regressions that do not include variables to represent school diplomas, we ..... traditional diplomas to those with a technical diploma, if they have at least.
Wage Determinants in C6te d'Ivoire Jacques van der Gaag Wim Vijverberg
Public Disclosure Authorized Public Disclosure Authorized
Public Disclosure Authorized
Public Disclosure Authorized
Living Standards Measurement Study Working Paper No. 33
FLE
LSMSWorking Papers No. 1
LivingStandardsSurveysin DevelopingCountries
No. 2
Povertyand Living Standardsin Asia: An Overviewof the Main Resultsand Lessonsof Selected HouseholdSurveys
No. 3
MeasuringLevelsof Living in LatinAmerica:An Overviewof Main Problems
No. 4
TowardsMore EffectiveMeasurementof Levelsof Living,and Reviewof Work of the UnitedNations StatisticalOffice(UNSO)Relatedto Statisticsof Levelsof Living
No. 5
ConductingSurveysin DevelopingCountries: PracticalProblems and Experience in Brazil,Malaysia,and the Philippines
No. 6
HouseholdSurveyExperience in Africa
No. 7
Measurementof Welfare:Theoryand PracticalGuidelines
No. 8
EmploymentDatafor the Measurementof LivingStandards
No. 9
Incomeand Expenditure Surveysin DevelopingCountries:SampleDesignand Execution
No. 10
Reflections on the LSMS GroupMeeting
No.II
ThreeEssayson a Sri LankaHouseholdSurvey
No. 12
TheECIELStudy ofHousehold IncomeandConsumptionin UrbanLatinAmerica:An AnalyticalHistory
No. 13
NutritionandHealthStatusIndicators: Suggestions for Surveysof the Standardof Living in Developing Countries
No. 14
ChildSchoolingand the Measurementof Living Standards
No. 15
MeasuringHealthas a Componentof Living Standards
No. 16
Procedures for Collectingand AnalyzingMortality Datain LSMS
No. 17
The LaborMarket and SocialAccounting:A Frameworkof Data Presentation
No. 18
Time Use Data and the Living StandardsMeasurementStudy
No. 19
The ConceptualBasisof Measuresof HouseholdWelfareand Their ImpliedSurveyData Requirements
No. 20
StatisticalExperimentation for HouseholdSurveys:Two CaseStudiesof HongKong
No. 21
The Collectionof PriceDatafor the Measurementof LivingStandards
No. 22
HouseholdExpenditureSurveys:Some Methodological Issues
No. 23
CollectingPanelData in DevelopingCountries:Doesit Make Sense?
No. 24
MeasuringandAnalyzingLevelsof Livingin DevelopingCountries:An AnnotatedQuestionnaire
No. 25
The Demandfor UrbanHousingin the Ivory Coast
No. 26
The C6ted'IvoireLiving StandardsSurvey:Designand Implementation
No. 27
The Roleof Employmentand Earningsin Analyzing Levelsof Living:A GeneralMethodologywith Applicationsto Malaysiaand Tkailand
(Listcontinueson the insidebackcover)
Wage Determinants in C6te d'Ivoire
The Living Standards Measurement Study
The Living Standards Measurement Study (LSMS)was established by the World Bank in 1980 to explore ways of improving the type and quality of household data collected by statisticalofficesin developing countries.Its goal is to foster increaseduse of household data as a basis for policy decisionmaking.Specifically,the LSMSis working to develop new methods to monitor progress in raising levels of living, to identify the consequencesfor householdsof past and proposed government policies, and to improve communicationsbetween survey statisticians,analysts, and policy makers. The LSMSWorking Paper series was started to disseminateintermediateproducts from the LSMS.Publicationsin the seriesincludecriticalsurveys covering differentaspects of the LSMS data collection program and reports on improved methodologies for using Living Standards Survey (LSS)data. More recent publicationsrecommendspecificsurvey,questionnaireand data processingdesigns, and demonstratethe breadth of policy analysisthat can be carriedout using LSSdata.
LSMSWorking Paper Number 33
Wage Determinants in C6te d'Ivoire
Jacques van der Gaag Wim Vijverberg
The World Bank Washington, D.C., U.S.A.
Copyright © 1988 The International Bank for Reconstruction and Development/THE WORLD BANK 1818 H Street, N.W Washington, D.C. 20433, U.S.A. All rights reserved Manufactured in the United States of America First printing May 1988 This is a working paper published informally by the World Bank. To present the results of research with the least possible delay, the typescript has not been prepared in accordance with the procedures appropriate to formal printed texts, and the World Bank accepts no responsibility for errors. The findings, interpretations, and conclusions expressed in this paper are entirely those of the author(s) and should not be attributed in any manner to the World Bank, to its affiliated organizations, or to members of its Board of Executive Directors or the countries they represent. Any maps that accompany the text have been prepared solely for the convenience of readers; the designations and presentation of material in them do not imply the expression of any opinion whatsoever on the part of the World Bank, its affiliates, or its Board or member countries concerning the legal status of any country, territory, city, or area or of the authorities thereof or conceming the delimitation of its boundaries or its national affiliation. The material in this publication is copyrighted. Requests for permission to reproduce portions of it should be sent to Director, Publications Department at the address shown in the copyright notice above. The World Bank encourages dissemination of its work and will normally give permission promptly and, when the reproduction is for noncommercial purposes, without asking a fee. Permission to photocopy portions for classroom use is not required, though notification of such use having been made will be appreciated. The most recent World Bank publications are described in the catalog New Publications,a new edition of which is issued in the spring and fall of each year. The complete backdistof publications is shown in the annual Index of Publications,which contains an alphabetical title list and indexes of subjects, authors, and countries and regions; it is of value principally to libraries and institutional purchasers. The latest edition of each of these is available free of charge from the Publications Sales Unit, Department F, The World Bank, 1818 H Street, N.W., Washington, D.C. 20433, U.S.A., or from Publications, The World Bank, 66, avenue d'I6na, 75116 Paris, France. Jacques van der Gaag was acting chief of the Living Standards Unit of the Development Research Department, the World Bank, when this paper was written. Wim Vijverberg was assistant professor at the University of Texas at Dallas and worked as a consultant for the Living Standards Unit. Library of Congress Cataloging-in-Publication Data Gaag, J. van der. Wage determinants in Cote D'Ivoire / Jacques van der Gaag, Wim Vijverberg. p. cm. -- (LSMS workino paper, ISSN 0253-4517 ; no. 33) Includes bibliographies. ISBN 0-8213-1058-5 1. Wages--Ivory Coast--Effect of inflation on--Econometric models. I. Vijverberg, Wim P. M. II. Title. III. Series. HD5096.9.G3 1988 331.2'96668--dcl9 88-14095
v
ABSTRACT
The following two papers present an analysis of wage determinantsin C8te d'Ivoire, using the standard Mincerian framework. The data used stem from the C6te d'Ivoire Living Standards Survey, conducted in 1985. This survey collected information on 1,600 households. Our sample consists of the 514 individualsin these householdswho reported a wage earning job during the seven days prior to the interview. The first paper uses the total sample and addresses the issues of credentialism and returns to years of schooling,by type of school. In the regressions that do not include variables to represent school diplomas, we find an unusual result: rates of return to one year of additional schooling increase with the level of schooling: almost 12 percent for elementary education, but 20 percent for high school and 22 per cent for university education. This pattern suggests a severe shortage of Ivorians with higher education. The results by age-cohort (presented in Appendix 2) seem to underscore this point: younger workers receive higher returns than their older counterparts. Apparently, the development of the Ivorian economy, and the correspondingincrease in the demand for better educated workers, has outpaced the supply of such workers. When diplomas acquired are added to the equation, the high returns to an additionalyear of schooling decrease substantiallywhile the diplomas show a large impact on the wage rates (40-50 percent). This suggest the existence of a certain amount of credentialismin the Ivorian wage sector. However, a pure credentialisticspecificationof the wage equation is rejected by the data. Appendix 2 to the first paper reports results by cohort, sex, nationality and region. The second paper reports the results for public and private workers separately. However, rather than relying on standard OLS results for each group, we develop a model that recognizes the endogeneity of the sector choice. We find that the OLS results are likely to be seriously biased. The overall dominance of public over private wages (indicatedby the OLS results) vanishes once the selection process is taken into account. Public wages are still somewhat higher for better educated workers, but the private sector offers higher wages than the government to workers with little education. We finally show the importanceof school diplomas as determinantsfor obtaining a job in the public sector.
