Ethnic Diversity and Perceptions of Government Performance

Transcription

1 Ethnic Diversity and Perceptions of Government Performance PRELIMINARY WORK - PLEASE DO NOT CITE Ken Jackson August 8, 2012 Abstract Governing a diverse community is a difficult task, often made more difficult by the absence of effective governance institutions. However, designing institutions to manage diverse societies requires an understanding of the mechanisms through which diversity may reduce government effectiveness. This paper analyzes the relationship between diversity and subjective assessments of local government effectiveness in Sub-Saharan Africa using the most recent wave of the Afrobarometer survey. The results indicate that, in most dimensions, diversity is associated with poorer assessments of government performance by both minority and majority groups in the community. Minority groups do report significantly worse outcomes in particular areas, notably corruption and citizen participation. Keywords: ethnic diversity, government effectiveness, Sub-Saharan Africa I would like to thank Wilfrid Laurier University and SSHRC for funding on this project. - kjackson@wlu.ca, Website - 1

2 1 Introduction Over the past 15 years, studies looking between countries, and within both developed and developing countries 1, have shown that diverse populations receive weaker governance, and that this leads to poorer outcomes in terms of public good provision and economic development. Understanding how ethnic diversity affects the governance process is critical to the design of local institutions and therefore the effective implementation of policy around the world. This is particularly important in Sub-Saharan Africa where ethnic diversity is very high and the problems of underdevelopment are particularly persistent. The existing research on governance and ethnic diversity has focused on the provision of measurable public services, looking either at government spending or household access, or at expert opinions of government quality. A complementary approach that I take in this paper is to focus on perceptions of the effectiveness of local government, as reported in surveys of local citizens. In this paper, I consider the relationship between diversity and subjective assessments of government performance in 18 Sub-Saharan African countries, using the fourth round of the Afrobarometer survey from 2008 and Figure 1 highlights a key result, showing the simple correlation between the level of diversity, measured at the district level, and the fraction of the community that approves of the 1 Easterly and Levine 1997 and La Porta et al.1999 focused on cross-country analysis, while Alesina et al.1999 (Miguel and Gugerty 2005, Jackson 2010 looked at the effects of diversity in the United States, Kenya, and sub-saharan Africa more generally. 2

3 performance of their local government official. In homogeneous communities, the approval rate is approximately 60%, while in the most heterogeneous communities, this drops below 50%. Similar baseline results are found in the context of corruption and the extent of trust in local government. A poorer average assessment of local governance in diverse communities could come from two sources. First, a diverse community will have, on average, a smaller fraction of the population in the largest ethnic group. Therefore, if members of the local minority have a lower opinion of their local government, this would show up as a negative effect of diversity. Alternatively, it could be that governments of diverse communities provide poorer service to all, and belonging to a minority or majority group has no bearing on the outcomes for an individual. 2 Previous Literature The initial papers on the economic effects of ethnic diversity demonstrated a strong correlation between diversity and both limited economic development and weak government institutions. Mauro 1995, Easterly and Levine 1997, La Porta et al.1999 were the initial cross-country studies that demonstrated a consistent correlation between ethnic diversity and poor outcomes, along a variety of measures. Alesina et al.2003 and Fearon 2003 extended this analysis using new measures of diversity, but with qualitatively similar results. At the sub-national level, Alesina et al.1999, demonstrated that this relationship also held for local governments in the US. Similarly, 2001 and Vigdor (2004) show that in the US these results extend to both subjective responses to welfare provision and to citizen choices to complete census forms. Miguel and Gugerty 2005 extended this to public good provision in Kenya, with similar results. More recent work, such as Algan et al 2011 focus on the issue of causality, with the key focus being the random assignment of households into more or less diverse housing blocks in France. Their results confirm that the previous work, which had to more or less assume the effective exogeneity of ethnic diversity, is likely finding effects that are real. The mechanisms behind these ethnic diversity effects are generally unclear. Alesina et al suggest that differences in the type of public good demanded may be the cause behind the relationship, while Vigdor 2004 interprets this as a lack of inter-group altruism. Miguel and Gugerty 2005 find evidence of differences in voluntary contributions by individuals, and interpret this as caused by differential sanctioning systems for majority and minority groups. Jackson 2010 focuses on ethnic diversity and public good provision across a wide group of countries within sub-saharan Africa, and finds that a key factor in identifying the critical mechanisms is the distribution of the benefits of public goods to local majority and minority 3

