Ethnic Favoritism: Micro Evidence from Guinea

  ∗

Ethnic Favoritism: Micro Evidence from Guinea

  †

Masayuki Kudamatsu

  

IIES, Stockholm University

July 28, 2009

Abstract

In an ethnically diverse country, does it matter to the welfare of ordinary citizens

which ethnic group is in power? This paper exploits a plausibly exogenous change

in the ethnicity of the president of Guinea in 1984 to identify the effect of having a

co-ethnic in power on welfare, measured by infant mortality. By using the retrospec-

tive fertility survey of women in a nationally representative sample of households in

1992-93, mother fixed effects estimation results show that, after 1984, infants born to

mothers of the new president’s ethnicity are not substantially less likely to die within

the first year of life than those born to mothers of other ethnicity. Since the govern-

ment had the capability to reduce infant mortality during the period of study (the

yearly average infant mortality rate declined rapidly under the new president’s rule),

the finding suggests that the new president did not favor his own ethnic group to a

large extent as far as ordinary citizens were concerned. (JEL codes: O55, P16, Z13.

Keywords: ethnic conflict; Africa; infant mortality; mother fixed effects estimation.)

  ∗ I thank Martina Bjorkman, Robin Burgess, Esther Duflo, Maitreesh Ghatak, Karl Moene, Rohini

Pande, Andrea Prat, and seminar participants at EOPP (LSE), IIES, Bocconi, NEUDC 2007, and ESOP

  (Oslo) for various comments. _† Postal address: IIES, Stockholm University, SE-106 91 Stockholm, Sweden. Telephone: +46 8 16 3070. Fax: +46 8 16 1443. Email address: masayuki.kudamatsu@iies.su.se.

1 Introduction

  Ethnic diversity is empirically associated with low economic growth (Easterly and Levine 1997; Alesina et al. 2003), poor quality of government (La Porta et al. 1999), and civil

  1

  wars (Montalvo and Reynal-Querol 2005). One possible mechanism for this association

  2

  is ethnic favoritism by the government. A conventional wisdom has it that policy-makers favor their own ethnic groups in the allocation of public funds. As a result, citizens support politicians from their own ethnic group even if these politicians may be less honest or less able than those from other ethnic groups (Banerjee and Pande 2007). As which ethnic group is in power is salient, citizens even resort to violence to have their co-ethnics in

  3

  power. However, there is a lack of systematic evidence that the ethnic group in power is indeed better off. This paper aims to test this conventional wisdom in a systematic and convincing way.

  4 Evidence on ethnic favoritism by the government in the literature is largely anecdotal.

  When statistics is provided, it is often the government expenditure by ethno-region (e.g. Barkan and Chege 1989). Given that government expenditures often do not reach the end- users of public goods in poor countries (Reinikka and Svensson 2004), it is not clear whether ethnic groups in power really benefit from more budget allocations to their regions. This paper looks at infant mortality as a measure of welfare each ethnic group actually enjoys.

  Comparing each ethnic group’s welfare cross-sectionally does not allow us to disentangle the effect of government favoritism from heterogeneity in unobservable characteristics across ethnic groups. To empirically show whether it matters which ethnic group is in power, we need to exploit a change in the ethnicity of political leaders and compare changes in welfare across ethnic groups. However, the ethnicity of leaders may be endogenous to changes in each ethnic group’s welfare. A certain ethnic group may accumulate economic power, which then allows it to seize political power as well. In this case, we would wrongly attribute _{1 2} See Alesina and LaFerrara (2005) for a survey.

  Another possible mechanism is that ethnic diversity exacerbates the problem of collective actions.

While this mechanism receives much attention in the economics literature (e.g. Miguel and Gugerty 2005), ethnic favoritism has so far been rather ignored by economists. ₃ Tishken (1994), in his review of a book on ethnic conflict, lists ethnic favoritism in state resource allocations as one reason for why many conflicts take on an ethnic dimension. ₄ Bates (1983) cites several examples from Africa. improvements in welfare for the ethnic group that gains power to the effect of having a co-ethnic in power. Alternatively, ethnicity in power may change because an ethnic group discriminated by the government seizes power out of grievance. As a result, the ethnic group newly in power becomes better off after the leadership change though this change in welfare has nothing to do with ethnic favoritism per se.

  In order to ensure the exogeneity of ethnicity in power, we need to look at a case where ethnicity in power is determined independently of relative welfare changes across ethnic groups. For this purpose, this paper focuses on Guinea, a country in West Africa with high ethnic diversity. The president who had ruled this country since independence in 1957 unexpectedly died in office in 1984. Only eight days later, a group of military officers who had been excluded from political power until then seized power with the officer most senior in rank becoming a new president. He is from a different ethnic group than his predecessor’s. As discussed in detail in section 3.1 below, the new president’s assumption of power is unlikely to have been made possible by changes in the welfare of his own ethnic group relative to others. Therefore, observed changes in welfare, measured by infant mortality, after the leadership change for the new president’s ethnic group relative to other groups give an unbiased estimate of the effect of having a co-ethnic in power.

