Decomposing Gender and Ethnic Earnings Gaps in Seven West African Cities

Transcription

1 DOCUMENT DE TRAVAIL DT/ Decoposing Gender and Ethnic Earnings Gaps in Seven West Arican Cities Christophe NORDMAN Anne-Sophie ROBILLIARD François ROUBAUD DIAL 4, rue d Enghien Paris Téléphone (33) Fax (33) E-ail : dial@dial.prd.r Site :

2 DECOMPOSING GENDER AND ETHNIC EARNINGS GAPS IN SEVEN WEST AFRICAN CITIES Christophe Nordan Anne Sophie Robilliard François Roubaud IRD, DIAL, Paris IRD, DIAL, Dakar IRD, DIAL, Hanoï Docuent de travail DIAL Octobre 2009 Abstract In this paper, we analyse the size and deterinants o gender and ethnic earnings gaps in seven West Arican capitals (Abidjan, Baako, Cotonou, Dakar, Loe, Niaey and Ouagadougou) based on a unique and perectly coparable dataset coing ro the Surveys conducted in the seven cities ro 2001 to Analysing gender and ethnic earnings gaps in an Arican context raises a nuber o iportant issues that our paper attepts to address, notably by taking into account labour allocation between public, private oral and inoral sectors which can be expected to contribute to earnings gaps. Our results show that gender earnings gaps are large in all the cities o our saple and that gender dierences in the distribution o characteristics usually explain less than hal o the raw gender gap. By contrast, ajority ethnic groups do not appear to have a systeatic avourable position in the urban labour arkets o our saple o countries and observed ethnic gaps are sall relative to gender gaps. Whatever the sign o the gap, the contribution o dierences in the distribution o individual characteristics varies arkedly between cities. Taking into account dierences in sectoral locations in the decoposition o gender earnings gaps provides evidence that within-sector dierences in earnings account or the largest share o the gender gap and that the dierences in sectoral locations are always ore avorable to en than to woen. By contrast, concerning ethnic earnings gaps, the ull decoposition indicates that sectoral location soeties plays a copensating role against observed earnings gaps. Looking at iner levels o ethnic disaggregation conirs that ethnic earnings dierentials are systeatically saller that gender dierentials. Key words: earnings equations, gender wage gap, ethnic wage gap, West Arica Résué Dans cette étude nous analysons le poids et les déterinants des diérentiels de réunérations entre genre et groupes ethniques dans sept étropoles d Arique de l Ouest (Abidjan, Baako, Cotonou, Dakar, Loé, Niaey and Ouagadougou), en obilisant une base de données unique et paraiteent coparable, provenant des enquêtes réalisées dans les sept villes en 2001 et Cette question soulève un certain nobre de questions éthodologiques que nous tentons de traiter en détail, notaent en tenant copte des diérences de coposition ethnique et de genre entre les secteurs public, privé orel et inorel qui sont susceptibles de jouer sur les écarts de revenus. Les résultats ettent en évidence l existence d un déicit systéatique de réunération pour les ees, les caractéristiques des eplois expliquant oins de la oitié de ces écarts. A contrario, les groupes ethniques ajoritaires ne seblent pas bénéicier d une situation avantageuse et les écarts de revenus suivant le groupe ethnique sont relativeent aibles par rapport à ceux que l on observe suivant le genre. Quel que soit le signe de ce diérentiel (positi ou négati), la contribution expliquée par les caractéristiques observées de l eploi varie très sensibleent d une ville à l autre. Les estiations ontrent qu une grande partie de l écart de revenu selon le genre provient de l allocation sectorielle, et que cette dernière est toujours déavorable aux ees. En revanche, dans le cas des écarts suivant le groupe ethnique, la distribution par secteur institutionnel joue parois de açon positive dans le sens d une réduction des écarts. Finaleent, une désagrégation plus ine des groupes ethniques, au-delà de la partition ajoritaire/inoritaire, conire que l entrée ethnique est systéatiqueent oins signiicative sur les revenus du travail que le genre. Mots-clé : Equation de gain, écart de salaire, décoposition, genre, Ethnie, Arique de l'ouest JEL Classiication: J31, J71, O15, O55 2

3 Contents 1. INTRODUCTION DATA, CONCEPTS AND METHODOLOGY Data and concepts Wage gap decoposition techniques Earnings deterination Oaxaca and Neuark s traditional earnings decopositions Earnings decopositions with saple selectivity A ull sectoral decoposition Earnings gap decoposition or ethnic groups RESULTS A Neuark decoposition o gender and ethnic earnings gaps A ull decoposition o the gender earnings gap Ethnic earnings dierentials CONCLUSION REFERENCES APPENDICES Appendix 1. Ethnicity in West Arican countries (1-2-3 Surveys) Appendix 2. Nuber o working individuals in the saple with non zero earnings List o igures Figure 1: Herindhal concentration indices o Ethnolinguistic ractionalization (ELF) List o tables Table 1: Neuark decopositions o gender and ethnic earnings gaps Table 2: Full decoposition o the gender earnings gap without correcting or selectivity Table 3: Full decoposition o the gender earnings gap accounting or selectivity Table 4: Full decoposition o the ethnic earnings gap without correcting or selectivity Table 5: Full decoposition o the ethnic earnings gap accounting or selectivity Table 6a: Ethnic Earnings Dierentials Table 6b: Ethnic Earnings Dierentials Control Variables

