工资差距与生产力差距：以加纳为证

toPS .I/6b POLICY RESEARCH WORKING PAPER 2168 Wage and Productivity Gaps Ghana's labor market is segmented, and the workforce adds more to Firm- Evidence from Ghana level productivity than its cost would suggest. The more Dorte Verner training and education workers have, the higher their wages and the greater their productivity. In short, investments in human capital improve productivity. The World Bank Africa Technical Families Human Development 3 August 1999  ! POI KCY REeSFA\RCH W RKING PAPE{R 2168 Summary findings Vernier- uses a unique data set (combining information * rhe more training and education workers have, the about individual workers wit'h information about the higher their wages and the greater their productivity. ftrms eniploving them) to jointly estimate production Productivity differences can be demonstrated for htnctions and wage equations. This approach allows her five levels of education completed. The productivity gap not only to assess the marginal impact on wages of is grcater than the wage gap. demographic and other characteristics but also to * Returns to education are similar across gender, compare how these variables affect productivity among sectors, and level of unionization, but they are lower for various groups of workers. Among her findings: unskilled workers than for skilled workers. Female employees are paid less than male * Training supplied by outside providers (as opposed employees, but this negative wage premium does not to in-house training) is associated with higher wages but reflect commensurately lower productivity. appears to have no (immediate) impact on productivity. Employees' experience is reflected equally in wages * Trade union members' wages are in line with and in productivity differentials over the worker's life productivity. Both wages and productivity are higher for cycle. Wages and productivity both increase, but at a union members than for non-union members. decreasing rate. This paper - a product of Human Development 3, Africa Technica.l Families -is part of a larger effort in the region to understand how labor markets work in Africa. Copies of the paper a e available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact Hazel Vargas, room 18-138, telephone 202-473-7871, fax 202-522-2 119, Internet address hvargasCtworldbank.org. Policy Research Working Papers are also posted on the Web at http:// www.worldbank.org/html,/dec!Publications/Workpapers/home.html. The author may be contacted at dverner v worldbank.org. August 1999. (49 pages) I he Policy Research Working l'aper Series disseminiates the finidings of vork in progress to encourage the exchange of ideas about 1 developrnent issues. An objective of the series is toget thefindings out quickl, evez if the presentationsare less thanfully 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 oat nzecessarily represent the view of the World Bank, its Executive Directors, or the contries they represent. Produced by the Policy Research Dissemination Center  Wage and Productivity Gaps: Evidence from Ghana Dorte Vemer The author thanks Helena Ribe for invaluable encouragement, support, and comments on this work; Tylor Biggs for supplying the data; Bernardo Kugler for comments on an earlier draft of the paper; John Elder for useful discussions; and Niels-Hugo Blunch for research assistance.   1. Introduction This paper studies labor market outcomes in Ghana. The analysis focuses on the formal manufacturing wage sector and, more specifically, on the determinants of wages and productivity for various groups of workers. It tests hypotheses that relate to the impacts of individual and enterprise characteristics on wages. Furthermore, it compares the marginal impact of each of these characteristics on wages with their respective impact on labor productivity. The results may indicate whether, for example, there exists a spot labor market, discrimination, and/or structural differences among sectors and groups of workers. The paper analyzes whether experience, training, and education impact wages and productivity. In recent years, analysts have paid a lot of attention to the impacts of education and labor force training. The rationale for investing in human capital is that a more skilled and educated labor force is more productive than a less educated one. Therefore, policymakers emphasize investment in human capital because they believe that, in general, it increases labor productivity. However, there is not have much evidence of this relationship in the Africa region.' This paper aims partially at filling this void by presenting evidence on the direct impact of education, training, and experience on productivity for different groups of workers using econometric regression analyses. It looks at whether Ghanaian labor markets are characterized by gender discrimination. It analyzes whether the labor markets 1 Glewwe (1996) finds that there is no return to human capital in Ghana. 3  are competitive. And it looks at whether union membership, manufacturing sector, and firm location affect labor market outcomes. I use data from the 1994 Regional Prograrn on Enterprise Development (RPED) survey on Ghana, which covers 215 manufacturing enterprises ranging from micro enterprises to very large enterprises. The questionnaire used in the data collection process has two related parts: one for management and one for workers. In addition, about 10 randomly selected workers from different occupational categories were interviewed in each enterprise. The RPED dataset matches employee and employer data and contains detailed information on both. The two types of information are merged from individual data on skills, education, and age, and from firm-level information about sector, production, sales, and employment. Empirical studies of the determinants of wages and earnings inequality have focused primarily on factors affecting labor suppl,y. Long-run labor supply factors include variables such as education, age, gender, and experience.2 Few studies have used variables controlled by the employers, the so-called demand factors. One reason is that in most industrial and developing countries more infbrmation has been collected on workers than on their employers. The lack of data on worker productivity constrains empirical research related to issues of wage determination. For example, without direct measures of productivity, 2 Groshen (1991la) mentions that education, age, occupation, ethnicity, gender, and union variables account for only 51 percent of the variation in the log of wages, analyzing (US) data from the Current Population Survey One Quarter Earnings Sample, 1986. Introducing demand-related factors is likely to explain part of the 49 percent of the variation in wages that workers' characteristics cannot explain. 4  discrimination by gender or ethnicity cannot be established correctly. Many studies of wage determination report positive coefficient estimates on the age of an employee, conditional on a variety of covariates. These estimates neither imply that older workers are more productive than younger ones, nor that wages rise faster with productivity because no bridge has been made between productivity and wages (see Hellerstein, Neumark, and Troshe 1996). These problems may be overcome by estimating the wage and productivity equations jointly and, thus, comparing wages and productivity for various groups of workers. Section two describes the methodology and data used in this study. Section three outlines the wage determination model used. Section four shows descriptive statistics and presents regression results. Section five presents conclusions. 2. Data and Methodology I carry out this study with the so-called RPED dataset from the 1994 survey in Ghana. The survey included 215 firms and interviewed about 1,20C' of their employees. This matched employee-employer dataset has many advantages compared with datasets of just employees. An employee dataset may contain information about the sector in which a worker is employed but little information about the firm. An employer dataset has information about the firms but limited information about the individuals actually employed in the firms, apart from aggregate wage costs and, in some cases, information 5  about aggregate education and training costs. The employer-employee dataset allows detailed analyses of hypotheses related to both firms and individuals. Unfortunately, many of the variables in the 1994 round of the Ghana RPED dataset have very few observations. This is the case, for example, for the share of exports in output, expenditures on research and development, and foreign licenses. Hence, these variables are excluded, so that the regression analysis includes relatively few enterprise- level controls. I use the so-called general-to-specific econometric methodology in the regression analysis. I apply this methodology by formulating a general model and then reducing it to a parsimonious specification by eliminating statistically insignificant variables one by one, while focusing mainly on the determination of wages. The more parsimonious specification includes only statistically significant variables in the wage equation. The general wage model contains explanatory variables in levels and some are also included in quadratic form, hence allowing for nonlinearities in the data. The natural log wage equation or the production function may be quadratic in variables such as experience. This way of modeling may capture that the return to experience is not constant but rather decreasing over the life cycle. Additionally, I introduce dunmmy variables that take the value of one if, for example, the worker is a member of a trade union and zero otherwise. Inclusion of this variable may reveal whether there is a wage or productivity premium related to union membership. I use simultaneous single-level statistical models. These models may cause aggregation biases equivalent to problems produced in the learning achievement literature. In this literature, workers are the units of observation and firms are included in 6  the individual vector of variables. It is worth noting that, in general, this methodology may cause aggregation biases in two ways. First, it may overestimate the group effect, that is, the firm effect, on productivity. Second, it may underestimate the individual effect on wages and productivity. Multi-level estimation takes aggregation biases into account. The firm-level effect impacts average production and the individual slope of, for example, gender. And the gender effect may be different in different firms. Furthermore, the group level effect may not be the same for every single worker. However, the hierarchical linear modeling software does not allow for missing observations. Additionally, small firms have too few employees to actually perform the multi-level analysis. Due to these data limitations, I report only the results obtained by estimating single-level models. 