From Bullets to Boreholes: A Disaggregated Analysis of Domestic Water Cooperation in Drought-prone Regions

Does water shortage incentivize cooperation? Case studies suggests that water scarcity can rarely, if at all, explain violence, instead such shortages rather facilitate cooperative actions around water. Another major ar-gument from qualitative research holds that water scarcity and armed conflict often occur side by side. These insights have rarely been tested empirically across cases on a sub-national level. Earlier quantitative work instead focused on basin or state level interactions. This article fills these gaps by using disaggregated data to analyze the effect of water scarcity on incidences of domestic water cooperation. Using event data covering the Mediterranean area and Northern Africa (1997–2009), this article first shows that water-related actions, co-operative or conflictual, in general are more frequent in water scarce areas. Second, the analysis demonstrates that water cooperation occurs in areas with difficult access to groundwater and with a history of violence. Third, the findings suggest that the relationship between water scarcity and water cooperation is conditional on levels of democracy. The presented results also differ depending on whether state or non-state actors collaborate in domestic water initiatives. Taken together, these findings provide crucial insights to our understanding of environmental peacebuilding and water security.


Introduction
Water, peace and security are inextricably linked.[…] Without effective management of our water resources, we risk intensified disputes between communities and sectors and even increased tensions among nations.[…] I commend this Security Council meeting for highlighting how water is and should remain a reason for cooperation not conflict.
842,000 people die in low-income countries each year as a result of inadequate access to drinking water and sanitation (WHO, 2014).Millions of people still lack safe drinking water, extreme water shortages are frequently associated with violent uprisings and water therefore remains a top international policy issue.Portraying the severity of water scarcity, researchers and journalists alike choose to focus on civil war dynamics, rioting over water prices, or farmer-herder conflicts (e.g.Abel et al., 2019;Raleigh et al., 2015;Buhaug et al., 2020;Koubi et al., 2016;Wischnath and Buhaug, 2014;Brochmann and Gleditsch, 2012;Vestby, 2019).Positive outcomes of water sharing rarely create headlines even with acts of organized violence remaining rare events.Some water scholars deem international cooperation the more likely result of water disputes (Kramer et al., 2013;Wolf, 1999;Conca, 2001).Relatively few large-N studies (Böhmelt et al., 2014;Bernauer and Böhmelt, 2014) focused on positive water initiatives, and almost none feature sub-national analysis.This article addresses this gap, moreover, by analyzing water scarcity and domestic water cooperation, the study contributes to three strands of literature on water security.Climate-conflict research has rigorously analyzed links between droughts and conflict, yet this literature rarely addresses cooperative outcomes (Koubi, 2019;Mach et al., 2020).While research of hydro-politics provides key insights into water cooperation, these studies too often neglect sub-national dynamics, focusing instead on inter-state relations (Bernauer and Böhmelt, 2020).Lastly, environmental peacebuilding scholars have frequently studied cooperative processes between different actors and scales, however, this literature rarely provides cross-case evidence (Ide, 2018).This study brings together these three literatures by providing a more fine grained picture of water cooperation and conflict.This article also addresses another issue within conflict research, namely its focus on conflict-ridden regions.Such samples can induce selection bias (Adams et al., 2018;Hendrix and Poinsatte, 2019).By comparing georeferenced data for Northern Africa, Southern Europe, and other Mediterranean countries (1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009), the study incorporates conflict regions as well as those less affected by violence.The results presented in this article hold both across all available data points and for a sample excluding Europe.
This article empirically demonstrates that water-related interactions -violent, neutral, or positive -are more frequent in water scarce areas.Furthermore, water scarce areas also see more cooperative engagement over water, particular where there is more difficult access to groundwater.This relationship is partially mitigated by democratic institutions.Lastly, areas that have witnessed armed conflict are more likely to see water cooperation involving state actors.
Most water cooperation involves a multitude of actors which deal with water issues at different levels of governance.Unless specified otherwise, I refer to water cooperation as actions that improve water quality or quantity.Such actions can be unilateral, but here I assume interactions to involve at least two actors.Since I focus on domestic water cooperation, initiatives between states are excluded from the main analysis.Analyzing different sets of outcome variables, this article juxtaposes cooperation between non-state actors with cooperation occurring between state and non-state actors.Separately analyzing nonstate actors allows to ascertain whether water scarcity -be it droughts or low groundwater -have an effect on water cooperation beyond governmental control.Thus, the results provide insights to a more local understanding of hydro-diplomacy which has become an important policy tool (Grech-Madin et al., 2018;Klimes et al., 2019).
This article is structured as follows.The subsequent section provides a cursory review of the literature on water vis-á-vis conflict and cooperation.The third section outlines the theoretical arguments, which is followed by the overview on the data and methodology.The main findings are presented in the Results and Discussion section.A brief summary and an outlook are provided in the final section.

