Especially since the refugee crisis in Europe in 2015, climate change has often been linked to conflict and migration, and different inferential strategies have been proposed to gain a better understanding of these relationships using data (see Hoffmann et al., 2020, for a thorough meta-analysis on the effects of environmental change on migration). The existing literature generally acknowledges five major categories of phenomena acting as determinants of migration flows, which can be summarized as economic, political, social, demographic and environmental drivers (Black et al., 2011; Lee, 1966). Most studies consider these factors simultaneously and do not focus on the pathway through which these drivers affect migration, as well as each other. While there is a strand of research analyzing how climatic change might foster migration and how this could lead to conflicts in countries that are receiving migrants (Reuveny, 2007), there is not a lot of empirical evidence supporting this narrative (Brzoska and Fröhlich, 2016; Raleigh and Urdal, 2007). In this contribution, we advance the literature by assessing empirically the role of violent conflict as a mechanism through which environmental shocks lead to internal migration flows. We do so by concentrating on the case of Colombia, a country that has experienced low-intensity war for a long period of time and for which data on internal mobility have recently become available (Siraj et al., 2020). Until the peace agreement of 2016, Colombia suffered from armed conflict for more than 50 years, which caused over 200,000 causalities. Our analysis thus contributes to the understanding of human mobility in the context of long-term conflict and examines the role played by environmental shocks as a driver of violent episodes leading to internal migration.

Abel et al. (2019) proposed a statistical approach that explicitly assesses the causal linkages surrounding the environment-conflict-migration pathway, and then rigorously tests, whether environmental change fuels the emergence of conflicts, which in turn promote forced migration. The focus of Abel et al. (2019) was on global international refugee flows, and their research showed that, at least for certain regions and points in time, this chain of events plays a significant role in explaining the connection between drought shocks, violent conflicts and human mobility. In particular, this pathway appears to be statistically relevant for the years 2010 to 2012 in Northern and Western Africa, whereas it seems to be less important in other time periods and parts of the world.

While Abel et al. (2019) focused on forced migration across countries, in this contribution we apply their approach to analyze internal migration between different areas within a given country. Our aim is to advance the research on this topic by providing a rigorous empirical quantification of the causal mechanisms linking environmental shocks to internal population mobility via the effect such shocks have on the probability of violent conflict. In particular, we concentrate on the assessment of internal migration patterns in Colombia between 2000 and 2005. For this case study we use municipality-level migration data based on Siraj et al. (2020), and combine them with information on conflicts and battle deaths from the Uppsala Conflict Data Program (UCDP), and with climate data based on the Standardised Precipitation-Evapotranspiration Index (SPEI). In addition, we employ a set of socioeconomic variables from the National Administrative Department of Statistics of Colombia (DANE) and the Integrated Public Use Microdata Series (IPUMS), as well as other controls. To test for a causal link between climatic shocks, conflict and internal migration in Colombia, we apply a gravity model that takes into account the potential lack of exogeneity in both the outbreak of conflict and the existence of non-zero migration flows between individual municipalities. Based on the method proposed by Kim (2006) and also employed by Abel et al. (2019), we have set up a sample selection model composed of three equations that are estimated in two steps. First, we estimate the impact of the climate on the probability of conflict. Second, we analyze how conflicts affect internal migration, while taking into account whether there is a non-zero flow of migrants, as well as the size of the flow. Thus, the analysis goes beyond the identification of partial correlation structures across these three variables (as was partly done for Colombia by Camargo et al., 2020) and instead concentrates on improving our understanding of the causal mechanisms linking environmental shocks to internal migration through changes in the occurrence of conflict.

Our novel results imply the existence of a significant positive effect both of the severity of droughts and the likelihood of conflicts and between conflicts and internal migration, revealing an indirect effect of environmental shocks on internal migration in Colombia. These results contribute to a better understanding of the linkages between climate, conflicts and migration at the subnational level and provide novel empirical evidence for migration and climate policy.

