Gender and Informal sector earnings gaps in Cameroon

Gender and Informal sector earnings gaps in Cameroon

KAPNANG Herrman Brice 1

NASSUF Hassani 2

AbstractIn this paper, we use the quasi-experimental evaluation involves examining gender and sector

earnings gaps in Cameroon using data from an original household consumption surveys. We investigatewhether there are gender differences in earnings between formal and informal workers. Given that thechoice to work in formal or informal sector may be related to outcomes, we use a Probit selection modeland a switching regression model with endogenous switching to control for potential sample selectionbias.We also use the most common approach : OaxacaBlinder decomposition to identifying sources ofgender wage gaps. The estimates are obtained using the log of hourly rather than monthly earningsto take account of the heterogeneity of working hours in different sectors. There is a very close linkbetween gender and the employment sector and earnings. Women have over 6 % likely more than mento work in the informal sector. Then, implementing switching regression model, we find that there isa gross wage penalty of a little over 11.8 % for women working in the informal sector, whereas thegender earnings gap in the formal sector is around 5.9 %. In other words,the earnings penalty tendsto be greater for women than for men in both sectors. Our analyses thus highlights the importance toface gender earnings in order to achieve development agenda.

JEL Classification : J460, O170, O550,

Keywords : wage, Informal Sector, bias selection,gender earnings gap, endogenous switching.

1. Institute of development economic and social studies :University of Paris 1 Panthéon-Sorbonne,[email protected]

2. Institute of development economic and social studies :University of Paris 1 Panthéon-Sorbonne

1

1 Introduction

The Constitution of Cameroon guarantees equality of opportunity in employment and directs theState to secure equal rights for livelihood, equal pay for equal work as well as just and humane conditionsof work for all. Despite the concerted efforts of the State, the economic status of women is laggingfar behind their male counterparts. Women work the most ; paradoxically they earn the least in life(Vandana Dave, 2012). A Majority of women work in unorganized sectors for low wages due to lowlevel of skills, illiteracy, ignorance and surplus labor and thus face high level of exploitation. Much workhas been done on the relationships between earnings and Human capital (Becker, 1964, and Mincer,1970, 1974). In the case of gender, the need for inequality reducing policies has been explicitly placed

on the poverty reduction agenda defined by the Millennium Development Goals (MDGs) in 2000, asMDG 3 is specifically aimed at promoting gender equality and empowering women. The design ofadequate policies to achieve this goal calls for an understanding as to whether differences in labourmarket outcomes stem from differences in workers’ characteristics or from differences in the returns totheir characteristics (Nordman al,2011).

The ILO(2003) defines informal employment as employment that does not provide coverage in theinstitution social protection scheme. Roubaud (2013) defined informal employment as all contributingfamily workers, all independent workers in the informal sector, and all employees without writtencontracts and not benefiting from social protection. In Camerron, we estimated at more than 70% theshare of employment in the informal sector and nearly 38% of the GDP (except agricultural sector).The often hidden role of informal institutions in shaping employement outcomes for women becomesfairly visible when the type of working mainly performed by women is considered, in particular inagriculture (Jütting Johannes, 2009)

Using data from household consumption surveys carried out in all cities in Cameroon,we shednew light on the issue of labour market disparities. The purpose of this paper is to measure andcompare gender labour market disparities in Cameroon. We address a number of questions. Are genderearnings gaps sizeable and similar ? Are there any labour market characteristics that explain variationsacross gender ? Traditional gender earnings decompositions rely on estimations of Mincer-type earnings

functions for men and women. Since the work of Gary Becker (1964) and Jacob Mincer (1974), severalwork succeed and showed that the wages increase with the training level and the acquired professionalexperiment throughout school and professional course. An often used methodology to study labor-market outcomes by groups (sex, race, and so on) is to decompose mean differences in log wages basedon linear regression models in a counterfactual manner. The procedure is known in the literature as theBlinder Oaxaca decomposition (Blinder 1973 ; Oaxaca 1973). It divides the wage differential betweentwo groups into a part that is explained by group differences in productivity characteristics, such aseducation or work experience, and a residual part that cannot be accounted for by such differences inwage determinants. This unexplained part is often used as a measure for discrimination, but it alsosubsumes the effects of group differences in unobserved predictors. Most applications of the techniquecan be found in the labor market and discrimination literature (Nordman et al, (2011) ; Nguetse etal(2013).

