The output gap between vertically mixed and homogeneous teams points to vertical discrimination: it appears that upstream workers are willing to accept lower own pay in order to lower the pay of non-coethnic co-workers. About 86 percent of the output gap between horizontally mixed and homogeneous teams is due to vertical misallocation and 14 percent due to horizontal misallocation. Because Kikuyu and Luo workers are of similar productivity on average, horizontal misallocation has little impact on total output. But the distribution of output across downstream workers is affected: in horizontally mixed teams, processors of the supplier’s ethnic group earn 27 percent more than processors of the other ethnic group. In the second main result of the paper, I find that the output gap between homogeneous and diverse teams nearly doubled when conflict between the Kikuyu and Luo political blocs began in early 2008. The reason appears to an increase in workers’ taste for ethnicdiscrimination. I estimate a decrease of approximately 35 percent in the utility-weight of non-coethnic co-workers when conflict began, through a reduced form approach. As also predicted by the model, there was a small but significant increase in the output of processors of the supplier’s ethnic group in horizontally mixed teams in early 2008. A back-of-the envelope calculation suggest that the decrease in productivity in mixed teams may have cost the farm half a million dollars in annual profit, had it not responded. It is clear from these results that the economic costs of ethnic diversity vary with the political environment. In the third main result of the paper, I find that the introduction of team pay for processors six weeks into the conflict period led to an increase in output in horizontally mixed team, cannabis grow racks returning the difference in output between homogeneous and horizontally mixed teams to pre-conflict levels.
The increase was likely due to a reduction in horizontal misallocation: a 32 percent output gap between coethnic and non-coethnic processors in horizontally mixed teams was eliminated when team pay was introduced, as predicted by the model. As a result, overall output increased, even though there was a modest decrease in output in homogeneous and vertically mixed teams. These results indicate that that firms are forced to adopt “second best” policies to limit the distortionary effects of ethnic diversity in the workforce when taste for discrimination is high enough. Figure 2 illustrates the evolution of output in teams of different ethnicity configurations during each of the three sample periods observed. This paper’s findings have important implications for theory and policy. Distortionary, taste-based discrimination in production appears to be the primary explanation behind my results. Theories of non-taste-based ethnic diversity effects are unlikely to simultaneously explain a differential fall in mixed teams’ output during conflict and equalization of downstream workers’ output under team pay. Distinguishing between different channels through which ethnic diversity may affect productivity is important. Higher output in homogeneous teams may be efficient if due to technological differences across diverse and homogeneous teams. But discriminatory preferences should lead to distortionary misallocation of resources in most joint production situations in which individuals influence the output and income of others. Interacting economically with individuals of other ethnic backgrounds is hard to avoid when urbanization and economic modernization brings larger groups of workers together, and large multiplier effects are associated with misallocation of intermediate goods . The contribution of taste-based discrimination in production to the lower incomes observed in diverse countries may thus be sizable.
The findings of this paper also suggest that relatively brief episodes of conflict can have a long-lasting impact on distortionary attitudes towards individuals of other groups. I find no reversion in ethnic discrimination in the nine months after conflict ended. It appears that the economic costs of ethnic diversity vary with the political environment because social preferences are affected by conflict, forcing firms to adjust their policies in conflictual environments. Entirely removing incentives to discriminate through contractual design is difficult, however. At the plant, biased upstream workers continued to derive less benefit from flowers supplied to pairs of processors that included non-coethnics under team pay. As a consequence, it appears, output in vertically mixed teams was 15 percent lower than in homogeneous teams after team pay was introduced.This paper contributes to and ties together several areas of research. Its results are to my knowledge the first to carefully identify and explain a negative effect of ethnic diversity on productivity in the private sector, perhaps because well-measured, micro-level output data from poor countries is rarely available. By showing that a taste for ethnic discrimination can lower output by leading to misallocation of intermediate goods, I also contribute to the literature on workplace favoritism initiated by Becker and the recent literature on social preferences at work . The difference between the findings of Bandiera, Barankay, and Rasul in the U.K. and my findings in Kenya are particularly interesting. The authors find that “upstream” supervisors at a fruit farm in the U.K., in their allocation of own effort and in their assignment of “downstream” workers to rows with different amounts of fruit, discriminate against workers to whom they are not socially connected only when doing so is costless to the supervisor. In contrast, this paper documents an upstream willingness to pay to lower the incomes of non-favored downstream workers, to my knowledge the first paper to do so in data on consequential choices made every day.