4 1. Introduction Many studies have shown that woen and ethnic inorities ay ace unequal treatent in the labour arkets o both developed and developing countries copared to en or ajority ethnic groups (Altonji and Blank, 1999). In the case o Arica, there is in act little known about inequalities in labour arket outcoes and enhancing the gender and ethnic gap literature on the poorest countries is iportant or several reasons. First, there are aniest shortcoings o studies on Arican countries, particularly due to the shortage o available data (Bennell, 1996). Second, gender and ethnic inequalities are likely to be greater when arkets do not unction eiciently and the states lack resources or introducing corrective policies. Third, understanding the roots o inequalities between the sexes and ethnic groups and reducing the gender and ethnic gap could help design poverty reducing policies in these countries. Under the Poverty Reduction Strategy Paper (PRSP) initiative that concerns over sixty o the world's poorest countries, policies designed to counter gender discriination are aong the recoended solutions to reduce poverty: Goal 3 o the Millenniu Developent Goals (MDG) is speciically aied at reducing gender inequalities. In order to put this recoendation into practice, one needs to understand whether dierences in labour outcoes ste ro dierences in characteristics or ro dierences in the returns to these characteristics. These would indeed require dierent sets o policies. In Sub-Saharan countries, the deterioration o the labour arkets as well as the partial reeze on public sector recruitent ro the id-1980s ay have accentuated the circustances (i.e. labour arket entry and exit) that could give rise to gender and ethnic inequalities in the labour arket. While there is a sizeable nuber o papers dealing with ethnic and gender wage gap in developed countries (Altonji and Blank, 1999; Blau and Kahn, 2000), we can iner ro Weichselbauer and Winter-Eber (2005) s eta-analysis that only 3 percent o the studies on gender wage gap ste ro Arican data out o all the epirical literature since the 1960s. The existing literature 1 on gender gaps indicates that there is a wide consensus on the 1 See, notably, Glewwe (1990) or Ghana; Cohen and House (1993) or Sudan; Milne and Neitzert (1994) and Agesa (1999) or Kenya; Glick and Sahn (1997) or Guinea; Lachaud (1997) or Burkina and Caeroun; Aritage and Sabot (1991) or Kenya and Tanzania; Appleton, Hoddinott and Krishnan (1999) or Uganda, Côte d'ivoire and Ethiopia; Iseonger and Roberts (1999) or South Arica; Siphabe and Thokweng-Bakwena (2001) or Botswana; Kabubo-Mariara (2003) or Kenya; Teesgen (2006) or Ethiopia; Kolev and Suarez Robles (2007) or Ethiopia; Nordan and Roubaud (2009) and 4

5 iportance o inequalities between en and woen, both or salaried and sel-eployed workers. For instance, in Guinea, Glick and Sahn (1997) ind that dierences in characteristics account or 45 percent o the ale-eale gap in earnings ro seleployent and 25 percent o the dierences in earnings ro public-sector eployent while, in the private sector, woen actually earn ore than en. Aritage and Sabot (1991) also ind that such gender inequality exists in the public sector o Tanzania but observed no gender discriination in Kenya's labour arket. The latter result holds true both or the public and private sectors o the Kenyan econoy. Siilarly, Glewwe (1990) ound no wage discriination against woen in Ghana. On the contrary, eales see better o than ales in the public sector. More recently, Siphabe and Thokweng-Bakwena (2001) show that in the public sector o Botswana ost o the wage gap is due to dierences in characteristics between en and woen and not to discriination. On the other hand, in the private sector, ost o the wage gap is due to discriination. Likewise, in Uganda and Côte d'ivoire, Appleton et al. (1999) ind evidence that the public sector practises less wage discriination than the private sector. A siilar result is obtained or Madagascar by Nordan and Roubaud (2009) who evidence a gender wage gap to the advantage o woen in the state sector. Fro all the existing studies, one can hardly conclude on the existence o a coon cross-country pattern in the relative agnitudes o the gender wage gaps in the public and private sectors. However, the ain reason or this diversity in results ight coe ro the great heterogeneity in the data sources used by the dierent authors (either labour orce or household surveys undertaken or other purposes than labour arket issues), in the period they consider and also in the ethodology they ipleent. Concerning the ethnic wage gap, the literature is even scarcer. Barr and Oduro (2000) ind or Ghana that a signiicant proportion o earnings dierentials between ethnic groups can be explained by standard observed workers characteristics. On the other hand, the question o the role o ethnolinguistic ractionalization on developent has received uch ore attention. For instance, Easterly and Levine (1997) conclude that Arica s growth tragedy is in part related to its high level o ethnic diversity, resulting in poor institutional unctioning. This result is, however, still debated (see Bossuroy, 2007 or a discussion). Nordan, Rakotoanana and Robilliard (2009) or Madagascar; Nordan and Wol (2009a) or Morocco and Nordan and Wol (2009b) or the oral sectors o Madagascar and Mauritius. 5

6 In this paper, we cast soe new light on these issues by using labour orce surveys carried out in seven econoic capitals o rancophone West Arica. The contribution o this study stes ro at least two ain advantages. First, the data used were collected using identical sapling ethod and virtually identical questionnaires in each city in the sae period o tie ( ), aking or totally coparable results. 2 Second, we analyse both gender and ethnic gap issues using the sae ethodological approach or each city. Given that these are the two ost iportant individual characteristics expected to give rise to discriination, we believe it is interesting to copare the agnitude o discriination, i any, against eales and against inority or other ethnic groups. The reainder o the paper is divided as ollows. Section 2 discusses the data, concepts and econoetric ethods used. In section 3 we coent on the results. Finally, in section 4, we draw together the ain indings and conclude. 2. Data, Concepts and Methodology In this section, we irst present the data and concepts used in this study beore discussing the ethodology o earnings decopositions, an essential aspect o our investigation o the gender and ethnic disparities in the West Arican labour arkets Data and concepts Our data are taken ro an original series o urban household surveys in West Arica, the Surveys conducted in seven ajor WAEMU 3 cities (Abidjan, Baako, Cotonou, Dakar, Loe, Niaey and Ouagadougou) ro 2001 to The surveys were carried out by the relevant countries National Statistics Institutes (NSIs), AFRISTAT and DIAL as part o the PARSTAT Project. 4 The surveys cover the econoic city, i.e. the adinistrative city and all the sall towns and villages directly attached to it and with which there are requent exchanges. As suggested by its nae, the Survey is a three-phase survey. The irst phase concerns individuals socio-deographic characteristics (including education and literacy) 2 For another coparative analysis with these data, see Kuepie et al. (2009). 3 WAEMU: West Arican Econoic and Monetary Union. The survey was not carried out in Guinea- Bissau. 4 Regional Statistical Assistance Prograe or ultilateral onitoring sponsored by the WAEMU Coission. 6

