Kudamatsu Democratization and Infant Mortality in Sub-Saharan Africa

  HAS DEMOCRATIZATION REDUCED INFANT MORTALITY IN SUB-SAHARAN AFRICA? EVIDENCE FROM MICRO DATA Masayuki Kudamatsu

  IIES, Stockholm University Abstract

Does democracy help babies survive in sub-Saharan Africa? By using retrospective fertility surveys

conducted in 28 African countries, I compare the survival of infants born to the same mother before

and after democratization to disentangle the effect of democracy from that of changes in population

characteristics, which is infeasible with country-level statistics on infant mortality. I find that infant

mortality falls by 1.2 percentage points, 12% of the sample mean, after democratization in the

post-Cold War period. Relevant aspects of democracy appear to be the combination of multiparty

elections and leadership change. (JEL: O15, O55, P16)

1. Introduction

  Does democracy promote development? Despite a large number of empirical studies of this question, the evidence remains inconclusive since it is difficult to establish causality running from democracy to development: democracy is likely to be endogenous to socio-economic factors that also affect development (Lipset 1959). As democracy at the national level is clearly not randomly assigned across countries, the empirical challenge is to disentangle the effect of democracy from other confounding factors to the largest possible extent. This paper revisits this question in the context of human development in sub-Saharan Africa. Specifically, I investigate whether the democratization sweeping the region in the 1990s has reduced infant mortality. For

  The editor in charge of this paper was Stefano DellaVigna

Acknowledgments: The previous version of this paper was awarded the 9th Moriguchi Prize by the

Institute of Social and Economic Research at Osaka University. I thank Maitreesh Ghatak, Tim Besley,

Robin Burgess, and Oriana Bandiera for numerous comments and encouragements at various stages of this

research project. I am also grateful to the editor, anonymous referees, Wiji Arulampalam, Iwan Barankay,

Aimee Chin, Stefan Dercon, Esther Duflo, Ray Fisman, Nicola Gennaioli, Hiro Ishise, Hisaki Kono,

Rocco Macchiavello, Ted Miguel, Rohini Pande, Torsten Persson, Steve Pischke, Emilia Simeonova, and

seminar and conference participants at Warwick, IDE-JETRO, Pompeu Fabra, SITE at Stockholm School

of Economics, IIES at Stockholm University, ISER at Osaka University, University of Hong Kong, the 4th

Development Economics PhD Seminar (Namur), CSAE Conference 2007 (Oxford), and Royal Economic

Society Conference 2007 (Warwick) for helpful suggestions and discussions. Sylvia Meek kindly guided

me through the public health literature. Christina L¨onnblad provided me with editorial assistance. Financial

support from the ERC and Handelsbanken’s Research Foundation is gratefully acknowledged.

  E-mail: masayuki.kudamatsu@iies.su.se Journal of the European Economic Association December 2012 10(6):1294–1317 this purpose, I use a cross-country micro panel dataset covering 28 countries in the region to remove confounding factors arising from demographic changes.

  How to confront underdevelopment in sub-Saharan Africa is one of the most important questions in economics today. Very few people, however, have paid attention to the impact on people’s lives of a wave of democratization experienced by the region since the early 1990s. It might be for this reason that the pessimism as concerns the quality of African government is deeply entrenched in any debate on African underdevelopment. Thus, the long-standing question of whether democracy promotes development gains additional importance in sub-Saharan Africa.

  As a measure of development, this paper focuses on infant mortality, defined as death within the first year of life. The survival of infants remains a huge concern in sub-Saharan Africa today, with nearly one in ten babies still dying before his or her first birthday in 2005 (see Web Appendix Figure A1). In addition, focusing on infant mortality has a methodological advantage: unlike other socio-economic outcomes such as personal income, data at the individual level across many countries over a long period of time are available from the retrospective fertility survey component of the Demographic and Health Surveys (DHS). In these surveys, women of childbearing age report when their children were born and whether, and when (if applicable), they died. As surveyed women in Africa on average give birth to four children during their lifetime, I observe a sizable number of mothers having babies both before and after democratization. This characteristic of the data allows me to estimate the effect of democratization by exploiting within-mother variation in the survival of babies, instead of cross-country or within-country variation. As a result, the estimated effect of democratization on infant mortality is robust to the possibility that changes in the composition of the population over time drive both democratization and changes in infant mortality with no direct relationship between the two.

