www.elsevier.com/locate/worlddev

World Development Vol. 29, No. 11, pp. 1905±1922, 2001 Ó 2001 Elsevier Science Ltd. All rights reserved Printed in Great Britain 0305-750X/01/$ - see front matter

PII: S0305-750X(01)00068-7

What is Behind Increasing Wage Inequality in Mexico? WILLY W. CORTEZ * El Colegio de la Frontera Norte, Mexico Summary. Ð We evaluate the impact of the educational expansion and changes in labor market institutions on wage inequality among Mexican workers using a simulation technique proposed by Knight and Sabot [The American Economic Review 73(5) (1983) 1132±1136]. We conclude that while increases in the relative rate of return of higher education would have induced an increase in wage inequality, changes in the composition of the educational distribution would have led to a stronger decline in wage inequality. Instead, increased wage inequality is largely explained by changes in Mexican labor market institutions; namely, unionization rate and minimum wages. Ó 2001 Elsevier Science Ltd. All rights reserved. Key words Ð wage inequality, education, gender, labor market institutions, Latin America, Mexico

1. INTRODUCTION There is overwhelming evidence that since the mid-1980s Mexico has faced increasing income inequality. 1 Among the di€erent sources of income, wage inequality has contributed signi®cantly to the increase in overall inequality. While the empirical evidence indicates that this is an international phenomenon, 2 the Mexican experience has been accompanied by a generalized decline in overall average wages. 3 In the literature there are three competing explanations about increasing wage inequality: changes in relative demand of skilled laborÐ caused by either trade liberalization or technical progressÐchanges in relative supply, and changes in labor market institutions. 4 The Mexican literature, however, has stressed changes in relative demand in its two versions: trade liberalization and increasing Foreign Direct Investment (FDI). To a large extent these latter two events coincided with the change in economic policy that the government began implementing during the mid-1980s. The trade argument can be summarized as follows: the opening of the economy induced the expansion of the export-oriented industries at the expense of the import-competing ones. Given that the export-oriented industries are intensive in skilled labor, this expansion induced higher demand for skilled labor relative to the demand for unskilled labor. Higher relative demand for skilled labor, in turn, induced positive changes in the relative return to higher education; thus, generating a widen-

ing wage gap between high and low educational attainment. The FDI explanation arises because of the inability of the trade argument to explain the fact that increasing wage inequality is largely due to changes not between industries (or sectors) as the theory would predict, but rather due to changes within industries. 5 In e€ect, Aitken et al. (1995), Feenstra and Hanson (1995), Hanson and Harrison (1995) and Meza (1999) ®nd that increasing wage inequality occurred in all industries alike. In an economy where the capital goods sector is underdeveloped, technical progress is achieved mainly through the increase in imported capital goods. In a sense then, the technical progress argument and the foreign direct investment one are the same. The FDI view argues that by bringing technology that is skill-biased, it contributed to the increasing wage gap between skilled and

* This

study was conducted while I was a visiting researcher at the Center for Economic Studies, El Colegio de Mxico (CEE-COLMEX) during February± July 2000. Financial support from El Colegio de la Frontera Norte (COLEF) is gratefully acknowledged. I would like to thank Maritza Sandoval from El Banco de Mexico for her assistance in some of the statistical tests and support throughout this investigation, to participants at the CEE-COLMEX Seminar for their comments to an earlier version of this paper and to an anonymous referee for helpful suggestions. Remaining errors are mine alone. Final revision accepted: 19 May 2001.

1905

1906

WORLD DEVELOPMENT

unskilled labor. This argument is consistent with a number of characteristics of the Mexican manufacturing sector that explain wage inequality in the manufacturing sector. First, foreign-owned ®rms pay higher wages than domestic ®rms. Second, foreign ®rms exhibit higher labor productivity than domestic ®rms. Third, foreign-owned ®rms tend to assemble in capital-intensive industries and in regions with more advanced infrastructure. Fourth, there is no wage spillover from foreign investment to domestic ®rms (Aitken et al., 1995). While FDI explains increasing wage inequality in the manufacturing sector, it is unable to explain changes in wage inequality in the rest of the economic sectors. In fact, the ¯ow of FDI is negatively correlated with wage inequality in agriculture, indicating that an increase in FDI would reduce inequality rather than increasing it. 6 Further, a decomposition of overall wage inequality by economic sectors shows that it varies greatly across sectors, suggesting that there might be other factors behind increasing wage inequality that have not been considered so far. 7 The core of the relative demand explanation rests upon the argument that the increase in overall wage inequality is in fact the result of the increase in the relative rates of return of higher education. Yet, the increase in the relative rates of return of higher education has occurred despite the substantial expansion of education that Mexico has enjoyed during the last decades. In e€ect, whereas the percentage of employed workers with less than secondary education declined from 60.2% in 1984 to 47.8% in 1996, the percentage of workers with at least some college education increased from 8.7% in 1984 to 12.2% in 1996. 8 Moreover, the expansion of the relative demand for skilled labor seems to have occurred despite the lackluster performance of the Mexican economy since the mid-1980s. In e€ect, during 1984±96 Gross Domestic Product grew at an annual average rate of 1.9%. 9 A further piece of evidence against the argument that changes in the relative rate of return of higher education are the leading factor behind the increasing wage inequality is the fact that wage inequality has increased within each educational category except among workers with no formal education. In fact, the highest growth rate of wage inequality occurred among workers with college education. 10 In other words, for the relative demand argument to hold and assuming some degree of labor

