Endogenous Wage Returns to Training in Manufacturing: the Kenyan and Senegalese Cases
Antoine Leblois∗ February 2008
Abstract Returns to on-the-job training initiated the treatment models literature and was studied by a large panel of works. We study in this paper the impact of formal training on wage levels in two African countries (Kenya and Senegal) using the Regional Program on Enterprise Development (RPED) database collected by the World Bank. Thanks to the employer-employees data, we observe the potential impact of current training of employees on their actual wages in the firm. We run a cross-sectional econometric analysis, based on an employer-employee matched database that allows us to control for firm level effects. We corrected selection bias in the earnings equation with the method developped by Heckman to correct such a bias that could be a consequence of the endogeneity of training in the Mincerian equation. We used IV method to correct endogeneity of the training variables in the model. The model we drew makes it possible for us to control for endogeneity and selection bias for training. Such an analysis validate the theoretical hypothesis of a share of training costs between the employer its employees. In this paper, we make a rapid overview of such an analysis and then try to draw a comparative study from statistics of two different countries that are Senegal and Kenya.
JEL classification code: O12. Keywords: training, treatment, Mincer regressions. ∗
CIRED (Centre International de Recherche sur l’Environnement et le D´eveloppement), leblois@centre-
cired.fr
1
Contents 1 The ‘labor economics’ of training
3
1.1
Training and firm specific skills . . . . . . . . . . . . . . . . . . . . . . . . .
3
1.2
Selection bias . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5
2 Data and model 2.1
2.2
6
Context an data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6
2.1.1
Description of surveys . . . . . . . . . . . . . . . . . . . . . . . . . .
7
2.1.2
Descriptive statistics of the workers . . . . . . . . . . . . . . . . . . .
9
2.1.3
Heterogeneity and training concentrations within firms . . . . . . . .
10
Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11
3 Motives and wage impact of undertaking a formal training 3.1
12
Determinants of training . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
12
4 Impact on earnings with endogeneity of training, an empiric evaluation
19
4.1
Raw differences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
19
4.2
Instrumental variables regressions . . . . . . . . . . . . . . . . . . . . . . . .
20
4.2.1
Firm-financed training received inside and outside the firm . . . . . .
25
4.2.2
Self-financed training . . . . . . . . . . . . . . . . . . . . . . . . . . .
25
5 Discussion and perspectives
30
Introduction There is a latent lack of labour demand in Africa but supply, in quantity, does not seem to be a real issue. However, the quality of labor supply could improve activity through the channel of productivity and dynamism in the labour market. A lack of work-related training affecting long-term performance and the poor quality of labor could have been part of the explaination for low growth rates of the continent. This is mostly the case in high value added products which require more skilled labour and different capacities for the workforce.
2
Training is a very qualitative and subjective notion. Indeed it is tied with the acceptation of certain norms, conventions and customs of each working place and also with experience and tenure. Thus initial literature estimating wage returns to training, was based on indirect training measures such as seniority or labor market experience (Mincer and Jovanovic, 1981; Bishop, 1989). New data sets allows us to add direct training variables. Following the hypothesis of decreasing returns in education, the most important change has to be triggered by a development of basic skills and thus elementary educational system. One could however question: are efforts needed to address a ‘skill-gap’ in Africa, particularly in manufacturing where first step innovations often took place in developed countries. Such question raised a second one, does such gap if it exists, come from a lack of investments from employers or a lack of training supply ? We will try to test in this article wether training has an large and significant impact on wage, controling for the motivation bias (section 1.2). We will also test for the potential share of its costs between the employer and the empoyee in the African context, comparing the case of Kenya and Senegal. The paper is organised as follows. The first section recalls the theoretical framework of training in economic literature. In a second section we describe the data and the model. We finally tried to extricate the major determinants of training in those countries before testing for its impact on earning levels. We conclude in a fourth and last section.
1 1.1
The ‘labor economics’ of training Training and firm specific skills
Firm specific skills is per definitionem not transferable to other workplaces. They do not allow to develop skills that could be easily valuable on the labor market. All sorts of training are added to general-purpose education, but more precise and specific training make the workers able to fit labor demand more precisely. Labor economists often oppose general training to vocational and on-the-job ones (Weiss and Rubinstein, 2007). The former is often undertaken by attending a scholarship, at a cost for the worker, and is more general; the latter allows acquiring branch or industry specific human capital and only obtained by 3
allowing time to it, only representing an opportunity cost. Specific skills are also benefiting to employers, who can gain from technological adaptation of their workers and from their improved capacities and productivity. They also tend to have a preference for it since it is less valuable in any other enterprise. There is thus an incentive for employers to share the cost (or even to pay for it) of more specific formal training that general ones, that are only and entirely profitable to the workers. This effect is enhanced by the incentive of the worker to quit the firm after its investment in human capital. Indeed wage returns to training are more easily valuable for job-switchers, as observed in France by Foug`ere et al. (2001). It is well known that on-the-job training can enhance future workers’ earnings but also diminish them by making the worker to accept lower wages. In conclusion, the theory assumes that the more specific the training, the bigger the firm-supported share of the training cost will be. There are also several more empirical reasons that explain the support of firms to purely general training of employees (particularly in Africa: Rosholm, Nielsen, Dabalen, 2007). Acemoglu and Pischke (1999) suggest that many types of non-observable skills exist. For instance those that could be learned on-the-job: by doing or advised by a mentor and which could be considered as a training for them. There are thus some sort of productivity gains, even stemmming from general training, that have no impact on the employee’s wage. Similar results can be found using bargaining models (e.g. Booth, Francesconi and Zoega, 1999). Rosholm, Nielsen and Dabalen also assumed that under imperfect competition the firm can set the wages under the marginal product level, to recover the cost of training. It creates an incentive for the firm to provide training courses to the worker (even general or non-specific ones) at a low level of costs. We have seen that under competition hypothesis, sharing costs of training should depend on the degree of the specificity of skills in the firm. Investment in specific skills must limit the turn-over, and improve the employee’s bargaining power in case of a negative transitory chock of demand. It is also probable that a worker to whom a firm is paying direct costs of a general training would actually accept a lower starting wage. Returns to training is thus highly tied with the background and the unobserved skills of workers. Empirical literature highlights the importance of unobserved heterogeneity between workers (Goux, 4
Maurin, 2000) and job-match (Barron et al., 1989) that enhance the probability of attending a training (follow a training program) and bias the impact of training on the starting wage. Self-educated people have a facility to learn by doing or by watching, and thus will have higher wages in average. Returns to specific skills like rewards and wage premiums are not necessarily related to productivity and competition does not always allocate premiums proportionally to productivity as seen by Biesenbrock (2007). More than measure problems, one would like to know to what extent wage growth correlates with productivity growth. Indeed wage and productivity gaps or premiums in developing countries are likely to differ in the long run, contrarily to recent evaluations in developped countries. The pertinence of evaluations of training returns depends on the acceptance of such an hypothesis. Before trying to see if the fact of initiating a training program can enhance earnings and work-related returns, it is quite logical to study the relation between earnings and productivity of employees.
1.2
Selection bias
The correlation between training and earnings is generally strong but causal impact is difficult to assess as for education that subit the same selection bias. Training could play an important role in narrowing the wage gap, however its impact on the relative wages of workers is more ambiguous. Wage returns to training are quite arduous to prove, indeed selection bias but also limits to quantification and definition of the notion are the main impediments. Training contents of a job offer is specified and chosen as such by a worker and thus reveals the desire of the worker to train (signal theory ` a la Spence). Hence the effect of training on earnings is associated with a possible selection bias that is proportional to the individual probability of undertaking a training. Indeed the will to undertake a training is an endogeneous variable in a Mincer equation and is tied with earnings. We should estimate the Our objective is to evaluate the impact of training on the earnings. The main obstacle to such a calculation is the potential endogeneity of training. The training variable underlies a choice variable, that supposes that the employer provides a training, and that the employee is willing to undertake such a formal training. Such assertions underlie unobserved characteristics by the econometrician and will give a biased estimation 5
of the causal effect of training on wages. We will test this endogeneity and control it by simultaneous equations and instrumental variables (IV) for the training. We control for firm fixed effects that must have an impact by means of its training policy on the probability to undertake one. There is a link between (in the literature??, but also in the data) the schooling level and the probability to receive a training. That is why we consider that the probability of receiving a training depends on the education and other characteristics of the individual. Unfortunately it seems rather clear that it does not only depend on individual’s observable characteristics. The same analysis could be drawn for self-educated people (unobservable that creates heterogeneity) or workers who are bright to learn alone and from their own mistakes. More than selection bias there is an endogeneous role of training into the Mincer equation due to such non-observable differences in personal characteristics and possible selfselection (Biggs, Shah et Srivastava 1995a). Biggs et al. (1995b) estimates the impact of training on output, that could have a repercussion on employee benefits. Employer-employee data are very useful to control for firm characteristics and to compare different groups (minority and majority groups) and their wages equations relatively to their productivity and to the firm policy. Indeed plant-level, input and output informations – even if the data set only consider the manufacturing sector and big firms – could be an good way to test the reality of wage premiums, and the hypothesis of efficient allocation of ressoures by the market competition. Cross-sectional data limit the study to the non-temporal and non-dynamic side analysis of productivity. Plant-level earnings equations and fixed effects permit to control for idiosyncratic behaviours of firms in term of training.