  Any discussion regarding democracy faces a thorny issue of what constitutes democracy. Adapted from the definition of democracy by Przeworski et al. (2000) and motivated by the political economy literature, a democracy in this paper refers to a country satisfying the following two conditions: (1) the chief executive of the government has been elected in multiparty elections with universal suffrage, without subsequently banning opposition parties, and (2) a new chief executive has assumed office by winning multiparty elections. I collected information on these requirements for a country to be democratic. This original data allow me to check how the results differ when the second requirement in the definition is dropped and whether leadership change without democratization reduces infant mortality. Thus, I make some progress in our understanding of what features of democracy contribute to development, which would be infeasible if I relied on widely used democracy scores to measure democracy.

  My findings are as follows. After democratization in sub-Saharan Africa since 1990, infant mortality drops by 1.2 percentage points (12% of the sample mean). This result is robust to controlling for country-specific linear trends in the birth year of babies, country-specific birth-order dummies, country-specific quadratic trends in the mother’s age at birth, and country-level covariates such as per capita GDP, the incidence of wars, and the amount of foreign aid. Except for a couple of outlying cases, there is no such reduction in infant mortality in countries where the dictator holds multiparty elections and stays in power by winning them or where leadership change takes place in a nondemocratic way.

  Due to the lack of appropriate data, I cannot provide conclusive evidence on the mechanism through which democratization has reduced infant mortality. However, suggestive evidence shows that democratization was associated with an increase in the use of some of the health inputs known to be effective in reducing infant mortality, while its association with affluence appears to have been absent. Case studies from two democratized countries suggest that democratically elected new governments indeed changed public health policies to improve maternal and child health care.

  This paper contributes to the large empirical literature that tries to identify the effect of democracy on development or other socio-economic outcomes. I am not

  1 aware of any studies using micro panel data to estimate the effect of democracy.

  Using country-level statistics, Ross (2006) finds no association between democracy and infant mortality. This finding may be subject to confounding factors at the country level, however.

  As an attempt at disaggregating a blunt concept of democracy in the estimation of its effect, this paper is also related to works summarized in Persson and Tabellini (2006). Miller (2008) looks at a particular aspect of democracy—the enfranchisement of women—and finds that this aspect of democracy reduced child mortality in US states in the early twentieth century.

  The rest of the paper is organized as follows. The next section presents two country case studies to motivate the present study. It also explains how I measure democracy in sub-Saharan Africa, discusses causes of democratization and their implications for empirical analysis, and then describes the individual-level data on infant mortality. Section 3 describes the empirical method and reports the main results. Section 4 provides suggestive evidence on possible pathways from democratization to the reduction of infant mortality. Section 5 concludes the paper.

2. Background and Data

2.1. Country Case Studies: Zambia and Nigeria

  Multiparty election for presidency was reintroduced in 1991 for Zambia, where opposition parties had been banned since 1972, and in 1999 for Nigeria, where military officers had successively ruled the country since 1983. In both cases, the former dictator stepped down and a new president took office. The new government in Zambia embarked upon “possibly the most radical and far-reaching reform of the health sector in sub-Saharan African region” (Lake and Musumali 1999, p. 254). In Nigeria, “the new democratic government has been eager to use the windfall (from

1. Aghion, Alesina, and Trebbi (2008) use cross-country industry-level panel data to estimate the effects of democracy that are heterogeneous across industries.

  1. Annual mean infant mortality rates in Zambia and Nigeria. Plotted are the sample mean

infant mortality rates by the year of birth for babies born to those mothers giving birth both before and

after the year of democratization (1991 for Zambia and 1999 for Nigeria), indicated by the vertical

solid line in the figure. Babies born within a year before the survey are dropped from the sample.

Sources: Zambia Demographic and Health Survey in 2001 and Nigeria Demographic and Health

Survey in 2003.

  IGURE F

  oil exports) to deliver so-called ‘democracy dividends’ to the people. In particular, spending on primary health care has increased substantially” (Das Gupta, Gauri, and

  2 Khemani 2004, p. 7). Figure 1 shows the average annual infant mortality rates for

  babies born to those mothers giving birth both before and after democratization in these two countries. Nigeria sees a clear drop in infant mortality. For Zambia, the pattern is less clear-cut, but the long-run increasing trend appears to be reversed shortly after the

  3 year of democratization.

  These two episodes might be exceptions. Figure 1 may be contaminated by confounding factors. In the analysis that follows, I argue that the decline in infant mortality after democratization is not limited to these two countries and that many, if not all, possible confounding factors do not explain this decline.

  

2. See Web Appendix Section A.1 for more details on these new health policies implemented by new

democratic governments in Zambia and Nigeria.