  Using the sample of infants created from the retrospective fertility survey of women in a nationally representative sample of households, conducted in Guinea from 1992 to 1993, I estimate the effect of having a co-ethnic as president on the survival of infants born to mothers of the new president’s ethnicity. I control for mother fixed effects in the estimation so that possible compositional changes of women giving birth in response to the leadership change in 1984 will not contaminate estimation results.

  Empirical results obtained in this paper do not provide evidence that the ethnic group in power is substantially better off than other groups in Guinea. As the sample mean infant mortality rates were on the sharp decline after the leadership change, these findings suggest that improvements in infant survival were more or less equally shared by all ethnic groups in Guinea in this period. Although we cannot rule out the possibility that the elite members of the ruling ethnic group benefit from their power or that the effect of favoritism on infant mortality is small but not zero (since infant mortality is a rare event and thus estimating the impact on it requires a large number of observations), this paper’s findings suggest that the ethnicity of the president did not matter substantially to the welfare of the average citizens in Guinea, where violence between different ethnic groups had been the key feature of its history.

  The only systematic evidence on ethnic favoritism in the literature that I am aware of is Kasara (2007), who shows that African leaders tax their co-ethnics more heavily than other ethnic groups, by exploiting time-series variation within each subnational ethno-

  5

  region across 30 African countries. Although her study has more external validity than this paper in terms of countries covered in the study, the endogeneity of changes in the ethnicity of leaders is not explicitly dealt with. In addition, unlike this paper, she does not directly look at welfare as an outcome.

  The paper is organized as follows. The next section provides the background to this study. Section 3 describes identification strategy and data. Section 4 reports empirical results and conducts robustness checks, followed by the concluding section.

2 Background

  This section provides background information on Guinea and its ethnic groups. After a brief discussion on the representativeness of Guinea for sub-Saharan Africa as a whole, I show how Guinean ethnic groups differ from each other and that the conflict among them is a persistent feature of the Guinean history to date. These pieces of information confirm that Guinea is an appropriate country to study on ethnic favoritism. Finally, I explain why we would expect ethnic favoritism to affect infant mortality, the outcome under investigation in this paper, in Guinea after 1984.

2.1 Ethnic Groups in Guinea

  Guinea is a country in West Africa with the population of over 7 million. Its GDP per capita in purchasing power parity terms is close to the average of 48 sub-Saharan African ₅ More precisely, she controls for country-crop fixed effects. As “both crop production and ethnic groups

  

are geographically concentrated” (Kasara 2007, p. 160) in Africa, country-crop combinations correspond to ethnic groups in each country. countries in 2000 though economic growth between 1960 and 2000 is among the worst in the

  6 7 region. Infant mortality per 1,000 live births has always been above the African average.

  8 Ethnic diversity is also higher than the African average. These statistics suggest that what

  9 we see in Guinea is likely to represent the basket case even by African standards.

  Guinea has six major ethnic groups: Sousou, Peulh, Malinke, Kissi, Toma, and Guerze,

  10

  with the last three groups often grouped together as “Foresters”. According to O’Toole and Baker (2005, p. 163), the estimated ethnic composition in 2000 was 40 percent Peulh, 30 percent Malinke, 20 percent Sousou, and 10 percent other groups.

  Members of each group speak different languages with the Sousou, Malinke, Toma, and Guerze languages more similar to each other (belonging to the Mande language group) than to the rest. Although French is the official language of Guinea, only about 20 percent of the population understand French (O’Toole and Baker 2005, p. 93). One can identify

  11 each other’s ethnicity from their surname to some extent.

  Sousou, Peulh, Malinke, and “Foresters” each predominate in one of the four topograph- ical regions: Lower Guinea (or Guinee Maritime), Middle Guinea (or Futa Jalon), Upper Guinea (or Haute Guinee), and Forest Guinea (or Guinee Forestiere), respectively. Each region has a slightly different climate pattern and thus people cultivate different crops, implying that a simple cross-sectional comparison of ethnic groups in terms of welfare can be misleading. ₆ Guinea’s real GDP per capita is 2,546 US dollars in purchasing power parity terms in 2000 while