mobility within each educational category, one would expect both an increase of the average wage rate and a reduction of wage dispersion within the skilled labor. The data indicate, however, that while the relative wage rate of a college-educated worker with respect to an uneducated worker has increased over time, their relative wage inequality has also increased over time. The increased wage inequality then cannot be explained exclusively in terms of shifts in relative demand. There is a general consensus that education plays a central role in income determination; however, its relative importance in the growing inequality among wage earners is still unclear. In fact, while education has induced changes on wage inequality in other countries, it is still far from providing a full account of increased inequality among Mexican wage earners. Alarc on and McKinley (1997), for instance, study the e€ect of education on wage di€erential among di€erent groups of workers but in a static framework. That is, they do not analyze changes either in educational attainment or in their rates of return over time. Bouillon et al. (1999), on the other hand, while allowing for changes in rates of return and in educational attainment, are unable to demonstrate directly the net impact of these changes on income (or wage) inequality. They reach their conclusion that education ``substantially'' contributed to increased inequality after comparing the educational gains of the lowest decile to the ones of the top decile with their respective monthly earnings. Another characteristic is that these studies have overlooked the impact of changes in labor market institutions upon wage inequality, for parallel to changes in economic policy, or rather as a result of these changes, there have been changes in the Mexican labor market that have resulted in an increasing ¯exibility. These changes have not been properly analyzed in the context of wage inequality. The purpose of this essay is therefore twofold. The ®rst objective is to measure the impact of the educational expansion on wage inequality among Mexican workers. The second is to evaluate the impact of changes in labor market institutions; namely, labor unions and minimum wages. A number of studies have illustrated not only the extent by which women's participation in the Mexican labor market has been increasing since the mid-1980s but also the extent by which they are discriminated against. 11 We, therefore, carry out the analysis for male and female workers, separately.

INCREASING WAGE INEQUALITY IN MEXICO

The paper is divided into ®ve sections. Section 2 describes the theoretical framework used for the analysis and the data sources. Section 3 shows the extent of wage inequality during the period of analysis. Section 4 describes some key characteristics of the Mexican educational expansion and its impact upon wage inequality. The conclusion from this section is that education has played a marginal role on increasing wage inequality. Section 5 turns the analysis toward some institutional changes in the Mexican labor market such as unionization rate and minimum wages. Section 6 concludes the paper. 2. THEORETICAL FRAMEWORK The theoretical roots of the method of analysis are found in Kuznets (1955). He argued that in the process of economic development, countries face initially increasing income inequality due to the industrialization process. But, as the economy reaches higher levels of economic development, the same industrialization process unleashes forces that induce the decline of inequality. Robinson (1976) formalized Kuznets' ideas of inequality into a simple framework. Assuming that wage dispersion remains constant and that inequality in the higher-income group is higher, he showed that in a two-sector case, as workers move from the lower to the higher income sector, inequality would tend to increase up to a point. 12 In a more general case and without the restrictions imposed, whether inequality increases or declines depend on the relative size of the kth sectors, these sectors' relative mean wages and their relative wage dispersion. 13 Knight and Sabot (1983) extend the framework to analyze the e€ect of an educational expansion on wage inequality. They argue that the overall impact of an expansion of education on wage dispersion can be decomposed into two e€ects: composition and compression. The composition e€ect is the one arising from changes in the educational distribution; that is, assuming that the relative rates of return to education remains ®xed, worker's mobility toward higher educational attainment would induce changes in wage dispersion. The compression e€ect, on the other hand, is the impact that arises from the change in the relative rate of return of educational attainment, holding the educational composition ®xed. Therefore, the net impact of an expansion of

1907

education on wage inequality would be sum of these two e€ects. It is clear then that in a two-sector case, an increase of the relative rate of return of higher education would not necessarily result in higher wage inequality if the composition e€ect has a negative and stronger e€ect on wage inequality. In a multiple-sector model, however, the compression and the composition e€ects are uncertain. A previous study by Vera and Bone (1995) applies the same methodology to the national survey of family's income and expenditure, 1977 and 1984. We depart from their study in three outstanding respects. First, we control for changes in the gender composition of employed workers, i.e., we carry out the analysis for male and female workers separately. Second, we not only evaluate the impact of educational expansion on increased wage inequality, but also the impact of changes in labor market institutions. Third, our study involves a di€erent period and is larger in scope since it covers the years of 1984, 1989, 1992 and 1996. (a) Methodology The standard procedure is to estimate a Mincerian earnings equation (Mincer, 1974), Ln w ˆ f …X ; Y †; where w is hourly wage rate (labor income only), X is a vector that includes an individual's personal characteristics (education and age), Y is a vector of other characteristics. The exercise includes an estimation of the exogenous variables' coecients to estimate the expected wage inequality derived from the model. Next, we calculate the compression and the composition e€ects. The overall impact of the educational expansion on wage inequality on two points of time (t and t ‡ s), would be the combination of the compression and the composition e€ects. The theoretical model is X X ai ei;t ‡ cj xj;t ; Ln wt ˆ where ai 's are the estimated coecients for each (ith) educational level, e represents educational level, cj 's represent the coecients of additional explanatory variables, xj . Between time t and t ‡ s the compression e€ect is calculated as follows: X X ~t ˆ …1† Ln w a^i;t‡s ei;t ‡ c^j;t xj;t :

1908

WORLD DEVELOPMENT

By the same token, the composition e€ect is estimated as X X t ˆ a^i;t ei;t ki;…t‡s†=t ‡ c^j;t xj;t : …2† Ln w Here k is the proportion of the educational distribution in year (t ‡ s) with respect to year t, i.e., represents the change in the educational distribution between the two years. The resulting wage inequality is the combination of both e€ects, thus, X X ˆ …3† Ln wALL a^i;t‡s ei;t k…t‡s†=t ‡ c^j;t xj;t : t The above equations imply that the year upon which the comparison is made is t; furthermore, s can be negative. Wage inequality for each e€ect is estimated as the Log Variance of the estimated hourly wage rate given by Eqns. (1)± (3), respectively. To be able to study the additional e€ect of changes in the rate of unionization on wage inequality we reformulate the Mincerian earning equation; thus, for a given year and gender, the equation to estimate is: Ln w ˆ