  

3. In support of this interpretation, Garenne and Gakusi (2006) report that the under-five mortality rate

in Zambia rose until 1992 and declined afterwards.

  2.2. Measuring Democracy in Africa

  The definition of democracy in this paper follows the one by Przeworski et al. (2000) with minor modifications to fit it into the context of African politics. Democracy is defined as a political system where (1) the chief executive of the national government has been elected in multiparty elections with universal suffrage, without subsequently banning opposition parties; and (2) a new chief executive has assumed office by winning

  4

  multiparty elections. As none of the existing democracy datasets allow me to measure democracy in that way, I consulted Nohlen, Krennerich, and Thibaut (1999), Europa Publications (various years), and some other sources to collect relevant information for 28 sub-Saharan African countries for which micro data on infant survival are

  5 available.

  This coding procedure reveals the following pattern of the evolution of democracy in the 28 countries (see Table 1). Among 24 countries that had become independent by the 1960s, 12 were democratic at the time of independence. However, all these countries experienced a collapse of democracy either by a military coup or by the banning of opposition parties by the early 1970s. Around 1980, three countries (Ghana, Nigeria, and Uganda) became democratic, but all of them saw military coups toppling democratic governments by the mid-1980s. Since 1990, eleven countries have been democratized within the period for which micro data on infant mortality are

  6 available.

  Web Appendix Figure A2 shows the geographic distribution of these eleven countries that were democratized after 1990. These countries are not concentrated in a particular area of Africa, thus ensuring that the estimated impact of democracy will not pick up an unobservable trend in infant mortality in a specific part of the continent.

  2.3. Difference of Democracy in Africa Before and Since the 1990s

  One might expect that the impact of democracy in Africa would differ before and after the end of the Cold War. During the Cold War period, attempts to subvert democracy in Africa were tolerated by the West so that African countries would not seek the support of the Soviet Union. After the end of the Cold War, however,

  

4. Web Appendix Section A.2 provides theoretical motivations behind the definition of democracy in the

present paper and discusses alternative definitions of democracy such as Polity IV. In their analysis of the

causes of democratization in Africa, Bratton and van de Walle (1997, pp. 10–13) adopt a similar definition

of democracy to the one in the present paper.

5. See Web Appendix Section B for more details.

  

6. Namibia and Zimbabwe became independent in 1990 and 1980, respectively, with the chief executive

elected in multiparty elections with universal suffrage. Opposition parties have been legal in both countries

up until today. However, it is impossible to disentangle the effect of democratization from that of

independence in these two cases. Therefore, I treat the two countries as control groups by dropping

babies born before the year of independence from the sample. Dropping these two countries from the

sample does not change the main result.

  K udamatsu D emocratization and Inf ant Mortality in Sub-Saharan A frica 1299

  ∗ Kenya 1963 1963–69, 2002- 1992–2002 1965–2002 0.074 Lesotho 1966 1966–70, 1993- - 1967,69–2004 0.072 Madagascar 1960 1960–65, 1993- - 1962–1996 0.105

  ∗ Rwanda 1962 1962–65 2003- 1963–1999 0.089 Senegal 1960 1960–63, 2000- 1978-2000 1961–1996 0.092 South Africa 1910 1994- - 1961–1997 0.065 Tanzania 1964 - 1995- 1968–2004 0.096 Togo 1960 1960–61 1993- 1960–1997 0.112

  ∗ Nigeria 1960 1979–83, 1999- - 1965–2002 0.115

  ∗ Namibia 1990 1990- - 1990–1999 0.050 Niger 1960 1993- - 1960–1997 0.140

  ∗ Mauritania 1960 - 1992- 1962–2000 0.078 Mozambique 1975 - 1994- 1975–2002 0.155

  ∗ Mali 1960 1992- - 1964–2000 0.152

  ∗ Malawi 1964 1994- - 1964–1999 0.133

  Gabon 1960 1960–61 1993- 1962–1999 0.067 Ghana 1957 1969–72, 1979–81, 2000- 1960–64, 1992–2000 1967–2002 0.082 Guinea 1958 - 1993- 1961–1998 0.142

  T ABLE

1. Periods of democracy and multiparty politics between 1950 and 2004.

Year of Periods of Periods of Sample Sample mean infant Country independence democracy multiparty politics period mortality before 1990