3. Modeling Wages Competing models of wage determination depend on the connections among wages, productivity, and employee-employer characteristics. Standard wage determination analyses (single-equation wage models) consider employee characteristics in the process of determining wages, but make no link to productivity. Standard analyses rarely consider employer characteristics. I analyze wages, productivity, and employee-employer characteristics simultaneously. Without direct measures of the relative productivities of employees, discrimination by gender, for example, cannot be based only on statistically 7  significant estimated coefficients on the respective dummy variables in the individual- level wage regression. Standard single-equation wage regressions report positive coefficients on age conditional on a variety of covariates. These regressions do not imply that wages increase faster than productivity or that older employees are more productive than younger ones. Therefore, it is important to analyze productivity and wage determination simultaneously. The regression analyses describe monthly wages and productivity in the formal manufacturing sector, conditional on individual and firm characteristics. The following equation explains the wage and productivity: lny=IpI+ZF6. The dependent variable (y) is a vector containing two variables: wages (w) and productivity (v), thus enabling a richer analysis than when applying the standard single- equation wage determination model. The vector of explanatory variables in the analyses consists of firm/employer characteristics (F), the so-called demand factors, and employee characteristics (I), the supply factors. When y is reduced to scalar w, the traditional wage determination model appears. ,3 and o are vectors of parameters revealing the marginal impacts of the explanatory variables on wages and productivity. The production function is estimated in a value-added form for two reasons. First, the input variables, such as materials, may be endogenous. By applying this specification, the analysis does not need to estimate the coefficients on materials. Second, this specification embeds contrasting production function specifications. For example, in 8  one specification, the elasticity of substitution is zero, so that: materials have to be used in fixed proportions. In another specification, the elasticity of substitution is infinite (see Grilliches and Ringstad 1971). There are several reasons for including various worker characteristics. A trained and educated workforce provides flexibility in adapting to changes in technology or other economic changes that a firm and an economy may face. Experience and years of schooling are widely used in analyses of wage determination and inequality (see Welch 1969 and Mincer 1974).3 The wage bargaining literature has emphasized institutions such as trade unions as an important factor in the process of determining wages. Gender and ethnicity variables reveal information on the female-male wage gap and wage differentials associated with different nationalities. If discrimaination exists, it may be indicated by these variables having significant parameters. Including the sector of employment enables testing competitive economic theories, some of which predict that the sector of employment has no impact on earnings differences for similar workers. And the location of an enterprise may be important because, for example, life in cities is generally more expensive than in towns. Workers may be compensated for the higher cost of living. Section three discusses each group in more detail. I estimate the wage equation and the value-added equation jointly by regressing them on the same independent variables. The estimated individual coefficients and standard errors produced by multivariate regression are identical to those that would be produced by estimating each equation separately. The difference is that, because 'See Levey and Murnane (1992). 9  multivariate regression is a joint estimator, it also estimates the between-equation covariances. Therefore, coefficients can be tested across equations to reveal whether costs are in line with productivity. The analysis assumes that employees with different characteristics are perfect substitutes in the production process, but with potentially different productivities. The individual-level wage and production equations are estimated simultaneously, allowing for comparison of the average productivities and wages of employees distinguished by various characteristics. This empirical setting may reveal new information on selected issues related to the determination of wages in Ghana. For example, it may provide evidence about gender discrimination in wages and the causes of increasing remuneration over the life cycle. When equality of the parameter estimates of a particular variable in the two equations cannot be rejected statistically, I interpret this result as evidence compatible with the existence of competitive spot labor markets. Rejection of equality of the parameter estimates points toward noncompetitive labor markets or discrimination. The analysis allows productivity to vary by type of employee and by type of enterprise by adding controls for enterprise characteristics, such as sector and region.4 4 There are numerous studies of the Ghanaian labor market. Canagarajah and Thomas (1997) analyze returns to education. Jones (1994) studies wages and productivity of workers in the manufacturing sector. Teal (1995) investigates whether the decline in real wages reflects the existence of a competitive labor market in Ghana. Velenchik (1995) focuses on apprenticeships in the manufacturing sector. Teal (1996) analyzes the possible existence of economic rents within the manufacturing sector. 10  4. Determinants of Wages and Productivity Appendix table A- I lists the variables and describes the construction of each variable. Appendix tables B- I through B-4 present the results of joint estimation of the wage equations and production functions, controlling for both firm and employee characteristics. The variables that cannot be rejected as statistically insignificant are eliminated one by one from the general model and the final selected wage model is presented in the tables in appendix B. 4.1 Differentials by Formal Education Human capital has proven important in enhancing long-term economic growth.5 A more educated workforce is likely to increase worker productivity, show innovative behavior, and facilitate the adoption and use of new technologies. The increasing speed of technological change that firms face today and increasing initernational economic integration call for workers to have higher skill levels in order for firms to be competitive. One reason is that more skilled employees can adjust more easily to changes in the economic and technological environment than less skilled workers. Knowledge about economic returns to human capital gives insights about the extent to which it is worth undertaking this particular investment compared with other types of investment. Therefore, it is of interest to estimate the impact on money wages of different kinds of education, training, and other experience. This analysis may indicate 11  areas where scarcity in training and education may exist because the existence of a wage differential due to training may be interpreted as rent (to skilled labor). Education in Ghana. The system of education in Ghana consists of up to 17 years of pre-university education: six years of primary school, followed by four years of middle school and seven years of secondary school. After pre-university education, higher education includes professional. polytechnic, or university education.6 Table 1. Education by Gender and Resiidence in Ghana (percent) Gender Residence Educational level Female Male Accra Other completed None 15.1 8.0 5.4 15.9 Primary school 1.6 3.8 2.9 4.3 Middle school 31.4 48.9 46.5 48.8 Secondary school 18.4 12.4 14.7 9.2 Vocational school 23.8 11.4 13.6 15.2 Professional and technical 8.7 14.3 15.2 6.1 schools University 1.1 1.3 1.8 0.6 Total Percent 100 100 100 100 Note: The values show the percentage of each category of workers with each level of education. For example, 15.1 percent of the females in the sample have not completed any level of education. Source: Round Three of the 1994 Regional Prograrn on Enterprise Development for Ghana. 5 See Barro (1991) and Mankiw, Romer, and Weil (1992). 6In 1987, the education sector was reformed and pre-tertiary education was reduced to a total of 12 years. 