WPS3771 Labor Market Distortions in Côte d'Ivoire: Analyses of Employer-Employee Data from the Manufacturing Sector Nicolai Kristensen* and Dorte Verner* World Bank 1818 H St, NW Washington, D.C. 20433 U.S.A. dverner@worldbank.org Abstract This paper investigates the extent and nature of distortions in the labor market in the Republic of Côte d'Ivoire (RCI) by using quantile regression analysis on employer-employee data from the manufacturing sector. We found that the labor markets in Côte d'Ivoire do not seem to be much distorted. Unions may influence employment through tenure but do not seem to influence wages directly except for vulnerable minorities that seem protected by unions. Establishment-size wage effects are pronounced and highest for white-collar workers. This may be explained by the efficiency wage theory, so that, even in the absence of unions, segmentation and inefficiencies will still be present as long as firms seek to retain their employees by paying wages above the market clearing level. The inefficiency arising from establishment-size wage effects can be mitigated by education. Furthermore, the premium to education is found highly significantly positive only for higher education, and not for basic education, indicating that educational policies should also focus on higher education. World Bank Policy Research Working Paper 3771, November 2005 The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they represent. Policy Research Working Papers are available online at http://econ.worldbank.org. ______________________________ *We wish to thank colleagues from the Africa regional labor market study, Niels-Hugo Blunch, Wendy V. Cunningham, and Norbert M. Fiess, for their helpful comments and suggestions; and Eliana Cardoso, Helena Ribe, and Arvil van Adams for their invaluable support. 1. Introduction The lack of labor demand is the main problem in Sub-Saharan Africa (SSA), while labor supply is not a big problem. This consideration along with recent adverse terms of trade movements for SSA key export products call for flexibility of labor markets. The labor market in Côte d'Ivoire has been deregulated but few studies analyze the possible existence of labor market distortions both before and after the start of reforms.1 Rama (1998) analyzes wage misalignment in RCI. He concludes that minimum wages are no source of distortion in the Ivorian labor market; real minimum wages in 1996 accounted for only about 35 percent of their level 20 years earlier.2 In this study we focus on two possible sources of labor market distortions in Côte d'Ivoire in the mid-late 1990s. First, we investigate the possible effects of unions on the individual income generation in the private manufacturing sector. Even though this sector involves only a small part of the labor force, it still plays an important role since a large part of exports originates from this sector, and since any export commodity dispersion is likely to come from this sector.3 Second, we analyze the possible existence of efficiency wages arising from employer-size wage effects. Furthermore, the paper aims at analyzing the characteristics of the small group of private manufacturing sector workers--who to some extent may be viewed as role models--and how they differ within and between occupational groups. This may have implications for educational policies, for example, and is an important complement to the household-based analysis by Verner (1999c), which found that union effects cannot be analyzed in the 1990s due to survey changes. One reason for the lack of studies that look at union effects in RCI is that the topic can only be satisfactorily analyzed by firm level data. This paper presents one of the first studies of employer-employee data, the so-called Regional Program on Enterprise Development (RPED) data set on Côte d'Ivoire from 1995 and 1996. This type of data represents a clear advantage to prior studies of the labor market, which rely on household surveys that are rather sparse on 1Appendix E contains an overview of labor market studies in RCI that makes use of regression techniques, based on individual data, to analyze wage/income generation. 2This drop is in line with market average wages which fell 60 percent in the 1985-95 period (Verner, 1999). 3Rama (1998) quotes ILO numbers, estimating 300,000 individuals as being union members in 1995 out of a total labor force of 7,063,000, which is equivalent to 4.25 percent of the entire labor force. These numbers are only estimates and highly uncertain. 2 information about firms. Likewise, the data at hand proves superior to macro or aggregate data since the latter do not allow for direct linking of employers and employees. The analysis is carried out using quantile regression techniques, which allow a more in- depth characterization of individuals across the entire wage distribution, than does the more standard technique of Ordinary Least Squares (OLS). The paper is organized as follows. Section 2 gives a brief overview of the Ivorian economy in the late 1990s. Section 3 outlines the economic model and explains the basic principles behind quantile regressions. Section 4 describes the surveys and the variables. Section 5 discusses a number of descriptive statistics and section 6 presents analytical results from quantile regressions. Finally, section 7 concludes with a summary of findings and policy implications and recommendations. 2. The Ivorian economy in the late 20th century As many Sub-Sahara African countries, RCI experienced a very severe recession throughout the 1980s, triggered by a dramatic fall in export prices for key commodities such as cocoa, coffee, and oil. At the same time, the CFA was pegged to the French franc, and the latter appreciated substantially against the US dollar, resulting in a deterioration of competitiveness and a drop in exports from RCI. Furthermore, the 1980s were characterized by a labor market heavily restricted by rules and regulations. For instance, as reported by Rama (1998), the government had monopoly over hiring decisions. All vacancies had to be reported to central offices, which also were responsible for registering all job searchers and making all placements, despite a tendency for these offices to be highly inefficient. Firing costs were high, too, since workers fired for economic reasons often managed to obtain several yearly salaries due to the courts' assessment that firing was "abusive". The 50 percent devaluation of the CFA in 1994 removed a major obstacle to economic adjustment in Côte d'Ivoire, and resulted in a major repatriation of flight capital from the early 1990s when the currency was overvalued. Afterwards, the economy has blossomed. Annual GDP growth has averaged more than 6 percent; investments almost doubled from 1993 to 1996; and the wage bill to GDP ratio declined by one third in the same period. The labor market was 3 also included in the general drive towards more democracy. Competition in collective bargaining was introduced and the monopoly of hiring and firing decisions was abandoned.4 Sub-Sahara Africa, including RCI, experienced a dramatic fall in the prices of its key commodities. The prices in the late 1990s were the lowest in 30 years, and the outlook for long- term real prices is not favorable. Exports from RCI constitute 28 percent of GDP, which is relatively high for a Sub-Saharan country but far below the level obtained in RCI in the late 1970s. Furthermore, exports from RCI is highly concentrated with 46 percent of total exports to OECD in 1997 being in one single commodity, cocoa, which makes the economy highly vulnerable to price fluctuations.5 3. Methodology 3.1. The economic model One could argue that wages and employment are determined jointly and hence constitute a system of equations to be estimated (Verner 1999a, b). But, Maloney and Ribeiro (1999) note that standard wage equations with employment omitted can be thought of as a reduced form and can be estimated using one-step estimators such as least squares or quantile regressions. Hence, the underlying economic model used in the analysis will simply follow Mincer's (1974) human capital earnings function extended to control for a number of other variables that relate to the firm. In particular, we apply a semi-logarithmic framework that has the form: ln yi = (xi, zi) + ui (1) where ln yi is the log of earnings or wages for individual i; xi is a measure of a number of personal characteristics, including human capital variables, ethnicity, etc.; and zi represents firm- specific variables, for instance, profit per employee in the firm. The functional form is left unspecified in equation (1). We choose to make extensive use of dummy variables in order to catch non-linearities in returns to years of schooling, tenure, and other quantitative variables. The last component, ui, is a random disturbance term that reflects unobserved characteristics. 3.2. Quantile regressions 4Verner (1999c) gives a more detailed historical description of the RCI economy. 5Source: UN Comtrade, World Bank (1998a, b). 4 The method applied in this paper is quantile regression. The idea is that one can choose any quantile and thus obtain many different parameter estimates. In this manner the entire conditional distribution can be explored. By testing whether coefficients for a given variable across different quantiles are significantly different, one implicitly also tests for conditional heteroskedasticity across the wage distribution. The method has many other virtues apart from being robust to heteroskedasticity. When the error term is non-normal, for instance, quantile regression estimators may be more efficient than least squares estimators.6 Furthermore, since the quantile regression objective function is a weighted sum of absolute deviations, one obtains a robust measure of location and, as a consequence, the estimated coefficient vector is not sensitive to outlier observations on the dependent variable.7 The main advantage, though, is the semi-parametric nature of the approach, which relaxes the restrictions on the parameters to be fixed across the entire distribution. Intuitively, quantile regression estimates convey information on wage differentials arising from non- observable characteristics among individuals otherwise observationally equivalent. In other words, by using quantile regressions, we can determine if individuals that rank in different positions in the conditional distribution (i.e., individuals that have higher or lower wages than predicted by observable characteristics) receive different premiums to education, tenure, or to other relevant observable variables. Labor market studies usually make use of conditional mean regression estimators, such as Ordinary Least Squares8. This technique is subject to criticism because of several, usually heroic assumptions underlying the approach. One is the assumption of heteroskedasticity in the distribution of the error terms. If the sample is not completely homogenous, this approach, by 6Ibid. 7That is, ify i - xi^ > 0, then yi can be increased toward + , or if y - x ^ < 0 i i , yi can be decreased toward -, without altering the solution ^ . In other words, it is not the magnitude of the dependent variable that matters but on which side of the estimated hyperplane the observation is. This is most easily seen by considering the FOC to n (3), which can be shown to be given as (see Buchinsky 1998) 1 1 1 n ( - + sgn( yi - xi^ )) xi = 0. 2 2 i=1 This can be seen both as a strength and a weakness of the method. In the present context, with data from Côte d'Ivoire, the advantages seem to outweigh the disadvantages, since the reliability of data from LDCs generally is questionable. However, to the extent that a given outlier represents a feature of the "true" distribution of the population, one would prefer the estimator to be sensitive to such an outlier--at least to a certain degree. 