2
Data and model
2.1
Context an data
Our database stem from two countries (Kenya and Senegal), it has been collected between 2003 and 2005. These countries are quite different regarding their background, their history as well as their size1 , but their levels of developpement are identical in terms of GDP (even 1
Kenya is three time bigger in terms of population
6
if the Senegalese one was growing faster during the last few years, see Table 1). Senegal has more urban dwellings (41.4 % vs 20.5 in Kenya) and less peasants, its health indicators are better but the literacy rate is low comparing to Kenya. Indeed Kenya is an English speaking country, and an old British colony, which can influence the development patternparticularly when compared to a former French colony. Senegal is a typical West-African French colony and to the end of handle with the whole population, the questionnaire was writen in French. Table 1: Average statistics and country specificitues of Kenya and Senegal Variable / Country Surface Area (sq. Km)
Kenya
Senegal
569 140
192 530
Population (millions)
33.47
11.39
Population growth (annual rate %)
2.22
2.37
Population density
58.8
59.14
Urban population (%)
20.5
41.4
Life expectancy at birth (years)
48.35
56.14
Infant mortality rate (per 1,000 live births)
78.5
77.6
GDP growth (annual %)
4.34
6.17
GDP per capita (constant 2000 US$)
426.56
461.19
Unemployment, total (% of total labor force)
40.00∗
48.00∗
Literacy rate, adult total (% of people above 14)
73.61
39.28
Inflation, consumer prices (annual %)
11.62
0.51
Foreign direct inv., net inflow (current US$ bn)
0.05
0.07
Internet users (per 1,000 people)
44.82
42.33
Mobile phone subscribers (per 1,000 people)
76.08
90.29
Personal computers (per 1,000 people)
13.18
21.25
Telephone mainlines (per 1,000 people)
8.94
20.58∗∗
Kikuyu 22%, Luhya 14%,
Wolof 43.3%, Pular 23.8%,
Ethnic breakdown
Luo 13%, Kalenjin 12%, Kamba 11%,
Serer 14.7%,Jola 3.7%,
Soninke 1.1%, Kisii 6%, Meru 6%,
Mandinka 3%, European
other African 15%, non-African 1%
and Lebanese 1%, other 9.4%
Source: WDI 2004, ∗ 2001 estimations, ∗∗ 2003 estimation
2.1.1
Description of surveys
The RPED, a multi-year study of the manufacturing sector in several African countries, was organized by the African Region Technical Department of the World Bank. The RPED is designed to provide an overview of the performance of manufacturing firms in the poststructural adjustment period, and focuses on a wide variety of aspects of firm behavior, governance and structural carcateristics. The Investing Climate Survey (ICS, referred as 7
ICA for Investing Climate Assessment before 2000) include several aspects of enterprises declined in different parts of the survey. In the manufacturing questionnaire we used a part dedicated to employee considerations. In each surveyed country, up to ten employees have been randomly sampled in each firm following the idea advocated by Mairesse and Green??an (1999). These surveys are based on the idea that workplace is a microdata unit where labour supply and demand are resolved. In that spirit, the ICA were conducted in 2003-2004 for the firms related part, and 2003-2005 for the workers related part in Kenya and Senegal. Those surveys collected data both on the firm characteristics and on a sample of employees in each workplace, offering a written questionnaire specifically tailored for each country. We must notify that the sample of firms is a selection of quite important sized firms that makes the sample of workers not representative of the labor force even restricted to the formal sector of each country. Such surveys enable cross-country comparisons as they are made of very similar questions. The information is drawn from a sample of firms (250 in Kenya and 254 in Senegal) based on a sampling plan made of 1645 formal companies. These firms were randomly selected using a stratification based on sector, size and localization and represented 59.6% of the formal sector in 2003 and 68.9% of its formal permanent jobs. The structure of the data allows building up matched worker-firm data sets. Note that all employees of small firms have been interviewed, while the sampling rate decrease with the size of the firms. The number of workers interviewed in Kenya and Senegal are respectively equal to 1771 and 1459. The country choice is made on the size of the panel of individuals and the quality of potential instruments to control for training auto-selection that could also be compared to an endogeneity of this variable. Frazis and Loewenstein (2003) showed that specifications including both a participation indicator and a training duration will have upward biased participation and downward biased duration coefficients as in the classical case. Conjunture did not seem to have a strong impact on the training returns given that none of those countries were affected by a macroeconomic choc during the period of the survey; we can thus assert that those data sets are comparable. Unfortunately we do not have precise enough questions about the place where happened former training, to allow us to study the 8
impact on growth of earnings. The training has, indeed, to be undertaken in the present firm to have a real impact on earnings. 2.1.2
Descriptive statistics of the workers
The questionnaires of the different surveys allow us to construct comparable human capital indicators for the workers in the two countries. We compute for each respondent the number of years of completed schooling, the number of years of experience off the current firm and years of tenure in the incumbent firm. We further construct dummy variables to indicate workers who received formal training in the past, during the last year and for those currently receiving such a training. We also create a dummy for gender. The information on the marital status was not provided in both countries, we thus chose not to take it as a control variable in our study, mainly for comparison reasons. Age variable is not informed for the entire sample, we thus decided not to take it into account. It does not give more information that the addition of tenure in and out the firm and the completed years of shooling. Table 2: Worker summary statistics in Senegal Kenya Variable
Senegal
Mean
Std. Dev.
Number of interviewed employees
8.577
1.997
Sexe (1 if female)
0.185
.
1771
0.168
.
1459
Education
8.981
7.874
1771
10.402
5.597
1459
1771
N Mean
Std. Dev.
N
7.847
2.446
1459
Tenure in the current firm
11.58
3.798
1771
8.523
7.835
1459
Actual experience off the firm
3.993
5.995
1771
4.736
6.449
1459
Temporary workers (1 if Yes)
0.194
.
1771
0.167
.
1459
Unionised workers (1 if Yes)
0.252
.
1771
0.289
.
1459
Managers (1 if Yes)
0.101
.
1771
0.068
.
1459
Days of last year training
46.759
75.413
114
111.687
135.710
67
Trained in the past (1 if Yes)
0.333
.
1771
0.354
.
1459
Self-financed (1 if Yes)
0.171
.
1771
0.251
.
1459
Firm-financed (1 if Yes)
0.156
.
1771
0.103
.
1459
0.071
.
1771
0.103
.
1459
Self-financed (1 if Yes)
0.033
.
1771
0.038
.
1459
Firm-financed (outside the firm, 1 if Yes)
0.015
.
1771
0.051
.
1459
Firm-financed (inside the firm, 1 if Yes)
0.023
.
1771
0.014
.
1459
Among which:
Currently trained (1 if Yes) Among which:
The pourcentage of trade union members is higher in the Senegalese than in the Kenyan 9
firm sample (Table 2). We also observe a negative impact on the share of unionised workers on wages, which is less significative in Senegal. Unionizing, whose negative impact on wages could be interpreted as a pejorative signal for employers that can offset the increase of remuneration. We finally have a quite comparable framework between the two countries concerning workers statistics except the education background and experience off the firm being higher in Kenya in spite of a younger labor force. The proportion of managers and women interviewed in Kenya could explain a greater share of trained workers, in share of the labor force but also in duration of trainings, if we assume that the probability of those groups to undertake a training is higher. Such effect will be confirmed by studying the determinants of training (section 3.1). Among current training the Kenyan workers seem to prefer training inside the firm (2.3% vs. 1.5 % outside) whereas Senegalese ones are more trained outside their firms (5.1% vs. 1.4% inside). Let us now investigate the composition of the firm samples in the two countries and explore the comparability of those descriptive statistics. 2.1.3
Heterogeneity and training concentrations within firms
Table 3: Firm summary statistics Kenya Variable
Senegal
Mean
Std. Dev.
N
Mean
Std. Dev.
N
Total workforce
232.553
448.36
264
129.969
425.475
260
Permanent workers
143.373
332.253
284
75.100
245.192
261
Training program undertaken (1 if Yes)
0.44
.
284
0.317
.
262
Share of unionised workers (1 if Yes)
0.369
.
284
0.283
.
262
Table 4 summarize incidence of training and its different types among the two countries. We can indeed see in that table the importance of taking into account the heterogeneity of firm in terms of training policy. The distribution of trainings among the firms is not proportionnaly allocated between firms, and it seems that only a limited share of the firm sample is concerned by training (75 in Senegal and 78 in Kenya for the current training that will be studied in the next section). Firm summary statistics are constructed from the employers part of the questionnaire. There can thus be some differences between the employees revelations and the employer side. 10
Table 4: Concentration of training in firms All
Passed
Current
Among
Firm-financed
Self-financed
trainings
training
training
which:
current training
current training
625
599
128
67
59
Kenya Number of trainees Number of firms concerned
194
191
78
43
47
3.222
3.136
1.615
1.558
1.255
Number of trainees
575
517
150
95
55
Number of firms concerned
201
197
75
36
46
2.861
2.624
2
2.639
1.528
Ratio of trainees per firm (on the 10 or less empl. interviewed) Senegal
Ratio of trainees per firm (on the 10 or less empl. interviewed)
We have much smaller firms in the Senegalese sample, with a comparable standard deviation (Such a difference is representative for the country differences according to WB 2005a). Much more women own those firms in Senegal, and the difference of training programs revealed by the manager are much larger in the Kenyan sample, such a difference reveals the importance of the firm level training policy. The rate of union members seems to be also overestimated by managers in Kenya. More than that, we found that large, public and exporting firms, invest more in trainings, a result which is consistent with Dabalen, Nielse and Rosholm (2007) and Biggs’s (1995) findings.