12  The distribution of education from the RPED data across gender and geographical regions reveals that in the formal manufacturing sector, the majority of workers have completed middle or secondary education (see tables 1 and 2). The data contain three notable features. First, middle school is the most conmmon level of completed education (46 percent). Second, almost twice as many employed females as males have not completed any education.7 Third, fewer of the workers employed in Accra (9 percent) have completed less than secondary education, compared with workers employed elsewhere in Ghana. Table 2 presents the percentage of workers in each sector that has completed each level of education. Most notably, the share of workers with no or primary education is quite small in the metal sector (6 percent) compared with the average share across all across sectors (13 percent). University graduates make up the largest share in the food sector (3 percent). Table 2. Education by Manufacturing Sector (percent) lvlanufacturing Sector Educational level completed Total Manufacturing Metal Textile Food Wood None 9.3 3.7 1]0.0 16.6 7.8 Primary school 3.3 2.0 3.8 3.9 3.5 Middle school 45.8 50.8 46.7 35.5 58.3 Secondary school 13.5 12.6 4L3.8 15.4 8.7 Vocational school 13.6 15.0 40.7 13.5 13.0 Professional and technical schools 13.3 15.6 13.9 12.0 7.9 University 1.3 0.3 [.0 3.1 0.9 Total Percent 100 100 [00 100 100 7Note that only 17.9 percent of the sampled workers are women and less than 0.5 percent are non-Africans. 13  Note: The values show the percentage of each category of workers with each level of education. For example, 9.3 percent of all manufacturing workers in the sample have not completed any level of education. Source: Round Three of the 1994 Regional Program on Enterprise Development for Ghana. Impact on wages. The estimated coefficients on the education variables in the wage equation reveal the impact of human capital obtained from education on wages controlling for other individual and firm characteristics (see appendix table B-1).8 The impact on wages of each of the six categories of completed education (primary, middle, secondary, vocational, technical, and university) is statistically significantly different from zero (no education completed being the category of reference). This result indicates that the more education a worker has completed, the higher wage the worker receives, conditional on a variety of individual and enterprise covariates, including occupation. The estimates show that the size of the wage premium to education increases rapidly with completed level of education. The wage gap is 51 percent for workers wh1o completed primary education, compared with those who did not complete any level of education. An employee who has completed middle school obtains a wage premiurn of 41 percent. The wage gap is 56 percent for secondary graduates and 186 percent for university graduates. These results are in line with those of Canagarajah and Thomas (1997), who find that workers in Ghana who have completed tertiary education earned 2.7 times more than illiterate workers in 1991. 8 The results reported in the text are exponential transformed estimates reported in appendix B 14  Hence, formal education has an impact on manufacturing wages, but it is not monotonically increasing. In fact, the analysis cannot reject the hypothesis that the estimated coefficients for primary, middle, and secondary education are of equal size (F(pri=sec)=0.04, Prob>F=0.83; F(sec=mid)=1.54, Prob>F=10.22; F(mid=pri)=0.16, Prob>F=0.69; and F(sec=voc)=2.84, Prob>F=0.09). Wage analyses of industrial countries usually find that the incremental returns to education fall as the level of schooling rises. However, the above results indicate that the returns do not fial in Ghana for the formal manufacturing sector. An employee who has completed technical or vocational education also receives a statistically significant premium in addition to that already captured by occupational controls and experience. The wage premiums are 48 percent and 20 percent, respectively (see appendix table B-1). These results indicate that education has spillover effects on productivity, meaning that education's aggregate contribution to output may be larger than its costs. The endogenous growth literature has emphasized these spillover effects or positive externalities that result from increasing returns to scale in the production function due to educational or learning externalities (see Lucas 1988). The spillover effects of education may result from entrepreneurial ability to speed up the adaptation of technology throughout the market. Impact on productivity. The regression results show that education enhances firm-level productivity in Ghana. The country's manufacturing sector has a positive and marginally significant production differential by completed education above primary level. Primary education does not contribute to productivity because it does not have a 15  statistically significant impact on value added. ThLe quality of primary education may be low or not sufficiently high so as to benefit value added in firms. This result may indicate that there is a private return to obtaining a diploma, which serves as a screening device when hiring labor. Workers who have completed rniddle school education contribute on average 3 72 percent more to value added than workers with no completed schooling. University and secondary school graduates contribute 660 percent and 287 percent, respectively. Furthermore, technically trained workers and employees who have completed vocational training provide 185 percent and 127 percent, respectively, more to value added than do workers with no completed education (see appendix table B-1). Wages and productivity. The wage differentials by education fall behind productivity differentials by education and significantly so for completed middle and secondary education. The test rejects the hypothesis of equality of the wage and productivity differentials. For completed middle and secondary school, the test rejects the hypothesis that the workers are paid according to their productivity in both cases (F= 14.2, Prob > F=0.00; F=6.24, Prob > F=0.0 1). Hence, the positive wage gap between workers with completed secondary education and no completed education can be completely justified by higher productivity. Furthermore, workers with completed secondary education could demand higher wages so as to match productivity. This result also holds for completed middle school education. Several reasons may explain why the workers do not demand higher wages. They may not be aware of the magnitude of the impact completed education has on productivity. Workers may not be aware of the extent and importance of knowledge 16  spillovers in the manufacturing sector. In addition, in Ghana, educated people have been loosing jobs in the last five to ten years. Thus, demanding higher wages may lead to more job insecurity. The findings may not indicate a shortage cf educated employees or a lack of demand because these results would be out of line with the standard assumption that wages reflect the productivity of labor. Another explanation may be that knowledge spillover effects, which are not instantaneously reflected in pay, indicate that social returns in a firm are higher than private returns. The cost of a university graduate seems to be in line with firms' benefits because the results do not reveal any discrepancy between productivity and returns to education. The test for equal coefficients in the wage equation and production function cannot be rejected at conventional levels of significance (F=1.39, Pirob > F=0.34). However, the small number of university graduates in the sample may affect the results. Another explanation may be that this group of workers obtains higher enumerations than revealed by wages. Also, earnings may include fringe benefits thali often are large in the manufacturing sectors in Africa. For both completed vocational and technical education, wages are not in line with productivity (F=4.48, Prob > F=0.04; and F=4.72, Prob > F=0.03). Productivity exceeds wages in both cases, implying that workers with completed vocational and technical education are under-compensated as compared with their productivity. When substituting the individual school attaimneint variables with a continuous variable for completed education, the other explanatory variables' impacts on wages are unchanged. There exists a significantly positive return to holding a permanent contract, but the impact on value added is insignificantly different from zero. The completed 17  education variable is statistically significantly different from zero. The education premium obtained from completing an additional level of education is 15 percent, conditional on a variety of individual and enterprise covariates (see appendix table B-2). The effect on productivity is statistically significant and the size of the impact is 54 percent. Education has a less pronounced influence on wages than it has on productivity, except for completed primary education. This result is illustrated by the significance and larger coefficient estimate of formal education in the productivity equation than in the wage equation. Therefore, researchers should be carefiul in interpreting the traditional earnings function approach as a means of estimating the effects of, for example, education on productivity. Policymakers often use results from earnings function estimation to make decisions about public expenditures. The findings here indicate that wages are not always a good proxy for the physical product. They suggest that enterprises in Ghana do not always "get prices right," ihat is, wages do not equal productivity. 