4 X iˆ1

a i ei ‡

4 X iˆ1

bi Ui ei ‡

2 X jˆ1

cj x j ;

where, a; b, and c are the parameters to be estimated, e's are the educational dummy variables de®ned previously, U's are union dummy variables (one for each level of education) that take the value of one if a worker is unionized within a given educational category and zero otherwise. Finally, x's are proxy's used to measure experience of workers, namely, age and its square. It should be noted that the model does not contain the constant term. Instead, it includes the dummy variables for all educational categories. 14 The subscript, ith, indicates the four di€erent educational levels (to be de®ned in the next section). In the above speci®cation, the b's represent the union wage premium for each educational level. In other words, for a given educational level ith, the average wage rate of a nonunionized worker would be eai , while the average wage rate of a unionized worker would be eai ‡bi , ceteris paribus. This formulation allows us to separate the e€ect of education from the unions on wage inequality. (b) Data The analysis is based on the National Survey of Households' Income±Expenditure (ENIGH)

that the National Institute of Statistics, Geography and Information Technology (INEGI) conducted the years 1984, 1989, 1992 and 1996. Unlike the Survey of Urban Employment (ENEU), ENIGH collects information about people working in agriculture and rural areas, which given their relative importance in the Mexican economy provides additional information about the extent of income and wage dispersion. A major drawback is that the sample size of ENIGH is much smaller than ENEU's. The analysis involves only those people who were working for a monetary income on a regular basis. In other words, workers whose main source of income came from own business, property rents, pension funds, and transfers were excluded from the analysis. In addition, workers older than 75 years of age were eliminated from the analysis. In a recent study, Hern andez-Licona (1997) illustrates the degree by which an increasing number of families face the general trend of declining real income by increasing the number of hours of work or by incorporating young and older family members and female labor to the labor market. Given these facts, and because they show important characteristics of the Mexican labor market, our study is not limited to full-time workers. The analysis focuses on wage inequality only. It is measured by the variance of the natural logarithm of the hourly wage rate. Hourly wage rate is calculated by dividing the monthly average income in real terms (1994 ˆ 100) by the total number of hours worked in a month. The monthly average wage income is the average for the previous six monthsÐto the moment in which the survey was done. Hourly wage rate thus refers to the case in which all other sources of income have been eliminated except labor income. After eliminating those observations in which there is no monetary wage income reported, the number of workers for each of the years considered were 4,618, 12,496, 10,969 and 15,007, respectively. The information about education is obtained from the same surveys. We classify workers into four educational categories: (i) no formal education (no ed); (ii) primary education; (iii) secondary education; (iv) college education. Primary education includes workers that have had between one and six years of formal education. Secondary education, in turn, involves workers who have had between seven and 12 years of formal education. Finally, college education includes workers who have had more

INCREASING WAGE INEQUALITY IN MEXICO

than 13 years of formal schooling including graduate school. 3. WAGE INEQUALITY The selection of inequality indicators is usually based on the desirable properties that they must have such as the principle of transfers and scale independence. 15 It is evident that the measurement of inequality also vary with sample size, de®nition of income and population selection (in terms of work status, age and gender). We estimate three of such indicators: the log variance of hourly wage rate (r2 ), the ratio of hourly wages between skilled and unskilled labor (x ˆ ws =wu ) and the Gini Coecient (G) estimated from the hourly wage rate. We use two de®nitions of skilled and unskilled labor based on workers' educational attainment. In the ®rst de®nition, skilled workers are those with secondary education while unskilled are those with primary education. The latter ratio determines the index x1 . In the second de®nition, skilled and unskilled workers are those that have some college education and secondary education, respectively. This index is represented by x2 . Table 1 presents our estimates of inequality between 1984 and 1996 among Mexican workers. All indexes indicate that wage inequality increased steadily throughout the period of analysis; moreover, it is evident that the rate of return of a college-educated worker grew faster than the rate of return of lower educational attainment. In e€ect, the index x2 grew by 55%, while the wage gap between secondary- and primary-educated worker, x1 , grew by only 5.2%. Both the log variance (r2 ) and the Gini Coecient present more moderate performance: 15.1% and 23.3%, respectively. These indicators also suggest that inequality had its greatest growth during the ®rst half of the 1990s. Other studies use a di€erent de®nition of skilled and unskilled labor. For example, Hanson and Harrison (1995), and Feenstra and

1909

Hanson (1995) de®ne relative wages as the ratio between white-collar and blue-collar workers. In Feenstra and Hanson (1995) wage inequality went from 1.93 in 1984 to 2.55 in 1990. Other estimates show a much larger increase in wage inequality. For example, Meza (1999) estimates the wage di€erential between the 90th decile and the 10th decile among Mexican urban workers and ®nds that during 1988±93 wage inequality increased at an annual rate of 20%. International comparison suggests that inequality in Mexico has been growing faster than most countries. For example, Freeman estimates that during 1984±94 the ratio of the earnings of men in the 90th decile to earnings of those in the 10th decile grew in the majority of developed countries at an annual rate that goes from 2.0% (United Kingdom) to 2.7% (United States) to 4.4% in New Zealand. 16 Checchi (1999), in turn, using a di€erent data set estimates the Gini index of income distribution for di€erent regions of the world. Compared to Checchi's estimates our estimated Gini Coecient is higher than the average estimates for most of the regions except Latin America. While Table 1 presents aggregated wage inequality, Table 2 shows wage inequality disaggregated by gender. It can be seen that there is some con¯icting evidence as to whether wage inequality is higher among female workers. On the one hand, r2 and x1 indicate that inequality is higher among female workers, while x2 and the Gini Coecient suggests the opposite. Over time, males' inequality indexes present a di€erent behavior from that of females'. For example, in the case of female workers all indexes but x2 present a cyclical behavior: after a slight decline in 1989, wage inequality increased steadily in 1992 and 1996. In contrast, in the case of male workers, two of the indexes, x2 and Gini Coecient, increased throughout the period of analysis. The other two indexes, r2 and x1 , while presenting a cyclical behavior, their periods of decline and increase do not coincide. Despite these di€erences there is a common pattern that holds for both male and female