2.2
Model
We draw an econometric model to assess the potential impact of earnings on wages as well as to test wether there is a share of training costs between the employer and his or her employee. We use a Heckman twostep approach (Heckman’s two-step consistent estimates: Heckman, 1979) to endogenize training. Let us begin by the second step which is the earnings2 determination of our basic Mincer equation. The first step equation is the following: Yi,j = β.X1,i + f irmj + ǫi estimated by a classical probit regression. X1 is a vector of variables, composed by completed 2
The variable used is the current hourly salary containing all premiums and gratifications
11
years of schooling, tenure in the current firm, actual experience off the firm, dummy variable, equals one if employee is a female, being a member of a worker union, holding a temporary job, or contract, 3 occupational categories: managers, professionnals, skilled workers, and non production workers3 and of course the type of training (we distinguished different types of current trainings that are: the firm-financed trainings, divided in training undertaken ouside the firm and inside the firm and the self-financed trainings). Z is a set of IV (the time to go to workplace in hours and the distance in Km.) and in the second step, we get the following equation: Ti,j = β.X2,i + γ.Zi + f irmj + ǫi X2 = X1 + squared of most important control variables. We controle for firm fixed-effects (in the two steps4 ), which seem (see table 6) to improve the capacity to explain the choice to train more than the significativity of potential determinants. We also control regressions by tenure in the current firm, experience off the firm, occupational status (unskilled workers is the reference group). The first step regressors are the same, without the squared variables that reduces the degrees-of-freedom without helping in regressions that explain only half of the variations in the training choice. We exceptionnaly took the squared completed years of schooling as a regressor in the comparison of probit and tobit regressions (Table 5) that explain the difference of the impact of the incidence of overtaking a training and its duration.
3
Motives and wage impact of undertaking a formal training
3.1
Determinants of training
The question asked to the respondants during the survey was “Did you received formal training ...” and there were three choices to this closed answer: self-financed and either firm3 4
The reference occupation is unskilled workers There is indeed no reason that specificity of firm explains only one of the two firm-behaviours that are
the salary policy and the training policy
12
financed inside the firm or firm-financed outside the firm. Education is a major determinant of training, we can also observe the importance of being a skilled worker, (in our classification) to accomplish a training. By running probit regression of the probability of undertaking a training (Tables 5 and 6), we can estimate the determinants of such a choice in the two countries. We separate the different types of training in these estimates, to the purpose of determining their impact on wages. Indeed the literature (efficiency wage literature, for example: Yellen, 1984) stresses the potential share of training costs between employer and employees, and it would be a different story in case of firm-financed and self-financed trainings. More than that, it seems that the place where the training program takes place can play a role in such a study (especially in the effect of IV we will study in the next section). The tenure in the current firm negatively influences the choice to undertake a training inside the firm, but such a result could show the impact of age that is not taken as regressor and taking into account the positive sign of off the firm experience coefficients in many specifications. The impact of experience seem not to explain the variation of dependant variable, and it is not significative in any cases. The firm fixed-effects improve the global explanation of variation dependant variable but do not improve the variables significativity. Temporary workers seem to be as well paid as the unionised workers in the two countries, but their probability to undertake a training is not clear at all (not significative in many cases, and thus not robust to different specifications). The fact of being unionised is not taken into account in Kenya because it was perfect predicator (it perfectly predicts the choice, i.e. no worker currently undertaking a training outside the firm is unionised). The variations of the dependant variable that are explained are very low, because of the high variability of the quality of training. Indeed when we regress (Table 6) the hours of last year training (duration model) and compare it to the impact of the incidence of undertaking a training (dummy), the impact of variables are not the same in amplitude as well as in significativity. It is indeed quite hard to quantify a training, especially by the proxy of the number of hours, that does not reflect the quality of the training and the profitability for the worker. We can then make the same regression with the self-financed training (Table 7), that is not 13
so much improved by the presence of firm fixed effect. It seems quite logical that such an effect is not worthwhile to control for, because of the low impact of firm training policy on the self-financed training that are (frequently, I suppose) decided by the worker him(– her)self.
14
Table 5: Probit of current firm-financed training, with and without firm fixed-effects Completed years of schooling
Kenya
Kenya
−0.0738
−0.240∗∗∗
(0.0596) Squared of completed years of schooling
0.00381∗∗ (0.00172)
Tenure in the current firm
−0.0106 (0.00905)
Actual experience off the firm
−0.00254 (0.0109)
Has received a training in the past
(either in the firm or outside this firm)
Trade Union member
15
Woman
0.877∗∗∗
(ingenieur, scientifique, economiste, etc.)
Technician, Supervisor, maintenance and repair
(0.0128) 0.000254 (0.0179) 0.906∗∗∗
−0.00403∗∗
(0.00141)
(0.00192)
−0.0000240 (0.00825) −0.00591
−0.00312
(0.0108)
(0.0110)
0.356∗
0.588∗∗
−0.0562
0.0607
(0.194)
(0.316)
(0.153)
(0.191)
0.509∗
−0.147 (0.182) 0.133
(0.512)
(0.687)
(0.382)
(0.370)
0.156
0.437
0.330
0.461
(0.283)
(0.433)
(0.282)
(0.328)
0.372
0.465
0.202
0.329
(0.326)
(0.455)
(0.237)
(0.329)
0.239
0.537∗
0.131
0.129
(0.295)
(0.164)
(0.190)
0.307
0.106
(0.228)
(0.333)
−0.233
−0.141 (0.473)
0.430∗∗
−1.996∗∗∗ (0.466)
Pseudo R2
−0.0915
0.474
(0.186)
Observations
(0.250)
(0.276)
Other (keepers, kitcheners)
Firm fixed-effect
(0.0137)
(0.194)
(0.260)
Constant
−0.0147
−0.566∗
(0.195) Non qualified office worker
−0.0176
(0.0450)
−0.000599
(0.241)
(0.157)
Cadre avec diplome universitaire
(0.00221)
(0.0366)
0.109∗∗
−0.403∗∗
−0.218
Employed manager
0.00805∗∗∗
0.0373
Senegal
(0.166)
0.316∗∗
Owner
(0.0762)
Senegal
−2.170∗∗∗ (0.573)
−2.112∗∗∗ (0.273)
(0.245) 0.742∗∗
0.385 (0.305) −3.004∗∗∗ (0.363)
No
Yes
No
Yes
1771
1771
1459
1459
0.167
0.504
0.050
Standard errors in parentheses, ∗ p < .1, ∗∗ p < .05, ∗∗∗ p < .01, Regressions are robust to clustering, and there is no case of currently trained worker from the last (Other) occupational category in Kenya
0.349
Table 6: Probit regression with and without firm fixed-effects, dependant variable: Firm-financed current training Inside the firm training Kenya Kenya Completed years of schooling
−0.0612 (0.0682)
Squared of years of schooling
0.00302 (0.00194)
−0.307∗∗∗
0.0392
(0.0822)
(0.0723)
0.00901∗∗∗
(0.0259)
0.00127
0.0117
(0.0120)
(0.0245)
0.0671
0.340
−0.0170
−0.00931
(0.212)
(0.337)
(0.175)
(0.243)
(0.0192)
16
Professionals
Skilled workers
0.800∗∗ (0.339) −0.968 (0.777)
(0.0157) −0.286
(0.0223) −0.258
0.212
0.277
0.300
0.0812
(0.246)
(0.534)
(0.222)
(0.402)
(0.244)
0.241
−0.0644 (0.362)
0.201
0.267
0.101
0.489
0.166
(0.320)
(0.553)
(0.413)
(0.696)
(0.407) 0.515
0.250
(0.387)
(0.496)
0.287
0.0266
0.0947
0.960∗
(0.378)
(0.609)
(0.396)
(0.501)
−0.121
(0.278)
(0.312)
0.371
0.0865
(0.369)
(0.335)
(0.411)
Non production workers
−0.482∗
−1.013∗∗
−0.0712
0.0406
0.0303
−0.0871
Constant
−1.684∗∗∗ (0.556)
Firm fixed-effect Observations Pseudo R2
(0.613)
(0.508)
(0.923)
(0.338) −4.195∗∗∗ (1.039)
0.0373
(0.550)
0.114
(0.452)
(0.292)
0.566∗