The remainder of this paper is structured as follows. First, we provide an overview of the empirical literature on the connections between climatic shocks, conflicts and migration, and review the existing work on the case of Colombia. Next, we describe the data employed in the analysis and the corresponding sources. We then provide information on our econometric framework and estimation method, and present the main results as well as the findings from our robustness checks. We close the paper with a discussion of our conclusions.

The academic debate concerning the role played by environmental shocks (and, to the extent that it affects their probability of occurrence and their size, by climate change) as a driver of conflict that can, in turn, lead to migration is not new. The discussion has gained momentum in the context of the conflicts leading to the Arab Spring and the start of the civil war in Syria in 2011 (Friedman, 2013). Empirically, this channel has been found to provide a partial explanation for the differences in migration rates between countries and subnational regions. The meta-analysis of the effect of environmental shocks on migration by Hoffmann et al. (2020) concluded that climatic shocks tend to affect human mobility more systematically in countries that have experienced adverse sociopolitical conditions and violent conflict. Similarly, Koubi (2019) reviewed recent scientific publications to examine the consensus on the climate-conflict link, and found that the existing evidence points to environmental change being one of the contributing factors to the emergence of conflicts. Droughts, natural hazards, sea level rise and the melting of glaciers are expected to have the biggest effects on climate change-induced migration in Latin America in the future (Kaenzig and Piguet, 2014).

Notwithstanding the problems associated with the measurement of the variables involved in the pathway leading from environmental change to migration through the channel of an increased likelihood of conflict, theoretical complexities can arise when conceptualizing the linkages between these phenomena. The climate system, natural resources, violent conflict and human mobility interact with each other in ways that make the empirical assessment of causal narratives particularly challenging (see, for instance, the analytical framework provided Scheffran et al., 2012). The complicated interactions arising in the climate-migration link (independently of whether conflict acts as an intermediate mechanism or not) have led to conceptualizations based on the existence of context-specific thresholds that make adaptation responses to climatic shocks inadequate (see the work by McLeman, 2018 or McLeman et al., 2021).

From a theoretical perspective, the vulnerability of populations to climatic extremes can be considered a function of their dependence on natural resources, the sensitivity of these resources to environmental change and the adaptive capacity particular human groups possess (see Barnett and Adger, 2007 or Von Uexkull et al., 2016). Adaptive capacity, in turn, is determined by socioeconomic development (which expands the set of potential income-generating activities) and past conflict experiences. Thus populations who are dependent on natural capital are more sensitive to climatic shocks that may affect their source of income generation.

In this context, several theoretical mechanisms linking climatic shocks to conflict events have been put forward in the literature. Koubi (2019) categorized the main climatic factors contributing to the emergence of conflicts as related to (a) direct effects and resource scarcity, and (b) indirect effects having to do with economic outcomes and migration. The direct effects include increased hostility and violent conflict following temperature anomalies due to physiological and psychological factors, a channel that has been widely disputed and discussed in the recent empirical literature (see Hsiang et al., 2013 or Buhaug et al., 2014, for instance). An additional direct mechanism is related to resource scarcity driven by negative environmental shocks. In the framework of agriculture-based economies, this situation would increase competition, thus leading to conflict and violence. Such a theoretical argument, the importance of which is reinforced by the relevance of the agricultural sector and the natural resources in the economy, is put forward by Homer-Dixon (2010), among others. In addition to this direct effect of climatic shocks on the likelihood of violence via competition over natural resources, there is a potential indirect mechanism through the negative income shocks associated with reductions in agricultural output caused by adverse environmental conditions (Dell et al., 2014). After outbreaks of violent conflict, migration emerges as a natural adaptation mechanism. Acknowledging the multidimensionality of human mobility phenomena, the relative importance of violence as a cause of migration flows (compared to that of mobility driven by environmental change and food security) is explored in Morales-Muñoz et al. (2020).

