Literature review

Our study dovetails several aspects of the existing literature on conflict and health. Here, we briefly review this literature and the context in which our research question arises.

Access to safe drinking water and waste disposal have large and well-documented effects on health. In fact, Magana-Arachchi and Wanigatunge (Reference Magana-Arachchi and Wanigatunge2020) argue that waterborne pathogens are the most common vector for disease globally. Dirty water is associated with the transmission of cholera, typhoid, hepatitis, dysentery, Guineaworm disease and polio, and also with diarrhea, which is a leading cause of childhood wasting and mortality (Centers for Disease Control, 2022; World Bank, 2023; Troeger et al., Reference Troeger, Blacker, Khalil, Rao, Cao, Zimsen and Reiner2018). With respect to childhood diarrhea, a majority of documented cases in Cameroon and Chad were directly attributable to dirty water (Ntouda et al., Reference Ntouda, Sikodf, Ibrahim and Abba2013; Yongsi, Reference Yongsi2010). Reconstitution of infant formula with unsafe water is a particular problem (Diwakar et al., Reference Diwakar, Malcolm and Naufal2019). Bisung and Elliot (Reference Bisung and Elliott2017) argue that the health consequences of dirty water go beyond physical health, and include psychosocial harm, with a disproportionate impact on women. In sum, the World Bank links access to clean drinking water with higher levels of education, a reduction in health expenditures, and more robust economic growth (World Bank, 2023).

The quality of drinking water is directly related to its source. The World Health Organization and UNICEF define an improved drinking-water source as a source that “by nature of its construction or through active intervention is protected from outside contamination” (World Bank, 2023). Shaheed et al. (Reference Shaheed, Orgill, Montgomery, Jeuland and Brown2014) note that, while “improved” drinking water systems do reduce health risks, the moniker can mask considerable variation in the safety and accessibility of water among improved systems.

The analysis in Shaheed et al. (Reference Shaheed, Orgill, Montgomery, Jeuland and Brown2014) suggests that, even for civilians who ordinarily have access to an improved public water system, conflict can disrupt the availability and safety of their drinking water. Gates et al. (Reference Gates, Hegre, Nygard and Strand2012) find from a study of 146 countries that an increase in conflict fatalities is associated with a statistically significant decline in access to potable water; specifically, a conflict with 2500 battle deaths deprives about 1.8% of the surrounding population of access.

The Middle East is replete with examples. Conflict in Gaza has damaged wells, pipelines and water treatment plants there, and the ongoing embargo often renders timely repairs impossible. As a result, wastewater contamination and overuse have led to irreversible damage to Gaza’s single aquifer and, by 2015, less than 5 percent of water from that aquifer “was deemed fit for human consumption” (Diep et al., Reference Diep, Hayward, Walnycki, Husseiki and Karlsson2017; Lazarou, Reference Lazarou2016; UNEP, 2009; Gostoli Reference Gostoli2016; World Bank 2016). Repercussions can extend beyond conflict zones. In Jordan, an influx of refugees has led to excess stress on water systems there, leading to physical damage to the piped water infrastructure (Diep et al., Reference Diep, Hayward, Walnycki, Husseiki and Karlsson2017). Households can respond to these stresses in a variety of ways, although the solutions may be costly. One study in southern Syria found that, between 2016 and 2017, the use of piped water as the main drinking water source for households declined from 22.0% to 15.3% over the survey rounds. This forced households to access water from private trucking networks, spending on average a fifth of their income on water (Sikder et al., Reference Sikder, Daraz, Lantagne and Saltori2018).

Turning specifically to Iraq, the World Resources Institute classifies Iraq as a country with high baseline water stress (Hofste et al., Reference Hofste, Reig and Schleifer2019). Water levels from the Tigris and Euphrates, Iraq’s main sources of water, risk continuing decline from prolonged drought and from damning of upstream water sources (Gavlak, Reference Gavlak2022; Reliefweb, 2022b). We should note here that there is considerable subnational variation, and a drinking water crisis in rural areas of Basra in 2018 led to violent protests and to at least 118,000 hospitalizations (Baker, Reference Baker2019). Overall, it has been estimated that 1.6 million internally displaced persons and returning refugees will require emergency water, sanitation and hygiene (WASH) services (Reliefweb, 2022a).

