications, have been deregulated. Moreover, in a still ongoing process of privatization, more than two-thirds of state enterprises have 'oeen sold, closed or spun off. Despite the rapid and far-reaching reforms, which have had a clea: impact on the labor market, unemployment has remained tairly low throughout the adjustment process. Official statistics drawn from the census and from employment surveys report a very low unemployment rate (2.8% in 1991). An analysis of the trends indicates that even in the worst years of the adjustment process, ave rage unemployment rates did not increase beyond 6%. These figures, surprisingly low by international standards, raise a number of questions about the nature and relative importance of unemployment in stexico. A first question is whether the official definition of unemployment adequately reflects tihe importance of the phenomenon. In other words, is unemployment properly meacured? A second issue relates to who bears the burden of unemployment. From a welfare perspective ic matters greatly whether the cost of unemployment is widely spread or whether it falls primarily on a few. Even if only a small fraction of the labor fc.rce is unemployed at any point in time, these individuals may have specific characteristics that would make them particularly and repeatedly vulnerable, and therefore deserving of special attention. What are, in fact, the characteristics of the unemployed? Can one identify population groups that are more vulnerable to unemployment? Within each population group, is the risk of unemployment concentrated on a small number of individuals who are repeatedly hit? Another issue revolves around the relative importance of long-term unemployment. Specifically, one would want to know whether most unemployment is associated with normal turnover (movements frcm one job to the next), or rather comprised primarily of individuals who are out of work for a long period of time. All these questions have important implications for the design of policies and programs aimed at the unemployed. In this report, we attempt to answer these questions using data drawn from two surveys. Our first source of data is the quarterly urban labor fPrce survey (ENEU). The ENEU, a household-based survey of sixteen main urban areas, elicits a wealth of information on sociodemographic characteristics, employment status, type of job, monthly salary and hours of work. For those individuals who are unemployed, it also reports the length of their unemployment spells up to the time of the survey. It is the main source of time-series household-based labor market data, having bee1, carried out continuously since 1983. A key feature of the ENEU is its panel structure. The survey uses a quarterly rotation system such that each rotation group (of households) remains in the survey for five consecutive quarters, and  2 then leaves the sample. By matching individual survey responses in successive quarters, flows between labor force states can be roughly estimated. This allows us to analyze certain aspects of unemployment --such as unemployment duration, persistence and turnover-- which could not be analyzed with purely cross-sectional surveys. The ENEU's main drawback, however, is its limited coverage: it provides no information on the populations from the rural areas or from smaller urban centers. Our second source of data is the National Employment Survey (ENE), which expands beyond the ENEU sample to cover in addition all other main urban areas and a sample of the rural population. The ENE is carried out every 2-3 years. The first ENE was fielded in 1988 and a second in 1991. The next ENE is planned for 1993. In its urban coverage, the ENE is very similar to the ENEU: the sampling schemes are alike and the survey questionaires identical. This allows us to work with both surveys, drawing on the larger ENE sample principally for the analysis of the characteristics and determinants of unemploymen. and on the ENEU, and its panel structure, for the analysis of the dynamics of unemployment. 11. STRUCTURE AND TR ENDS OF UNEMPLOYMENT A. General Trends .n Unemployment Official unemployment figures for Mexico are reported quarterly by the National Strtistics Institute (INEGI) on the basis of the ENEU. These figures reflect the official definition of unemployment, which consider s an individual as unemployed if he/she participates in the labor force and fulfills the following conditions: - worked for less than one hour during the week preceding the survey - was not sick, on paid vacation or waiting to return to work wvithin the following month - was actively searching for a job during the month before the survey Table I reports these official unemployment rates for the 1980-91 period, separately for men and women.3' The trends should be interpreted with some care since the coverage of urban areas in the survey has increased over the period, from 12 in 1980 to sixteen from 1985 to date. The table shows that the aggregate urban unemployment rate in Mexico is quite low. In 1990, the unemployment rate stood at 2.8%, and even in the worst years of the ac justment process, the average rate did not rise beyond 6. 1 %. The table also reveals that the unempioyment rate has consistently been higher among women than among men. In 1983, the unemployment rate for men peaked at 5.3%, while that for women stood a full two percentage points higher at 7.6%. B. Characteristics and Structure of Unemplovment Table 2 presents unemployment figures by age and education categories for men and women. These figures were obtained using individual response data from the 1988 National Employment Survey (ENE). The numbers show that unemployment rates are highest for the young, particularly 3/ Figures for 1980-82 are from the Continuous Survey on Occupation, the predecessor of the ENEU.  3 for those 16 to 25 years of age. In 1988, the unemployment rate for males aged 16 to 20 stood at 8.4%, while . at for males 21 to 25 was 5.3%, as compared to an average male unemployment rate of 3.4%. Similarly, for women aged 16 to 20, the unemployment rate in 1988 stood at 14%, and for those aged 21 to 25 percent it stood at 8.9%, while the female average was 6.3%. By educational attainment categories, the highest rates for males correepond to those with either incomplete or complete low secondary (7-9 yrs of schooling). For women, the highest rates correspond to those with either complete low secondary (9 yrs) or higher secondary (10-12 yrs) levels of education. The above 4.