8See Velenchik (1997) and Verner (1999a,b). 5 forcing the parameters to be the same across the entire distribution of individuals may be too restrictive and may hide important information. A simple solution and yet a powerful tool is to make use of quantile regression techniques. The method of quantile regression amounts to minimizing the absolute sum of errors rather than, as in least squares, minimizing the sum of their squares. Formally the method, first developed by Koenker and Basset (1978), can be formulated as9 yi = xi + ui = Quant(yi | xi) = xi (2) where Quant(yi | xi) denotes the th conditional quantile of y given x, and i denotes an index over all individuals, i = 1,...,n. In general, the th sample quantile (0 < < 1) of y solves min = 1 | yi - xi | + (1 - ) | yi - xi | (3) n i : yi xi i: yi < xi Buchinsky (1995) examines various estimators for the asymptotic covariance matrix and concludes that the design matrix bootstrap performs the best. In this paper, the standard errors are obtained by bootstrapping using 200 repetitions, in line with the literature. 4. Data description 4.1. The Surveys The data used in this study are drawn from surveys of manufacturing firms in Côte d'Ivoire conducted in 1995 and 1996. The surveys are part of the RPED, a multi-year study of the manufacturing sector in several African countries (Cameroon, Côte d'Ivoire, Ghana, Kenya, Rwanda, Burundi, Tanzania, Zambia, and Zimbabwe). The RPED was organized by the Africa Region Technical Department of the World Bank. The RPED is designed to provide an overview of the performance of manufacturing firms in the post-structural adjustment period, and focuses on a wide variety of aspects of firm behavior. The survey instruments include several modules covering creation of the enterprise; the enterprise in general; interior competition; labor markets; financial markets; solutions to 9See Buchinsky (1998). 6 conflicts; infrastructure; regulations; adjustments; investors; use of help from (public) institutions; employees; and apprentices. In Côte d'Ivoire, data were collected for two successive years, covering 234 and 230 manufacturing enterprises in 1995 and 1996, respectively. A moving panel structure was applied, which implies that some of the firms surveyed in 1995 also were surveyed in 1996. Likewise, some workers surveyed in 1995 were also included in the sample in 1996. Problems of sample attrition and missing values for some key variables haunt the surveys and have seriously reduced the sample size, both in terms of number of observations and variables, requiring that we pool the two samples. By doing so, we implicitly assume that the relatively few firms (26) and workers (95) that are counted twice are no source of bias; that is, firms and workers present in both years do not, in any systematic fashion, deviate from firms and workers not present in both years.10 The wages of the pooled sample are deflated, using 1995 as the base year. The samples do not contain information on regional locality, presumably because the vast majority of firms are located in Abidjan. Hence, no regional or spatial deflation is feasible. To the extent that some firms are from outside Abidjan, results may lead to a downward bias in the salaries of workers employed in these firms. Furthermore, women represent only 10 percent of the observations. This may mirror the true picture; but the sample is deemed too small for meaningful analysis of women, and these observations are, therefore, excluded from the analysis. Hence, both sample and any policy recommendations are limited to male workers employed in the manufacturing sector. The number of observations is 891 males that work in 128 different manufacturing enterprises.11 Despite the abovementioned problems, the data set remains highly interesting, since it is among the first data collected on manufacturing enterprises in Côte d'Ivoire. For the selected group consisting of male workers employed in the manufacturing sector, we are able to perform meaningful analysis despite the limitations discussed above. 10In the regression analysis, we include a dummy that equals 1 if the year is 1996 and zero otherwise. This dummy almost always turns out to be significant. Our interpretation is that, while the samples may differ, none can, a priori, be said to be more representative of the "true" population. Pooling the samples can perhaps average out any non- representativeness, resulting in a pooled sample that may even be more representative. 11Ideally, one would take into account the potential (self)-selection taking place here, and estimate, e.g., a selection model in the vein of Heckman (1979) and others. However, it is increasingly recognized that this method requires assumptions that are not likely to hold. For instance, assumptions on what is determining the selection mechanism has to be made, and the results in the wage-equation are often very sensitive to these assumptions. Therefore, we choose not to apply any selection method. Given that we only study males, the issue of self-selection may be less of a problem as compared to biases, had the analysis included females, where selection tends to be most prevalent. 7 4.2. Variables Dependent variable: We use wages as our dependent variable. Throughout the paper, these are calculated and reported as real monthly wages. In the quantile regressions, the dependent variable is the natural log of monthly wages.12 Explanatory variables: Age dummies: yrgroup1 includes all workers aged 15-25; yrgroup2 includes workers aged 26-45 years; and yrgroup3 includes workers aged 46-65. These variables proxy for general experience (firm-specific experience is captured by tenure (see below)). Educational dummies: edunone includes all workers with no education; edubasic includes workers who have obtained primary or secondary education diplomas; and eduhigh includes workers who have obtained a higher degree than secondary diploma. Union variables: Two union variables are included in the analysis. The first variable union takes on the value of 1 if the worker himself indicates he is a union member, and zero otherwise. The second variable density is a proxy for the degree of unionization in each establishment. It is constructed as the ratio of interviewed workers in a given firm that state they are members of a union relative to all interviewed workers in the firm. Tenure dummies: tenure1 is a dummy variable that takes on the value of 1 if tenure in the current job is 0-5 years and zero otherwise; tenure2 equals 1 if tenure in current job is 6-12 years and zero otherwise; and tenure3 equals 1 if tenure is longer than 12 years and zero otherwise. Occupational dummies: The workers are split into 5 different categories: manager includes management, supervisors, and foremen; admsales includes administration and sales personnel; techmain includes technicians and maintenance; qprod includes qualified production workers; and support is a dummy variable for support staff. Industrial dummies: The workers are split into 5 broad industrial categories, which are textile, food, wood, metal, and a group with "other industry" or "industry not stated" called otherin. 12Any measure of in-kind benefits is not part of this variable and is not included in the analysis. Admittedly, these may be important but there are too many missing values for them to be included. 8 Firm size: The number of employees in each firm was split into 4 groups: microf includes firms with less than 17 employees; smallf includes firms with 17-40 employees; mediumf includes firms with 41-99 employees; and largef includes firms with 100 or more employees. Furthermore, we constructed a variable, firm40, which takes on the value of 1 if the number of employees in the firm is 40 or more, and zero otherwise. Other variables: We include a dummy variable for nationality/ethnicity, Ivorian, that equals 1 if the worker's nationality is Ivorian and zero otherwise; a dummy variable for the nationality of the owner, franlib, that takes on the value of 1 if there is a majority of French or Lebanese owners and zero otherwise. Finally, we include a variable, profitper, to measure profit per employee in each firm. 5. Descriptive statistics A little more than half of the workers included in the sample are trade union members (table A1 in appendix A). Not surprisingly, larger firms tend to be more unionized than smaller firms. About 72 percent of the men working in a firm with 40 employees or more report union membership, while the corresponding number for firms with less than 40 employees is 31 percent. Since 57 percent of the firms with 40 or more employees are owned by a majority of either French or Lebanese stockholders, naturally the same picture emerges when one looks at unionization across ownership. Sixty-seven percent of the workers in French/Lebanese owned firms are in a union as compared to 43 percent of the workers in firms not owned by a majority of French/Lebanese stockholders. However, the causality between rate of unionization, size of firm, and nationality of ownership is not clear. Workers with no education (edunone) are less prone to be organized than workers with basic education, and even less prone than workers with high education (42 percent, 63 percent, and 52 percent, respectively). Occupational groups such as managers, administrative and sales personnel, technicians, and maintenance staff are relatively more often members of a union than are qualified production workers and, especially, than are support personnel. Almost half of the workers report tenure of less than 6 years in the firm where they are currently working, while the rest are equally distributed in the `6-12' and `more than 12' years of tenure brackets. Tenure is generally much longer in French/Lebanese owned firms. These firms 9 also tend to have more educated workers, so again there is an issue of causality.13 French/Lebanese-owned enterprises are not so active in the food, wood, or textile industries, but are mainly operating in the metal industry (47 percent). Very few of the workers (6 percent) report they have received training after entering the firm (in the form of either on-the-job training or training outside the enterprise). With so few observations, it is very hard to generalize these findings. However, it seems as if firms with more than 40 employees are undertaking more training (9 percent) than smaller firms (2 percent)--corresponding to more training in firms with the higher educated workforce. Presumably, the higher the education the more "trainable" is the individual and, hence, the return to training increases with the level of education. The data do support this hypothesis (table A4), but the number of observations is very low. The monthly real wages for the three different tenure groups are surprisingly closely distributed (see figure A1), whereas the three educational groups are widely dispersed. Highly educated workers earn much higher wages than workers with no education or only basic education. The earnings of highly educated workers are also very heterogeneous within the group, but they vary much more across the percentiles than do the wages for the other two educational groups. The graph also indicates that wages of workers that are union members may be a little more homogeneous than wages of non-union members, since their wage curve is slightly more flat across the percentiles. This corresponds to the calculations of decile ratios given in table A5. This indicates that in the upper part of the distribution (50th to 90th), union and non-union member wage distributions differs the most. In the low end (10th to 50th), the ratio 50th/10th is almost the same for the 2 groups (union = 1.96; nonunion = 2.14), while the 90th/50th ratio differs more (union = 2.45; nonunion = 4.60). Finally, the firm size seems to influence the level of the wage distribution. Large firms pay higher wages (figure A1), but there is less inequality within each group of firm size. The inequality is almost identical within the two groups of firms (table A5). 