Journal of Health Economics 35 (2014) 1–19

Contents lists available at ScienceDirect

Journal of Health Economics journal homepage: www.elsevier.com/locate/econbase

The child health implications of privatizing africa’s urban water supply Katrina Kosec ∗ Stanford University, Graduate School of Business, 655 Knight Way, Stanford, CA 94305, United States

a r t i c l e

i n f o

Article history: Received 10 January 2013 Received in revised form 4 November 2013 Accepted 22 January 2014 Available online 31 January 2014 JEL classification: I18 L33 L95 O12

a b s t r a c t Can private sector participation (PSP) in the piped water sector improve child health? I use child-level data from 39 African countries during 1986–2010 to show that PSP decreases diarrhea among urban-dwelling, under-five children by 2.6 percentage points, or 16% of its mean prevalence. Children from the poorest households benefit most. PSP is also associated with a 7.8 percentage point increase in school attendance of 7–17 year olds. Importantly, PSP increases usage of piped water by 9.7 percentage points, suggesting a possible causal channel explaining health improvements. To attribute causality, I exploit time-variation in the private water market share controlled by African countries’ former colonizers. A placebo analysis reveals that PSP does not affect respiratory illness, nor does it affect a control group of rural children. © 2014 Elsevier B.V. All rights reserved.

Keywords: Water Child health Privatization Infrastructure Education

Can private sector participation (PSP) in the urban piped water sector improve child health? This question is critically important given the global burden of water-related disease. Each year, one in ten child deaths—roughly 800,000 in total—is the direct result of diarrhea (UNICEF and WHO, 2009). An estimated 88% of deaths from diarrhea could be prevented by ensuring access to safe, improved water and sanitation supplies (Black et al., 2003). In 2000, the United Nations issued the Millennium Development Goals (MDGs). Among goals set were: (a) “to halve, by 2015, the 1990 proportion of the population without sustainable access to safe drinking water and basic sanitation,” and (b) “to reduce by two thirds, between 1990 and 2015, the under-five mortality rate.” Most Africa counties will not reach these goals if current trends persist (United Nations, 2011). As of 2010, 39% of Sub-Saharan Africans lacked safe drinking water, and only 19 of 50 countries in the region were on track to meet the MDG drinking water target (UNICEF and

WHO, 2012).1 Today, Africa accounts for about 15% of the world’s population, but for over half of child deaths (Batholomew and Oot, 2005; You et al., 2010). One potential solution is more extensive use of PSP in the water sector, but little is known about how it would affect child health and progress on the MDGs. Allowing the private sector to provide basic infrastructure such as piped water is controversial. Theory suggests that PSP can increase utility efficiency if it is accompanied by direct competition (Shapiro and Willig, 1990; Schmidt, 1996), or by competition for the market, through a competitive bidding process (Demsetz, 1968). However, water provision is a natural monopoly in which firms are not subject to direct competition. Furthermore, there have been fewer bidders in the water sector than in others including energy, telecommunications, and roads (Iimi, 2008). In developing countries, competition may be further encumbered by corruption, collusion, a lack of transparency, and regulatory capture (Beato and Laffont, 2002). This raises the question of whether efficiency gains can translate into greater access to water, or better

∗ Current address: IFPRI, 2033 K Street NW, Washington, DC 20006, United States. Tel.: +1 202 421 3393; fax: +1 202 467 4439. E-mail addresses: [email protected], kosec [email protected]

1 The world as a whole met the MDG drinking water target in 2010—five years early. Nonetheless, more than 780 million people—one in ten—still lacked safe drinking water.

1. Introduction

0167-6296/$ – see front matter © 2014 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.jhealeco.2014.01.006

2

K. Kosec / Journal of Health Economics 35 (2014) 1–19

health outcomes, in the developing world. Also, in developing countries it may be relatively more costly if a private firm fails to internalize the positive externalities of using piped drinking water, since many people may realistically turn to unsafe alternatives. This paper uses child-level data from 39 African countries during 1986–2010 to test whether PSP affects the prevalence of diarrhea in urban-dwelling children under age five. This is the period during which nearly all African countries with PSP in water implemented those arrangements. The data come from 99 Demographic and Health Surveys (DHS) and a novel sub-national region-level database of PSP in water. I show that PSP decreases diarrhea by 2.6 percentage points, which is a 16% decrease in its mean prevalence. Also, among areas with PSP in water, the prevalence of diarrhea is decreasing in the total number of years under PSP. Importantly, PSP reduces diarrhea most among children from the poorest households. Children from households with secondary school-educated mothers and more assets see the smallest reductions. I also show that PSP is associated with a 7.8 percentage point increase in school attendance of 7–17 year olds—an 11% increase over the mean attendance rate. Examining potential causal mechanisms for child health and education outcomes, I find that PSP leads to a 9.7 percentage point increase in reliance on piped water (either in the home or from a public tap) as the primary drinking water source. This implies a 14% increase over the mean rate of access to piped water. Those with piped water have higher-quality water at the source, use larger quantities of water, and are less likely to store water in containers, all of which help reduce the prevalence of diarrhea.2 PSP in water is also associated with greater access to flush toilets, which often require piped water. I also show that PSP is associated with modest but statistically significant reductions in time (round-trip) taken to collect water, equivalent to a 26% decrease in the mean time. The analysis faces a challenge to identification: the possibility of time-varying, unobserved covariates correlated with both PSP in water and child health. I address this issue using an instrumental variables (IVs) strategy. Specifically, I exploit variation over time in the share of privately run water taps worldwide (but excluding Africa) run by an African country’s former colonizer (France, the UK, Belgium, Portugal, Germany, or Italy). The exclusion restriction requires that changes in the water market share of the colonizer only affect changes in child health in the African country by changing the likelihood that the African country has PSP in water. I show that a European country’s water market share is not correlated with the strength of its economy, and that the strength of African economies and their propensity to invest in health are correlated with neither the economic strength of their colonizer nor their colonizer’s water market share. This time-varying measure of relative water market expertise of a country’s former colonizer can thus be interpreted as a supply-side shock to PSP in Africa. I show that when a former colonizer gains an additional 10% of the nonAfrican water market, this increases the probability of PSP in the African country by 4.4 percentage points. Instrumenting for PSP in Africa with the current year water market share of the former colony, I find that PSP in water may reduce diarrhea even more than indicated by the OLS estimates—though the IV estimates are not statistically distinguishable (at conventional levels) from the OLS point estimates. A possible concern is that time variation in PSP in water is somehow correlated with some time-varying factor that improves health. In that case, PSP might spuriously appear to reduce diarrhea, and OLS estimates would be unreliable. I explore this possibility