Table 1. Wage inequality: 1984, 1989, 1992 and 1996 r2 x1 x2 G

1984

1989

1992

1996

0.93 1.576 1.842 0.43

0.91 1.413 2.076 0.47

0.95 1.526 2.562 0.49

1.07 1.658 2.857 0.53

Source: Own estimates based on INEGI (various years).

1910

WORLD DEVELOPMENT Table 2. Wage inequality by gender: 1984, 1989, 1992 and 1996 1984

1989

1992

1996

Male r2 x1 x2 G

0.885 1.497 1.952 0.431

0.922 1.397 2.187 0.478

0.915 1.491 2.696 0.489

1.05 1.604 3.091 0.535

Female r2 x1 x2 G

1.026 1.811 1.582 0.359

0.862 1.513 1.774 0.356

1.039 1.684 2.287 0.403

1.119 1.832 2.454 0.439

Source: Own estimates based on INEGI (various years).

workers: the relative rate of return of higher education (w2 ), substantially increased throughout 1984 and 1996. In e€ect, during the period it grew by 58.3% and 55.1% among male and female workers, respectively. 4. EDUCATIONAL EXPANSION, RATES OF RETURN AND WAGE INEQUALITY (a) Educational expansion During the last 60 years Mexico has experienced three reforms of its educational system: Federal Laws of Education of 1941, 1973 and 1992. These reforms have resulted in important advancements of the educational level of the Mexican population. The highest growth in enrollment rates occurred during the 1950s, 1960s and 1970s. In e€ect, the annual growth rates of enrollment of primary education during those decades were 6%, 5.6% and 4.7%, respectively. The expansion of secondary education was very impressive for its growth rate during the same period were 13%, 17.9% and 9.5%, respectively. 17 It should be noted that while the expansion of basic education is recognized by all, there is less agreement about

the homogeneity in the quality of education across regions. 18 The expansion of education during the 1980s was not as strong as the previous three decades, perhaps in part as a result of the limits that the coverage that the educational system had reached by then. Table 3 shows the educational distribution of workers included in the ENIGH sample for 1984, 1989, 1992 and 1996 and the degree of wage inequality within each educational category (in parentheses). Table 3 shows several trends. First, the proportion of workers with no formal education and primary education has steadily declined over time: from 60% in 1984 to 47.6% in 1996. The percentage of workers with secondary and college education, on the other hand, increased. It went from 40% in 1984 to 52.4% in 1996. Second, the largest changes in the distribution of education occurred in the second half of the 1980s, while the 1990s are characterized by small changes. Third, despite the improvements, the bulk of workers still have low educational levels. With regard to the degree of wage inequality, there are two notable outcomes. First, wage inequality has barely grown among workers with low educational attainment. In fact, it

Table 3. Educational distribution and wage inequality: 1984±1996 No ed Primary Secondary College

1984

1989

1992

1996

9.5 (0.898) 50.5 (0.780) 31.2 (0.717) 8.8 (0.488)

8.2 (0.897) 40.8 (0.784) 38.1 (0.710) 12.8 (0.546)

8.3 (0.891) 43.0 (0.759) 38.0 (0.709) 10.6 (0.680)

6.7 (0.847) 40.9 (0.818) 40.0 (0.823) 12.3 (0.682)

Source: Own estimates based on INEGI (various years).

INCREASING WAGE INEQUALITY IN MEXICO

declined among workers with no formal education. Second, inequality among workers with higher educational achievement grew signi®cantly. In particular, inequality among collegeeducated workers grew by more than 39%. A comparison with other countries shows that Mexico's performance in terms of educational expansion is rather poor. Table 4 presents the educational distribution in four countries including Mexico during the 1980s and 1990s. It can be seen that the di€erence that existed during the 1980s between Mexico and South Korea and Sweden widened during the 1990s. In e€ect, compared to South Korea, the reduction in the percentage of low-education workers is lower, while the increase in the percentage of workers with higher education is also smaller. (b) Rates of return of education Rates of return were estimated for each year using a Mincerian log linear model. 19 In order to evaluate the di€erential impact of educational expansion on wage inequality among male and female workers, we carry out the simulation analysis separately. As explained, four educational dummy variables were usedÐ No Ed, Primary, Secondary, and CollegeÐ which had a value of 1 if the worker had obtained that particular educational level and 0 otherwise. Age and the square of age were used as proxies for experience. 20 The model includes four additional dummy variables to consider the e€ect of unionization on wages (one for each educational level). The results are presented in Table 5. No problems of multicollinearity among the exogenous variables are expected because by