13Is the tenure longer because French/Lebanese firms treat their employees differently/better or because they hire more educated staff and tenure is positively related to the level of education? The data do not support the last hypothesis: eduhigh generally has a lower tenure related to it than does edubasic/edunone (see table A4). One should not put too much emphasis on these very partial descriptions, though. 10 6. Analytical results 6.1. General results Table B1 presents the quantile regression results and, as a reference, OLS results. Estimates for the 10th, 25th, 50th, 75th, and 90th quantiles are presented to identify differences between low earners and high earners conditional on observables. Most of the variables enter with the expected sign at all quantiles. In the following, each variable's impact on wages is discussed. Age and tenure. Both age dummies yrgroup2 and yrgroup3 are positive and highly statistically significant at all quantiles (10th, 25th, 50th, 75th, and 90th), hence, impacting wages in RCI positively. The age dummies proxy for general, as opposed to firm-specific experience. The regressions show that workers with general experience receive a premium, and the premium increases with increased skill level. Tenure is also positive and the estimated coefficients are non-constant as they increase with tenure. Tenure over 12 years (tenure3) is significant for all quantiles, while 6-12 years of tenure (tenure2) is highly insignificant in the upper quantiles (75th and 90th). At the 90th quantile, the coefficient even becomes negative, though it is statistically insignificant. One would expect tenure to have a positive effect on wages as it captures a pay-off to experience. In Côte d'Ivoire this pay-off does not seem to be systematic and, therefore, possibly delinks experience from wages.14 Education. Completed basic education with diploma generally enters with the expected positive sign but is statistically insignificant at all quantiles (with respect to no completed education). This may be explained by the fact that we only analyze wages in the manufacturing sector (see more below). Completed higher education relative to no completed education is positive and significantly different from zero at the 1 percent level across all quantiles. The premium obtained by highly educated workers ranges between 33 percent (10th) and 105 percent (90th), and increases almost linearly with the quantile, vis-à-vis workers with no education.15 Why is there not a premium to basic education as compared to no completed education-- a result that strongly differs from prior expectations and from previous results in the literature (see survey-table in appendix E)? In Verner (1999c) the return to basic education is generally found significant and positive. When restricting the 1995 sample to males only (as we have 14The result is robust to different splines. In particular, introducing a dummy for 0-1 years of tenure and a dummy for 2-5 years of tenure along with tenure2 and tenure3 does not change the result. 11 here), Verner also gets an insignificant return to basic education at the 10th and 25th quantiles. This could be a sign of a `structural' shift in the wage setting since the return to basic education in the 1980s is found to be statistically significant. One explanation could also be that our sample is highly selective: we do not look at the population as a whole but only at those privileged males employed in the manufacturing sector. This selection may be more pronounced for individuals with no education since, relatively, these individuals are less prone to be in the manufacturing sector. An analysis of the entire population could result in a significant impact on wages from having completed basic education.16 It is worth noticing that basic education impacts wages significantly in the OLS regression and, hence, analysis based on this method would have given an entirely different conclusion. Our interpretation of the finding is that education is important for the individual wage formation, but that basic level education (primary and secondary diplomas) is not enough to affect the wage setting process. It may serve as a screening device in the hiring process. For education to have a significant impact on wages in the manufacturing sector, a higher level of education is required. Employer size effects. There is a very significant positive premium to being employed in large firms--the group of comparison is firms with less than 17 employees (microf). The relationship is not linear since firms with 17-40 employees (smallf) get a higher premium than firms with 41-99 employees (mediumf). This employer-size wage effect may arise from simple market based factors and merely be a sign of an efficient market. It could, however, also be a sign of inefficiency--through efficiency wages, if it is not warranted by inferior working conditions or other factors. This issue will be taken up in section 5.3. Industrial sectors. The industry dummies are generally insignificant except for the metal industry, where employees receive a wage premium ranging between 15.2 percent (25th) to 35.1 percent (10th) vis-à-vis the textile industry. The metal industry is primarily run by French/Lebanese owners but this is already controlled for in the regression--the dummy variable for French/Lebanese ownership (franlib) is also positive and highly significant. 15To obtain these premiums, one has to calculate (exp (coefficient estimate) - 1) * 100. 16Theoretically, another possibility is that school quality could have fallen. We have no information on this issue. 12 Union. When all the abovementioned factors are taken into account, the union membership variable becomes statistically insignificant at all quantiles. Hence, unions do not appear to have any measurable impact on workers' wages. One exception is the 75th quantile, where union membership impact wages significantly negatively. The "return" to union membership at the 75th quantile is ­13 percent.17 This finding would be highly surprising for an analysis of labor markets in any OECD country. For an LDC, it is less surprising due to the high level of underemployment, and our finding is also in line with the findings of several other studies.18 An interpretation of the finding is that workers who earn a relatively high wage (that is, are at the 75th quantile), given measured characteristics, are "punished" by the unions. Assume a worker with non-observable characteristics, such as motivation and reliability that are very "good". He ought to be paid a higher wage than other workers with the same observable characteristics; but union membership will push him down the wage ladder towards the average. It could also be an indication of a social awareness among the union member workers in the 75th quantile, since they accept the unions' negative impact on their wage outcome--maybe in order to help secure the firms' competitiveness and their own future employment in a world of rapid technological change. This interpretation points to a very different, more positive effect stemming from unions than they are normally attributed in the literature, which mainly focuses on OECD countries. In the theoretical literature on unions, it has long been recognized that unions may influence factors other than wages such as, for example, security in employment (tenure) or less variation in wages. Still, a negative coefficient is unexpected. Maloney and Rebeiro (1999) obtain similar results for the impact of union density on skilled-worker wages in Mexico. They argue it could be due to more successful bargaining over firm rents by unskilled workers (received in forms other than wages), or that it is related to a desire to reduce the wedge between skilled and unskilled remuneration for equity reasons. Such equity considerations may also apply here. The issue of union influence and power will be further analyzed in the next section. Occupation. The occupational variables turn out to be highly significant for all occupational groups and for all quantiles. The reference group is support staff. The wage premium to manager as compared to support staff increases over the quantiles from 100 percent 17Substituting union membership with union density in the model specification, following the assumption that the union power depends on this, does not in any way alter the above results. 13 (10th) to 200 percent (90th). Not surprisingly, the premium decreases as one goes down in the "worker hierarchy." The technician/maintenance group receives premiums ranging from 43 percent (10th) to 110 percent (90th); the administration/sales group receives a premium of some 26 percent (10th) up to 96 percent (90th); and finally the qualified production workers receive premiums ranging from 7 percent (25th) to 22 percent (90th). Generally, these numbers do only reveal substantial between-group heterogeneity; they also suggest extreme within-group heterogeneity across the distribution of male manufacturing workers. The within (occupational) group heterogeneity is largest in the top of the "hierarchy" and decreases gradually.19 The same pattern of increasing premiums over the quantile-level is found for education: highly educated workers are much more heterogeneous than basic educated workers. This is illustrated graphically for both education and occupation in figure B1, where 90 percent confidence limits also are given. The figure suggests that not only are the coefficients changing across the distribution but the changes also seem to be statistically significant. Indeed, simple one-sided t-tests confirm that coefficients change for occupational groups and higher education and that these changes generally are significant at the 1 percent level (table B2). The standard regression techniques (OLS and 2SLS) hide many insights that are statistically important. In the same vein, the t-tests show that for almost all variables there is too much heterogeneity to restrict the coefficient of a variable to be the same across the entire wage distribution. 6.2. Separate analysis of union and non-union members In the following, we split the worker sample into two groups: unionized and non- unionized. We split the sample to analyze if unions cause wage differentials or if they are more prone to be present in certain firms that also happen to pay higher (or lower, if the estimated coefficient to union is statistically significant and negative) wages. In the quantile regressions including all observations, tenure2 is insignificantly different from zero (table B1). When the samples are split into unions being present at the workplace versus unions not being present, the tenure2-coefficient becomes significantly different from zero across quantiles in the union sample and remains insignificant in the non-union sample (table C1). Therefore, constraining the tenure2 coefficient to be equal for both sub-groups could 18See Verner (1999a) and Rama (1998). 