(0.289)
−4.158∗∗∗
−0.410
0.461∗
0.272
(0.329)
(0.0111)
−0.0969
(0.373)
−2.776∗∗∗
(0.0130) −0.000340
(0.202)
0.237
(0.511)
(0.00212) −0.0160
(0.173)
(0.201)
−1.224∗∗
(0.0110)
0.888∗
(0.173)
(0.285)
(0.00868) −0.00168
(0.423)
(0.299)
−0.202
0.00151
(0.259)
(0.189)
−0.285
(0.00144)
(0.0495) −0.00423∗∗
(0.0108)
−0.0160
(0.00372)
(0.0350) −0.00157
(0.0253)
(0.0109) −0.0193
(0.00298)
0.000355
−0.00584
(0.0123)
Managers
(0.142)
−0.00657
0.00744
Trade Union member
(0.116) −0.00284
0.112∗∗
0.00332
(0.0149)
(0.378)
(0.103) −0.00270
0.0543
(0.00375)
(0.0111)
Not permanent full time job
0.103
(0.00245)
−0.00521
−0.918∗∗
0.181
Senegal
−0.000964
Actual experience off the firm
(0.164)
0.0869
Senegal
(0.00246)
−0.0102
0.521∗∗∗
0.000404
Senegal
−0.0328∗∗
Tenure in the current firm
Woman
Senegal
Outside the firm training Kenya Kenya
−0.130
(0.451) −4.794∗∗∗ (1.291)
0.402∗∗ (0.198)
0.421 (0.300)
0.236
0.261
(0.161)
(0.193)
0.473∗∗∗ (0.176) −2.273∗∗∗ (0.281)
0.318 (0.290) −2.782∗∗∗ (0.406)
No
Y es
No
Y es
No
Y es
No
Y es
1771
1771
1459
1459
1771
1771
1459
1459
0.100
Regressions are robust to clustering
0.552
0.061
0.462
0.114
0.492
0.035
0.331
Table 7: Probit and Tobit of last year training with and without firm fixed-effects (Kenya)
(Kenya)
(Senegal)
(Senegal)
(Kenya)
Probit of received a formal training last year Completed years of schooling
0.0668∗∗∗
−0.00262
(0.0222)
(0.0414)
Tenure in the current firm
−0.0174∗
−0.0385∗
Actual experience off the firm
−0.0133 (0.00941)
(0.0130)
Trade Union member
−0.483∗∗∗
−0.525∗∗
Not permanent full time job
−0.734∗∗∗
(0.0100)
(0.143)
(0.196) Woman
0.286∗∗
17
(0.129) Managers
Professionals
Skilled workers
Non production workers
Constant
0.336
(0.0202) −0.0131
(0.254) −1.046∗∗∗ (0.295) 0.413∗∗ (0.204) 0.739∗∗
(0.228)
(0.364)
0.473∗
0.647∗
(0.245)
(0.388)
0.0517∗∗∗
0.0532∗∗ (0.0264)
−0.00678
−0.0359∗∗
(0.00803)
(0.0118) 0.420∗∗∗ (0.123) 0.375∗∗
(Senegal)
(Senegal)
Tobit of last year formal training in hours
(0.0178)
−0.0112
(Kenya)
(0.0170) −0.00444
9.350∗∗∗ (1.772) −1.087 (0.929) −1.430
(0.0165)
(1.147)
0.658∗∗
−54.13∗∗∗
(0.291) 0.741∗∗
(19.17) −89.59∗∗∗
6.756∗∗∗ (2.049) −1.250 (1.063) −2.076 (1.309) −37.11∗ (21.44) −99.94∗∗∗
13.00∗∗∗ (4.054) −2.088 (2.450) −4.491
7.373∗∗ (3.558) −5.725∗∗ (2.590) −1.694
(2.940)
(2.479)
97.15∗∗
92.94∗∗
(39.26)
(37.27)
66.21
88.53∗
(0.175)
(0.293)
(25.38)
(31.44)
(49.78)
(46.10)
−0.0416
0.141
28.08∗
29.28∗
−1.429
22.89
(0.168)
(0.278)
(15.96)
(16.73)
(44.85)
(39.21)
0.411
0.906∗
13.72
30.93
88.47
166.4∗∗
(0.298)
(0.542)
(25.13)
(28.70)
(72.65)
(69.24)
38.61
39.44
191.1∗∗∗
234.8∗∗∗
(26.13)
(28.27)
(60.28)
(59.29)
0.706∗∗∗ (0.268)
1.379∗∗∗ (0.448)
0.275
0.425
0.123
0.422
18.78
17.95
26.66
68.12
(0.193)
(0.302)
(0.245)
(0.334)
(20.63)
(23.25)
(51.57)
(51.67)
−0.0525
0.0890
0.122
0.244
(0.201)
(0.314)
(0.231)
(0.343)
−2.253∗∗∗ (0.355)
−2.678∗∗∗ (0.700)
−2.630∗∗∗ (0.253)
−4.504∗∗∗ (0.662)
−9.772
17.44
57.02
(23.20)
(26.43)
(55.57)
(53.36)
−285.9∗∗∗
−310.7∗∗∗
−661.8∗∗∗
−583.7∗∗∗
(36.96)
(42.17)
(93.95)
(85.50)
121.0∗∗∗
100.4∗∗∗
261.7∗∗∗
189.2∗∗∗
(9.352)
(7.592)
(27.26)
(18.78)
−17.56
sigma Constant
Firm fixed-effect Observations Pseudo R2
No
Y es
No
Y es
No
Y es
No
Y es
1771
1771
1459
1459
1771
1771
1459
1459
0.134
Regressions are robust to clustering
0.453
0.111
0.467
0.054
0.160
0.049
0.148
Table 8: Probit with and without firm fixed-effects regarding the source of financement (Kenya)
(Kenya)
(Senegal)
(Senegal)
(Kenya)
Probit of formal current training financed by the firm Completed years of schooling
−0.0297 (0.0669)
Squared of years of schooling
0.00271 (0.00193)
Tenure in the current firm
Squared of tenure in the firm
0.0349
Squared of actual experience
Woman
18
Trade Union member
Not permanent full time job
Professionals
Skilled workers Non production workers
Observations Pseudo R2
(0.00216) 0.0526
−0.000856
−0.00400∗∗
(0.00139)
(0.00182)
0.0486
−0.00274
−0.00194
−0.000157
−0.00369
(0.0217)
(0.0467)
(0.0283)
(0.00173)
0.0196
0.0349
−0.0398∗
−0.0490∗
(0.0486)
(0.0204)
(0.0270)
−0.00117
−0.00145 0.744∗∗
(0.101)
−0.0514∗
(0.0283)
0.400∗∗
0.0831 (0.0773)
−0.243∗∗∗
(0.0501)
(0.00205)
0.122 (0.137)
−0.125∗∗∗
−0.00226
(0.00115)
0.130∗
0.0353
(0.0318) (0.00106)
0.00272∗∗∗ (0.000640) −0.0368 (0.0313)
0.000952
(0.00266)
(0.00116)
(0.00160)
−0.671∗
−0.0358
0.130
−0.314∗
(0.188)
(0.342)
(0.154)
(0.175)
(0.167)
−0.124
0.185 (0.244)
−0.207 (0.296)
−1.320∗∗∗ (0.339)
0.196
0.450
0.394
0.419
0.346
(0.294)
(0.473)
(0.266)
(0.331)
(0.318)
0.466
0.431
0.332
0.393
(0.327)
(0.490)
(0.211)
(0.282)
(0.326)
0.212
0.416
0.216
0.227
−0.00344
(0.196)
(0.338)
(0.165)
(0.176)
(0.305)
(0.704)
(0.0402)
(0.00176)
−0.468∗∗
−2.131∗∗∗
−0.0447
(0.000840)
(0.172)
−2.052∗∗∗
(0.0264)
(0.000698)
−0.0496
(0.419)
−0.0433 0.000859
(0.222)
−0.412
(0.0564)
0.00127 (0.00133)
0.000173
−0.0543
−0.301
−0.0888
(0.000878)
(0.0464)
−0.000391
(0.152)
(0.339)
(0.00125)
0.00121
(0.00319) −0.0450
0.00203∗∗
(0.299)
−0.219
0.00570∗∗∗
(0.00262)
0.216∗∗
0.00149∗∗
(0.156)
−0.288
(Senegal)
(0.0750)
(0.00396)
−0.00166
(0.00183)
(Senegal)
(0.00236)
(0.0449)
(0.557) Firm fixed-effect
0.00857∗∗∗
0.125∗∗∗ (0.0443)
−0.00250
(0.245) Constant
(0.0343)
(0.0286)
(0.206) Managers
0.0558
(0.0766)
−0.00167 (0.00115)
Actual experience off the firm
−0.255∗∗∗
(Kenya)
Probit of self-financed formal current training
0.422∗∗ (0.177) −2.368∗∗∗ (0.309)
0.292 (0.281) −3.039∗∗∗ (0.509)
0.0213
0.0465 (0.290) −2.169∗∗∗ (0.586)
−0.190 (0.298) −0.494 (0.379) −2.383∗∗∗ (0.572) 1.180∗ (0.611) 0.175 (0.628) 0.0139 (0.537) 0.327 (0.576) −3.804∗∗∗ (1.209)
−0.0760
−0.0885
(0.206)
(0.283)
0.426∗∗∗ (0.148) 0.270 (0.197)
0.428∗ (0.251) 0.508∗ (0.271)
0.130
0.267
(0.386)
(0.553)
0.606∗∗ (0.281)
0.711∗ (0.390)
0.143
0.436
(0.235)
(0.316)
0.0847 (0.250) −2.790∗∗∗ (0.529)
0.272 (0.353) −4.992∗∗∗ (0.793)
No
Y es
No
Y es
No
Y es
No
Y es
1771
1771
1459
1459
1771
1771
1459
1459
0.105
Standard errors in parentheses ∗ p < .1, ∗∗ p < .05,∗∗∗ p < .01 Regressions are robust to clustering
0.500
0.047
0.352
0.182
0.561
0.164
0.436
.8 .6 Density .4 .2 0 0
2
4 6 Log hourly earnings Untrained
8
10
Trained in the past
Figure 1: Distribution of log wages for trained and untrained workers in Kenya (in local currency)
4
Impact on earnings with endogeneity of training, an empiric evaluation
4.1
Raw differences
Figure 1 and 2 show the distribution of wages in the two countries and the raw impact of training; the shift in the distributions setting indicates that trained workers are more paid in average. The distribution of trained workers earnings are more flattened than those of non-trained ones in the two countries. We can observe raw differences in wages between treated (the group of trainees) and untreated individuals (Table 9). Looking at the results, we can stress the importance of the difference of wage differentials between starting and current earning. If the current earning differential is bigger than the starting earnings one we can assert that there is a potential impact of current training on wages. The observable and unobservable characteristics of employees is not taken into account here (we tried to take them into account in the next section). According to those very preliminary observation, firm-financed current training seem to have a positive impact on growth of earnings in Kenya whereas it is more true of self-
19
1 .8 Density .4 .6 .2 0 2
4
6 8 Log hourly earnings Untrained
10
12
Trained in the past
Figure 2: Distribution of log wages for trained and untrained workers in Senegal (in local currency) financed (unobservables are very important here, because of a relative degree of motivation of workers) current training in Senegal. Table 9: Raw Differences in Log Wages Variable:
Self-financed
(current training)
Firm-financed
Starting earnings
Current earnings
Starting earnings
(current training)
In the firm
Outside the firm Current earnings
Starting earnings
Current earnings
Country Kenya
0.924378
0.527177
0.694387
0.951038
0.705782
0.688384
Senegal
0.325981
0.430563
0.111444
0.331354
0.242839
0.218282
Raw differences in wages between treated and untreated groups are larger in Kenya than in Senegal in all types of current trainings. Such differences can be explained by the level of education (or other characteristics, like the share of managers in the sample) of those participating or beneficiating from the training courses.