-mrnployment figures are oased upon a strict definition of unemployment that defines an individLsi as unemployed only if he/she is actively searching for a job. However, research on other countries suggests that the distinction between "unemployment" and "not in the labor force" based on intensity of search is often very weak.y For examp!e, in their study of U.S. unemployment, Clark and Summers (1979) find that repeated spells of unemployment interrupted only by brief spells outside the labor force are very comnmon. They find that approximately 50% of all unempioyment spells for males aged 16 to 20 end in withdrawal from the labor farce. About 8G% of those, however, return to employment within 2 months. Although less pronounced, the patterns are similar for individuais 20 and over. These findings, and those for other countries, underscore the importance of looking beyond the official definition of unemployment to transitions in and out cf the labor force in understanding unemployment patterns. -.ible 3 presents some characteristics of labor force withdrawal and reentry based on data from the 1990-1991 ENEUs. As discussed above, the ENEU uses a quarterly rotation system such that each rotation group (of households) remains in the survey for five consecutive quarters, and then leaves the sarnple. We obtained panel data for the rotation group that remained in the survey trom the third quarter of 1990 to the third quarter of 1991, matched individual survey responses in the successive quarters, and used these data to estimate flows between labor force states. Table 3 shows that 25% of all unemployment spells for men and 53% of all unemployment spells for females, end in withdrawal from the labor force. As is the case in the U.S., these fractions are hig,a2r for those under 20 years of age: 37% of unemployment spells for mnales aged 16 to 20 end in withdrawal from the labor force, while the comparable figure for women is 55%. A large fraction of those who withdraw (55% of males and 41% of females), reenter the labor force within 3 months. This suggests that at any point in time the disticntion between unemployed" and "out of the labor force" is quite fuzzy for certain groups of workers and implies that the official definition of unemployment, which includes only those individuals who report to be actively searching for a job, will tend to underestimate the true number of people wr., are in fact jobless. Our analyses of the individual survey responses from the ENE and the ENEU reveal a large fraction of men who report to be idle -- these individuals are out of work, able to work, not studying, and not taking care of the household. The panel structure of the ENEU allows us to follow them over time, revealing that many of them subsequently find employment, and often do so without going through a spell of what is officially defined as unemployment -ie. without reporting to have been actively searching for a job. This raises a question: how should one treat these apparently idle workers? As out of the labor force? As discouraged unemployed workers who have given up searching actively but will nevertheless take a job if the opportunitv arises? 4/ See Clark and Summers, 1979 for an analysis of unemployment and labor force transitions in the U.S..  4 Table 4 presents an alternative definition of unemployment for males, computed using the 1988 ENE, which includes those who appear to be idle -that is, those individuals who are not working, studying, taking care of the household or otherwise occupied, but who are too younr. to he retired and are physical'y able to woLk. We do not compute similar alternative unemployment rates for femaies since their labor force attachment patterns are necessarily more complex because of their household and child-rearing responsibilities. Column (I) in Table 4 shows the official unemployment rate, calculated to include only those who have actively looked for a job during the month preceding the survey. Column (2) shows the unemployment rate computed including thosc who looked for a job sometime during the two months preceding the survey. Finally, co;umn (3) presents an alternative definition of unemployment, which includes those who appear to be idle TUsing the alternative definition of unemployment increase the average rate significantly frot., 3.4% to 6.4%. The most ir.-eresting feature of the table is that the largest increases correspond to t.,ose under 20 and to those with little education. This suggests that the choice of definition can lsave important implications for the analysis of the structure and characteristics of unemployment. Tables 5 and 6 report the distribution of unemployment by age and education categories. Using the standard definition of unemployment (corresponding to the ofticial rate and to column (1) in Table 4), we find that for men Ps much as 60% of total unemployment is accounted for by individuals below the age of 25. The comparable fraction for ferrmales is even higher at about 77%. As regards education, 53% of total male unemploynment and 62% of total fernale unemployment corresponds to individuals with some form of secondary education. Individuals with completed secondary education (9 years of schooling) a-count for 20% of total male unemployment and about 19% of total female unemployment. Those with higher secondary education (10-12 years of schooling) account for an additional 20% of male unemployment and a stunning 35% of female unemployment. These figures indicate that unemployment is concentrated among those with a certain level of education and not, as could be expected, among the least educated. This su,gests, in turn, that the reservation wage and, possibly, family income are important determinants of unemployment: more educated individuals will tend to have both a higher reservation wage and more family income, which would allow them to afford longer job