-
vi
-
ACKNOWIEDGMENTS
These two papers benefited greatly from numerous comments made by colleagues working in various parts of the World Bank: for the first, special thanks go to George Psacharopoulosfor his insightful and detailed comments. We like to thank Morty Stelcner for his useful comments and suggestionson the second paper.
We also thank Kalpana Mehra for her excellent programming
assistance, Carmen Martinez for cheerfully typing the various drafts and Brenda Rosa for her competencein putting together this final version.
-
vii
-
TABLE OF CONTENTS I.
WAGE DETERMINANTSIN COTE D'IVOIRE: EXPERIENCE, CREDENTIALS AND HUMANCAPITAL Page
1.
Introduction ..........................................
2.
The Basic Model; Returns to Schooling and Experience.....o........ 3 Table 2.1: Regression Results Basic Log Wage Equation ....... 5
3.
Returns to Schooling by Type of Schooling and Experience........O...7 Table 3.1: Years of Schooling and Experience by Type.........7 Table 3.2: Regression Results Extended Log Wage Equation ....8
4.
Do Diplomas Mte?....................................l Table 4.1 Diplomas Acquired..........................12 Table 4.2: Regression Results Extended Model, Including Variables on Diplomas Received .............................13
5.
Summary and Conclusion....*...*.*..**
.........................15
NOTES .................................. .........17 Appendix 1.............................................. 19 ASppendix 2 .............. ......*.0.*.0...................... Rveferences
..........................
II.
.........
O..*
......
0.......
.....
O..*
....
19
A SWITCHING REGRESSION MODEL FOR WAGE DETERMINANTS IN THE PUBLIC AND PRIVATE SECTORS OF A DEVELOPING COUNTRY
1,
Introduction.......
..... ...... ..*...*........................
2.
Who Gets the Public Job? .... ..................................... 30 Table 2.1: Definitions and Summary Statistics of the Variables by Public and Private Sector Employment* ........934
3.
Estimation
Rsls.................................. .0.0-....35
3.1 The Wage Equation................................. ......... 36 Table 3.1: FIML and OLS Estimates of Log Wage Equations for the Public and Private Sectors ......es ............. 37 Table 3.2: FIML Estimates of Log Wage Equations for the Public and Private Sector, with Restrictions on Diploma and Basic Skill Variables.. ............ .. .. ... 39 3.2 The Switching Equation........... *. ............. ........................ . 40 Table 3.3: Estimates of the Switching Equation..............41
-
viii
-
4. Summary and Discussion.............................................42 Table 4.1: Variationsin the Probabilityof Obtaining a Public Job and in the Differences in the Log Wage Offers, Mean Observed Log Wages and Overall Average Log Wages among Ivorian Employees.... . . . . . . .. . . .. . . .. .... ... .. 43 Table 4.2: Non-Wage Benefitsfor Private and Public
NOTES
............................................................ 47
References.....
..
.. o...
.... ..
o.
-*o *.
*.o ...
*o -o-49
I.
WAGE DETERMINANTS IN COTE D'IVOIRE: EXPERIENCE, CREDENTIALS AND HUMAN CAPITAL
1. Introduction This paper investigatesthe determinantsof wages in Cote d'Ivoire, using the well-known Mincerian framework. As such it presents yet another piece of evidence regarding the importance of education and experienceas determinantsof an individual'sproductivity,which, using a conventional assumption, is measured by the hourly wage rate. At the same time, it suffers from a number of shortcomingsusually present in the economic literatureon this topic. First of all, the sample consists of wage earners only. Thus, all results should be interpretedas conditionalupon having a wage earning job and extrapolationsto other economic activities (selfemployment,agriculture) should be avoided. Secondly, following the standardMincerian framework,the analysis takes the two key variables,education and experience,as given. A more structuralanalyses would, for instance, treat education as endogenous and investigatethe factors that determine schooling enrollment. Still despite these shortcomings,we expect the results presented to be of interest for a variety of reasons. The general importanceof the role of human capital in issues of developmentand distributionis sufficientlyrecognized. (See King, 1980 for an excellent summary of this issue. Also World Development Report, 1980.) Perhaps more importantly,our study is one of the few that uses data from a sub-Saharancountry. Of course, within the sub-Saharan region, Cote d'Ivoire is not a "typical" country. Its real growth in GDP increased 7.9 percent per year from 1965 to 1975, during which period economic growth was driven by rapid expansionof export crops. This exceptionalgrowth patternswas continued from 1975 to 1980 (6.4 percent GDP increase per year), but now fueled by rapid expansion of public investment. And though this
-2-
unprecedented"miracle ivorien"was followed by a severe recession, its per capita GNP of U.S.$720 (1983) is well above that of its neighbors. Still, to have estimates of, for example, returns to schooling for that part of the world, is likely to benefit the discussions regardinggrowth oriented policies in general, and public policies on education in particular. With respect to public policies on education, Cote d'Ivoire presents a particularlyinterestingcase: more than 40 percent of the recurrent governmentbudget is spent on education. Thus even though our analysis is only partial, it will shed some light on the rationale of this exceptionally strong public emphasis on investmentsin education. The outline of the paper is as follows: in the next section we discuss the data, give background informationon schooling and literacy in C6te d'Ivoire and present the estimation results of a basic Mincerian log wage equation. This basic model is extended in Section 3 to allow for differential effects by type of schooling and by type of experience. We also address the question of whether elementary schooling per se, or the acquired cognitive skills (as measured by literacy and numeracy) are the determining factors of wage differentials. In Section 4 we look at the relative importanceof years of schoolingversus diplomas obtained. Here we will enter the discussionson screeningand credentialismthat are put forward as alternativesto the human capital model. In Section 5 we discuss the results and conclude.!'
In Appendix 2 we present and briefly discuss re-estimatesof the general model of Section 4, by cohort, sex, nationalityand region.
-3-
2. The Basic Model; Returns to Schooling and Experience Data used in this study are drawn from the C6te d'Ivoire Living Standards Survey (CILSS). This multipurposesurvey, which aims at measuring socioeconomicfactors relevant to the living standards of Ivorian households, started in February 1985. During the first 12 month period, 1600 households are interviewed,688 in urban and 912 in rural areas. These 1600 households form a random sample of the Ivorian population. Non-Africanexpatriatesare excluded from the sample. In the second year 50 percent of the households will be reinterviewedand 800 new households will replace the other half of the sample. The survey is scheduled to be conducted on a permanent basis. The current analysis is based on first year data only. 1/ The vast majority of the economicallyactive populationin C6te d'Ivoire is self-employed,both in agriculturaland non-agricultural activities. Only a small percentageof the populationreports having a wage earning job. Not surprisingly,wage earners are concentratedin the urban areas: in the capital, Abidjan, 17.2% of the labor force are wage earners, 9.7% and 1.4% in other urban areas and in rural Cote d'Ivoire, respectively. The CILSS collects data on primary, secondary and tertiary jobs, using recall periods of one week and one year. In this paper we restrict the sample to all individualswho report a wage earning job as their primary activity during the past 7 days. 2/ Reported earnings (generallyreported per month) were divided by hours worked per day times days worked per month 3/ to obtain an hourly wage rate. Wages include the cash value of in-kind income. 4/ In our sample the total hourly compensationaverages CFA 836, with a standarderror of 1354. The natural log of the hourly wage will be the dependent variable throughoutthis study.
- 4 -
The analysis focuses on two sets of exogenous variables: education and experience. As stated in the introduction,education is of major public concern in C6te d'Ivoire, absorbing 42 percent of the recurrent government budget. Still the overall literacy rate is only 37.6 percent, but the data indicate that progress has been made. For younger cohorts, e.g. between 15 and 24 years of age, literacy ranges from 78.3 percent in the capital, to 59.3 percent in the villages. Years of schoolingaverages just 2.6 over the entire population. Younger cohorts have almost 7 years of education in Abidjan, but just 1.5 years in the villages. Given the national averages, our sample of wage earners shows a fairly high education level: 6.87 years. In addition the sample averages 1.03 years of technical training. The literacy rate is 74 percent. Clearly education encourages the decision to seek employment in the wage sector, underscoringour previouswarning to interpret the results as conditional upon being a wage earner. In the subsequentanalyses we will use various measures of experience. The most general one has been defined as age minus years of schooling (includingtechnical training) minus 5. Thus calculated,experience averages 21.33 years, in a sample with an average age of 33.20 years. The basic equation used to explain differences in the observed hourly wages reads:
2
ln Y = ao + a
S + a2 E + a3 E
with Y, hourly wage rate S, years of schooling E, experience.
- 5 -
Table 2.1 shows estimation results of this equation.