14 give rise to differences in the union coefficient. The differing tenure2 coefficient estimates may indicate, for example, union power that leads to wage promotions according to tenure.20 The problem sketched here is one of selection bias in measuring the union premium, which perhaps can be mitigated by including the density variable in the union regressions. Recall that density measures the share of workers interviewed at a given work plant that are union members, over all workers interviewed at the plant. This variable is used as a proxy for unionization in the entire enterprise. A Mann-Whitney test for equality of sub-populations (workers in unionized firms are equal to workers in non-unionized firms) strongly rejects equality at the 1 percent level, which also suggests that union and non-union workers may be fundamentally different.21 On these grounds the sample was split into the two sub-samples: one for workers from firms where part of the workforce are union members--that is, where density is different from zero; and one for workers from firms with no union members--that is, where density is equal to zero. A number of insights are obtained by splitting the samples by unionized firms (table C1). The coefficient to the indicator variable for Ivorians shows that Ivorians earn a much higher income at the lower quantiles than non-Ivorians, conditional on workers that are employed in non-unionized enterprises. The premium to Ivorians is as high as 52 percent at the 25th quantile. In enterprises with some degree of unionization, Ivorian is insignificant except at the top quantiles (75th and 90th) where it becomes statistically significant and negative. This suggests that union-membership or union presence at a work-site protects non-Ivorian workers from wage discrimination. Then, non-Ivorian workers who earn low wages (given other measured characteristics)--maybe, perhaps, due to a low degree of motivation, reliability, or other unobservable characteristics--are, therefore, pushed by non-Ivorian unions towards the average income class. Undoubtedly, this group is primarily (if not solely) comprised of other African 19Manager includes both management, supervisors, and foremen. However, if one looks at management only, the heterogeneity still prevails. 20See Borjas (1996). Maloney and Ribeiro (1999) find similar results for Mexico. 21Given the low number of observations, we face some limitations. Therefore, in what follows, we concentrate on robust results. 15 nationalities and does not include, say, French citizens. This group probably belongs in the top of the distribution that leads to the reverse in the sign.22 Just as union was insignificant when all workers were included in the regressions, density turns out insignificant here, except at the 90th quantile where it is significantly negative. This hardly suggests any union power in the wage bargaining process--a surprising result that supports our previous findings in section 5.1. Union power can, as mentioned in the last section, have different shades. If it does not affect wages directly, maybe it has power along various other dimensions--one of them seemed to be to protect non-Ivorians in the lower end of the wage distribution as we discussed previously. Other channels may be tenure, profit per employee, or occupational group, each of which we will consider in turn in the following. Tenure: As mentioned in the introduction to this section, with respect to 0-5 years of tenure, we find 6-12 years of tenure (tenure2) to be statistically significant (at least at the 10-15 percent significance level) for the 10th-50th quantiles in the regressions for unionized firms and generally insignificant and/or with lower premium in the non-unionized firm regressions. This may suggest that career paths in the unionized world are more rigid, following rules of promotion and wage increases that are based on tenure and not so much individual performance. However, the story could also simply be that certain enterprises tend to train their employees more, and the same enterprises happen to have some degree of unionization. Neither story can be ruled out. Whether unionized firms train their workers more than non-unionized firms is difficult to tell given the low number of workers who have received training. However, as table A1 suggests, this may be the case. Profit per employee: In the unionized firm regressions, the profit per employee is significantly positive across the entire distribution (the coefficient is very low though, about 1.1E-08), while it is insignificant across the entire distribution for non-unionized firms. This indicates that union power leads to profit-sharing, and thus, indirectly, unions may affect wages. Occupation: The occupational groups have a vast impact on wages in the unionized firms, while they generally have a much smaller and often insignificant impact on wages in non- unionized firms. A priori, one would expect such a pattern, and it may be taken as evidence of union power that results in more rigid wage patterns between groups. Thus, the between- 22No quantile regressions were run with the three groups--Ivorians, other Africans, and other nationalities--since this would lead to too few observations in certain brackets. 16 (occupational)-group heterogeneity seems to originate, at least partly, from the presence of unions, and it shows the result of the detailed grid wage bargaining that takes place in RCI. What about the within-(occupational)-group heterogeneity? One would expect it to be less pronounced in the unionized enterprises since unions tend to average out differences. This does not seem to be the case here, though, as becomes evident when looking at the distribution of coefficients across quantiles for each occupational group (figure C1). The occupational coefficients vary much more across quantiles for workers in unionized enterprises than they do for workers in non-unionized enterprises, which indicate a higher degree of within-group heterogeneity. Not only are the coefficients higher and more significant for workers in unionized enterprises, they also vary much more. This is surprising, and does not conform to a high degree of union power. 5.3. Establishment-size effects The preceding paragraph suggests that union power particularly affect tenure. In the literature, tenure has also been associated with establishment-size premiums; that is, that larger firms pay higher wages to their workers as compared to observationally equivalent workers in smaller establishments.23 The question here is whether unions succeed in negotiating a higher job security, which results in longer tenure, or whether the long tenure arises because large establishments pay relatively higher wages, which results in a low turn-over rate? Unfortunately, this cannot be analyzed in depth by the data at hand. That establishment-size effects may play a very large role in Côte d'Ivoire seems likely given our previous results that suggest intrinsic differences between small and large firms. A Mann-Whitney test for whether the individual earnings for workers employed in firms with less than 40 employees are equal to firms with 40 employees or more, strongly rejects the hypothesis of equality. On this basis, we split up the sample once more.24 The results so far (section 6.1) show a non-linear relationship in firm-size premiums where small firms (17-40 employees) at some quantiles receive a greater size-premium than medium size firms (41-99 employees). That 23For LDCs see Schaffner (1998) and Velenchik (1997). 24An alternative would be to simply include interaction-effects between variables. Since this, however, would not lend the same flexibility for all coefficients to vary (except if interaction terms are included for every single variable), we choose not to do that. 17 we choose to cut the sample at firm size 40 is, therefore, based more on considerations related to equal size of sub-samples since we have relatively few observations. The results indicate that, indeed, tenure may be strongly related to establishment-size effects (see table D1). Tenure is more often significant and generally receives a higher premium in large firms (40 or more employees) after controlling for union density. Thus, it could be that unions just happen to be more present in large establishments and that these pay an employer- size wage effect that lead to a lower degree of turn-over. Union density enters significantly positive in the small-establishment regressions at the lower quantiles (10th and 25th), while it enters significantly negative at the upper quantiles in the large-establishment regressions. The first result corresponds to what one would expect and indicates that unions protect the low-income workers who, given measured human capital and other variables, are in the low end of the income distribution. These workers are pushed upwards by the union. For the upper quantiles, though, why would anybody be a union member if the pay-off is negative? An explanation could be that job security increases, which also partly may explain the difference between tenure premiums in large vs. small firms, since large firms tend to be more unionized. Why does the employer size-wage effect occur? A number of explanations have been discussed in the literature. It could simply be an efficient market-determined pay-back to inferior working conditions. Why, then, should working conditions be more inferior in large establishments? One explanation is that larger firms may have to hire workers from a geographically broader area, which will increase commuting costs. Unfortunately, we do not have data to control for this. However, it seems highly unlikely that this should be the whole story.25 The market-based explanation of the employer-size wage effect is probably not enough. Other economic explanations include problems and costs associated with monitoring shirking, hiring costs, and costs connected with the screening of applicants combined with the need for high quality workers. Non-economic explanations have also been launched. A sociological explanation is that an excessive wage will increase the morale and will be viewed as a gift that must be paid back. A political explanation argues that large firms want to maintain some 18 monopoly power and be on good footing with local government representatives. Of these possible explanations, the shirking and/or turnover explanations may be the most promising, but why should these costs increase with firm size? Higher capital-labor ratios, more sophisticated technology, higher ratios of workers to owners, and deeper hierarchies in larger establishments may increase costs of monitoring, hiring, training, screening or failing to obtain workers of high quality. Another almost "stylized fact" in the literature is that the establishment-size effect on wages is higher for white-collar than blue-collar workers.26 This is also the case for Côte d'Ivoire. When we run quantile regressions for white-collar workers and blue-collar workers separately and include firm40 as the only firm-size dummy, the result that emerges is consistent with the literature (see figure D1). This pattern can also be explained by the efficiency wage story if it is harder to monitor white-collar workers; if they are more prone to quitting their job voluntarily; or if white-collar workers receive more (firm-specific) training than blue-collar workers. With the data at hand, we cannot verify nor reject these possibilities. The bottom line is that part of what appears to be union power probably can be explained by establishment-size effects, and that both have a non-negligible effect on wages. 6.4 Spillover effects from unionized to non-unionized establishments The explanation for the absence of a positive union/non-union wage differential could be that wages secured by unionized workers spill-over to raise wages in establishments that are not unionized. Pencavel (1995) set up a model that control for this possibility. He introduces three sectors, only one of which is unionized. In addition to the unionized sector, there is another relatively high wage sector that is influenced by wage-setting practices in the unionized sector. Firms in this sector may pay high wages to discourage unionization or to reduce turnover. Lastly, there is a third sector, which is not influenced by the unionized sector, and where wages are low. According to this model, unions raise wages not only for their members but also for the high-wage non-unionized workers indirectly influenced by the union negotiations. This leads to 25 Velenchik (1997) controls for the existence of electricity, phone, etc. as proxies for good or bad working conditions. This seems fruitless since establishments with more than 100 employees surely all have these facilities (in compliance with the findings of Velenchik). 