Ethnic antagonism may be of greater importance to workers in Kenya than social connections are to workers in the U.K. Burgess, Jedwab, Miguel, Morjaria, and i Miquel and La Ferrara show that Africans belonging to a different ethnic group than “upstream” decisionmakers have less access to economic resources in other contexts, suggesting that distortionary discrimination may be a common phenomenon in Africa. If individuals have discriminatory preferences, output is likely to be lower in diverse production units in most production situations in which co-workers affect each other’s income. I begin to address how the productivity effects of ethnic diversity are likely to vary across time and space by studying how workplace discrimination responds to increased ethnic conflict in society, and how firms respond to distortionary discrimination. I follow an innovative paper by Krueger and Mas in exploring worker behavior during conflict, but my focus is on a poor country characterized by frequent, ethnic tensions. I follow Ksoll, Macchiavello, and Morjaria in studying Kenyan flower farms during the political crisis of 2008, but focus on the effect of conflict on distortionary attitudes towards non-coethnics. As such, this paper also adds to an emerging literature investigating how social preferences are shaped . How firms respond to distortions due to ethnic diversity and how to optimally organize production in the presence of discriminatory attitudes is an exciting venue for future research. Prendergast and Topel provides a theoretical analysis of the influence of favoritism on optimal compensation and extent of authority for managers. In studying the motivation behind the introduction of team pay at the plant, this paper is particularly re-lated to La Ferrara who shows that ethnically diverse cooperatives are more likely to adopt group-pay. I also investigate why the plant chose not to segregate Kikuyu and Luo workers. Finally, there are interesting connections between this paper’s results on within-firm misallocation and the literature in macroeconomics on across-firm misallocation of capital and intermediate goods in poor countries . First, some of the distortionary policies studied by macroeconomists may exist in part as a means for politicians to skew the distribution of resources towards their own ethnic groups and thus ultimately arise from biased preferences upstream. Second, firms whose output suffers from internal misallocation due to ethnic diversity distortions may survive due to macro-level misallocation of capital. Jones points out that to understand development we need to understand both why misallocation occurs and the intermediate goods and linkages through which its effects are amplified. The paper is organized as follows. In section 2, I describe the setting and the organization of production at the plant, cannabis drying racks outline the data used, and test for systematic assignment to teams. The model of upstream discrimination is presented in section 3, and its predictions for the three sample periods observed tested in section 4. Section 5 explores the extent to which other ethnic diversity mechanisms may explain my results. Section 6 investigates the response of distortionary attitudes towards non-coethnics to conflict in more depth, and section 7 the plant’s response to discrimination. Ethnic divisions have influenced Kenyan society and politics since independence and contributed to periodical violence. The country’s biggest tribe, the Kikuyu, was favored by Kenya’s British colonizers, a fact that has had long-lasting influence on tribal relations. The Kikuyu has also been the most economically successful and politically influential tribe during most periods of the post-independence era.
Although the relationships between different tribes have varied over time, the other major tribes have typically defined themselves politically in opposition to the Kikuyu. In recent years the opposition has been led by the second biggest – and persistently politically active – tribe, the Luo. Most Kenyan tribes have aligned themselves with one of the two associated camps. I therefore categorize workers according to the tribal coalition to which their tribe is seen to belong – the “Kikuyu” and the “Luo” .3 An interesting case study in the context of ethnic divisions is Kenya’s vibrant floriculture sector, which brings together large numbers of workers of different backgrounds. A rapid expansion of the sector began in the 1980s; Kenya is now the third-largest exporter of flowers in the world and supplies approximately 31 percent of flowers imported into Europe . Around 50,000 Kenyans are employed in floriculture, and 500,000 in associated industries. Flower farms are part of the fastest growing sub-sector of the Kenyan economy . Production takes place on large farms that typically sell their product through auctions in The Netherlands. Most flower farm employees work either in greenhouses or packing plants . On some farms, including the one I focus on, workers reside on farm property in gated communities. Such farms essentially constitute a miniature society – complete with schools, health clinics and other amenities – in which groups of individuals from different ethnic backgrounds live and work together. Flower farm jobs are considered relatively desirable.The sample farm primarily produces roses. Plant workers are roughly equally divided across three halls. Packing takes place in three-person teams, as depicted in figure 1a. One upstream “supplier” supplies two downstream “processors” working on separate tables. The supplier brings flowers arriving from the greenhouses to her worktable and throws out poor quality flowers. She then sorts flowers of different lengths/types into piles that are placed on the worktable of one of the processors. The processors remove leaves, cut flowers down to the right size, and finally create bunches that are labeled with the worker’s ID number. Nearly all workers are observed in both positions . My primary data source is records of daily processor output from 2007 and 2008. There are 924 packing plant workers in total. The quantities produced were recorded on paper by the farm for remuneration purposes and subsequently converted to electronic format by the research team. A survey provides additional information about workers’ experience, ethnicity, birthplace and other background information. Summary statistics are in table 1. 59 percent of workers are female and 46 percent Kikuyu. The average worker is 35 years old and has five years of tenure at the factory. These figures are similar for Kikuyu and Luo workers. On average, workers are observed working for 22 days followed by two leave days. When a worker takes leave, another worker returning from leave joins the two remaining workers. Teams are observed for 10 consecutive days on average, but because there is substantial variation in the length of individual work spells, the same is true for team spells. The length of work spells is statistically unrelated to characteristics of workers and teams. 28280 different teams are observed during the sample period. Individual workers are observed on 90 different teams on average. Suppliers are paid a piece rate w per rose finalized by the processors supplied throughout the sample period. In 2007, the first year of the sample period, each rose finalized by a processor earned her a piece rate 2w. In February 2008 the factory began paying the two processors based on their combined output, which led to a change in suppliers’ incentives that I exploit in section 4.