4 groups. This paper extends this discussion results to more subjective assessments. Bratton 2011 analyzes the same data, and while they point to a variety of factors correlated with the apparent performance of government, he does not consider local ethnic diversity. 3 Data The data for this study comes from the 4th wave of the Afrobarometer survey, conducted in 20 countries of Sub-Saharan Africa over 2008 and The two smallest countries, Cape Verde and Lesotho, are dropped from our sample, leaving the 18 countries highlighted in Figure 3. In Cape Verde, the question on ethnic identity is distinctly different from the remainder of the continent, reflecting the very homogeneous population. In Lesotho, the survey does not report the district in which each household lies, which is the scale of ethnic diversity and majority/minority status closest to the local governments that are the focus of this analysis. Countries for Analysis - Afrobarometer 2008 Cape Verde Mali Cape Verde Senegal Liberia Burkina Faso Ghana Benin Nigeria Uganda Kenya Tanzania Zambia Malawi Mozambique Zimbabwe Madagascar Namibia Botswana South Africa The sample of countries included in the survey is not random, as it tends to include 4

5 the more democratic, peaceful and wealthier countries of Sub-Saharan Africa. As a long and largely inconclusive literature has attempted to address the issue of whether the effects of ethnicity increase or decrease with economic development, one should be cautious in generalizing the results to other situations. Casual inspection of the map indicates an obvious categorization based on West African countries and South or East African countries. Alternatively, the countries could be divided based on wealth, colonial origin, or on some other basis. In general, the results are not obviously different across groups of countries, when divided along these lines. 3.1 Measuring Government Performance The survey offers a wide variety of questions that are related to government performance, including specific questions focused on, for example, whether government officials are qualified to do their job, whether local government officals listen to members of the community, or whether governments officials allow sufficient citizen participation. The first question of focus for this analysis is a holistic question on government performance. Do you approve or disapprove of the way that the following people have performed their jobs over the past twelve months, or haven t you heard enough about them to say? (Your Local Government Councilor) The description of the local councilor varies by country. In Ghana, it is Your Elected Assembly man/woman, while in Namibia it is... Responses vary between strong disapproval, disapproval, approval or strong approval. In addition, survey participants are asked to rate the level of corruption in local government: How many of the following people do you think are involved in corruption, or haven t you heard enough about them to say: Local Government Councilors? with the description of the local council again varying across countries. Valid responses are none, some of them, most of them or all of them. This variable is therefore coded opposite to the previous one, with effective government institutions coded as a lower score, and corrupt local institutions receiving a higher score. Finally, participants are asked to rate their level of trust in a variety of groups and individuals, government and otherwise. How much do you trust each of the following, or haven t you heard enough about them to say (Local Government Council): 5