19 an important insight: the premium received by unions is not mirrored in the wage differential between the unionized and non-unionized high-wage sectors. In the absence of unions, wages in both these sectors would decrease and employment in the two sectors would increase, while the third low-wage sector would remain unaffected. In the analysis above, we have not included such a third sector since the data do not allow it. Instead, we have what can be considered as the two high-wage sectors. Hence, to the extent that such a spillover effect exists, our measure of the impact from unions must be considered a lower bound, since any such spillover effect will lead to a downward bias in the returns to unionization. The idea of a third sector may also explain part of the establishment-size wage effect since, presumably, small non-unionized firms are less prone to have their wage setting tied to the wage setting in the unionized firms. This does not explain the difference between the premium to blue vs. white-collar workers that we observed above and, hence, cannot be the only reason for the establishment size wage effects. 7. Conclusion Summary of findings The objective of this paper has been to analyze possible channels of distortions in the labor markets in Côte d'Ivoire by using enterprise micro-level data. The tool kit used is the technique of quantile regressions, which has proven its usefulness by providing a much more detailed analysis across the distribution than least square techniques could have accomplished. The main findings are the following: (1) Basic education: The wage premium related to basic education is insignificant across all quantiles when the entire sample is used as well as for all sub-samples. (2) Higher education: The wage premium from higher education is very significant and positive at all quantiles for all sub-samples, except for firms with less than 40 employees, and the premium is relatively high. (3) Occupational groups: The effect from occupational groups on wages, on the other hand, are generally very significant, and the heterogeneity both within and between these groups 26See Schaffner (1998). We define blue collar workers as quality production workers or support staff, while white collar workers include managers, adm/sales, and tech/maintenance. 20 is found to be substantial, and much larger in unionized enterprises than non-unionized, a finding that questions the actual power of unions. (4) Union: Union power does not seem to influence wages directly. Union is insignificant in the analysis that includes all observations. When unionized firms are analyzed separately, and a proxy for the degree of unionization is included, the proxy (density) turns out to be statistically insignificant. Union influence and power seem mostly to come through tenure. Tenure is very important in unionized enterprises and not at all so in non-unionized enterprises. Union power also seems to be protecting "vulnerable" minorities in the Ivorian society-- probably mostly "other Africans." The union-premium may be significantly higher if there is a spill-over effect to the non- unionized firms included in our sample, since this will bias the union premium downwards. Data that include manufacturing firms along with more informal sector firms are welcomed to cast light over this issue. (5) Establishment-size: Effects from establishment size on wages are pronounced. Workers with comparable observable human capital characteristics earn widely different wages, depending on firm size (in a non-linear fashion). The effect is found much higher for white- collar workers than for blue-collar workers. Both results are in sync with the literature, and may be explained by efficiency wage considerations. As previously mentioned, these results do not extend to the entire economy but only hold for the limited and highly selected group of workers employed in the manufacturing sector. Policy implications and recommendations Basic education is not sufficient, and does not lead to higher wages among the "golden league" of workers that the manufacturing sector employees constitute. But since both high education and occupational groups are found very important in generating personal income, formal education is highly recommendable, but should go beyond basic education. Intellectual capital is important, and more of it may lead to a reduction in wage inequality. Further, setting aside for a moment the possible significant spill-over effects from unionized to non-unionized firms (and hence downward bias in the union premium), labor market distortions do not seem to be of primary concern in the manufacturing sector. The unions show no sign of "monopoly union" power, where--in the extreme case--the unions quote 21 the wages. Instead, unions seem more concerned with maintaining and securing jobs. However, unions do lead to inefficiency if the payback to tenure represents rigidly enforced seniority-based promotions that are not based on an increase in the actual level of human capital. This may well be the case since, presumably, if they were based on human capital considerations, tenure would also turn out important in non-unionized enterprises. A part of the story about tenure, however, is probably related to efficiency wage considerations related to establishment-size wage effects. This means that, even in the absence of unions, segmentation and inefficiencies will still be present as long as firms seek to retain their employees by paying wages above the market clearing level. Education will, in this case, also improve the income distribution through efficiency wage channels, reducing the individual firm's transferable human capital investment which, according to this theory, will reduce the efficiency wage premium. In any event, the recent decentralization and introduction of enterprise-level bargaining will make it more difficult for union confederations to pursue any aggressive rent-seeking policies, and will reduce the degree to which collective bargaining is politicized. Hence, this policy should be further encouraged. 22 References Borjas, G. (1996). Labor Economics, New York: McGraw-Hill. Buchinsky, M. (1998). "Recent Advances in Quantile Regression Models ­ A Practical Guideline for Empirical Research," The Journal of Human Resources, Vol. XXXIII, No. 1, pp. 88-126. Buchinsky, M. (1995). "Estimating the Asymptotic Covariance Matrix for Quantile Regression Models: A Monte Carlo Study," Journal of Econometrics, 68:303-38. Grootaert, C. (1990). "Returns to Formal and Informal Vocational Education in Côte d'Ivoire: The Role of the Structure of the Labor Market," Economics of Education Review, Vol. 9, No. 4, pp. 309-319. Heckman, J. J. (1979). "Sample Selection Bias as a Specification Error," Econometrica, Vol. 47, No. 1, pp. 153-161. Hoddinott, J. (1996). "Wages and Unemployment in an Urban African Labour Market," The Economic Journal, 106, 1610-1626. Koenker, R. and G. Basset, Jr. (1978). "Regression Quantiles," Econometrica, Vol. 46, No. 1, pp. 33-50. Komenan, A. G. and C. Grootaert (1990). "Pay Differences Between Teachers and Other Occupations: Some Empirical Evidence from Côte d'Ivoire, Economics of Education Review, Vol. 9, No. 3, pp. 209-217. Maloney, W. F. and E. P. Ribeiro (1999). "Efficiency Wage and Union Effects in Labor Demand and Wage Structure in Mexico: An Application of Quantile Analysis," Working Paper, The World Bank. Mincer, J. (1974). "Schooling, Experience and Earnings," NBER Working Paper, New York. Pencavel, J. (1995). "The Role of Labor Unions in Fostering Economic Development," Policy Research Working Paper, No. 1469, The World Bank, Washington, D.C. Rama, M. (1998). "Wage Misalignment in CFA Countries: Are Labor Market Policies to Blame?," Policy Research Working Paper, No. 1873, The World Bank, Washington D C. Schaffner, J.A. (1998). "Premiums to employment in larger establishments: evidence from Peru," Journal of Development Economics, Vol. 55 (1998), pp. 81-113. Van der Gaag, J. and W.P.M. Vijverberg (1989). "Wage Determinants in Côte d'Ivoire: Experience, Credentials, and Human Capital," Economic Development and Cultural Change, No. 37, pp. 371-381. 23 Van der Gaag, J. and W.P.M. Vijverberg (1988). "Switching Regression Model for Wage Determinants in the Public and Private Sectors of a Developing Country," Review of Economics and Statistics, 70:244-252. Van der Gaag, J. and W.P.M. Vijverberg (1991). "The Private Sector in Côte d'Ivoire: Homogeneity or Heterogeneity?," Southern Economic Journal 58(2): 406-21. Vijverberg, W.P.M. (1994). "Educational Investment and Returns for Women and Men in Côte d'Ivoire," The Journal of Human Resources, XXXVIII (4). Vijverberg, W.P.M. (1989). "Labor Market Performance as a Determinant of Migration," LSMS Working Paper, No. 59, The World Bank, Washington D.C. Velenchik, A.D. (1997). "Government intervention, efficiency wages, and the employer size wage effect in Zimbabwe," Journal of Development Economics, Vol. 53 (1997), pp. 305-338. Verner, D. (1999a). "Are Wages and Productivity in Zimbabwe Affected by Human Capital Investment and International Trade?," Policy Research Working Paper, No. 2101, The World Bank, Washington D.C. Verner, D. (1999b). "Wage and Productivity Gaps: Evidence from Ghana," Policy Research Working Paper, No. 2168, The World Bank, Washington, D.C. Verner, D. (1999c). "What happened to wages in Côte d'Ivoire in the 1980s and 1990s? An application of quantile regressions," Mimeo, The World Bank, Washington, D.C. World Bank (1995). "Labor and Growth Crisis in Sub-Saharan Africa," Regional Perspectives on World Development Report 1995, The World Bank, Washington, D.C. World Bank (1998a). Global Development Finance, Washington, D.C. World Bank (1998b). World Development Indicators, Washington, D.C. 24 Appendix A Table A1. Basic descriptive statistics All union members not union members french-lebanese not french- lebanese firm size 40 empl firm size < 40 empl freq perc freq perc freq perc freq perc freq perc freq perc freq perc all 891 100.0 496 100.0 395 100.0 465 100.0 426 100.0 528 100.0 363 100.0 Ivorian 704 79.0 416 83.9 288 72.9 380 81.7 324 76.1 445 84.3 259 71.4 other african 173 19.4 79 15.9 94 23.8 72 15.5 101 23.7 77 14.6 96 26.5 other nation 14 1.6 1 0.2 13 3.3 13 2.8 1 0.2 6 1.1 8 2.2 not union 395 44.3 NA NA NA NA 151 32.5 244 57.3 146 27.7 249 68.6 union 496 55.7 NA NA NA NA 314 67.5 182 42.7 382 72.4 114 31.4 yrgroup1 (15-25 yrs) 87 9.8 18 3.6 69 17.5 18 3.9 69 16.2 20 3.8 67 18.5 yrgroup2 (26-45 yrs) 679 76.2 395 79.6 284 71.9 374 80.4 305 71.6 424 80.3 255 70.3 yrgroup3 (46-65 yrs) 125 14.0 83 16.7 42 10.6 73 15.7 52 12.2 84 15.9 41 11.3 tenure1 (0-5 yrs) 412 46.2 167 33.7 245 62.0 177 38.1 235 55.2 191 36.2 221 60.9 tenure2 (6-12 yrs) 245 27.5 177 35.7 68 17.2 141 30.3 104 24.4 159 30.1 86 23.7 tenure3 (more than 12 yrs) 234 26.3 152 30.7 82 20.8 147 31.6 87 20.4 178 33.7 56 15.4 edunone (no education) 223 25.0 94 19.0 129 32.7 88 18.9 135 31.7 94 17.8 129 35.5 edubasic (basic education) 503 56.5 316 63.7 187 47.3 295 63.4 208 48.8 313 59.3 190 52.3 eduhigh (higher education) 165 18.5 86 17.3 79 20.0 82 17.6 83 19.5 121 22.9 44 12.1 manager 114 12.8 65 13.1 49 12.4 66 14.2 48 11.3 83 15.7 31 8.5 administration/sales 136 15.3 79 15.9 57 14.4 76 16.3 60 14.1 90 17.1 46 12.7 technicians/maintenance 88 9.9 58 11.7 30 7.6 57 12.3 31 7.3 61 11.6 27 7.4 qualified production workers 281 31.5 167 33.7 114 28.9 147 31.6 134 31.5 154 29.2 127 35.0 support staff 272 30.5 127 25.6 145 36.7 119 25.6 153 35.9 140 26.5 132 36.4 not french-lebanese owned 426 47.8 182 36.7 244 61.8 NA NA NA NA 226 42.8 200 55.1 french-lebanese owned 465 52.2 314 63.3 151 38.2 NA NA NA NA 302 57.2 163 44.9 year 95 579 65.0 307 61.9 272 68.9 217 46.5 362 85.0 319 60.4 