4.2 Differentials by Occupation and Skill Table 3 shows some differences in the occupations of employees due to gender. More women than men work in administrative activities and more men than women hold managerial positions. The table also shows that more than half of both male and female 18  employees work in production. Only professionals and commercial workers each account for less than 10 percent of the total sample. Table 3. Gender and Occupation (percent) Full Occupation sample Female Male Administration 11.8 25.4 8.8 Commercial 6.7 7.6 6.5 Manager 10.0 5.4 11.0 Production 54.9 54.1 55.1 Professional 3.5 1.6 3.9 Support 13.2 6.0 14.7 Total Percent 100 100 100 Note: The values show the percentage of each category of workers with each level of education. For example, 25.4 percent of all female manufacturing workers in the sample work in administration. Source: Round Three of the 1994 Regional Program on Enterprise D)evelopment for Ghana. Table 3 divides the employees into six groups by occupation: manager (encompassing managers, supervisors, and foremen); administration (administrative workers and clerks); commercial (commercial workers, salespeople, maintenance workers, and technicians); support (support staff); production (production workers); and professional (engineers, accountants, and others). The regression analysis reveals that only the impact of managers is statistically significant anid positive in the wage equation, controlling for other individual and firm characteristics. The reference group for the analysis is workers in maintenance, skilled production workers, other production workers, support staff, and trainees. The calculated occupational wage premium is 33 percent for managers and -40 percent for support staff (see appendix table B-1). Surprisingly, the estimated productivity of managers is not statistically significantly different from that of 19  the reference group. Productivity differentials by occupation do not fall behind wage differentials, as revealed by the rejection of all tests oi equality between the two estimates (F(jobsupp)=2.67, Prob > F=0.1O and F(jobman)=O.01', Prob > F=0.93). I use the occupational data to test for the possibility of a gap in returns between skilled and unskilled labor in Ghana. Skilled workers are those in the manager, administration, and professional, occupations. Unskilled workers are those in the commercial, production, and support occupations. The skilled employees may receive higher returns to experience, sector, training, education, unionization, occupation, and location than unskilled employees. The results are based on the inclusion of interaction variables in the analyses. A dummy variable that takes the value one if the employee is unskilled is interacted with the variables presented in appendix table B-1. The findings indicate that the returns to education and, in particular, to having completed middle and secondary education, are lower for unskilled than skilled workers (see appendix table B-4). The comparison group is workers who have not completed any level of education. The unskilled workers who have completed middle school earn returns 23 percentage points lower than skilled workers do(see appendix table B-4). The productivity analysis reveals that unskilled workers are not statistically less productive than skilled workers. The economic return to completed secondary education is 24 percentage points lower for unskilled than for skilled workers. 4.3 Differentials by Trade Unions 20  Twenty-nine percent of the employees in the sample belong to trade unions. The data show that female workers are not less unionized than male workers. In contrast, the data show an education-unionization gap; workers who have completed university education and those with just primary education fall below the average level of unionization (see table 4). Table 4. Union Membership and Education (percent) Total Educational level completed Union Non-union Percent Number Full sample Percent 28.7 71.3 100 None 23.2 76.8 100 Primary school 14.7 85.3 100 Middle school 27.4 72.6 100 Secondary school 28.1 71.9 100 Vocational school 37.9 62.1 100 Professional and technical schools 34.3 65.7 100 University 0.0 100.0 100 Note: The values show the percentage of each category of workers vvith each level of education. For example, 14.7 percent of the manufacturing workers in the sample who have completed primary school belong to unions. Source: Round Three of the 1994 Regional Program on Enterprise Development for Ghana. Table 5. Union Membership and Occupation (percent) Total Occupation Union Non-union Percent Number Full sample Percent 28.7 71.3 100 Administration 40.2 59.8 100 Commercial 33.3 66.7 100 Manager 26.2 73.8 100 Production 24.1 75.9 100 Professional 16.7 83.3 100 Support 40.4 59.6 100 21  Note: The values show the percentage of each category of workers with each level of education. For example, 40.2 percent of the manufacturing workers in the sample who work in administration belong to unions. Source: Round Three of the 1994 Regional Program on Enterprise Development for Ghana. Workers in administration and support occupations have the highest share of workers in unions; workers in manager and professional occupations have the lowest share (see table 5). It appears that unionization makes a difference in wages because the union premium in the Ghanaian manufacturing sector is statistically significant and positive (see appendix table B-1). Union members earn more than nonunion members do, when controlling for firm and individual characteristics in the wage equation. In addition, union members have statistically significantly higher productivity than nonmembers do. The test of the hypothesis that the wage and productivity gaps are of equal size is highly rejected (F=68.44, Prob > F=O.00). The estimated wage differential associated with being a union member is approximately 16 percent (see appendix table B-1). This outcome is in line with results found for industrial countries, which usually reveal that union members earn more than nonunion members. These results may provide evidence that trade unions do not reduce wage differentials and, therefore, that unions negatively affect the distribution of income. The finding for Ghana may be surprising because other African studies of the impact of unions on wages report a negative association. Both results for CFA countries applying a cross-section dataset of African countries (Rama 1997) and results from wage analyses in Zimbabwe (Verner 1998) show that trade unions impact wages negatively. The presence of trade unions is generally associated with more firm-provided training. Trade union bargaining at the firm level may increase social welfare by counterbalancing 22  a firn's monopsonistic power in wage determination (see, for example, Booth and Chatterji 1997).9 4.4 Differentials by Employer-Led Training In industrial countries, governments have increasingly ermphasized the importance of employer-led training. The benefit of training should be in the form of higher output. Training provides workers with the skills necessary for improving competitiveness, adaptability, and growth. Furthermore, skill acquisition may reduce wage inequalities. In addition to informal on-the-job training by supervisors and co-workers, employers may supply formal training, either in-house or by outside providers. The incidence of training in the Ghanaian manufacturing seci:or is relatively high compared with industrial and semi-industrial countries. Around 64 percent of the employees receive formal structured training. In the sample, 28 percent of the workers received in-house training (see table 6). This value is in line with the data for Taiwan, where 38 percent of large firms and 4 percent of small firms train employees in-house (see Tan and Batra 1995). Furthermore, 36 percent of the Ghanaian employees received training outside the firn. 'Booth and Chatterji (1997) show that long-term contracts are socially optimal when workers are trained because the training reduces workers' incentives to quit and leave with the skills obtained. Alternatively, wage bargaining by firms and local unions may ensure that the post-training wage is set sufficiently high to defer inefficient quits, and thus to ensure that the number of trainees the firm takes on is near the socially optimal number. Hence, it would be expected that the results here would show higher post- training wages. 23  Table 6. Gender, Residence, and Training (percent) Gender Residence Training Full sample Female Male Accra Other In-house 28.0 33.0 27.0 25.1 34.2 By outside providers 36.2 24.7 38.7 38.0 35.4 None 35.8 42.3 34.3 36.9 30.4 Total Percent 100 100 100 100 100 Note: The values show the percentage of each category of workers with each level of education. For example, 28.0 percent of the manufacturing workers in the sample received in-house training. Source: Round Three of the 1994 Regional Program on Enterprise Development for Ghana. The incidence of training is only marginally h:igher for firms in Accra, compared with firms outside the capital (see table 6). Fewer fem,ale employees receive training outside the firm and more receive training inside the firm than male workers do. Forty- two percent of the female workers are not trained, conapared with only 34 percent of the male workers. Investigating sector differences in the incidence of training reveals that the wood sector trains relatively fewer workers than the other sectors. Only 24 percent of all employees do not receive any training, while 32-37 percent of workers in the wood sector do not receive training . More support staff receive training than workers in other occupations do. Table 7. Occupation and Training (percent) Administr Comme Professi Product Support Manager Full Training ation rcial onal ion sample In-house 11.6 14.7 8.6 35.7 29.6 20.4 28.0 24  By outside providers 47.9 29.4 45.7 30.5 44.4 42.7 36.2 None 40.5 55.9 45.7 33.8 26.0 36.9 35.8 Total Percent 100 100 100 100 100 100 100 Note: The values show the percentage of each category of workers with each level of education. For example, 11.6 percent of the manufacturing workers in the sample who work in administration received in- house training. Source: Round Three of the 1994 Regional Program on Enterprise Development for Ghana. Only 26 percent of support staff do not receive any training, compared with the average of 35.8 percent across all occupations (see table 7). Employees in commercial and sales occupations receive the least training; 55.9 percent of this group does not receive any kind of training (table 7). University graduates receive the most training, 76.9 percent (see table 8). 