4.2
Instrumental variables regressions
Chosen IV (Table 10) are the time to go to the workplace in hours and the distance to reach the workplace from home. We thought that living far from the working place could inhibit 20
the will to undertake a training, especially if the journey to workplace is very long. More than that the journey to workplace, all things being equal, could be correlated to the worker motivation which could also highly increment his (her) probability to undertake a training. Such a variables could thus capture a part of non observables like motivation or desire to work in this area and be positively linked to the choice of undertaking a training. On the other hand, the workers who attend a training inside the firm tend to be more motivated if they live far from their home, but being far from their workplace must be negatively linked to the desire of attending a training at the workplace. Table 10: IV var. summary statistics Variable
Mean
Std. Dev.
Min.
Max.
N
Time to go to workplace
0.532
0.411
0
3
1771
distance
7.314
8.451
0
70
1771
Time to go to workplace
0.63
0.483
0
4
1459
distance
8.882
7.478
0
55
1459
Kenya
Senegal
It seems that the Senegalese workers are working further away from their home, and so take a bit more time to reach their workplace. Average density of the two countries’ population are exactly the same at the coutry level (58.8 in Kenya vs. 59.14 in Senegal according to WDI 2004), informations that could imply that the spread of manufaturing in Kenya is better distributed. We then ran probit regression with IV (Table 11), for current training. We placed the squared values of these IV to estimate their marginal effect on training occurence. Treatment regressions (Table 12) for the whole sample of currently trained workers (without separating self-financed trainings from firm-financed trainings that could be done inside or outside the firm), the results are quite unsatisfactory. The non-significativity of the current training coefficient could be due to a different sign for the self-financed trainings and the firm-financed training on the one hand and the firm-financed outside the firm trainings on the other. The other two trainings are not attended at the workplace and can thus be influenced positively by the distance (or time) to reach the workplace.
21
Table 11: Probit of current training, with IV variables and with and without firm fixed-effects (Kenya) Completed years of schooling
0.0532∗∗∗ (0.0183)
(Kenya) 0.00800 (0.0333)
Actual experience off the firm
−0.0188∗∗ (0.00913)
(0.0144)
Tenure in the current firm
−0.0291∗∗∗
−0.0418∗∗
Woman
(0.0101)
(0.0184)
0.216∗
0.455∗∗
(0.122) Not permanent full time job
−0.737∗∗∗ (0.186)
Trade Union member
−0.465∗∗∗ (0.138)
Managers
Professionals
Skilled workers
(0.190) −0.968∗∗∗ (0.302) −0.630∗∗∗ (0.233)
0.0552∗∗∗ (0.0125) −0.0128 (0.00948) −0.00484 (0.00747) −0.127
(Senegal) 0.0611∗∗∗
(Senegal) 0.0615∗∗∗
(0.0143)
(0.0147)
−0.00978
−0.00970
(0.0107)
(0.0107)
−0.0264∗∗
−0.0260∗∗
(0.0111) −0.128
(0.0113) −0.141
(0.144)
(0.210)
(0.210)
0.221
−0.0389
−0.0413
(0.197)
(0.227)
(0.225)
0.149
0.271
0.277
(0.128)
(0.177)
(0.186)
0.289
0.545
0.341
0.389
0.375
(0.240)
(0.363)
(0.251)
(0.350)
(0.353)
0.387
0.568
(0.264)
(0.389)
(0.189)
(0.245)
(0.248)
0.130
0.163
0.227
0.271
0.276
(0.278)
(0.144)
(0.167)
(0.168)
0.307
0.312
(0.251)
(0.251)
(0.182) Non production workers
−0.0178
(Senegal)
−0.119 (0.206)
−0.0764 (0.300)
0.567∗∗∗
0.361∗∗ (0.157)
0.606∗∗
0.596∗∗
Journey to go to workplace (in hours)
−0.413 (0.321)
Squarred of journey
0.196 (0.168)
−1.060∗∗ (0.426) 0.521∗∗ (0.213)
−0.102
0.255
(0.279)
(0.355)
0.0440
−0.0940
(0.106)
(0.154)
Distance to workplace
0.0313 (0.0263)
Squarred of distance
−0.00124 (0.00103)
Constant
−1.677∗∗∗
Firm fixed effect
No
Y es
No
Y es
Y es
1771
1771
1459
1459
1459
(0.301)
Observations Pseudo
R2
0.129
−2.065∗∗∗ (0.595)
0.411
Reg. are robust to clustering
22
−2.103∗∗∗ (0.231)
0.084
−2.842∗∗∗ (0.373)
0.334
−2.872∗∗∗ (0.340)
0.335
Table 12: Treatment regressions of current training, with and without firm fixed-effects
Completed years of schooling
Kenya
Kenya
Kenya
Senegal
Senegal
Senegal
Senegal
(OLS)
(Treatment)
(Treatment)
(OLS)
(Treatment)
(Treatment)
(Treatment)
0.0645∗∗∗ (0.0138)
Squared of years of schooling Tenure in the current firm
−0.00107∗∗
Actual experience off the firm
Not permanent full time job
23 Trade Union member
Managers
0.0332∗∗∗
0.0376∗∗∗
0.0374∗∗∗
0.0354∗∗∗
0.0359∗∗∗
−0.000448∗∗∗
(0.00695) −0.000649∗∗∗
(0.00536) −0.000452∗∗∗
(0.00730)
(0.00777)
(0.00679)
−0.000413∗
−0.000525∗∗
−0.000532∗∗ (0.000207)
(0.000204)
(0.000154)
(0.000227)
(0.000250)
(0.000207)
0.0462∗∗∗
0.0463∗∗∗
0.0465∗∗∗
0.0262∗∗∗
0.0252∗∗∗
0.0258∗∗∗
−0.000994∗∗∗
−0.0423
(0.00782) −0.00109∗∗∗ (0.000330) −0.0291
(0.00575) −0.000996∗∗∗ (0.000241) −0.0511
(0.00676)
−0.000550∗∗
(0.000167)
0.0259∗∗∗
(0.00607)
(0.00715)
(0.00557)
(0.00554)
−0.000324
−0.000492∗
−0.000334
−0.000333
(0.000239) 0.0499
(0.000284) −0.0395
(0.000219)
(0.000218)
0.0435
0.0449
(0.0405)
(0.0534)
(0.0383)
(0.0453)
(0.0570)
(0.0424)
(0.0420)
−0.538∗∗∗
−0.570∗∗∗
−0.524∗∗∗
−0.327∗∗∗
−0.368∗∗∗
−0.330∗∗∗
−0.329∗∗∗
(0.0480)
(0.0612)
(0.0459)
(0.0533)
(0.0581)
(0.0496)
(0.0491)
−0.357∗∗∗
−0.219∗∗∗
−0.349∗∗∗
−0.120∗∗
−0.107∗∗
−0.110∗∗
(0.0426)
(0.0504)
(0.0402)
(0.0536)
(0.0500)
(0.0496)
1.146∗∗∗ 0.841∗∗∗ 0.244∗∗∗ 0.192∗∗∗
0.0985 3.070∗∗∗
Lambda (hazard)
1.237∗∗∗ (0.0838) 1.179∗∗∗ (0.100) 0.226∗∗∗ (0.0529) 0.196∗∗∗ (0.0586) −0.299 (0.452) 2.545∗∗∗ (0.126) 0.296
IV var
Observations
0.00137∗∗∗
0.0424∗∗∗
(0.111)
Firm fixed effect
0.00143∗∗∗
0.0324∗∗∗
(0.0606) Constant
0.00228∗∗∗
(0.000404)
(0.0474) Formal current training
0.00120∗∗∗
(0.000405)
(0.0432) Non production workers
0.0148∗ (0.00797)
(0.000515)
(0.0740) Skilled workers
0.0144∗ (0.00798)
(0.000427)
(0.0632) Professionals
0.0155 (0.0103)
(0.000469)
(0.000261) Woman
−0.00109∗∗
0.0154∗ (0.00877)
(0.000624)
(0.00620) Squared of experience off the firm
−0.00105∗
0.0649∗∗∗ (0.0128)
(0.000508)
(0.00575) Squared of tenure
0.102∗∗∗ (0.0156)
1.129∗∗∗ (0.0602) 0.819∗∗∗ (0.0709) 0.239∗∗∗ (0.0403) 0.192∗∗∗ (0.0440)
0.824∗∗∗ (0.0726) 0.607∗∗∗ (0.0663) 0.300∗∗∗ (0.0451) 0.116∗∗ (0.0468)
0.206∗∗∗ (0.0478) 0.865∗∗∗ (0.0911) 0.860∗∗∗ (0.0941) 0.331∗∗∗ (0.0548) 0.329∗∗∗ (0.0602)
0.837∗∗∗ (0.0680) 0.648∗∗∗ (0.0661) 0.308∗∗∗ (0.0424) 0.125∗∗∗ (0.0440)
0.321∗
0.00633
−0.863∗
−0.416∗
(0.194)
(0.0599)
(0.487)
(0.250)
2.784∗∗∗ (0.524)
5.727∗∗∗ (0.0695)
5.505∗∗∗ (0.0750) 0.472∗
−0.134
6.367∗∗∗ (0.478) 0.259∗
0.834∗∗∗ (0.0673) 0.640∗∗∗ (0.0654) 0.307∗∗∗ (0.0419) 0.123∗∗∗ (0.0435) −0.331 (0.246) 6.370∗∗∗ (0.478) 0.207
Journey
Journey
Journey
Journey
Distance
Journey2
Journey2
Journey2
Journey2
Distance2
Yes
No
Yes
Yes
No
Yes
Yes
1771
1771
1771
1459
1459
1459
1459
4.2.1
Firm-financed training received inside and outside the firm