260 71.6 year 96 312 35.0 189 38.1 123 31.1 248 53.5 64 15.0 209 39.6 103 28.4 firm size < 40 employees 363 40.7 114 23.0 249 63.0 163 35.0 200 47.0 NA NA NA NA firm size 40 employees 528 59.3 382 77.0 146 37.0 302 65.0 226 53.0 NA NA NA NA never received training 832 93.6 456 92.3 376 95.2 431 93.1 401 94.1 478 90.9 354 97.5 received training 57 6.4 38 7.7 19 4.8 32 6.9 25 5.9 48 9.1 9 2.5 food industry 216 24.2 122 24.6 94 23.8 76 16.3 140 32.9 121 22.9 95 26.2 textile industry 109 12.2 30 6.1 79 20.0 32 6.9 77 18.1 68 12.9 41 11.3 wood industry 232 26.0 120 24.2 112 28.4 97 20.9 135 31.7 133 25.2 99 27.3 metal industry 263 29.5 192 38.7 71 18.0 218 46.9 45 10.6 172 32.6 91 25.1 other industry/not stated 71 8.0 32 6.5 39 9.9 42 9.0 29 6.8 34 6.4 37 10.2 25 Table A2. Summary Statistics, workers variable all union non union french/lebanese non french/lebanese mean std mean std mean std mean std mean std monthly wages, all 128,766 186,957 124,113 131,001 134,610 239,443 159,452 224,500 95,271 126,553 managers 263,112 278,876 202,863 133,901 343,035 384,378 287,760 311,371 229,222 225,555 adm/sales 189,978 239,870 179,938 206,101 203,893 281,428 226,934 269,796 143,167 187,447 tech/main 191,142 217,369 168,669 167,471 234,590 289,238 237,032 255,823 106,762 59,159 qual. prod work. 85,326 88,393 94,225 78,587 72,290 99,998 98,888 95,624 70,448 77,364 support staff 66,551 117,809 68,034 27,767 65,252 159,504 82,846 174,940 53,877 24,840 firm size (# empl) 169.4 535.1 211.4 675.4 116.6 262.1 136.9 187.7 204.82 747.5 value added/empl 3,927,864 8,088,673 3,106,221 6,661,085 4,959,597 9,493,680 4,110,592 7,526,535 3,728,407 8,665,307 export share 24.75 34.83 27.47 36.43 21.30 32.39 26.11 35.17 23.24 34.42 union density 0.54 0.42 0.84 0.23 0.16 0.26 0.68 0.37 0.40 0.42 (share in union) Table A3. Summary Statistics, enterprises variable All french/lebanese not french/lebanese firm size 40 firm size < 40 mean std mean std mean std mean std mean std export share 21.8 34.7 31.4 38.3 15.3 30.7 40.24 40.00 4.48 15.09 value added/empl 3,350,537 8,165,914 4,312,062 8,378,298 2,713,682 8,013,524 3,769,156 7,596,123 2,957,289 8,707,023 union density (share in union) 0.42 0.44 0.65 0.40 0.26 0.40 0.68 0.38 0.17 0.34 Table A4. edunone edubasic eduhigh freq perc freq perc freq perc all 223 100.00 503 100.00 165 100.00 tenure1 117 52.47 205 40.76 90 54.55 tenure2 54 24.22 149 29.62 42 25.45 tenure3 52 23.32 149 29.62 33 20.00 training 5 2.25 36 7.17 16 9.70 no training 217 97.75 466 92.83 149 90.30 26 Table A5. Some percentiles and ratio's of percentiles for monthly wages percentiles and ratio's all union non-union french/lebane not french/leb firm size 40 firm size < 40 percentiles 10 36,000 46,000 28,100 49,355 29,268 46,406 29,561 50 75,000 90,000 60,000 92,683 58,537 92,683 58,537 90 236,798 220,294 276,073 278,049 195,122 270,732 158,702 ratio's 50/10 2.08 1.96 2.14 1.88 2.00 2.00 1.98 90/10 6.58 4.79 9.82 5.63 6.67 5.83 5.37 90/50 3.16 2.45 4.60 3.00 3.33 2.92 2.71 27 Figure A1. Distribution of real monthly wage, various groups 500000 350000 firm>=40 firm < 40 union nonunion 450000 300000 400000 350000 250000 300000 200000 250000 150000 200000 150000 100000 100000 50000 50000 0 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 percentile percentile 500000 600000 tenure1 tenure2 tenure3 edunone edubasic eduhigh 450000 550000 500000 400000 450000 350000 400000 300000 350000 250000 300000 200000 250000 200000 150000 150000 100000 100000 50000 50000 0 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 percentile percentile 28 Appendix B, General results (using the entire sample) Table B1. Quantile regressions and OLS, all observations n=891 quantile regressions OLS 0.1 0.25 0.50 0.75 0.90 coef pvalue coef pvalue coef pvalue coef pvalue coef pvalue coef pvalue constant 9.3556 0.000d 9.7860 0.000d 10.0287 0.000d 10.2785 0.000d 10.5778 0.000d 9.9651 0.000 d individual characteristics yrgroup2 0.4546 0.002d 0.2257 0.033 c 0.2198 0.002d 0.2442 0.000d 0.2238 0.024 c 0.2732 0.000d yrgroup3 0.5529 0.001d 0.3434 0.002d 0.3868 0.000 d 0.3756 0.000d 0.3650 0.008d 0.4473 0.000d Ivorian 0.0980 0.166 0.1763 0.002d 0.0921 0.149 a -0.0143 0.821 -0.1838 0.199 -0.0426 0.408 union 0.0193 0.739 0.0192 0.663 -0.0645 0.303 -0.1441 0.056b -0.0812 0.352 -0.0868 0.074b edubasic -1.5E-8 1.000 0.0494 0.337 0.0621 0.270 0.0503 0.368 0.0927 0.261 0.1077 0.042 c eduhigh 0.2825 0.018c 0.4001 0.000d 0.5569 0.000d 0.6310 0.000d 0.7155 0.002d 0.6208 0.000d tenure2 0.0324 0.650 0.0684 0.204 0.1265 0.055b 0.0572 0.347 -0.1067 0.245 0.0585 0.249 tenure3 0.2727 0.000d 0.2461 0.000d 0.2039 0.002d 0.1861 0.010d 0.0541 0.600 0.1831 0.001d french/leban 0.1070 0.071b 0.1348 0.001d 0.1425 0.022 c 0.2277 0.000d 0.2075 0.017 c 0.1448 0.003d occupation manager 0.6907 0.000d 0.5443 0.000d 0.7100 0.000d 0.9154 0.000d 1.1033 0.000d 0.7401 0.000d adm/sales 0.2300 0.031 c 0.2842 0.002d 0.4467 0.000d 0.6524 0.000d 0.6715 0.000d 0.4780 0.000d tech/mainten 0.3607 0.001d 0.3417 0.000d 0.5996 0.000d 0.7028 0.000d 0.7461 0.000d 0.5750 0.000d qualified prod. 0.1542 0.018 c 0.0689 0.127 a 0.1515 0.002d 0.1272 0.000d 0.2009 0.022 c 0.1430 0.005d industry food 0.1125 0.383 0.0615 0.454 0.2300 0.004d 0.2718 0.000d 0.2954 0.002d 0.1856 0.009d wood 0.1384 0.267 0.0550 0.455 0.0968 0.210 0.1290 0.060b 0.1877 0.040 c 0.1121 0.114a metal 0.3006 0.008d 0.1411 0.057b 0.1941 0.007d 0.2099 0.005d 0.2459 0.033 c 0.2629 0.000d other/no stated 0.1667 0.213 0.0169c 0.855 0.0205 0.864 0.1877 0.047 c 0.1524 0.293 0.1317 0.151 firm size small firm 0.3477 0.000d 0.2975 0.000d 0.2439 0.002d 0.3523 0.000d 0.5534 0.000d 0.4012 0.000d medium firm 0.1334 0.251 0.2270 0.002d 0.2481 0.008d 0.4127 0.000d 0.5323 0.000d 0.3327 0.000d large firm 0.4067 0.000d 0.3572 0.000d 0.2930 0.000d 0.3996 0.000d 0.5272 0.000d 0.4956 0.000d value added 1.1E-8 0.000d 1.1E-8 0.001d 8.0E-9 0.012 c 1.1E-8 0.002d 1.0E-8 0.101 a 1.2E-8 0.000d year dummy 0.1111 0.126 a 0.1808 0.001d 0.1943 0.000d 0.1901 0.001d 0.2382 0.012 c 0.1981 0.000d (1996=1) Note: a--significance at the 15% level; b--10%; c--5%; d--1%. 29 Table B2. Tests of equality of coefficients across quantiles, all observations Tests are one-sided t-tests, H0: coefficient(QA) = coefficient(QB). The t-statistic is given as the numerical value. Data are pooled over the two years 1995 and 1996 and are for men only. a-significant at the 15% level b-significant at the 10% level c-significant at the 5% level d-significant at the 1% level yrgroup2 yrgroup3 Ivorian union QA QB t-stat p-value t-stat p-value t-stat p-value t-stat p-value 0.1 0.25 3.23 0.0363c 2.04 0.0768b 1.29 0.1283a 0.00 0.4995 0.1 0.50 2.70 0.0504b 0.92 0.1688 0.01 0.4712 1.16 0.1408a 0.1 0.75 1.88 0.0855b 0.95 0.1645 1.46 0.1135a 3.50 0.0309c 0.1 0.90 2.01 0.0783b 0.93 0.1677 3.27 0.0355c 0.93 0.1682 0.25 0.50 0.00 0.4753 0.14 0.3519 1.86 0.0866b 2.19 0.0696b 0.25 0.75 0.03 0.4343 0.05 0.4125 6.44 0.0057d 4.68 0.0154c 0.25 0.90 0.00 0.4943 0.02 0.4508 6.19 0.0065 1.17 0.1401a 0.50 0.75 0.15 0.3471 0.01 0.4540 2.62 0.0529b 1.50 0.1107a 0.50 0.90 0.00 0.4841 0.02 0.4413 3.88 0.0246c 0.03 0.4271 0.75 0.90 0.05 0.4149 0.01 0.4657 1.90 0.0842b 0.65 0.2105 edubasic eduhigh tenure2 tenure3 QA QB t-stat p-value t-stat p-value t-stat p-value t-stat p-value 0.1 0.25 0.62 0.2149 1.29 0.1281a 0.35 0.2784 0.21 0.3253 0.1 0.50 0.78 0.1888 3.66 0.0281c 1.28 0.1287a 0.73 0.1971 0.1 0.75 0.40 0.2625 6.51 0.0055d 0.09 0.3804 0.99 0.1598 0.1 0.90 0.81 0.1841 2.95 0.0432c 1.65 0.0999b 3.94 0.0238c 0.25 0.50 0.06 0.4038 2.07 0.0752b 0.93 0.1681 0.58 0.2234 0.25 0.75 0.00 0.4943 3.91 0.0242c 0.03 0.4338 0.79 0.1875 0.25 0.90 0.22 0.3195 1.83 0.0880b 3.26 0.0358c 3.32 0.0344c 0.50 0.75 0.04 0.4165 0.41 0.2604 1.29 0.1281a 0.07 0.3932 0.50 0.90 0.13 0.3580 0.47 0.2464 5.84 0.0080d 2.09 0.0744b 0.75 0.90 0.43 0.2550 0.19 0.3327 4.14 0.0211c 2.22 0.0681b manager admsales techmain qprod QA QB t-stat p-value t-stat p-value t-stat p-value t-stat p-value 0.1 0.25 1.11 0.1464a 0.32 0.2847 0.03 0.4272 2.12 0.0731b 0.1 0.50 0.01 0.4563 3.32 0.0344c 4.49 0.0172c 0.00 0.4842 0.1 0.75 1.60 0.1030a 13.65 0.0001d 4.46 0.0176c 0.11 0.3675 0.1 0.90 3.07 0.0401c 5.98 0.0074d 4.47 0.0174c 0.19 0.3301 0.25 0.50 2.36 0.0623b 2.62 0.0528b 8.34 0.0020d 2.69 0.0508b 0.25 0.75 11.21 0.0004d 12.69 0.0002d 6.52 0.0054d 0.73 0.1963 0.25 0.90 8.21 0.0022d 4.96 0.0131c 6.59 0.0052d 1.89 0.0846b 0.50 0.75 3.96 0.0234c 6.04 0.0071d 0.70 0.2018 0.21 0.3228 0.50 0.90 4.29 0.0193c 2.19 0.0698b 1.00 0.1586 0.34 0.2790 0.75 0.90 1.23 0.1343a 0.02 0.4398 0.09 0.3832 1.13 0.1440a yeard96 franlib valadper smallf QA QB t-stat p-value t-stat p-value t-stat p-value t-stat p-value 0.1 0.25 1.21 0.1356a 0.22 0.3214 0.06 0.4020 0.32 0.2848 0.1 0.50 1.17 0.1399a 0.20 0.3276 0.77 0.1902 0.96 0.1639 0.1 0.75 0.79 0.1875 2.37 0.0622b 0.04 0.4215 0.00 0.4844 0.1 0.90 1.14 0.1428a 0.99 0.1604 0.02 0.4437 1.83 0.0883b 0.25 0.50 0.07 0.3978 0.02 0.4431 0.79 0.1865 0.52 0.2352 0.25 0.75 0.02 0.4443 2.32 0.0642b 0.00 0.4900 0.34 0.2792 0.25 0.90 0.29 0.2944 0.64 0.2123 0.00 0.4833 3.33 0.0341c 0.50 0.75 0.01 0.4707 1.64 0.1002a 0.59 0.2216 1.82 0.0890b 0.50 0.90 0.23 0.3154 0.45 0.2508 0.13 0.3587 5.60 0.0091d 0.75 0.90 0.33 0.2831 0.07 0.3954 0.00 0.4868 2.77 0.0482c 30 Table B2 (continued). Tests of equality of coefficients across quantiles, all observations Tests are one-sided t-tests, H0: coefficient(QA) = coefficient(QB). The t-statistic is given as the numerical value. Data are pooled over the two years 1995 and 1996 and are for men only. a-significant at the 15% level b-significant at the 10% level c-significant at the 5% level d-significant at the 1% level mediumf largef food wood QA QB t-stat p-value t-stat p-value t-stat p-value t-stat p-value 0.1 0.25 0.81 0.1840 0.39 0.2651 0.17 0.3387 0.49 0.2411 0.1 0.50 0.74 0.1944 1.26 0.1308a 0.74 0.1956 0.11 0.3724 0.1 0.75 4.44 0.0178c 0.00 0.4734 1.25 0.1316a 0.00 0.4722 0.1 0.90 6.48 0.0056d 0.93 0.1680 1.45 0.1144a 0.11 0.3710 0.25 0.50 0.07 0.3982 0.76 0.1913 4.68 0.0154c 0.34 0.2805 0.25 0.75 4.11 0.0215c 0.19 0.3333 5.35 0.0105c 0.80 0.1858 0.25 0.90 5.71 0.0086d 1.94 0.0819b 4.07 0.0219c 1.50 0.1108a 0.50 0.75 4.17 0.0207c 1.94 0.0821b 0.32 0.2849 0.20 0.3282 0.50 0.90 4.45 0.0176c 3.99 0.0231c 0.39 0.2670 0.77 0.1902 0.75 0.90 1.12 0.1452a 1.53 0.1085a 0.07 0.3980 0.41 0.2600 metal other industry QA QB t-stat p-value t-stat p-value 0.1 0.25 2.34 0.0632b 1.40 0.1182a 0.1 0.50 0.87 0.1759 0.86 0.1770 0.1 0.75 0.53 0.2325 0.02 0.4454 0.1 0.90 0.14 0.3540 0.01 0.4696 0.25 0.50 0.61 0.2179 0.00 0.4870 0.25 0.75 0.62 0.2154 2.21 0.0689b 0.25 0.90 0.67 0.2059 0.72 0.1977 0.50 0.75 0.05 0.4125 2.32 0.0640b 0.50 0.90 0.19 0.3320 0.66 0.2091 0.75 0.90 0.11 0.3702 0.08 0.3918 31 Figure B1 m a n a g e r , 9 0 % C I b a n d (b o o ts tr a p p e d 2 0 0 ) a d m /s a le s , 9 0 % C I b a n d (b o o ts tra p p e d 2 0 0 ) 1 .6 1 .8 1 .4 1 .6 1 .2 1 .4 1 1 .2 0 .8 1 0 .6 0 .8 0 .6 0 .4 0 .4 0 .2 0 .2 0 1 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0 9 9 0 - 0 .2 1 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0 9 9 q u a n t ile -0 .2 q u a n tile t e c h m a i n , 9 0 % C I b a n d ( b o o t s t r a p p e d 2 0 0 ) q u a lifie d p r o d u c tio n w o r k e r , 9 0 % C I b a n d (b o o ts tr a p p e d 2 0 0 ) 1 . 6 1 .6 1 . 4 1 .4 1 . 2 1 .2 1 1 0 . 8 0 .8 0 . 6 0 .6 0 .4 0 . 4 0 .2 0 . 2 0 0 1 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0 9 9 1 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0 9 9 - 0 .2 - 0 . 2 q u a n tile q u a n t i l e e d u b a s ic , 9 0 % C I b a n d (b o o ts tr a p p e d 2 0 0 ) e d u h i g h , 9 0 % C I b a n d ( b o o t s t r a p p e d 2 0 0 ) 1 .6 1 . 6 1 .4 1 . 4 1 .2 1 . 2 1 1 0 .8 0 . 8 0 .6 0 . 6 0 .4 0 . 4 0 .2 0 . 2 0 1 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0 9 9 0 -0 .2 1 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0 9 9 q u a n tile - 0 . 