2 See Blum and Feachem (1983), Esrey et al. (1991), Black et al. (2003), Fewtrell et al. (2005), and Kremer and Zwane (2007).

using two placebo tests. First, I test whether PSP affects symptoms of acute respiratory illness in the same children. I find that PSP does not affect the prevalence of coughing. Health improvements seem to be driven by changes in the prevalence of water-borne illnesses rather than broad improvements in healthcare or sanitary conditions. Second, I test whether PSP in urban parts of a given sub-national region affects child diarrhea in rural areas not directly impacted by PSP policies. I find no significant effect of PSP on this natural control group. The paper is organized as follows: Section 2 offers background information on water and human health. Section 3 describes the empirical strategy. Section 4 presents the main empirical results, quantifying the impacts of PSP in water on child health and school attendance outcomes. Section 5 discusses potential causal channels through which PSP may affect child outcomes. Section 6 presents the two placebo analyses. Finally, Section 7 concludes. 2. Background 2.1. Water and child health Health has a profound impact on individual wellbeing and productivity. Poor health in childhood lowers school enrollment (Alderman et al., 2001), reduces learning productivity (Glewwe et al., 2001), increases absenteeism (Miguel and Kremer, 2004), and lowers test scores (Paxon and Schady, 2007), all of which have lasting impacts. Yet young children have both the least knowledge of how to avoid exposure to disease, and the least resistance to disease (Burstrom et al., 2005; Kremer and Zwane, 2007). This is especially problematic in Africa—a continent that accounts for 15% of the world’s population, but for over half of child deaths (Batholomew and Oot, 2005; You et al., 2010). The health indicator on which I focus is the prevalence of diarrhea in children under age five. Each year, about 800,000 child deaths are the direct result of diarrhea (UNICEF and WHO, 2009). Diarrhea accounts for more under-five deaths than malaria and HIV, combined (WHO, 2007). Diarrhea also contributes to morbidity and delayed development; it reduces growth, fitness, and cognitive function by reducing appetite, altering feeding patterns, and decreasing absorption of nutrients.3 Children under age five are at far higher risk of death from diarrhea then are older children or adults (WHO, 2002), and risk of death appears to decrease with child age (Yassin, 2000; Fikree et al., 2002). Unsafe water is the major cause of diarrhea. An estimated 88% of deaths from diarrhea could be prevented by ensuring access to safe, improved water and sanitation supplies (Black et al., 2003). However, as of 2012, 39% of Sub-Saharan Africans lacked access to improved drinking water (UNICEF and WHO, 2012).4 Piped water is the premier improved water source, and ensuring access to it can help combat diarrhea for several reasons. First, alternative sources of water are relatively more exposed to diseasecausing contamination (Tonglet et al., 1992; Fewtrell et al., 2005). Second, piped water is relatively easy to collect; it comes from a tap in the home or yard, or from a (usually) nearby public standpipe. Having piped water can contribute to the use of larger quantities

3 Diarrheal diseases include cholera, colitis, gastritis, gastroenteritis, enteritis, intestinal inflammation, and typhoid fever, among others. These originate from pathogens such as rotavirus, adenovirus, and Norwalk Virus (Burstrom et al., 2005). 4 Improved means protected from excreta, contaminants, and insects. Water sources likely to be improved include: piped water into the dwelling; piped water to a yard/plot; a public tap/standpipe; a tube well/borehole; a protected dug well; a protected spring; and rainwater. Unimproved sources include: an unprotected dug well; an unprotected spring; a cart with a tank/drum; a water tanker-truck; and surface water of any kind (UNICEF and WHO, 2006).

K. Kosec / Journal of Health Economics 35 (2014) 1–19

3

of water, and reduces the need to store water in containers. Households that use more water tend to practice better hygiene. Also, storage of water in containers has itself been linked to contamination and diarrhea.5 In addition to diarrhea, environmental (tropical) enteropathy is another condition associated with contaminated water. It occurs when ingested fecal bacteria infect the small intestine, reducing its capacity to absorb nutrients. It may be a leading factor in the persistence of malnutrition despite interventions targeting diarrhea and supplementing children’s diets (Humphrey, 2009; McKay et al., 2010). Thus, improving water quality could have wider-ranging impacts on nutrition beyond any effects on diarrhea morbidity and mortality. The empirical evidence on whether increased access to piped water improves health is mixed. Gamper-Rabindran et al. (2010) show that increasing access to piped water in Brazil decreases infant mortality. However, Bennett (2012) finds that access to cleaner, piped water in the Philippines is correlated with decreased sanitation in the community as a whole. People substitute clean, piped water for sanitary behavior (which has large health externalities). Devoto et al. (2012) do not find reductions in waterborne diseases, including child diarrhea, as a result of a program in urban Morocco providing subsidized credit to obtain an individual water tap over a free communal one.