We can make the assumption of a share of training costs, between a employer and his or her employees who directly beneficiate from the training program. Indeed only firm specific trainings are beneficent to the employer, because the worker cannot value it outside the firm. Thus employer could value the greater productivity the employee get after beeing trained by an increase of its remuneration that do not equalize this increase. By doing that he would then limit the earning gain facilitated by the training, and thus share the costs of it. We thus will estimate that such behaviour occur if the returns to training is negative. There is no evidence of sharing training costs between employer and employee in Kenya, while there is a net share of these costs in Senegal which is not significative (Table 14). Education negatively influences the probability of undertaking a training program in Kenya, which either due to a lack of data concerning inside the firm training or difficult to explain. It is also certain that the probability to undertake a training are not the same in Kenya (where it seem to be increasing with eduation) and Senegal, where it is decreasing with the number of completed years of education. The high impact of undertaking a firm-financed training inside the firm in Kenya (shift of .8 of earnings, mean value: 4.126) indicates the feeble occurence of shares of training costs. Outside trainings do not seem to follow the same rule, on the other hand such trainings are largely and significatively (in Senegal anyway) financed by a transfer of cost from the employer to the employee. In Kenya the negative impact of firm-financed training outside the firm is not very significative and of a limited scope, thus the hypothesis of share of training costs is a little bit tenuous in this case. Finally, there seems to be such a cost share in the case of training inside the firm, and such a hypothesis is less significant in the outside training regressions. More than that the explaination of the choice is less convincing (pseudo R 2 indeed do not exceed 40 % in the outside training cases, Cf. Table 15 ) and that could explain the feeble impact of endogeneous training in Kenya. 4.2.2
Self-financed training
For the self-financed training, the regressions are not very significant (contrarily to probit regressions), it could be due to the impact of self-motivation that is an unobservable variable 24
Table 13: Probit of current firm-financed training inside the firm, with IV variables, with and without firm fixed-effects (Kenya) Completed years of schooling
(Kenya)
0.0218
−0.0748
(0.0296) Actual experience off the firm
Tenure in the current firm
Woman
(0.0492) 0.0115
(0.0155)
(0.0308)
−0.00957
−0.0334∗∗
−0.000151
0.00453
(0.0111)
(0.0144) 0.767∗∗
(0.166)
Trade Union member
−0.293
(0.305) −1.178
(0.397)
Non production workers
Journey to go to workplace (hours)
(0.728) −0.288
−0.0188
(0.0112) −0.302
−0.0421
(0.0280) −0.228
(0.255)
(0.436)
0.201
0.495
(0.247)
(0.513)
0.258
−0.0575
(0.190)
(0.269)
(0.173)
(0.321)
0.109
0.0843
0.0830
0.243
(0.329)
(0.500)
(0.421)
(0.916)
0.263
0.137
0.0965
1.018∗
(0.385)
(0.563)
(0.398)
(0.565)
0.188
0.175
0.126
0.321
(0.197)
(0.335)
(0.296)
(0.398)
−0.538∗
−0.800∗
−0.0654
(0.281)
(0.456)
(0.318)
0.00255 (0.507)
−1.499∗∗∗
−0.516 (0.429)
Squarred of journey
0.0430 (0.0401)
(0.0167)
−0.976∗∗
Skilled workers
(0.0255)
(0.0124)
Not permanent full time job
Professionals
0.0505∗∗
(Senegal)
−0.00219
0.521∗∗∗
Managers
(Senegal)
(0.531)
0.322∗
0.756∗∗∗
(0.188)
(0.210)
Distance to workplace
0.0102 (0.0357)
Squarred of distance
−0.000968
0.226∗∗ (0.115) −0.00931∗∗
(0.00158)
(0.00475)
−2.834∗∗∗
−5.102∗∗∗
Constant
−2.038∗∗∗
Firm fixed effect
No
Y es
No
Y es
1771
1771
1459
1459
−1.562∗∗
(0.426)
Observations Pseudo
R2
(0.614)
0.095
0.507
Regressions are robust to clustering Standard errors in parentheses ∗
p < .1,
∗∗
p < .05,
∗∗∗
p < .01
25
(0.434)
0.067
(1.012)
0.480
Table 14: Earning regressions with endogeneous current firm-financed training inside the firm, with IV variables and with and without firm fixed-effects (Kenya)
(Kenya) 0.0651∗∗∗
(Senegal)
(Senegal)
Completed years of schooling
0.0642∗∗∗
Squared of years of schooling
−0.00104∗∗
−0.00102∗∗
(0.000508)
(0.000463)
(0.000427)
(0.000386)
0.0316∗∗∗
0.0320∗∗∗
0.0376∗∗∗
0.0376∗∗∗
(0.0138)
Tenure in the current firm
(0.0127)
(0.00574) Squared of tenure
Actual experience off the firm
(0.00533)
−0.000430∗∗
−0.000424∗∗∗
0.0155∗
(0.00877)
(0.00794)
0.00121∗∗∗
(0.00729) −0.000549∗∗
0.00120∗∗∗
(0.00660) −0.000549∗∗∗
(0.000167)
(0.000154)
(0.000227)
(0.000205)
0.0461∗∗∗
0.0460∗∗∗
0.0262∗∗∗
0.0262∗∗∗
(0.00621) Squared of experience off the firm
0.0155∗
(0.00573)
−0.000996∗∗∗
−0.000996∗∗∗
(0.000261)
(0.000240)
(0.00607)
(0.00550)
−0.000324
−0.000324
(0.000239)
Woman
−0.0405 (0.0406)
(0.0389)
(0.0453)
(0.0410)
Not permanent full time job
−0.542∗∗∗
−0.522∗∗∗
−0.326∗∗∗
−0.326∗∗∗
(0.0479)
(0.0456)
(0.0533)
(0.0483)
Trade Union member
−0.360∗∗∗
−0.355∗∗∗
−0.119∗∗
−0.119∗∗
(0.0426)
(0.0402)
(0.0535)
(0.0485)
Managers
−0.0611
1.151∗∗∗
1.147∗∗∗
(0.0631) Professionals
(0.0596)
0.848∗∗∗
0.838∗∗∗
(0.0740) Skilled workers
(0.0696)
0.245∗∗∗
0.242∗∗∗
(0.0432) Non production workers
(0.0407)
0.194∗∗∗
0.210∗∗∗
(0.0475) Current formal training
(0.0451) 0.880∗∗∗
0.0678 (0.106)
Constant
(0.284)
3.079∗∗∗
2.786∗∗∗
(0.111)
(0.533)
Lambda (hazard)
0.0497
(0.000217)
0.825∗∗∗ (0.0726) 0.609∗∗∗ (0.0660) 0.300∗∗∗ (0.0451) 0.116∗∗ (0.0468)
Observations
(0.0658) 0.608∗∗∗ (0.0602) 0.300∗∗∗ (0.0409) 0.116∗∗∗ (0.0424) −0.0394
(0.131)
(0.418)
5.728∗∗∗ (0.0694)
6.380∗∗∗ (0.476) −0.0282
(0.158)
Firm fixed effect
0.825∗∗∗
−0.0885
−0.487∗∗∗
IV var
0.0497
(0.230)
Journey
Distance
Journey2
Distance2
No
Y es
No
Y es
1771
1771
1459
1459
26
Table 15: Probit of current firm-financed training outside the firm, with IV variables and with and without firm fixed-effects (Kenya) Completed years of schooling
0.0959∗∗∗ (0.0265)
(Senegal)
(Senegal)
0.0226
0.0229
(0.0146)
−0.00286 (0.0196)
(0.0110)
(0.0112)
Tenure in the current firm
−0.00265
−0.0237∗∗
−0.0241∗∗
(0.0148)
(0.0108)
(0.0103)
0.280
0.00121
0.0113
Woman
(0.298) Not permanent full time job
0.210 (0.305)
Managers
Professionals
Non production workers
Journey to go to workplace (hours)
(0.264) −0.416 (0.294)
(0.268) −0.426 (0.299)
0.625
0.347
0.372
(0.386)
(0.381)
0.872∗∗
0.248
0.277
(0.307)
(0.305)
0.677
0.221
0.226
(0.432)
(0.197)
(0.195)
0.275
0.282
0.288
(0.494)
(0.292)
(0.299)
0.324
0.715∗∗
(0.902) Squarred of journey
−0.000339
(0.522)
(0.443) Skilled workers
0.000141
(0.0143)
Actual experience off the firm
(0.312) −0.272∗∗
−0.532 (0.748)
Trade Union member
Distance to workplace
(0.131) 0.157
0.149
(0.191)
(0.181)
0.0639∗∗ (0.0291)
Squarred of distance
−0.00216∗ (0.00123)
Constant
−4.660∗∗∗ (0.702)
Observations Pseudo
R2
1771
−2.670∗∗∗ (0.273) 1459
0.398
Regressions are robust to clustering
27
0.349
−2.657∗∗∗ (0.351) 1459 0.348
Table 16: Treatment regressions of current outside the firm training, (with and without firm fixed-effects) (Kenya) Completed years of schooling
0.0639
(Kenya) ∗∗∗
(0.0138) Squared of years of schooling Tenure in the current firm
Actual experience off the firm
Woman Not permanent full time job
Trade Union member
Managers
(0.000427)
(0.000385)
0.0320∗∗∗
0.0376∗∗∗
0.0360∗∗∗
−0.000436∗∗∗
(0.000208)
0.0463∗∗∗
0.0465∗∗∗
0.0261∗∗∗
0.0262∗∗∗
(0.00575)
−0.00100∗∗∗
−0.00101∗∗∗
(0.000261)
(0.000241)
−0.0376
−0.0337
(0.000386) 0.0358∗∗∗ (0.00679) −0.000548∗∗∗ (0.000210) 0.0260∗∗∗