The schooling
variable is split into years of academic (i.e. general curriculum) schooling and years of technical training.
Nationality and sex are added as
Sixteen percent of the sample is non-Ivorian, nineteen percent
regressors.
of the wage earners are female.
Table 2.1:
Regression Results Basic Log Wage Equation; (T-Values in Parenthesis)
Intercept
3.363
(18.84)
NAT
, nationality; 0 = Ivorian, 1 = Other
-.120
(1.59)
SEX
, 0 = male, 1 = female
-.002
(.02)
YRSCH
,
.207
(22.76)
YRSTECH
, years of technical training
.113
(4.63)
GEXPER1
, general experience
.053
(4.38)
GEXPER1Q
, general experience squared, *1000
-.082
(.36)
years of schooling
R2 , adjusted
.585
The estimation results show a familiar picture: schooling and experience are important determinants of wage differentials. additional year of schooling are very high, 20 percent.
Returns to an
Psacharopoulos (1985)
reports an average of 13 percent for studies that used data from African countries.
These countries (Ethiopia, Kenya, Morocco and Tanzania) are, with
one exception (Morocco), considerably poorer than Cote d'Ivoire.
Furthermore,
Psacharopoulos shows that returns to education have a tendency to decline with economic development (as measured by growth in per capita income).
Thus the
- 6-
Ivorian results seem to be out of line with "average" rate-of-return estimates. Technical training shows an 11 percent return for each year of training, well below the result for academic schooling,as expected, but still of considerablemagnitude. The experience variables show an essentially flat 5 percent increase in the hourly wage rate for each year of experience. The coefficientsfor Nationalityand Sex are not significantlydifferent from zero. In the next Section we will show how these results hold up when years of schooling is differentiatedby level of schooling and when general experienceis broken down into two components: occupation-specificand other experience.
- 7-
3. Returns to Schooling by Type of Schooling and Experience The traditional schooling system in C6te d'Ivoire includes 6 years of elementary school, 4 years of junior high school, 3 years of senior high school and a university program. Table 3.1 presents the average years of schooling in our sample for each level of education. Of the 6.87 years of total schooling,only a very small part (.39 years) is universitytraining, while most of its (4.20 years) is elementaryeducation.
Table 3.1:
Years of Schoolingand Experienceby Type
Mean
Standard Deviation
YRSED-EL
, years of schooling,elementary school
4.20
2.67
YRSED-H1
, years of schooling,junior high school
1.73
1.84
YRSED-H2
, years of schooling,senior high school
.54
1.10
YRSED-UN
, years of schooling,university
.39
1.38
YRSTECH
, years of technical training
1.03
1.63
YRSAPP
, years of apprenticeship
.78
1.87
GEXPERM2
, general experienceunrelated to current occupation
11.76
9.34
8.93
8.05
EXPERM2
, experience in current occupation
The data allowed us to differentiatetotal experienceinto experience related to the current occupationand other general experience. Occupationspecific experience is broader than tenure on the current job, as it includes work experiencein previouslyheld jobs that have the same job description as the current one. The results indicatea fair amount of job mobility, with
more than 11 years of experience (out of a total of 21) not related to the current occupation.
We now reestimate the log wage equation to detect the
differential impact of type of schooling and type of experience.5/ As before, the experience measures are also included in quadratic form (EXPERM2Q and GEXPERM2Q).
Results are given in Table 3.2, column 1.
Table 3.2: Regression Results Extended Log Wage Equation (T-values in Parenthesis)
_______ _______
1 )
______(
(2)
Intercept
3.840
(21.12)
3.813
(20.90)
NAT
-.137
(1.34)
-.117
(1.15)
SEX
.021
(.23)
.021
(.23)
YRSED-EL
.119
(5.90)
.074
(2.10)
YRSED-Hl
.209
(6.35)
.208
(6.31)
YRSED-H2
.208
(4.25)
.208
(4.25)
YRSED-UN
.227
(6.79)
.227
(6.79)
RRR
-
.104
(1.53)
YRSTECH
.085
(3.35)
.085
(3.34)
YRSAPP
.001
(.08)
-.005
(.25)
EXPERM2
.112
(8.65)
.111
(8.58)
-1.973
(4.29)
-1.957
(4.26)
GEXPERM2
.021
(1.74)
.020
(1.67)
GEXPERM2Q * 1000
.105
(.35)
.122
(.41)
EXPERM2Q * 1000
R2 , Adjusted
_
.620
.621
- 9 -
Surprisingly,rates of return to one year of schooling increase with the level of schooling:almost 12 percent for elementaryeducation,but 20 percent for high school and 22 percent for university education. These results, which are estimated quite precisely, run counter to much of the evidence that the largest productivitydifferentialis between primary school graduates and illiterates,rather than between high school and primary school graduates. Of course, literacy and numeracy can be acquired without attending elementary school, while some persons with elementaryeducation may remain illiterate. We reestimatethe equation, controlling for literacy and numeracy as measured by ability to read, write and do simple arithmetic (the three R's). The variable, RRR, is zero if the individualcannot write or read or do arithmetic, and increaseswith I for each of the skills acquired. 61
The
estimation results (Table 3.2, column 2) show a positive effect of these cognitive skills on the wage rate, though the effect is not measured very precisely. 71 Note that the point estimate indicatesa 31 percent differentialbetween illiterates(RRR = 0) and those with all three skills (RRR = 3). 3/ At the same time returns to elementaryschoolingreduce from 12 percent to 7 percent per year. Thus the overall effect remains the same: completed elementary schooling (6 years) producesa 73 percent overall wage differential. Compared to those without formal schooling,but with reading and writing skills,however, the differentialis only 42 percent. All other estimates remain the same. The coefficientsfor experienceindicate that the overall flat 5 percent per year increase (estimatedin section 2) is a combinationof an 11 percent return on occupation-specificexperienceand 2 percent on general experience. Moreover, there is some curvature in the occupation-specific
-
10
-
experience profile; the peak is reached after almost 30 years, an EXPERM2 level reached by only 8 individuals is our sample. Thus, we find a relatively steep experiencecurve that flattens out over time. It is worthwhile to take a closer look at this experience profile. Many authors have shown that experience profiles are steeper for the educated workers, and have proposed various explanations(e.g. Knight and Sabot, 1983). To test for this, we interactedtotal years of education with the two experiencemeasures that enter linearly and quadraticallyin the equations. The results indeed indicate steeper experienceprofiles for those with more schooling,at least during the early years. 91 We finally note the importanceof distinguishingbetween occupational-specific and general experience. The results indicate that someone who looses his job and has to start a career in a new occupation, suffers an income loss during the early years, experiencesa fairly long period necessary to catch up and an income gain at the end of the working life span. Though this may suggest that job mobility pays at the end, present value calculations(with a 10 percent discount rate) always indicateda welfare loss for the person who changes professions. As mentioned above, the experiencedata show a fair amount of occupationalmobility. Our calculations indicate that this is likely to be the result of involuntaryjob losses, rather than a profitable strategy to maximize lifetime income.
-
11 -
4. Do Diplomas Matter? In the previous section we measured schoolingas years of education. Obviously, for the human capital model, output measures of the education process (skills acquired) are the preferred variables,but such variables are only rarely available. Boissiere et al. (1985), who do have test scores in addition to wage information,conclude that acquired skills rather than years of schoolingcause the observed wage differentials. Our results of Section 3 show the relative importanceof literacy and numeracy. While the importanceof cognitive skills supports the human capital model, evidence of a separate impact of school diplomas receivedmay lend credit to the screeningor credentialistexplanationsthat are often put forward as alternativesto the human capital model. (See for instance Layard and Psacharopoulos,1974, and Riley, 1979). Completed schooling can indicate ability and motivation, and thus schooling is used by employers to screen for these desirable attributes. Productivitydifferentialsassociatedwith differences in schooling are then thought to be caused by these basic attributes rather than by schooling investmentsin human capital. In the absence of data from explicit skills tests that accuratelymeasure human capital, we are not able to conduct a conclusivetest to distinguishbetween the human capital and the screeninghypothesis. However, by including in the equation informationon completed schooling (i.e. diplomas acquired) in addition to years of schooling,we can show the relative importanceof these variables in explainingwage differentials. If years of schooling can be viewed as a proxy for investmentin human capital,while diplomas are the "signals" to employers regardingability and motivation, then the regression
- 12 -
results show the relative merits of the human capital and the screening hypothesis. For these reasons we include in the regressionsa set of dummy variables indicatingwhether the individualhas one of the four following diplomas: elementaryschool, junior high school, any higher diploma 10/ and/or a technical diploma. Note that the first three diplomas are cumulative,e.g., if you have a junior high school diploma you also have an elementary school diploma. Those with a technical diploma may or may not have one of the other diplomas. Unfortunatelythe data do not contain enough informationto be absolutely sure about this. We have assigned the traditionaldiplomas to those with a technicaldiploma, if they have at least the appropriateyears of traditionalschooling to make that plausible. Table 4.1 gives summary statisticsof these new variables.