  

the African average is 2,633 dollars. The Guinean economic growth rate between 1960 and 2000 is -0.47

percent, which is the fourth lowest among 33 African countries with data available. All the figures are

based on Penn World Table 6.2 (Heston et al. 2006). ₇ Infant mortality per 1,000 live births for Guinea and for the African average is 215 versus 160 in 1960, 162 versus 105 in 1985, and 112 versus 95 in 2000 (World Development Indicators, September 2006). ₈ The ethnic fractionalization index (Alesina et al. 2003) is 0.74 for Guinea and 0.66 on average for 47

sub-Saharan African countries. Guinea’s ethnic polarization index (Montalvo and Reynal-Querol 2005) is

the highest in sub-Saharan Africa (0.84). ₉ Indeed, the 2007 Failed State Index, compiled by the Fund for Peace and Foreign Policy magazine,

ranks Guinea as the 9th most fragile state in the world (see “The Failed State Index 2007,” Foreign Policy,

  July/August 2007, pp. 54-63). ₁₀ There are various spellings for the names of ethnic groups in Guinea (Susu or Sosso for Sousou; Fulbe,

Fula, Fulani, or Peul for Peulh; Maninka, Mandinka, or Manding for Malinke; Loma for Toma; Kpelle for

Guerze). I follow the spelling in the codebook of the Demographic and Health Survey conducted in Guinea

in 1992-93 (the dataset used in this paper). ₁₁ Guineans with family names such as Bah, Balde, Barry, Diallo, Sow, Tall, and Thiam are generally

Peulh while family names Camara, Conde, Diawara, Fofana, Kante, Kourouma, Kouyate, Soumaoro, and

  Traore indicate the Malinke people (O’Toole and Baker 2005, pp. 96 and 139).

  12 The six ethnic groups in Guinea are socially and culturally distinctive. Peulh is

  particularly different from the rest. Its subsistence economy depends on animal husbandry a lot more than the other groups in Guinea. Core membership of Peulh kin groups is confined to a single community while lineages for the other groups comprise residents of more than one community. Settlement patterns differ, too. For Peulh, a community comprises a nucleated village or town with outlying homesteads or satellite hamlets. For other groups, a community is just a nucleated village or town only. Finally, Peulh traditional society is more politically complex: there is a jurisdictional level above local communities. For other ethnic groups, each community is traditionally independent.

  Non-Peulh ethnic groups also exhibit differences. In terms of family organization, Sousou, Malinke, and Kissi have large extended families while Toma and Guerze fami- lies are independent. Polygyny is general for all groups, but co-wives live together for Sousou and Guerze, but occupy separate quarters for the rest. For rules for inheritance of real property, all groups are patrilineal. But a man’s property is inherited by his sons in Kissi and Guerze societies while the other groups designate other patrilineal heirs than sons. In Kissi society, the inheritance is equally distributed among all sons. In Guerze, on the other hand, the senior son inherits real property. For inheritance of movable property, all but Sousou follow the same rule as for real property. Sousou society now allows a man’s sons to equally inherit.

  Do these ethnic groups in Guinea differ in socio-economic characteristics? Table 1 shows the shares of women aged 15-49 (in the sample of the Demographic and Health Surveys conducted from 1992 to 1993) who have attended primary school (Panel A), have access to electricity (Panel B), or own a television set (Panel C), by ethnicity (Kissi, Toma, and Guerze are grouped together as Foresters) and by natural region plus Conakry, the capital city. Column (1) looks at the whole country. For education, Foresters (and Sousou to a lesser extent) are significantly more educated while Peulh are significantly less educated. In terms of access to electricity and television ownership, Sousou are significantly better off while Foresters (and Peulh to a lesser extent) are significantly worse off. This partly ₁₂ The following two paragraphs are derived from entries in columns 7, 14, 20, 30, 32, 74, and 76 for

  

Af2 (Kissi), Af11 (Toma), Af15 (Kpelle), Ag6 (Futajakonke), Ag9 (Malinke), and Ag26 (Susu) in tables of Murdock (1967). reflects the differences across regions, because each ethnic group predominates in one region (see Panel D). Even within each region and the capital city of Conakry, however, there are statistically significant differences across ethnic groups in most of the cases (although ethnic groups that are minority in each region tend to be better off, perhaps due to the tendency of migrating Guineans to be richer).

  To sum up, ethnic groups in Guinea are distinctive to each other in terms of culture and socio-economic status, suggesting that ethnic divisions could result in conflicts in this country.

2.2 History of Ethnic Rivalries

  Ethnic groups in Guinea have a long history of rivalry. During pre-colonial days, Peulh rulers oppressed Sousou people while Foresters fought against the invasion by Malinke

  13

  people. Under the French colonial rule starting at the end of the 19th century, the four ethnic groups could not agree on the location of a new secondary school financed by

  14

  the colonial authority in 1947. Most political parties formed after 1945 were ethnically-

  15

  based. Prior to independence, there were riots against Peulh people, especially by Sousou

  16 people.