(0.00607)
(0.00558)
(0.00561)
−0.000317
−0.000325
−0.000321
(0.000239)
(0.000218)
(0.000219)
0.0496
0.0538
0.0533
(0.0377)
(0.0453)
(0.0437)
(0.0440)
−0.541∗∗∗
−0.540∗∗∗
−0.327∗∗∗
−0.354∗∗∗
−0.355∗∗∗
(0.0479)
(0.0445)
(0.0533)
(0.0522)
(0.0525)
−0.361∗∗∗
−0.366∗∗∗
−0.119∗∗
−0.109∗∗
−0.108∗∗
(0.0427)
(0.0398)
(0.0535)
(0.0503)
(0.0507)
1.152∗∗∗ 0.850∗∗∗ 0.246∗∗∗ 0.192∗∗∗
−0.109 3.080∗∗∗
Lambda (hazard)
1.155∗∗∗ (0.0587) 0.867∗∗∗ (0.0699) 0.249∗∗∗ (0.0402) 0.191∗∗∗ (0.0441) −0.638∗ (0.373) 2.813∗∗∗ (0.525)
0.826∗∗∗ (0.0726) 0.611∗∗∗ (0.0660) 0.301∗∗∗ (0.0451) 0.117∗∗ (0.0468) −0.0927 (0.0992) 5.729∗∗∗ (0.0695)
IV var
0.846∗∗∗ (0.0705) 0.623∗∗∗ (0.0635) 0.313∗∗∗ (0.0438) 0.131∗∗∗
0.849∗∗∗ (0.0710) 0.625∗∗∗ (0.0640) 0.314∗∗∗ (0.0441) 0.132∗∗∗
(0.0455)
(0.0458)
−0.927∗∗∗
−0.959∗∗∗
(0.343) 6.389∗∗∗ (0.491) 0.504∗∗
0.291 (0.195)
Observations
0.00128∗∗∗
(0.0405)
(0.111)
Firm fixed effect
−0.000553∗∗∗
(0.000227)
(0.124) Constant
−0.000555∗∗
(0.00674)
(0.000155)
(0.0474) Current training
−0.000442∗∗∗
(0.00729)
0.00127∗∗∗
(0.000167)
(0.0432) Non production workers
(0.00801)
(0.000475)
(0.0741) Skilled workers
(0.00798)
0.0318∗∗∗
(0.00532)
0.00118∗∗∗
0.0158∗∗
0.0159
(0.000508)
(0.0631) Professionals
(0.00878)
(Senegal) ∗∗
−0.000903∗
(0.00621) Squared of experience
(0.0128)
0.0159
(Senegal) ∗
−0.00102∗∗
(0.00574) Squared of tenure in the firm
0.0622
(Senegal) ∗∗∗
(0.198)
(0.343) 6.393∗∗∗ (0.493) 0.526∗∗∗ (0.199)
Distance
Distance
Journey
Distance2
Distance2
Journey2
No
Y es
No
Y es
Y es
1771
1771
1459
1459
1459
28
and could jeopardize the potential impact of IV on earnings. Twostep regressions are not very robust for IV variables runnings that are not really correlated to output, although the self-financed trainings seem to be more lucrative (in term of salary) in Senegal (with no significativity).
5
Discussion and perspectives
We controled for the establishment effect (or the employer’s propension to offer trainings if it is not idiosyncratically distributed within the employees), elaborating a firm fixed-effects model under the hypothesis that the treatment model corrects for both the employer and employee selection. This is a strong hypothesis because an employer generally does not offer training with the same probability to every employee, but such a statement is a bit hard to control for. It is impossible to observe the same individual in both states at a given time, but we could make propensity scores and evaluate such an effect by one of the matching methods5 to improve our argumentation.In the purpose of identifying a significative wage premiums and assign it to training we could also use production fonction estimation as Hellerstein et al., 1999 (and applied to Africa by Biesenbroeck, 2007). We must underline that trade union members are less payed, ceteris paribus, it could be explained by the fact that they are discriminated. This explanation seems to be coherent whis the higher proportion of declared unionised workers in Senegal, where such a discrimination seems to have a more little amplitude. Firm financed trainings are anyway very rare in the two countries and even in most of the RPED countries we studied in addition6 , fact that limits the scope of our study. The possible share of costs between employers and employees stresses the existence of a burden for workers, and become a counter incentive for them given their budgetary constraint. Such a conclusion underlines the importance of a country level program which facilitates training programs initiatives for the employer and for the employees. Based on the decreasing returns (which are more true in Kenya than in Senegal) to education we can assert that less educated workers are in a higher need for trainings, and this reasoning is more true for manufacturing, 5 6
Which give a (maybe too) large role to observables In Benin, Madagascar, Mauritius, Morocco and Uganda
29
Table 17: Earnings with endogeneous current self-financed training, with and without firm fixed-effects in Kenya OLS Completed years of schooling
Sq. of years of schooling
Tenure in the current firm
Actual experience off the firm
−0.0510∗
−0.175∗∗∗
−0.0511∗
(0.0307)
(0.0338)
(0.0283)
0.0230∗∗∗
0.00872∗∗∗ (0.00238)
0.0229∗∗∗ (0.00266)
0.00872∗∗∗ (0.00220)
Treatment ML −0.0521 (0.0392) 0.00879∗∗∗ (0.00310)
−0.000572∗∗∗
−0.000227∗∗∗
−0.000574∗∗∗
−0.000227∗∗∗
−0.000228∗∗∗
(0.0000619)
(0.0000540)
(0.0000611)
(0.0000499)
(0.0000731)
0.0495∗∗∗
0.0368∗∗∗
0.0519∗∗∗
0.0366∗∗∗
−0.000750∗∗∗
(0.00580) −0.000545∗∗∗
(0.00682) −0.000764∗∗∗
(0.00537) −0.000547∗∗∗
(0.000201)
(0.000168)
(0.000201)
(0.000155)
0.0472∗∗∗
0.0461∗∗∗
0.0492∗∗∗
0.0459∗∗∗
(0.00759) Sq. of experience off the firm
Treatment Two
(0.0338)
(0.00667) Sq. of tenure in the current firm
Treatment Two
−0.177∗∗∗
(0.00268) Cubic years of schooling
OLS
−0.00102∗∗∗ (0.000324)
(0.00617) −0.000970∗∗∗ (0.000259)
(0.00775) −0.00104∗∗∗ (0.000326)
(0.00571) −0.000968∗∗∗ (0.000239)
0.0365∗∗∗ (0.00885) −0.000550∗ (0.000306) 0.0456∗∗∗ (0.00679) −0.000966∗∗∗ (0.000255)
Woman
−0.0506 (0.0500)
(0.0402)
(0.0507)
(0.0371)
(0.0374)
Not permanent full time job
−0.502∗∗∗
−0.519∗∗∗
−0.466∗∗∗
−0.522∗∗∗
−0.526∗∗∗
(0.0519)
(0.0479)
(0.0574)
(0.0446)
(0.0596)
Trade Union member
−0.178∗∗∗
−0.346∗∗∗
−0.164∗∗∗
−0.347∗∗∗
−0.349∗∗∗
(0.0455)
(0.0424)
(0.0470)
(0.0393)
(0.0485)
Managers
1.055∗∗∗ (0.0796)
Professionals
0.951∗∗∗ (0.0928)
Skilled workers
0.162∗∗∗ (0.0512)
Non production workers
0.152∗∗∗ (0.0568)
Self-financed current training
0.101 (0.102)
Constant
3.349∗∗∗ (0.148)
−0.0439
1.087∗∗∗ (0.0642) 0.782∗∗∗ (0.0747) 0.226∗∗∗ (0.0431) 0.175∗∗∗ (0.0472)
−0.0513
1.019∗∗∗ (0.0840) 0.932∗∗∗ (0.0947) 0.159∗∗∗ (0.0519) 0.143∗∗ (0.0578)
0.191∗∗
0.904∗
(0.0821)
(0.529)
3.411∗∗∗ (0.138)
Completed years of schooling
3.298∗∗∗ (0.152) 0.0453
−0.0440
1.092∗∗∗ (0.0600) 0.786∗∗∗ (0.0694) 0.227∗∗∗ (0.0398) 0.177∗∗∗ (0.0437)
1.098∗∗∗ (0.0837) 0.792∗∗∗ (0.0923) 0.228∗∗∗ (0.0481) 0.179∗∗∗ (0.0554)
0.117
−0.00635
(0.172)
(0.274)
3.154∗∗∗ (0.524)
∗∗
−0.0441
0.0870
3.162∗∗∗ (0.193)
∗∗∗
0.0809∗∗
Actual experience off the firm
−0.0383∗∗∗
−0.0633∗∗∗
−0.0711∗∗
Tenure in the current firm
−0.0471∗∗∗
−0.0617∗∗∗
−0.0632
Woman
−0.0296
−0.0274
Not permanent full time job
−1.147∗∗∗
−1.876∗∗∗
−1.795∗∗∗
Trade Union member
−0.365∗
−0.691∗
−0.602
0.0116
Managers
0.370
0.980∗∗
1.045∗
Professionals
0.174
0.655
0.704
Skilled workers
0.0109
0.0218
0.0567
Non production workers
0.0878
0.276
0.223
0.838∗
Journey (hours)
−0.157
Journey by distance
−0.000986
distance
0.00783 ∗∗∗
Constant
−1.857
Lambda (hazard)
−0.331 (0.259)
−0.0759∗∗∗ 0.0710∗∗∗ −4.856
∗∗∗
0.850∗ −0.0726∗∗ 0.0624∗ −4.740∗∗∗
0.0503 (0.104)
Constant (athrho)
0.269 (0.355) −0.684∗∗∗
Constant (lnsigma)
(0.0434) Firm fixed-effect(2 steps) Observations
No
Y es
No
Y es
Y es
1771
1771
1771
1771
1771
30
Table 18: Earnings with endogeneous current self-financed training, with and without firm fixed-effects in Senegal OLS Completed years of schooling
−0.0349∗ (0.0194)
Squared of years of schooling
Tenure in the current firm
Actual experience off the firm Squared of experience off the firm
Not permanent full time job
Trade Union member