1911

de®nition these variables are exclusive. We conducted several tests to see if there was evidence of heteroskedasticity and rejected the null hypothesis. We also tested if the estimated parameters were statistically di€erent over time and rejected the null hypothesis that the estimated coecients were not di€erent over time. 21 From the regression results we can infer the following about the rates of return of education. First, the wage premium increases as education increases. This fact remains unchanged for both male and female workers and over time. Second, in 1984 the rate of return to education among women was higher than among men, but this changed in subsequent years. Third, with the exception of 1996, there is a clear tendency for the wage premium of the lower end of education to decline, while the premium for college education improved. Fourth, experience (measured by age and its square) plays an important role in wage premium. There is parabolic relationship between earnings and age. On average, the turning point for male has declined to about 45 years of age while that of female has increased to about 50 years of age. All these results are consistent with other studies' ®ndings about education's rates of return in Mexico. 22 There are three additional conclusions that deals with the union wage premium. First, the union wage premium is positive and statistically signi®cant for each educational level except for college-educated workers. In 1996, union wage premium for college-educated workers becomes statistically signi®cant for both male and female workers. Second, the union wage premium for females is higher than for males. Third, after losing some ground by

Table 4. International comparison of educational attainmenta (%) Mexicob

Sweden

South Korea

United States

Mexicoc

Sweden

1980 Less than high school High school Some college

South Korea

United States

1990

60

45

48

19

48

30

28

15

32 9

40 15

33 19

39 42

40 12

46 24

47 25

39 46

Source: Data for Mexico comes from INEGI (various years). Data for Sweden, South Korea and United States come from Topel (1997, Table 1, p. 66). a In the case of Mexico less than high school corresponds to less than secondary education. High school includes secondary education (incomplete and complete) and pre-college education (incomplete and complete). Some college includes college education (incomplete and complete) and graduate school. b Data correspond to 1984. c Data correspond to 1996.

1912

WORLD DEVELOPMENT Table 5. Log earning equations 1984

1989

1992

1996

Coe€

t-stat

Coe€

t-stat

Coe€

t-stat

Coe€

t-stat

Male No ed Primary Secondary College U no ed U Primary U Secondary U College Age Age sq. Adj. R2

)1.46 )0.88 )0.38 0.26 0.77 0.51 0.36 0.08 0.10 )0.001 0.786

)13.66 )9.09 )4.05 2.32 4.62 10.77 6.51 0.84 19.06 )16.44

)1.25 )0.78 )0.27 0.38 0.26 0.29 0.17 0.03 0.09 )0.001 0.767

)18.46 )12.62 )4.56 5.39 2.45 8.59 5.16 0.53 26.77 )22.18

)1.09 )0.63 )0.12 0.74 0.49 0.44 0.27 )0.01 0.08 )0.001 0.750

)15.71 )9.82 )1.95 9.97 3.69 11.33 7.43 )0.16 22.18 )18.05

)1.50 )1.19 )0.70 0.27 0.71 0.59 0.50 0.21 0.09 )0.001 0.647

)23.28 )20.33 )12.15 3.93 3.96 13.62 14.18 3.97 27.41 )23.05

Female No ed Primary Secondary College U Primary U Secondary U College Age Age sq. Adj. R2

)1.37 )0.74 )0.10 0.48 0.75 0.59 0.23 0.09 )0.001 0.786

)7.20 )4.41 )0.58 2.26 7.83 8.03 1.41 8.93 )7.57

)1.35 )0.93 )0.42 0.14 0.31 0.29 0.12 0.10 )0.001 0.774

)11.50 )8.85 )4.23 1.19 4.92 6.81 1.60 16.46 )13.70

)1.49 )1.11 )0.53 0.22 0.41 0.38 0.11 0.10 )0.001 0.750

)11.90 )9.82 )4.84 1.68 5.19 7.51 1.30 14.69 )11.09

)1.75 )1.43 )0.89 )0.10 0.60 0.58 0.43 0.10 )0.001 0.691

)16.36 )15.29 )9.59 )0.94 7.42 13.26 6.84 17.70 )13.90

Source: Own estimates based on INEGI (various years).

the end of the 1980s, unions regained importance in the second half of the 1990s. (c) Simulation results Following the methodology outlined in Section 2, we carried out the simulation exercise. Table 6 shows the simulation results for both male and female workers separately. Let us ®rst focus on the estimated wage inequality resulting from our regression equations (last row). We notice that the estimated values (^ r2 ) are well below the actual values (Table 2). In

general, it explains about 33% of overall inequality in the case of males and about 44% in the case of females. The estimated values, however, follow the general tendency of actual inequality. For our purpose, the model, though very simple, can help us explain the overall tendency of wage inequality. At the outset, we argued that whether changes in the composition of the educational distribution result in increases or decreases of wage inequality would depend on the relative size of the di€erent groups, the wage dispersion within each group, and their respective wage

Table 6. Simulated wage inequality: educational expansion 1984

1989

1992

1996

Male Compression Composition Net r^2

0.299 0.299 0.299 0.299

0.289 0.270 0.275 0.281

0.328 0.252 0.305 0.290

0.329 0.235 0.283 0.369

Female Compression Composition Net r^2

0.408 0.408 0.408 0.408

0.326 0.368 0.255 0.256

0.379 0.409 0.334 0.370

0.364 0.272 0.238 0.462

Source: Own estimation based on INEGI (various years).