  Guinea became independent from France in 1958 after Guineans voted for independence in a referendum. Ahmed Sekou Toure, a Malinke, became president and established one- party rule immediately after independence. He was accused of ignoring Middle Guinea

  17

  (and thus Peulh people living there) in the first development plan for 1960 to 1962. His

  18 ethnic group, Malinke, was allegedly overrepresented in the army and in top leadership.

  19 In 1976, Peulh leaders were alledged to attempt a coup against President Toure. _{13 14} See O’Toole and Baker (2005), pp. xxxvii, xxxvix, 81-2. ₁₅ See Adamolekun (1976), p. 125. ₁₆ See O’Toole and Baker (2005), p. 160. ₁₇ See ibid., pp. xl-xli, and Adamolekun (1976), p. 125. ₁₈ See Adamolekun (1976), p. 132.

  See ibid., pp. 131-2, 172-3. See also Everett (1985), p. 23. Yansane (1990, footnote 48) provides a

different view by noting that “Toure certainly did not favor any ethnic group except for his family.” As

his family members are, by definition, all Malinke, this “family” favoritism may have been seen as ethnic

favoritism. ₁₉ See O’Toole and Baker (2005), p. 96.

  20 Sekou Toure died in office unexpectedly on March 26, 1984, at the age of 62. Upon

  Sekou Toure’s sudden death, Lansana Beavogui, a Toma who had been prime minister under Toure’s rule since 1972, became interim president. Only eight days later, Colonel Lansana Conte, a Sousou, staged a coup against the interim government and became president. (It is this leadership change that I exploit in the empirical analysis below.) His presidential

  21

  guards were said to be mostly Sousou. When Diara Traore, a Malinke military officer who was number two in Conte’s military government, attempted a coup in 1985, there was

  22 looting against Malinke people in the capital city of Conakry.

  In 1990, when the first local elections were held, violent clashes between ethnic factions

  23

  errupted in some areas. When opposition parties were legalized in 1992, most newly

  24

  formed political parties were ethnically based. In 1993, the first multiparty presidential election since independence was held in which Lansana Conte won with 52 percent of valid votes (Brune 1999, p. 457). Conte was re-elected in 1998 and 2003, and died in office in December of 2008. During the last years of Conte’s rule, there was fear of civil wars between ethnic groups after his death. Peulh people are reported to demand presidency

  25 after almost five decades of rules by other ethnic groups.

  The above description of Guinean history shows that ethnicity has been salient in this country for a long time. Whether or not the ethnicity of the president matters for each ethnic group’s welfare, however, is an empirical question which this paper aims to answer.

2.3 Possible Means of Favoritism in Guinea after 1984

  To estimate the impact of the leadership ethnicity change, I look at infant mortality as an outcome. An obvious question is why we expect infant mortality to change as a result of ₂₀ Adamolekun (1976) and Riviere (1977) argue that Sekou Toure was successful in integrating ethnic

  

groups in Guinea. Gardinier (1988), however, points out that it is not clear how Toure managed to integrate

Guineans while his education policy led to a situation where “primary and secondary school classes were

taught only in local dialects” (Everett 1985, p. 23). _{21 22} See Schissel (1986), p. 23.

  See O’Toole and Baker (2005), p. 203. Presumably, non-Malinke citizens in Conakry saw this coup as

Malinke’s attempt to seize power back. As will be discussed in Section 3.1 below, though, a closer look

reveals that this attempted coup does not appear to be ethnically motivated. _{23 24} See ibid., p. xlvi. ₂₅ See ibid., p. 161.

  See Sillah (2007). ethnic favoritism. Possible mechanisms in the context of Guinea in the mid-1980s are the selective dismissal of civil servants and the selective revamp of primary health care systems in a way to favor Sousou people.

  Under Sekou Toure’s rule, Guineans were impoverished while the bureaucracy was hugely bloated. The number of public sector employees in November 1985 was 140,000,

  26

  above 2 percent of the entire population. Civil servants engaged in ‘moonlighting’ by sell- ing goods from state warehouses to the black market (Graybeal and Picard 1991, p. 288). By the time Sekou Toure died, Guinean national debt was accumulated to the level of 62 percent of its GNP (Ibid., p. 282). Lansana Conte needed to accept IMF conditionalities, including the streamlining of the public sector, to obtain loans. It could be conceivable that Conte may have chosen whom to be fired based on ethnicity. To the extent that Sousou civil servants have an informal risk-sharing network with their co-ethnics, then we would see the survival of babies more likely for Sousou than for the other ethnic groups, because only Sousou people could thus afford sufficient nutritional intakes and access to health care.