contracts are incomplete in practice. As a result, competition for the market cannot fully substitute direct competition. Empirically, there have been fewer bidders in the water sector than in other sectors including energy, telecommunications, and roads—possibly due to already relatively high market concentration and technical complexity in this industry (Iimi, 2008).7 Empirical evidence on the consumer impacts of PSP in water is mixed. While some studies find that PSP can improve efficiency and consumer outcomes,8 others are less optimistic or find null results.9 Most econometric studies focus on utility efficiency or access to piped water rather than health. Existing studies have also tended to focus on single countries given data limitations, raising questions of external validity. Galiani et al. (2005) uniquely examine the health implications of PSP. They find that child mortality from water-related diseases in Argentina (an upper-middle income country) fell following PSP, and poor municipalities benefited more than richer ones. However, it is unclear whether this would apply in a developing country context. Institutional and governance challenges abound in developing countries, which could lead to poorer performance by public or private firms. Also, in developing countries it may be relatively more costly if a private firm fails to internalize the positive externalities of using piped water, since many people may realistically turn to unsafe alternatives.

2.2. The effects of PSP on consumers

2.3. Where does PSP in water come about and why?

Private sector participation (PSP) includes concessions, leases, affermage contacts, build-operate-transfer contracts, and management and service contracts. It has been motivated by theoretical efficiency gains, fiscal crises, and the requirements of lenders. The water industry has several features that make PSP less common than in other industries. First, a clean and reliable water supply generates large, positive health externalities. Second, allowing a private firm to profit from something necessary for life is politically controversial. This makes PSP risky for governments as well as for private firms that fear contract cancellation.6 Third, governments can often profit heavily from public control by delaying investments (usually longer than they can in other industries) and earning quasi-rents (Noll, 2002). They often use quasi-rents to reduce water tariffs, thus intensifying public resistance to PSP as it would likely end the subsidies. Finally, provision of piped water is a natural monopoly. It is not desirable to duplicate the water provision network, and fragmentation can limit economies of scale. Theory suggests that PSP can increase utility efficiency if accompanied by competition. Shapiro and Willig (1990) use the distortions in the objectives of public managers (a malevolent government) to show the benefits of private ownership under incomplete contracting. Schmidt (1996) eliminates the assumption of a malevolent government and shows that the real threat of bankruptcy faced by private sector firms, combined with competitive conditions, generate gains from PSP. However, direct competition is difficult with natural monopolies. Demsetz (1968) proposes a possible solution: competition for the market, via a bidding process. However, Goldberg (1976) and Williamson (1976) identify problems with this approach: collusion, asymmetric information, problems in the pricing of assets, and other incumbent advantages may make it anti-competitive. Moreover, any resulting

Why does PSP in the water sector come about in some areas and not in others? Kosec (2012) sheds light on this question using a database of all major PSP contracts in the water sector, worldwide, during 1990–2010 (from Envisager Limited, 2011). Contracts are defined by the firm name and the country in which it is headquartered, the market country (where the PSP is carried out), and the number of water customers (taps) served. Two factors strongly positively associated with a firm signing a new contract in a country are: (a) its existing share of the global market for private, piped water (a measure of relative global experience), and (b) whether there is a colonial relationship between the two countries. Kosec (2012) also shows that the headquarters country’s Gross Domestic Product (GDP), foreign direct investment (FDI), and payment of remittances are not significant predictors of the number of water contracts a firm from that country signs.

5 See Blum and Feachem (1983), Esrey et al. (1991), Black et al. (2003), Fewtrell et al. (2005), and Kremer and Zwane (2007). 6 An AfroBarometer survey of 12 African countries undertaking economic reforms between 1999 and 2001 revealed that only 35% of people preferred private to state ownership.

3. Empirical strategy Lacking a randomized experiment, I control for time-invariant, unobserved heterogeneity by using panel data to estimate the following fixed effects model dijkt = ıpjt + ␥Xijkt + j + ˛k + ˇt + uijkt

(1)

where an observation is a child indexed by i, j indexes the subnational region of residence,10 k indexes the month he is surveyed, and t indexes the year. dijkt is a dummy for whether a child experienced diarrhea at some point during the last two weeks. pjt is a dummy for PSP in the water sector in sub-national region j in year t.

7 On average, there are 3.6 bidders for water and sewerage contracts, 3.8 for energy, 4.7 for roads, and 5.4 for telecommunications (Iimi, 2008). 8 See Galiani and Petrecolla (1996), Ménard and Saussier (2000), Noll et al. (2000), Kirkpatrick et al. (2004), Galiani et al. (2005), Nellis (2005), and Barrera-Osorio et al. (2009). 9 See Byrnes et al. (1986), Tynan (2000), Saal and Parker (2001), Estache and Rossi (2002), Wallsten and Kosec (2008), Clarke et al. (2009), and Kremer et al. (2011). 10 A sub-national region is a province, state, or region within a country; the exact terminology varies across countries.