7 and labour arket integration. The second phase covers the inoral sector and its ain productive characteristics. The third phase ocuses on household consuption and living conditions. The sae ethodology and virtually identical questionnaires were used in each city, aking or totally coparable indicators. Our study uses only the Phase 1 data. Phase 1 o the Survey is a statistical eployent survey designed to (i) provide the ain indicators to describe the situation o individuals and households on the labour arket. It covers household eployent and econoic activities, especially in the inoral sector; (ii) serve as a ilter survey to identiy a representative saple o inoral production units, which are then surveyed in Phase 2. The saple surveyed in Phase 1 has a total o 93,213 individuals (17,841 households) with country saple sizes varying ro 9,907 individuals in Togo to 19,065 individuals in Senegal. All individuals are asked about their ethnic group. The groups obviously dier between countries: the nuber o groups taken into consideration in the questionnaire varies ro 9 in Benin and Niger to 40 groups in Togo. However, in order to haronize the data and the nuber o categories considered, the 40 Togolese groups and the 18 Ivorian groups were reduced to 6 groups. Non response appears to be the exception with only 665 issing or Does not know answers Wage gap decoposition techniques In the ollowing, we discuss earning equations estiations and present ethods that are traditionally used to decopose gender wage gaps. A ull decoposition ethod à la Appleton, Hoddinott and Krishnan (1999) is also presented. The application o these ethods to decoposing ethnic wage gaps is then discussed Earnings deterination Traditional gender earnings decopositions rely on estiations o Mincer-type earnings unctions or en and woen o the or: where ln w = β + ε (1) i x i i ln wi is the natural logarith o the observed hourly earnings or individual i, x i is a vector o observed characteristics, β is a vector o coeicients and ε i is a disturbance ter with an expected value o zero. 7

8 Earning unctions are irst estiated separately or ales and eales, and also or the dierent sectors. There is no universally accepted set o conditioning variables that should be included or describing the causes o gender labour arket outcoes dierentials. Yet, the consensus is that controls or productivity-related actors such as education, labour arket experience and arital status should be included. However, it is debatable whether job characteristics, occupation and industry should be taken into account: i eployers dierentiate between en and woen through their tendency to hire into certain occupations, then occupational assignent is an outcoe o eployer practices rather than an outcoe o individual choice or productivity dierences. 5 In this paper, it is not possible to account or the workers actual experience in the labour arket, but only or potential experience which can be viewed as relecting the gross tie that individuals have spent while in the labour orce (easured as age inus years o schooling inus six the legal age at school entry). This is a possible liitation o our study since, as argued in the epirical literature, dierences in labour orce attachent across gender are iportant to explain the extent o the gender wage gap. Indeed, easures o woen s work experience are particularly prone to errors given their discontinuity in labour arket participation (or child baring and care or instance). Using proxy easures such as potential experience ay thus lead to overestiate the aount o experience or eales, while it ight be a good approxiation o true experience or en with higher labour orce attachent (Nordan and Roubaud,2009). 6 Concerns arise over possible saple selection biases in the estiations. Strictly speaking, there are two sources o selectivity bias involved. One arises ro the act that earnings are only observed when people work, and not everyone is working. The second coes ro the selective decision to engage in public wage eployent rather than private wage eployent or the inoral sector. Here, we address both issues using Lee s two-stage 5 Conversely, one can argue that analyses that oit occupation and industry ay underestiate the iportance o background and choice-based characteristics on labour arket outcoes (Altonji and Blank, 1999). 6 Regan and Oaxaca (2006) show that using potential versus actual experience in earnings odels is best viewed as a odel isspeciication proble rather than a classical errors-in-variable raework. Instruental variable techniques are the traditional approach taken to correct classical easureent error. Then, as underline Regan and Oaxaca (2006), in the absence o actual experience easures, instruenting potential experience would not solve the odel speciication proble. 8