  By the end of Sekou Toure’s rule, health systems in Guinea collapsed severely. Kaba (1977, p. 40) reports that despite a rapid population growth of Conakry, no new hospital was constructed in the city. Inexperienced individuals were appointed to hospital admin- istration. Medicine shortage was endemic. Toure even denied a cholera epidemic in 1973. Knippenberg et al. (1997, pp. S30-S31) describes the state of the Guinean health system right after the death of Toure in 1984. Storage and distribution of vaccine and drugs was inadequate due to scarcity of spare parts for refrigerators and of fuel for vehicles. The lack of financial resources due to Toure’s economic mismanagement exacerbated the unavail- ability of drugs. Access to health services was limited due to a long distance to clinics. Health staff could not travel to villages for outreach activities because of lack of transport or fuel. The quality of health service, if provided, was perceived as poor by Guineans.

  After seizing power, Lansana Conte initiated the revitalization of health systems in 1986 by formulating a new health policy (World Bank 2005, p. 1) and by developing primary health care centers throughout the country with an emphasis on child and maternal care ₂₆ See Graybeal and Picard (1991, p. 287). As a benchmark, consider the number of civil servants in

  

Cote d’Ivoire, a neighboring country of Guinea whose population is nearly twice as large. It was 80,000 at

that time (Africa Contemporary Record, 1983-84, p. B446).

  (Glik et al. 1989, p. 423). As we will see below, infant mortality started dropping sharply after 1986. Conte could have deliberately allocated public funds and human resources for revamping health care provision to communities where his co-ethnics predominate. Given the extremely poor conditions of the health care system after the death of Sekou Toure as described above, such selective attempts to improve the system in favor of Sousou people could have resulted in an immediate change in infant mortality.

3 Identification Strategy and Data

3.1 Exogenous Change in President’s Ethnicity?

  As indicated above, the ethnicity of the Guinean president changed from Malinke to Sousou in 1984. I exploit this change to estimate the effect of having a co-ethnic as president on individual welfare measured by infant mortality. An obvious issue on this identification strategy is whether the seizure of power by Conte, a Sousou military officer, is exogenous to changes in determinants of the welfare of Sousou people over time.

  The welfare of Sousou people is unlikely to be correlated with the fact that Conte seized power and stayed in office for four reasons. First, Conte had not been politically powerful before the coup. Momoh (1984, p. 756) describes him as belonging to “the less privileged sector of Guinean armed forces”. As a result, it is unlikely that his position in the government before 1984 both brought about improvements in Sousou people’s welfare and allowed him to seize power.

  Second, the military coup does not seem to have been ethnically motivated, suggesting that Sousou’s economic power was unlikely to be crucial for Conte to seize power. Several non-Sousou military officers participated in the military coup. As mentioned above, the

  27

  number two figure in Conte’s government, Diara Traore, is a Malinke. Among the other 16 original members of the military junta (Comite Militaire de Redressment National), one major, two captains, and one lieutenant are Peulh, and four majors are Malinke, judging

  28

  from their surnames. Another member of the military junta, Captain Jean Traore, is _{27 28} According to Kaba (1985, p. 178), Traore was the “main force” behind the coup.

  See Momoh (1984) for the list of members of the military junta. I rely on O’Toole and Baker (2005, pp. 96, 117, and 139) for which surname is typical for which ethnic group. from Forest Guinea, where ‘Foresters’ (Kissi, Toma, Guerze) reside, and he is thought to

  29

  be one of the closest to Conte. On the other hand, the only Sousou politician among top political leaders under Toure’s rule, N’Famara Keita (see Adamolekun 1976, pp. 173-4),

  30 was arrested after the coup and died in prison a year after.

  Third, the 1985 attempted coup by Traore does not appear to have been a clash between Sousou and Malinke ethnic groups, suggesting that Sousou’s economic power was unlikely to be decisive for Conte to stay in power. Ousmane Sow, whose surname indicates that he is a Peulh (O’Toole and Baker 2005, p. 96), led a battalion to first counter-attack Diara Traore’s soldiers during the 1985 coup attempt. In addition, not all Malinke officers

  31 supported Traore.

  Finally, Conte became president because he was the most senior in rank among the coup plotters (Hodonou 2004), and the reason for his senior position does not appear to have been his ethnic background but his military talent. He was a sergeant at the time of independence. In 1970, when Portugal invaded Conakry, the capital city, to attack the headquarters of the independence movement for Guinea-Bissau and Cape Verde (which were Portuguese colonies at that time), Conte was in charge of the defense of Conakry and successfully repelled the Portuguese invasion. Afterwards, he was named captain for

  32

  the exceptional service to his country. This military background appears to have allowed Conte to lead the military junta.

