Does SENA matter? Some preliminary results on the impact of Colombia's national training system on earnings

THE WORLD BANK Discussion Paper EDUCATION AND TRAINING SERIES Report No. EDT2 Does SENA Matter? Some Preliminary Results on the Impact of Colombia's National Training System on Earnings Emmanuel Jimenez and Bemardo Kugler August 1985 Education and Training Department Operations Policy Staff The views presented here are those of the author, and they should not be interpreted as reflecting those of the World- Bank.  Discussion Paper Education and Training Series Report No. EDT2 DOES SENA MATTER? SOME PRELIMINARY RESULTS ON THE IMPACT OF COLOMBIA'S NATIONAL TRAINING SYSTEM ON EARNINGS Emmanuel Jimenez B. Kugler Research Division Education and Training Department August, 1985 The World Bank does not accept responsibility for the views expressed herein which are those of the author and should not be attributed to the World Bank or to its affil-iated organizations. The findings, inter- pretations, and conclusions are the results of research or analysis supported by the Bank; they do not necessarily represent official policy of the Bank. The designations employed, the presentation of material, and any maps used in this document are solely for the convenience of the reader and do not imply the expression of any opinion whatsoever on the part of the -World Bank or its affiliates concerning the legal status of any country, territory, city, area, or of its authorities, or concerning the delimitation of its boundaries, or national affiliation.  ABSTRACT Previous studies of Colombia's Servicio Nacional de Aprendizaje (SENA), one of the developing world's more extensive national job training programs, have been only partially useful because of limited data. This study contributes to the existing literature by utilizing a more detailed data set collected by SENA from 1979 to 1981 and by addressing a broader range of issues. This data set contains extensive information on earnings and socioeconomic characteristics of SENA graduates and a control group of non-SENA workers. The ultimate aim of the research is to provide guidelines for future public investment in alternative modes of education and training. This first paper attempts to quantify the impact of SENA on earnings. The basic methodology is to compare the earnings of SENA graduates with those of nongraduates, with experience, schooling and socioeconomic background held constant. Although it is premature to draw any strong policy recommendations, some general conclusions have emerged from the preliminary analysis: * SENA has a positive effect on earnings. e Impact on earnings varies according to SENA course length. * SENA complements other human capital investments. * Complementarity depends on SENA course length.  I. INTBODUCTION It is widely recognized that investment in human capital is a prerequisite for economic development. The formation of skills through vocational or in-service training is an important and growing component of such investment in developing countries. The magnitudes devoted to specialized training programs outside of the formal school system is a reflection of a perceived gap in the spectrum of skills imparted by academic schools (too general) and on-the-job experience (too specific), relative to the skills demanded by a changing labor market. Nowhere in developing countries is the commitment to vocational and technical training more established than in Latin America, where such investment is termed "professional formation,." to emphasize on the one hand, its link with the labor market, and on the other, its supposedly formative character. By 1975, all countries in the region, with the exception of El Salvador and the Caribbean countries, had established national training programs. (Castro, Kugler and Reyes). Despite the large investments implied above, the economic impact of training institutions and methods has traditionally received little research attention-in developing countries, especially in comparison with the studies on formal education. (Metcalf, Lee). As a result, key issues remain fertile areas for analysis, such as the magnitude of the impact of training on workers' incomes and their occupational mobility; and the rate of return to alternative training investments in human capital. The program  -2- of research that this paper initiates will contribute to redressing the imbalance through an analysis of these key issues for one of the oldest and largest of these schemes, Colombia's Servicio Nacional de Aprendizaje (SENA). SENA was instituted in 1957, spends an equivalent of about one tenth of the public budget for education, is spread over the whole country and has trained about 15% of Colombia's urban labor force. The Board of Directors of Sena seats representatives of the main production associations and labor unions and government representatives of the labor, education and planning ministries. In Colombia, SENA is considered to be a successfully operated public venture.l/ There have been several evaluations, both in/house and external, of the SENA program (SENA-Holanda, Puryear, Maldonado, Reyes and Gomez, Gomez and Librero.s). While these studies have been helpful in outlining an. initial view of SENA's effectiveness and efficiency, they have not generated a robust set of results which would allow the derivation of general policy implications. Some of the studies have employed a relatively rigorous methodology on a limited data set (such as Puryear and Maldonado). Another makes some limited use of pieces of information on SENA within a general sample (Reyes and Gomez). Gomez and Libreros, utilize a very detailed data set to answer a limited set of questions. The aim of I/ A brief institutional description of SENA is found in the appendix.  -3- this paper is to extend the analysis of the detailed data set used by Gomez and Libreros to a wider range of issues and with a more extensive methodological tool kit. In particular, this paper will address the following questions: (1) What is the effect of different types of SENA courses on earnings? (2) Are different types of SENA courses complementary to more traditional forms of academic education in influencing workers' earnings? (3) Are different types of SENA courses complementary to on- the-job experience in influencing workers' earnings? Aside from having direct policy implications, the conclusions of this paper will be used for subsequent work on estimating the rate of return to SENA investments and on the consequences of SENA for general job mobility. The second section will outline the methodology and the third, the data to be used. The fourth section will discuss the main empirical results and relate them to previous findings. The fifth and final section will summarize the paper and discuss the future work program.  - 4 - II. METHODOLOGICAL ISSUES The basic methodology is a simple and well known one: to compare the earnings (y) of SENA graduates with those of non-graduates from a cross-section data base. However, in order to infer that such a comparison reflects the impact of SENA-type vocational training, it is necessary to account for other variables which may also affect earnings. The most important measurable ones in the human capital model are schooling (s) and experience (x), both measured in terms of years. In addition, there may be other variables (u) which may not be adequately measured by the data but which may nonetheless affect earnings, such as innate ability. The relationship between earnings and its determinants can be summarized in a general earnings function of the form:. y f(s,x,v,u) (1) where v is an index of participation in a SENA vocational training course. In the human capital model developed by Mincer (1974) and used widely in most earnings functions analyses, equation (1) is specified as: ln y X ao + ai s + a2 x + a3 x2 + a4 v + u, (2)  -5- Equation (2) can then be estimated by standard statistical techniques, such as ordinary least squares, to obtain estimates of the a-parameters. The value of the coefficient of v, a4, can be interpreted as the unbiased effect of SENA on earnings, provided the following additional assumptions are valid: (a) There is only one type of SENA course so that v = I for participants and is equal to zero otherwise. (b) The impact of schooling and experience on earnings is the same for participants and non-participants in SENA. (c) The variable u can be assumed to be a random normal variable which is sytemtically uncorrelated with any of the other explanatory variables in (2). Complications arise when any of these assumptions are violated, as they are likely to be in the sample to be examined. Adjustments to the basic model in (2) would then need to be accommodated. Different types of SENA courses: Since one of the main purposes of this paper is to estimate the differential impact of alternative types of SENA courses, assumption (a) is violated. The model (2) can be easily adjusted by redefining the variable, v. There are many types of SENA courses (up to 1,000). A natural division would be to divide them into apprenticeship, promotional, complementary, qualifying and mobile courses. (See Appendix 1). These can be further aggregated into short (the last two) and long (the first three) types. In this paper, v is simply interpreted as a vector of indices of participation in short and long types of SENA courses.  -6- Different earnings functions for sample subgroups: It may not be valid to restrict the coefficients of the earnings function (2) for participants and non-participants, as assumed in (b). For example, the impact of schooling on earnings may be greater (or less) for SENA graduates. To test for this possibility in the sample, statistical methods are used to respecify (2) by allowing interaction terms between v and the other variables. Biases in the error term: This potentially serious problem can arise because there may be other determinants of earnings which cannot be measured adequately but which may nevertheless have an effect on the way that training influences income. Two possible candidates are innate ability and socio-economic background. If SENA is selective in choosing trainees, then, only-the most able are picked to attend. But innate ability would also affect earnings positively, either directly, or because ability is positively correlated with schooling. If so, equation (2) would overestimate the  -7- SENA effect, since differences in earnings between attendees and non- attendees may be due, not only to having had SENA courses, but also to pure ability. Interviews with SENA staff indicate that this first situation described SENA's first decade. However, in the last ten or fifteen years, due perhaps to the rapid expansion of formal education, SENA has not been able to select its entrants. Enrollment has been driven by demand. Moreover, if SENA were considered as remedial and attracted