9 approach to take into account the possible eect o endogenous paid-work participation and sector allocation on earnings (Lee, 1983) 7. In the irst stage, ultinoial logit odels o individual i s participation in sector j are used to copute the correction ters λ ij ro the predicted probabilities P ij. The dierent odalities considered in the ultinoial logit are: non-paid work participants, public sector workers, oral private sector workers and inoral sector workers. A potential proble is that the ultinoial logit ay suer ro the Independence o Irrelevant Alternatives assuption (IIA), which in ost cases is questionable. We perored Hausan-type tests (Hausan and McFadden, 2004) or each city and sector which assively provide evidence that the IIA assuption is not violated, with the exception o the inoral sector in BaakoIn Lee s procedure, identiication is achieved by the inclusion o additional individual variables in the irst stage selection equations which are oitted in the second stage earnings regressions: a set o duies indicating relationship to the household head, the dependency ratio (nuber o non working age individuals divided by the total nuber o individuals in the household), and the household size. 8 Our assuption is that these variables have arguably no reason to inluence earning levels Oaxaca and Neuark s traditional earnings decopositions The ost coon approach to identiying sources o gender wage gaps is the Oaxaca- Blinder decoposition. In this approach, two separate standard Mincerian log earnings equations are estiated or ales and eales. The Oaxaca decoposition is: where ln w ln w = β ( x x ) + ( β β ) x (2) w and w are the eans o ales and eales earnings, respectively; x and are vectors containing the respective eans o the independent variables or ales and eales; and β and β are the estiated coeicients. The irst ter on the right hand side captures the earnings dierential due to dierent characteristics o ales and eales. The x 7 Following Tunali (1986), an alternative approach would be to eploy a sequential selection rule (nested ultinoial logit) rather than a cobined one. This eans controlling or sel-selection into the paid-work group and then dierent endogenous choices between the public, oral private and inoral sectors. This technique requires inding a least one variable aecting the decision to enter the paid-work group but not the sector choice in order to achieve identiication via the use o restriction exclusion. Unortunately, in our data, it sees ipossible to ind variables that ay be used in the irst stage selection equation and arguably be excluded ro a second selection equation o sector allocation. 8 Siilarly, in the sae context o a two-step sectoral selection correction, Appleton et al. (1999) use the proportion o children in the household as an identiying instruent. 9

10 second ter is the earnings gap attributable to dierent returns to those characteristics or coeicients. It can be argued that, under discriination, ales are paid copetitive wages but eales are underpaid. I this is the case, the ale coeicients should be taken as the nondiscriinatory wage structure, as in equation (2). Conversely, i eployers pay eales copetitive wages but pay ales ore (nepotis), then the eale coeicients should be used as the non-discriinatory wage structure. Thereore, the issue is how to deterine the wage structure β that would prevail in the absence o discriination. This choice poses the well-known index nuber proble given that we could use either the ale or the eale wage structure as the non-discriinatory benchark. While a priori there is no preerable alternative, the decoposition can be quite sensitive to the selection ade. The literature has proposed dierent weighting schees to deal with the underlying index proble. In this paper, we rely on the general decoposition proposed by Neuark (1988) which can be written as ollows: ln w lnw = β ( x x ) + [( β β ) x + ( β β ) x ] (3) This decoposition can be reduced to Oaxaca s two special cases i it is assued that there is no discriination in the ale wage structure, i.e. β = β, or i it is assued that β = β. Neuark shows that β can be estiated using the weighted average o the wage structures o ales and eales and advocates using the pooled saple to estiate β. The irst ter is the gender wage gap attributable to dierences in characteristics. The second and the third ters capture the dierence between the actual and pooled returns or en and woen, respectively Earnings decopositions with saple selectivity Neuan and Oaxaca (2004) show that saple selection coplicates the interpretation o earnings decopositions. They oer several alternative decopositions, each based on dierent assuptions and objectives. We use one o the that consist in considering selectivity as a separate coponent. This technique has the advantage o not calling or any prior hypothesis regarding the links between individual characteristics and selectivity. An additional ter in the decoposition easures the contribution o selection eects to the observed gender earnings gap, ˆ θ ˆ λ ˆ θ ˆ λ, where λˆ and θˆ denote respectively the ean 10

11 correction ter (generalised Mill s ratio) and its estiated coeicient ro each regression by sex. Hence, in the ull sectoral decoposition that ollows, when trying to account or saple selectivity, we will consider the decoposition o oered earnings instead o actual earnings, i.e. earnings net o the selection eects ˆ θ ˆ λ (see Reier, 1983) A ull sectoral decoposition While the iproveent proposed by Neuark s decoposition is attractive, it is not iune ro coon criticiss o decoposition ethods in general. One o the is that, without evidence that eployers care only about the proportion o each type o labour eployed, it is not clear that the pooled coeicient is a good estiator o the nondiscriinatory wage structure. Appleton et al. (1999) s ull sectoral decoposition takes into account sectoral structures dierences between genders by using a siilar approach to that o Neuark and decoposing the gender earnings gap into three coponents. Let W and W be the eans o the natural logs o ale and eale earnings and p j and p j be the saple proportions o en and woen in sector j respectively. Male and eale ean earnings can be written as the su o sectoral earnings weighted by the proportion o workers in each sector : W W = = 3 j= 1 3 j= 1 W W j j p p j j As a result, one can decopose the dierence in ean earnings into intrasectoral earnings dierences and dierences in proportions eployed in the dierent sectors. In order to overcoe the index proble, Appleton et al. (1999) assue a sectoral structure that would prevail in the absence o gender dierences in the ipact o characteristics on sectoral choice. Let p j be the proportion o workers in sector j under this assuption. They then decopose the dierence in ean earnings such as: W (4) W = p j ( Wj W j ) + Wj ( pj p j ) + j= 1 j= 1 j= 1 W j ( p j p j ) The irst ter can be decoposed using the Neuark decoposition presented earlier. The second and third ters can urther be decoposed in order to set apart dierences arising 11