0.00574∗∗∗ (0.00177)
(0.0000540)
(0.0000611)
(0.0000488)
(0.0000529)
0.0383∗∗∗
0.0371∗∗∗
0.0372∗∗∗
(0.00728)
(0.00776)
(0.00663)
−0.000457∗
−0.000565∗∗
−0.000421∗
(0.000248)
(0.000227)
(0.000250)
(0.000206)
0.0284∗∗∗
0.0276∗∗∗
0.0265∗∗∗
0.0274∗∗∗
−0.000537∗∗∗
0.0378∗∗∗ (0.00854) −0.000553∗∗ (0.000261) 0.0275∗∗∗
(0.00690)
(0.00607)
(0.00709)
(0.00552)
(0.00603)
−0.000546∗
−0.000368
−0.000532∗
−0.000377∗
−0.000374
(0.000285)
(0.000217)
(0.000239) 0.0345
−0.0416
−0.0536
(0.000239)
0.0322
0.0332
(0.0527)
(0.0456)
(0.0549)
(0.0415)
(0.0488)
−0.387∗∗∗
−0.318∗∗∗
−0.367∗∗∗
−0.309∗∗∗
−0.313∗∗∗
(0.0524)
(0.0533)
(0.0550)
(0.0488)
0.182∗∗∗
−0.121∗∗ (0.0534)
0.806∗∗∗
0.804∗∗∗ (0.0729)
0.726∗∗∗
0.575∗∗∗ (0.0668)
0.286∗∗∗
0.281∗∗∗ (0.0456)
0.273∗∗∗
0.105∗∗
(0.0464) 0.804∗∗∗ (0.0868) 0.808∗∗∗ (0.0846) 0.295∗∗∗ (0.0529) 0.278∗∗∗ (0.0545)
0.0855
−1.101∗∗
(0.0802)
(0.457)
5.560∗∗∗ (0.0753)
0.209∗∗∗
(0.0469)
0.00297 (0.0950)
Constant
(0.00164)
(0.0151)
0.0394∗∗∗
(0.0524) Self-financed current training
0.00576∗∗∗
−0.0214
(0.0000629)
(0.0508) Non production workers
(0.00206)
(0.0151)
−0.000138∗∗∗
(0.0748) Skilled workers
0.00793∗∗∗
−0.0214
−0.000137∗∗∗
(0.0835) Professionals
(0.00181)
(0.0195)
Treatment ML
−0.000171∗∗∗
(0.0434) Managers
0.00574∗∗∗
−0.0340∗
Treatment Two
−0.000140∗∗∗
(0.000280) Woman
(0.0167)
Treatment Two
−0.000187∗∗∗
(0.00761) Squared of tenure in the firm
−0.0214
0.00801∗∗∗ (0.00209)
Cubic year of schooling
OLS
5.773∗∗∗ (0.0718)
Completed years of schooling
5.572∗∗∗ (0.0776) 0.0821
(0.0486) 0.803∗∗∗ (0.0663) 0.597∗∗∗ (0.0621) 0.284∗∗∗ (0.0415)
∗∗∗
(0.0645) −0.119∗ (0.0614) 0.804∗∗∗ (0.0827) 0.587∗∗∗ (0.0660) 0.283∗∗∗ (0.0473)
0.106∗∗
0.106∗∗
(0.0426)
(0.0502)
−0.210
−0.0748
(0.191)
−0.0271∗
Experience off the firm
−0.116∗∗
5.007∗∗∗ (0.476) 0.120
∗∗∗
−0.0282
(0.156) 5.001∗∗∗ (0.123) 0.124∗∗∗ −0.0285
Tenure in the current firm
−0.0162
−0.00957
−0.0119
Woman
−0.0834
−0.0735
−0.0591
Not permanent full time job
0.299
0.515∗
0.529∗∗
Trade Union member
0.393∗∗
0.414
0.472∗∗
Managers
0.106
0.190
0.158
Professionals
0.597∗∗
0.695∗
0.654∗
Skilled workers
0.143
0.462
0.415
Non production workers
0.0931
0.282
0.248
−0.0184
−0.0190
Distance
−0.0102
Journey by distance
−0.000180
0.00459
0.00429
0.0677
0.00769
0.0428
−2.900∗∗∗
−4.467∗∗∗
−4.523∗∗∗
Time to go to workplace (hours) Constant
(0.321) 0.535∗∗
Lambda (hazard)
(0.216)
(0.580)
(0.521)
0.178∗ (0.106)
Constant (athrho)
0.212∗
Constant (lnsigma)
−0.771∗∗∗
Firm fixed-effect(2 steps) Observations
No
Y es
No
Y es
Y es
1459
1459
1459
1459
1459
31
where less educated workers play a major role. Finally, we can assert that public policy must create more incentives for firms to invest in low level trainings, for lower educated people particularly in local, little sized and non-exporting firms; and also make it easier for firms to invest in a low level of training’s specialization, even if it could be a cost for workers (by temporarily reducing the salaries and wages levels for example).
References [1]
J. Abowd, F. Kramarz [1999], “Econometric Analysis of Linked EmployerEmployee Data”, Labour Economics, No. 6,53-74.
[2]
J. Abowd, F. Kramarz, D. Margolis [1999], “High-Wage Workers and High-Wage Firms”, Econometrica, Vol. 67, No. 2, 251-333.
[3]
Acemoglu D. and Pischke J-S. [1999], “Beyond Becker: Training in imperfect Labor Market”, The Economic Journal, Vol. 109, pp. 112-142.
[4]
Biesebroeck J. Van [2007], “Wage and Productivity Premiums in Sub-Saharan Africa”, W.P. N 291, University of Toronto.
[5]
Barron J. M., Black D. A., Loewenstein D. M. [1989], “Job Matching and On-the-Job Training”, Journal of Labor Economics, Vol. 7, No. 1, pp. 1-19
[6]
Biggs T. [1995], “Training, Technology and Firm-Level Efficiency in Sub-Saharan Africa”, RPED paper No. 48.
[7]
Biggs T., Shah M. & Srivastava P. [1995], “Capabilities and Learning in African Enterprises”, The World Bank, Private Sector Development Department.
[8]
Bishop J. H. [1989], “On-the-job training and new hires”, CAHRS W.P. No. 89-10, Cornell University.
[9]
Booth, Francesconi & Zoega [1999], “Training, Rent-Sharing and Unions”, CEPR discussion Paper.
32
[10]
Destre G., Levy-Garboua L. & Sollogoub M. [2005], “On-the-job learning ´ and earnings”, Cahiers de la Maison des Sciences Economiques bla05022, Universit´e Panth?on-Sorbonne (Paris 1), revised. Fougre et al., 2001 More for Job Switchers Goux et Maurin E. [2000] Heckman, J. J. [1979], “ Sample Selection Bias as a Specification Error ”, Econometrica, Vol. 47, No. 1, pp. 153-161. Hellerstein K., Neumark D., Troske K. [1999], “Wages, Productivity, and Workers Characteristics: Evidences from a Plant-Level Production function and Wage Equations”, Journal of Labor Economics, Vol. 17, N3. Kristensen, N. & Verner, D. [2005], “ Labor Market Distorsion in Cˆote d’Ivoire: Analyses of Employer-Employee Data from The Manufacturing Sector ”, World Bank working paper, No. 3771. J Mincer & B Jovanovic [1981], Studies in Labor Markets. Kuckulenz A. & Zwick T. [2003], “The Impact of Training on Earnings Differences Between Participant Groups and Training Forms”, ZEW D. P. No. 03-57. Leuven E. [2004], “A review of the wage returns to private sector training”, OECD W.P. The Impact of Training on Earnings Differences Between Participant Groups and Training Forms” Leuven E. An alternative approach to estimate the wage returns to private-sector training Journal of Applied Econometrics. [2008]. forthcoming. [with Hessel Oosterbeek). Leuven E., Lindahl M., Oosterbeek H. & Webbink D. [2007], “The effect of extra funding for disadvantaged students on achievement”, Review of Economics and Statistics, Vol. 89, No. 4, pp. 721-736. Leuven E. The economics of private-sector training: A review of the literature Journal of Economic Surveys 19[1). [2005]. 91–111. Reprinted in: Recent Developments in 33
the Economics of Training. International Library of Critical Writings in Economics, editor: F. Green. Edward Elgar Publishing, 2007. Nordman C.[2003], “Formal and Informal Training: a Critical Review of its Economic Features and Measures”, preliminary WP, University of Oxford: Modelling Vocational Learning System for the 21st Century. Nordman C., Hayward G. [2007], “Returns to On-the-Job Training: Do Skill Usage, Tasks and Workstation Matter?”, WP. [38]
Rosholm M., Nielsen H., Dabalen A. [2007], “Evaluation of Training in African Enterprises”, Journal of Development Economics N 84, pp. 310-329. Shone P. [2004], “Firm-financed training: Firm-specific or general skills?”, Empirical Economics, Vol. 29, No. 4, pp. 885-900. Sicilian P. [2001], “On-The-Job Training and Starting Wages”, Journal of Labor Research, Vol. 22, No. 22, pp. 809-816.
[39]
R. Walther [2003], “Les m´ecanismes du financement de la formation professionnelle”, AFD.
[40]
Weiss Y. and Rubinstein Y. [2007], “Post Schooling Wage Growth: Investment, Search and Learning”, Handbook of the Economics of Education, Vol. 1, Chapter 1.
[41]
Wooldridge M. [2000], “Introductory Econometrics: A Modern Approach”, SouthWestern.
[42]
Yellen, J. L. [1984], “Efficiency Wage Models of Unemployment, ” The American Economic Review, Vol. 74, No. 2, pp. 200-205.
34