Table 4.1:
Diplomas Acquired
Standard
Mean
Deviation
DIP-EL, ElementarySchool
.311
.463
DIP-HI, Junior High School
.089
.285
DIP-UPP, Senior High School and Above
.042
.202
DIP-TECH, Technical Training
.313
.464
Regressionresults are presented in Table 4.2 column 1. To ease comparisonwe repeat (in column 2) the results of Section 3.
- 13
-
Table 4.2: Regression Results Extended Model, Including Variables on Diplomas Received (T-Values in Parenthesis)
(3)
(2)
(1)
3.859 (22.84)
Intercept
3.774
(21.07)
3.813
(20.90)
NAT
-.117
(1.15)
-.118
(1.15)
-.098
(.93)
SEX
.011
(.12)
.021
(.23)
.059
(.65)
YRSED-EL
.023
(.59)
.074
(2.10)
-
-
YRSED-Hl
.088
(2.07)
.208
(6.31)
-
-
YRSED-H2
-.032
(.31)
.208
(4.25)
-
-
YRSED-UN
.208
(5.96)
.227
(6.79)
-
-
RRR
.113
(1.70)
.104
(1.53)
.141
(3.05)
DIP-EL
.494
(3.05)
-
-
.753
(5.78)
DIP-Hl
.594
(3.60)
-
-
.840
(6.96)
DIP-UP
.536
(1.80)
-
-
.979
(7.55)
DI-TECH
-.011
(.10)
-
-
-.042
(.43)
YRSTECH
.072
(2.56)
.085
(3.39)
-
_
-.002
(.11)
-.005
(.25)
-
-
.107
(8.46)
.111
(8.58)
.111
(8.48)
-1.909
(4.23)
-1.957
(4.26)
-2.034
(4.36)
.026
(2.15)
.020
(1.67)
.016
(1.35)
.020
(.06)
.122
(.41)
.187
(.62)
YRSAPP EXPERM2 EXPERM2Q
1000
*
GEXPERM2 GEXPERM2Q
1000
*
R2 , Adjusted
.635
.621
.607
- 14 -
Comparing column (1) and (2) we see that the effect of years in elementary school is no longer significant,if an elementary school diploma is added to the equation. All that remains is the effect of cognitive skills (RRR, 11 percent) and the diploma (DIP-EL,49 percent). 11!
The effect of a
year in primary high school drops to 8 percent, but the coefficientfor highschool diploma is .594. Additional years of universityeducation continues to pay off (20 percent, only slightly lower than in column 2), while higher diplomas are apparentlyvery valuable. Technical diplomas do not make any difference. While these results lend credibilityto credentialism(showing large bonuses for obtained diplomas),returns to basic cognitive skills (the three R's) and years in junior high school (YRSED-Hl),to years of technical training (YRSTECH)as well as years of university training (YRSED-UN)are also rewarded. These results, assuming that years of schooling indeed reflect increases in human capital, support the validity of the human capital model. The third regression (column 3), using only diplomas obtained as the explanatory schoolingvariables,explains the wage rate significantlyless than the one that includesyears of schoolingas well (F = 6.92, significant at 1 % level), thus rejectinga pure credentialistinterpretationof the results.
-
15 -
5. Summary and Conclusion In this paper we presented the first estimates available for returns to education in the Republic of C6te d'Ivoire. The analyses was restrictedto wage earners only. Not surprisingly,we found that the two variables usually associatedwith wage differentials,experienceand education, turned out to be good predictorsin the Ivorian case as well. What is surprising,however, is the order of magnitude of the estimated returns to education. For instance, the 20 percent per year returns for higher education is almost twice as high as usually found. Furthermore,higher education showed a higher pay-off than elementary schooling,a result that also runs counter to usual findings. Since the wage sector is only a small part of the total labor market, it is perhaps prudent not to draw strong policy conclusionsfrom these observations. Still, they do point in the direction of a shortage of schooled personnel,especiallywith regard to higher education. To the extent possible, we disaggregatedthe returns to schooling into two effects: returns to skills and returns to credentials. The latter turned out to be very high, though a pure credentialistinterpretationof the data was rejected. If indeed there is a shortageof schooled personnel,it comes as a surpriseto find a bonus for diplomas; employers who compete for the scarce schooledemployees should reward any augmentationin productivity. Whether schooling is completed should not be the dominant factor. Note, however, that 41 percent of the Ivorian wage sector is government employment. If governmentwage policiesare such that those with completed schoolingare rewarded, without regard to the marginal increase in productivitydue to the last year of schooling,our data will reflect this. 12/ Though the government'semphasis on education,especiallyhigher
- 16 -
education, seems to adequatelyaddress the apparent shortage of schooled workers, a policy of paying rents for educationalcredentials is economically inefficientfor the public wage sector, while at the same time hampering the optimal distributionof scarce skilled labor resources in the labor market as a whole. A more detailed study of this issue is likely to have a high payoff, especially in light of the government'sstrong public policy on education. Using our most extensive model we showed that it pays to hold on to one's job, or at least not to change occupations. Given the large amount of job mobility observed in the data, these results did highlight the individual'swelfare loss associatedwith losing a job. We conclude by recalling a warning mentioned in the introduction: all results are conditionalupon being a wage earner. As such they refer to a small part of the Ivorian labor market only. An analysis of the other sectors of the labor market, including the question of how people sort themselvesinto the various sectors (e.g. Summer, 1981; Vijverberg,1986), will give a more complete picture of the role of education and experiencebut is beyond the scope of this paper.
- 17-
NOTES To be precise, the current analysis draws from data obtained on 1588 households interviewedbetween February 15, 1985 and February 15, 1986. Data on 12 households scheduled to be interviewedduring this period is missing.
2/ Secondary jobs are mostly agriculturalactivities or work in nonagriculturalfamily enterprises. Calculatingthe returns to labor of these activities is notoriouslydifficult and well beyond the scope of this paper. 31 Or by the relevant period if earnings were reported per day, week, quarter or year. 41 Obviously,reporting errors and miscodings in hours and/or earnings can lead to some highly implausiblehourly wages. Of the 522 individualswith wage earning jobs, 7 showed hourly wages of less than 10 CFA, and one reportedlyworked as a teacher for 13082 CFA per hour for 4 hours a week. These individualswere deleted after they were identifiedas outliersusing Cook's test. The total sample thus obtained includes 514 individuals. 51
For completenesssake we also includedyears of apprenticeshipin the equation, as an alternativeto technical training.
6/ These skills are selfreported. The correlationsbetween them are very high: .86 between arithmeticand writing, and exceeding .90 for the other cases. We estimated some test regressionsthat include these skills separately,but the multicollinearitypreventedus from separatingthe effects. 7/ The T-value of 1.53 indicates a 12% - significancelevel. 8/
It is customary to refer to the coefficientsin log wage equations as "percentage"differences,though this is not quite correct for discrete variables. See Halverson and Palmquist for the exact interpretationand correctionfactors.
9/ Full estimationresults are presented in Appendix 1. 10/ We did not separate senior high diploma and the various university degrees,as these groups become very small. / Note once again that the point estimates for completedelementary schoolingadd up to a 73 percent wage differentialwith illiterates.(11 + 49 + 6 * 2.3)
21 A comparativestudy of wage determinantsin the private and public sector separatelycan shed further high on this issue. However, since being in one sector rather than the other is endogenousto the wage earner, such a study is beyond the scope of this paper (See van der Caag and Vijverberg, 1986).
- 18
-
APPENDIX 1 Regression results with experience - education interaction terms. Table Al: Regression Results; Log Wage Equation with Education - Experience Interaction Terms (T-Values in Parenthesis)
Intercept
3.235
(12.58)
NAT
-.147
(1.46)
SEX
.041
(.45)
YRSED-EL
.205
(5.97)
YRSED-H1
.240
(6.53)
YRSED-H2
.235
(4.59)
YRSED-UN
.233
(6.75)
YRSTECH
.085
(3.36)
YRSAPP
.012
(.60)
EXPERM2
.081
(4.17)
-.870
(1.27)
.080
(4.00)
GEXPER2Q 1/
-.945
(2.31)
YRS 2/ * EXPERM2 1/
4.072
(1.53)
YRS * EXPERM2Q
-.153
(1.51)
-6.888
(2.55)
.088
(.77)
EXPERM2Q 1/ GEXPER2
_/
YRS * GEXPER2 1/ YRS * GEXPER2Q R 2, adjusted
_/
.639
1/
-'- 1000
2/
YRS =total years of formal schooling.