INCREASING WAGE INEQUALITY IN MEXICO

mean values. Our results indicate that all these elements worked in favor of declining wage inequality: in both cases, male and female workers, educational expansion should have led to a decline of wage inequality. In the case of male workers, the compression e€ect, i.e., the increases in the relative rate of return of higher education, resulted in higher wage inequality. The composition e€ect, on the other hand, led to lower wage inequality. To the extent that the composition e€ect attests mobility from lower to higher educational achievements, it is not surprising that wage inequality declined as a result. The combined impact of both e€ects resulted in lower wage inequality suggesting that the impact of labor mobility was stronger than the changes in the relative rate of returns, since it went from 0.299 in 1984 to 0.283 in 1996. In the case of female workers, both the falling of the rates of return over time and labor mobility to higher educational levels had negative e€ects upon simulated wage inequality. As a result, their combined impact was an even stronger drop of inequality than in the case of male workers: 41.7% in the case of female workers against 5.3% in the case of male workers. Our simulation results therefore indicate that wage inequality for both male and female workers should have declined as a result of the educational expansion. In the case of male workers, this result occurs despite the increases in the relative rate of return of higher education, a fact that has been pointed out by a number of studies. Its impact was o€set and even reversed by the impact of labor mobility toward higher educational achievements. In the case of female workers, both e€ects worked in the same direction i.e., toward a declining wage inequality. If education does not explain increasing inequality, what then explains it? This apparent paradox can be explained in terms of our estimated regression equations for which we return to Table 5. The regression equation includes other variables that did not intervene in our simulation analysis: experience and unionization. The estimated coecients on age and its square barely changed in the case of male workers while it remains constant in the case of female workers. Thus, their likely impact upon wage inequality changes is small. On the other hand, the coecients of unions are not only higher in magnitude but also their changes over time are rather signi®cant.

1913

5. LABOR MARKET INSTITUTIONS AND WAGE INEQUALITY Several authors have argued that changes in the returns to education are the leading cause of raising wage inequality. Most of their results are based, however, on models that do not consider labor market institutions. In particular, changes in unionization and in minimum wages have been overlooked. (a) Changes in unionization rates For many years, labor unions were important in setting working conditions for a signi®cant number of workers. 23 Unions o€ered workers higher wages, job stability, better social bene®ts and bonuses. But, beginning in the early 1980s and throughout the 1990s there have been a number of changes that have limited the scope for action of labor unions. In what follows, we estimate the impact of those changes on wage inequality. From the results presented in Table 5, we conducted the same simulation exercise considering the changes in the union wage premium, or rate of return to unionization, (compression e€ect) and the changes in unionization rates (composition e€ect). Table 7 presents the resulting wage inequality for male and female workers, separately. In the case of male workers, following a decline during 1989, changes in union wage premium led to a signi®cant increase in wage inequality. The impact of changes in the composition of unionization rates, though smaller in magnitude, also worked in favor of an increase in wage inequality. Thus, the overall impact of unionization was an increase in inequality. In the case of female workers, the combined impact of the compression and the composition e€ects also resulted in an increase in wage inequality. There is an important di€erence between the simulation results for male and female workers though. In the former case, the compression e€ect is stronger than the composition e€ect, whereas in the latter case it is the opposite. In other words, in the case of males, changes in union wage premiums were the leading cause behind increasing wage inequality, while in the case of female workers, changes in unionization rates and not in union wage premiums were the main force. In order to explain the mechanisms by which changes in unionization a€ected wage inequal-

1914

WORLD DEVELOPMENT Table 7. Simulated wage inequality: unionization 1984

1989

1992

1996

Male Compression Composition Net

0.300 0.300 0.300

0.267 0.301 0.268

0.280 0.305 0.284

0.329 0.308 0.349

Female Compression Composition Net

0.408 0.408 0.408

0.313 0.391 0.311

0.334 0.434 0.349

0.421 0.441 0.486

Source: Own estimates based on INEGI (various years).

ity we now take a closer look at unionization rates, union wage premiums, and wage inequality between unionized and nonunionized workers. Table 8 shows the union and nonunion average wage rate, union and nonunion wage inequality and the percentage of unionization among male and female workers. It shows several characteristics of the Mexican labor market. First, on average, unionized workers earn higher wage rates than nonunionized ones. The wage rate of unionized males is about 57% higher while that of a unionized female is about 93% higher. Moreover, the wage di€erence between union and nonunion 24 increased substantially during the 1990s after a signi®cant decline in the late 1980s. In e€ect, between 1984 and 1996, the wage di€erence had grown by 17.1% among male workers and by 9.3% among female workers. Second, the degree of wage inequality

is much higher among nonunion workers for both males and females. Over time, however, wage inequality among union workers grew at a higher rate than wage inequality among nonunion workers. In the case of male workers, it went from 0.328 in 1984 to 0.548 in 1996, while in the case of female workers, it went to 0.538 from 0.277 during the same period. Third, there is a continuous decline in unionization rates for male and female workers. A decomposition of unionization by educational levels indicates that unionization rates increases with educational level for both male and female workers. In 1984, for example, the unionization rates among workers with no formal education, primary, secondary and college were 5.2%, 17.9%, 33.9% and 37.7%, respectively. Furthermore, even though the actual rates of unionization have declined within each educational category, this charac-

Table 8. Average wage rate (W) and wage inequality (r2 ) between nonunion and union workers Male Nonunion % W r2 Union % W r2 Female Nonunion % W r2 Union % W r2

1984

1989

1992

1996

78.6 5.05 0.933

77.5 5.47 0.960

83.0 5.07 0.933

86.5 3.94 1.022

21.4 7.97 0.328

22.5 7.73 0.611

17.0 7.81 0.541

13.5 7.30 0.548

70.9 4.34 1.022

72.4 4.9 0.912

77.9 4.82 1.022

79.2 3.56 1.020

29.1 9.32 0.277

27.6 7.45 0.440

22.1 8.78 0.613

20.8 8.36 0.538

Source: Own estimates based on INEGI (various years).