6 Here the responses are coded as not at all, just a little, somewhat and a lot. The match between the question of the local government councilor, which suggests an individual, and either the local government councilors, plural, or local government council, which suggests a collective, is unfortunately not perfect. People could easily be satisfied with the performance of their own councilor, while not trusting the overall local government council to provide effective governance. In particular, this could be true of members of minority groups who are able to elect a member of their own ethnic group to serve as one councilor on a larger council dominated by other groups. 3.2 Measuring Ethnic Differences The Afrobarometer survey asks each participant, What is your tribe? You know, your ethnic or cultural group., and then presents a selection of ethnic groups to choose from or participants can define their own answer. In addition, participants are asked about their home language. Either could be used in a measure of social grouping, though with slightly different interpretations. For the purposes of this paper, the results are not sensitive to this choice, as ethnicity and language are strongly correlated in much of Sub-Saharan Africa. In general, the distinction between ethnic and linguistic diversity is far more important in places such as Asia or South America, where linguistic homogeneity may exist even in the presence of significant ethnic heterogeneity. Social groups are differentiated by country, as the same ethnic group in different countries may have had remarkably different experiences with government. While the measurement of ethnicity within the Afrobarometer survey is generally consistent with national measures produced in other contexts (cite Fearon and Alesina), it does present some missing data. In particular, 3% of the sample is recorded as missing, refusing to answer or identifies only with a national identity. The two obvious options are to either include these as separate identities, or drop the individuals from the sample. In the results presented below, these individuals are dropped from the analysis, but none of the results change qualitatively if they are included in the sample. The second concern with the measurement of ethnic diversity is that the choice of indicators is not innocuous. The most common measure in the literature is fractionalization, a measure that represents the likelihood that two random individuals are of different ethnicities. Alternative measures, such as polarization, or the size of the largest group in the community, may also be logical, depending on the underlying mechanism behind why ethnic diversity matters. Complicating the analysis is the fact that these measures are highly correlated. In most of what follows, I follow the literature and focus on fractionalization, but the choice is not trivial. Further analysis (available upon request), demonstrates that polarization is likely not the relevant measure of diversity, while there is effectively no difference between 6

7 fractionalization and the population share of the largest group. Similarly, choosing a measure of minority status is not simple. The size of an individual s own ethnic group, a binary measure of minority status, or the size of one s own group relative to the average local group are possible choices. The last of these has the advantage of being orthogonal to the level of diversity as measured by fractionalization. Fractionalization controls for the size of the average group, while the size of one s own group relative to the average controls for the position of the ethnic group in the community. Alternative approaches vary in interpretation, but not qualitatively in the results. Individual/Household Variables Additional variables available in the survey include the gender, age, educational status and employment status of the respondent, their evaluation of their level of living and whether they are the household head. For age, educational status and the level of living, the responses are collected into multiple categories, and included as dummy variables in the regression analysis. The data on employment status is limited, and is simply coded as a dummy variable taking the value 1 if the respondent works for a cash income. Aggregated Variables The survey includes some variables coded by the surveyor, whether the primary sampling unit is urban or rural, and the presence of a wide variety of quasi-public goods, such as whether the sampling unit is connected to the electricity grid, whether there is a school in the district, etc. These are identical for everyone within the same sampling unit, though generally not for everyone in the same district. The average or median level of living reported by individuals in the district is also controlled for in all regressions along with average educational attainment. Regional and Ethnic Controls The focus in this analysis is on differences correlated with diversity, not ethnic identity. For this reason, all regressions include dummy variables for each ethnicity and for each region (and therefore country). Empirical Specification The results reported below are all the outcomes of OLS analysis, for ease of interpretation. As discussed above, the survey responses are generally coded on a four-point scale, and therefore an ordered logit analysis would also be appropriate. All results are similar under such a specification. 7

8 Sampling Design and Reweighting Observations The Afrobarometer survey incorporates a stratified sampling design, intended to get a representative cross-section of the population within each country, but also to ensure that all areas of the country are represented in the survey. Sample weights are then included with the data to accommodate the different sampling rate across communities, and additionally across countries. Finally, the survey methodology is to sample random individuals in eight households in each primary sampling unit. With a very limited number of individuals, measuring ethnic diversity, or the share of the local population in a particular ethnic group will generate measurement error. For just over half of the survey population, fewer than 20 individuals are included from their district. A potential solution to this problem is to overweight responses from more heavily sampled districts, on the grounds that the accuracy of local diversity measures will improve with size. The analysis that follows incorporates the within country weights, but not the between country weights, nor the alternative weights based on the size of the sample in the district. None of the results are qualitatively changed in any of the various possible weighting schemes. 4 Primary Results Table 4 contains the primary regression results for the three key variables. Controlling for the ethnicity of respondents and region, diverse communities have, on average, lower approval for government councilors, lower trust in government and higher perceived levels of corruption. Among these variables, relative group size only has a statistically significant correlation with the perception of corruption, with minority groups being more likely to observe the local council as corrupt. All covariates discussed above are included in the analysis, with the results for only a small number reported. Of those not reported, household living conditions are correlated with responses suggesting better governance, as is the average reported living conditions in the district. The results on education may be viewed as discouraging, as people that have acquired more education view government as less effective, more corrupt and less trustworthy. Women and household heads do not respond significantly differently from men, or other members of the household. Households in rural areas and people employed for cash income do report higher levels of trust in local government, results that are consistent with existing research on generalized trust. 8