Unsurprisingly, persistent conflict in Iraq has only aggravated this problem. A study drawing on a meta-analysis of unpublished literature on Basra explored the impact of armed conflict on drinking water service from 1978 to 2013. The authors observed a steep decline in service quality of water, which they attributed to a lack of water treatment chemicals and spare parts, in addition to a brain drain among water service staff (Zeitoun et al., Reference Zeitoun, Elaydi, Dross, Talhami, de Pinho-Oliveira and Cordoba2017). Also with respect to Iraq, Dowdeswell and Hania (Reference Dowdeswell and Hania2014) discuss the limited access to potable drinking water and adequate sewage systems that many communities in Iraq face because of the ongoing war there, and note in particular that many public water sources are contaminated. Combatants specifically targeted water pipelines during battle for Mosul in 2016, which left 650,000 people without access to piped water; the authors specifically noted that continuing unrest following the battle made it impossible for staff to assess and repair damage (Diep et al., Reference Diep, Hayward, Walnycki, Husseiki and Karlsson2017).

Violent conflict is known to have severe consequences for the health of affected populations, spanning physical and mental health, and increased risk of death and disability from a host of causes. Furthermore, these effects are potentially persistent for a decade or more after the conflict’s end, with the costs for women and children especially large (Kadir et al., Reference Kadir, Shenoda and Goldhagen2018; Ghobarah et al., Reference Ghobarah, Huth and Russett2003). Given the extensive documentation of the damage that conflict can do to WASH infrastructure, coupled with primacy of access to clean drinking water for health, it seems obvious that water is a natural linking construct for the relationship between conflict and health. Nevertheless, Gates et al. (Reference Gates, Hegre, Nygard and Strand2012) claimed that the literature on this point is “at best scarce”. One study finds that, in 16 conflict-affected countries, including Iraq, almost three times as many children under the age of 15 died from inadequate WASH facilities than from violence that was directly linked to conflict (UNICEF, 2019). Certainly, understanding the relationship between armed conflict and WASH resources is an important element of understanding the long-term development consequences of war.

We should briefly discuss the broader context for this study. While our paper focuses on how conflict can affect access to water, the converse has become a major focal point for the Middle East. The United Nations has called the Middle East the “most water-scarce region” in the world (Food and Agriculture Organization of the United Nations,2018), and it contains 17 countries that fall below the “water poverty line” (Scott, Reference Scott2019). Gleick (Reference Gleick2014) has notably emphasized the role of water in Syria’s ongoing civil war. Haddadin (Reference Haddadin2001) and Salameh et al. (Reference Salameh, Alraggad and Harahsheh2021) engage in an extensive analysis of the multifaceted array of political, economic, and social consequences of water stress in the Middle East, including the security implications and the potential for future violent water-related conflicts. Sullivan and Jemmali (Reference Sullivan and Jemmali2014) discuss the economic dimensions of this problem, particularly for oil-wealthy countries that suffer from water scarcity. Note that this analysis introduces the possibility of two-way effects: increasing water scarcity at the national and subnational level can increase the prevalence of water-related conflict, and this conflict in turn affects the condition and availability of water-related infrastructure.

Data and methods

Our paper uses microlevel data from the MICS, which is a nationally representative data set of Iraqi households collected by a joint effort between the Iraqi government and the United Nations International Children’s Emergency Fund (UNICEF). We use the three waves of data that are publicly available: 2006, 2011 and the recently added 2018 wave. Households are selected using stratified random sampling, in order to be representative of the governorates that make up Iraq.Footnote ¹ The response rate to the survey is quite high, 98% for the 2006 wave and 99% for the other two years, and the sample size is large. Our unit of observation is the household, and the survey includes 17,739 households in 2006, 35,662 households in 2011 and 20,206 households in 2018, for a total sample size of 73,608 households. MICS undertakes these surveys approximately every five years. In 2018, certain districts in two provinces were excluded because of insecurity; we discuss the potential bias and implications for our findings in the next section. Because MICS uses stratified random sampling, it could be possible for a household to be surveyed in more than one round. However, survey administrators assign a unique household ID each year, and so there is no way to use the data to create a panel.

