Civil war's impact on the environment and on access to water and sanitation services: the case of Colombia

No more recent versions of the Prio-Uppsala Conflict Datasets are available.

Two different concepts of war are used throughout this paper: war and civil war (also called internal armed conflict, civil conflict, or armed strife). In a civil war, rebel groups fight the army of the state; civil wars are featured by violation of human rights through use of guerrilla war tactics by the rebels and groups with different ideologies. In a civil war, the rebels seek to overthrow the government.

Justino (2009) envisions circumstances in which households may actively participate in a civil war. This paper assumes that households are victimized, since civil wars are devastating for life, health, living standards, and for human and physical capital (Miguel and Blattman 2010).

For full expanded literature review, threats of identification and tables of results, visit Ortiz-Correa and Dinar (2018) available at http://​spp.​ucr.​edu/​publications/​civil_​war_​impact.​pdf.

This paper is about the effect of the Colombian conflict on the access to water and sanitation services and not about the dynamics and historical evolution of the conflict itself. Other papers exploring issues related to civil wars and their effects (Akresh and de Walque 2008; Akrest et al. 2011) provide only an overview of the conflict. An in-depth analysis of the conflict is beyond the scope of this paper.

The utility maximization framework has been used in previous research, such as: child mortality and fertility (Wolfe and Behrman 1982; Ben-Porath 1976; Rosenzweig and Evenson 1977), labor supply and intrahousehold bargaining and behavior (Rosenzweig 1980; Boulier and Rosenzweig 1984; Ravallion and Dearden 1988; Behrman and Deolalikar 1993), and policy evaluation (Rosenzweig and Wolpin 1986).

Water quality level is not directly addressed, since such measure at the state level is not available. Any negative impact of civil war on access to sanitation services, like lack of toilet connection to the sewerage, will reduce the quality of water, by increasing the probability of contact between the excreta and water sources (as defined by WHO).

The DHS is extensively used as a dataset in research on developing countries, even though, in most cases, it is not a panel dataset. The sampling strategy guarantees national representativeness. The concepts of access to water and sanitation services are taken as defined by the DHS for the Colombian waves. There might be a concern about the representativeness of the survey if households could not be surveyed in the most violent places. The results have to be taken as a lower bound of the real effect of civil conflict on the access to water and sanitation.

The Police codes the criminal events per year and per police units. Most of the police units are of state-level jurisdiction. The police units with national jurisdiction were discarded for the purpose of this paper. Data from police units with jurisdiction in the capital city of a state were added to the data of the state in which the city is located. Some police units have jurisdiction in more than one state; these units are usually located in violent regions covering bordering regions of two or more states. Information was gathered from police and military personnel to assign a weight for each of the states that have a share of its territory covered by one of these special police units.

Auxiliary regressions (not reported in this paper but available upon request from the corresponding author) indicate that the education data does not seem to be impacted by any of the measures of the stock violence.

It is entirely possible that some variables that change over time (such as pattern of urbanization, population dynamics, changes in technology, among others) have an effect on both access to water and sanitation, and the conflict intensity. Such a possibility is not ruled out in this paper. On the contrary, it is assumed that such an effect takes place at the state level at a linear fashion, and that is why, state-level specific time trends are included.

Results can be interpreted as casual, as long as the fixed effects remove those variables that do not change over time and that have an impact on conflict intensity and access to water and sanitation at the state level. Fixed effects may aggravate any problem of measurement error, such as measurement error related to conflict intensity. Results have to be interpreted as a lower bound.

Only linear state-specific time trends are considered in this paper. Even though non-linear state-specific time trends may take place, this remains to be explored in future research.

The estimation does not distinguish between direct or indirect target of the violent events, whether the targets are households where the attacks took place or in neighboring places within the state.

The unit of analysis is the household (the household connection to water and sanitation services), but the treatment (the intensity of the conflict) as well as the quality of institutions variables are taken at the State level. Municipality-level figures for conflict intensity are not used due to the high degree of within state migration that might invalidate the identification strategy.

Generally, the R-squared is not very informative in panel data analysis. The analysis in panel data regressions relies on the significance of each individual variable as well as the overall significance of the model. In panel data, the R-square tends to be low due to heterogeneity of the various cross sections included, in our case, the states and the households. As the relevant variables are significant, as well as the overall significance of the regressions, the real problems that can be expected would be specification bias and multicollinearity. A high R-squared would be only relevant if forecasting the access to water and sanitation in the presence of conflict, but this is not the objective of the research in this paper.

Tables with full results are available in Appendix 2 (of Ortiz-Correa and Dinar 2018).

Individual-level regressions (available at Ortiz-Correa and Dinar, 2018) for the incidence of diarrhea and fever in children aged 5 years or younger at the households in the sample also showed contradictory signs on the coefficients of the institutional variables. For instance, GDP per capita growth rate increases the probability of diarrhea in the regressions for extortions, terrorist attacks, kidnappings, and attacks against the police. However, there is an opposite sign and the same indicator reduces the incidence of diarrhea in the regression for mass-murder victims. Although a higher GDP implies that households have more resources to cope with conflict and to offer better food and better health care to their children, it can also be true that conflict causes higher disruptions when households are more economically included. The secondary education students-to-teacher ratios are positively related to the incidence of fever, which suggest that there is a higher exposure to pathogens and contagion as more children go to school and reside in the same classroom.

Since the Colombian conflict has not been initiated or driven by lack or scarcity of water and sanitation services, the estimations do not suffer from reversed causality. For instance, it is not the case that the lack of access to water and sanitation services obliges households to kill, to kidnap or to carry out terrorist attacks.

Regressions (not presented here, but available at Ortiz-Correa and Dinar 2018) using the children aged 5 years or younger of the households in the sample indicate that armed conflict harms the health of children. Two variables related to health are available in the DHS: incidence of fever and incidence of diarrhea. The relationship between these infectious diseases and the conflict indicators seems to depend on the type of conflict indicator and the length of the conflict data considered. For the incidence of diarrhea, the number of the extortions in the year before the survey increases the likelihood of diarrhea by 7.4%. However, when aggregating the number of extortions over 10 years, the probability is reduced by 10%. The terrorist attacks in the year before the survey reduced the probability of diarrhea by 1%. The 10 year averaging of attacks against the police also reduced the probability by 11% and 8%, for 10% and 20% averaging. In relation to the incidence of fever, the aggregation or averaging of the terrorist attacks for 5 and 10 years reports an increase of the fever probability around 2%. The number of mass-murder victims in the year before the survey reduces the probability of fever (close to 3%), but the aggregation for 5 and 10 years consistently increases it up to 19%. Overall, the differences in sign and in magnitudes can be an indication of how households may react to different conflict intensity information and (for instance, adjusting the time for parental supervision) and how they can help their children cope with the post-traumatic stress disorder (Nersisyan 2006).