4

K. Kosec / Journal of Health Economics 35 (2014) 1–19

Xijkt is a vector of child characteristics that may affect susceptibility to disease, including dummies for the child’s age and gender, the mother’s age, the mother’s education level, the martial status of the child’s parents, the number of household members, and whether the child’s house has a natural (earth/clay/dung, etc.) floor, electricity, a refrigerator, a radio, a television, a bicycle, a motorcycle, and a car. j are sub-national region fixed effects, ˛k are month fixed effects, and ˇt are year fixed effects. Sub-national region fixed effects control for time-invariant geographic, institutional, and social features of each region.11 Year fixed effects account for the global state of access to medicine, vaccines, and finance, for the strength of the global economy, and for the state of global experience with private water provision.12 My analysis thus compares the change in outcomes in regions with PSP before and after PSP was introduced to the change in outcomes in a control group of regions that never had PSP. The prediction that PSP in water reduces diarrhea implies ı < 0. I test this prediction in Section 4. There, I also examine whether PSP has heterogenous effects on children at different levels of poverty, and whether PSP in water affects child attendance at school. In Section 5, I explore potential causal channels by examining the effects of PSP on access to piped water, time taken (round trip) to collect water, and access to a flush toilet. Finally, in Section 6, I present two placebo analyses: one exploring PSP’s effect on illnesses it should not affect (acute respiratory illness in the same set of 0–4 year olds, as evidenced by coughing) and another examining PSP’s effect on a group of children it should not affect (those living in rural areas). 3.1. Identification The main threat to identification is the existence of timevarying, unobserved covariates correlated with both PSP in water and child health. This could arise for several reasons. First, governments may be more likely to privatize a public water utility after years of under-investment in infrastructure, when health is already deteriorating.13 Second, people may be more likely to accept or even demand PSP when their government has already revealed incompetence in handling the water sector. The government’s incompetence, however, may also coincide with a deterioration in public healthcare and early childhood development policies. Third, PSP might be part of a larger set of “good governance” reforms that themselves improve public health. Finally, private water utilities may be more likely to invest in the water sector when they suspect that it will be profitable—which could be more likely when there are no major health problems. The possibility of simultaneity is also a potential concern. Even a competent government that has invested heavily in its water infrastructure may get a bad draw which causes child health to suffer (for example, an outbreak of cholera). Addressing this negative shock might require a private investor with more experience in doing so. Donors and lenders observing water sector mismanagement or health problems might also demand PSP in water. Fig. 1 shows the average prevalence of diarrhea for several types of observations: sub-national regions that never undergo PSP

11 Some regions have deserts while others have forests, which can affect water sources and human health. Different regions also have different colonial histories, which have affected modern-day institutions (Acemoglu and Robinson, 2006). 12 On average, we have four years of data on each locality. Given the unbalanced nature of the 1986-2010 panel, year fixed effects index the three-year period in which the observation falls (1986-1988, 1989-1991, etc). 13 For example, Clarke and Cull (2009) show that bank privatizations in Argentina are more likely in times of fiscal and economic crisis, when the costs of raising money through taxes are relatively high. And Galiani et al. (2005) show that poorer municipalities are likely to privatize water in Argentina.

Fig. 1. Average prevalence of diarrhea among water utilities that will undergo PSP, by years private. Notes: The prevalence of diarrhea is the average prevalence (share of children experiencing diarrhea in a two week period). The horizontal line shows the average prevalence of diarrhea among water utilities that are always public. PSP areas are places that undergo PSP at some point; I show the prevalence of diarrhea in places that have not yet undergone PSP (0 years under a contract) and for different numbers of years under a PSP contract. Nine years is the 95th percentile of years under a PSP contract. Source: Author’s calculations based on data from DHS (1986–2010), Pinsent Masons (1999–2011), Envisager Limited (2011), and World Bank WDI (2011).

during 1986–2010, sub-national regions that undergo PSP in the future but are currently public, and sub-national regions that have been under PSP for one, two, three,..., or nine years. The average prevalence of diarrhea among regions that are always public is about 17% (i.e. 17 out of 100 children experienced diarrhea sometime during the last two weeks). In contrast, the average prevalence of diarrhea is about 21% in regions that will eventually undergo PSP but have not yet done so. In the first and second years of a PSP contract, the prevalence of diarrhea lowers to 20%. In years 4, 5, 6, 7, and 8, it drops further to 18%, 16%, 15%, 13%, 13%, and 10%, respectively.14 Thus, PSP may be more likely when diarrhea is more problematic, and the prevalence of diarrhea appears to decrease in years under PSP. The IV strategy, described below, shows robustness of the OLS findings to accounting for time-varying, unobserved covariates driving both PSP and child diarrhea. 3.2. Instrument for private sector participation in water 3.2.1. IV strategy To address threats to identification posed by time-varying, unobserved covariates correlated with both PSP and child health, I construct an instrumental variable. To do so, I first identify the European country that most recently colonized each African country. Appendix Table A.1 (Cannon, 2004) presents a list of colonizers. There are six colonizers of the 39 African countries in the sample: France (16), the United Kingdom (UK) (15), Belgium (3), Portugal (2), Germany (1), Italy (1), and never colonized (1).15 I then use the former colonizer’s time-varying share of the world market for private, piped water (ignoring contracts covering African countries) as an instrument for PSP in the African country. In short, I posit that African countries are relatively more likely to undergo PSP in water during years when their former colonizer’s world, nonAfrican water market share is relatively high.

14 In year nine, it increases slightly. Nine years is the 95th percentile of years under PSP. 15 Liberia is classified as never colonized.

K. Kosec / Journal of Health Economics 35 (2014) 1–19

5

Italy and Germany’s shares are both relatively small, and peaked in the early 2000s. Portugal and Belgium have been largely absent from this market. I estimate the following first stage equation using the same child-level dataset, where pjt is a dummy for PSP in sub-national region j, in year t: pjt = mjt + Xijkt + j + k + t + εijkt

Fig. 2. Market shares of selected countries in the private, non-African world water market, 1990–2010. Notes: Market shares are based on the country’s share of total private water connections (taps) located in countries outside of the African continent. The connections (taps) of all companies headquartered in a given country are said to pertain to that country. The category “Other” includes any country other than these five European countries (even African countries) whose companies have contracts outside of Africa. Source: Pinsent Masons (1999–2011) and Envisager Limited (2011).