search periods. These conclusions have to be modified slightly when the alternative definition is used. Taking into account those in, .viduals who report to be idle sli;htly alters the age pattern, giving more importance to those between the ages of 12 and 16. More importantly, using the alternative definition also tends to increase the fraction of males with less than secondary education in total unemployment. This reflects the fact that many of those individuals who report to be idle have little education. The above tables suggest that a large fraction of the unemployed are young. This raises the question of how many of these young unemployed individuals are first-time job seekers or new entrants into the labor force? Table 7 presents the fraction of total unemployment accounted for by new entrants from 1983 through 1991. nte table shows that, although it is still significant, the proportion of new labor market entrants among the unemployed has actually declined steadily since 1983, from nearly 30% to about 19% in 1991. III - THE DETERMINANTS OF UN IEMPLOYMENT The previous section analyzed both unemployment rates and the distribution of unemployment by sex, age and education levels. However, further statistical analysis is needed to take into  5 consideration other possible determinants of unemployment and to ascertain the joint effect of different variables as well as the interaction between them, To this effeLt, this section presents the results of a multivariate analysis which aims at estimating the effect of different variables - age, education, geog-aphical location and marital status - on tde probability of t ing unemployed in a given week. Probit models are estimated for men and won,cn, using data on urban areas drawn from the 1988 National Employment Surve, (ENE). Results are presented in Tables 8 to 10. Our estimates show that sex, educa ;on levels, age, marital status and geographical location are important determinants of the probability of unemployment. They also show that the effect of some of these variables is somewhat differet t from what was observed earlier when interactions between variab!es were not taken into account. A. Results for Males Results obtained for men are presented in tables 8 4nd 9. Table 8 reports probit estimates obtained with the two definitions of unemployment. Table 9 considers only the broader definition but reports estimates ob.ained by dividing the sample by age and education groups. Table 8 shows that while the effect of schooling on the proba'>ility of being unemployed is positive when only active job seekers are considercd as unemployed, the effect is reversed when a broader defirition of unemployment is used. This result is consistent with what was observed earlier in the previous section. Movements in and out of the labor force over short time spans, as well as low search intensity, are more prevalent among low educated workers. When this is taken into account, the probability of being unemployed appears higher at lower levels of education. Table 9, which reports estimates obtoined by dividing the sample by education levels, indicates that the probability of unemployment declines with education up to 6 years of schooling; beyond, it remains constant. Age decreases the likelihood of being unemployed up to the age of about 45. The effect is not linear: the decline is much stronger at an early stage of working life (between ages 12 and 20) than afterwards; it is also sharper among secondary graduates that for other levels of el' cation (see table 9). Being married or cohabitating reduces significan.ly the p.obability of being unemploye.d, which is consistent wiu the hypothesis thar family responsabilities induce greater labor force attachment among men. Regional differences also appear to be signiticant. Estimates obt.ined for the whole sample suggest that the Drobability of unemployment is higher in Mexico City (the DF) than in other parts of the country. H 2ver, when the sample is divided into education groups (as shown in Table 9, column 6), this effect disappears for those with at least 10 years of schooling, suggesting that more highly educated workers are not any more likely to be unemployed in the D.F. than in other regions of Mexico. Similarly, while the probability of unemployment does not appear higher in the northern states of Mexico than in the center or southern states for the sample as a whole, distinguishing according to age and education shows that the likelihood of being unemployed is higher in the North for teenagers (and similar to what is observed in the DF). It is also higher for individuals with less  6 than 10 years of schooling There is obviously a strong demand for skilled workers both in the DF and in the N4orth which makes them less vulnerable to unemployment than other education groups. B. Results for Females. Estimates obtained for women (Table 10) refer only to the standard definition of unemployment. As explained in the previous section, it is difficult to construct an appropriate braoder definition of unemployment for women because of their more complex patterns of labor force participation, resulting from their g'eater household and family responsibilities. The effect of schooling on the likelihood A4 being unemployed appears positive and significant (the same effect was found for men when a restr tive definition of unemployment was used). This positive effect, however, only occurs up to 6 years of schooling. Beyond, the effect is no longer statistically significant. As for men, the effect of age is negative and significan: up to the age of 45. Thbe decline is stronger for vomen with at least 7 years of schooling than for women wi;h lower 'evels of education. When the sample is divided into three age groups, the negative effect of age only appears atter the age of 20, while for men, the decline is sharp even among teenagers. Significant differences can be observed between men and women regarding the effect of marital status. While marriage can be clearly associated with lower risk of unemployment for men, it increases the probability of unemployment for women at low levels of education, but decreases it at higher l3vels. Cohabitation, which appears