Water: conflict and cooperation
Links between water, conflict, and cooperation have been studied at the interstate, intrastate and local levels.There is ample empirical evidence that water scarcity neither explains wars between states, nor the onset of civil wars.Yet, researchers continue to study possible links between environmental factors and the dynamics of armed conflict.Such work often focuses on water by analyzing possible connections between violence and weather-driven phenomena such as meteorological drought, flooding, or more general variations in rainfall.
Considerable agreement exists about the potential of water scarcity as a conditioning or amplifying factor when explaining fighting between groups in ongoing conflicts (e.g.O'Loughlin et al., 2014;Ide, 2015;Raleigh et al., 2015;von Uexkull et al., 2016;Koubi, 2019;Döring, 2020).However, the research on violence and water can only provide a limited foundation for our understanding on the peaceful interactions around natural resources.Studying peaceful interactions over water within countries, this article addresses a gap that points to broader trends within the environmental research agenda.
For a long time, researchers have faced challenges when explaining peace beyond the absence of violence (Galtung, 1969) and systematic literature on outcomes of peace still remains piecemeal compared to the study on the causes of war (Höglund and Söderberg Kovacs, 2010).This is surprising regarding water security because peacefully sharing water is far more common than fighting over it.Water sharing has the potential to decrease mistrust between actors in post-conflict settings (Swain, 2016).Related research also finds cooperation in form of peace talks to be more common after disasters (Kreutz, 2012;Walch, 2014).Yet, studies specifically dealing with water security overwhelmingly focuses on interstate relations.
Early quantitative studies found water scarcity to increase armed conflict (Hauge and Ellingsen, 1998;Toset et al., 2000).However, such findings were not replicated by later research.Instead, collaboration is more likely, for example between countries in river basins with established treaties on water allocation or where institutional solution are stipulated (Dinar et al., 2015;Ovodenko, 2016;Giordano et al., 2014;Owsiak and Mitchell, 2017).A large body of research sought to identify those river basins most prone to conflict, and how institutions mitigate water disputes (De Stefano et al., 2017;Yoffe et al., 2004;Bernauer and Böhmelt, 2014;Barquet et al., 2014;Swain, 2004;Wolf, 1999;Böhmelt et al., 2014).While water treaties can be conducive to peace, such processes may require third-party support (Ide, 2018a;Krampe, 2017) or enforcement mechanisms within the treaties (Mitchell and Zawahri, 2015;Karreth and Tir, 2018).In addition, democratic institutions take a key role as they prevent dispute escalation via information provisions or through conflict resolution mechanisms (Gizelis and Wooden, 2010;Tir and Stinnett, 2012).These state-based insights contribute even to our understanding of subnational cooperation because governments can mediate local level cooperation.Yet, the interplay between cooperation and conflict can be complex, for example if government collaboration diminishes local-level achievements (Fatch and Swatuk, 2018;Zeitoun and Mirumachi, 2008).This means that water cooperation and conflict can intertwine, thereby making binary understandings of interstate relations less meaningful when incorporating civil society or grass roots organizations.
Clearly, interactions over water are never binary, i.e, completely peaceful or solely violent.While contention over water is frequent, armed conflict over water is rare.Thus, existing patterns of violence and non-violent cooperation may exist at the same time, even within otherwise intense intrastate conflict.For instance, collaborative transboundary ecosystems services in DRC, Rwanda, and Uganda have coexisted alongside severe security concerns (Martin et al., 2011).Ide and Fröhlich (2015) show how local and state actors in Israel and Palestine cooperate over water resources while still not refraining from extreme hostilities.Within the same context, Reynolds (2016) finds cooperative actions are driven only by minorities within conflict parties.This also applies for instigators of violence, who usually constitute a small fraction of groups they claim to represent.The more effective conflict resolution effort should prevail when either side can "win over" those in the more neutral center.Such resolution processes can be facilitated through institutions building.
Seminal work has highlighted how irrigation communities can effectively share water (for examples in the Philippines, or in Spain), but also how such projects fail swiftly without the proper institutional foundation (Ostrom, 1990).Such processes do not necessarily require aid from state institutions.For example, using informal, self-governing irrigation networks that adapt to climate change challenges is key in rural communities (Ostrom, 1990;Berhe et al., 2017).The perhaps most cited examples for violence over water disputes are for the local level, e.g.narratives about communal conflicts and especially in regard to farmer-herder disputes.
However, such stories often tell only half of the story because cooperative, non-violent engagement is far more common in inter-group relations.Gravesen and Kioko (2019) show how such "cooperation" can go as far as joint cattle raids.Research on non-state groups has documented mostly peaceful resource sharing in Burkina Faso (Breusers et al., 1998), Eastern Ethiopia (Bogale and Korf, 2009;Beyene, 2014), Kenya (Koehler et al., 2018;Schrijver and Lenkaina, 2017), Niger (Turner et al., 2012) or between Bedouins and Israeli settlers (Tubi and Feitelson, 2016).Such resource sharing can even reach beyond highly securitized borders as witnessed through "people to people diplomacy" across the borders in the Sudans (Abdalla, 2013).A recent summary of resource-based conflicts also finds conflict resolution practices are far more common in communal disputes across Sub-Saharan Africa (Seter et al., 2016).
In-depth studies have provided detailed descriptions and valuable understanding of the different levels of water cooperation or waterrelated land rights.Especially the emerging field of environmental peacebuilding has shifted focus to positive outcomes of natural scarcity (Conca and Dabelko, 2003;Dresse et al., 2018;Ide and Detges, 2018;Ide, 2018).While there is a great diversity of case studies that describe cooperative actions in water scarce areas from different angles, there are still fewer insights from quantitative research on the same topic.In part, this can be explained by the lack of high-quality environmental data which has only started to become more accessible.Earlier research almost entirely relied on "national-aggregate data, which maske[d] internal inequalities and subnational processes" (Conca, 2002, p.7;c. Gleditsch, 1998).Thus, more cross-case analyses are needed to connect local water issues highlighted by case studies.This study contributes to closing this gap, bringing together qualitative and quantitative insights on water cooperation.