  Ethiopia - - 1995- 1963–1999 0.131 ∗

  Cote d’Ivoire 1960 2000- 1990–99 1962–1998 0.100 ∗

  Comoros 1975 1975, 1990–99, 2002- - 1975–1995 0.111 ∗

  Cameroon 1960 1960–65 1992- 1968–2003 0.086 Chad 1960 1960–62 1996- 1962,65,67-2003 0.112 ∗

  Burkina Faso 1960 - 1978–80, 1998- 1966–2002 0.105 ∗

  Benin 1960 1960–61, 1991- - 1964–2000 0.118 ∗

  ∗ Uganda 1962 1962–69, 1980-85 1996- 1964–2000 0.091 Zambia 1964 1964–72, 1991- - 1965–2001 0.087 Zimbabwe 1980 1980- - 1980–1998 0.048 Notes: See the text for the definition of democracy. The periods of democracy in bold numbers are what Democracy since 1990 refers to in Tables 3 to 7. Periods of multiparty politics are defined as years in which the chief executive has been elected by multiparty elections with opposition parties being legal, but either the former dictator or the single ruling party has continuously been in power or suffrage was not universal (Ghana between 1960 and 1964). See Web Appendix Section B for the details of the coding procedure. The sample period refers to the years of birth for babies in the sample used in Tables 3 to 5. For the final column, the 14 countries whose sample mean infant mortality before 1990 is higher than the rest are indicated by an asterisk.

  7 Western donor countries became less willing to accept the collapse of democracy. As

  a result, African policymakers may no longer expect to receive foreign aid if democracy

  8 collapses.

  To stay in power, policymakers in a democracy either need to improve the welfare of citizens to win multiparty elections or ignore citizens’ welfare and, whenever they would otherwise be voted out, terminate democracy by buying off a few politically powerful people. It can plausibly be assumed (1) that the cost of choosing the latter option is smaller than that of choosing the former and (2) that the benefit from staying in power, in the African context, depends on the amount of foreign aid from which policymakers can extract rent. These two assumptions, together with the above- mentioned change in how foreign donors respond to the collapse of democracy after the end of the Cold War, imply that there should have been a significant drop in the payoff from ignoring citizens’ welfare and, if necessary, terminating democracy to remain in power after the end of the Cold War. Therefore, in Africa, democratic institutions have been more likely to shape the incentive of policymakers since the 1990s than before. Consequently, I estimate the impact of democracy separately between the eleven cases

  9 since 1990 and the other cases during the Cold War.

2.4. Causes of Post-Cold War Democratization in Africa

  As democratization is hardly exogenous, it is essential to understand the causes of democratization in Africa after the end of the Cold War, in order to understand what may bias the estimation of the impact of democratization. In a widely cited study from the political science literature on the causes of democratization in Africa in the early 1990s, Bratton and van de Walle (1997) find that democratization is cross-sectionally correlated with military intervention in politics in favor of democracy, more official development assistance (ODA), more frequent political protests, and the presence of

  10

  cohesive opposition to the incumbent dictator. While military intervention in politics is unlikely to affect infant survival, ODA is potentially a confounding factor for

  

7. Indeed, military coups against democratic governments in the 1990s were quickly suppressed by

foreign intervention (the Comoros in 1995, Lesotho in 1994 and 1998) or immediately followed by fresh

multiparty elections due to donor pressure (Niger in 1996).

  

8. Consistent with this possibility, Dunning (2004) finds that it is only after 1986 (when the Soviet Union

withdrew its financial support to African countries) that the amount of ODA is positively correlated with

the degree of democracy in Africa.

  

9. An anonymous referee points out that there might have been a dramatic increase in global investments

in child health since 1990, which may have interacted with democracy in Africa. According to Rosenfield

and Min’s (2009) account of the history of international cooperation in maternal and child health, it was in

1982 that the “child survival revolution” took place, after which immunization and oral rehydration therapy

(used to treat dehydration caused by diarrhea) were heavily promoted in developing countries. Due to the

lack of democratization episodes in Africa after 1982 and before 1990, we cannot rule out the possibility

that democratic governments in Africa before 1982 could not improve child survival because of the lack

of international assistance even if they wanted to.

10. This finding is broadly confirmed for the sample of countries in this paper (see Web Appendix Section A.3).

  democracy if more aid-dependent countries tend to be democratized due to donor pressure. In the analysis that follows, I will check if the main result is robust to controlling for the amount of ODA. Frequent political protests and opposition cohesiveness may have resulted from an increased ability for African citizens to solve collective action problems, which may also have helped the survival of babies (by, for example, improving the accountability of local health care staff as in Bj¨orkman and Svensson 2009). Due to the lack of time-variant data on collective action, we cannot exclude the possibility that the estimated impact of democratization in this paper picks up this effect.