Besides this microeconomics approach, Macroeconomics model use Harris and Todaro function.Several authors used the approach of Harris and Todaro to built the equations of labour force offer inEGC models , Agénor and El Aynaoui (2003), Marouani and Robalino (2012). Indeed, a bias of selectionresulting from the probabilities that has the worker to find himself in formal or informal sector estimateby OLS (GATIGNOL, Roy 1999 ; ZAIBI Fakher, 2007 ; Roy, Paul-Martel and Paul Bodson, 1992). TheMincer earnings equation estimated using an ordinary least squares (OLS) approach. The potentialpresence of a correlation between the matrix of regressors and the error term has been shown to leadto inconsistent (and biased, in the small sample case) coefficient estimates (Card 1999 and 2001). Inthe case of the Mincerian earnings equation, there are several potential sources of correlation between

2

the regressors and the error term. Our estimates of decomposition terms may be directly affectedwhen workers of informal and formal self-select into the labor market differently. Thus, controllingfor selection based on observables and unobservables is necessary to guarantee point identificationof the decomposition terms. When labor market participation is based on unobservables, correctionprocedures for the mean wages are also available. In these procedures, a control variate is added as aregressor in the conditional expectation function. The exclusion restriction that an available instrumentZ does not belong to the conditional expectation function also needs to be imposed. Khitarishvili (2008)uses the instrumental variables approach to test for the presence of endogeneity in the educationvariable and does not find sufficient evidence to reject the hypothesis of exogeneity of education. Inthis study we test and correct for another potential source of correlation : sample selection bias using" switching regression model ".

The studies which test the outputs of the investment in human capital by comparing two segmentsof the labour market start, in general, by dividing the sample into two sectors. After them test anequation of wages for each sector using OLS in order to analyze the differences. This approach wouldbe suitable if the selection of individuals into the category of wage earners is random, the coefficientestimates in the wage equation can be biased if not. But research supports that the informal sector is notuniform and that certain individuals voluntarily choose this sector and not under the constraint. The

rest of this paper is organised as follows. Section 2 the econometric methods used, Section 3 examinesthe data and some demographic and labour market statistics. Finally, in Section 4, we comment onthe results and draw together the main findings and conclude.

2 Methods

In this article, we describe the implementation of the maximum likelihood (ML) algorithm tofit the endogenous switching regression model. Models with endogenous switching can be fitted oneequation at a time by either two step least squares or maximum likelihood estimation. However, bothof these estimation methods are inefficient and require potentially cumbersome adjustments to deriveconsistent standard errors. The movestay command, on the other hand, implements the full-informationML method (FIML) to simultaneously fit binary and continuous parts

In this model, a switching equation sorts individuals over two different states (with one regimeobserved). The econometric problem of fitting a model with endogenous switching arises in a varietyof settings in labor economics, the modeling of housing demand, and the modeling of markets indisequilibrium.

Consider the following model, which describes the behavior of an agent with two regression equa-tions and a criterion function, Ii, that determines which regime the agent faces.

Ii = 1 if γZi + ui > 0Ii = 0 if γZi + ui ≤ 0

Regime1 : y1i = β1X1i + ε1i if Ii = 1 [1]Regime2 : y2i = β2X2i + ε2i if Ii = 0 [2]

Here,yji are the dependent variables in the continuous equations ; X1i and X2i are vectors of weaklyexogenous variables ; and β1 , β2 and γ are vectors of parameters. Assume that ui, ε1i and ε2i have atrivariate normal distribution with mean vector zero and covariance matrix.