12 ro dierences in observable characteristics and dierences arising ro dierences in returns to observable characteristics. In order to do so, one can derive the average probability to be eployed in a given sector or ale and eale workers ro the estiation o pooled and separate ultinoial logit odels or en and woen. These ean probabilities are denoted by p j and p j respectively. Ebedding the sel-selection process in (4), the ull decoposition can then be written in the ollowing way: W W = p j ( xj x j ) β j + p j xj ( βj β j ) + j= 1 j= 1 j= 1 3 j= 1 W j ( p j p j ) W j ( p j p j ) + j= 1 j= 1 W j p ( p j j x j ( β β ) p j j ) + 3 j j= 1 W j ( p j p j ). (5) The irst three ters are siilar to Neuark decopositions o within-sector earnings gaps. The ourth and ith ters easure the dierence in earnings due to dierences in distribution o ale and eale workers in dierent sectors. The last two ters account or dierences in earnings resulting ro the deviations between predicted and actual sectoral copositions o en and woen not accounted or by dierences in characteristics Earnings gap decoposition or ethnic groups Extending decoposition ethods developed and traditionally used to analyse possible discriination against woen to the study o earnings dierentials between ethnic groups is not straightorward. One o the ain probles is related to the deinition and easureent o ethnicity: what deines an ethnic group? In developed countries, there exist conlicting views and dierent traditions regarding the collection o data on ethnic origin: while Anglo- Saxon societies are used to easuring and analysing data on so-called racial or ethnic groups, a nuber o countries reuse to categorize individuals using ethnic or racial criteria 9 and, as a result, do not collect statistical data on ethnic origin. In Arica, the notion o ethnicity also raises a nuber o questions that have been extensively debated aong social scientists (see or instance Bayart, 1989). Works by anthropologists have indeed shown that, contrary to a naïve a priori, ethnic groups are not characterized by the genetic hoogeneity o their ebers. Depending on countries and contexts, the constitution o ethnic groups appears to be ore or less recent and their deinition is oving. While soe groups have their origin in a coon yth and/or ancestor, others only share a coon language and culture, and soe have been constructed ro outside, i.e. by other groups, either upon a 9 In France, the collection o data on ethnic origin is subject to the authorization o a governent body and is not granted systeatically. Recently, a survey designed to study racial discriination in the labour arket gave rise to a strong opposition ro French public opinion. 12

13 igration or invasion event, or through an exogenous categorization constructed and iposed by colonial rulers. Despite their various origins, it is widely aditted that the notion o ethnicity plays a certain role in the social relations o any Arican countries. There is, or instance, strong evidence o high levels o endogay, not only in rural areas where ethnic hoogeneity is oten observed at the local level, but also in urban areas where dierent ethnic groups usually cohabit. In the past ten years, econoists have seized the ethnic issue around the question o its ipact on developent and growth. The seinal paper is Easterly and Levine s contribution (1997) that concludes that Arica s growth tragedy is in part related to its high level o ethnic diversity, resulting in poor institutional unctioning. However, this result is still debated nowadays (see Bossuroy, 2007 or a discussion). In this paper, we ocus on the ipact o ethnicity on labour arket outcoes easured through earnings. In order to apply the ethods developed or the analysis o the gender earnings gap, one is inclined to construct a dichotoous variable identiying either a possibly avoured or discriinated against ethnic group. Data collection on ethnicity at the household or individual level is coon in Arica: ost household and eployent surveys include a variable indicating the ethnic group. However, given the diversity o national contexts, two diiculties arise: the irst one is related to identiying a priori a discriinated ethnic group: should one consider the ajority ethnic group as avoured? Or should one consider instead the group related to the head o state? The second diiculty arises because o our coparative raework: how does belonging to the dierent groups copare across countries? For instance, i one considers ajority ethnic groups in the cities o the Surveys, is it the sae to be a Mossi in Ouagadougou (76.6 percent o the population) and a Babara in Baako (34.0 percent o the population)? Although we do not attept to answer this question in the paper, we try to consider various aspects o possible ethnic discriination on urban labour arkets while keeping in ind the dierent national contexts. 13

14 3. Results 3.1. A Neuark decoposition o gender and ethnic earnings gaps In this section, gender and ethnic earnings gaps are analysed using traditional decoposition approaches. As entioned earlier, in order to apply these ethods to decopose the ethnic earnings gap, one is inclined to construct a dichotoous variable identiying either a possibly avoured or discriinated against ethnic group. For that purpose, we identiy a ajority ethnic group in each city. Descriptive statistics indicate that these ajority ethnic groups represent an absolute ajority o the capital s population in three countries out o seven. 10 More precisely: - the Fon represent 60.9 percent o the population o Cotonou (Benin); - the Mossi represent 78.2 percent o the population in Ouagadougou (Burkina Faso); - the Akan represent 34.2 percent o the population in Abidjan (Cote d Ivoire); - the Babara represent 34.4 percent o the population o Baako (Mali); - the Djera represent 49.5 percent o the population o Niaey (Niger); - the Wolo represent 40.4 percent o the population o Dakar (Senegal); - the Ewe-Mina-Wachi represent 74.2 percent o the population o Loe (Togo). In six cities out o seven, the ajority ethnic group corresponds to the ajority group at the national level. The only exception is Niger where the ajority ethnic group in the capital is the Djera while it is the Haoussa at the national level (54 percent o the population). A irst look at earnings gap decopositions based on gender and ajority ethnic groups is provided in Table 1 which reports a decoposition o earnings gaps based on Neuark s approach (see section 2.2.2). A nuber o results are worth ephasizing. Raw gender earnings gaps are large, signiicant and vary ro 50.0 in Niaey to 79.2 in Abidjan: these igures indicate that eales in Niaey (resp. in Abidjan) earn on average 50.0 percent (respectively 20.8 percent) o ale earnings. Gender dierences in the distribution o characteristics related to productivity such as education and experience usually explain less than hal o the raw gender gap in six cities out o seven: Loe is an exception with dierences in characteristics explaining alost 55 percent o the gap. Including variables related to the type o occupation decreases soewhat 10 See Appendix 1 or ore details on the ajority ethnic group in each country. 14