  While Bratton and van de Walle (1997) downplay the role of economic factors, Br¨uckner and Ciccone (2011) find that negative rainfall shocks in the previous year increase the probability of democratization in Africa during the period 1980–2004.

  As income mostly depends on rain-fed agriculture in Africa, they interpret this result as the effect of a temporally low opportunity cost of revolution, the key determinant of democratization according to Acemoglu and Robinson (2000). Since unusually low precipitation is found to increase infant mortality in arid climate zones of Africa (Kudamatsu, Persson, and Str¨omberg 2010), I will check the robustness of the main result to controlling for annual rainfall and its one-year lag.

2.5. Micro Data on Infant Mortality

  The micro data on infant mortality (and all other characteristics of babies used in this paper) are obtained from the Demographic and Health Surveys (DHS). The DHS questionnaire is standardized so that researchers can compile cross-country micro datasets. In each DHS survey, a nationally representative sample of women of child- bearing age (15 to 49) are interviewed about the date of birth, and the date of death if applicable, of up to 20 children to whom they gave birth in the past. From this recall data, a panel dataset of mothers can be constructed where the time dimension is the year of child birth given by each mother. Therefore, as long as at least one round of survey was conducted in a country, a panel dataset of mothers is available for that country.

  For the purpose of the present study, I select 28 countries in sub-Saharan Africa where at least one DHS survey has been conducted since 1996 and its results were made available to researchers by 2006. If more than one survey is available in a country since 1996, I select the latest survey to maximize the number of post-democratization years covered in the sample. Web Appendix Tables A.2 and A.3 (columns (1) and (2)) show that these 28 countries are representative for the whole of sub-Saharan Africa on a variety of dimensions except that they depended less on foreign aid in 1990 and were subject to more political protests between 1985 and 1994. The estimation results in this paper may therefore not be applicable to those African countries that were heavily dependent on foreign aid or with fewer political protests around 1990.

  After dropping babies born either before the year of independence of their country or within twelve months before their mother’s interview from the sample (to avoid measurement error in infant mortality), the 28 DHS surveys provide a sample of 643,846 children born to 161,876 mothers with the year of birth of the children spanning from 1960 to 2004. To measure individual-level infant mortality, a dummy is constructed for whether a child dies before the age of one year. Since the literature suggests that determinants of infant death within the first month of life (known as neonatal mortality) differ from those for the rest of the first year of life (e.g. Razzaque et al. 1990), a dummy for whether a child dies before the age of one month is also constructed.

  Web Appendix Section A.4 discusses in detail data issues such as recall bias, mortality selection bias, and fertility selection bias. It shows that for those 21 countries where the previous round of the DHS survey is available, the difference across surveys in the reported annual infant mortality rates for a particular birth cohort of women is not significantly associated with democratization. It also provides evidence that neither changes in fertility of the interviewed women nor changes in the mortality of sisters of the interviewed women after democratization differ across different observable characteristics such as education, affluence, and the area of residence (urban or rural).

  The sample average infant mortality rate is 10% of live births. The sample mean neonatal mortality rate is 4.7%, indicating that nearly half of the infant deaths occur

  11 within the first month after birth.

  With mother fixed effects being controlled for, it is those mothers giving birth both before and after the year of democratization that contribute to the identification of the democracy effect. There are more than 27,000 such mothers in the sample (about 17%

  12

  of the total number of mothers in the sample). Compared to the other mothers of the same age in the same country, these mothers are more likely to be uneducated, poor (in terms of consumer durable ownership), and living in rural areas (see Web Appendix Section A.6). Column (1) of Table 2 shows that for these mothers, the sample mean infant and neonatal mortality rates fall after democratization. Column (2), on the other hand, shows that the sample mean mortality rates are higher for babies born to those mothers giving birth only after democratization than for babies born to those giving birth only before democratization. Column (3), where we exclude mothers giving only one birth from those in column (2) so that the fertility rate is more comparable to column (1), strengthens this observation, suggesting that the composition of mothers in the eleven democratized countries appears to have changed after the democratization. These observations suggest the importance of controlling for the mother fixed effects to estimate the impact of democracy on infant survival.

  11. See Web Appendix Section A.5 for how sample infant mortality rates changed over time.

  12. Among these mothers, 10,377 women experienced the death of at least one of their children.

  ABLE T 2. Mean infant and neonatal mortality rates in democratized countries.