25  Table 8. Education and Training (percent) Educational level completed Training None Primary Middle Second Vocati Professio Universit Full school school ary onal nal and y sample school school technical schools In-house 30.9 23.5 24.8 37.9 13.4 14.7 0 28.0 By outside 35.1 41.2 30.7 32.7 40.3 44.9 76.9 36.2 providers None 24.0 35.3 44.5 29.4 46.3 40.4 23.1 35.8 Total Percent 100 100 100 100 100 100 100 100 Note: The values show the percentage of each category of workers with each level of education. For example, 30.9 percent of the manufacturing workers in the sample with no completed level of education received in-house training. Source: Round Three of the 1994 Regional Program on Enterprise Development for Ghana. In-house training. The hypothesis that training has no impact on wages is not rejected for training within the enterprise. This result indicates that in-house training does not directly contribute to measurable wages. In addition, formal training obtained within the firm is not significantly more productive than the training workers obtain from colleagues in an informal way. The results hint that in-house training is not productive; however, it seems more likely that the impact is not instantaneously revealed in wages and productivity. Initially, there is no measurable effect of in-house training in the data, but effects may be measurable at a later date. Training by outside providers. The analysis rejects the hypothesis that training supplied by outside providers has no impact on wages. Workers who have undergone formal, structured training outside the firm earn 16 percent more than workers who have not been trained (see appendix table B-i). However, the regression results indicate that training is not associated with higher firm-level productivity. In terms of the level of 26  parameter estimates and in a statistical sense, the productivity-enhancing effects of training are not important for training obtained outside or inside the firm. The productivity effects of skilled worker training are estimated to be not significantly different from zero. This result is surprising because studies of other low- income countries have found a statistically significant impact. For example, Tan and Batra (1996) find significant coefficients of 1.43 for Indonesia and 0.39 for Columbia. Verner (1998) finds a statistically significant coefficient estimate of 0.33 for Zimbabwe. I find that training by outside providers is not associated with increased productivity in Ghana, possibly because the impact of training is not instantaneously reflected in productivity. Underinvestment in training. From the regression analysis, it appears that Ghana underinvests in training. Evidence of underinvestment is suggested by significant returns to outside training, reflecting its relative scarcity. Training activities generate skills and knowledge that are employee specific in the sense that the employee may quit and take the accumulated human capital away from the firm. This possibility makes trained employees more valuable to the firm than other workers, including workers from other firms. Acemoglu and Pischle (1997) show what happens when the current employer has superior information about the worker's ability relative to other firms. This information advantage gives the employer an ex post monopsony power over the worker, which encourages the firm to provide training. Acemogil and Pischle's model can lead to multiple equilibria. In one equilibrium, quits are endogenously high; as a result, 27  employers have limited monopsony power and are willing to supply little training. In another equilibrium, quits are low and training is high. It wvould seem to be valid to apply this model in the African context, where job mobility and the number of quits are low in the formal sector. Firms in Ghana provide little or no training for several reasons. The country has imperfect capital markets, limited access to information, and other market failures. The results for employer-led training seem to imply that policies that encourage increased enterprise training will lead to larger productivity gains for the economy as a whole. However, the results here show that in-house training does not pay off in terms of higher wages. It may be that the effect is not captured by the available data. The failure of training to increase productivity would cause a disincentive to invest in training. The policy implication is presumably that Ghana should promote training by outside providers. 4.5 Differentials by Experience Labor market experience is not easy to measure. However, the labor market literature has used the age of an employee as a proxy. The worker's age is included in the wage equation both in levels and squared to allow for possible nonlinearities. Both age variables are statistically significantly different from zero and have the expected signs (see appendix table B-1). Experience affects wages positively and the relationship is nonlinear, controlling for other individual and enterprise characteristics. More 28  experienced workers have higher levels of labor productivity; the productivity curve is steeper at the early stage of a worker's life than later. Put differently, the estimated life cycle wage profile in Ghana has the usual quadratic shape known also from industrial countries. The positive effect of experience on wages is increasing for younger workers and continues for older workers at a decreasing rate. The productivity profile in the Ghanaian manufacturing sector follows the pattern of wages; it increases at a decreasing rate over the life cycle. However, the wage gap and the productivity gap are not equal in size (F=4.41, Prob > F=0.04; F=1.42; Prob > F=0.22). These results indicate that the increase in productivity obtained from a more experienced workforce is larger than the costs. 4.6 Differentials by Apprenticeship Apprenticeships are widespread in labor markets in Sub-Saharan Africa (see Velenchik 1995). In Ghana, 10.2 percent of the sampled workers have been apprentices in the firm in which they currently work. Moreover, the share of apprentices does not seem to differ significantly across gender. The trainees often pay for their training in Ghana. The economic rationale is that an apprenticeship raises future productivity and thereby increases future earnings by more than the current fees. The results of the empirical analysis show that employees who have been apprentices do not obtain higher wages than others do. T'he analysis detects no positive return to apprenticeships, controlling for formal education and other individual and firm 29  characteristics (see appendix table B-1). The point estimate of the wage differential is - 0.31 and the wage gap is significantly different from zero, controlling for individual and firm characteristics. For the value-added specificatiorn, the estimate of the productivity differential is negative but statistically insignificant. HIence, the negative wage premium is out of line with productivity. 4.7 Differentials by Gender Empirical research on gender pay gaps has traditionally focused on the role of gender- specific factors, particularly gender differences in quailifications and differences in the treatment of otherwise equally qualified male and female workers (that is, labor market discrimination)." This section explores the determinanits of the gender pay gap and argues ' Modem explanations of the gender wage gap rest on two pillars: the human capital interpretation and models of labor market discrimination. The human capital explanation was developed by Mincer and Polachek (1974) and others. This idea is based on productivity differences between men and women and, therefore, differences occur in economic outcomes in the form of pay and occupation. The human capital model builds on the traditional division of labor in the family. According to this theory, women have fewer incentives to invest in human capital than men because women anticipate fewer work weeks over their life span than men. The result is that women have lower human capital investment than men do and consequently wornen have lower wages. The same arguments are thought to cause gender differences in occupation, as females choose occupations where human capital investment is less relevant and penalties for spells of labor force interruptions are smaller. Models of labor market discrimination builds on work originating from studies on ethnic discrimination. For example, Becker (1957) conceptualized discrimination as taste or personal prejudice against a specific subset of a larger group. A related type of model is the so-called overcrowding model (see Bergmann 1974). It builds on exclusion of a group, for example, women, from "male" occupations and, therefore, leads to excess supply of workers in "women's" occupations. Despite equally productive workers, the excess supply suppresses wages in "women's" occupations. 