9 Performance Corruption Trust (1) (2) (3) Ethnic Fractionalization (.051) (.039) (.052) Relative Size (District) (.039) (.040) (.045) Rural (.025) (.019) (.025) Female (.013) (.012) (.017) Household Head (.017) (.017) (.019) Employed (.016) (.017) (.020) Primary (.020) (.019) (.021) Secondary (.025) (.023) (.029) Post-Secondary (.031) (.025) (.033) Obs R Table 1: Performance, Corruption and Trust in Local Government 5 Are Local Officials Qualified? Greater action by the local majority, including the fact that they are more likely to vote, could translate into different opinions of the qualifications of people in local government. The survey includes questions on these qualifications along four dimensions; education, caring about the community, honesty and experience. Table 5 suggests that people in diverse communities have lower opinions of their government officials in general, but with the exception of education there is no distinction between assessments of qualifications by the minority and majority groups. In particular, while the minority group sees more corruption, this doesn t appear to be associated with a lower opinion of the honesty of officials or whether they care about the community. I don t have an immediate interpretation of the education result. 9

10 Education Care Honest Experience (1) (2) (3) (4) Ethnic Fractionalization (.052) (.053) (.059) (.058) Relative Size (District) (.047) (.046) (.051) (.060) Rural (.021) (.024) (.025) (.023) Female (.017) (.015) (.015) (.015) Household Head (.020) (.019) (.020) (.019) Employed (.020) (.020) (.021) (.019) Primary (.023) (.022) (.025) (.022) Secondary (.028) (.026) (.029) (.027) Post-Secondary (.032) (.030) (.034) (.031) Obs R Table 2: Qualifications of Government Officials 6 Transparency If the differences between government in homogeneous areas and diverse areas is not one of outcomes, it may be about process. Table 6 shows the relationship between diversity, minority status and the process of government. The first two columns refer to local government communication, both generally, and specifically on budget issues. The next three columns are related to inclusive, whether they are effective at including local citizens, consulting other groups, and handling complaints. Finally, column (6) asks how well the local government guarantees that revenues are used for public services and not for private gain. The result here indicates no difference in responses for those in the minority or majority locally. This qualifies the previous result regarding corruption, or at least reduces the expectation that the form of corruption that is critical is the expropriation of funds. 10