  (1) (2) (3) Both before Only before Only before or only after and after or only after with 2+ births Infant death before democratization 0.120 0.101 0.103 after democratization 0.094 0.109 0.123

  Neonatal death before democratization 0.056 0.047 0.048 after democratization 0.042 0.054 0.061 Number of live birth observations before democratization 98,330 77,717 73,757 after democratization 47,929 25,893 16,855

  

Notes: Live birth observations for the eleven democratized countries are included in the sample of this table.

Each column represents a subsample by the type of mothers: column (1) for those giving birth both before and

after democratization; column (2) for those giving birth either only before or only after democratization; column

(3) for those giving birth more than once but either only before or only after democratization. Infant death is the

indicator for death at the age of less than twelve months; Neonatal death is the indicator for death at the age of

less than one month.

3. Empirical Analysis

3.1. Method

  To investigate whether democratization has reduced infant mortality, I estimate the

  13

  following linear probability model: = + +

  

y α β γ D ∗ 1(t ≥ 1990) + γ D ∗ 1(t < 1990)

imact m at 1 ct 2 ct

  (1)

  $

  • .

  δ TREND x θ + ε

  c ct imact imact

  The dependent variable y is a dummy that equals one if baby i who was born to

  imact

  mother m of birth cohort a in country c in year t dies before reaching the age of one year (or one month if the outcome of concern is neonatal mortality). Mother fixed effects α and mother’s birth cohort by child’s birth year fixed effects β are controlled for so

  m at

  that we estimate the impact of democracy by first taking a difference in infant mortality for the same mother over time and then taking a difference in these differences between mothers of the same birth cohort across countries. By allowing child birth year fixed effects to differ across different birth cohorts of mothers, I nonparametrically control for the effect of a mother’s age at birth that is allowed to change over time due to the continent-wide improvements in the survival of babies (see Web Appendix Figure A.1).

  D is a dummy variable equal to one if country c is a democracy throughout ct year t. Therefore, this dummy is turned on a year after the year of democratization.

  

13. Conditional fixed effects logit estimation requires no serial correlation in the error term for

consistency, which is unlikely to hold in the present context (see Zenger 1993, for example). In addition,

the coefficient estimates in fixed effects logit models are difficult to interpret.

  1(·) is an indicator function that equals one if the argument holds true. The term

  14

  δ TREND represents a linear time trend specific to country c. The vector of

  c ct

  exogenous covariates, x , includes a dummy for girls, a dummy for multiple births

  imact

  15

  (i.e. twins, triplets, or quadruplets), and dummies for each birth order from the second

  16 to the ninth as well as the tenth or higher (with the first birth as the omitted category).

  Standard errors are clustered at the country level to take into account any arbitrary correlations of the error term, ε , over space and time within each country. As the

  imact

  number of countries is only 28 in the sample, I also report the p-value obtained from Cameron, Gelbach, and Miller’s (2008) wild cluster bootstrap-t procedure for the main result.

  The parameters of interest, γ and γ in equation (1), measure the impact of

  1

  

2

  democracy since 1990 and before 1990, respectively. For a consistent estimation of γ

  1

  and γ , the error term ε must be strictly exogenous; that is, it must be uncorrelated

  2 imact

  with the democracy dummies not only contemporaneously but also in the past and the future when the same mother gives birth. It must also be uncorrelated with the democracy dummies for other countries where mothers of the same birth cohort give birth in the same year. After the main estimation results have been presented in what follows, I check the validity of this identifying assumption.

3.2. Main Results

  Table 3 reports estimated coefficients on the two democracy dummies. To see how controlling for mother fixed effects affects the estimation results, column (1) controls for country fixed effects instead of mother fixed effects in addition to cohort-year fixed effects and exogenous covariates. The coefficient on democracy since 1990 is negative and statistically insignificant. Column (2) controls for mother fixed effects instead of country fixed effects. The post-Cold War democracy coefficient becomes larger in absolute terms and is now significantly different from zero at the 5% level.

  Why does the post-Cold War democracy coefficient become larger in absolute terms once mother fixed effects are controlled for? Column (3) replaces mother fixed effects with country fixed effects that are allowed to differ across different birth cohorts of mothers. The post-Cold War democracy coefficient is now estimated to be of similar

14. As there are plenty of observations before the years of post-Cold War democratization, linear trends are unlikely to pick up the post-treatment trends (Wolfers 2006).