We derive our primary dependent variables from the response to the following question: “What is the main source of drinking water used by members of your household?”.Footnote ² The survey offers 14 options for the participant (mainly the head of household) to select. Table 1 lists the full set of choices available to respondents, along with their overall frequencies in the data. Each respondent is prompted to select one option from among the 14 given. We group these responses into five categories:

1. 1. Piped water, which includes water piped directly into the dwelling or into the household’s yard or plot

2. 2. Public water, which includes public taps, wells of all types, and springs of all types

3. 3. Tank on truck/cart water, which includes water carried in tanker trucks or on carts

4. 4. Bottled water, which includes packaged water and bottled water

5. 5. Other water sources, including surface water and any other source not listed

Table 1. Main source of drinking water used

Note: Averages over all survey years.

The survey data do not provide any additional detail about the fifth category (about 12% of the responses), and so we treat these observations as missing data since we lack information about how the household obtains drinking water. In the next section, we do explore the robustness of our results to the inclusion or non-inclusion of these observations.

To supplement our main findings, we also consider the determination of whether a family travels to obtain water. Specifically, the survey asks: “How long does it take for members of your household to (go there) get water and come back?”. One option provided to respondents is that members of the household do not travel to acquire drinking water. For households that do travel, the survey records the number of minutes required for the trip. We operationalize this as a dummy variable equal to 1 if the household travels to obtain drinking water and equal to 0 otherwise.

To better understand the implications of our findings, we should discuss the dynamics of these various water sources for households. Piped water is generally thought to be the safest and most convenient option, as it does not require leaving the home to acquire drinking water. However, crucially, as we discussed in the previous section, piped water may not necessarily be the highest in quality, especially during times of conflict. Public water sources, by contrast, require household members to travel in order to obtain water, potentially to crowded areas. Tankers carrying water travel to households using the road network and pump water into water tanks. Bottled water can either be purchased or delivered to the dwelling, implying that households may or may not have to travel in order to acquire it.

Conceptually, the effect of conflict on the relative merits of these various means of obtaining water is uncertain. On one hand, one would expect higher conflict intensity to hinder traveling and thus reduce the likelihood that a household member would travel to obtain water from a public source or a store or that tanker-delivered water would be easily accessible. On the other hand, infrastructure damage could reduce the quality of piped water. Finally, if conflict disrupts the economy and stresses household budgets, bottled water may become unaffordable. The ambiguity of these effects motivates an empirical analysis of the question.

Our primary explanatory variable of interest is the intensity of armed conflict, which we derive from the number of conflict-related civilian deaths per 1000 population in the governorate in which the household is located. The source of these data is the IBC, a comprehensive source of conflict data in Iraq, and the curators of this database have maintained these records continuously since 2003 (Carpenter et al., Reference Carpenter, Fuller and Roberts2013). Specifically, our main measure of conflict intensity is the conflict-related death rate in the household’s governorate one year prior to survey administration. As a check on the robustness of our results, we alternatively consider death rates two years prior to survey administration and the contemporaneous death rate in the survey administration year. We then use these death rates to construct a treatment variable that is equal to one if a governorate experienced high conflict during the respective period and equal to zero if the governorate did not experience high conflict.Footnote ³ Following Diwakar et al. (Reference Diwakar, Malcolm and Naufal2019), Malcolm et al. (Reference Malcolm, Diwakar and Naufal2020) and Naufal et al. (Reference Naufal, Malcolm and Diwakar2019), we define high conflict as the 75^th percentile of conflict-related civilian death rates within the distribution of death rates across all governorates for that specific year. In other words, the treatment variable is one for a household residing in a governorate where the number of conflict-related civilian deaths per 1000 residents is higher than the death rate for the 75^th percentile of all governorates for that year. We also employ standard household-level controls: the head of household’s age, the head of household’s gender (1 for male), the head of household’s level of education, the number of bedrooms in the primary dwelling per household member, the total number of household members and whether the household lives in an urban area (1 for urban). Number of bedrooms per household member is our proxy for household wealth, as there is no direct measure of household wealth or income available for all three survey rounds.Footnote ⁴

Table 2 presents full summary statistics on all variables, aggregated across the full sample. Table 3 presents means for all variables, but separated by year and by treatment status. Household characteristics are roughly comparable across treatment and control groups and across the three survey waves.