11 7 Participation in Government The questions in the Afrobarometer survey can be loosely divided between questions about actions taken by the respondent, such as whether they voted or attended a community meeting, and questions asking for an opinion on the actions of government officials. In both Miguel and Gugerty 2005 and Jackson 2010, the models predict greater contributions from local majority groups, even though those contributions benefit everyone. Table?? presents the results of the regression analysis over three measures of respondent participation; have they voted, have they attended a community meeting, and are they a member of a community group. In all three cases, people are less likely to take part in the governance process if they live in a diverse community, and they are less likely to take part if they come from a relatively smaller group. Men, household heads and people in rural communities or with more education, outside employment or wealth (not reported) are generally more likely to participate, with the exception that education does not seem to generate higher levels of voting. Limited participation could manifest itself as a perception of low government responsiveness. Bratton 2011 surveys the correlates of perceptions of responsiveness across the sample, though diversity and ethnicity are not considered. Table?? indicates the relationship between diversity, minority status and four measures of perceptions of citizen voice in governance; do local government officials listen, do they permit citizen participation, do people have a voice between elections, and can ordinary people solve problems. In all cases, perceptions of citizen voice are lower in diverse communities, but there is no relationship with minority status. Men and those with outside employment generally identify greater participation in governance, as do people with a higher reported level of living. The results on citizen action and citizen voice are consistent with an environment where anyone could make their voice heard, but only the majority group chooses to do so. 8 Outcomes The perception that officials in local government are less qualified in diverse communities does not appear to translate into poorer perceptions of the actions of local government. Table 8 shows the regressions results for six dependent variables, on how well the local government is handling maintenance of local roads, local markets, health standards in restaurants, community cleanliness, and collecting license fees and taxes. In only one of the six variables is there a marginally significant relationship between diversity and quality of service, while in three of six there is marginal evidence that minority groups actually perceive better service. One interpretation of this could be that local minority 11

12 groups are more likely to use local roads, markets and pay taxes and therefore have a more informed opinion on these issues. Overall, there is no evidence that the perceived performance of local government on specific outcome issues is related to diversity. 9 Corruption Extending the analysis of corruption further, table 9 extends the analysis to whether assessments of corruption by minority groups are driven by direct experience with corruption. The survey includes questions on whether individuals have had to pay a bribe in order to get documents or permits, get water or sanitation services or to avoid a problem with the police. As reported in table 9, the results show no relationship between minority status and paying bribes for government services, though minorities are more likely to have paid a bribe to avoid a problem with the police. The secondary variables are consistent with expectations, as richer households, more educated households and men are more likely to be the targets of corruption. Column 4 of this table includes direct experience with corruption in a regression with the assessment of corruption as the dependent variable. People that experienced corruption, particularly over documents and the police, are significantly more likely to report that government is corrupt, however even after controlling for this affect, minorities remain more likely to report higher levels of government corruption. This lack of an effect is caused both by no difference in the reported experiences with corruption, and no difference in the extent to which experienced corruption is correlated with poorer assessments of corrupt government (regression not shown). 10 Conclusion People surveyed from diverse districts are less likely to approve of the performance of their local government, less likely to trust their local government, and more likely to see their local government as corrupt. For approval and trust, this result does not depend on whether the respondents are in the local majority or minority group, a result that is consistent with government performance as a community good that benefits everyone equally. Conversely, and perhaps not surprisingly, minority groups are far more likely to report their local government officials as corrupt. Supporting evidence is consistent with the theory proposed by Miguel and Gugerty 2005 on how diversity affects the provision of community goods, with people in minority groups taking a less active position in the governance system, including voting or attending community meetings. However the benefits of effective governance do appear to spread to all groups in 12

13 society. While the lack of participation in the system may not lead to differential outcomes with respect to public provision by government, it may be a factor in the results on corruption. If minority groups find themselves further outside the governance system, they may interpret the actions of government as more corrupt than they would otherwise. References Alesina, A., R. Baqir, and W. Easterly (1999): Public Goods and Ethnic Divisions, Quarterly Journal of Economics, 114(4), Alesina, A., A. Devleeschauwer, W. Easterly, S. Kurlat, and R. Wacziarg (2003): Fractionalization, Journal of Economic Growth, 8, Algan, Y., C. Hémet, and D. Laitin (2011): Diversity and Public Goods: A Natural Experiment with Exogenous Residential Allocation,. Bratton, M. (2011): Citizen Perceptions of Local Government Responsiveness in Sub- Saharan Africa, World Development. Easterly, W., and R. Levine (1997): Africa s Growth Tragedy: Policies and Ethnic Divisions, The Quarterly Journal of Economics, 112(4), Fearon, J. D. (2003): Ethnic and Cultural Diversity by Country, Journal of Economic Growth, 8, Jackson, K. (2010): Public Good Provision, Diversity and Distribution, Mimeo. La Porta, R., F. Lopez-de Silanes, A. Shleifer, and R. Vishny (1999): The quality of government, Journal of Law, Economics, and organization, 15(1), 222. Luttmer, E. (2001): Group Loyalty and the Taste for Redistribution, Journal of Political Economy, 109(3), Mauro, P. (1995): Corruption and Growth, Quarterly Journal of Economics, 110(3), Miguel, E., and M. K. Gugerty (2005): Ethnic Diversity, Social Sanctions and Public Goods in Kenya, Journal of Public Economics, 89, Vigdor, J. L. (2004): Community Composition and Collective Action: Analyzing Initial Mail Response to the 2000 Census, The Review of Economics and Statistics, 86,