-
19
-
APPENDIX 2 Wage Equations by Age Group, Sex, Nationalityand Region
-/
In this Appendix we present estimates of the same wage equation discussed in Section 4 (Table 4.2), for age cohorts (under and over 32 years), males, females, Ivorians,non-Ivoriansand by region (Abidjan,other urban and rural). In Table A.1 we present F-statisticsfor the test whether the results differ significantlybetween the subgroupsdistinguished. For the 84 nonIvorians in the sample we find the wage equations to be the same as for Ivorians. However, pairwisedifferences between the other groups are statisticallysignificantat the 5% level.
Table A.1: F-Statisticsfor Testing the Equality of Wage Equation Between Subsamples F
Significancelevel
Age groups (over and under 32)
1.602
5%
Sex
1.974
5%
Nationality
1.319
Region
2.868
not significant 5%
We will now briefly discuss the main differencesdetected.
/ We are indebted to Hailu Mekonnenwho preparedall the statisticalresults of this Appendix.
-
20 -
Cohort Effects Table A.2 shows estimates of the wage equations for young (less than 32 years) and older (32 and over) wage earners. Out of the sample of 514 wage earners, 252 were below 32 years of age.
This group has an average of 7.60 years of schooling, 4.11 years of
current experience and 7.82 years of general experience.
The corresponding
figures for the remaining 262 wage earners above 32 years of age are 6.18 years, 13.57 years and 15.55 years.
Table A.2:
Wage Equations for Young and Old Workers
OLD
YOUNG INTERCEPT NAT SEX YRSED-EL YRSED-H1 YRSED-H2 YRSED-UN BASICED DIPCEPE1 DIPBEPC1 DIPUPPR1 DIPTECH YRSTECH YRSAPP EXPERM2 EXPRM2Q GEXPERM2 GEXPM2Q R2 , Adjusted
3.351 .088 .213 .023 .115 .066 .204 .102 .666 .660 .300 -.109 .068 .0007 .174 -.006 .021 .001
(10.67) (.51) (1.58) (.34) (1.72) (.39) (3.77) (.82) (2.80) (2.54) (.65) (.58) (1.43) (.02) (3.46) (1.31) (.55) (.67) .644
(12.23) (3.14) (2.36) (1.21) (1.24) (.91) (4.21) (1.33) (1.13) (2.11) (2.18) (.005) (2.41) (.55) (4.72) (2.68) (.35) (.98)
4.371 -.365 -.287 .058 .068 -.113 .194 .096 .240 .434 .810 -.0006 .081 .015 .083 -.001 .006 .0003 .668
- 21 -
Investmentin formal schooling has been emphasized by the Ivorian government as one of the strategiesfor sustainedgrowth. Government expenditureson education currently exceed 40 percent of the total budget, more than found in any other country. Education of the younger cohort now averages 7.58 years, while the literacy rate for this cohort is about 83 percent, compared to 73 percent for the older workers. Therefore, it is interestingto see whether this increased supply of trained people has reduced the returns to education,or whether the development of the Ivorian economy has lead to an even greater shortageof educated personnel. As stated above, the wage rate regressionsfor young (below 32 years of age; N = 252) and older (32 and over; N = 262) wage earners are significantlydifferent. However, the differencesappear in a somewhatunexpectedway: despite the considerable increase in education levels, rates of return to education for the younger cohort are generallyhigher than for the older ones. This is especially true for elementaryand junior high school diplomas. Older workers, however, benefit more from diplomas of higher education. Another differencebetween the wage equations for the two cohorts is that between male and female wage earners. While young women tend to be somewhat favored as comparedto their male counterpartsolder women receive 20 percent less (T = 2.35). In other words, while the overall picture did not show any significantgender discrimination,the cohort specific equations show a particularlyinterestingone.
- 22 -
Differentials by Gender Of the 514 wage earners in our sample, 100 are women.
Separate
estimates for male and female wage earners are presented in Table A.3.
Wage
equations estimated for male and females separately are statistically different at a 5 percent level. show a mixed picture.
The differences among individual coefficients
Male non-Ivorians face discrimination on the basis of
their nationality (-.21; T = 1.96), while female foreigners are favored (.56; T = 2.03).
Returns to junior high school are larger for women.
experience patterns are distinctly different:
The
while males show the "overall"
pattern (i.e. 12 percent per year in the early years, with a peak after about 3- years), females receive a flat 9 percent a year return to job specific experience. Table A.3:
Wage Equations for Male and Female Workers
Males INTERCEPT NAT YRSED-EL YRSED-Hl YRSED-H2 YRSED-UN BASICED DIPCEPE1 DIPBEPC1 DIPUPPR1 DIPTECH YRSTECH YRSAPP EXPERM2 EXPM2Q GEXPERM2 GEXPM2Q R2 Adjusted
3.537 -.214 .041 .079 .074 .211 .109 .528 .523 .223 .059 .073 .004 .117 -.002 .041 -.0003 .673
Females (17.94) (1.96) (.99) (1.67) (.64) (5.64) (1.58) (3.07) (2.75) (.68) (.47) (2.32) (.19) (8.51) (4.27) (2.63) (.64)
4.554 .558 -.113 .164 -.260 .136 .109 .683 .615 1.489 -.173 .102 .128 .093 -.002 -.016 .0006 .655
(11.36) (2.03) (.81) (1.72) (1.20) (1.41) (.46) (1.52) (1.88) (2.08) (.76) (1.61) (1.03) (2.79) (1.77) (.63) (1.31)
-
23 -
Wage equation by region Separate wage equations for Abidjan, other urban areas and rural areas are presented in Table A.4. Regional Effects The regional distributionof wage earners is that 280 live in Abidjan, 179 live in other cities and 55 live in rural areas. Wage earners in Abidjan have an average of 7.30 years of schooling,8.99 years of current experienceand 11.47 years of general experience. The correspondingfigures for other cities is 6.98 years, 9.67 years and 11.52 years, respectively. In rural areas, the wage earners had 4.34 years of schooling,6.24 years of current experienceand 14 years of general experience. None of the wage earners in the rural sample had universityeducation. As shown by the F-statistic in Table A.1 the wage equations for the three regions are significantlydifferent. It should also be noted that none of the coefficientsfor the rural wage equation are significant. This may be attributedto the relativelysmall rural sample. Comparisonswill therefore be made between the wage equation for Abidjan and that for other cities. Both in Abidjan and other cities, university educationhas a rate of return of 23 percent. The most striking result is the rate of return for advanced diplomas in Abidjan which is 89 percent. This variable has an insignificantcoefficientfor other cities. It is also noteworthy that the rate of return to a CEPE diploma is 96 percent for other cities and only 38 percent for Abidjan. It appears that the market responds to apparent shortages in well-trainedemployers in cities other than Abidjan.
-
Table A.4:
Wage Equations for Workers in Abidjan, Other Cities and Rural Areas
Abidjan INTERCEPT NAT SEX YRSED-EL YRSED-H1 YRSED-H2 YRSED-UN BASICED DIPCEPE1 DIPBEPC1 DIPUPPR1 DIPTECH YRSTECH YRSAPP EXPERM2 EXPM2Q GEXPERM2 GEXPM2Q R2 , Adjusted
24 -
4.009 -.174 -.062 .055 .021 -.169 .234 .024 .377 .700 .890 -.100 .111 -.024 .109 -.002 .035 -.0002
(16.03) (1.27) (.53) (.98) (.34) (1.19) (5.03) (.26) (1.68) (3.08) (2.12) (.64) (3.13) (.85) (6.14) (3.30) (1.86) (.45) .639
Other Cities 3.247 .285 .156 -.034 .157 .122 .233 .205 .962 .414 .087 .133 .047 .043 .098 -.002 .034 -.0001 .730
(10.40) (1.79) (1.13) (.55) (2.55) (.86) (3.58) (2.22) (3.58) (1.70) (.22) (.85) (.91) (1.24) (5.26) (2.34) (1.55) (.14) .730
Rural Areas (5.95) 3.893 (1.60) -.755 (.01) -.006 (.62) -.131 (.82) .220 (.83) .489 (.90) .385 (.53) .393 (.22) .199 (.50) -.846 (.88) .563 (.03) .004 (.33) -.042 (1.31) .092 (.33) -.001 (.66) .031 -.00009 (.11) .633
- 25 -
Finally, we present in Table A.5 the wage equations for Ivorians and nonIvorians.