  I cannot entirely exclude the possibility that Conte seized power because the Sousou- dominated region was becoming relatively better-off than others, however. Lower Guinea, where the Sousou predominates and his home village is located, has major bauxite mines in operation during the 1970s and the 1980s (see Campbell 1991, pp. 34-39). It is not clear to what extent local people benefited from bauxite mines as local processing of bauxite was limited (Campbell 1991). Still, given that Guinea possesses about one-third of the world’s highest-grade bauxite deposits and has been the world’s leading exporter, bauxite mining could have been a huge source of economic power. In addition, Conte was a commander of the Boke military region in Lower Guinea (Hodonou 2004), where there was one bauxite _{29 30} Africa Contemporary Record , 1984-1985, p. B470; O’Toole and Baker (2005), p. 203. ₃₁ See Keesing’s Record of World Events, p. 33710 (July 1985) and O’Toole and Baker (2005, p. 124). ₃₂ See “Guinea: Diarra Traore’s Attempted Comeback,” West Africa, 15 July 1985, pp. 1412-3.

  See O’Toole and Baker (2005, pp. 55 and 164-5). mine in operation since 1973. He might have accumulated personal wealth from bauxite export, which could have allowed him to buy support for his presidency. In fact, the military’s support for Conte, which is likely to be crucial for political survival in non- democratic politics, appears to have been based on improvements in living conditions among

  33

  officers and soldiers in the army. This might not have been possible if Conte’s regional base was a poor area.

  To partly deal with this concern, I will in some specifications control for any arbitrary region-specific trends by allowing year fixed effects to differ across regions. By doing so, I compare welfare changes across those Guineans living in the same region.

3.2 Data

  The data source used in this paper is the Demographic and Health Survey (DHS) con- ducted in Guinea from February 1992 to March 1993. In the survey, 450 communities (villages and town districts) were first sampled with selection probability proportional to the population. In each community, 16 households were randomly sampled. The head of a sampled household reported every household member’s ethnicity. In section 4.2 below, I use this data to identify communities where the Sousou ethnic group accounts for more than a half of the population.

  34 The survey then samples one woman aged between 15 and 49 from each household and

  interviews her on, among others, their ethnicity, educational attainment, migration date, their children’s birth date and, if applicable, age at death in months. From these interviews of 6,065 women in total, I construct an unbalanced panel dataset of mothers with the time dimension being the birth year of their children. In the original sample, there are 20,897 babies born to 4,980 mothers at least one year before the survey (we do not observe whether babies born within a year before the survey will actually survive beyond their first birth day). From this sample, I drop babies whose birth year is not reported or whose mother’s ₃₃ According to Momoh (1984, p. 757), under Sekou Toure’s rule, “[w]ages for the armed forces had been

  

poor while housing was short and mostly in deplorable conditions.” On the other hand, Conte ensured that

the army would be shielded from the public sector payroll cut under the structural adjustment (Africa

Contemporary Record ₃₄ , 1984-85, p. B473).

  This way of sampling women results in the under-sampling of women from large households. Whether this sampling method biases the estimate on ethnic favoritism will be discussed in Section 4.3 below. ethnicity is not reported. The resulting sample contains 18,932 babies born to 4,706 women

  35 from 1954 to 1992.

  From information on the age at death, I create a dummy variable for infant death (death within the first year of life) as the dependent variable in the following analysis. For exogenous controls, I also create dummy variables for birth orders, baby girls and babies born in multiple birth (twins, triplets, and quadruplets). In the sample, babies of lower birth order are more likely to die, so are baby boys and babies born in multiple birth.

  These variables are included as regressors to increase the precision of coefficient estimates.

  There is one data issue that may bias the estimated effect of favoritism towards zero: mortality selection of mothers. Babies born to women who are dead at the survey date are missing from the sample. Section 4.3 below discusses whether this sample selection is serious enough to bias estimation results.

3.3 Summary Statistics Overall, 14.5 percent of live births lead to death within the first year of life in the sample.

  Figure 1 plots yearly average infant mortality rates over time. The top graph shows that the mortality rates had been stable in the range between 15 and 20 percent until 1986, after which there was a remarkable decline to below 10 percent. This trend break is consistent with the health care system reform initiated by President Conte in 1986 as described above, implying that Conte’s health care reform worked. One would argue that we will not be able to detect evidence for ethnic favoritism even if President Conte did have an intention to favor his own ethnic group, if the government did not have the capability to affect the welfare of any ethnic group. The fact that the health care reform did seem to have an impact on the reduction in infant mortality, however, makes this argument less likely to be plausible. ₃₅ There is no statistically significant difference in infant mortality between those born to mothers with

  

and without ethnicity known. Babies without their birth year reported are more likely to die in the first

year of life. From their birth order and the birth year of their siblings, however, we know, for most of these

babies, whether they were born by or after 1984. Among babies born after 1984, there is no statistically

significant difference between Sousou and non-Sousou in the proportion of those with their birth year

missing. Among babies born by 1984, the proportion of those without their birth year reported is higher

for Sousou than for non-Sousou. Dropping these babies, therefore, does not bias the estimated effect of

favoritism towards zero.