Theoretical considerations
This articles analyzes incidences of domestic water cooperation in water scarce regions.Water cooperation is understood as any action enhancing water quality or quantity, and which results from an interaction between two actors.These actors can either be a government of a state, or a non-state actor.Water cooperation inherently means at least some improvement of the status quo in terms of water provision.Here it is important to consider different sources of water scarcity.
Water scarcity more generally refers to the physical lack of water resources in a specific geographic area, but scarcity should also always be understood through the perspectives of the user.On the one hand, water scarcity is the direct result of human incompatibilities over the quantity, the quality and (or) the control of water (Swain, 2012).On the other hand, scarcities can also be expressed through uncertainty, for example concerns over the timing of precipitation, or market prices (Ostrom, 1990).Although shortages can be the result of natural phenomena such as geological or climatic processes (e.g.erosion, precipitation, or sea water intrusion), only human activity ultimately displays apparent scarcities.Furthermore, agriculture, manufacturing, or urbanization can accelerate erosion, pollution, or other processes that lead to water scarcity.
Existing norms and practices on water use are other factors that explain water scarcity.This means water availability is based on subjective restraints evident through the existing norms in a region (Homer-Dixon, 1994).Such norms can shape over longer time periods and often reflect adaption processes specific to the existing geography.Responses to water scarcity appear first through individual coping strategies while medium and long term adaption processes are often ascribed to a higher group level.Engaging with other communities is partially explained through the expected value of future interactions, which is amplified by increasing the number of potential cooperation partners (Fearon and Laitin, 1996).This suggests that, over time, groups should be inclined to adopt to changes and cooperate more often with others.
Yet, there are physical limits to such processes.Adaption to water scarcity is negatively affected by increased water demands through demographic development and rising living standards.Interaction over water issues are generally more frequent due increased salience of such issues in regions where water is scarce (Hensel et al., 2008).As a consequence, communities, corporations, or authorities start to engage over water issues with explicit interventions.Such actions can be positive, i.e., they might result in the construction of water infrastructure or solutions for water sharing.Water scarcity can also lead to interactions that are less positive in nature, like the illegal appropriation or diversion of water resources.The latter might lead to violent disputes between communities.Whether positive or negative, areas suffering from water scarcity should generally see more activities around water issues compared to those without water problems.This leads to the first hypothesis: H(1): Water-related interactions (cooperative, neutral, or violent) are more likely to occur in water scarce areas.
Until now, the theoretical argument was more general on any type of water interactions.I now shift the focus on water cooperation.Further specifying the theoretical argument, I posit that one specific part of interaction, namely cooperation over water is more likely where such resources are scarce.This should especially be the case in postconflict settings.One cause of water scarcity is in fact armed conflict.Examples in South Sudan, Nigeria, or Syria show that armed conflict frequently destroys infrastructures for agricultural, industrial, or domestic use (UN Water, 2019;UNEP, 2010).Recently, peacebuilding efforts have increasingly incorporated improvements for the water sector.While it is acknowledged that such peacebuilding requires a multi-actor approach, governments remain the first entry point for postconflict water projects (Bruch et al., 2020).Even when armed conflict persists, people can come together over water issues.For instance, at the height of tensions in Cyprus, the sewage system in the city of Nicosia was established with help from both sides of the UN buffer zone.Whether or not violence was initiated or influenced by water shortages, scarcities create incentives to improve water conditions.Such improvement can take the form of physical structures, such as wells, irrigation canals or treatment facilities; this can also include water agreements as well as the creation of institutions.
As the discussion above shows, several pathways linking scarcity to conflict are similar to those explaining cooperation.This is not necessarily puzzling.While the theoretical argument assumes cooperation to be the more likely result of experienced scarcity, violence might still occur over shared resources.Yet, for states and non-state actors, violent altercations are costly and will not benefit resource allocation or sharing in the long-term.Communities should thus have a particular interest in cooperation after incidences of violence.This also dovetails with Ostrom (1990Ostrom ( , 2002) ) who argues that repeated interactions, even if conflictual, will eventually lead to more cooperation over shared resources.
Moreover, cooperation over water could serve as a reconciliation mechanism, whereby positive actions help to overcome previous wrongdoing.Here one should note that organized violence can severely impair or even destroy infrastructures, especially within sanitation, irrigation or other water sectors.Rebuilding and possibly cooperating over resources are more often urgently required after fighting.In sum, incidences of water cooperation should be more visible following organized violence: H(2): Water cooperation is more likely to occur in areas that experienced armed conflict.
Existing norms on water usage are particularly important in water scarce areas where communities have the ability to regulate water intake.Such norms facilitate adaption processes to new environmental conditions.An important feature of human nature involves to adapt to emerging challenges, for instance climate change (Eriksen et al., 2015).Many pastoralist groups for example develop norm-based grazing rights (Schrijver and Lenkaina, 2017) specifying how existing water resources are being used (Foster, 2017).Such norms can also be summarized as institutions, even if they are less formalized.This also alludes to the insight that institutions can take many forms, but most importantly they define specific rules that regulate decision making, procedures, and payoff structures (Ostrom, 1986).A way of such regulation is actively rationing water quantity and penalizing over-exploitation.Once enough people follow norms on sharing and rationing water, it is more likely that free-riding will be sanctioned by the norm-following majority (Schlüter et al., 2016), especially in more homogeneous groups (Fearon and Laitin, 1996).This also means that with lower levels of norm-followers, sanctioning is not effective, ultimately aggravating water scarcity (Schlüter et al., 2016).This is even visible with unequal access to resources, for instance where a more wealthy minority controls a significant portion of the available water (Ostrom, 1990).While the specific underlying norms may vary, the outcomes for communities in water scarce areas should be similar.Actors in such regions should be more apt at solving shortages, which should make cooperation over water more likely:

H(3): Water cooperation is more likely to occur in water scarce areas.
Conflictual outcomes can also be prevented if the involved actors agree on third parties to mediate or promote cooperation.While such actors can be states, their responses vary across countries.Less democratic state institutions are often less involved, leading to more autonomy and self-ruling.Thus, non-state groups need to cooperate more often where democratic governance is absent.
On average, democracies are more likely to respond to water shortages due to a higher interest in responding to crises and gradually establishing institutions.Democracies are also often countries that have seen more economic growth and development (Acemoglu et al., 2015), hence those countries are more likely to have water-regulating institutions in place.Natural disasters such as drought might question the capability of people in power and democratic states are more likely to provide sufficient aid to their citizens (de Mesquita and Smith, 2009;Kahn, 2005).While all leaders are generally interested in minimizing political threats to their office (de Mesquita and Smith, 2009), democratic regimes are more wary as their stability is under higher threat.As constituents scrutinize the state's ability to manage water scarcities when they occur, they can indirectly or directly influence the actions of elected politicians.The latter is not necessarily the case for regimes built on low levels of democracy.Even if voters might know that political leaders are not responsible for a drought or a flood, they will still punish politicians for merely presiding over such instances (Achen and Bartels, 2016).Furthermore, free press and media are important factors that contribute to better informed citizens and higher accountability in democracies (Kahn, 2005).This gives democratically elected leaders clear incentives to support institutions which can prevent and sufficiently react to disasters (Quiroz Flores and Smith, 2013).
The crucial flip-side of this argument assumes communities to become more independent in places with weak democratic structures.While communities in participatory democracies can sway decisions to more cooperative outcomes, less democratic countries rely more on governing elites.Autocratic regimes also have a smaller interest in providing public goods (Olson, 1993) or they allocate services only to specific groups and regions.For example ethnic favoritism explains infrastructural support to those areas aligned with the regime (Burgess et al., 2015).Even when conflicts arise, certain government interventions discriminate by political interests (Elfversson, 2015).This can lead to inefficient interventions or might even discourage governments from intervening at all.Ostrom (1990) for instance provides several examples of local, non-state resource sharing institutions that emerged because they were located far away from government control.Similar findings show that communities with less government support are more likely to adapt to climate change (Paul et al., 2016).Again, this does not mean issues always become cooperative, though as Ostrom (1990) argues, cooperative solutions more often prevail as they are more sustainable.Hence, adaption to water scarcity would involve water cooperation among non-state actors.
In sum, one would expect very democratic countries to have functioning institutions which address water issues before they become salient.Thus, the effect of water scarcity on cooperation would not change.However, for countries with low levels of democracy, non-state actors (local networks, organizations, but also NGOs) are expected to be more active, filling voids left by the state.Thus, the last hypothesis states: H(4): The effect of water scarcity on non-state water cooperation decreases with higher levels of democracy.
Several mechanisms resemble latent processes which cannot be directly measured on a disaggregated scale across a large geographic sample.The following section lays out the research design and explains how different measures of water scarcity provide proxies that allow for empirical testing.

Material and methods
This article analyzes how water scarcity affects domestic water cooperation.As a unit of comparison, I rely on grid-cell years.The chosen unit of analysis allows direct comparison to other studies that use a similar approach (e.g.Landis et al., 2017;Daskin and Pringle, 2018;van Weezel, 2020).The grid-cell observations are derived from PRIO-GRID, a global .5 by .5 decimal degree geographic fishnet (Tollefsen et al., 2012;Tollefsen et al., 2015).This means the measurement units are about 55 km 2 measured at the Equator.The Supplementary material shows results of robustness tests with larger grids, which demonstrates that the findings are not only due to the level of aggregation and the applied modifiable areal units.The selected level of analysis has the advantage to study units within countries or river basins.Grids were chosen over political divisions because cell boundaries are constant; moreover, the units allow the use of hydrology data at much better resolution than country variables.
All main models were estimated using a two-level mixed-effect logistic estimation and with clustered standard errors for the respective grid cell countries.Mixed effects models allow to account for model intra-cluster correlation, thereby using grid cells as nested clusters of random effects to account for unobserved variation.The number of integration points was set to 10.The main findings are robust to different model specifications and the use of other regression models (see Supplementary material).