INCREASING WAGE INEQUALITY IN MEXICO

teristic relating unionization rates to educational levels has remained stable over time. As a matter of fact, between 1984 and 1996 unionization rates declined faster the lower the level of educational attainment. The rates of decline among workers with college, secondary, primary and no formal education were 4.8%, 43.9%, 53.1% and 55.8%, respectively. The fact that workers with higher educational achievement also have higher rates of unionization may explain an additional characteristic: wage di€erential between unionized and nonunionized workers increases as we move from workers with no formal education to workers with college education. This wage di€erential has tended to grow at a faster rate among workers with secondary and college education for both males and females. Keeping in mind these characteristics, we can explain the increase in wage inequality by the following events. On the one hand, as an increasing number of workers moved away from unions, they moved to a sector characterized by very ¯exible wages and thus by higher wage inequality. Furthermore, a large percentage of these workers were workers with low educational attainment. On the other hand, the decline of unionization rate led to a drop in the bargaining power of unions as suggested by the increase in wage inequality within the unionized sector. Table 8 also shows that the average wage rate declined for both unionized and nonunionized workers throughout the 1990s. The overall drop of real wages is expected given that a key component of Mexico's economic policy since the 1982 economic crisis has been to institute a restrictive wage policy to control in¯ation. In December 1987 labor bureaucracy agreed to the Economic Solidarity Pact that further accentuated the downward trend of real wages. The decline in unionization can be explained by political and economic events that have taken place since the mid-1980s. Garcõa (1993), for example, argues that workers became increasingly disappointed in their labor unions for several reasons. First, workers disliked the anti-democratic means that their union leaders used to elect themselves. Second, workers felt that the organization leadership no longer represented their interests since it negotiated agreements with the government and private sector without considering their opinions. Third, the unwillingness and inability of the leadership to change in accordance with the ongoing economic transformation. Among

1915

workers who have decided to remain unionized, however, a signi®cant portion reacted to these issues by constituting more democratic and representative labor unions (Bensus an, 1999; La Botz, 1988). 25 A discreditization of Mexico's largest labor confederations has occurred in the midst of the industrial restructuring that has accelerated the process of ¯exibilization of the labor market. Zapata (1992) notes two examples. First, the privatization of state enterprises disbanded a large number of unions. 26 Second, increasing pressure on domestic exporting ®rms to achieve international price competitiveness obliged them to subcontract important segments of their production and to hire temporary and part-time workers to reduce labor costs. The increasing ¯exibility has also manifested itself in the form of extended hours of work, easiness to lay-o€ workers and lower severance payments among other changes (Bensus an, 1998). The increased ¯exibility has occurred without changing the Federal Labor Law and without much opposition of the largest union confederations' bureaucracy (Bensus an, 1998, 1999). (b) Changes in minimum wages The decline in unionization as the explanation for increasing wage inequality is incomplete however. Empirical studies indicate that in some economies minimum wages represent an e€ective ¯oor for a large number of workers, while there are cases in which they have little impact on labor market performance. One would expect that whenever minimum wages represent an e€ective ¯oor, wage dispersion would be lower than otherwise. By the same token, if minimum wages do not represent an e€ective ¯oor then changes in the former would not have an impact on either wage structure or employment. In the Mexican case, several studies indicate that minimum wages have an important impact on wage structure and wage inequality while their e€ect on employment is very limited. Feliciano (1995) reports that 18.1% of workers earned around the minimum wage or less while the remaining 81.9% earned more than the minimum. After an analysis of the wage distribution, she concludes that minimum wages might not be an e€ective wage ¯oor either because they are too low relative to the average wage or they are not enforced. Paldam and Riveros (1987) have pointed out that in Mexico, as in Latin America, while the

1916

WORLD DEVELOPMENT

minimum wage legislation establishes workers' needs as the basic criterion for setting minimum wages, in practice they have been ®xed by political assessment and expediency. Given that workers earning minimum wages or less are young, less educated, mostly female, more likely to work in the service sector than in the manufacturing sector, their e€ect on wage inequality would be through the lower end of the wage distribution. Figure 1 illustrates the distribution of workers by hourly wage rate in 1984, 1989, 1992 and 1996 and the hourly real minimum wage. We observe two phenomena over time: ®rst, the entire wage distribution moves leftward, and second, there is a growing clustering around the value of the minimum wage, meaning that it is becoming an e€ective ¯oor for an increasing number of workers. Two common measures used to describe a probability distribution are skewness and kurtosis. 27 In 1984 the wage distribution had skewness equal to 2.747 while the kurtosis was 12.612. By 1996 they had jumped to 3.886 and 21.459, respectively, indicating a leftward movement of the distribution (or an increase of the right-hand tail), while, at the same time, the thickness of the tail shrank considerably. This result seems to contradict ®ndings by other authors in the sense that minimum wages are not binding in Mexican labor market. To some extent, this result is expected since our data include workers located not only in urban areas but also in rural areas. Some may argue that the increase in wage inequality is due to the worsening of working conditions among agricultural workers

compared to nonagricultural ones. We found evidence that suggests, however, that while agricultural workers remain in a very bad situation, the increase in inequality is rather due to changes occurring in the nonagricultural sector. Table 9 shows the wage rate di€erential (c) and wage inequality di€erential (q) between nonagriculture and agriculture for male and female workers, separately. 28 It can be seen that between 1984 and 1996, the average wage rate di€erential between nonagricultural and agricultural workers increased signi®cantly for male and female workers. They went from 2.5 and 2.3 in 1984 to 2.7 and 2.5 in 1996, respectively. Parallel to the increases in wage di€erential, however, wage inequality among nonagricultural workers has been growing faster than among agricultural workers, with the highest growth occurring among male workers. In other words, changes in wage inequality have mainly occurred within the nonagriculture sectors, more speci®cally, among male nonagricultural workers. A decomposition of wage distribution by economic sectors shows that, in e€ect, minimum wages are not binding in manufacturing, ®nancial and service sectors, while they are important in agriculture. Figures 2±5 show the wage distribution in the agriculture, manufactures, commerce and service sectors, respectively. These are the four largest economic sectors since they employ about 80% of workers. From these ®gures, we can infer that while a large proportion of workers in agriculture

Figure 1. Wage distribution: 1984, 1989, 1992 and 1996.