  The bottom graph in Figure 1 compares the trend in infant mortality between Sousou (the solid line) and non-Sousou (the dashed line). Sousou babies are in general more likely to survive throughout the sample period. While the trend break for non-Sousou babies appears to be 1986 as well, that for Sousou seems to be 1987, suggesting that Sousou people benefited from the health care system reform later than the other ethnic groups.

4 Estimation and Results

4.1 Individual-level Favoritism?

  To investigate the possibility of individual-level favoritism, I estimate the following equa- tion: _′ θ y ∗ _{imt = α m + β t + γS m 1(t > 1984) + x + ǫ imt , (1) imt}

  The dependent variable, y imt , is a dummy indicating whether baby i born to mother m in year t dies within the first year of life. We control for the mother fixed effect, α m , so that the compositional change of child-bearing women for each ethnic group does not affect our estimate of the favoritism effect. If the Conte administration treats Sousou people better than other groups, less healthy women (whose infants are thus more likely to die) from the Sousou group may give birth more often than those from other groups, biasing the estimated favoritism effect towards zero. Controlling for mother fixed effects avoids this bias. We also control for the birth-year fixed effect, β t , so that the country-wide trend in infant mortality seen in Figure 1 is non-parametrically controlled for. Since the country- wide infant mortality was on the decline after 1986, we therefore identify favoritism from the difference in the speed of a decline in infant mortality between Sousou and other groups.

  The coefficient γ on the interaction term of a dummy for Sousou mothers, S , and _m the indicator for years after 1984 is our measure of the degree of ethnic favoritism. It is interpreted as changes in infant mortality for babies of a Sousou mother, relative to babies of a mother of other ethnicity, after Conte seized power.

  A vector of exogenous controls, x imt , includes dummies for whether baby i is a girl, whether baby i is born in multiple birth, and the birth order of baby i. Controlling for these variables is expected to increase the precision of estimates.

  Standard errors are clustered at the ethnicity-by-period level where by “period” we

  36 mean years until 1984 and years after.

  Panel A of Table 2 reports the estimated γ in equation (1). Column (1) uses the full sample and shows that babies born to Sousou mothers are more likely to die after 1984 by 1.6 percentage points. Since the estimate is not significantly different from zero, we cannot reject the possibility that there is no difference between Sousou and the other ethnic groups. However, the positive coefficient implies that Sousou people’s welfare in terms of child survival did not improve more than that of the other ethnic groups due to the change in ethnicity of the president.

  This finding may, however, reflect the migration of non-Sousou people to Sousou com- munities with health care services improved by favoritism after 1984, which biases the estimated effect of favoritism towards zero. To investigate this possibility, I restrict the sample to those babies conceived after their mother migrated to the surveyed community. Column (2) estimates equation (1) with this restricted sample, showing that the result is more or less similar to the one with the full sample. Therefore, migration does not seem to

  37 bias the estimated γ in column (1).

  Another concern on the results in column (1) is that those areas where most Sousou women live, namely Conakry and Lower Guinea, may have had worsening disease environ- ments while other areas did not, biasing the estimated γ in equation (1) upwards. Also, as discussed in section 3.1, Conte’s assumption of power may have reflected the export of bauxite mined in his region of birth, Lower Guinea. Column (3), using the same restricted sample as in column (2), replaces year fixed effects in equation (1), β , with region-by-year _t fixed effects so that we compare Sousou and non-Sousou mothers giving birth in the same region. Now the estimated γ in equation (1) is negative although it is not significantly ₃₆ Forest people (Kissi, Guerze, Toma) are grouped together as one ethnic group, and minority groups

  

and foreigners are also grouped together as the fifth ethnic group. To avoid the bias due to a small number of clusters, I cluster standard errors at the ethnicity-by-period level instead of at the ethnicity level. ₃₇ After Lansana Conte seized power, some of more than one million Guineans (which is up to 20 percent

of the population) who had fled the country due to the repressive nature of Sekou Toure’s rule returned to

Guinea, though the number of such returnees was thought to be not very large (O’Toole and Baker 2005,

p. 171). The result in column (2) suggests that this return of those in exile does not seem to bias the estimated γ in column (1), either. different from zero. Columns (4) and (5) further restrict the sample to urban and rural areas, respectively, to see if favoritism may have a different impact on Sousou women living in urban areas from those in rural areas (because, for example, the support from urban residents is more important for a non-democratic ruler). The results show that the negative coefficient in column (3) reflects a decline in infant mortality for Sousou mothers after 1984 in urban areas.