Outcome variables
I first name the three outcome variables before explaining the coding and data sources in more detail.For a brief overview please refer to Table 1 and Table 2. To properly account for the different scope conditions for water-related interactions, the analysis builds on three different outcome variables, namely water-related incidence, water cooperation (non-state), and water cooperation (state based).This is important.Merely lumping together all water-related events into one variable would mask crucial differences in how water cooperation takes place between different actors.For all variables, events at the government level are excluded.An example of such an omitted event is for The use of "x" denotes inclusion for the respective variable.
Note that for the last row, non-state actors are only included if they are the target of state actors' actions.
example the signing of interstate water agreements or actions between the United Nations and the government of a state.Such events are outside of the scope of this article.The underlying data for the outcomes variables comes from the WARICC dataset (Bernauer et al., 2012) which includes water-related events for over 30 states bordering the Mediterranean Sea as well as the Sahel region between 1997 and 2009.Thus, the analysis is restricted to this time range.While geographic coverage does not fully overlap with previous studies on conflict, it allows for comparisons of conflict-prone regions with areas that have been more peaceful.A detailed description of the WARICC data and the coding procedures can be found in Bernauer et al. (2012). 1  For the first hypothesis, the research question relates to whether or not water scarce areas generally witness more activities over domestic water issues.Thus, the analysis requires a more inclusive definition that comprises different activities around water.Using WARICC data, I can take advantage of the coding scale which ranks events from conflictual to cooperative.The outcome variable water-related incidences takes the value "1" for all grid-years with at least one water-related event.These can be either cooperative, neutral or conflictual.For grid cell-years with no such event, the variable is coded "0".The variable water-related incidences thus includes domestic events initiated by both government and non-state actors.
For Hypotheses 2-4, the inquiry shifts to solely include cooperative water events.This necessitates to further disaggregate by the involved actors.The outcome variable water cooperation (non-state) is binary; the value "1" represents all grid cell-years with at least one cooperative water event, or with "0" respectively for non-cooperative grid years.
Here only cooperative incidences between non-state actors are included.The coding for water cooperation (state based) follows the same setup; here actions initiated by state actors (the central government, or sub-national authorities) are included whereas actions initiated by nonstate actors are excluded.An important difference is that the outcome variable water cooperation (non-state) only includes events between nonstate actors, while water cooperation (state) can still include incidences where the government (or another sub-national authority) initiates cooperation with non-state actors on water projects.
These definitions suggest that at least two sides are required for cooperation.For the water cooperation (non-state) outcome measure, only events are included that were initiated by non-state actors such as grassroots organizations, firms or individuals, but even interventions by national or international NGOs are included as long as they do not target cooperation with the government.For water cooperation (nonstate), some of the included events are e.g. the building of small dams by villagers to preserve water during the rainy season in Darfur, or an initiative by a local company improving the water supply in Southern Morocco.State-based water cooperation includes events such as the repair of an irrigation canal in Somalia, or the inauguration of a dam in Algeria.These events can appear as more unilateral, but here it is assumed that general public acts as implicit recipient of government actions. 2  There are some limitations with WARICCs original geo-coding and with potential reporting biases.The following paragraphs outline these issues and explain how they were addressed.First, in the original structure, events without a clear location were assigned to the capital of the state.Many actions are also assigned to the state capital for less peculiar reasons.When governments announce actions on water-related issues, for example increasing spending on pipelines or implementing new quality standards, the event occurs more often in the capital or its vicinity.This applies also to various government or regional agencies as well as NGOs which often have their offices in capitals.To address this circumstance, all regression models include a control for capitals.This also relates to an idiosyncrasy that can be observed in the data (see for example Fig. 1), namely the high amount of observations in areas that were part of former Yugoslavia.The breakup of the country led to the establishment of several national water governing agencies which dealt with the modernization and rebuilding of the existing water infrastructure.Many of these areas also saw increased privatization of the water sector (c.Krampe, 2016).A dummy variable for areas from the former Yugoslavia accounts for this clustering.
Another problem arises through potential reporting bias.Research demonstrated how reporting or coding procedures impinge on the data quality, especially for conflict events (Croicu and Kreutz, 2017;Eck, 2012;Gleditsch et al., 2014;Otto, 2013;Weidmann, 2016).Some of these issues are transferable to the study of water cooperation because hydro-politics often include issues sensitive to governments and their partners (Grech-Madin et al., 2018).Ultimately, almost all event data builds at least in part on journalistic reports.Whereas well-functioning institutions in Spain or Greece will not generate a lot of news when they mitigate water problems, regimes in for example Ethiopia or Egypt will make sure that even relatively small policy actions on water are put in the limelight.The latter can also be said about international or local aid organizations that sometimes only operate in a few key countries.To counter problems with state censoring, all regression models in the analysis include a measure on media bias.
Furthermore, other research argues that WARICC might "be more useful in tracing variations […] over time by country […]" (Bernauer et al., 2012, p. 541) because water events are less frequently featured in news.Being more hesitant about the quantity and magnitude of events, focusing on incidences of cooperation resembles a more appropriate measure.Explaining the cooperation incidences also seems more 1 The focus of this study are domestic events describing water cooperation.
The original event data is coded on a 11-step scale ranging from very cooperative to very conflicting actions.The scale in the WARICC data ranges from most conflictive (−5) to most cooperative (5); neutral events are coded as (0).A very cooperative event refers to actions by an actor (government, international organizations, firms, etc.) "trying to initiate or implement policies, programs, or actions that substantially improve the quality or quantity of water".Robustness test with a multinomial model using a simplified scale (conflictive, neutral, cooperative) show similar results.Very conflictive actions are those with "a strongly negative impact on the water quality/quantity of a country, for instance overt violence precipitated by governments, groups, institutions, or individuals in connection with water resources." 2The WARICC data contains events coded as unilateral.For state-based cooperation, this can include the building of a new dam or the repair of a treatment plant.While I assume constituents and general populations to be (footnote continued) recipients of such actions, it is crucial to distinguish unilateral water events for the non-state outcome variable.For the latter, I coded actions as unilateral if for instance local NGOs start one-sided interventions or if organizations provide humanitarian support as part of disaster relief.I identified 97 such events among the non-state cooperation entries.The robustness tests show that the results remain the same when excluding observations with these events.I thank Reviewer 1 for pointing this out.
reliable than trying to model the water actions on an 11-point scale for the same unit of analysis.Improvement in the water sector can vary meaningfully between regions.An area that is very rural might improve their water supplies significantly by the drilling of a few boreholes whereas such action would not necessarily enhance the water supplies in a more urban area.Thus, coding cooperative incidences also seems more appropriate because it creates no comparison in relation the scale of water quality improvement across regions.