The complex interactions between environmental phenomena, conflict and migration flows make it particularly difficult to empirically identify causal mechanisms. Abel et al. (2019) examined whether - and, if so, to what degree - environmental change might fuel regional conflicts, and thus result in an increase in forced international migration. Their work analyzed data on asylum applications from 2006 to 2015 for 157 countries, and focused on the identification of causal effects. The results in Abel et al. (2019) confirm that only a few conflicts leading to an increase in forced migration can be causally linked to environmental shocks, particularly in Northern Africa and Western Asia between 2010 and 2012 (and then only for certain countries). These findings imply that climatic shocks do not affect conflicts and thus migration in the same ways or to a similar degree all over the world. It thus appears that climatic shocks related to climate change, and the governmental responses to such shocks, may fuel existing conflicts, particularly in countries that undergo socioeconomic and political changes like those we have seen in Syria over the last decade.

In contrast to the focus on international migration by Abel et al. (2019), Ash and Obradovich (2020) examined the impact that the stress caused by droughts and environmental changes had on conflicts within Syria. Their findings indicate that climate change had at least an indirect effect on these conflicts, which resulted in increased internal migration that itself fueled conflicts. The primary indirect effect in this context was the contribution of environmental change to harvest outcomes and agricultural yields, which highlights the indirect role of income shocks as a driver of violent conflict in agricultural societies. Changes in agriculture and progressive urbanization can add to already subliminal conflicts within countries. Ash and Obradovich (2020) found that the risk of protests increased not only due to drought conditions, but also in response to additional precipitation in relatively cooler districts. In their study they focused on the effects of environmental change anomalies in certain subdistricts of the country. As no migration data were available for the time period under study, with the biggest drought occurring between 2006 and 2010, Ash and Obradovich (2020) used nighttime light intensity data to measure changes in population.

As laid out by Abel et al. (2019) and Ash and Obradovich (2020), it appears that there are certain underlying circumstances in which climatic changes are likely to ignite and feed conflicts. This risk is especially high in regions that have large agricultural sectors, and are thus dependent on harvests and efficient agriculture, as well as in regions characterized by political instability and/or low levels of economic development (Koubi, 2019). These results are also in line with those reported in the meta-analysis by Hoffmann et al. (2020).

Most of the empirical literature linking climatic shocks to migration relies on the use of high frequency data on temperature and precipitation anomalies, droughts and floods. However, the potential effects of climate change on conflict and migration may be of a long-term nature, and could therefore remain hidden when only short-term horizons are investigated. Furthermore, the lack of a consistent definition of the term “migrant” or “conflict” throughout these studies is a common criticism of the existing empirical literature. As there are several different types of conflicts (including interpersonal conflict, inter-group conflict, civil conflict, inter-communal violence, low-intensity conflict, social conflict and political repression), not all results seem comparable or interpretable in the same manner (Koubi, 2019). Brzoska and Fröhlich (2016) criticized the lack of empirical evidence to support the simplistic assumption that climatic shocks lead to resource scarcity, which, in turn, leads to conflict and migration.

In our analysis, we employ the methods used in Abel et al. (2019) to analyse the causal links between drought, conflict and internal migration for the case of Colombia. Given the long history of low-intensity violent conflict in the country, the focus on this particular country offers us insights into the potential role of environmental shocks in causing outbreaks of violence in a context of latent insecurity. The existing literature on internal migration in Colombia has emphasized the role of conflict as a push factor for human mobility, and has concluded that migrants are more willing to relocate to distant locations when confronted with extreme violence (Lozano-Gracia et al., 2010). In addition, Ibáñez and Vélez (2008) found that very sizeable welfare losses are associated with such forced migration flows. Engel and Ibáñez (2007) presented a conceptual theoretical framework for the empirical assessment of the determinants of migration decisions and applied it to data for Colombian households. The results of the study show that while economic incentives are important for explaining mobility decisions, violence affects the relevance of economic factors. Ibáñez et al. (2019) found that the economic impact of weather shocks in Colombia can lead to migration, and that the mobility response is modulated by the existence of particular wartime institutions in the origin location.