30  for the importance of an additional factor in the analysis, namely productivity. Furthermore, it takes into account the possible difference in prices set for labor market skills and the rewards received for employment in favored sectors. In empirical research on the evidence of discriminiation, traditional single- equation models rely on the existence of a residual gender pay gap that cannot be explained by gender differences in measured qualifications. This is seen to accord well with the definition of labor market discrimination, that is, wage gaps between groups that are not explained by productivity differences, see Blau 1996. Blau notes that pay differences between groups may also reflect group differences in unmeasured qualifications or compensating differentials. If men are more highly endowed with respect to these omitted variables, then discrimination would be overestimated. However, if some of the controls (for example, occupation) themselves reflect the impact of discrimination, labor market discrimination will be underestimated. The analysis here reduces these caveats by comparing estimated differences in wages and productivity and controlling for the same measure of qualifications in both equations. Under perfect competition in the capital and labor markets, equivalent employees at equivalent jobs are compensated equally, that is, there is no discrimination. The analysis concludes that labor force discrimination exists in Ghana only if both the following hypotheses are rejected: (1) the impact on wages and productivity of being a woman is zero; and (2) the estimated impacts on wages and productivity are of equal size. Female workers comprise about 17 percent of the interviewed workers in the sample. The regression analyses reveal that, on average, females are paid statistically 31  significantly less than male workers, controlling for differences in qualifications and individual and other characteristics. The wage gap between men and women is notable at a general level in the Ghanaian manufacturing sector. Women are paid 17 percent less than men (see appendix table B-1). It seems dubious that this result occurs because women are less productive than men, taking into account segregation by gender across labor market sectors and the level of qualifications such as education and experience. The analysis controls for sector of occupation, so that it takces into account the possibility of larger rents received by workers in favored sectors, imiplying that it would be harder to detect possible discrimination. The point estimate of the productivity differential is found to be statistically different from zero and negative. Female workers are 37 percent less productive than male workers, accounting for qualifications, occupation, and sector of occupation. Does that mean that there is gender discrimination in the Ghanaian manufacturing sector? Both the wage and productivity gaps are significant and the former is lower than the latter. The analysis tests whether the coefficients are statistically the same in the two equations and rejects the hypothesis of equality (F=1.08, Prob > F=0.30). This result is not consistent with parts of the wage gap being attributable to discrirnination against women. In fact, the results indicate that women receive more compensation in terms of wages than their male colleagues. Further analysis shows that the lower wages that Ghanaian women earn do not seem to originate from being employed in the textile sector. The results show that women working in textiles do not earn less than their male colleagues, controlling for the textile sector and other covariates. 32  The analysis looks at whether the wage structure is important in the determination of wages for men and women in Ghana. I test the hypotheses that some sectors have larger returns to skills, all else equal, leading to a larger gender gap; these hypotheses are all rejected. I also consider the possibility that women may receive lower returns to experience, sector, training, education, unionization, occupation, and location than their male colleagues. The outcome indicates that there are no gender differences in returns to the four levels of formal education or technical education, unionization, or occupation. The analysis yields statistically insignificant parameter estimates of interaction variables of the female dummy variable with each of the explanatory variables. However, the results indicate that women receive lower returns to experience than men (around 1 percentage point). The results hint that increased female experience impacts productivity less than increased male experience (see appendix table B-3). The results across sectors show that women are paid more than men in the wood sector (142 percent), despite not being significantly more productive. The analysis now considers the possibility of a gap in returns between male and female labor in Ghana. I analyze whether male employees receive higher returns to experience, sector, training, education, unionization, occupation, and location than do female employees. The education premium is neither larger for males than for females nor does there exist any particular sector premium related to education. The Analysis, including education variables for both males and females, shows that females do not obtain a significantly lower return than do men in any of the six levels of education. Surprisingly, the gender-wage differentials by level of education are not very strong and, 33  4.10 Differentials by Contract There are good economic reasons why, in many circumstances, it is in the interests of both the employer and the employee to form a long-nm employment relationship, thereby helping to build and retain firm-specific skills. The estimated wage differential associated with permanent employment in a firm is estimated to be statistically insignificant. Hence, there does not seem to be a trade-off between enhanced job security and wages. The contrary view that job security endows insider workers with more bargaining power that makes it possible to achieve higher wages does not seem to hold in Ghana. The result that greater job security seems not to lead to lower wages suggests that insider workers benefit from trade unions. The productivity differential associated with a permanent contract is positive but not statistically significantly different from zero. Moreover, temporary workers are not less productive than permanent workers, indicating that the former may be working hard in an attempt to obtain a permanent contract. 4.11 Differentials by Marriage or Ethnicity The marriage premium is not statistically significantly different from zero. This finding is surprising because results from wage analyses in industrial countries usually show that married workers earn higher wages than unmarried workers. This finding reflects that 36  married workers are more productive-whether because of selection or a true productivity effect-or that firms discriminate in favor of married workers. No good theoretical argument explains why married workers should be paid more. However, the Ghanaian manufacturing sector data show that married workers are more productive than unmarried workers, even though the difference is not revealed in earnings. I could not analyze whether differentials by ethnicity exist in manufacturing in Ghana because too few Europeans or other ethnic groups were interviewed (less than 0.5 percent of the interviewed workers are non-African). 5. Conclusion I used a unique dataset from Ghana that combines data on individual workers with data on workers' employers to estimate plant-level production functions and wage equations. I compared relative productivities and relative wages for various groups of workers and jointly estimated wages and output functions. This approach allows not only for assessing the marginal impact of demographic or other characteristics on wages, but also comparisons of the impact of these variables on productivity. The analysis found that wage differentials match productivity differentials for certain groups of workers, while for others they do not. Female employees are generally paid less than male employees and the negative wage premium females receive does not match the corresponding negative productivity premium. Females' returns to schooling and training are not different from those of males. 37  Employees' experience, as measured by age, is reflected equally in wages and productivity differentials over the life cycle. Both wages and productivity increase, but at a decreasing rate. However, the impact of experience on productivity is significantly higher than on wages. These results indicate that the increase in productivity obtained from a more experienced workforce is larger than the costs. However, other compensations may not be reflected in wages, and these may be distributed unequally across the life cycle, for example, allowances for children's education. In the formal manufacturing sector, the majority of workers have completed middle or secondary education. The size of the wage premium to education increases rapidly with completed level of education and with occupation. The size of the positive returns to education for primary, secondary, and micldle school are not statistically significantly different. Furthernore, the analysis reveals productivity differences for five levels of completed education: middle; secondary; vocational; technical; and university. Primary education does not contribute to increased productivity. University graduates are worth the high returns and, in fact, the productivity differential is in line with the wage premium. This is not the case for middle, secondary. technical, and vocational education, where the productivity gap is significantly larger than the wage gap. Hence, the positive wage gap between workers with completed middle, secondary, vocational, technical, or university education and workers with no completed primary education can be justified because the former groups are also more productive than the latter. For workers in management, the analysis establishes productivity and wage differentials that are in line with economic theory. The analyses show that returns to education are not different across gender, unionization, or sector. However, they