  

15. See Klasen (1996) and references therein for available evidence on gender bias in infant mortality

in Africa. Pison (1992) reports that in sub-Saharan Africa, twins are three to four times as likely to die

within the first year of life as singletons. Although there is no a-priori association with democratization,

controlling for these exogenous characteristics of babies reduces the error variance and thus increases the

precision of the democracy coefficient estimates.

  

16. Children of higher birth orders may be more likely to die because they need to compete for household

resources with many other children (e.g. Behrman 1988). Alternatively, children of lower birth orders,

especially the first child, may be more likely to die because mothers are not experienced in child-bearing or

because labor tends to be prolonged during the first birth-giving, predisposing to birth injury and respiratory

distress syndrome of babies. The estimated coefficients on birth order dummies from equation (1) show

that the higher the birth order, the smaller the infant mortality rate in Africa.

  K udamatsu ABLE T

3. Infant mortality drops after democratization since 1990.

  D (1) (2) (3) (4) (5) (6) (7) (8) emocratization Dependent Infant Infant Infant Infant Neonatal Infant Infant Infant variable death death death death death death death death

  ∗∗ ∗ ∗∗ ∗∗∗ ∗∗ ∗∗ ∗ Democracy since 1990 − 0.009 − 0.013 − 0.013 − 0.012 − 0.006 − 0.012 − 0.012 − 0.010 (0.006) (0.005) (0.006) (0.005) (0.002) (0.005) (0.006) (0.006)

  ∗ [0.076] and

  ∗ Democracy before 1990 0.001 0.001 0.000 0.006 0.001 0.007 0.006 0.007 − − − −

  Inf (0.005) (0.005) (0.005) (0.004) (0.002) (0.004) (0.004) (0.004)

  ∗∗ ∗ ∗∗∗ ∗∗∗ ∗∗ F -test

  0.83

  1.86

  2.44

  7.57

  3.91

  7.92

  8.27 6.03 ant [0.371] [0.184] [0.130] [0.010] [0.058] [0.009] [0.008] [0.021] Mortality Exogenous covariates YES YES YES YES YES See notes See notes YES Cohort-year fixed effects YES YES YES YES YES YES YES See notes Country fixed effects YES NO See notes NO NO NO NO NO in

  Mother fixed effects NO YES NO YES YES YES YES YES Sub-Saharan

  Country-specific linear trends NO NO NO YES YES YES YES YES Number of countries

  28

  28

  28

  28

  28

  28

  28

  28 Number of mothers 161,876 161,876 161,876 161,876 161,876 161,876 161,876 161,876 Observations 643,846 643,846 643,846 643,846 643,846 643,846 643,846 643,846 A

  Notes: Standard errors clustered at the country level are reported in parentheses. Infant death is the indicator for death at the age of less than twelve months; Neonatal death is frica the indicator for death at the age of less than one month. For column (4), reported in brackets is the p-value obtained from Cameron, Gelbach, and Miller’s (2008) wild cluster bootstrap-t procedure for the null that the coefficient on Democracy since 1990 is zero. The F-test row reports F-statistics (and its associated p-value in brackets) for the null that coefficients on Democracy since 1990 and Democracy before 1990 are the same. The Exogenous covariates row indicates whether birth order dummies and dummies for girls and multiple births are controlled for. Column (3) controls for country fixed effects that differ across mother cohorts. Column (6) controls for country-specific birth order dummies as well as girl and twin dummies. Column (7) controls for country-specific mother age quadratic trends as well as birth order, girl and twin dummies. Column (8) controls for cohort-year fixed effects that differ between countries with high and low pre-1990 mortality rates.

  ∗ ∗∗ ∗∗∗ Significant at 10%; significant at 5%; significant at 1%.

  1305 magnitude to the mother fixed effects estimate in column (2). This result suggests that the cohorts of women who give birth only after democratization may be unhealthier and thus face a higher chance of their infant’s death. Their health status during adulthood might be negatively affected by the early childhood exposure to the lack of nutrition

  17

  (e.g. Maccini and Yang 2009). The result in column (3) also suggests that the mother fixed-effect estimate of the democracy effect may be applicable to a broader category of mothers, namely, the cohorts of women that have their fertile period around the year of democratization.

  Column (4) estimates equation (1) by controlling for country-specific linear trends in addition to the set of controls in column (2). The post-Cold War democracy coefficient remains almost the same and statistically significant at the 5% level. The p-value from Cameron, Gelbach, and Miller’s (2008) wild cluster bootstrap-t procedure is 0.076. This result shows that the coefficient estimate in column (2) does not reflect a steeper declining trend in infant mortality in democratized countries than in nondemocratized countries. Democratization since 1990 is followed by a reduction in the infant mortality rate by 1.2 percentage points, which is as much as 12% of the sample mean.