Our contribution expands this body of empirical work for Colombia by providing evidence on the causal link between environmental shocks and migration, which is mediated by the emergence of conflict. While other contributions assessing the role of climatic shocks in conflict and migration have relied on information on population density changes to identify mobility (see Ash and Obradovich, 2020, for instance), we employ in our analysis actual migration flow estimates that have recently been developed by Siraj et al. (2020). The use of these data allows us to directly assess the role played by conflict as a transmission mechanism of the effect of environmental events on mobility, without the need for the (potentially unrealistic) assumptions regarding population dynamics that are required to interpret variation in population density as migration changes. In contrast to the analysis in Ibáñez and Vélez (2008), which concentrated on quantifying the welfare losses associated with forced displacement in Colombia, our study focuses on the particular role played by drought risks as a causal determinant of conflict, and thus of internal migration flows in the country. In addition, the particular methodological framework used in our analysis, which aims at providing a rigorous empirical identification of the causal linkages in the hypothesized theoretical mechanisms, has not been used in the context of internal migration until now.

Our method is based on Abel et al. (2019), and uses a gravity equation model to assess the determinants of internal migration in Colombia. This setting builds on the work of Cohen et al. (2008), and integrates an estimation technique developed by Kim (2006) to additionally address the statistical issues of endogenous selection and non-random treatment in conflict occurrence. In particular, it considers potential biases that might be introduced by a lack of randomness in the selection of municipalities affected by conflict and municipalities with non-zero migration flows. Since conflict affects both the existence of migration flows and their size, the method appears to be particularly adequate for the research question at hand. In addition, the emergence of conflict is potentially endogenous and might depend on climatic conditions (and other factors). In contrast to Abel et al. (2019), where the target variable was forced international displacement, in our application we assess the response in terms of internal migration.

Using the model setting in Kim (2006), we set up a sample selection model consisting of three equations, which are estimated in two steps. First, we analyze the impact of climate on conflict occurrence in Colombia. In the remaining equations of the specification, we estimate how conflicts affect internal migration, considering both a model for the existence of non-zero flows of migrants and another model for the size of the flow, conditional on its existence. The specification used is thus given by

where the indicator function I(x) takes a value of one if x is true, and a value of zero otherwise. Equation (1) describes the occurrence of conflict (ci=1) in municipality i in the period of interest (2000 to 2005). Equation (2) specifies whether a non-zero flow of migrants can be observed from origin municipality i to destination municipality j (sij=1), and equation (3) addresses the size of the observed flow of migrants in logs (aij) from municipality i to municipality j for origin-destination pairs with non-zero flows. Because of the lack of exogenous randomness in the selection of municipalities with non-zero outflows of migrants, we assume the error terms εc,i, εs,ij and εa,ij to be jointly multivariate normally distributed and potentially correlated.

The vectors Zc,i, Zs,ij and Za,ij contain the control variables for each of the equations, which are chosen based on existing models for migration and conflict. In order to account for the persistence of past conflict, as well as for the potential effect of environmental shocks, which is at the center of our analysis, Zc,i includes information on the number of battle deaths in the corresponding municipality from 1995 to 1999, the number of conflicts and the average level of the SPEI in the same period. If conflict is long-lived and geographically concentrated, the effect of the first two variables on the probability of subsequent conflict is expected to be positive. Furthermore, Zc,i also contains a variable measuring the shares of black and indigenous persons in the total population (to control for ethnic fractionalization, which is often found to affect the likelihood of armed conflict), the shares of agriculture and mining in total GDP (to control for the sectoral composition of the region), as well as the number of years that have passed since the last conflict (as of the year 2000), which allows us to account for recurring conflicts. Zs,ij contains variables measuring the distance between the municipalities of origin and destination, and the distance squared and two dummy variables, with one indicating whether origin and destination share a common border and the other indicating whether they are located in the same department. These variables are routinely used as covariates in gravity models for migration, and are thus assumed to affect the existence of positive migration flows across pairs of municipalities. Finally, Za,ij contains all variables included in Zs,ij as well as information on total population, GDP per capita and its growth rate, and the shares of agriculture and mining in both origin and destination. As such, this set of controls is typical of gravity models of migration (see Cohen et al., 2008 or Abel et al., 2019). In this setting, we assume that the equation-specific covariates included only in Zc,i do not to affect our main dependent variable directly, but rather through the mediating conflict variable, which allows us to interpret the links between climate, conflicts, and migration in a causal manner.