are 38  lower for unskilled than skilled workers and the former are not less productive. Furthermore, returns to experience are lower for women than for men. The results suggest that the educational curriculum above the primary level in Ghana is providing workers with productive skills. The study suggests that wages are not always a good measure of productivity for public expenditure decisions. The incidence of training in the Ghanaian manufacturing sector is relatively high compared with industrial and semi-industrial countries. The incentive for investing in training is that it leads to higher output. The results reveal the existence of an asymmetry in training: training supplied by outside providers is associated with higher wages, while there does not seem to be any (instantaneous) effect on productivity. In-house training does not have a productivity enhancing effect and it is nol: instantaneously rewarded in the form of higher wages. Therefore, a scarcity of trained workers may exist to some degree. The results show that policies encouraging increased training and education will lead to larger productivity gains for the economy. Apprentices obtain a negative wage premium, but they are neither more nor less productive than their colleagues who have not been apprentices. Workers employed in an enterprise located in Accra are both more productive and paid higher wages than workers in other locations. Furthermore, the productivity differeni:ial is significantly larger than the wage differential. Employees in the wood sector are paid lower wages than employees in the textiles, metal, and food sectors. Employees in the wood sector are less productive than employees in other sectors. However, the impacts on wages and productivity are of equal magnitude. 39  Long-term contracts do not help firms to retain and build firm-specific skills. The analysis does not indicate that there is a trade-off between enhanced job security and wages or that job security endows insider bargaining power. The trend in manufacturing employment has been increasing slowly since 1982, which further strengthens this result. However, the productivity differential associated with a permanent contract is not different from zero, indicating that neither temporary nor permanent workers lack the incentive to work hard. Temporary workers are just as productive as permanent workers. And trade union members' wages are in line with productivity. Wages and productivity for union members are higher than for nonunion members. Altogether, the results indicate that the labor mnarket in Ghana is not perfectly competitive, but segmented. For example, wages are not equal for workers with similar skills employed in different sectors. Thus, the analysis finds structural differences across sectors. The analysis compared coefficients on education from the production function with coefficients from the wage function. It showed that fimns behave competitively and pay workers with completed university education according to their productivity. However, for all other levels of education, firms do not behave competitively; they pay either more or less than workers' productivity. The former category encompasses primary education and the latter middle and secondary education. The results obtained in this paper indicate that the workforce employed in the manufacturing sector in Ghana adds more to firm-level productivity than the costs it involves. Thus, wages are out of line with productivity, possibly due to externalities or spillover effects, for example, in skills. Hence, the le vel of wages is not too high to 40  compete internationally. Furthermore, the analysis provides evidence that increased human capital, through training or education, as it benefits both productivity and wages. Economic policies aiming at increasing productivity in Ghanaian enterprises and, hence, long-run economic growth, should emphasize the training and education of the employees. 41  References Acemoglu, D. and Pischke, J.-S. (1997) "Why Do Firms Train? Theory and Evidence," NBER Working Paper 5605, National Bureau of Economic Research, Cambridge, Mass. Barro, R. (1991) "Economic Growth in a Cross Section of Countries," Quarterly Journal of Economics 56 (2): 407-43. Blau, F. D. (1996) "Where are we in the Economics of Gender? The Gender Pay Gap," NBER Working Paper 5664, National Bureau of Economic Research, Cambridge, Mass. Booth, Alison L. and Monojit Chatterji (1997) "Training and Unions," CEPR Discussion Paper 1573. Canagarajah, Sudharshan and Saji Thomas (1997) "Ghana's Labor Market (1987-92)," Policy Research Working Paper 1752, World Bank, Washington, D.C. Glewwe, Paul (1996) "The Relevance of Standard l,stimates of Rates of Return to Schooling for Education Policy: A Critical Assiessment," Journal of Development Economics 51: 267-90. G-rilliches, Zvi and Vidar Ringstad (1971) "Economies of Scale in the Form of the Production Function," Amsterdam: North-Holland Publishing Company. Groshen, Erica L. (199la) "Five Reasons Why Wages Vary Among Employers," Industrial Relations 30 (3). Hellerstein, Judith K; David Neumark, and Kenneth R. Troshe (1996) "Wages, Productivity, and Worker Characteristics: Evidence From Plant-level Production Functions and Wage Equations," NBER Werking Paper 5626, National Bureau of Economic Research, Cambridge, Mass. Jones, Patricia (1994) "Are Manufacturing Workers Really Worth Their Pay?" Working Paper WPS/94-12, Centre for the Study of African Economies, Oxford University. Levey, Frank and Richard J. Murnane (1992) "U.S. Eamings Levels and Earnings Inequality: A Review of Recent Trends and Proposed Explanations," Journal of Economic Literature 30: 1333-81. Lucas, Robert (1988) "On the Mechanics of Economic Development," Journal of Monetary Economics 22: 3-42. 42  Mankiw, Gregory N; David Romer and David N. Weil (1992) "A Contribution to the Empirics of Economic Growth," Quarterly Journal of Economics 107 (2): 407-37. Mincer, Jacob (1974). "Schooling, Experience, and Earning,s," New York: Columbia University Press. Mincer, J. and Polachek, S. (1974) "Family Investment in Human Capital: Earnings of Women," Journal of Political Economy 82 (2): S76-S108. Rama, Martin (1997) "Are Labour Market Policies to Blame?" Working Paper, World Bank, Washington, D.C. Tan, Hong W. and Geeta Batra (1995) "Enterprise Training in Developing Countries: Incidence, Productivity Effects, and Policy Implicaitions," World Bank, Washington, D.C. Teal, Francis (1995) "Real Wages and the Demand for Labour in Ghana's Manufacturing Sector," Working Paper WPS/95-7, Centre for the Study of African Economies, Oxford University. Teal, Francis (1996) "The Size and Sources of Economic R'ents in a Developing Country Manufacturing Labour Market," The Economic Journal 106: 963-76. Velenchik, Ann D. (1995) "Apprenticeship Contracts, Small Enterprises, and Credit Markets in Ghana," The World Bank Economic Review 9 (3). Verner, D. (1998) "Are wages and productivity affected by human capital investment and international trade in Zimbabwe?" Policy Research Working Paper 2101, World Bank. Welch, Finis (1969) "Linear Synthesis of Skill Distribution," Journal of Human Resources 4 (3): 311-27. 43  Appendix A Table A-1. Variable Definitions Variable Definition Accra I if individual is working in a firm in Accra Africa I if employee is African and zero otherwise Age age of an employee Appr I if employee has been in an apprentice in the finm and zero otherwise educom completed level of education Empl number of employees in the firm Europea I if employee is European and zero otherwise Exportsh export share in produced output Female I if employee is a female and zero when a male Food I if individual is working in the food sector Foreigli 1 if firm holds foreign licenses Found 1994 - year business was founded Jobamnd I if employee is currently doing administrative or clerical work Jobcom 1 if employee is currently doing commercial/sales work, working in maintenance, or as a technician jobman 1 if employee is currently working as a manager or supervisor/foreman Jobprod I if employee is currently working as a production worker jobprof I if employee is currently a professional (engineer, accountant, etc.) Jobsupp I if employee is currently working as support staff Kumasi 1 if individual is working in a firm in Kumasi Married I if employee is married and zero otherwise Metal 1 if individual is working in the metal sector mid I if individual has completed middle school Mowage monthly wages Mowork number of months employed in the firm (tenure) Non I if individual has no completed education Numexpat number of expatriates in the firm Other I if individual is working in a firm in another location Permem I if employee is permanent and zero otherwise Pri I if individual has completed primary school Proftech I if individual has completed technical/polytechnic o:r professional education Sec I if individual has completed secondary school Textile 1 if individual is working in the textile sector Train I if employee has been trained inside the firm and zero otherwise Training I if employee has been trained and zero otherwise Trainou I if employee has been trained outside the firm and zero otherwise Uni 1 if individual has completed university Union 1 if employee is member of a labor union and zero otherwise Valad value added: total value of sales minus costs of raw material inputs and indirect costs (electricity etc.) Voc I if individual has completed vocational education Wood I if individual is working in the wood sector Note: The prefix I indicates that it is the natural logarithm of a variable and d that it is a dummy variable which takes the value one or zero. The suffix 1 indicates it is the first wave of the RP]ED data and sq denotes that it is the square of the variable. Variable Interaction variables 44  Ageeuro age* traiou Agetrao