2 q u a n t i l e 32 Appendix C, Results when splitting union and non-union members Table C1. Regression results, Workers in unionized firms. n=634 quantile regressions OLS # firms=68 0.1 0.25 0.50 0.75 0.90 coef pvalue coef pvalue coef pvalue coef pvalue coef pvalue coef pvalue constant 9.8539 0.000d 9.9580 0.000 d 10.2771 0.000d 10.5651 0.000d 10.7920 0.000d 10.2709 0.000d individual characteristics yrgroup2 0.0862 0.453 0.1034 0.254 0.1685 0.053b 0.1950 0.191 0.3969 0.049 c 0.2606 0.021 c yrgroup3 0.2221 0.104a 0.1807 0.118a 0.2735 0.032 c 0.3881 0.037 c 0.5514 0.018 c 0.4298 0.001d Ivorian 0.0485 0.540 0.0493 0.490 -0.0279 0.758 -0.2590 0.024 c -0.4328 0.007d -0.2119 0.002d union density -0.0735 0.548 -0.0321 0.745 -0.0299 0.790 -0.0432 0.771 -0.4554 0.083b -0.1857 0.080b edubasic -0.0263 0.694 1.3E-8 1.000 0.0136 0.797 0.0919 0.172 0.1282 0.197 0.1265 0.057b eduhigh 0.4805 0.000d 0.5243 0.000d 0.5309 0.000d 0.6734 0.000d 0.7233 0.001d 0.7130 0.000d tenure2 0.1479 0.104 a 0.0764 0.162 0.1972 0.005d 0.0740 0.395 -0.1449 0.294 0.0862 0.165 tenure3 0.3460 0.000d 0.2654 0.000d 0.2568 0.001d 0.2836 0.004d 0.0811 0.557 0.2245 0.000d french/leban 0.0800 0252 0.0716 0.127 a 0.1244 0.071b 0.0902 0.291 0.1359 0.194 0.1120 0.065b occupation manager 0.5693 0.000d 06025 0.000d 0.8250 0.000d 1.0304 0.000d 1.2901 0.000d 0.8420 0.000d adm/sales 0.2930 0.016c 0.3521 0.000d 0.6199 0.000d 0.8772 0.000d 1.0357 0.000d 0.6368 0.000d tech/mainten 0.2779 0.017 c 0.2562 0.004d 0.6087 0.000d 0.6879 0.000d 0.8715 0.000d 0.5603 0.000d qualified prod. 0.1296 0.097b 0.1032 0.024 c 0.1676 0.009d 0.2696 0.001d 0.4232 0.000d 0.2231 0.001d industry food 0.1358 0.599 0.2410 0.045 c 0.1581 0.169 0.3192 0.002d 0.5583 0.000d 0.2292 0.019 c wood 0.1816 0.472 0.1012 0.373 -0.0815 0.444 0.1195 0.250 0.4199 0.007d 0.1238 0.219 metal 0.3776 0.131 a 0.2610 0.029 c 0.0875 0.411 0.1721 0.082b 0.4128 0.007d 0.2522 0.010d other/no stated 0.2193 0.435 0.1586 0.296 0.1468 0.391 0.3227 0.016 c 0.5641 0.008d 0.2310 0.082b firm size small firm 0.2202 0.122 a 0.3024 0.009d 0.2689 0.030 c 0.2265 0.099b 0.4744 0.107 a 0.2877 0.009d medium firm -0.0130 0.929 0.0751 0.520 0.1386 0.316 0.1308 0.411 0.2837 0.381 0.0895 0.414 large firm 0.2816 0.044 c 0.3665 0.001d 0.2301 0.067b 0.1543 0.286 0.3661 0.240 0.2847 0.009d value added 1.0E-8 0.020 c 1.3E-8 0.000d 8.3E-9 0.031 c 1.0E-8 0.055b 1.3E-8 0.129 a 1.2E-8 0.000d year dummy 0.1551 0.066b 0.2948 0.000d 0.2213 0.000d 0.1962 0.005d 0.0919 0.425 0.2203 0.000d Workers in non-unionized firms. n=257 quantile regressions OLS # firms=59 0.1 0.25 0.50 0.75 0.90 coef pvalue coef pvalue coef pvalue coef pvalue coef pvalue coef pvalue constant 9.2417 0.000 d 9.5993 0.000 d 9.9875 0.000 10.2812 0.000 d 10.5652 0.000d 9.9802 0.000 d individual characteristics yrgroup2 0.5072 0.052b 0.4157 0.017 c 0.1772 0.122 a 0.2609 0.001d 0.2535 0.002d 0.2458 0.012 c yrgroup3 0.6670 0.009d 0.4244 0.079b 0.3629 0.015 c 0.3827 0.024 c 0.3430 0.191 0.4824 0.001d Ivorian 0.1598 0.296 0.4157 0.002d 0.2696 0.012 c 0.0610 0.428 -0.0506 0.722 0.1532 0.044 c edubasic 0.1034 0.476 -4.0E-8 1.000 -0.0418 0.630 0.0356 0.623 -0.0989 0.352 0.0208 0.805 eduhigh 0.1324 0.618 0.0078 0.962 0.1567 0.435 0.5257 0.074b 0.4892 0.134 a 0.3565 0.006d tenure2 0.1636 0.378 0.0165 0.901 0.1518 0.090b 0.0780 0.374 0.0927 0.336 0.1041 0.237 tenure3 0.3528 0.106 a 0.2664 0.077b 0.3414 0.018 c 0.1738 0.135 a 0.2469 0.099b 0.2607 0.028 c french/leban 0.1949 0.286 0.1380 0.312 0.3314 0.009d 0.2376 0.015 c 0.2579 0.066b 0.3186 0.000d occupation manager 0.0203 0.957 0.5545 0.057b 0.4414 0.063b 0.5263 0.100b 0.8175 0.025 c 0.4460 0.003d adm/sales 0.0310 0.889 0.0223 0.902 0.0723 0.663 0.2276 0.294 0.5095 0.009d 0.1554 0.206 tech/mainten 0.6118 0.015 c 0.5076 0.003d 0.4521 0.007d 0.7168 0.004d 0.7832 0.019 c 0.6479 0.000d qualified prod. 0.2064 0.259 0.0515 0.651 -0.0040 0.965 0.0313 0.685 0.0814 0.421 0.0237 0.773 industry food 0.0316 0.884 0.1503 0.325 0.2329 0.045 c 0.2062 0.039 c 0.0615 0.596 0.1340 0.229 wood -0.0332 0.870 0.0564 0.685 0.1385 0.268 0.0903 0.326 0.0995 0.384 -0.0178 0.855 metal 0.2611 0.348 0.2395 0.290 0.2797 0.138 a 0.2275 0.102 a 0.1533 0.729 0.2389 0.124 a other/no stated 0.1665 0.423 0.1092 0.499 0.1762 0.244 0.1199 0.469 0.1676 0.460 0.1232 0.301 firm size small firm 0.2346 0.278 0.1011 0.510 -0.0951 0.407 -0.1321 0.277 0.1134 0.527 0.0212 0.854 medium firm 0.3674 0.054b 0.4200 0.003d 0.4778 0.000d 0.4661 0.001d 0.3864 0.013 c 0.5590 0.000d large firm 0.7515 0.006d 0.3879 0.082b 0.6619 0.010d 0.9066 0.014 0.8511 0.219 0.8478 0.000d value added 1.3E-8 0.182 6.9E-9 0.286 4.6E-9 0.391 4.7E-9 0.343 3.3E-9 0.928 6.2E-9 0.160 year dummy -0.0336 0.818 0.1047 0.423 0.2046 0.041 c 0.2214 0.017 c 0.3363 0.018 c 0.1823 0.052b Note: a--significance at the 15% level; b--10%; c--5%; d--1%. 33 Table C2. Tests of equality of coefficients across quantiles, unionized enterprises only Tests are one-sided t-tests, H0: coefficient(QA) = coefficient(QB). The t-statistic is given as the numerical value. a-significant at the 15% level b-significant at the 10% level c-significant at the 5% level d-significant at the 1% level yrgroup2 yrgroup3 Ivorian union density QA QB t-stat p-value t-stat p-value t-stat p-value t-stat p-value 0.1 0.25 0.02 0.4384 0.11 0.3674 0.00 0.4959 0.16 0.3450 0.1 0.50 0.45 0.2508 0.12 0.3663 0.57 0.2252 0.10 0.3768 0.1 0.75 0.45 0.2521 0.77 0.1905 6.00 0.0073 0.03 0.4348 0.1 0.90 1.72 0.0949 1.72 0.0951 7.38 0.0034 1.81 0.0897 0.25 0.50 0.46 0.2491 0.57 0.2246 0.96 0.1639 0.00 0.4913 0.25 0.75 0.43 0.2568 1.57 0.1056 7.41 0.0034 0.01 0.4717 0.25 0.90 1.93 0.0827 2.45 0.0589 8.29 0.0021 2.66 0.0518 0.50 0.75 0.05 0.4129 0.65 0.2095 4.64 0.0158 0.01 0.4610 0.50 0.90 1.24 0.1327 1.46 0.1140 5.78 0.0083 2.85 0.0461 0.75 0.90 1.16 0.1415 0.58 0.2236 1.46 0.1141 3.39 0.0330 edubasic eduhigh tenure2 tenure3 QA QB t-stat p-value t-stat p-value t-stat p-value t-stat p-value 0.1 0.25 0.21 0.3223 0.17 0.3390 0.92 0.1689 0.95 0.1652 0.1 0.50 0.38 0.2700 0.12 0.3666 0.25 0.3082 0.77 0.1902 0.1 0.75 1.52 0.1089 1.50 0.1109 0.45 0.2515 0.28 0.3000 0.1 0.90 2.19 0.0696 1.16 0.1412 2.88 0.0452 2.42 0.0600 0.25 0.50 0.07 0.3981 0.00 0.4750 3.62 0.0288 0.01 0.4557 0.25 0.75 1.13 0.1442 1.33 0.1247 0.00 0.4894 0.03 0.4284 0.25 0.90 1.72 0.0948 0.91 0.1696 1.86 0.0864 1.34 0.1234 0.50 0.75 1.05 0.1534 1.39 0.1193 1.95 0.0814 0.09 0.3823 0.50 0.90 1.45 0.1142 0.74 0.1957 4.30 0.0193 1.31 0.1368 0.75 0.90 0.18 0.3359 0.07 0.3982 2.47 0.0582 2.42 0.0601 manager admsales techmain qprod QA QB t-stat p-value t-stat p-value t-stat p-value t-stat p-value 0.1 0.25 0.08 0.3881 0.28 0.2977 0.04 0.4225 0.20 0.3287 0.1 0.50 2.66 0.0516 4.67 0.0156 5.18 0.0012 0.24 0.3105 0.1 0.75 6.99 0.0042 14.71 0.0001 7.15 0.0039 2.14 0.0721 0.1 0.90 11.29 0.0004 13.75 0.0001 11.65 0.0004 5.00 0.0129 0.25 0.50 3.83 0.0254 6.05 0.0071 10.4 0.0007 1.45 0.1145 0.25 0.75 8.90 0.0015 19.53 0.0000 10.73 0.0006 4.29 0.0194 0.25 0.90 13.64 0.0001 15.95 0.0001 14.10 0.0001 7.09 0.0040 0.50 0.75 2.56 0.0550 6.02 0.0072 0.41 0.2604 2.26 0.6650 0.50 0.90 6.25 0.0064 5.95 0.0075 2.50 0.0572 4.89 0.0137 0.75 0.90 2.79 0.0478 1.37 0.1214 1.75 0.0931 2.38 0.0618 34 Table C2 (continued). Tests of equality of coefficients across quantiles, unionized enterprises only Tests are one-sided t-tests, H0: coefficient(QA) = coefficient(QB). The t-statistic is given as the numerical value. a-significant at the 15% level b-significant at the 10% level c-significant at the 5% level d-significant at the 1% level yeard96 franlib valadper smallf QA QB t-stat p-value t-stat p-value t-stat p-value t-stat p-value 0.1 0.25 4.24 0.0200 0.02 0.4382 0.48 0.2442 0.37 0.2710 0.1 0.50 0.65 0.2099 0.33 0.2842 0.17 0.3417 0.10 0.3739 0.1 0.75 0.21 0.3226 0.01 0.4571 0.00 0.4905 0.00 0.4869 0.1 0.90 0.21 0.3244 0.21 0.3217 0.09 0.3830 0.81 0.1837 0.25 0.50 1.54 0.1078 0.80 0.1856 1.48 0.1125 0.08 0.3888 0.25 0.75 1.53 0.1080 0.05 0.4111 0.18 0.3365 0.21 0.3237 0.25 0.90 2.45 0.0589 0.33 0.2829 0.01 0.4701 0.40 0.2637 0.50 0.75 0.16 0.3461 0.26 0.3068 0.18 0.3347 0.09 0.3824 0.50 0.90 1.17 0.1400 0.01 0.4583 0.23 0.3164 0.61 0.2172 0.75 0.90 0.91 0.1705 0.19 0.3303 0.12 0.3668 1.07 0.1512 mediumf largef food wood QA QB t-stat p-value t-stat p-value t-stat p-value t-stat p-value 0.1 0.25 0.35 0.2759 0.42 0.2583 0.25 0.3077 0.16 0.3469 0.1 0.50 0.85 0.1787 0.12 0.3648 0.01 0.4643 1.27 0.1301 0.1 0.75 0.49 0.2429 0.47 0.2474 0.48 0.2439 0.06 0.4009 0.1 0.90 0.94 0.1669 0.08 0.3856 2.01 0.0785 0.65 0.2104 0.25 0.50 0.24 0.3134 1.20 0.1371 0.50 0.2402 3.16 0.0380 0.25 0.75 0.11 0.3726 1.74 0.0940 0.27 0.3011 0.02 0.4457 0.25 0.90 0.53 0.2333 0.00 0.4996 2.46 0.0586 2.78 0.0478 0.50 0.75 0.00 0.4796 0.31 0.2878 2.32 0.0643 3.34 0.0340 0.50 0.90 0.26 0.3056 0.26 0.3037 5.36 0.0105 8.05 0.0024 0.75 0.90 0.38 0.2701 0.77 0.1902 3.13 0.0387 4.49 0.0173 metal other industry QA QB t-stat p-value t-stat p-value 0.1 0.25 0.35 0.2781 0.07 0.3993 0.1 0.50 1.55 0.1072 0.06 0.4010 0.1 0.75 0.71 0.1998 0.12 0.3621 0.1 0.90 0.02 0.4498 1.02 0.1568 0.25 0.50 2.37 0.0619 0.00 0.4722 0.25 0.75 0.43 0.2572 0.78 0.1891 0.25 0.90 0.75 0.1933 2.78 0.0479 0.50 0.75 0.66 0.2077 1.17 0.1397 0.50 0.90 4.10 0.0217 3.18 0.0375 0.75 0.90 3.56 0.0298 1.95 0.0816 35 Table C3. Tests of equality of coefficients across quantiles, non-unionized enterprises only Tests are one-sided t-tests, H0: coefficient(QA) = coefficient(QB). The t-statistic is given as the numerical value. a-significant at the 15% level b-significant at the 10% level c-significant at the 5% level d-significant at the 1% level yrgroup2 yrgroup3 Ivorian QA QB t-stat p-value t-stat p-value t-stat p-value 0.1 0.25 0.21 0.3243 1.05 0.1528 3.40 0.0333 0.1 0.50 2.05 0.0767 1.44 0.1158 0.42 0.2590 0.1 0.75 1.13 0.1439 1.14 0.1432 0.32 0.2859 0.1 0.90 1.12 0.1457 0.73 0.1970 1.13 0.1445 0.25 0.50 2.55 0.0559 0.09 0.3832 1.34 0.1244 0.25 0.75 0.89 0.1727 0.03 0.4340 5.89 0.0080 0.25 0.90 0.90 0.1725 0.05 0.4107 7.21 0.0039 0.50 0.75 0.60 0.2195 0.01 0.4546 4.73 0.0154 0.50 0.90 0.40 0.2632 0.00 0.4747 4.86 0.0142 0.75 0.90 0.01 0.4628 0.02 0.4397 1.10 0.1476 edubasic eduhigh tenure2 tenure3 QA QB t-stat p-value t-stat p-value t-stat p-value t-stat p-value 0.1 0.25 0.57 0.2262 0.26 0.3048 0.80 0.1854 0.24 0.3112 0.1 0.50 0.82 0.1836 0.01 0.4700 0.00 0.4753 0.00 0.4775 0.1 0.75 0.18 0.3349 1.08 0.1496 0.19 0.3338 0.78 0.1888 0.1 0.90 1.28 0.1293 0.72 0.1989 0.12 0.3647 0.24 0.3140 0.25 0.50 0.14 0.3563 0.47 0.2461 1.42 0.1174 0.28 0.2986 0.25 0.75 0.08 0.3879 3.14 0.0389 0.21 0.3222 0.35 0.2764 0.25 0.90 0.42 0.2576 1.87 0.0863 0.26 0.3038 0.01 0.4571 0.50 0.75 0.90 0.1722 2.20 0.0698 0.85 0.1784 1.92 0.0836 0.50 0.90 0.25 0.3071 1.08 0.1502 0.31 0.2894 0.39 0.2671 0.75 0.90 2.55 0.0560 0.02 0.4500 0.03 0.4301 0.33 0.2818 manager admsales techmain qprod QA QB t-stat p-value t-stat p-value t-stat p-value t-stat p-value 0.1 0.25 3.12 0.0393 0.00 0.4825 0.27 0.3024 1.17 0.1407 0.1 0.50 1.3 0.1274 0.03 0.4309 0.35 0.2782 1.55 0.1071 0.1 0.75 1.24 0.1337 0.46 0.2481 0.11 0.3678 0.95 0.1657 0.1 0.90 2.58 0.0548 2.64 0.0529 0.16 0.3471 0.42 0.2599 0.25 0.50 0.20 0.3275 0.09 0.3850 0.10 0.3781 0.28 0.2988 0.25 0.75 0.01 0.4686 0.78 0.1891 0.76 0.1928 0.03 0.4327 0.25 0.90 0.46 0.2499 3.85 0.0254 0.50 0.2406 0.05 0.4152 0.50 0.75 0.10 0.3764 0.73 0.1977 1.93 0.0830 0.17 0.3404 0.50 0.90 1.31 0.1269 4.59 0.0166 0.86 0.1778 0.58 0.2238 0.75 0.90 0.95 0.1660 2.43 0.0602 0.04 0.4241 0.34 0.2798 36 Table C3 (continued). Tests of equality of coefficients across quantiles, non-unionized enterprises only Tests are one-sided t-tests, H0: coefficient(QA) = coefficient(QB). The t-statistic is given as the numerical value. a-significant at the 15% level b-significant at the 10% level c-significant at the 5% level d-significant at the 1% level yeard96 franlib valadper smallf QA QB t-stat p-value t-stat p-value t-stat p-value t-stat p-value 0.1 0.25 0.82 0.1835 0.13 0.3604 0.50 0.2402 0.54 0.2314 0.1 0.50 1.69 0.0976 0.52 0.2350 0.75 0.1938 2.37 0.0624 