  Column (5) reports the result for neonatal death in the same specification as in column (4). The probability that a mother sees her baby die within the first month of life falls by 0.6 percentage points after democratization (statistically significant at the 1% level), about 13% of the sample mean. A sizable portion of the fall in mortality within the first year of life after post-Cold War democratization is therefore due to the fall in mortality within the first month of life.

  The coefficient on the dummy for democracy during the Cold War is, on the other hand, closer to zero and statistically insignificant in columns (4) and (5). The difference between the coefficients on democracy before 1990 and from 1990 and onwards is significant at the 5% level for infant mortality (column 4) and at the 10% level for neonatal mortality (column 5). Although this result is not conclusive due to the limited number of observations exposed to democracy during the Cold War, democracy in

  18 Africa appears to have an impact only after the end of the Cold War.

3.3. Validity and Robustness Checks

  To validate the identifying assumption, I estimate year-wise changes in infant mortality before and after democratization. Specifically, I replace D ∗ 1(t ≥ 1990) in

  ct

  equation (1) with the set of year-wise dummy variables which equal one if n years have passed since the year of post-Cold War democratization, where −4 ≤ n

  19 2, and another dummy variable equal to 1 if three years or more have passed.

  17. See Web Appendix Section A.7 for some suggestive evidence for this possibility.

  

18. The number of mothers exposed to democracy during the Cold War for at least one child birth is

3,096 (1.9% of the total number of mothers in the sample).

  

19. Ghana only has two years of observations after democratization since 1990. In addition, three more

democratized countries (Madagascar, Nigeria, and South Africa) only have three years of observations

  2. Dynamics of infant mortality before and after democratization. The horizontal axis

measures the number of years since democratization. The plots connected by the solid line indicate

changes in infant mortality compared to the period until five years before democratization conditional

on mother fixed effects, cohort-year fixed effects, dummies for girls, multiple births, and birth orders,

the Cold War democracy dummy, and country-specific linear trends. See Web Appendix Table A.9

for the exact numbers of these point estimates. The dotted lines indicate the 95% confidence intervals

where standard errors are clustered at the country level. The comparison between two years since

democratization and three years or more should be interpreted with caution because the composition

of democratized countries is different (only two years after democratization are observed for Ghana).

  IGURE F

  Figure 2 plots the estimated coefficients on these dummies and shows the 95% level

  20

  confidence intervals. These are interpreted as changes in infant mortality, relative to nondemocratized countries, as compared to the period until five years before post-Cold War democratization. The point estimates show that there was no hike or dip in infant mortality before post-Cold War democratization and that the decline in infant mortality started only after democratization. Although the decline after democratization is not statistically significant at the conventional level, these point estimates encourage the

  21 interpretation that democracy has reduced infant mortality.

  Columns (6) to (8) in Table 3 deal with three further concerns on the identifying assumption. Columns (6) and (7) control for country-specific birth order dummies and country-specific mother’s age quadratic trends, respectively, to see if the main result is

  

after post-Cold War democratization. Therefore, I lump together three years or more after democratization

in order to prevent changes in the composition of democratized countries from affecting the result.

  20. The regression table for this result can be found in Web Appendix Table A.9.

  

21. Although a very large confidence interval for years after democratization may suggest a large

heterogeneous effect of democracy, I do not find the democracy coefficient to be significantly different by

mother characteristics such as education, affluence, and living in urban or rural areas (see Web Appendix Table A.10). driven by steeper declining trends of birth order or the mother’s age in infant mortality

  22

  for democratized countries. In column (8), I interact β in equation (1) with an

  at

  indicator variable for the 14 countries whose sample mean infant mortality before 1990 is higher than the rest (see the last column in Table 1) so that changes in the outcome

  23

  could be dependent on a country’s initial level in the absence of democratization. In all cases, there is little change in the size of the estimated democracy coefficient.

  Table 4 reports estimation results when possibly confounding variables for democracy are controlled for in equation (1). The first two columns control for some of the causes of democratization in Africa found in the literature (see Section 2.4). Column (1) controls for total yearly rainfall at the country level and its one-year lag, obtained from the Africa Rainfall Dataset by Miguel, Satyanath, and Sergenti (2004).

  24 Column (2) controls for total ODA as a percentage of GDP (World Bank 2008). In