14 Communicate Budget Participate Consult Complaint Revenue (1) (2) (3) (4) (5) (6) Ethnic Fractionalization (.053) (.050) (.052) (.052) (.053) (.050) Size Relative to Avg. Group (.042) (.041) (.045) (.044) (.042) (.045) Rural (.025) (.023) (.023) (.026) (.025) (.024) Female (.014) (.015) (.014) (.015) (.015) (.016) Household Head (.018) (.017) (.018) (.018) (.018) (.018) Employed (.020) (.019) (.020) (.021) (.021) (.021) Primary (.021) (.020) (.020) (.023) (.021) (.021) Secondary (.026) (.027) (.027) (.029) (.027) (.028) Post-Secondary (.030) (.030) (.028) (.031) (.030) (.030) Obs R Table 3: Government Transparency 14

15 Vote Meeting Member Listen Participate Between Ordinary (1) (2) (3) (4) (5) (6) (7) Ethnic Fractionalization (.019) (.059) (.044) (.052) (.052) (.053) (.056) Size Relative to Avg. Group (.020) (.057) (.037) (.041) (.045) (.044) (.046) Rural (.009) (.027) (.020) (.023) (.023) (.022) (.024) Female (.007) (.020) (.015) (.013) (.014) (.016) (.016) Household Head (.008) (.022) (.017) (.017) (.018) (.021) (.020) Employed (.008) (.022) (.018) (.018) (.020) (.023) (.022) Primary (.008) (.026) (.018) (.020) (.020) (.024) (.026) Secondary (.012) (.035) (.026) (.027) (.027) (.032) (.032) Post-Secondary (.013) (.038) (.031) (.029) (.028) (.035) (.036) Obs R Table 4: Citizen Participation in Governance 15

16 Roads Market Restaurants Clean Licenses Taxes (1) (2) (3) (4) (5) (6) Ethnic Fractionalization (.057) (.051) (.046) (.049) (.056) (.060) Size Relative to Avg. Group (.042) (.039) (.044) (.042) (.047) (.052) Rural (.026) (.025) (.023) (.024) (.026) (.026) Female (.014) (.013) (.014) (.014) (.016) (.018) Household Head (.017) (.017) (.017) (.017) (.020) (.021) Employed (.017) (.017) (.016) (.018) (.024) (.026) Primary (.019) (.020) (.019) (.020) (.024) (.025) Secondary (.024) (.026) (.025) (.026) (.030) (.032) Post-Secondary (.029) (.029) (.030) (.031) (.035) (.036) Obs R

17 Documents Water Police Corruption (1) (2) (3) (4) Ethnic Fractionalization (.015) (.013) (.015) (.038) Size Relative to Avg. Group (.015) (.013) (.013) (.040) Rural (.007) (.007) (.007) (.019) Female (.006) (.004) (.005) (.013) Household Head (.007) (.005) (.006) (.017) Employed (.007) (.006) (.007) (.017) Primary (.008) (.006) (.007) (.019) Secondary (.010) (.008) (.009) (.023) Post-Secondary (.013) (.010) (.013) (.024) Bribe for Documents.092 (.022) Bribe for Water.035 (.028) Bribe Police.109 (.025) Obs R