Some brief words on African colonization are in order. By the mid-1870s, Europe had only established a few coastal trading posts in Africa, and the colonies of Algeria and South Africa. However, as Pakenham (1991) describes: “Suddenly, in half a generation, the scramble gave Europeans virtually the whole continent: including thirty new colonies and protectorates, 10 million square miles of new territory, and 110 million dazed new subjects, acquired by one method or another. Africa was sliced up like a cake, the pieces swallowed by five rival nations—Germany, Italy, Portugal, France, and Britain.” Under colonial rule, legal systems were imposed and Africans were taxed to fuel the colonizer’s empire. Africans were also exposed to the languages and religions of their colonizers. By the mid 1960s, the majority of African countries had declared independence. However, many African countries still share a special relationship with their former colonizer, as they are linked by language, legal customs, and conventions. Piped water PSP arrangements in Africa typically involve large multinational firms, and frequently involve a company from the former colonizer country—likely because of advantages conferred by these commonalities, which make former colonies relatively receptive customers. The number of customers a multinational water firm serves can expand merely due to population growth. However, growth in its market share indicates expansion relative to other firms and relative to the global prevalence of PSP in water (which is itself captured by year fixed effects). As a former colonizer’s world, non-African water market share grows, its firms are actively expanding their capacity outside of Africa, and are thus more relatively more likely to aggressively pursue new contracts in Africa. The market share of each colonizer changes over time, as shown in Fig. 2. The UK’s share has declined over the last 20 years, France’s share has increased, and the shares held by countries other than the UK, France, Germany, Italy, and Portugal have also increased markedly. The decline in the UK’s share is not due to a reduction in the number of private water customers of UK firms; indeed, these increased by 15% during 1990–2010. Instead, it is due to UK companies’ failure to keep pace with expansions in France and other countries. For example, France added 21 times more private water customers during 1990–2010 than did the UK, primarily through contracts signed by its largest three firms: Suez, Veolia, and SAUR.

(2)

mjt is the water market share of the European colonizer of subnational region j in year t, which ranges from 0 to 1. j are sub-national region fixed effects,  k are month fixed effects, and t are year fixed effects. Xijkt is a vector of child characteristics, as previously defined. To further ensure that the results are not driven by a correlation of PSP or colonizer water market shares with foreign aid or investment, I control for several African country-level variables: (i) net inflows of foreign direct investment (FDI), (ii) net official flows from all UN agencies, (iii) official development assistance (ODA) from the former colonizer country—all in 1000s of constant 2000 USD per capita—and (iv) the average annual amount of World Bank water and sewerage sector loans and grants over the previous five years (in constant 2000 USD per capita). Shortly, I show that their inclusion has little effect on the sign, magnitude, or significance of ı in Eq. (1). 3.2.2. Exogeneity of the excluded instrument For the proposed instrumental variable to be valid, the exclusion restriction requires that changes in the water market share of the former European colonizer only affect changes in child health in the African country by changing the likelihood that the African country has PSP in water. It cannot be the case that the water market share of the former European colonizer has a direct effect on child health in Africa, or that it has an effect running through omitted variables that themselves affect child health. The exclusion restriction might fail if the water market share of European colonizers were strongly correlated with the strength of their economy. Strong European economies might invest more heavily in their former colonies, which could itself improve child health in Africa. Table 1, Panel A explores this possibility using a panel dataset of the six European colonizer countries in the sample during 1986–2010. This is the same period for which I analyze child health data. Column (1) contains the results of three separate regressions: in each case, the water market share of the European colonizer is regressed on one of three key independent variables as well as on country and year fixed effects and country-specific time trends. The three key independent variables capture the strength of the European country’s economy and include: GDP (billions of constant 2005$), GDP per capita (1000s constant 2005$), and the growth rate of GDP per capita over the previous year. I cluster standard errors at the European country level. In none of these regressions is the European country’s water market share correlated with the strength of its economy. A European country having a greater share of the world water market does not appear to be somehow capturing the strength of its economy. The exclusion restriction could also fail if the strength of the African country’s economy or its investments in health were strongly correlated with either the strength of their colonizer’s economy or with its water market share.16 Table 1, Panel B looks for such correlations using a panel dataset of the 38 African countries in

16 This could occur, for example, if colonizers with large water market shares had other companies simultaneously privatizing other industries (e.g., telecommunications, electricity, natural gas, transportation) and thus boosting growth in the former colony’s economy.

6 Table 1 OLS results, showing a lack of correlation between European countries’ world water market shares and their economies, and showing a lack of correlation between African countries’ economies and those of their former colonizer (1986–2010). Panel A: analysis at the European country level (1) Water market share (0–1)

GDP (billions $)

0.00007 [0.00021] 0.001 [0.010] −0.023 [0.015] 150

GDP per capita (1000s $) Growth rate of GDP per capita since last year Observations

Panel B: analysis at the African country level (1)

(2)

(3)

(4)

(5)

(6)

Dependent variable

GDP (billions $)

GDP per capita ($)

Growth rate of GDP per capita since last year

Health expenditure per capita, constant 2005 US$

Health expenditure as a share of GDP

Public health expenditure per capita, constant 2005 US$

GDP (billions $) of country’s colonizer

0.007 [0.008] 0.382 [0.474] 5.490 [3.561] 933

−0.177 [0.299] 0.311 [16.036] 32.080 [95.016] 933

−0.001 [0.005] 0.002 [0.336] 2.096 [1.312] 930

−0.036 [0.024] −2.106 [1.573] 15.634 [24.957] 593

−0.004 [0.004] −0.244 [0.231] 2.807 [2.893] 599

−0.005 [0.020] 0.062 [1.527] 12.771 [18.088] 593

GDP per capita (1000s $) of country’s colonizer Water market share of country’s colonizer (0–1) Observations

Source: Pinsent Masons (1999–2011), Envisager Limited (2011), and World Bank WDI (2011). Notes: Each cell comes from a separate regression of the dependent variable (column heading) regressed on the main independent variable (row heading). All specifications include country and year fixed effects and countryspecific time trends. Robust standard errors are in parentheses and clustered at the country level. In Panel A, an observation is a European colonizer country–year (six in total: Belgium, France, UK, Portugal, Italy, and Germany). In Panel B, an observation is an African country–year (38 in total, representing all African countries in the sample used for analysis which have a former European colonizer). Dollar amounts are in constant 2005 $.