Explanatory variables: water scarcity, armed conflict and democracy
To study the effect of different types of water scarcity, all models include measures for groundwater depth, drought, and precipitation.In addition, models accounting for surface water are found in the Supplementary material.To account for the prevalence of drought, I use the Standardized Precipitation and Evapotranspiration Index, SPEI-3, from the SPEIbase (Beguería et al., 2010) which allows to ascertain the drought characteristics of a given grid cell during the respective rainy season (Schneider et al., 2015) compare. 3The binary variable drought takes the value "1" if SPEI values were equal or smaller than −1.5 and 0 if otherwise.
The presence of surface water is tested through robustness checks, showing no diverging findings.This variable is constructed by overlaying shape-files for the reservoir extent of lakes, rivers, and dams as well as single drainage-lines for lakes and rivers.This data comes originally from World Data Bank 2 and was downloaded from Natural Earth in 1:10 m scale.Extent data is version 4.0.0 and line data version 3.0.0,respectively.These include only major, perennial water bodies which account for stable sources of surface water only.If a grid cell contains such a body of water it is coded as a "1", or otherwise as a "0".Enabling a comparison to earlier studies, the analysis also includes measures for rainfall.The variable precipitation (ln) gives the annual sum of each cell's precipitation in millimeter, which was log-transformed.The data is based on monthly meteorological observations through the Global Precipitation Climatology Centre (Schneider et al., 2015).Furthermore, all models control for temperature to avoid biases that might occur from solely relying on rainfall estimates (Auffhammer et al., 2013).The variable temperature reflects an area's yearly average measured on the Celsius scale (Fan and van den Dool, 2008).
To measure water scarcity in terms of groundwater, I use the average depth in meters (logged) to the groundwater layer within a grid cell for a given year.The data has been generated and validated by de Graaf et al. ( 2015) (see also de Graaf et al., 2019).Here it is assumed that access to groundwater will be more common in areas with a lower depth to the groundwater table.The groundwater measure also allows to proxy for non-perennial streams or other surface water bodies because values for the variable will be lower in cells with lakes, rivers, or smaller streams.
It is argued that more cooperative action should follow incidences of previous violence.Thus, the model includes a dummy variable that takes the value "1" if there had been at least one violent conflict event in a grid cell during the previous year, or "0" if otherwise.This data comes from UCDP Geo-referenced Event Dataset (UCDP GED) (Sundberg and Melander, 2013;Croicu and Sundberg, 2016).An armed conflict event is defined as an incident where armed force was used by an organized actor against another organized actor, or against civilians, resulting in at least 1 direct death at a specific location and a specific date (Croicu and Sundberg, 2016). 3The mean value for SPEI is 0, and the standard deviation is 1.SPEI values are spatially and temporally comparable.For each month, the index reflects the deviation from long-term normal rainfall during the three preceding months.The SPEI estimation technique based on the Penman-Montheith method is considered better than the frequently used Thornthwaite estimation (Tollefsen et al., 2015).
The proposed theory considers active participation in democracy as a crucial conditional variable.To measure democracy I rely on V-Dem's participatory democracy index which aims to reflect the participatory principles of a democracy.The variable is measured on an interval scale with higher values representing higher emphasis on active participation by citizens in all political processes (Coppedge et al., 2018;Pemstein et al., 2018).This index also takes into account the level of electoral democracy as well as engagement of civil society organizations (ibid.).
The maps in Figs. 1 and 2 provide a quick overview of the area under study as well as the spatial variation of the outcome variable.For the map in Fig. 2, cooperative water events from WARICC (Bernauer et al., 2012) are overlaid with data on armed conflicts for the analyzed time period.While such figures cannot easily display temporal variation, it is clear that many grid cells both exhibit a history of violence, and also see incidences of water cooperation.Similarly, Fig. 1 shows water-related events with data on the depth to the available groundwater.Many water-related interactions do seem to cluster in areas with more difficult access to groundwater.However, there are also a considerable amount of events located in areas which do not require deep drilling.In this way the maps show that water scarcity (or previous armed conflict) alone cannot explain water cooperation.Other factors are key intermediaries which ought to be accounted for as controlling factors.

Control variables
To account for possible spurious relationships all models include a set of control variables.The availability of water resources is changed by human activity.For example, groundwater exploitation through industry and urban users alters water tables.Dense urban areas also put more pressure on available water resources and water-cooperative events should therefore be more frequent in more populated places.Therefore all models account for population density (Goldewijk et al., 2011;Goldewijk et al., 2010).As mentioned above, it is crucial to control for national capitals and consequently all iterations include a dummy variable controlling for whether a country's capital lies within a grid cell.Responses from NGOs, civil society or other local actors can be impacted if media misrepresents water-related interventions as well as water availability.To control for media bias, I rely on V-Dem data which accounts for impartiality of media coverage towards the government and political candidates (Coppedge et al., 2018).Every model is tested with a binary variable for those grid cells that lie in the former Yugoslavia in order to show that the results are not contingent on the aforementioned clustering.WARICC data also includes coding on international presence which describes whether or not an event was impacted by a third-party actor from outside of the country.This could be financial or technical aid.It is important to control for such events as international actors can influence both water scarcity as well as water cooperation.
Level of development and local wealth are predictors for the capacity to address environmental scarcity.Variation in economic activity also explains both water use and the need to cooperate over scarce resources.All models account for calibrated night-time light emissions based on Elvidge et al. (2014). 4Night-time light emissions can be useful as a proxy for state capacity (Koren and Sarbahi, 2018) and local wealth (Weidmann and Schutte, 2017); both affect the proposed relationship.The Supplementary material provides alternative models with Nordhaus data for gross cell product in US dollars using purchasing-power-Fig. 2. Cooperative water events and areas with a history of armed conflict.Data from Bernauer et al. (2012) and the Uppsala Conflict Data Program (Sundberg and Melander (2013)). 4Image and data processing by NOAAs National Geophysical Data Center.DMSP data collected by US Air Force Weather Agency.Values are standardized to be between 0 and 1, where 1 is the highest value in the time-series, and 0 is the lowest.Previous research suggest that calibrated values can overcome issues inter-satellite differences and issues with inter-annual sensor decay (Tollefsen et al., 2015).parity (Nordhaus, 2006).With the exception of the media bias variable, all control variables were obtained through PRIO-GRID (Tollefsen et al., 2015;Tollefsen et al., 2012).
Lastly, the analysis relies on spatial and temporal controls.To account for serial auto-correlation, the model applies Carter-Signorino polynomials of spells without the respective outcome of interest (Carter and Signorino, 2010;Beck et al., 1998).Furthermore, possible spatial dependence is captured by separately controlling for whether armed conflict events or the respective water incidence occurred in a neighboring grid cell (roughly equivalent to radius of 150 km 2 ).To address endogeneity all right-hand variables are lagged by one year.

Results and discussion
This section discusses the findings on the linkages between water scarcity and domestic water cooperation.Odds ratios are reported for the possible direct links, whereas analyses for conditional relationships are displayed as marginal effects plots.The latter is particularly warranted for models with interactive terms (Brambor et al., 2006;Braumoeller, 2004).Complete tables for all models and robustness checks can be found in the Supplementary material.