15 the unexplained share o the raw gender gap. This decrease appears to be substantial in Ouagadougou, Abidjan and Loe. Contrary to the systeatic avourable position o en with respect to woen, ajority ethnic groups do not appear to have a systeatic avourable position in the urban labour arkets o our saple o countries. It is only in Abidjan and Dakar that the gap appears both signiicant and avourable or the ajority ethnic group: in Abidjan, the Akan earn on average 28.0 percent ore than other ethnic groups while in Dakar, the Wolo earn on average 6.8 percent ore than other ethnic groups; on the contrary, ajority ethnic groups in Ouagadougou, Baako and Loe earn signiicantly less on average than other ethnic groups. Concerning the decoposition o ethnic earnings gaps, results dier arkedly. In Abidjan, the results indicate that dierences in the distribution o individual characteristics explain ore than 85 percent o the gap so that little is let or what could be labelled discriination (the unexplained share) against non ajority ethnic groups. In Dakar, on the contrary, 100% o the gap is let unexplained until job characteristics related to occupation and sector are introduced. In Ouagadougou, where the ajority ethnic group (Mossi) receives lower earnings than other groups, the gap is also in large part explained by dierences in the distribution o observable characteristics such as education and experience; as a result, the unexplained share is low at 20.0 percent; in Baako, the unexplained share o the gap against the ajority ethnic group (Babara) is uch higher: there, dierences in returns to characteristics account or 43.4 percent o the gap (down to 39 percent once occupation status duies are included in the regressions) A ull decoposition o the gender earnings gap It is widely acknowledged that there are at least our types o labour arkets in ost developing countries: rural (or agricultural), public, oral private and inoral. These arkets each have their own characteristics, such as job seasonality, uncertainty o deand, nature o contracts and structure o wages and earnings. As a result, gender and ethnic labour allocation between these sectors can be expected to contribute to earnings gaps. Following Appleton et al. (1999) and Nordan and Roubaud (2009), we provide coparable estiates o the size and deterinants o gender earnings gaps using the decoposition ethod described in section Given that we are analysing urban labour arkets, only three types o labour arkets are taken into consideration: public, oral private and 15

16 inoral. Results are reported in Tables 2 and 3, without and with correction or selectivity o participation and sectoral allocation (see section 2.2.3). Within-sector dierences in earnings account or the largest share o the gender gap with contributions ranging ro 60.2 percent in Abidjan to 74.8 percent in Cotonou. The reainder can then be attributed to gender dierences in proportions o workers in each sector. The positive su o these three ters or all cities iplies that the dierences in sectoral locations are ore avourable to en than to woen. For instance, the gender earnings gap would have been 40 percent saller respectively in Abidjan i en and woen had been equally distributed across the three sectors. This is because ewer woen than en are located in the higher paying sectors such as the public and private oral sectors. Dierences attributable to characteristics only account or a relatively sall share o the within-sector dierences in earnings: their contribution varies ro 10 percent in Dakar to 41 percent in Loe (as a share o the contribution o within-sector dierences; not shown in the table). Conversely, dierences attributable to characteristics account or a very large share o the sectoral location dierences between genders: their contribution varies ro 65 percent in Dakar to 85 percent in Cotonou and Baako. Concerning dierences attributable to deviation in ale and eale returns, their contribution to within-sector dierences in earnings are o the sae order, indicating that both discriination against woen and nepotis in avour o en contribute to the gender earnings gap; both discriination against woen and nepotis in avour o en also contribute to dierences in sectoral location but at a uch lower level. Taking into account selectivity leads to analysing the decoposition not o actual earnings but o oered earnings. These are coputed using the coeicients o the selection ter in the earnings equations (see section 2.2.3). Results in Table 3 show that oered earnings gaps are uch higher in Cotonou, Baako and Dakar, while they are lower in the other cities. Higher earnings gaps when sectoral selectivity is accounted or are not systeatically associated with higher contribution o sectoral location dierences however. Except in Niaey, withinsector earnings dierences reain the ain contributor to gender gaps. Concerning ethnic earnings gaps, our results in Table 4 (without correcting or selectivity) indicate that: In Ouagadougou the gap can alost evenly be attributed to within-sector earning dierence (46.7%) and to sectoral location (53.2%). In Abidjan, it is dierences in sectoral location that explain the highest share o the gap (86.1 percent) o which 75 percent 16

17 are accounted or by dierences in characteristics. In Baako, within-sector dierences in earnings account or 77.4 percent o the earnings gap out o which 33.3 percent are attributable to dierences in characteristics; both nepotis (15.9 percent) and discriination (28.2 percent) signiicantly contribute to the gap through their contribution to within-sector dierences in earnings. On the contrary, sectoral location dierences are alost entirely explained by dierences in characteristics. In Loe, the gap is also explained by sectoral location dierences but, contrary to Baako, the deviation in the eect o characteristics on location explains a big share o sectoral location dierences. Contrary to the results obtained or gender, where sectoral location systeatically increases the gap against woen, it is in soe cities the case that sectoral location plays a copensating role against observed earnings gaps. Results reported in Table 5 show that taking into account selectivity leads to reassessing soe easures o the gaps. The gap decreases or Ouagadougou, Abidjan and Dakar and increases or Baako. For these cities, the decoposition results appear however relatively stable. In Loe, the gap is actually reversed, a possible indication that the ajority ethnic group is oered on average higher earnings than the other ethnic groups. This result is soewhat puzzling and would require urther investigation. For instance, in order to understand the eatures o earnings negotiations, one would need to know the ethnic group o the eployer Ethnic earnings dierentials In this section, we exaine earnings dierentials between ethnic groups. As entioned earlier, several ethnic groups can be dierentiated in each capital. The highest nuber o groups is in Baako (11 groups), ollowed by Ouagadougou (10 groups), Cotonou, Niaey and Dakar (9 groups) and Abidjan and Loe (6 groups). Figure 1 reports two Herindhal s concentration indices or ethnolinguistic ractionalization (ELF) in each country: the irst one is coputed at the national level while the second is coputed at the level o capitals using the Surveys. Levels are siilar across countries except or Burkina Faso where the ELF index appears to be uch lower in the capital than at the country level. This could ste ro the act that the ethnic ajority group (Mossi) represents 78.2 percent o the population in Ouagadougou and only 50 percent at the national level. This points to a actor that can explain why ajority ethnic groups are not systeatically avoured in the labour arkets o 17