As proposed by Kim (2006), this sample selection model is estimated as a combination of the bivariate probit and the type II tobit model, which contains the endogenous conflict indicator in both the selection equation and the final equation on the number of migrants. Hence, we control for the endogeneity of ci and the bias caused by the non-random selection of sij. Following Kim (2006), we first estimate equations (1) and (2) as a bivariate probit model. Next, we estimate the outcome equation by applying a Generalized Method of Moments (GMM) estimator. This enables us to obtain estimates of the parameters of the model as if it was a type II tobit model with a bivariate selection and parameter restrictions.

Environmental factors are generally seen as being among main causes of migration. Together with economic, political, social and demographic drivers, these factors are considered in a large number of studies analyzing the triggers of migration flows. In many of these studies, the effects of different causes are evaluated simultaneously, without taking into account the potential causal linkages between them.

Following the methodological setting used in Abel et al. (2019), this paper focused on the interaction between environmental and political factors, and on how this correlation can affect internal migration. In particular, our aim was to make this connection in the context of internal migration in Colombia in the years 2000 to 2005. To establish a causal linkage between climate, conflicts and migration dynamics on the municipality level, we used a sample selection model proposed by Kim (2006). This gravity model considered the issue of endogenous selection regarding the probability of observing a conflict, and regarding the likelihood of non-zero migration flows.

We found a significant positive link between droughts and conflicts, as well as between conflicts and migration, indicating the existence of a causal linkage between climate, conflicts and internal migration in Colombia. Our robustness checks confirmed these results and revealed further details about the observed effects. First, we found that droughts tend to be more closely related to smaller conflicts for which the number of battle-related deaths is below 10. Second, we showed that the effect of drought on the emergence of conflict tends be larger in areas that have not witnessed another conflict in previous years. Our analysis advances the literature by making use of a novel dataset on internal migration flows in the country, which was recently developed by Siraj et al. (2020). The results of our analysis complement the existing results for international migration (see Abel et al., 2019, for example), and confirm that in countries that have experienced violent conflict, human mobility, is more prone to react to environmental shocks than it is in countries with a stable political setting (Hoffmann et al., 2020). In addition, our results provide estimates that can be used to inform estimates of the potential welfare losses from climatic shocks within climate scenarios that incorporate migration (building, for instance, upon the work of Ibáñez and Vélez, 2008).

Providing empirical evidence for one of the pathways through which climatic change can affect migration may be expected to support the design of evidence-based policy measures aimed at mitigating the negative consequences of environmental change trends in the future. Unfortunately, the existing data did not allow us to discern the actual reason for migration, and further data collection efforts should concentrate on improving the information quality of internal mobility data for the country. This is particularly important for differentiating conflict migrants from migration driven exclusively by climate hazards and thus for refining the predictive power of econometric models such as the one employed in this study.

Our analysis leaves room for future research focusing on other countries and different periods of time. Additional evidence might then allow the community to gain an even better understanding of the linkages between climate, conflicts and migration and could make it easier to assess the scope of these dynamics in different parts of the world. Measuring the actual contribution of anthropogenic climate change to changes in the probability of drought episodes in Colombia would require a full-fledged climate attribution study (see Simpson et al., 2021 or van Oldenborgh et al., 2021, for example). This exercise is beyond the scope of the current contribution, but future research that combines the results of attribution studies with the quantitative insights of causal inference models will prove particularly important. Such an interdisciplinary effort would provide extremely valuable insights that can be used to obtain reliable predictions of future migratory pressures.