  Although the coefficient estimates in columns (3) and (4) are not significantly different from zero, it could just be due to the insufficient number of observations given that the size of the coefficient is large relative to the sample mean infant mortality, which is 14.5 percent. The negative point estimate, on the other hand, could also reflect an accidental surge in infant mortality among Sousou mothers just before Conte seized power in 1984.

  To check the latter possibility, I estimate the following equation for the urban sample (the one used in column (4) of Table 2):

  1991

  X _′ θ y γ S ∗

  = α + imrt m + β rt j m 1(t = j) + x + ε imrt (2) _{j imrt}

  =1977

  where subscript r refers to the region where baby i was born. γ j measures the change in

  38

  infant mortality in year j compared to the period before 1977. If President Conte favors < his own ethnic group, we should see γ j = 0 for j ≤ 1984 and γ j 0 for j > 1984. Figure 2 plots the estimated γ j ’s in equation (2). It reveals that the negative point estimate in column (4) of Table 2 reflects a surge in infant mortality among Sousou mothers in 1982. It is therefore difficult to interpret the negative coefficient in column (4) of Table

  39 2 as evidence for favoritism.

  The above results, however, may mask the heterogeneity among Guineans. Ethnic favoritism may matter only to the better-off segment of the population because they may be politically more powerful in a non-democratic country. Alternatively, ethnic favoritism ₃₈ Since there are very few babies born in 1992 in the sample, they are treated as being born in 1991 in this estimation. ₃₉ One may argue that coefficients reported in Figure 2 are not comparable to the one in column (4)

of Table 2 because the composition of mothers contributing to the identification of these coefficients is

  

different. If we restrict the sample to those mothers giving birth both before and after 1984, however, we

obtain similar results to those in Figure 2. Other figures reported below are also robust to the same way of restricting the sample. may matter more to the poorer who need government assistance. Panels B and C of Table 2 estimate equation (1) for the subsample of educated and uneducated mothers, respectively, where having attended primary school is defined as educated.

  For babies born to educated mothers (Panel B), the results are similar to Panel A. Over- all, Sousou mothers, if anything, suffer more from the leadership change in 1984 (columns (1) and (2)). Once region-by-year fixed effects are controlled for, those in urban areas appear to benefit though the estimate is imprecise (column (4)). However, estimating equation (2) for this subsample of babies reveals that the infant mortality for educated mothers in urban areas has been more or less flat before and after 1984 with an exception of a hike in 1983 (see Figure 3).

  For babies born to uneducated mothers (Panel C), once we compare Sousou babies to non-Sousou babies born in the same region, the former appear to be less likely to die after 1984 (columns (3) to (5)). Estimating equation (2) for the subsamples in columns (4) and (5) indicates that these negative point estimates do not reflect the decline in infant mortality only after 1984, however (see Figure 4).

4.2 Community-level favoritism?

  The failure of detecting ethnic favoritism in the previous subsection may be driven by the fact that ethnic favoritism in Guinea operated not at the individual level but at the com- munity level. As Bates (1983) argues, ethnicity may matter because of its correlation with space, and Sousou communities, rather than Sousou individuals, may have been targeted in the provision of public health services under Conte’s rule. As it is difficult to imagine that non-Sousou people in the Sousou community are denied access to a community health clinic, they would have benefited from spatial ethnic favoritism, which biases the estimated impact of being born to Sousou mothers towards zero in the previous subsection.

  In this subsection, I estimate the impact of being born in Sousou communities after 1984. For this purpose, we need to identify Sousou communities in Guinea. Using the ethnicity of all the members of the (randomly) sampled households, I obtain the percentage of Sousou residents in the (sample) population for each surveyed community. I define Sousou communities as those with Sousou’s population share exceeding 50 percent. The choice of

  50 percent as the cutoff to be a Sousou community is certainly arbitrary, but using different threshold values turns out to yield qualitatively similar results.

  Table 3 shows the distribution of Sousou communities by region. Overall, Sousou ac- counts for more than half the population in 17.8 percent of the communities in the sample. But they are concentrated in Lower Guinea and Conakry. There is no Sousou community in Middle and Forest Guinea. This table gives an idea about how controlling for region- by-year fixed effects instead of year fixed effects changes the source of identification for the effect of being born in Sousou communities.

  The following equation is estimated: _′ θ y ∗ _{imct = α m + β t + δS c 1(t > 1984) + x + ǫ imct (3) imct} where subscript c refers to a community in which baby i was born, S is a dummy variable _c equal to one if the population share of Sousou people in community c exceeds 50 percent.

  Coefficient δ measures changes in infant mortality for babies born to a mother living in Sousou communities after 1984 relative to those born in the other communities. To avoid the measurement error in S c , the sample is restricted to those babies born in the surveyed

  40

  community (the sample used in columns (2) and (3) of Table 2). Standard errors are clustered at the community level.