18 our saple: indeed, in the case o Burkina, where Mossi have lower average earnings than other ethnic groups, it could be the case that only the better peroring non-mossi actually igrate to the capital. This is consistent with the results o the Neuark decoposition o the ethnic earnings gap in section 3.1. where we ind that the gap against the Mossi is ainly explained by dierences in characteristics. Coeicients o the duies indicating each ethnic group in city-level earnings equations regressions are reported in Table 6a. In the irst colun, ethnic group duies are the only regressors while a set o usual controls is introduced in the speciication reported in the second colun (coeicients or these variables are reported in Table 6b). Results show two things: irst, there is at least one signiicant coeicient on ethnic duies in all the cities o the saple eaning that there exist dierences in average earnings between ethnic groups. However, ost o these dierences diinish and, in soe cases, vanish once other observables characteristics are controlled or. In the case o Cotonou, both the Dendi people and the Yoruba appear avoured ceteris paribus with respect to the ajority ethnic group (Fon), while the Yoa have lower earnings ceteris paribus than the Fon. In Ouagadougou, the group o other andingues as well as the Senouo people are avoured ceteris paribus copared to the Mossi. In Abidjan, both the Volta people and the natives ro Burkina Faso have lower earnings ceteris paribus than the ajority ethnic group (Akan). In Baako, both the Peul and the Sarakole are avoured with respect to the ajority ethnic group (Babara). In Niaey, once control variables are included, only the Haoussa appear less avoured copared to the Djera. In Dakar, both the Serere and Diola people have lower earnings ceteris paribus than the Wolo. 11 Overall ajority ethnic groups see not be avoured on the labour arket once one controls or productivity related individual characteristics. On the contrary, soe inority groups actually have higher earnings ceteris paribus. This is the case in Benin, Burkina, and Mali. However, none o the avoured groups see to be related to the ethnicity o the head o state at the tie o the survey Despite the act that soe earnings dierentials hold ceteris paribus, one should note that in soe cases, the groups considered represent very sall shares o the population (see Appendix 2). Consequently, the question o the size o our saples or analysing the characteristics o these groups can be raised. This is one o the reasons why we did not ipleent decoposition ethods at this level o ethnic disaggregation. 12 Head o state ethnicity at the tie o the survey is provided in the dataset put together by Fearon, Kasara and Laitin (2007). 18

19 4. Conclusion In this paper, we analyse the size and deterinants o gender and ethnic earnings gaps in seven West Arican capitals. The study is based on a unique dataset taken ro an original series o urban household surveys in West Arica, the Surveys conducted in seven ajor WAEMU cities (Abidjan, Baako, Cotonou, Dakar, Loe, Niaey and Ouagadougou) ro 2001 to Analysing gender and ethnic earnings gaps in an Arican context raises a nuber o iportant issues that our paper attepts to address. First, international coparisons o earnings gaps are still scarce in Arica. Our surveys use identical ethodologies and virtually identical questionnaires in each city, aking or totally coparable results. Second, we address the issue o saple selectivity due to endogenous sector choices (public, private oral and inoral sectors) as gender and ethnic labour allocation between these sectors can be expected to contribute to earnings gaps. Following Appleton et al. (1999), we then provide coparable estiates o the size and deterinants o gender and ethnic earnings gaps using decoposition ethods that address the sectoral allocation issue. The results show that gender earnings gaps are large in all the cities o our saple and that gender dierences in the distribution o characteristics usually explain less than hal o the raw gender gap. By contrast, ajority ethnic groups do not appear to have a systeatic avourable position in the urban labour arkets o our saple o countries and observed gaps are sall relative to gender gaps. Moreover, none o the inority avoured groups see to be related to the ethnicity o the head o state at the tie o the survey. Whatever the sign o the gap, the contribution o dierences in the distribution o individual characteristics varies arkedly between cities. Taking into account dierences in sectoral locations in the decoposition o gender earnings gaps provides evidence that within-sector dierences in earnings account or the largest share o the gender gap and that the dierences in sectoral locations are always ore avourable to en than to woen. By contrast, concerning ethnic earnings gaps, the ull decoposition indicates that sectoral location soeties plays a copensating role against observed earnings gaps. Looking at iner levels o ethnic disaggregation conirs that ethnic earnings dierentials are systeatically saller that gender dierentials. 19