Ω =

σ2u σ1u σ2uσ1u σ21 .σ2u . σ22

where σ2u is a variance of the error term in the selection equation, and σ21 and σ22 are variances of

the error terms in the continuous equations. σ1u is a covariance of ui and ε1i, and σ2u is a covarianceof ui and ε2i. We can assume that σ2u = 1 (γ is estimable only up to a scalar factor).

3

Given the assumption with respect to the distribution of the disturbance terms, the logarithmiclikelihood function for the system of (12) is

lnL =i(Iiwi [ln F (η1i)+ ln F (ε1i/σ1)/σ1]) + (1-Ii)wi [ln 1− F (η2i)+ ln F (ε2i/σ2)/σ2]

where F is a cumulative normal distribution function, f is a normal density distribution function,wi is an optional weight for observation i.

After estimating the models parameters, the following conditional and unconditional expectationscould be calculated :

Unconditional expectations :

E(y1i|x1i) = x1iβ1 [3]E(y2i|x2i) = x2iβ2 [4]

Conditional expectations :

E(y1i|Ii = 1, x1i) = x1iβ1 + σ1ρ1f(γZi)/F (γZi) [5]E(y1i|Ii = 0, x1i) = x1iβ1 − σ1ρ1f(γZi)/ 1− F (γZi) [6]

E(y2i|Ii = 1, x2i) = x2iβ2 + σ2ρ2f(γZi)/F (γZi) [7]E(y2i|Ii = 1, x2i) = x2iβ2 − σ2ρ2f(γZi)/ 1− F (γZi) [8]

When the explanatory variables in the regressions are the same and there is only one dependentvariable, only one equation need be specified. Alternatively, both equations must be specified when theset of exogenous variables in the first regression is different from the set of exogenous variables in thesecond regression or when the dependent variables are different between the two regressions.

We will illustrate the use of the movestay command by looking at the problem of estimatingindividual earnings in the informal and formal sectors. Movestay is implemented as a d2 ML evaluatorthat calculates the overall log likelihood along with its first and second derivatives. The commandallows for weights and robust estimation, as well as the full set of options associated with Statasmaximum likelihood procedures. A typical specification might be the following :

lnw1i = Xiβ1 + ε1i [9]lnw2i = Xiβ2 + ε2i [10]

I∗i = δ(lnw1i − lnw2i) + γZi + ui [11]

Here I∗i is a latent variable that determines the sector in which individual i is employed ; wji is the wageof individual i in sector j ; Zi is a vector of characteristics that influences the decision regarding sectorof employment. β1, β2, and γ are vectors of parameters, and ui,ε1 and ε2 are the disturbance terms.The observed dichotomous realization Ii of latent variable I∗i of whether the individual i is employedin a particular sector has the following form :

Ii = 1 if I∗i > 0Ii = 0 if otherwise [12]

The assumption that is often made in this type of model is that the sector of employment is endogenousto wages. Some unobserved characteristics that influence the probability to choose a particular sectorof employment could also influence the wages the individual receives once he is employed. Neglecting

these selectivity effects is likely to give a false picture of the relative earning positions in both the

4

public and private sectors. The simultaneous ML estimation (9-12) corrects for the selection bias insectoral wage estimates.

In our study, the sector choice indicator informal takes value 1 if the individual is employed in theinformal sector and 0 if in the formal sector.

The wage equations (9-10) estimate log of monthly individual earnings. The exogenous variables inthe wage regressions (9-10) are based on a typical Mincers type specification and include such individualcharacteristics as age, age2, education and sex.

In addition to these variables, the sector selection equation (11) includes two variables to improveidentification. An area of residence and sector activity are believed to influence an individuals choiceof the sector of employment but not affect the wages.