Water-related interaction
I first report results from models analyzing all water-related interactions (Hypothesis 1).This outcome variables measures any domestic water-related activity, i.e., it includes events coded as conflictual, neutral or cooperative.Fig. 3 displays the odds ratios for the three main variables of interest.The coefficients show that increasing depth to groundwater increases the probability of water-related incidences while controlling for other factors.A one-unit increase of the groundwater variable on average increases the odds of water-related incidence by about 46%.This translates to for instance increasing depth to groundwater from 30 m to about 55 m.The effect is even more pronounced in the sample that exclude areas in Europe.In all models, this relationship is statistically significant for the .99confidence levels.
There is however a different picture for the measures on precipitation and drought.The results for both variables run counter to the proposed hypothesis, but the estimates are both very close to zero, suggesting no relationship to the outcome variable.In addition, neither coefficients are statistically significant at any conventional level.Thus only when measured with groundwater accessibility, there is support for the first hypothesis.These results show that it is important to distinguish between different sources of scarcity to obtain a more nuanced picture on where water interaction takes place.The results also provide a foundation showing that water scarcity enables interaction over water.This exemplifies that water scarce regions lead to responses over water issues.The analyses in the next subsection further disaggregate these interactions by focusing on different types of domestic water cooperation.

Water cooperation initiated by state actors
Now turning to the results for incidences of water cooperation actuated by the state (see Table 3 and Fig. 4).Again this variable captures Fig. 3. Incidences of Water-related action (1997Water-related action ( -2009)).This figure shows the exponentiated coefficients (odds ratios) with .95confidence intervals for the full sample and for a sample excluding all European countries.The outcome variable is incidence of water-related actions (cooperative, neutral or conflictual).The complete table can be found in the Supplementary material.
only incidences of domestic water cooperation initiated by the state, but such events can occur between the government and non-state actors (e.g.grassroots organizations, firms, or individuals).It also subsumes cooperation between the state and sub-national authorities, but it excludes cooperation that is initiated by non-state actors.The focus is on variables that explain state-based actions on water issues.
Interpreting the estimates of the armed conflict variables, I find support for Hypothesis 2 which held that areas with a history of violence are more likely to see water cooperation.Having experienced violence in the previous year increases the odds of water cooperation by over 60%, while controlling for other variables in the model.This relationship is statistically significant across all models and employed samples.Furthermore, I find some evidence that armed conflict in neighboring areas increases the probability of water cooperation, even while controlling for previous violence and other variables.A counterargument could be that areas affected by armed conflict simply receive more foreign aid in the following year.However, events initiated by international NGOs are not included for this outcome variable.It is still possible that such aid might be channelled through governments.I also conducted robustness test accounting for the presence of international actors which does not change the results.State agencies surely have an interest in restoring vital infrastructure following events of organized violence and the results suggest that water is central to such efforts.These findings on armed conflict and water cooperation also align well with insights from case studies which have long suggested that cooperation and conflict more than often coexist.
The results on the proposed links with water scarcity reveal a similar picture as the estimates for more general water-related interactions.The coefficients for drought and precipitation are close to zero and also here do not yield any statistical significance.On the other hand, the estimate for groundwater indicates that areas with deeper groundwater are more likely to see water cooperation.This relationship is also statistically significant for the .99confidence level.Thus, with lower groundwater levels, the probability of state-based water cooperation increases, holding everything else equal.

Water cooperation between non-state actors
This sub-section considers the results on water cooperation between non-state actors.The main coefficients are displayed graphically (Fig. 5) as well as in Table 4. Full tables are found in the supplementary material.For this sub-analysis the outcome variable only includes cooperative incidences involving non-state actors.Such cooperation between non-actors includes actions over water resources between grassroots organizations, local residents or NGOs.
For this sub-analysis, I do not find support for the second hypothesis, i.e., areas which had experienced incidences of armed conflict are not more likely to witness water cooperation between non-state actors.This is an interesting non-finding considering that the same hypothesis is supported for cooperation initiated by state actors.The finding also highlights the importance to differentiate between levels of organization.An explanation can be that government actors are more likely (than non-state actors) to respond to water issues by committing additional resources.Another important factor can be trust.In post-conflict settings, decision making processes are altered due to the trauma and other wartime experience.This relates both to trust among groups which were involved in armed group, but also more generally towards other actors.Lastly, non-state actors often also lack capabilities to reshape or rebuild water infrastructure after armed conflict, whereas governments on average should have more response capacities and state agencies often also receive more help from outside.
For the water scarcity variables, I find results similar to the previous analyses.The coefficients show that deeper groundwater increases the probability of non-state water cooperation when controlling for other factors.Here the effect of groundwater seems to be even more pronounced when compared to the results from state-based cooperation or general water-related interaction.Increasing depth to groundwater by one unit on average increases the odds of water cooperation by about 62%.Again, this can be understood as lowering available groundwater from 30 m to about 55 m depth.In all models, this relationship is statistically significant for the 99.9% level.These results suggest that groundwater is even more important in areas where government actions are less visible.In such areas, citizens become more involved or other organization fill voids left by missing infrastructure.
Neither precipitation or drought estimates support the proposed hypothesis in this analysis on non-state water cooperation.The The outcome variable here is incidence of cooperative events between non-state actors.See Table 3 for the full list of estimates.respective coefficients are not statistically significant at any conventional level.Thus only when measured with groundwater accessibility, there is support for the first hypothesis.Some alternative arguments should be addressed when comparing the three main indicators.First, drought might not lead to further water cooperation because there are many other possible responses to drought (or lower rainfall) which are not captured through the data generating process in this analysis.Such responses can be temporary or longer term resettlement, or shifting to other sources of water allocation.Another factor is the timing of drought.The effect of drought could be more visible when comparing areas over a longer period of time.It is also possible that annual observations mask the actual temporal extant of droughts.