20 Reerences Agesa, Richard U. (1999), The Urban Gender Wage Gap in an Arican Country: Findings ro Kenya, Canadian Journal o Developent Studies, 20, no. 1: Altonji, Joseph G., Rebecca M. Blank (1999), Race and Gender in the Labor Market, in Orley Ashenelter and David Card, Handbook o Labor Econoics, Volue 3C, North Holland, Asterda, pp Appleton, Sion, John Hoddinott, Praila Krishnan (1999), The Gender Wage Gap in Three Arican Countries, Econoic Developent and Cultural Change, 47, no. 2: Aritage, Jane, Richard Sabot (1991), Discriination in East Arican's Urban Labor Market, In Unair Advantage: Labor Market Discriination in Developing Countries, ed. Nancy Birdsall and Richard Sabot, Washington D.C., The World Bank. Barr, Abigail, Abena Oduro (2000), Ethnicity and Wage Deterination in Ghana, World Bank Policy Research Working Paper No. 2506, Washington D.C., The World Bank. Bayart, Jean-François (1989) Le théâtre d'obres de l'ethnicité, in L'Etat en Arique, la politique du ventre, Fayard Paris, pp Bennell, Paul (1996), Rates o Return to Education: Does the Conventional Pattern Prevail in Sub-Saharan Arica?, World Developent, 24, no. 1: Blau, Francine, Lawrence Kahn (2000), Gender Dierences in Pay, Journal o Econoic Perspectives, 14: Blinder, Alan S. (1973), Wage Discriination: Reduced For and Structural Estiates, The Journal o Huan Resources, 8, no. 4: Bourguignon, François, Martin Fournier and Marc Gurgand (2007), Selection Bias Correction Based on the Multinoial Logit Model: Monte-Carlo Coparisons, Journal o Econoic Surveys, 21, no. 1: Bossuroy, Thoas (2007), Ethnicity as a Resource in Social Capital, ieo DIAL, Paris. Brown, Randall S., Marilyn Moon, Barbara S. Zoloth (1980), Incorporating Occupational Attainent in Studies o Male-Feale Earnings Dierentials, The Journal o Huan Resources, 15, no. 1:3-28. Cohen, Barney, Willia J. House (1993), Woen's Urban Labour Market Status in Developing Countries: How Well Do They Fare in Khartou, Sudan?, Journal o Developent Studies, 29, no. 3: Easterly, Willia, Ross Levine (1997), Arica s Growth Tragedy: Policies and Ethnic Divisions, Quaterly Journal o Econoics, 111, no. 4: Fearon, Jaes D. (2003), Ethnic and Cultural Diversity by Country, Journal o Econoic Growth, 8, Fearon, Jaes, Kiuli Kasara, David Laitin (2007) Ethnic Minority Rule and Civil War Onset. Aerican Political Science Review 101 (1). Glewwe, Paul (1990), Schooling, Skills and the Return to Education: an Econoetric Exploration Using Data ro Ghana, Living Standards Measureent Working Paper, no. 76, Washington D.C.: The World Bank. Glick, Peter, David E. Sahn (1997), Gender and Education Ipacts on Eployent and Earnings in West Arica: Evidence ro Guinea, Econoic Developent and Cultural Change, 45, no. 4:

21 Hausan, Jerry. A., Daniel MacFadden (1984), Speciication Tests or the Multinoial Logit Model, Econoetrica, 52, no. 5: Iseonger, A.G., Neil Roberts (1999), Post-entry Gender Discriination in the South Arican Labour Market, Journal or Studies in Econoics and Econoetrics, 23, no. 2:1-25. Kabubo-Mariara, Jane (2003), Wage Deterination and the Gender Wage Gap in Kenya: Any Evidence o Gender Discriination?, Research Paper 132, Arican Econoic Research Consortiu, Nairobi. Kolev, Alexandre, Pablo Suarez Robles (2007), Addressing the Gender Pay Gap in Ethiopia: How Crucial is the Quest or Education Parity?, ieo, joint AFD-World Bank research project. Kuepie, Mathias, Christophe J. Nordan, François Roubaud (2009), Education and Earnings in Urban West Arica, Journal o Coparative Econoics, orthcoing. Lachaud, Jean-Pierre (1997), Les Fees et le Marché du Travail Urbain en Arique Subsaharienne, Editions l Harattan, 233 pages. Lee, Lung-Fei (1983), Generalized Econoetric Models with Selectivity, Econoetrica, 51, no. 2: Milne, Willia, Monica Neitzert (1994), Kenya, in Labor Markets in an Era o Adjustent: Issues Papers 1: , ed. Susan Horton, Ravi Kanbur and Dipak Mazudar, Washington D.C: EDI Developent Studies, World Bank. Neuan, Shoshana, Ronald Oaxaca (2004), Wage Decopositions with Selectivity- Corrected Wage Equations: A Methodological Note, The Journal o Econoic Inequality, 2, no. 1:3-10. Neuark, David (1988), Eployers Discriinatory Behavior and the Estiation o Wage Discriination, The Journal o Huan Resources, 23: Nordan, Christophe J., François Roubaud (2009), Reassessing the Gender Wage Gap in Madagascar: Does Labour Force Attachent Really Matter?, Econoic Developent and Cultural Change, 57, no. 4: Nordan, Christophe J., Faly Rakotoanana, Anne-Sophie Robilliard (2009), Gender Disparities in the Malagasy Labor Market, orthcoing in Gender Disparities in Arica s Labor Markets, Washington DC: The World Bank. Nordan, Christophe J., François-Charles Wol (2009a), Is there a Glass Ceiling in Morocco? Evidence ro Matched Worker-Fir Data, Journal o Arican Econoies, orthcoing. Nordan, Christophe J., François-Charles Wol (2009b), Islands Through the Glass Ceiling? Evidence o Gender Wage Gaps in Madagascar and Mauritius, orthcoing in Labor Markets and Econoic Developent, Ravi Kanbur and Jan Svejnar (eds), Routledge. Oaxaca, Ronald (1973), Male-Feale Wage Dierentials in Urban Labor Markets, International Econoic Review, 14, no. 3: Regan, Tracy L., Ronald L. Oaxaca (2006), Work Experience as a Source o Speciication Error in Earnings Models: Iplications or Gender Wage Decopositions, IZA Discussion Papers 1920, Bonn: Institute or the Study o Labor. Reiers, Cordelia W. (1983), Labour Market Discriination Against Hispanic and Black Men, The Review o Econoics and Statistics, 65, no. 4: