American Economic Review: Papers & Proceedings 2015, 105(5): 457–461 http://dx.doi.org/10.1257/aer.p20151003 100271 The Impact of Ethnic Diversity in Bureaucracies: Evidence from the Nigerian Civil Service† By Imran Rasul and Daniel Rogger* We document the correlation between the between public service delivery and the ethnic workplace diversity of bureaucracies and pub- diversity of civil service organizations.1 lic services delivered. We do so in the context The impact of bureaucratic diversity on of the Federal Civil Service in Nigeria, the most public service delivery in this context is not important government bureaucracy operating obvious. While the management literature sug- in a highly ethnically fractionalized society gests it might have positive impacts if it helps in which ethnicity is a salient form of identity bureaucrats solve problems etc., an estab- (Eifert, Miguel, and Posner 2010). lished macroeconomic literature documents An established literature documents the poten- a negative correlation between societal diver- tially beneficial impacts of workplace diversity sity and ­ economy-wide outcomes. Beginning on organizational performance. Theoretically, with Easterly and Levine (1997), a body of diversity can positively impact on organizations ­ cross-jurisdiction evidence has suggested that by matching workers of a variety of perspectives more ethnically diverse societies have lower or experiences, enhancing team creativity or growth, choose worse policies, and are more problem solving capacities. Lazear (1999) and likely to experience conflict (Alesina and La Hong and Page (2001), among others, develop Ferrara 2005). Our analysis begins to shed light models along these lines to make precise when on whether the positive channels through which workplace diversity beneficially impacts organi- diversity operates, as stressed in the manage- zational performance. Shore et al. (2009) over- ment literature (e.g., increased skill comple- view the evidence, much of which stems from mentary of workers in the production function), private sector firms in high income settings, and dominate the negative channels through which consider diversity along age-race-gender lines. diversity might operate, as stressed in the macro We provide novel evidence extending the empir- literature (e.g., divisions in preferences, free-rid- ical literature along all three margins: in the con- ing on other groups’ contributions, ineffective text of Nigeria, we document the relationship social sanctions, etc.). Understanding the role of diversity in gov- ernment bureaucracies is important because, as the state capacity literature has emphasized, the effective functioning of government bureau- * Rasul: University College London, Gower Street, cracy matters for poverty, inequality, and eco- London, WC1E6BT (e-mail: i.rasul@ucl.ac.uk); Rogger: nomic growth (Besley and Persson 2010). In World Bank, Washington, DC 20433 (e-mail: drogger@ worldbank.org). We gratefully acknowledge financial sup- earlier work, Rasul and Rogger (2014)—hence- port from the Government of Nigeria; the International forth, RR—documented how the management Growth Centre (RA-2009-11-018), the Economic and Social practices bureaucrats operate under correlate Research Council (ES/G017352/1), the Institute for Fiscal with public service delivery. We build on that Studies, and the Royal Economic Society. We thank the Presidency and the Office of the Head of the Civil Service of analysis to examine the relevance of a key fea- Nigeria for their support. We are grateful to the many gov- ture of Nigerian society: that it is highly ethni- ernment officials who have assisted us during this project cally fractionalized. but are too numerous to name individually. We thank Arun Advani, Oriana Bandiera, and Nicholas Bloom for useful discussions. All errors remain our own. † Go to http://dx.doi.org/10.1257/aer.p20151003 to visit In Kenyan private sector settings, Hjort (2014) and 1  the article page for additional materials and author disclo- Macchiavello and Morjaria (2014) show ethnic divisions sure statement(s). impact productivity due to worker discrimination.  457 458 AEA PAPERS AND PROCEEDINGS MAY 2015 I.  Data and Results tree. A substantial fraction of civil servants are from the Niger-Congo family of ethnicities, but To study the link between public service are then split across ​ 38​ sub-families. Following delivery and ethnic diversity, we merge data on the ethnic diversity literature, for each tier of the outputs of government bureaucracies with ethnicity, we construct a standard measure of a survey to bureaucrats. Both data sources are ­ ethnolinguistic fractionalization in organization​ described in RR and here we just recap the core n​ , ​ EL​ F​n​    , (one minus the Herfindahl index of details. ethnicities). This is interpreted as the probabil- On outputs, we use project level data for​ ity that two randomly selected bureaucrats have 4,700​projects implemented by ​ 63​organizations different ethnicities, and so a higher ​ EL​F​n​ mea-   in the Nigerian civil service, including govern- sure implies a more diverse organization. We do ment ministries and other federal agencies. We so for each organization using these three tiers, exploit data coded from the OPEN initiative as well as based on Otite’s (1990) list. As shown in Nigeria, that traced, by project, the use and at the foot of Table 1, the average fractionaliza- impact of 1​ 0​percent of all federal Government tion of organizations is ​ 0.19​if we use the highest social sector expenditures approved in 2006– tier of family ethnic groupings, ​ 0.38​ using the 2007. Monitoring teams visited project sites 18 sub-family definition, ​ 0.60​ using the sub-sub- months after they were approved and recorded family definition, and ​ 0.65​ using Otite’s (1990) whether the project had started, and its stage of most disaggregated list. completion. Our outcome variable is a contin- Two points are of note. First, these latter two uous zero to one measure of project comple- indices are very close in magnitude to Posner’s tion rates. Projects are of 1 ​ 1​ types, including (2004) ​ ELF​ measure for Nigeria (​ 0.66​) that construction (boreholes, buildings, etc.) and is based only on “politically relevant” eth- ­nonconstruction types (procurement, training, nic groups. Second, the Nigerian bureaucracy etc.). appears far less diverse than the population, We measure the ethnic diversity of bureau- even using the most disaggregated ethnic group- crats through a survey fielded to a representative ings. For Nigeria as a whole, the E ​ LF​ is typi- sample of 4 ​ ,100​ civil servants, corresponding cally measured to be over 0​ .80​ (Alesina and La to ​ 13​ percent of all bureaucrats in the 6 ​ 3​ orga- Ferrara 2005). nizations we study. Civil servants were asked to The management literature stresses any ben- name their ethnicity directly. Such self-reports eficial impacts of workplace diversity on orga- have the advantage of measuring the identity nizational performance likely operate through most salient to the individual. However, such enhanced skill complementarity of workers. To direct elicitations do not necessarily conform probe this idea a little in our data, we examine to a common dictionary. To convert responses whether there are significant differences across to comparable ethnolinguistic groupings, we bureaucrats of different ethnicities, in their proceed in two steps. We first utilize Otite’s self-reported experiences on-the-job, beliefs (1990) comprehensive assessment of Nigeria’s about the job, and own motivation to originally ethnic groups, that outlines ​ 374​ unique eth- join the Nigerian civil service. nicities. In our data we are able to match On experiences, as part of the survey bureaucrat responses to 1 ​ 28​ ethnic groups administered to bureaucrats, individuals were from Otite’s (1990) list. Second, to determine asked, “think about recent projects and/or how ethnic groups relate to each other, we use programmes you worked on for this organiza- Blench (2012) to build an ethnolinguistic tree: tion. In what proportion of the projects have online Appendix Figure A1 represents the con- you had to face the following difficult chal- structed family tree. Each individual’s ethnicity lenges?” Responses were ordered as Never can thus be defined at three tiers: t ​ hree​families (​1​), ​ 1–25​ percent (​ ), through to 7 2​ ​ 5−100​ per- (Niger-Congo, ­Afro-Asiatic, Nilo-Saharan), ​ 17​ cent (​ ). We then use an OLS model to regress 5​ sub-families, and ​ 51​sub-sub-families. these responses on bureaucrat characteristics Table 1 describes how bureaucrats in the and a set of ethnicity dummies, as defined at Nigerian civil service are distributed across the sub-sub-family level. We then test whether families, sub-families, and sub-sub-families the ethnicity dummies are jointly signifi- of ethnicities according to this ethnolinguistic cant, conditional on these other individual VOL. 105 NO. 5 ETHNIC DIVERSITY IN BUREAUCRACIES 459 Table 1—Ethnicities of Civil Servants Family Sub-family Sub-sub-family Otite Niger-Congo 3,172 West Benue-Congo 2,477 6 26 Cross River 189 4 10 Plateau 147 9 19 Southern Bantoid 139 4 14 Atlantic 117 1 1 Adamawa 58 6 10 Kainji 27 2 8 Jukunoid 11 2 3 Kwa 3 1 1 Mande 2 1 1 Ijoid 1 1 1 Mambiloid 1 1 1 Afro-Asiatic 558 Chadic West A 476 4 18 Biu-Mandara A 64 5 10 Chadic West B 13 2 3 Semitic 5 1 1 Nilo-Saharan 43 Saharan 43 1 1 Number 3 17 51 128 ELF 0.19 0.38 0.60 0.65 (SD) (0.19) (0.26) (0.29) (0.28) Notes: The sample of civil servants described here are those who gave their ethnicity during the Civil Servants Survey, and their self-reported ethnicity could be located within the Otite (1990) categorization of ethnicities. This covers approximately 90 percent of all civil servants inter- viewed. The ethnolinguistic family structure relies on the coding of Blench (2012). We construct measures of ethnolinguistic fractionalization in organization n, for each level of aggregation of eth- nicities (by family, sub-family, sub-sub-family, and based on Otite’s (1990) disaggregated group- ings). This measure is one minus the Herfindahl index of ethnicities, and is interpreted as the probability that two randomly selected bureaucrats have different ethnicities. Hence, a higher eth- nolinguistic fractionalization measure implies a more diverse organization. ­characteristics.2 We just highlight those areas rules of government” (  p-value ​ ). Finally, 0.05​ of experience that significantly differ across we consider a proxy for civil servant’s intrinsic ethnicities. For example, bureaucratic reports motivation: we asked bureaucrats which factor significantly differ by ethnicity relating to had most influenced them to originally enter the whether officials have been under pressure to civil service. We define those that answered “the divert funds (the p-value on the joint F-test of chance to serve Nigeria” as being intrinsically ethnicity dummies is 0​ ), or to change proj- .01​ motivated. Around a third of officials give this ect specification ( p-value ​0.07​). answer, and we find this original motivation to Focusing next on bureaucrat beliefs, and join the civil service to differ significantly across again just highlighting a few areas where there ethnicities ( p-value ​0.02​). are significant differences, we find that bureau- These dimensions provide suggestive evi- crats significantly differ by ethnicity on their dence on how ethnic diversity might be bene- belief that “officials should work within a team ficial for bureaucratic organizations, implying to achieve duties” (  p-value ​ ), or their belief 0.01​ project completion rates might be higher in that it is “sometimes right to go around the strict more ethnically diverse organizations if such bureaucracies bring together civil servants with divergent experiences, beliefs, and motivations. 2  Individual controls include whether the official is male, Our empirical specification has as its unit of whether they are a manager, years of schooling, years in the observation project ​ i​of type ​ n​ j​in organization ​ . civil service, and years at the organization.  We estimate the following OLS specification, 460 AEA PAPERS AND PROCEEDINGS MAY 2015 Table 2—Ethnic Diversity and Project Completion Rates Family Sub-family Sub-sub-family Otite (1) (2) (3) (4) ELF (z-score) 0.09** 0.11*** 0.09*** 0.09*** (0.04) (0.03) (0.03) (0.02) Controls Project and organizational Fixed effects Project type Observations (clusters) 4,721 (201) Notes: Standard errors in parentheses are clustered by project type within organization. Project type fixed effects relate to whether the primary project classification is a financial, training, advocacy, procurement, research, electrification, borehole, dam, building, canal, or road proj- ect. Project controls include the budget, whether it is new or a rehabilitation project, and a measure of its complexity. Organizational controls include the logs of number of employees, total budget, and capital budget, the share of the workforce with degrees, with postgraduate qualifications, and management practice indices related to autonomy, incentives/monitoring, and other practices. Noise controls include interviewer dummies, indicators of the seniority, gender, and tenure of the managers who responded, the day of the week and time the interview was conducted, a dummy for whether the interview was conducted during Ramadan, the inter- view duration, and an indicator of the reliability of the information as coded by interviewers. *** Significant at the 1 percent level.  **  Significant at the 5 percent level.   *  Significant at the 10 percent level. where ​ y​ijn​ is the project completion rate and   Table 2 presents our results and shows ethnic ​​ ​ LF E n​ is the ethnolinguistic fractionalization in   diversity in bureaucracies matters: a one stan- the organization, dard deviation increase in the ethnolinguistic fractionalization of bureaucrats is associated y​ ​   γ1   =  ​ ijn​  ​ ​ F​ EL​ n​   β1  + ​  ​ ​ C​  P​  + ​ ijn​   β2​  ​ O​ Cn ​ ​   with around ​ 9​ percent higher project comple- tion rates. To put this in context, we note the λ​ + ​  + ​  j​ ϵijn ​ ​    .​ average project completion rate is ​ 46​ percent, and ​38​ percent of projects have a zero comple- P​ ​ C​  ijn​ and ​ O​ ​   ​ include project and organiza- Cn tion rate. Hence, in line with evidence from the tional characteristics, respectively.3 As many management literature based largely on private organizations implement project type ​ ­   , we con- j​ sector organizations operating in high income trol for project type fixed effects ​ λ  . Standard ​j​ economies, here we also find the ethnic diversity errors are clustered by project type-organization. of public sector organizations is positively cor- To ease interpretation, ​ EL​ F​   n​is normalized into a related with their performance.4 γ1 z-score so that ​   ​is the effect size of diversity on ​ project completion rates. II. Discussion We contribute to the literature on the impact of workplace diversity on organizational per- 3 ​ P​   ijn​ includes project complexity, log project bud- C​ formance. We do so in a public sector setting, ​ ​ get, and whether the project is a rehabilitation or not. O ​  ​ Cn considering organizations in the Federal Civil includes the log number of staff, log total organization bud- get, log capital budget, the proportions of officials with a Service in Nigeria, a society in which ethnicity college and postgraduate degree. We also control for three is the most salient dimension of diversity. Ethnic dimensions of management practice in organizations consid- ered in RR: the autonomy/flexibility given to bureaucrats in their daily task management; the provision of incentives/ 4  monitoring of bureaucrats; all other practices compiled into On the extensive margin, we find comparable impacts a third index. We also include “noise” controls related to the of diversity on whether a project starts. The main results are management interview.  robust to using a generalized linear model.  VOL. 105 NO. 5 ETHNIC DIVERSITY IN BUREAUCRACIES 461 diversity in bureaucracies is found to be a force Besley, Timothy, and Torsten Persson. 2010. for good: more diverse organizations have sig- “State Capacity, Conflict, and Development.” nificantly higher project completion rates. Econometrica 78 (1): 1–34. The literature linking ethnic fragmentation Blench, Robert. 2012. An Atlas of Nigerian Lan- and macroeconomic outcomes has emphasized guages. 3rd ed. Cambridge, UK: Kay William- that poor policy choices are a channel through son Educational Foundation. which ethnic fragmentation lowers economic Easterly, William, and Ross Levine. 1997. “Afri- growth (La Porta et al. 1999). In this paper ca’s Growth Tragedy: Policies and Ethnic we have taken as given the set of projects the Divisions.” Quarterly Journal of Economics Nigerian bureaucracy is tasked to implement. 112 (4): 1203–50. Our focus has been, given this portfolio of Eifert, Benn, Edward Miguel, and Daniel N. Pos- projects, does the ethnic diversity of bureau- ner. 2010. “Political Competition and Ethnic crats matter for the implementation of public Identification in Africa.” American Journal of projects? Clearly, future work should study the Political Science 54 (2): 494–510. assignment of projects, and also whether the Francois, Patrick, Ilia Rainer, and Francesco implementation of projects is impacted by the  orthcoming. “How is Power Shared Trebbi. F ethnic fractionalization among the communities in Africa?” Econometrica. being served by the project. Hjort, Jonas. 2014. “Ethnic Divisions and Pro- Finally, in contrast to diverse societies, in duction in Firms.” Quarterly Journal of Eco- diverse organizations it might be much harder nomics 129 (4): 1899–1946. for certain groups to become segregated from Hong, Lu, and Scott E. Page. 2001. “Problem others, and it becomes much easier for groups Solving by Heterogeneous Agents.” Journal of to be exposed to each other and thus poten- Economic Theory 97 (1): 123–63. tially offset biases held against members of La Porta, Rafael, Florencio Lopez-de-Silanes, other groups. Both channels can be expected to Andrei Shleifer, and Robert Vishny. 1999. improve performance, all else equal (Alesina “The Quality of Government.” Journal of and Zhuravskaya 2011). This suggests it is Law, ­ Economics, and Organization 15 (1): important to more closely study the processes 222–79. by which bureaucrats are recruited into service, Lazear, Edward P. 1999. “Globalisation and the and the use of specific human resource strategies Market for Team-Mates.” Economic Journal employed by bureaucracies in diverse societies, 109 (454): 15–40. say related to assignments to teams and jobs, as Macchiavello, Rocco, and Ameet Morjaria. 2014. well as job rotation, for the effective delivery of “The Value of Relationships: Evidence from a public services.5 Supply Shock to Kenyan Rose Exports.” http:// www2.warwick.ac.uk/fac/soc/economics/ REFERENCES staff/rmacchiavello/revmaua.pdf. Otite, Onigu. 1990. Ethnic Pluralism and Ethnic- Alesina, Alberto, and Eliana La Ferrara. 2005. ity in Nigeria. Ibadan: Shaneson. “Ethnic Diversity and Economic Perfor- Posner, Daniel N. 2004. “Measuring Ethnic Frac- mance.” Journal of Economic Literature 43 tionalization in Africa.” American Journal of (3): 762–800. Political Science 48 (4): 849–63. Alesina, Alberto, and Ekaterina Zhuravskaya. Rasul, Imran, and Daniel Rogger. 2014. 2011. “Segregation and the Quality of Govern- “­Management of Bureaucrats and Pub- ment in a Cross Section of Countries.” Ameri- lic ­ Service Delivery: Evidence from the can Economic Review 101 (5): 1872–1911. Nigerian Civil Service.” http://www.ucl. ac.uk/~uctpimr/research/CSS.pdf. Shore, Lynn M., Beth G. Chung-Herrera, Michael A. Dean, Karen Holcombe Ehrhart, Don I. Jung, Amy E. Randel, and Gangaram Singh. 5  2009. “Diversity in Organizations: Where On the executive branch of government, Francois, Rainer, and Trebbi (forthcoming) show political cabi- Are We Now and Where Are We Going?” net appointments are often made to be inclusive of many Human Resource Management Review 19 ethnicities.  (2): 117–33.