3 Results

3.1 Data

The data used within the framework of this work come from the third investigation of consumptionnear households in Cameroon(ECCAM). It was carried out into 2008 by the National Institute ofStatistics (INS). The fourth edition is under development. It constitutes with the general census of thepopulation the two greater activities of data acquisition in Cameroun.

The sample used to analyze the determinants of the wages consists of 6095 heads of householddivided on all the national triangle. The variable of interest selected is the monthly income of the mainthing activity. Several other variables were collected, in particular those relating to the human capitalthe such age of the head of the household, the age of entry on the labour market, the number of yearof education, the level of schooling, the sex, just as that relating to the localization of the household,namely in rural zone or urban zone, those relating to the type of employment, temporaire/permanent,informel/formel and privé/public.

5

Figure 1 – Comparative earnings gap

Comparative earnings gap, Note : estimates stemming from Kernel density distribution of log hourly earningsincluding a set of individual characteristics

Therefore, there is no clear-cut evidence that women are systematically more likely than men to bein informal employement. However, when the decomposition of earnings is considered, there is evidencethat men are more remunerated than women.

There is an additional element of gender segregation emphasised by Chen et al.(1999) that alsocontributes to the aggregate gender earnings gap, and that is within these six informal job statuses.From this Chen (2006) concludes that :

"women tend to work in different types of activities, associated with different levels of earnings, thanmen-with the result that they typically earn less even within specific sectors of the informal economy".

These gender differences exist to a large extent because women still bear the brunt of the unpaid butunavoidable domestic tasks of daily life, such as childcare and housework. In less developed countries,young women are more likely than men to be neither in employment nor in education or training,and when they enter the labour market, women are more likely confined to the most vulnerable jobs,frequently in the informal sector.

3.2 Result of Probit selection model

The dependent variable is whether the individual is in formal or informal sector. In this householdsample, two-thirds worked in informal, and one-third in formal. The mean age is 40, 77 percent aremale. To control for broad geographic differences in access to labor market, we include the fact that

6

whether the person lives in rural or urban area. We fitted a probit model, although the results for logitwould be virtually identical.

The table below presents marginal effects of the estimate carried out. The first result is that ,predicted probability to work in informal sector in Cameroon is around 82%.

Figure 2 – marginal effet result of probit selection model

Source : ECCAM survey , authors calculations

The analysis of the marginal effects reveal that female have 6% likely more than male to work ininformal sector. In the same way, compared to the workers of the agricultural sector, the passage inthe industrial sector reduced from approximately 30% the probability to work in the informal, against8% only in the trade and nearly 40% in the services. Indeed, the fact of residing in rural zone increasefrom approximately 10% probability of working in the Informal sector.

3.3 Estimation of income equation

The exogenous variables in the wage regressions are based on a typical Mincers type specification(Mincer 1974) and include such individual characteristics. In addition to these variables, the sector se-lection equation includes many variables to improve identification. The results of the wage regression inthe formal and informal sector are reported in the table bellow. Three common approach are presented :OLS regression, OAXACA-Blinder decomposition and endogenous switching regression model.

Figure 3 – Estimation of income equations (Dependent variable : logarithm of the hourly income)

Source : ECCAM survey , authors calculations: significant at 10 % ; ** : significant at 5 % ; *** : significant at 1 %

7

Model 1, as shown in Figure 3 (OLS) uses this approach to predict log-hourly wages, controllingfor education, hours worked, demographic groups, work experience, and job tenure. If there is a genderwage penalty, then the variable "woman" should have a negative coefficient. Using this somewhat naivemodel, gender wage penalty is measured and appears to associated with an approximate 10% declinein wages in formal sector and around 17% in informal sector. However, using the simple approach ofpooled OLS is unsatisfying as it does not account for any of the potential biases discussed in section2. Accordingly, more complex models will be estimated to account for these potential biases.

Endogenous switching models are used in other parts of the labour economics field for their abilitydeal with endogenous selectivity issues, similar to that of the choice to work in informal sector. Byusing this model, the endogenous choice of whether or not to work in formal sector can be controlledfor and estimates robust to endogenous selectivity obtained for a woman’s wages. This model estimatesthat there is a 12% wage gap between women and men in informal sector and 6% in formal sector.Additionally, a Wald test (Appendix 1) determining whether the estimated formal and informal wage

"regimes" are independent reveal that the two wage equations are significantly independent from oneanother (p-value = 0.00). This result suggests that there are two distinct labour markets, one for formaland one for informal.

3.4 The Blinder Oaxaca Full decomposition of the gender earnings gap

A widely used method for studying differences in wages between two separate groups has beendeveloped in both Oaxaca (1973, p.696) and Blinder (1973, p.438). This "Oaxaca" method decomposeswage differences between groups into two parts, an "explained" portion and an "unexplained" portion.The explained portion is the component of the difference in wages between groups which can beattributed to differences in characteristics such as education levels and work experience. For this reason,a two step Heckman (1979) correction has been made in both models and a "selectivity" term addedto the Oaxaca decomposition.

Figure 4 – The Blinder Oaxaca decomposition for linear regression models

Source : ECCAM survey , authors calculations1 USD= 572 FCFA

In Figure 3, it is shown that women on average 51 FCFA lower wages than men.The decompositionoutput reports the mean predictions by gender and their difference in the first sample with selectionbias. In Cameroon, the mean of log wages (lnwage) is 5.28( 197 FCFA) for men and 4.98( 145 FCFA)for women, yielding a wage gap of 0.29 (51FCFA). In labor-market research, it is common to include acorrection for sample-selection bias in the wage equations based on the procedure by Heckman (1976,1979). Oaxaca used with Heckman, allows the decomposition to automatically adjust for selection.Informal sector allocation is modeled as a function of living area, gender, economy branche. Comparingthe results with the output in the first estimation reveals that the uncorrected wages of women arehighly biased downward (4.98 versus the selectivity-corrected 4.67), and the wage gap is somewhatunderestimated (0.29 versus the corrected 0.60). Finally the decomposition with informal selectivity-corrected output reports the mean predictions by gender and their difference. In Cameroon, the meanof hour wage is 197 FCFA for men and 106 FCFA for women, yielding a wage gap of 90 FCFA.

8

Conclusion

This paper has provided robust quantitative evidence for the existence of a gender wage penalty inCameroon. In particular, using data from the nationally representative household survey dataset, ourempirical estimates show that there is a positive and statistically significant 12% gender difference inthe wages in informal sector and around 6% in formal sector. Furthermore, it has been found that theseresults remain consistent and statistically significant after controlling for potential endogeneity of thechoice to work in formal or informal , the self-selection to work self-selection to work and unobservedheterogeneity.

Using Cameroon data, it was found that educational level directly influence the gender wage penalty.Then, mainstreaming the gender perspective at all levels of policy is one aspect of efficiently enhancinggender equality. Gender equality is not just about economic empowerment. It is a moral imperative,it is about fairness and equity, and includes many political, social and cultural dimensions. Genderequality, however, is also a key factor in self-reported well-being and happiness across the world.

Substantial international research on the gender wage penalty has been conducted in seven WestAfrican cities (Nordman and al, 2011). They estimates of actual earnings gap=log (male earning)-log(female earning.) consistently ranging from 0.77 in Cotonou to 0.55 in Dakar (see Appendix 2). Weapply the same methodological approach to the analysis of gender earnings gaps in central Africa weobtain 0.608 in Cameroon. The design of effective support policies for women entrepreneurship in deve-loping countries requires a deeper understanding of the heterogeneous landscapes where women operateas entrepreneurs. The distinction between formal and informal is meaningful for policy design, becausefemale owners of formal and informal businesses have different profiles and their businesses have dif-ferent needs and growth potentials. Female owners in the informal sector of developing countries havemuch less education, start their business because of necessity and have very low earnings from theirbusiness. Several countries have tried to push firms into the formal sector, primarily by reducing thecosts associated to registration. Formalisation programs can have a relatively large impact on women,given that the burden of complying with government regulations is heavier for the low-scale businesseswhere female owners are prevalent. However, reducing the costs of registration without significantlyimproving the business environment is unlikely to turn millions of informal micro-businesses into com-petitive small and medium enterprises in the formal sector. Then, it seem important to increase boththe quantity and quality of data by gender and improve evaluation of public policy and ensure equalaccess to finance for male and female entrepreneurs.

9

Références

Andreas Peichl, Nico Pestel Earnings Inequality ,IZA Policy Paper No. 89, July 2014

Ben Jann , ETH Zûrich The BlinderOaxaca decomposition for linear regression models,The StataJournal (2008) 8, Number 4, pp. 453479

Cahuc Pierre et al. Dualisme des contrats de travail et rotation de la main-d’oeuvre ,Revue françaised’économie, 2012/1 Volume XXVII, p. 47-64.

Christophe J. Nordman , Anne-Sophie Robilliard , François Roubaud Gender and ethnic earningsgaps in seven West African cities ,Labour Economics 18 (2011) S132S145

Harris, R., Todaro, M. Migration, Unemployment and Development : a two sector analysisAmericanEconomic Review, 60, 126-142, 1970.

Jütting Johannes, de Laiglesia Juan R. L’emploi informel dans les pays en développement Une nor-malité indépassable ?OECD Publishing, 20 avr. 2009 - 170 pages

Le Duigou Sarah L’effet de l’âge sur la distribution des salaires , Revue française d’économie, 2012/1Volume XXVII, p. 163-185.

Lewis, W.A Economic development with unlimited suplies of labour , The Manchester School of Eco-nomic and Social, 22, 139-191, 1954.

Michael Lokshin, Zurab Sajaia Maximum likelihood estimation of endogenous switching regression mo-dels ,The Stata Journal (2004) 4, Number 3, pp. 282-289.

Mohamed A. Marouani and David A. Robalino Assessing interactions among , education, socialinsurance and labour market policies in Morocco ,Applied Economics, 2012, 44, 3149-3167

NGUETSE Pierre, BEM Justin Impact of gender wage differentials on poverty and inequalities inCameroon : a distributional approach ,Preliminary version, 2010,

Oaxaca, R. L., and M. R. Ransom IOn discrimination and the decomposition of wage differentials,Journal of Econometrics 61 : 521,1994

Philippe De Vreyer,Francois Roubaud Urban Labor Markets in Sub-Saharan Africa ,World Bank Pu-blications, 7 juin 2013 - 460 pages

Todaro, M. Urban job expansion, induce migration and rising unemployment. A formulation andsimplified empirical test for LDC’s Journal of Development Economic, 3, 165-187, 1976

VANDANA DAVE Women Workers in Unorganized Sector WOMENS LINK, VOL. 18, NO. 3, 2012

Yongjian H. Essays on Labour Economics : Empirical Studies on Wage Differentials Across Categoriesof Working Hours, Employment Contracts, Gender and Cohorts Volume 357 de Research series /Tinbergen Institute, Rozenberg Publishers, 2005

10

Appendix

Appendix 1 : Switching model result

Figure 5 – Appendix 1 : Switching model result

Source : ECCAM survey , authors calculations

11

Appendix 2 : Gender and ethnic earnings gaps in seven West African cities

Figure 6 – Appendix 2 : Gender and ethnic earnings gaps in seven West African cities

Source : Christophe J. Nordman a, Anne-Sophie Robilliard b, François Roubaud

Appendix 3 : The Blinder Oaxaca decomposition Estimation

Figure 7 – Appendix 1 : Switching model result

Source : ECCAM survey , authors calculations

12

Appendix 3 : Selectivity bias adjustment for Blinder Oaxaca decomposition

Figure 8 – Selectivity bias adjustment

Source : ECCAM survey , authors calculations

13