Based on the F-test we could not reject the hypothesis that both
equations are statistically the same.
Table:
Table A.5
Wage Equations for Ivorian and Non-Ivorian Workers
Ivorian INTERCEPT SEX YRSED-EL YRSED-H1 YRSED-H2 YRSED-UN BASICED DIPCEPE1 DIPBEPC1 DIPUPPRI DIPTECH YRSTECH YRSAPP EXPERM2 EXPM2Q GEXPERM2 GEXPM2Q R2 , Adjusted
3.811 -.100 .028 .080 -.020 .211 .109 .479 .575 .526 -.022 .074 -.023 .117 -.002 .022 .0002
Non-Ivorian (19.03) (1.05) (.63) (1.83) (.19) (5.96) (1.41) (2.68) (3.38) (1.77) (.19) (2.48) (.87) (8.23) (4.27) (1.72) (.59)
.647
3.150 .940 .023 .175 .148 .088 .118 .247 .576 .286 .123 .070 .073 -.001 .078 -. 001
(6.10) (2.97) (.23) (.70) (.70) (.43) (.81) (.52) (.62) (.53) (1.10) (1.67) (2.33) (.85) (1.84) (1.25) .597
- 26
-
REFERENCES
Boissiere,M., J. B. Knight, R. H. Sabot. Earnings, Schooling,Ability and Cognitive Skills. American Economic Review Vol. 75, 5, 1985, pp. 10161030. King, T. (Ed.). Education and Income. World Bank Staff Working Paper, No. 402, July 1980. Knight, J. B. and R. H. Sabot. Why Returns to Experience Increase with education. World Bank, Mimeo, 1983. Layard, R. and C. Psacharopoulos. The Screening Hypothesisand Returns-toEducation,Journal of Political Economics,82, September/October1974, pp 985-998. Riley, John C. Testing the Educational Screening Hypothesis,Journal of Political Economy, 87, October 1979, pp. S227-S252. Psacharopoulos,G. Returns to Education: A Further InternationalUpdate and Implications.Journal of Human Resources Sincerely yours, , Volume 20, Number 4, Fall 1985, pp. 583-597. Sumner, Daniel A. Wage functions and OccupationalSelection in a Rural Less Developed Country Setting, Review of Economics and Statistics,November 1981, pp. 513-519. Van der Gaag, Jacques and Wim P.M. Vijverberg (1986). A SwitchingRegression Model for Wage Determinantsin the Public and Private Sectors of a DevelopingCountry,Mimeo. Vijverberg,Wim P.M. ConsistantEstimates of the Wage Equation when IndividualsChoose Among Income-EarningActivities, Southern Economic Journal 52, 4, April 1986, pp. 1028-1042. World DevelopmentReport, 1980. The World Bank, August 1980.
II.
A SWITCHING REGRESSION MODEL FOR WAGE DETERMINANTS IN THE PUBLIC AND PRIVATE SECTORS OF A DEVELOPING COUNTRY
-
28 -
1. Introduction Ever since education has been recognizedas an investment in human capital, economistshave attempted to estimate the rate of return to this investment. Occasionallysuch studies take a broad view, either by looking at the contributionof education to the overall growth of the economy (e.g. Krueger, 1968), or by taking a very comprehensiveview of the private benefits of education that accrue to the individual (e.g. Haveman and Wolfe, 1984). More often, however, rates of return to education are estimated as the marginal contributionto an individual'sproductivityof one year of education. Psacharopoulos'recent update of internationalstudies in this l area contains studies from no less than 61 countries (Psacharopoulos,1985). This Mincerian approach, which centers around the estimationof al wage or earnings function, is based on the assumption that wages are set equal to the marginal productivityof the wage earners, though--ofcourse--thereare many reasons why this may not be the case. Non-competitivemarket forces may influence the wage structure in many ways. Unionized workers are likely to) receive wages that differ from their non-unionizedcounterparts. Minimum wage legislationdrives a wedge between marginal productivityand the wage rate. Governmentsmay pursue employment,distributionalor other political policies using the wage scale for public employeesas a policy instrument. In general, the larger the institutionalor regulatoryinfluence on the labor market, the more likely there is a differencebetween the observed wage rate and the worker'smarginal productivity. This problem is especiallyimportant in developingcountries,where it is common that the wage sector is dominated by public employment. Consequently,the public wage bill forms a major part of the government budget
- 29 -
and comes under intense scrutiny in times of recession and adjustments. Furthermore,if distortionsexist, the working of the labor market in general, and developmentpolicies that depend on a flexible market in particular,may be seriously hampered by spillovereffects of public employment policies on the private sector. The first question to answer - and the one we address in this paper is whether wage differentialsbetween the public and the private sector exist. This questionhas spun a sizable literature in the industrialized world (e.g. Smith, 1977, Ganderson, 1979), but few studies exist for developingcou.itries.Moreover the results of such studies are mixed. Lindauer and Sabot (1983) conclude that a substantialwage premium exists for public employees (in Tanzania),but Corbo and Stelcner (1983) find a small bonus in the private sector (in Chile). Psacharopoulos(1985) reports the differencein returns to schooling between the "competitiveprivate" sector and the "non-competitivepublic" one to be 3 percentage points on average (13 and 10 percent, respectively). Unfortunately,a common shortcomingof these studies is that they attempt to estimate wage differentialsby using one or more dummy variables to indicate the sector where the individualis employed, or by estimating separatewage equations for each sector. If the labor market is segmented in sectors that give different awards for human capital, one of these sectors will be preferred by most workers and entry in that sector is likely to be determinedby factors other than productivity. Ignoring the endogeneityof being in one sector or the other may bias the estimates that are based on sector-specificsamples.
-
In this paper we will contributeto the Mincerian returns to education literatureand focus on public-privatesector differentials. Data
-
30 -
stem from a recent survey conducted in the Republic of C6te d'Ivoire, where 41.1 percent of all wage earners work in the public sector. The model developed in section 2 is similar to the one now commonly used to study union/non-unionwage differentials(e.g. Lee, 1978; Robinson and Tomes, 1984). It consists of two wage equations and a "switching"equation that determines in which sector the employee is working. Among other things, the model shows the relative importanceof human capital (years of education, experience)versus credentials (diplomasacqui-red)in obtaining a public or private job and in determining the wage. In section 2 we also describe the data. In section 3 we present and discuss the estimation results. Sector 4 concludes.
2.
Who Gets the Public Job?
If the labor market is segmented into a public and private sector, there will be a shortage of jobs in the preferred sector, and non-price rationingwill determine entry into this sector. The selection process has two steps: first an individualwill determine whether or not to try to obtain a public job.
Secondly he may or may not be chosen for the job. The
likelihoodof not being chosen forms a cost to the prospectiveemployee, that he compareswith the expectedbenefits. The probabilityof obtaining a public job depends on characteristicsof the individualthat are used by the employer to choose a worker from the queue. We assume, for the time being, that expected benefits are equal to the difference in the (log) wage rates between the two sectors. Thus, an individualwill be in the public sector if
(1)
(in w1 - Qn w2 ) > X 1 a1
e1
-
wI
where
31 -
the wage rate in the public sector
,
W2 ^ the wage rate in the private sector Xi , a vector of characteristicsthat are associatedwith the probabilityof obtaininga public job el
, a disturbanceterm.
Note that this one equation summarizesthe two-stepprocess: first, the expectedwage differentialhas to be large enough for the individualto make it worthwhileto try to obtain a public job and secondlythe employee needs to be chosen from the list of prospectivecandidates. We now assume that wages in each sector are determinedas follows: En w1 = Z YZ + u
(2)
n w2 =Zy2 +u
(3)
where
Z
2
is a vector of wage determining variables.
Substituting (2) and (3)
into (1), and assuming that all wage determiningvariables also influence the probability of obtaining a public job, we have:
I = 1 if XB + e > 0 (i.e., the individualhas a public sector job)
(4)
I = 0 otherwise
Where e = ul - u 2 - E
and X absorbs all exogenousvariables in Xi and Z.
Assuming normality of e, u1 and u2, maximum likelihoodestimates of the parametervectors of interest,y1 and y2, can be obtained. Note that OLS regressionson the public and private sample separatelyimplicitlyassume coV
(ul,
E)
=
coV
(u 2 ,
E)
= 0 and thus
are generally
biased.
-
32 -
The wage equations have the standardMincerian form (Mincer, 1974), regressing the log of the hourly wage rate on a set of education,experience and other exogenous variables. Education is measured by years of schooling at each level of the schooling system. We also include years of technical training and apprenticeshipin the equation. Total work experience is divided into experiencerelated to the current occupationand other work experience. Both terms are also entered in the regressionin quadraticform. In the literatureon returns to education there is much debate on the interpretationof the education effect. Does education indeed increase productivity(the human capital school) or does schooling,especially completed schooling,serve as a signal to employers about the innate ability and motivationof the worker (the screeninghypothesis). In order to test whether completed schooling or years of schooling is the relevantwage determiningvariable,we add to the wage equationswhether or not the individualhas one or more of the followingdiplomas: elementaryschool, junior high school, a higher diploma from traditional(i.e. general curriculum)education,or a technicaldiploma. The first three diplomasare cumulative. Note that adding both years of schooling and diplomas acquired, does not constitutea strong test of the relativemerits of the human capital and screening hypothesis. For instance,governmentworkers may receive a bonus for a diploma for reasons not directly related to a perceived difference (by the employer) in ability or motivation. On the other hand, able students may decide not to finish their traditionalschoolingin order to take advantage of better opportunities. Direct measurementsof a worker's abilities and skills are usually lacking in studies of this kind, which probably accounts for the lack of
- 33 -
consensus in the literatureon the relativemerits of the human capital and the screening model. Riley (1979) report results that support the screening hypothesis, but Boissiere et al. (1985) in what is perhaps the only study for the developingworld that has such measures, show strong support for the human capital theory, i.e. skill (productivity)differences rather than differences in schooling input, determinewage differentials. Our data allow us to include one measure of basic cognitive skills: reading, writing or simple arithmetic. The measure takes the value zero for illiteratesand increasesby 1 for every skill. The human capital model would predict that this variable, rather than years of elementary schoolingwill determine wage differentials.3/ We finally add sex and nationality to the equation to test for discriminationon the basis of these attributes.4/ Data used in this study stem from the C6te d'Ivoire Living Standards Survey (CILSS). Details on this nationwide survey, which collected informationon 1600 households in 1985, can be found in Van der Gaag and Vijverberg (1986) and Ainsworth and Munioz(1986). The survey contains informationon 513 individualswho report a wage earning activity as their main job during the seven days prior to the interview. Our analysis is based on this sample of wage earners. Reported earnings (generallyreported per month) were divided by hours worked per day times days worked per month to obtain an hourly wage rate. Wages include the cash value of in-kind income. Summary statisticsof the variables are presented in Table 2.1, for public and private workers separately.
-
34
-
Table 2.1: Definitionsand SummaryStatisticsof the Variables by Public and Private Sector Employment
Private Sector
Public Sector
N=301
N=212
Standard Symbol
NAT SEX DIP-EL DIP-HI DIP-UPP DIP-TECH RRR YRS-EL YRS-H1 YRS-H2 YRS-UN YRS-TECH YRS-APP EXPOCC GEXPER YRSCH AGE LNW
Definition
Mean
.275 Nationality: O=lvorian,1=other .149 O=male, 1=female 0/1 if no/yes elementaryschool diploma .478 .182 0/1 if no/yes junior high school diploma .089 0/1 if no/yes higher diploma .202 0/1 if no/yes technicaldiploma Reading,writing and arithmeticskills 1.973 3.561 Years elementaryeducation Years junior high school 1.215 Years senior high school .322 Years university .169 .734 Years technicaltraining Years apprenticeship 1.166 Occupationspecificexperience 7.399 13.135 Generalwork experience Total years of schooling 5.269 Age in years 32.554 Log of hourly wage rate 5.557
deviation
.44 .35 .50 .38 .28 .40 1,37 2.84 1,69 .89 .83 1.58 2.20 7.58 9.36 4.94 10.16 1.29
Standard Mean
deviation
.000 .259 .830 .491 .236 .472 2.637 5.132 2.472 .859 .717 1.462 .241 11.116 9.705 9.179 36.566 6.577
.00 .44 .38 .50 .43 .50 .94 2.08 1.80 1.29 1.87 1.61 1.08 8.23 8.85 5.26 8.70 .99
These summary statisticsforeshadowto some extent the estimation results of the switchingequation. Public employees are on average better educated, showing 9.179 years of education versus 5.269 in the private sector. Furthermore,the concentrationof schoolingdiplomas is much higher in the public sector. There are no non-Ivoriansin the public sector, and 25.9 percent of the governmentlabor force is female versus only 14.9 percent in the private sector. Total experience,measured as age minus formal schoolingminus technical trainingminus 5, averagesabout 20 years in both sectors. But occupationspecificexperienceis much lower in the private than
- 35 -
in the public sector, showing the importanceof job tenure in the latter and of job mobility in the former. Note also that, with an average age of 32 years, 20 years of experience indicates very young entry in the private sector. Worthy of special notice is also the differencein the wage rates: the difference in the means of lnW is 1.020 in favor of public sector employees. Finally note that wages in the private sector show more variation than in the public sector. In the next section we will first present the estimationresults of the log wage equations for the two sectors. Then, using the expected wage differenceas an instrumentalvariable, we will estimate the structural switching equation (i.e. equation (1)) as a probit equation.
3. EstimationResults Full informationmaximum likelihood(FIML) 5/ estimates of the model consistingof equations (2), (3) and (4) were obtained using the assumption that (ul, u2, E) are N(0, z) where the covariancematrix E equals
=
a 11
a 12 ale
012
022
a2e
All variables that enter equations (2) and (3) are also included in equation (4), with two exceptions: first, to avoid multicollinearity,years of schoolingby level of school have been aggregatedto total years of schooling.61 Secondly, since both experiencevariables are job specific, it is somewhat tautologicalto use them in the switchingregression. Therefore we decided to replace the experiencevariables by the age (and age-squared)of
- 36 -
the worker. If there is queuing we expect a positive effect of age on the probabilityof obtaining a public job, though the effect may diminish at higher age levels.
3.1 The Wage Equations Results of the FIML estimationfor the two wage equations are presented in Table 3.1, togetherwith OLS estimates for the public and private sector separately. We see, for both sectors, that diplomas are important in determiningwage differentials,with the exception of technical diplomas. The effect is very large, for instance a 42 percent wage increase in the public sector and a 61 percent increase in the private sector for a junior high school diploma. The effect of the three basic cognitive skills (RRR) is barely significantin the private sector, and not at all in the public sector. 7/ Note, however, that the effect of an elementaryschool diploma (highly correlatedwith basic skills) is larger in the public sector. The main question addressed in this study is whether the public and private wage equations are different,in part or overall. We divided the set of explanatoryvariables into three groups, related to (1) diplomas and RRR, (2) years of schoolingand technical training, (3) years of apprenticeshipand experience. We reestimatedthe model under the assumptionthat the coefficientsfor any one or a combinationof these groups of variables were different for the public and private sector. Likelihood ratio test were used to test whether the unrestrictedmodel (Table 3.1) performedbetter than any of the restrictedones. We could not reject the hypothesis that the diploma variablesand RRR, years of apprenticeshipand the experiencevariables have
-
37 -
Table 3.1: FIML and OLS Estimatesof Log Wage Equations for the Public and Private Sectors (t-valuesin parentheses)
FIML-Estimates* Private sector Public sector
INTERCEPT NAT
SSX
2.841 -
(7.87)
3.482 (14.65) .285
-
(2.20)
OLS-regressions Private sector Public sector
4.819 (14.59) _
-
3.320 (14.14) -.124
(-1.05)
-.125
(-1.04)
.141
(.97)
-.319
(-3.06)
.312
(2.17)
DIP-EL
.801
(2.50)
.395
(1.92)
.195
(.54)
.552
(2.75)
DIP-Hi
.424
(2.14)
.617
(2.40)
.179
(.95)
.806
(3.00)
DIP-UPP
.621
(2.10)
.221
(.45)
.551
(1.83)
.175
(.32)
DIP-TECH
.002
(.02)
.031
(.17)
-.036
(-.32)
.071
(.35)
RRR
.108
(.93)
.147
1.82)
.195
(1.44)
.091
(1.15)
YRS-EL
.035
(.60)
.018
(.37)
.002
(.02)
.048
(1.00)
YRS-H1
.205
(4.08)
.012
(.21)
.152
(2.85)
.047
(.76)
YRS-H2
-.040
(-.42)
-.101
(-.60)
-.010
(-.10)
-.067
(-.35)
YRS-UN
.205
(5.66)
.300
(4.21)
.160
(4.74)
.307
(3.91)
YRS-TECH
.036
(1.34)
.098
(2.42)
.020
(.60)
.112
(2.47)
YRS-APP
.067
(1.85)
-.008
(-.31)
.014
(.27)
.010
(.38)
EXPOCC
.087
(4.79)
.116
(7.32)
.053
(2.70)
.127
(7.47)
-2.258 (-4.14)
-.513
(-.79)
EXPOCCQx 100
-.868 (-1.43)
-2.321 (-3.82)
GEXPER
.016
(.94)
.024
(1.43)
-.023
(-1.28)
.053
(3.15)
GEXPERQ x 100
.711
(1.73)
-.089
(-.25)
1.007
(1.95)
-.489
(-1.31)
a
.704
(8.33)
.814
(9,57)
.948 (34.18)
-.746
(9.48)
11
i E
Log-Likelihood 2
R
.389
.692
-780.80 .631
.609
*Estimatesof parametersin the associatedswitchingequationare reported in Table 3.3.