Conditional effects
Lastly, I am turning to the fourth hypothesis which considers the conditional effect of water scarcity over different values of participatory democracy.Figs.6a and 6b show the effect of groundwater depth on non-state water cooperation over different values of participatory democracy.In Hypothesis 4, it was theorized that the effect of water scarcity on non-state water cooperation decreases with higher levels of democracy.The results lend support to this argument.With higher levels of participatory democracy, the effect of groundwater on water cooperation decreases.This relationship is statistically significant at the .95level for participatory democracy values between 0.1 and about 0.34 for the full sample (and up to approximately 0.2 for the sample excluding Europe).After these thresholds the direction of relationship still holds, but I can no longer reject the null hypothesis.Nonetheless, for both samples about 75% of all observations are actually within the thresholds of statistical significance.It is also not surprising that the effect of water scarcity on water cooperation will level off after specific thresholds.Upon reaching a certain level, democracy does not impact the relationship any longer because areas or countries with higher levels of democracy will have institutions or mechanisms that mitigate problems with water scarcity.This findings provide some evidence that non-state groups in less democratic countries become more independent, cooperating over water issues without the central government.

Other findings and additional testing
This subsection briefly discusses relevant control variables and provides an overview of additional testing.Including other control variables does not change the findings and most control variables show the expected relationship patterns.The presence of surface water significantly increases the probability of water cooperation or water-related events, everything else equal.However, the presence of large bodies of surface water does not change over time, making it less suitable for panel analyses.It is also not surprising to find areas close to rivers or lakes to show more water-related interaction.
Population density and proximity to the nation's capital are robust predictors for water-related actions or water cooperation across all models and outcome variable specifications.This is in line with the expectation that more populated areas would also see a higher likelihood of human interaction.The dummy variable for areas in the former Yugoslavia is also a strong predictor for water-related events.Additional robustness checks can be found in the supplementary material.This extra material includes results using models with logit, mixed effect logit without clustered standard errors, and multinomial models, among others.I also conducted tests with grid cells measuring 1 by 1 decimal degree cells showing that the findings are not merely to the applied modifiable areal units.

Conclusion
This article empirically tested the relationship between water scarcity and incidences of water-related interactions, with a specific focus on water cooperation.While many studies focus on water scarcity and violent conflict, this analysis shows water scarce areas are also more likely to witness instances of water cooperation.Everything else equal, more difficult access to groundwater is found to increase the likelihood of water cooperation, both between non-state actors but also between the government and other state or non-state actors.The findings go beyond samples of particularly conflict prone areas as the results cover different samples for Northern Africa and other countries bordering the Mediterranean Sea.The article further shows that the relationship between water scarcity and non-state water cooperation is stronger in areas within less democratic countries.This suggests that in less democratic countries, actors find solutions to water scarcity without help from their central government.These results hold even when taken into account the direct involvement from international actors.The study provides cross case empirical support to arguments of environmental peacebuilding scholars.These findings also contribute to climate-conflict research showing that focusing too much on armed conflict neglects the potential of water cooperation.
The analyses connect to case studies which have long suggested that water scarce regions also have potential for peaceful interaction.A relevant share of areas included in this study have witnessed armed conflict.Yet, the same areas also show incidences of cooperative water events.In fact, I find that areas with armed conflict in the previous year to be more likely to witness incidences of state-initiated water cooperation.The same relationship is however not found for non-state cooperation as an outcome variable.This suggests that the international community ought to pay more attention to non-state actors after armed conflict, in particular in countries with a weaker civil society and fewer active grass-root organizations.This also highlights the potential of strengthening civil society organizations in areas where the willingness of governments might be low.Investing in the water sector can also have several benefits in post-conflict societies.Such support can build trust between communities and with the government.Moreover, a functioning water infrastructure for sanitation, agriculture or industry can be the foundation towards poverty alleviation.
This study can be extended in different ways.A first avenue is to expand the collection of water events data both spatially and temporally.While there are many data providers for different types of violent events, there is a dearth of such data-sets on environmental issues, particularly for cooperation.An extension of water-related events data should also aim to improve the precision of geographic location in order to better match events to climate data.It is vital to gather more data on cooperation.If all we do is study conflict, we overlook cooperation and peace.Further research should thus study more closely the interaction between water cooperation arising after violent events, or whether conflict can also arise after incidences of cooperation.Future studies could consider more hybrid units of observation, for instance by combining data on river basins with environmental events data.Remote sensing in particular provides high resolution information that is comparable globally and still useful for analyses that make use of subnational data.Moreover, such data should be made easily accessible to a variety of users including, grass roots organizations, policy makers, and scholars.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Fig. 4 .
Fig. 4. Incidences of water cooperation, state actors (1997-2009) -This figureshows the exponentiated coefficients (odds ratios) with .95confidence levels for the full sample and for a sample excluding all European countries.The outcome variable is incidence of cooperative events between non-state actors.See Table3for the full list of estimates.

Fig. 5 .
Fig. 5. Incidences of water cooperation, non-state actors (1997-2009) -This figure shows the exponentiated coefficients (odds ratios) with .95confidence intervals for the full sample and for a sample excluding all European countries.The outcome variable here is incidence of cooperative events between non-state actors.See Table3for the full list of estimates.

Fig. 6 .
Fig. 6.Average marginal effects -These figures show the probability of non-state water cooperation incidence as a function of groundwater depth and participatory democracy.The average probability on incidence of water cooperation for different values of depth to groundwater overlaid with the sample distribution of the democracy measure; other variables are held at their means.The shaded area represents the .95confidence interval.