0.1 0.75 1.61 0.1030 0.04 0.4181 0.62 0.2156 2.49 0.0578 0.1 0.90 3.21 0.0372 0.08 0.3905 0.68 0.2054 0.21 0.3232 0.25 0.50 0.62 0.2167 2.41 0.0611 0.19 0.3313 2.06 0.0765 0.25 0.75 0.61 0.2182 0.46 0.2499 0.11 0.3694 1.76 0.0928 0.25 0.90 2.02 0.0784 0.50 0.2411 0.20 0.3273 0.00 0.4780 0.50 0.75 0.02 0.4380 0.65 0.2100 0.00 0.4942 0.08 0.3864 0.50 0.90 0.88 0.1752 0.25 0.3087 0.04 0.4233 1.25 0.1326 0.75 0.90 1.00 0.1594 0.03 0.4308 0.06 0.4031 2.52 0.0568 mediumf largef food wood QA QB t-stat p-value t-stat p-value t-stat p-value t-stat p-value 0.1 0.25 0.09 0.3821 2.51 0.0572 0.38 0.2701 0.25 0.3077 0.1 0.50 0.30 0.2917 0.06 0.3995 0.85 0.1784 0.69 0.2039 0.1 0.75 0.17 0.3415 0.19 0.3313 0.56 0.2267 0.33 0.2840 0.1 0.90 0.01 0.4696 0.02 0.4452 0.02 0.4506 0.35 0.2762 0.25 0.50 0.19 0.3308 1.01 0.1574 0.35 0.2776 0.43 0.2568 0.25 0.75 0.07 0.3957 2.84 0.0467 0.11 0.3724 0.05 0.4100 0.25 0.90 0.03 0.4318 0.47 0.2471 0.22 0.3203 0.07 0.3936 0.50 0.75 0.01 0.4660 0.75 0.1935 0.07 0.3950 0.22 0.3192 0.50 0.90 0.29 0.2948 0.09 0.3848 1.54 0.1082 0.09 0.3796 0.75 0.90 0.28 0.2974 0.01 0.4644 1.48 0.1123 0.01 0.4656 metal other industry QA QB t-stat p-value t-stat p-value 0.1 0.25 0.01 0.4647 0.10 0.3750 0.1 0.50 0.00 0.4746 0.00 0.4834 0.1 0.75 0.01 0.4576 0.03 0.4297 0.1 0.90 0.05 0.4116 0.00 0.4985 0.25 0.50 0.04 0.4227 0.14 0.3528 0.25 0.75 0.00 0.4811 0.00 0.4812 0.25 0.90 0.04 0.4221 0.04 0.4171 0.50 0.75 0.11 0.3728 0.12 0.3653 0.50 0.90 0.11 0.3727 0.00 0.4855 0.75 0.90 0.04 0.4181 0.06 0.4058 37 Figure C1 Manager adm /sales 1.5 1.5 non-union union non-union union 1.3 1.3 1.1 1.1 0.9 0.9 0.7 0.7 0.5 0.5 0.3 0.3 0.1 0.1 -0.1 1 10 20 30 40 50 60 70 80 90 99 -0.1 1 10 20 30 40 50 60 70 80 90 99 quantile quantile technician/maintenance quality production worker 1.5 1.5 non-union union non-union union 1.3 1.3 1.1 1.1 0.9 0.9 0.7 0.7 0.5 0.5 0.3 0.3 0.1 0.1 -0.1 1 10 20 30 40 50 60 70 80 90 99 -0.1 1 10 20 30 40 50 60 70 80 90 99 quantile quantile 38 Workers in firms with 40 employees or more n=528 quantile regressions OLS 0.1 0.25 0.50 0.75 0.90 coef pvalue coef pvalue coef pvalue coef pvalue coef pvalue coef pvalue constant 10.0154 0.000d 10.0864 0.000d 10.3631 0.000d 10.6268 0.000d 10.5598 0.000d 10.3209 0.000d individual characteristics yrgroup2 0.1567 0.454 0.3140 0.070b 0.2424 0.070b 0.3673 0.000d 0.5275 0.000d 0.3513 0.013c yrgroup3 0.2469 0.287 0.3971 0.041 c 0.4094 0.010d 0.5612 0.000d 0.8206 0.000d 0.5571 0.000d Ivorian 0.0151 0.867 0.0831 0.287 0.0047 0.964 -0.1551 0.116 a -0.2030 0.308 -0.1086 0.150 a union density -0.1276 0.307 -0.1098 0.319 -0.1419 0.138 a -0.1353 0.236 -0.1135 0.377 -0.1978 0.026 c edubasic -0.0142 0.870 0.0204 0.764 0.0516 0.460 0.0904 0.251 0.1987 0.064b 0.1262 0.102 a eduhigh 0.3362 0.004d 0.3818 0.001d 0.4994 0.001d 0.6598 0.000d 0.8077 0.001d 0.6918 0.000d tenure2 0.2184 0.076 0.1526 0.056b 0.1347 0.060b 0.0163 0.865 -0.0943 0.463 0.1154 0.101 a tenure3 0.4085 0.001d 0.3583 0.000d 0.2646 0.003d 0.2020 0.032 c 0.0779 0.556 0.2490 0.000d french/leban 0.0977 0.275 0.1080 0.114 a 0.0788 0.281 0.0694 0.361 0.0843 0.399 0.1321 0.048 c occupation manager 0.4634 0.001d 0.5465 0.000d 0.7618 0.000d 0.9502 0.000d 1.2093 0.000d 0.7178 0.000d adm/sales 0.2504 0.020c 0.4276 0.002d 0.6761 0.000d 0.8310 0.000d 0.8517 0.000d 0.5695 0.000d tech/mainten 0.2489 0.094b 0.3708 0.002d 0.6497 0.000d 0.8059 0.000d 0.9944 0.000d 0.6114 0.000d qualified prod. -0.0097 0.903 1.1E-8 1.000 0.0906 0.170 0.1987 0.014 c 0.3678 0.001d 0.1030 0.149 a industry food 0.1237 0.612 0.0954 0.444 0.3038 0.007d 0.3610 0.001d 0.4675 0.003d 0.2757 0.004d wood 0.3911 0.065b 0.1847 0.076b 0.1806 0.098b 0.1880 0.048 c 0.3234 0.019 c 0.2250 0.020 c metal 0.4459 0.047 c 0.1906 0.087b 0.2647 0.013 c 0.1736 0.088b 0.3121 0.009d 0.2387 0.014 c other/no stated 0.8005 0.004d 0.3887 0.023 c 0.4758 0.001d 0.3558 0.001d 0.3684 0.064b 0.4173 0.002d value added 6.5E-9 0.265 1.1E-8 0.007d 8.0E-9 0.015 c 8.4E-9 0.088b 1.4E-8 0.074b 1.0E-8 0.006d year dummy -0.0311 0.775 0.1871 0.016 c 0.1731 0.025 c 0.1920 0.011 c 0.1495 0.288 0.1969 0.005d Workers in firms with less than 40 employees n=363 quantile regressions OLS 0.1 0.25 0.50 0.75 0.90 coef pvalue coef pvalue coef pvalue coef pvalue coef pvalue coef pvalue constant 9..3356 0.000d 9.8308 0.000d 10.040 0.000d 10.5103 0.000d 10.7841 0.000d 10.0582 0.000d individual characteristics yrgroup2 0.3755 0.056b 0.1264 0.152 0.2499 0.006d 0.1821 0.049 c 0.1310 0.204 0.1859 0.035 c yrgroup3 0.5561 0.009d 0.3556 0.015 c 0.3817 0.001d 0.2750 0.028 c 0.0888 0.531 0.3400 0.006d Ivorian 0.0986 0.433 0.1473 0.103 a 0.1318 0.115 a -0.0736 0.399 -0.2961 0.052b 0.0105 0.883 union density 0.2637 0.022 c 0.2264 0.010d 0.0816 0.329 0.2633 0.040 c 0.3698 0.052b 0.2994 0.001d edubasic -4.5E-8 1.000 -0.0015 0.985 -0.0192 0.759 -0.0542 0.403 -0.0422 0.668 0.0197 0.798 eduhigh -0.0458 0.775 0.1510 0.383 0.3850 0.061b 0.6730 0.023 c 0.6452 0.119 a 0.4909 0.000d tenure2 0.1698 0.178 0.1706 0.084b 0.1745 0.031 c 0.1401 0.121 a 0.0489 0.669 0.1373 0.078b tenure3 0.1506 0.199 0.1586 0.178 0.2133 0.017 c 0.1737 0.110 a 0.2031 0.194 0.2571 0.009d french/leban 0.2769 0.008d 0.2855 0.000d 0.3037 0.000d 0.2781 0.002d 0.1882 0.133 a 0.2813 0.001d occupation manager 0.8511 0.000d 0.7674 0.000d 0.7216 0.000d 0.8598 0.003d 0.1603 0.001d 0.8440 0.000d adm/sales 0.0439 0.766 0.1588 0.209 0.2092 0.154 0.4511 0.053b 0.8284 0.006d 0.4462 0.000d tech/mainten 0.6707 0.000d 0.4749 0.000d 0.3628 0.031 c 0.7276 0.003d 0.6454 0.008d 0.5545 0.000d qualified prod. 0.3495 0.001d 0.1921 0.026 c 0.1573 0.028 c 0.1080 0.113 a 0.1563 0.112 a 0.1533 0.041 c industry food 0.2118 0.178 0.1393 0.270 0.2102 0.060b 0.1056 0.300 0.1830 0.128a 0.0831 0.469 wood -0.1287 0.428 -0.0320 0.810 0.1355 0.217 0.0766 0.451 0.2333 0.077b 0.0102 0.925 metal 0.2407 0.076b 0.1751 0.131 a 0.2921 0.011 c 0.1963 0.127 a 0.7059 0.009d 0.3335 0.005d other/no stated -0.2032 0.153 -0.2362 0.028 c -0.2142 0.039 c -0.2480 0.146 a -0.1251 0.571 -0.1693 0.198 value added 2.4E-8 0.000d 1.6E-8 0.000d 9.3E-9 0.008d 7.6E-9 0.176 9.8E-9 0.176 1.4E-8 0.000d year dummy 0.2623 0.003d 0.2229 0.009d 0.1835 0.008d 0.2325 0.012 c 0.3319 0.025 c 0.2636 0.000d Note: a--significance at the 15% level; b--10%; c--5%; d--1%. 0.4 white-collar blue-collar 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 10 15 20 25 30 40 50 60 70 75 80 90 quantile Appendix E, Literature overview of studies of the labor market in Côte d'Ivoire Author(s) Year Purpose of study Data Dependent Method Coefficient of key variables Significant variable ** = 5% Variable Coeff * = 10% Grootaert, C. 1990 To compare graduates of the formal CILSS log Heckman's yrs primary edu 0.031 vocational and technical education 1985 monthly selection model yrs second. edu 0.127 ** system in Côte d'Ivoire with those of earnings yrs higher edu 0.130 ** informal apprenticeships explicitly yrs current job exp 0.084 ** considering the structure of the labor yrs current job exp, sq -0.0015 ** market diplomas (different types) varies none rural dummy -0.289 non-ivorien dummy -0.042 ** female dummy 0.079 Komenan, A.G. and 1990 To analyze pay differences between CILSS log Wage regressions yrs of schooling 0.1678 ** Gootaert, C. teachers and other wage earners in 1985 monthly (also on different diploma 0.1228 ** Côte d'Ivoire controling for various total wages sub-groups ­ not experience 0.1151 ** personal characteristics reported here) experience, sq -0.002 ** log monthly hours worked 0.1302 government dummy -0.0224 female -0.1782 * ivorien 0.1545 abidjan dummy -0.092 teacher dummy 0.008 Hoddinott, J. 1996 To analyze the relationship between CILSS log hourly 2 stage method; a lagged rate of unempl. -0.7492 ** wages and unemployment; with focus 1985, 1986, nominal logit for potential exp 0.0677 * on the urban labor market 1987, wages participating in potential exp, sq -0.0026 pooled the labor market completed grades 0.1283 ** sample head of household -0.1002 married 0.1094 non-ivorien -0.0875 occupation service worker -0.1216 skilled worker -0.0256 teacher/principal 0.5723 ** white collar 0.3420 ** other tech/prof 0.0903 location abidjan or bouake 0.1190 eastern region town -0.3127 ** western region town -0.3908 ** 41 Appendix E, Literature overview of studies of the labor market in Côte d'Ivoire Coefficient of key variables Vijverberg, W.P.M. 1991 To challenge the usual assumption of CILSS log hourly ML two-stage yrs schooling 0.077 and Van der Gaag, J. a homogenous wage sector 1985 wage rate method with both yrs schooling, sq/100 0.240 (including OLS and GLS in experience in current job 0.106 ** the cash the structural exp. current job, sq/100 -0.214 ** value of in- model (GLS general experience 0.048 * kind reported here) general experience, sq/100 -0.085 income female dummy 0.420 ** non-ivorien -0.089 Vijverberg, W.P.M. 1988 To contribute to the Mincerian CILSS log wages Full information Public sector: and Van der Gaag, J. returns to education literature and 1985 for public maximum yrs elem sch 0.035 focus on public-private sector and private likelihood yrs junior high 1 0.205 ** differentials sector (FIML) diploma elem sch 0.801 ** workers diploma junior high 1 0.424 ** respectivel higher diploma 0.621 ** y current job experience 0.087 ** curr job exp, sq*100 -0.868 non-ivorien --- --- female -0.125 RRR 0.108 Private sector: yrs elem sch 0.018 yrs junior high 1 0.012 diploma elem sch 0.395 * diploma junior high 1 0.617 ** higher diploma 0.221 current job experience 0.116 ** curr job exp, sq*100 -2.258 ** non-ivorien 0.285 ** female 0.141 RRR 0.147 * 42 Appendix E, Literature overview of studies of the labor market in Côte d'Ivoire Coefficient of key variables Vijverberg, W.P.M. 1992 To seek answers to the questions: Do Pooled log hourly Two-stage Abidjan, men: women in the labor market enjoy the CILSS earnings in estimation yrs elem school 0.099 ** same returns to their human capital 1985, 1986, wage procedure (self- yrs junior high 0.121 ** investments as men do? Is the 1987 employmen selection model yrs senior high 0.162 ** different treatment of women in the t with two decision yrs university 0.196 ** labor market a cause for the lower variables: where occupational exp 0.083 ** educational attainment so often to live and what occup exp, sq/100 -0.121 ** observed among women in the third work (if any) to prior exp 0.026 ** world? choose. prior exp, sq/100 -0.005 ** non-ivorien -0.110 Abidjan, women: yrs elem school 0.020 yrs junior high 0.286 ** yrs senior high 0.230 ** yrs university 0.246 ** occupational exp 0.102 ** occup exp, sq/100 -0.158 ** prior exp 0.049 ** prior exp, sq/100 -0.057 ** non-ivorien 0.080 Vijverberg, W.P.M. 1989 To analyze whether migrants are CILSS log hourly Self-selection yrs elem school 0.104 ** more productive workers than 1985-1986 wages model with a yrs junior high school 0.102 ** nonmigrants panel data choice of activity yrs senior high school -0.010 yrs university 0.223 ** diploma elememtary 0.402 ** junior high 0.608 ** beyond junior high 0.642 ** RRR 0.159 ** yrs occup. specific exp 0.106 ** yrs squared*100 -0.217 ** yrs general exp 0.075 ** yrs squared*100 -0.078 ** abidjan dummy 0.160 * other urban dummy 0.114 non-ivorien -0.033 female -0.056 43 Appendix E, Literature overview of studies of the labor market in Côte d'Ivoire Coefficient of key variables Vijverberg, W.P.M. 1989 To investigate wage determinants CILSS log hourly OLS yrs elem school 0.023 and Van der Gaag, J. using the Mincerian framework; with 1985 wage yrs junior high school 0.088 ** a special eye on the role of credentials yrs senior high school -0.032 yrs university 0.208 ** diploma elememtary 0.494 ** junior high 0.594 ** beyond junior high 0.536 * RRR 0.113 * yrs occup. specific exp 0.107 ** yrs squared*1000 -1.909 ** yrs general exp 0.026 ** yrs squared*1000 0.020 ** non-ivorien -0.117 female 0.011 44 Hazel M. Vargas L:\VERNER\Labor Markets\QuantileRPED-3.doc 01/28/00 5:44 PM 45