K. Kosec / Journal of Health Economics 35 (2014) 1–19

Dependent variable

K. Kosec / Journal of Health Economics 35 (2014) 1–19

the sample during 1986–2010.17 Through 18 separate regressions, Panel B regresses each of the six African country-level outcomes on each of three variables describing their colonizer: the colonizer’s GDP (billions of constant 2005$), the colonizer’s GDP per capita (1000s constant 2005$), and the colonizer’s water market share. Once again, each regression controls for country and year fixed effects and country-specific time trends, and I cluster standard errors at the African country level. Column (1) shows that GDP in the sample African countries is neither correlated with the GDP nor GDP per capita of its colonizer, and is also uncorrelated with the colonizer’s water market share—the excluded instrumental variable from Eq. (2). Columns (2)–(6) similarly show that a number of other characteristics of African country economies are uncorrelated with these features of their colonizer. They include, respectively: GDP per capita (constant 2005$), the growth rate of GDP per capita over the previous year, health expenditure per capita (constant 2005$), health expenditure as a share of GDP, and public health expenditure per capita (constant 2005$). It does not appear to be the case that the strength of African economies or their propensity to invest in health can be explained by the economic strength of their colonizer or their colonizer’s water market share (the excluded instrument). These findings strengthen confidence that the exclusion restriction is satisfied. 3.3. Data I use an unbalanced, pseudo-panel dataset that spans 1986–2010 and includes samples of under-five children from 39 African countries. An observation is a child, described by his country and sub-national region of residence, and the month and year he is surveyed. Children are not followed over time. There are between one and seven years of data from each sub-national region, with an average of four and a standard deviation of 2. Data come from several sources, matched at the sub-national (where possible) or national level. 3.3.1. Child-level data My primary data source is a set of 99 Demographic and Health Surveys (DHS) conducted in Africa during 1986–2010. This is the universe of publicly available and useable, standard DHS surveys from Africa.18 They cover 39 countries—33 of which have data from multiple years.19 I do not exclude data from countries with only one round of data so as to allow these countries to influence the intercepts of the regressions.20 I combined the data into a single, child-level dataset comprised of all children aged 0–4 from urban areas. These children come from 372 different sub-national regions. Throughout the paper, I cluster standard errors at the sub-national region level, as PSP varies at this level. Thus, there are a total of 372 urban clusters in the analysis. DHS surveys are a leading source of information on health and nutrition. They collect data on household member characteristics (age, education level, marital status, household size, minutes to get

17 This number is not 39 because Liberia was not colonized and thus has no former colonizer. 18 I drop only two publicly available standard DHS surveys from Africa: the 1988 Egypt survey, as there was not a variable describing whether children experienced diarrhea in the previous two weeks, and the 1999 Senegal survey, as only raw data were available. 19 The six countries for which I have one year of data are: Burundi, Comoros, Democratic Republic of the Congo, Republic of the Congo, Sierra Leone, South Africa, and Tunisia. 20 While this choice also affects the standard errors, the baseline OLS and IV estimates of the coefficient on PSP are statistically significant at the same levels when I drop observations from countries for which there is only one year of data and re-estimate the standard errors. These results are available upon request.

7

to a water source and return, whether children attend school, and whether children aged 0–4 experienced diarrhea or a cough in the last two weeks),21 household characteristics (whether there is a natural floor, electricity, access to a flush toilet, and access to piped water), and assets owned (including a refrigerator, radio, television, bicycle, motorcycle, and car). A household is coded as having piped water if the primary drinking water source comes from a pipe—whether in the home or from a public standpipe.22 Appendix Table A.2 shows the countries, sub-national regions, and years for which data are available.23 PSP arrangements in water usually affect only urban areas, so I exclude rural households from most analysis. Among urban-dwelling children under age five, about 35% live in capitals or large cities, 35% live in small cities, and 30% in towns. I study several child-level outcomes. The primary health outcome is a dummy for a child aged 0–4 experiencing diarrhea at some point during the previous two weeks.24 This captures both the extensive and intensive margins of diarrhea. When diarrhea is more intensive, it is present longer. This makes a child more likely to have recently experienced diarrhea when the household is surveyed. Whether a child experienced an illness with a cough during the last 2 weeks provides information on whether the child suffers from acute respiratory illness. Finally, if PSP affects child health, access to piped water or flush toilets, or time taken to collect water, it may have additional implications for children’s time use. Thus, I explore any impacts of PSP on whether children aged 7–17 (i.e. of school-going age) currently attend school. 3.3.2. Notes about the imbalance of the panel The pseudo-panel dataset is unbalanced due to the fact that DHS surveys are not carried out in every country and in every year, for obvious logistical reasons. A few notes on the imbalance of this dataset are warranted. That data are missing for a large number of countries and years could bias estimates of the impacts of PSP on child health. If data (for particular countries and years) were missing only due to completely random reasons, the imbalance of the panel would not generate bias. However, if there would have been systematically different values of the covariates for missing country-year combinations, then having an unbalanced panel would generate bias. Fortunately, DHS surveys are carried out (or not) due to a lot of considerations other than prevailing trends in child health or changes in water sector management, increasing the plausibility that missing data will not generate bias. However, it is a distinct possibility that data are more likely to be missing given particular values of the covariates. To be conservative, it is