- 38 -
the same effects on the wage rates in the public and the private sector. The years of schooling variables,however, showed a statisticallysignificant 2 differencebetween both sectors (X = 15.53, significantat 1 percent level). Estimates of the thus restrictedmodel are presented in Table 3.2. The results of years of elementary schooling show no effect in both sectors. Note, however, that the basic cognitive skills acquired in elementary school show a significanteffect (.125, T = 2.03). Junior high school years are rewarded in the public but not in the private sector, university training is valuable in both sectors but technical training pays only in the private sector. The results reported in Tables 3.1 and 3.2 clearly show the selectivitybias in the OLS estimates. OLS equations estimate wages in both sectors very well in view of the high R2 and high t-statistics. They produce an estimated wage profile for public sector employees that lies substantially above that in the private sector. 8/ The FIML estimates correct for selectivitybias and show that this bias is large. The (marginal)wage profilesnow lie at the same level. The estimated variance rises, especially that for the public sector equation, which is an indicationthat the sample of public sector employees is a severelycensored one. Finally, the correlation coefficientsbetween the disturbance in the switching equation (E) on the one hand and those in the two wage equations (ul and u2) on the other, are highly significant. Their signs conform to what one would expect based on the definitionof e = u 1
- u2
-
el. They imply that someone with, for some
unobservedreason, a high public (private)wage is also more likely to obtain employmentin the public (private) sector. Again, this effect of unobserved factors is the reason for selectivitybias in the OLS results. As is well-
- 39 -
Table 3.2: FIML Estimates of Log Wage Equations for The Public and Private Sector, with Restrictionson Diploma and Basic Skill Variables* (t-valuesin parentheses)
Public Sector
INTERCEPT
3.169
Private Sector
(12.03)
_
NAT SEX
-.112
(-.93)
DIP-EL**
.450
(2.88)
DIP-HI**
.487
(3.37)
DIP-UPP**
.482
(2.07)
3.412
(18.27)
.288
(2.10)
.148
(1.06)
DIP-TECH**
.001 (.01)
RRR**
.125
(2.03)
YRS-EL
.035
(.86)
.032
(.84)
YRS-H1
.189
(4.24)
.041
(.83)
YRS-H2
-.013
(-.16)
-.145
(-1.39)
YRS-UN
.207
(6.16)
.295
(4.32)
YRS-TECH
.037
(1.35)
.107
(2.92)
YRS-APP**
.009
(.43)
EXPOCC**
.100
(8.06)
EXPOCCQ x 100**
-.153 (-3.59)
CEXPER**
.021
(1.84)
CEXPERQ x 100 **
.016
(.62)
aii
.670
(7.41)
.821
(9.11)
,lie Log-Likelihood
.925
(25.18)
-.741
(-8.91)
* **
-785.99
Estimates of parametersin the associated switchingequation are reported in Table 3.3. Coefficientsfor these variablesare restrictedto be the same in both sectors.
- 40 -
known (e.g. Goldberger,1983) correctionsfor such bias may be sensitive to the distributionalassumptionsadapted, as well as to the specification(and consequentidentification)of the switching equation. Still, given the prevalenceof OLS over FIML in this type of research, our conclusions sound a strong warning against the use of results obtainedwithout taking the endogeneityof sector choice into account.
3.2 The Switching Equation Given the estimationresults we can now calculate expected log-wage rates in both sectors and, using their differencesas an instrumental variable, estimate the structuralswitchingequation as a probit equation. Togetherwith the FIML estimates of the switchingequation related to Tables 3.1 and 3.2, respectively,these probit results are presented in Table 3.3, in column 3. Note that the first two columns combine the effect of the personal characteristicson wages with their impact on the likelihoodof obtaining a public sector job. As became already apparent from the descriptive statistics,women are relativelyfavored over men in the public sector. The schoolingvariables play an important role in determiningwhether an individualobtains a public or private job. That by itself is not a very surprisingresult: the skill mix necessary in the governmentsector is more likely to show a need for higher educatedworkers than that in the private sector. What is important, however, is which of the schoolingvariables are dominatingthe selection process. Elementaryand high-schooldiplomas are extremely important,but higher and technical diplomashave no significanteffect.
- 41 -
Table 3.3: Estimates of the Switching Equation* (t-values in parentheses)
INTERCEPT SEX DIP-EL DIP-Hl DIP-UPP DIP-TECH RRR YEARS SCHOOLING AGE AGEQ * 100 Zyl
'
-
(1) FIML-Estimates
(2) FIML-Estimates
(3) Probit-Estimates
-4.384 (-6.27) .423 (2.70) .615 (2.19) .372 (1.72) -.219 (-.79) .016 (.10) -.124 (-1.21) .048 (1.30) .165 (4.47 -.158 (3.26)
-4.731 (-6.85) .414 (2.67) .525 (2.26) .444 (2.50) -.324 (-1.38) .028 (.21) -.116 (-1.38) .048 (1.53) .191 (5.19) -.199 (-4.13)
-4.872 (5.10) .456 (2.21) .824 (2.54) .645 (2.31) -.185 (-.63) -.007 (-.04) -.144 (-1.26) .036 (.86) .183 (3.55) -.177 (-2.59)
ZY2
-.307
(-.81)
Columns (1) and (2) are associatedwith the estimated wage equations reported in Tables 3.1 and 3.2, respectively. Perhaps the most important result is that years of schooling do not
show any significanteffect on the probabilityof obtaining a public job. All that seems to matter is whether or not a diploma is obtained. Thus, screening on the basis of diplomas obtained, seems to be the dominant case in Cate d'Ivoire. Age shows the positiveeffect that can be expected if there is queuing for the public sector. Only at fairly high ages (around 50) becomes the effect negative. In the structuralprobit equation the diploma effects become even more pronounced,but the effect of the expectedwage differential is not significantlydifferent from zero. The probit equation predicts the probabilityof being in one sector or the other quite well. Over 71 percent of the observationsare predictedcorrectly.
-
4.
42 -
Summary and Discussion The estimationresults of Tables 3.2 and 3.3 (column 2) allow us to
examine the large observedwage gap between the public and private sector. For an average Ivorian employee, choosing to work in the public sector, we predict a log wage of 6.246 (CFA 516). An identicalemployee who chooses to work in the private sector would earn CFA 471 (log wage is 6.154). Clearly, the larger differencesin wage rates that are observed between both sectors reflect to a large extent differencesin educationalattainmentand experience. We present the effect of changes in educationalattainment and occupation-specificexperienceon an otherwise average Ivorian employee, in Table 4.1. The first column shows differencesin the probabilityof obtaining a public job. The impact of educationand age is very pronounced. Columns 2 and 4 show the differencesin the offered log wage 91 in the public and private sector, respectively. For example, someone with reading, writing and arithmetic skills (RRR = 3) and 6 years of elementaryschooling (and having average characteristicsotherwise)will be offered a log wage of 4.714 in the public sector. If this person obtains an elementary school diploma, the offer becomes 5.164. Wage offers in the private sector are higher for low education levels, but lower for high education levels, quite different from what the OLS results suggest. In columns (3) and (5) we present the expected value of the log wage rates conditionalupon being in the public or the private sector. For the public sector this is defined as E(Qn w|I=l) = E(Un wle
>
- X6) = ZY1+ Pic
1/2
(X)
-
43 -
and for the private sector E(Qn w|I = 0) = E(Qn w|e