age*traiou Ageunio age*union Apprmeta appr*metal Apprtext appr*textile Apprtrao appr*traiou Apprunio appr*union Apprvala appr*lvalad Apprwood appr*wood Educfe female*educom Empleuro lempl*europea Emplmeta lempl*metal Empltext lempl*textile Empltrao lempl*traiou Emplunio lempl*union Emplvala lempl*lvalad Emplwood lempl*wood Eurotrao europea*traiou Eurounio europea*union Expeuro exportsh*europea Expmeta exportsh*metal Exptext exportsh*textile Expunio exportsh*union Expvala exportsh*lvalad Expwood exportsh*wood Femtex female*textile Folieuro foreigli*europea Folimeta foreigli*metal Folitext foreigli*textile Foliunio foreigli*union Folivala foreigli*lvalad Foliwood foreigli*wood Foodexp food*exportsh Founeuro found*europea Founmeta found*metal Fountext found*textile Fountrao found*dtraiou Foununio found*union Founvala found*lvalad Founwood found*wood Metaeuro metal*europea Metalexp metal*exportsh Metatrao metal*traiou Metaunio metal*union Mowktrai mowork*train Nexpeuro numexpat*europea Nexpmeta numexpat*metal Nexptext numexpat*textile Nexptrao numexpat*traiou Nexpunio numexpat*union Nexpvala numexpat*lvalad Nexpwood numexpat*wood Permeuro permem*europea Permtrao permem*traiou 45  Permunio permem*union Texteuro textile*europea Textiexp textile*exportsh Texttrao textile*traiou Textunio textile*union Unioeuro union*europea Uniometa union*metal Uniotext union*textile Uniotrao union*traiou Uniovala union*lvalad Uniowood union*wood Valaeuro Ivalad*europea Valameta Ivalad*metal Valatext Ivalad*textile Valatrao Ivalad*traiou Valaunio Ivalad*union Valawood lvalad*wood Valfem Ivalad*female Woodeuro wood*europea Woodexp wood*exportsh Woodtrao wood*traiou Woodunio wood*union Appendix B Table B-1. Wages and Productivity Log value added Log monthly wages Variable Coefficient Standard t-statistic P >tl Coefficient Standard t-statistic P>Itj error error Accra 0.856 0.205 4.174 0.000 0.167 0.068 2.454 0.015 Age 0.146 0.045 3.224 0.001 0.050 0.015 3.308 0.001 Age squared -0.001 0.001 -2.030 0.043 -0.001 0.000 -2.482 0.013 Apprentice -0.153 0.288 -0.529 0.597 -0.364 0.095 -3.811 0.000 Constant 10.248 0.872 11.757 0.000 8.991 0.289 31.162 0.000 Female -0.408 0.240 -1.698 0.090 -0.155 0.079 -1.947 0.052 Manager 0.255 0.334 0.763 0.446 0.284 0.111 2.567 0.011 Middle school 1.552 0.316 4.907 0.000 0.344 0.105 3.289 0.001 Primary school 0.348 0.546 0.638 0.524 0.410 0.181 2.270 0.024 Secondary school 1.353 0.358 3.778 0.000 0.447 0.119 3.769 0.000 Support staff -0.992 0.295 -3.367 0.001 -0.504 0.098 -5.172 0.000 Technical school 1.347 0.433 3.113 0.002 0.394 0.143 2.750 0.006 Trainingbyoutsideproviders 0.092 0.211 0.439 0.661 0.149 0.070 2.136 0.033 Union 2.087 0.232 9.015 0.000 0.146 0.077 1.911 0.057 University 2.028 0.818 2.480 0.014 1.052 0.271 3.885 0.000 Vocational school 0.820 0.299 2.746 0.006 0.180 0.099 1.818 0.070 Wood sector -0.720 0.289 -2.496 0.013 -0.327 0.096 -3.426 0.001 Note: See appendix table A-I for variable definitions. Source: Author's calculations. 46  Table B2. Wages and Productivity Including a Continuous Variable ifor Education Log value added Log monthly wages Variable Coefficient Standard t- P>ItI Coefficient Standard t-statistic P>jtj error statistic error Accra 0.776 0.204 3.814 0.000 0.161 0.067 2.423 0.016 Age 0.161 0.046 3.540 0.000 0.046 0.015 3.083 0.002 Age squared -0.001 0.001 -2.484 0.013 -0.001 0.000 -2.355 0.019 Apprentice -0.086 0.291 -0.294 0.769 -0.367 0.095 -3.857 0.000 Completed level of education 0.434 0.069 6.318 0.000 0.137 0.022 6.077 0.000 Constant 9.632 1.015 9.486 0.000 8.587 0.332 25.835 0.000 Female -0.542 0.237 -2.287 0.023 -0.182 0.077 -2.346 0.020 Manager 0.266 0.330 0.806 0.421 0.295 0.108 2.736 0.007 Permanent employee 0.415 0.689 0.602 0.548 0.425 0.226 1.882 0.061 Support staff -0.851 0.295 -2.891 0.004 -0.504 0.096 -5.224 0.000 Training by outside providers 0.036 0.213 0.168 0.867 0.131 0.070 1.876 0.061 Union 2.234 0.229 9.758 0.000 0.146 0.075 1.945 0.053 Wood sector -0.512 0.288 -1.778 0.076 -0.273 0.094 -2.894 0.004 Note: See appendix table A-1 for variable definitions. Source: Author's calculations. 47  Table B3. Wages and Productivity Including Female Interaction Variables Log value added Log monthly wages Variable Coefficient Standard t-statistic P>ItI Coefficient Standard t- P>}tj error error statistic Accra 0.887 0.205 4.327 0.000 0.192 0.067 2.886 0.004 Age 0.157 0.045 3.484 0.001 0.055 0.015 3.784 0.000 Age squared -0.001 0.001 -2i.212 0.028 -0.001 0.000 -2.802 0.005 Apprentice -0.104 0.289 -0.359 0.720 -0.358 0.094 -3.816 0.000 Constant 10.060 0.843 11.929 0.000 8.897 0.274 32.422 0.000 Female x age -0.020 0.008 -2.585 0.010 -0.009 0.003 -3.526 0.000 Female x wood sector 0.700 0.717 0.977 0.329 0.882 0.233 3.784 0.000 Manager 0.218 0.334 0.655 0.513 0.268 0.109 2.467 0.014 Middle school 1.546 0.313 4.932 0.000 0.328 0.102 3.216 0.001 Primary school 0.336 0.542 0.621 0.535 0.400 0.176 2.265 0.024 Secondary school 1.371 0.358 3.834 0.000 0.437 0.116 3.752 0.000 Support staff -1.028 0.294 -3.495 0.001 -0.516 0.096 -5.389 0.000 Technical school 1.366 0.431 3.169 0.002 0.400 0.140 2.852 0.005 Training by outside providers 0.060 0.210 0.283 0.777 0.127 0.068 1.861 0.063 Union 2.080 0.231 9.017 0.000 0.141 0.075 1.880 0.061 University 2.080 0.815 2.553 0.011 1.081 0.265 4.077 0.000 Vocational school 0.803 0.298 2.690 0.007 0.153 0.097 1.573 0.117 Wood sector -0.864 0.316 -2.735 0.007 -0.494 0.103 -4.803 0.000 Note: See appendix table A- 1 for variable defiitions. Source: Author's calculations. 48  Table B4. Wages and Productivity Including Unskilled Worker Interaction Variables Log value added Log monthly wages Variable Coefficient Standard t-statistic P>Itl Coefficient Standard error t-statistic P>Itl error Accra 0.851 0.205 4.159 0.000 0.178 0.067 2.652 0.008 Age 0.146 0.045 3.232 0.001 0.052 0.015 3.468 0.001 Age squared -0.001 0.001 -2.034 0.043 -0.001 0 -2.603 0.010 Apprentice -0.079 0.290 -0.271 0.786 -0.410 0.095 -4.308 0.000 Constant 10.251 0.869 11.791 0.000 8.952 0.286 31.347 0.000 Female -0.504 0.241 -2.090 0.037 -0.117 0.079 -1.480 0.140 Manager 0.465 0.346 1.346 0.179 0.172 0.114 1.515 0.131 Middle school 1.371 0.334 4.107 0.000 0.462 0.110 4.217 0.000 Primary school 0.337 0.542 0.621 0.535 0.414 0.178 2.323 0.021 Secondary school 0.991 0.389 2.550 0.011 0.554 0.128 4.339 0.000 Support staff -0.971 0.297 -3.273 0.001 -0.537 0.097 -5.509 0.000 Technical school 1.338 0.430 3.112 0.002 0.396 0.141 2.807 0.005 Training by outside providers 0.143 0.210 0.681 0.496 0.124 0.069 1.798 0.073 Union 2.115 0.230 9.185 0.000 0.137 0.076 1.810 0.071 University 2.147 0.818 2.625 0.009 1.046 0.269 3.891 0.000 Unskilled x middle school 0.383 0.263 1.460 0.145 -0.265 0.086 -3.069 0.002 Unskilledx secondary school 0.972 0.432 2.250 0.025 -0.277 0.142 -1.955 0.051 Vocational school 0.821 0.298 2.757 0.006 0.169 0.098 1.728 0.085 Wood sector -0.772 0.287 -2.686 0.008 -0.305 0.094 -3.228 0.001 Note: See appendix table A-1 for variable definitions. Source: Author's calculations. 49     Policy Research Working Paper Series Contact Title Author Date for papei WPS2149 Income Gains to the Poor from Jyotsna Jalan July 1999 P. Sade Workfare: Estimates for Argentina's Martin Ravallion 33902 Trabajar Program WPS2150 Who Wants to Redistribute? Russia's Martin Ravallion July 1999 P. Sader Tunnel Effect in the 1990s Michael Lokshin 33902 WPS2151 A Few Things Transport Regulators Ian Alexander July 1999 G. ChenrtKRmi-, Should Know about Risk and the Cost Antonio Estache 36370 Of Capital Adele Oliveri WPS2152 Comparing the Performance of Public Antonio Estache July 1999 G. Chanet>- and Private Water Companies in the Martin A. Flossi 36370 Asia and Pacific Region: What a Stochastic Costs Frontier Shows WPS2153 The Mystery of the Vanishing Martin Ravallion July 1999 P. Sadere Benefits: Ms. Speedy Analyst's 33902 Introduction to Evaluation WPS2154 Inter-industry Labor Mobility in Howard Pack August 1999 H. Sladovich Taiwan, China Christina FPaxson 37698 WPS2155 Lending Booms, Reserves, and the Barry Eichengreen August 1999 S. Kpunde,h Sustainability of Short-Term Debt: Ashoka Mody 39591 Inferences from the Pricing of Syndicated Bank Loans WPS2156 How Has Regionalism in the 1990s isidro Solcaga August 1999 L. Tabana Affected Trade? L. Alan Winters 36896 WPS2157 How Regional Blocs Affect Excluded Won Chang August 1999 L. Tabada Countries: The Price Effects of L. Alan Winters 36896 MERCOSUR WPS2158 The Effect of Foreign Entry on George R. G. Clarke August 1999 P. Sntim-Ad.i2 Argentina's Domestic Banking Robert Cull 38526 Sector Laura D'Arnato Andrea Molinari WPS2159 Provincial Bank Privatization in George R. G. Clarke August 1999 P. Sintim-Ac2cc .a Ye Argentina: The Why, How, and Robert Cu I 38526 "So What?" WPS2160 Protecting the Poor from Francisco Ferreira August 1999 PREM Macroeconomic Shocks Giovanna Prennushi 87735 Martin Ravallion  Policy Research Working Paper Series Contact Title Author Date for paper WPS2161 Will the Real "Natural Trading Maurice Schiff August 1999 M. Kasilag Partner" Please Stand Up? 39081 WPS2162 Quantifvinc the Fiscal Effects Shantayanan Devarajan August 1999 H. Sladovich of Trade Reform Delfin S. Go 37698 Hongyi Li WPS21 63 Coverage under Old-Age Security Estelle James August 1999 M. Leenaerts Programs and Protection for the 84264 Uninsured - What Are the Issues? WPS2164 Challenging El Salvador's Rural Maureen Lewis August 1999 M. Lewis Heaith Care Strategy Gunnar S. Eskeland 39080 Ximena Traa-Valerezo WPS21 65 The Russian City in Transition: Martha de Melo August 1999 H. Sladovich The First Six Years in 10 Volga Gur Ofer 37698 Capitals WPS2166 Seeking Votes: The Political Norbert R. Schady August 1999 N. Schady Economy of Expenditures by the 88247 Peruvian Social Fund (FONCODES). 1991-95 WNPS2167 Social Capital and the State: Deepa Narayan August 1999 B. Jones Complementarity and Substitution 39475 /