more frequently among the less educated, also tends to increase the likelihood of unemployment. Both effects most likely reflect patterns of labor force attachment which are consist nt with what is observed in most countries. Marriage, and to some extent cohabitation, are correlated with family responsabilities which increase the demand for women's time spent at home. Regional variables confirm the greater probability of unemployment in the DF than in other parts of the country. Although the effect is weaker at higher levels of education, it still persists. In the northern states, the lower probability of unemployment found for older and more educated women seems consistent with the hypothesis that the demand for labor in the North is biased towards educated and experienced workers. Overall, these results seem consistent with human capital theory. Theory predicts that, as work experience is acquired, specific human capital stock is built up, iointly financed by the worker and the firm, and which neither employer nor employee wish to loose. This reduces incentives for both quits and layoffs. As a result, one can expect a decrease in the probability of unemployment with age. The fact that, for women, this effect only appears after the age of 20 and not earlier as for men, suggests that the job matching process takes more time for them. In the same way, firms are expected to invest more in more educated workers as human capital acquired on the job is highly complementary with education. This also induces a higher probability of unemployment for workers with low levels of education and little specific on-the-job human capital. IV. DURATION OF rNEMPLOYNIENT  7 The analysis in sections 11 and III suggests that there are some important differences in unemployment rates among demographic groups. In this section, we extend our analysis of unemployment differentials, with particular attention to the dynamics of unemployment and how they differ by sex, age and education. Recent research in labor economics suggests that unemployment should be viewed not as a static phenomenon affecting a stagnant pool of job seekers, but rather as the result of individuals flowing in and out of unemployment, each experiencing jobless spells of varying length. At any point in time, observed unemployment may comprise a number of individuals experiencing very short spells of unemployment as they move from one job to another (the "churning" or "normal turnover" component of unemployment), as well as a smaller number of people who are out of a job for a long time. The relative importance of these two components in explaining observed unemployment has important welfare implications. If most of unemployment is associated with normal turnover, the burden will be widely spread and few individuals suffer greatly. If, however, most unemployment is associated with a few individuals remaining unemployed for extended periods of time, the hardship associated witn unemployment will fall primarily on a few. This perspective on unemployment emphasizes the difference between frequency and duration of unemployment, and suggests that the measured unemployment rate, in itself, contains relatively little information: a similar rate of unemployment could reflect a large proportion of the labor force being unemployed for a short period of time, or a small group experiencing long spells of unemployment. Better insights into the nature of unemploynent in Mcxico may be gained by explicitly exploring the dynamics of unemployment and labor market behavior. A. Analysis of the Distribution of Spells of Unemplovment. We begin our analysis by examining the distribution of the duration of completed unemployment spells. We calculate this distribution using individual data for the 1990-91 quarterly urban laDor force survey (ENEU). The questions we seek to answer are: (i) what is the mean duration of a completed spell?; and (ii) what is the relative importance of long-term unemployment? The procedure we used to calculate the distribution of unemployment spells is the following. We first constructed a dataset comprising two ENEU cohorts. The first cohort included individuals wh.o were unemployed in the third quarter of 1990. These individuals were then observed at discrete three-month intervals over the follcw.ng twelve months. The second cohort included individuals wno became unemployed in the fourth quater of 1990 and were not in the first cohort. For this latter cohort, individuals were followed for only nine months. As discussed above, most individuals in our dataset report the duration of their (incomplete) unemployment spell up to the time of the survey. We calculate complete unemployment spells by tracking individuals over time, identifying their transitions to employment, and adding time elapsed until the job was found to the incomplete unemployment spell reported in the initial quarter. Unfortunately, the ENEU does not include a question on starting date for the current job. As a result, if an individual moves from unemployment one quarter to employment the next, it is impossible to know exactly when in the intervening three months the transition took place --in other words, knowing which quarter an individual finds a job does not allow us to compute the exact length of the unemployment spell. To obtain a fairly smooth distribution of spells over the quarter, we assume that rifty percent of those who found jobs exited unemployment within one month, thirty  8 percent exited in the second month, and the remaining twenty percent exited in the third month. 1' If an individual remains unemployed at the end of one year, his/her unemployment spell is truncated at twelve months (53 weeks). Some basic features of the distribution ot completed unemployment spells are presented in Table 11 for males and Table 12 for females. For males, we present separate statistics for the two alternative definitions of unemployment presented in Sections II and III above. The first five rows of Table 11 show some basic duration statistics obtained using the standard definition of unemployment (ie. excluding idle/discouraged workers). These numbers suggest that the typical unemployment spell is not long.