21 Data on other symptoms of acute respiratory illness were not available consistently across DHS surveys. 22 Since Jalan and Ravallion (2003) find that the impact of piped water on the prevalence of diarrhea is largely unaffected by whether households obtain piped water from an in-home tap versus a public tap (standpipe), I do not differentiate between the two. 23 In a few countries, the sub-national regions identified in the DHS vary over time. To address such inconsistencies, I aimed to maximize the length of the panel while still ensuring that the same code never refers to two different (even if quite similar) entities in different years. In the case of two sub-national regions given separate codes in one year, but lumped together (i.e. identified by a single code) in another year, I re-coded the data to lump them together in all years. For the vast majority of countries with inconsistent sub-national region coding over time, this was sufficient to crate a harmonized set of sub-national region identifiers. When a simple re-coding could not be used to construct comparable sub-national regions over time, I simply allowed each region to carry a unique code, which resulted in fewer observations on each region. All coding is available upon request. 24 This is the outcome studied by Günther and Fink (2010). I use this measure since the causes and consequences of diarrhea are well-known, and because data on the prevalence of water-borne diseases and on mortality from them are limited for most African countries.

8

K. Kosec / Journal of Health Economics 35 (2014) 1–19

safer to conclude that the empirical results reflect the impacts of PSP in the particular countries and years covered by the DHS—not the impacts of PSP in Africa as a whole during 1986–2010, nor the effects of PSP in the 39 sample countries during 1986–2010. 3.3.3. Sub-national-level data on PSP in water I cross-referenced several sources to construct a sub-national panel dataset spanning 1986–2010 on locations of PSP in water: Pinsent Masons World Water Year Books (1999–2011), Envisager Limited (2011), the World Bank Private Participation in Infrastructure (PPI) Database (2010), Hall et al. (2002), and Lexis Nexis Academic (2011). Appendix Table A.4 summarizes these PSP arrangements. Overall, 20 countries experienced PSP in water in at least some sub-national region during the sample period, while 19 countries always had publicly run water. In 14 countries, PSP occurred at the national level, affecting all urban water networks.25 In six countries, PSP only affected select sub-national regions, or occasionally only select cities in a region.26 I use this information to code a dummy for PSP in each sub-national region—year.27 Following the same procedure, I also constructed a panel dataset on locations of PSP in sanitation. I found no regions with PSP in sanitation but without PSP in water. However, I found that 32% of sample observations with PSP in water were accompanied by PSP in sanitation, while 68% involved only PSP in water. As private water firms are generally involved in only water or sanitation, their involvement in the water sector is unlikely to lead to them pursuing PSP in some unrelated industry. Envisager Limited (2011) also provided a database of the size of all PSP contracts in the water sector, worldwide, during 1985–2010.28 They report the country of the company awarded the contract, the country in which the project took place, and the number of customers (water taps) served by each contract. I used these data to compute the annual market shares of each European colonizer in the world (non-African) private water market during 1986–2010, for use as an instrumental variable. 3.3.4. National-level control variables I also collected data on national-level controls. From the World Development Indicators (WDI) databank, I gathered data on per capita net inflows of FDI, net official flows from all UN agencies, and ODA (World Bank, 2011a). From a listing of all World Bank water and sewerage sector grants and loans, I computed the average annual amount of sector funding per capita given to each African country over the previous five years (World Bank, 2011b).29 3.3.5. Summary statistics Table 2 summarizes the means and standard deviations of key variables. The average urban household has electricity, a nonnatural floor, piped water, and a radio, but no flush toilet, television, bicycle, motorcycle, or car. About 16% of under-five children experienced diarrhea in the last two weeks. Diarrhea is most common among the children of less-educated mothers and those with fewer assets and amenities. Among children whose mothers have no education, 18% experienced diarrhea in the last two weeks; among those whose mothers have secondary or more education, only

Table 2 Summary statistics. Variable Dummy – child experienced diarrhea during last two weeks Dummy – diarrhea last two weeks (mom – no education) Dummy – diarrhea last two weeks (mom – primary educ or less) Dummy – diarrhea last two weeks (mom – primary educ) Dummy – diarrhea last two weeks (mom – primary + educ) Dummy – diarrhea last two weeks (mom – secondary + educ) Dummy – diarrhea last two weeks (1st quartile – poverty index) Dummy – diarrhea last two weeks (2nd quartile – poverty index) Dummy – diarrhea last two weeks (3rd quartile – poverty index) Dummy – diarrhea last two weeks (top quartile – poverty index) Dummy – child had a cough during the last two weeks Dummy – PSP in water Years under a PSP contract Non-African water market share of final colonizer Non-African water market share of original colonizer Dummy – child is a boy Number of household members Dummy – parents are married Mother’s age Dummy – mother has no education Dummy – mother has primary education Dummy – secondary or more education Dummy – HH has electricity Dummy – HH has a radio Dummy – HH has a television Dummy – HH has a refrigerator Dummy – HH has a bicycle Dummy – HH has a motorcycle Dummy – HH has a car Dummy – HH has a natural floor Dummy – HH has piped water Dummy – HH has a flush toilet Minutes spent collecting water World Bank water grants/ loans per capita (constant 2000 USD) Net UN aid per capita (100s constant 2000 USD) FDI (net inflows) per capita (100s constant 2000 USD) ODA from final colonizer per capita (100s 2000 USD) Dummy – child currently enrolled in school (age 7–17)

Mean

S.D.