the less able, equation (2) would understimate the SENA effect. There were no measures of innate ability available for this study. Thus, it is not possible to correct for this source of possible selectivity bias. This potential underestimate of the SENA effect is likely to be small in magnitude, according to other studies of earnings functions. (Griliches). Socioeconomic background may be important for a similar line of reasoning. Those from backgrounds with a higher socioeconomic status may be more able to bear the private costs of investing in both training and education. If so, then the impact of SENA will be overestimated. Socioeconomic background variables are used in this paper to partially correct for this source of bias by being included as explanatory variables in (2). However,t this is only a partial correction for the following reason: if, for example, only high status individuals had access  -8- to SENA because of failures in credit markets, then there would be no low-status individuals who would be observed to have been in SENA in the sample. This is called a 'truncation bias" in the economics literature. Correcting for it requires fairly sophisticated statistical techniques and will be done in a separate paper. The Estimating Equation: As a result of the adjustments required above, the final estimating equation is: ln y - ao + al s + a2 x + a3 x2 + a4 vl + as v2 + 11 aj bi + a12B + u- (3) i-6 where vl and v2 are indices for long and short courses and bi and B are socioeconomic background (SEB) variables.  -9- III. THE DATA The survey raised by Sena between 1979 and 1981 was designed to evaluate in detail the results of each Sena training activity in each of the regional subsidiaries" (See Gomez & Libreros, pp.253-255. More information on characteristics of the survey can be found in SENA- Holanda, pp.150-155). There were five different questionnaires. (i) The main survey was addressed to Sena alumni who were asked questions on their socioeconomic status, formal education history, SENA training history, labor force activity, past and current income, and attitudes towards SENA (see Appendix III for a copy of the questionnaire). A sample of the work places where SENA alumni were found was then drawn at random and three more surveys were administered. (ii) One was administered to each staff manager, who was asked for characteristics of the firm, with special references to its labor force. (iii) Another survey was filled by the supervisors of the identified alumni to assess the quality of the alumni's work. (iv) A fourth survey interviewed other workers, under the same supervisors as the SENA alumni, but who had not attended Sena. (This survey was similar to Appendix III). (v) One further survey which obtained the same information as (i) and (iv), was conducted for alumni of technical schools. The first (Sena alumnus) and fourth (similar worker) surveys have been made available to the Bank and have been used as the data base for the "experimental" and 'control' groups, respectively, in this paper. We have worked with the surveys for the Bogota subsidiary, which is the largest branch of Sena, but the methodology can be repeated with the surveys of any other subsidiary. Although the rate of finding alumni seems low, 50% for Bogota, the surveyors believe that the sample is  - 10 - representative for the subsidiaries covered within the selected cohorts, 1968-78 for 'long courses' and 1975-77 for "short courses". Apparently they expected such a retrieval rate, based on previous experience of Sena students follow-ups.)] The retrieval rates decrease for older cohorts and the effect is magnified if there are address changes. There still remains the question as to what extent alumni who are not found would statistically differ from those found, but without further information it appears reasonable to assume no special biases. Efforts have been made in order to have the control group as comparable as possible to the SENA group. However, some differences will always persist. Fortunately econometric methods have been devised to deal with this kind of problem (See Maddala or Heckman & Robb). Our control groutp is defined as 'employed labor force members performing similar jobs within similar firms", as SENA alumni. However, the following are under-represented in the control: nonworking, unemployed and selfemployed persons. Testing from resulting biases can be done only by referring to some complementary samples. This paper focuses only on male earnings functions. The reason is purely methodological. Selection biases in earnings functions are much more important for females than males because of the lower labor force participation rate of females. This rate is close to unity for males. 2/. In the appendix, Table Al, appears the total size of the survey, and in Table A2, some data related to the sampling and follow-up processes.  - 11 - Table 1 presents the means and standard deviations of the relevant variables to be used in the analysis. Some clear distinctions can be made between SENA and non-SENA populations. The earnings data are probably more reliable and complete than would usually be ayailable from micro-level surveys. The questionnaire included an exhaustive account of earnings from several jobs and from sources different from regular wages and salaries (see Appendix III for the questionnaire and a description on the use of the data for finding earnings). While there is no statistically significant difference in the mean earnings of the SENA and non-SENA groups, the former exhibit a higher variance. This can be partly be explained by the presence of self employed within the SENA subsample. Non-SENA persons seem to be slightly younger (Experience is defined as Age-School years-6. This variable appears to be tore reliable than other experience measures available from the questionnaire. In subsequent versions we will refine this specification to take length of SENA training explicitly into account.) They also have more formal schooling than former Sena students, who also appear to come from lower socioeconomic background households. Formal schooling is slightly higher for Non-Sena persons. This has traditionally been interpreted as the result of substitution between Sena and formal education. Socioeconomic background variables show Non-Sena persons with a higher status.  - 12 - TABLE 1. VARIABLES USED - Mean Values, Standard Deviations in Parentheses Code Meaning SENX NON-SENA Total ln y Natural logarithm of yearly 12.10 12.03 12.07 earnings (pesos). (.941) (.579) (.810) s Total number of years of formal 9.100 9.764 9.379 school attendance. (3.50) (3.95) (3.71) x Years of experience (measured 15.61 12.58 14.34 as age - schooling [s] - 6) (9.26) (9.28) (9.38) v Binary variable - 1, when 1.000 .5799 attended SENA (0.00) (.494) VI Binary variable - 1, when .7542 .4374 attended SENA's long courses. (.431) (.496) V2 Binary variable - 1, when .2458 .1425 attended SENA's short (.431) (.350) courses only. v3 Binary variable - 1, when .1381 .0801 attended SENA's industry (.345) (.272) apprentices courses. bl Binary variable - 1, when .2266 .1550 .1967 father had some (but not all) (.419) (.362) (.398) primary school. b2 Binary variable - 1, when .3669 .3270 .3503 father had complete (.482) (.470) (.477) primary (only). b3 Binary variable - 1, when .1586 .1800 .1676 father had some (but not all) (.366) (.384) (.374) secondary. b4 Binary variable - 1, when .0935 .1240 .1064 father had complete (.291) (.330) (.308) secondary (only). b5 Binary variable - 1, when .0446 .1080 .0710 father had more than complete (.207) (.310) (.257) .secondary. b6 Binary variable - 1, when .0205 .0390 .0283 father's education not (.142) (.194) (.166) reported.. B Index of father's occupation 5.267 4.954 5.136 prestige (based on the ILO 1- (1.26) (1.47) (1.36) digit occupational classification with 1 -iHgh Prestige. Number of-Cases 1412 1023 2435  - 13 - IV. RESULTS This section will present the main results of the estimation of the earnings functions. As stated earlier, only the male earnings functions are presented in the text. The reason is that female earnings functions are hypothesized to differ significantly from male earnings functions and that standard statistical techniques are unlikely to be sufficient to estimate the former because of selection biases. The difference between male and female earnings functions is statistically confirmed for the SENA sample. The details of this test are in Appendix Table A3. The analysis does not differentiate between the employed and self-employed, in order to determine the SENA effects on working males as a whole. However, since the control sample does not include the self-employed, some of the SENA effects may be attributable to the fact that SENA graduates may be self-employed. Indeed, the estimates do vary as suggested by the results of the regressions when self employed are omitted (see Appendix Table A.4). We expect to analyze this issue in another paper, which will utilize information from another (more general) sample on specific aspects of self employed.3/ 3/ SENA's overall effects are stronger than for the whole group, but when short and long courses are considered, their effects are milder when considering the limited sample.  - 14 - The main results of the multivariate regression analysis are presented in Table 2. Each of the questions posed in Section 1 is examined in turn. Does SENA have an effect on earnings? The first hypothesis to be tested is that, without distinguishing among types of course, SENA has a significant effect in shifting the earnings function of Colombian males, all other effects being restricted to be fixed. The v dummy variable of equation (2) in Table 2 is significant at the 90% level and shows that earnings for Sena alumni are about 5% higher on the average after controlling by human capital variables and socioeconomic background.4/ In order to compare our results with those previously found in other studies, some regressions with restricted samples of SENA students have been estimated and appear in Appendix Table A.5. Reyes and Gomez considered only students which had taken courses three months long ( or longer) within a labor force sample. They found a somewhat higher but less significant effect on earnings ( .187, t- 4.4) than we have ( .143, t- 5.3). Chow tests appearing in Table 3 show that it may be too restrictive to estimate earnings functions which are the same, except for a constant term, for SENA participants and non-participants. Equation (3) of Table 2 allows for more flexible functional forms. The results show strong significant differences on all the coefficients tested. The impact of SENA 4/ Log earnings differences can be read from the regression equations. Taking the exponential of the differences and subtracting one, the percentage difference is obtained.  - 15 - 7wD 2. RH Nan i aXF Ilff KR K F9J (t-sm I In II p esffIu rs) V