INCREASING WAGE INEQUALITY IN MEXICO

1917

Table 9. Average wage rate and wage inequality between agriculture and nonagriculture sector 1984

1989

1992

1996

Male c q

2.532 0.707

2.654 0.730

2.514 0.835

2.722 1.016

Female c q

2.342 0.766

2.638 0.703

2.833 0.837

2.548 0.843

Source: Own estimates based on INEGI (various years).

Figure 2. Wage distribution in agriculture.

Figure 3. Wage distribution in manufactures.

1918

WORLD DEVELOPMENT

Figure 4. Wage distribution in commerce.

Figure 5. Wage distribution in service sector.

depends on minimum wage, its behavior over time does not explain the change in overall wage distribution. In fact, the leftward movement is largely explained by the behavior in the rest of the economic sectors, in particular, by the shifting of the entire wage distribution of the service sector. To what extent is the leftward movement of the wage distribution independent of the decline in unionization rates? We argue that these phenomena are interrelated: to the extent that labor unions are becoming more ine€ec-

tive at negotiating nominal wages, workers' wages are increasingly being determined by market forces. Minimum wages are therefore becoming an e€ective ¯oor over time. The fact that an increasing number of women and youngsters are entering the labor market, while older workers have increased their average number of hours worked, is also related to the steady decline of real wages which has been accelerated by industrial restructuring. In other words, the erosion of institutional factors, by having a more signi®-

INCREASING WAGE INEQUALITY IN MEXICO

cant e€ect on the lower end of the wage distribution, has contributed to increasing wage inequality. This is particularly important among nonagricultural workers. 6. CONCLUSIONS The model used in this essay is a very simple one and yet it provides some insights about the leading forces behind increasing wage inequality. We conclude that the impact of the increases in the relative rate of return to higher education were outpaced by the e€ects of labor mobility that occurred simultaneously, so that the overall e€ect of educational expansion on wage inequality was a decline. Instead, what seem to be behind the increasing wage inequality are changes in labor market institutions. In this essay we have considered changes in the degree of unionization and changes in minimum wages. We found that the wage dispersion is largely explained by changes in the composition and in the rate of return of unionization. The changes in the Mexican labor market are not independent of the change in economic policy that took place during the mid-1980s. For parallel to trade liberalization there has been an increasing ¯exibility in the labor market. This ¯exibility has

1919

taken several forms. We stressed two of them: (a) the decline of unionization rates across educational levels that have led to the weakening of the union workers' bargaining position and (b) the steady increase of the minimum wage as an e€ective ¯oor in most economic sectors. Since the early 1990 the Mexican labor force seems to be going through a generational change: an increasingly educated labor force that is breaking with old ways of labor organization. This may provide a turning point for changing the nature of the relationship between the state and the labor unions. It is unclear yet if more democratic and representative unions alone would suce to reverse the current trends of wage inequality. In any event, it may lead to increasing uneasiness about current economic policies that, at some point, would have to be tackled. By the same token, the results suggest that policy prescriptions that focus on the expansion of the upper end of the educational distribution as a way to reduce wage inequality, without considering the changes in the labor market institutions, may be ine€ective. Without reducing the importance of education, the focus should be a closer evaluation of the labor market institutions, and how we can induce changes so that inequality declines.

NOTES 1. See, for instance, Aitken, Harrison, and Lipsey (1995), Alarc on and McKinley (1997), Bouillon, Legovini, and Lustig (1999), Feenstra and Hanson (1995), Feliciano (1995), Hanson and Harrison (1995) and Meza (1999).

5. An important characteristic of the changes in Mexican trade policy was that these changes were implemented across sectors and not within industries or ®rms. 6. Estimated correlation was )0.937.

2. Some OECD countriesÐparticularly the United States, United Kingdom, and CanadaÐas well as some developing countries face the same problem. In the case of United States see, for example, Fishlow and Parker (1999), Freeman (1999), Gottschalk (1997) and Topel (1997). For the United Kingdom see Haskel and Slaughter (1999) and for Canada see MacPhail (2000).

7. Between 1984 and 1996, average annual growth rate of the log variance of hourly wage rate in Construction, Electric, Mining Financial and Transport sectors were 5.8%, 3.8%, 3.2%, 2.9% and 2.8%, respectively, while the national average growth rate was 1.2%.

3. Between 1984 and 1996 average hourly wage declined from 5.7 new Pesos to 4.4 new Pesos (1994 ˆ 100).

8. It should be noted that these percentages correspond to employed workers (INEGI, 1984, 1996). Thus, the larger unemployment or subemployment across educational levels the larger the underestimation of actual rates of educational attainment.

4. See Aghion, Garcõa-Pe~ nalosa, and Caroli (1998), Fishlow and Parker (1999), Fortin and Lemieux (1997), Mincer (1995) and Topel (1997) among others.

9. Measured in constant terms (1993 ˆ 100). Source: INEGI (2000).

1920

WORLD DEVELOPMENT

10. See Table 3.

imperfections in Mexican markets, particularly in the capital market, this may not be a proper assumption.

11. See, for instance, Bracho and Zamudio (1994), Hern andez-Licona (1997) and Pagan and Ullibarri (2000). 12. The maximum value of wage inequality is determined by the value of r21 r22 1 ‡ ; k ˆ 2…Y1 Y2 †2 2 where r2 and Y are the log variance and log mean of income, respectively. The subscripts indicate the sectors. 13. If there are kth sectors, overall wage dispersion is P P P kk kj Cov…Uk Yj †, where given by r2 ˆ k2k r2k ‡ 2 k