0.160

0.367

0.180

0.384

0.178

0.383

0.177

0.381

0.151

0.358

0.128

0.334

0.199

0.399

0.174

0.379

0.145

0.352

0.121

0.326

0.287 0.289 2.666 0.282 0.264 0.504 7.470 0.767 28.571 0.311 0.324 0.365 0.544 0.762 0.463 0.298 0.212 0.131 0.102 0.225 0.701 0.315 5.277 0.100

0.452 0.453 7.369 0.247 0.248 0.500 4.671 0.423 6.590 0.463 0.468 0.481 0.498 0.426 0.499 0.458 0.409 0.337 0.302 0.418 0.458 0.464 17.830 0.108

0.022 0.192 0.129 0.715

0.026 0.327 0.361 0.451

Source: DHS (1986–2010), Pinsent Masons (1999–2011), Envisager Limited (2011), and World Bank, WDI (2011). Notes: Nearly all data are summarized for all urban-dwelling children aged 0–4 during 1986–2010 (N = 165, 543). School enrollment is summarized for all urbandwelling children aged 7–17 during 1986–2010 (N = 74, 999).

13% experienced diarrhea (almost 30% less diarrhea). About 29% of under-five children experienced a cough in the last two weeks. Of the regions and years covered by DHS surveys, about 29% are marked by a PSP arrangement, and the average length of time under PSP is 2.7 years. 4. Effect of PSP on child health and education outcomes

25

This was the case for: Burkina Faso, Cameroon, Central African Republic, Gabon, Ghana, Guinea, Ivory Coast, Mali, Mozambique, Namibia, Niger, Republic of the Congo, Rwanda, and Senegal. 26 This was the case for: Egypt, Kenya, Morocco, Tanzania, Uganda, and Zambia. 27 A sub-national is coded as being under PSP if its largest city is under PSP. In most cases, all cities in a sub-national region either are or are not under PSP. 28 Only contracts serving 10,000 or more, are included. 29 The specific major sector I used is: General Water, Sanitation, and Flood Protection.

4.1. Diarrhea prevalence: main OLS results Table 3, columns (1)–(4), presents OLS estimates of the effect of PSP in water on diarrhea in under-five children. The results suggest that PSP has a robust, negative effect on diarrhea (ı < 0). All specifications include both sub-national region and year fixed effects, and other controls are introduced sequentially. The coefficient on PSP,

K. Kosec / Journal of Health Economics 35 (2014) 1–19

9

Table 3 OLS results, showing the effect of PSP in water on the prevalence of diarrhea in under-five children. (1)

(2)

(3)

(4)

−0.022** [0.010]

−0.022** [0.010]

−0.026*** [0.010]

(5)

(6)

(7)

(8)

−0.037*** [0.012] 0.009 [0.015]

−0.035*** [0.013] 0.007 [0.015]

−0.036*** [0.013] 0.009 [0.015] 0.006* [0.004] −0.014*** [0.004] −0.013*** [0.004] −0.013*** [0.003] −0.011*** [0.003] −0.010*** [0.004] −0.002 [0.003] 0.001 [0.004] −0.004 [0.004] 0.015*** [0.004]

160,342 0.032 No

160,342 0.064 Yes

160,342 0.067 Yes

−0.037*** [0.013] 0.001 [0.014] 0.006* [0.004] −0.014*** [0.004] −0.012*** [0.004] −0.013*** [0.003] −0.011*** [0.003] −0.010 [0.004] −0.002 [0.003] 0.001 [0.004] −0.004 [0.004] 0.015*** [0.004] 0.038 [0.027] 0.233** [0.095] 0.015 [0.019] 0.004 [0.030] 160,342 0.067 Yes

Dependent variable: dummy – child experienced diarrhea during last two weeks Dummy – PSP in water

−0.023** [0.010]

Dummy – PSP in water, not sanitation Dummy – PSP in water and sanitation Dummy – mother has primary education

0.007* [0.004] −0.014*** [0.004] −0.012*** [0.004] −0.013*** [0.003] −0.011*** [0.003] −0.01*** [0.004] −0.002 [0.003] 0.001 [0.004] −0.004 [0.004] 0.015*** [0.004]

Dummy – secondary or more education Dummy – HH has electricity Dummy – HH has a radio Dummy – HH has a television Dummy – HH has a refrigerator Dummy – HH has a bicycle Dummy – HH has a motorcycle Dummy – HH has a car Dummy – HH has a natural floor World Bank water grants/ loans per capita Net UN aid per capita FDI (net inflows) per capita ODA from former colonizer per capita Observations R-squared Dummies – age, child gender, marital status, # HH members

160,342 0.03 No

160,342 0.06 Yes

160,342 0.07 Yes

0.007* [0.004] −0.014*** [0.004] −0.012*** [0.004] −0.013*** [0.003] −0.011*** [0.003] −0.01*** [0.004] −0.002 [0.003] 0.001 [0.004] −0.004 [0.004] 0.015*** [0.004] 0.044 [0.028] 0.266*** [0.096] 0.022 [0.018] 0.008 [0.030] 160,342 0.07 Yes

Source: Author’s calculations based on data from DHS (1986–2010), Pinsent Masons (1999–2011), Envisager Limited (2011), and World Bank WDI (2011). Notes: Robust standard errors are in parentheses and clustered at the sub-national region level. All specifications include sub-national region, month, and year fixed effects. Age dummies are for both mother and child. World Bank water and sewerage sector loans and grants is the total over the previous five years in constant 2000 USD per capita. Net UN aid is total received from any UN agency, in 100s of constant 2000 USD per capita. FDI (net inflows) is the sum of equity capital, reinvestment of earnings, other long-term capital, and short-term capital from other countries in 100s of constant 2000 USD per capita. ODA is official development assistance from the country’s final colonizer, in 100s of constant 2000 USD per capita. As there are no instances of PSP in sanitation without PSP in water, the base group in specifications (4)–(8) is public operation of both water and sanitation. *** indicates p