Table 2. Descriptive statistics

Note: Summary statistics calculated over all survey years. Age is measured in years. Gender = 1 for male. Primary education = 1 if head of household completed primary education but does not hold a secondary credential. Secondary+ education = 1 if head of household completed secondary education. Number of bedrooms and number of members in household are counts. Urban = 1 if residence is in an urban area. Main drinking water sources are given as a proportion. Conflict-related deaths are the number of civilian conflict-related deaths in the governorate in which the respondent lives, per 1000 persons in the governorate, lagged the respective number of years from survey administration.

Table 3. Means by group and year

Note: Arithmetic means by group and year. See Table 2 notes for variable definitions.

Regarding statistical methodology, an important piece of context surrounding our choice of model is that the nature of armed conflict in Iraq changed dramatically from the first data wave in 2006 to the most recent wave in 2018. In 2006, most of the conflict in Iraq was related to the 2003 US invasion and to the removal of Saddam Hussein’s regime. By contrast, the main source of violent conflict in 2018 was the aftermath of the fall of the Islamic State of Iraq and the Levant (ISIL). These various conflicts affected different geographical areas in different ways. For example, the Kurdish-dominated region in northern Iraq was relatively conflict-free during the first survey wave but was heavily affected by ISIL-related conflict. What this means for us is that certain governorates are high-conflict governorates in some years, but not in others, which necessitates an extension of the standard difference-in-differences approach. The generalized difference-in-differences estimator described by Hansen (Reference Hansen2022) allows for more than two periods and also allows for governorates to switch between the treatment and control groups across periods, which gives us the opportunity to better exploit the evolution in the nature of the Iraqi conflict over time as an explanatory variable. Note that our model expands on existing work on armed conflict in Iraq, which uses simple difference-in-differences (Cetorelli, Reference Cetorelli2015; Malcolm et al., Reference Malcolm, Diwakar and Naufal2020; Naufal et al., Reference Naufal, Malcolm and Diwakar2019). Explicitly, the baseline model is:

(1) $${Y_{i,t}} = {\beta _0} + {\beta _1}{T_{i,t}} + {\beta _2}H{H_{i,t}} + {\alpha _g} + {\delta _t} + {\varepsilon _{i,t}}$$

Here, ${Y_{i,t}}$ is the outcome (type of water source) for household $i$ in period $t$ . The dummy variable ${T_{i,t}}$ is the treatment status of the governorate in which household $i$ resides in period $t$ . Importantly, this treatment status can vary within groups (governorates) across time. A governorate might be part of the high-conflict treated group in one period, but a part of the low-conflict control group in another. The vector $H{H_{i,t}}$ constitutes our household-level control variables. The dummies ${\alpha _g}$ are a set of governorate indicators and ${\delta _t}$ are a set of period indicators. These are just the general versions of the treatment and post-year indicators in the conventional difference-in-differences setup. If the model is properly specified, then ${\beta _1}$ is the effect of treatment (conflict) on the household’s respective source of drinking water.

Given the binary nature of the outcome variable (main drinking water source), we adapt the generalized difference-in-differences estimator in Equation (1) by incorporating a probit linking function:

(2) $$\Pr \left( {{Y_{i,t}} = 1} \right) = {\it \Phi} \left( {{{\hat \beta }_0} + {{\hat \beta }_1}{T_{i,t}} + {{\hat \beta }_2}H{H_{i,t}} + {\alpha _i} + {\delta _t}} \right)$$

where ${Y_{i,t}}$ is a binary variable reflecting whether a household primarily uses the respective source of drinking water, and ${T_{i,t}}$ is treatment status as described above.

As a robustness check on our main results, we also adopt a multinominal logistic approach, which examines the choice across all categories simultaneously.