Growing crops or growing conflicts? : Climate variability, rice production and political violence in Vietnam

(1)

Growing crops or growing conflicts?

Climate variability, rice production and political violence in

Vietnam

William Sundelin

Thesis, 30 ECTS (hp)

Political Science with a focus on Crisis Management and Security Master’s Programme in Politics and War

Autumn 2020

Supervisor: Simon Hollis Word count: 17 876

(2)

i

Abstract

This thesis contributes to research on climate change and violent conflict by testing the theory of a causal relationship between climate variability, agricultural production and political violence in the case of Vietnam 2010-2019. Climate-related negative shocks to agricultural production in developing countries are expected to lower the opportunity cost of violence through an income effect. This increases the risk of violent conflict. The thesis draws on a framework that combines climate-conflict research, civil war theory and research on how climactic factors affect rice cultivation in Southeast Asia. It tests the hypotheses emerging from the framework using mixed-effect models and a counterfactual comparison. Minimum temperature increases in the growing season for rice have been found to decrease rice yields, while maximum temperature increases have a positive effect on yield.

The results show that minimum temperature increases are averse to Vietnamese rice production and have a positive relationship with political violence in the following year. Maximum temperature however is not significantly related to either rice production or violence. These results are in line with the hypotheses drawn from the framework. The minimum temperature effect on political violence is small compared to some of the covariates but robust to several different model specifications. The results provide evidence of a climate-conflict link through agricultural production in contemporary Vietnam which is similar to the findings in existing case studies in Southeast Asia. However, more research will be needed to decisively identify the causal mechanism and the specifics of how it works.

Keywords: climate change, climate variability, security, agriculture, food security, political violence, Vietnam

(3)

ii

1. Introduction

That climate change1 might lead to violent conflict has gained wide recognition in international politics, with the United Nations (UN) declaring it a ‘danger to peace’ (United Nations, 2020). For almost two decades the link between climate change has also been the subject of intense research and scholarly debate within International Relations (IR) and security studies (Raleigh and Urdal, 2007, p. 675, Abel et al., 2019, p. 239).

For a long time, scholars have disagreed whether there is evidence for such a climate-conflict link, with some claiming that there is strong evidence (Burke et al., 2009; Hsiang, Burke and Miguel, 2013) while others argue that there are few results that are robust across cases (Salehyan, 2014; Buhaug, 2015). In recent years there has been a growing consensus among experts that climate change can influence conflict risks. Today climate change is often described as a ‘threat multiplier’ (Mobjörk et al., 2016, p. 14) but the causal pathways and contextual conditions for when and how this happens remain largely unknown and contested (Koubi, 2019, p. 344; Mach et al., 2020, p. 1; Ide, 2020, p. 2).

While the past provides no simple prediction of the future, there are certainly lessons to be learned from it. If and how climate-related environmental change has influenced human conflicts in recent history are key questions related to adaptation and mitigation of climate change during coming decades (Nordqvist and Krampe, 2018, p. 1). But despite intense research and increasingly refined datasets and methodologies, definitive answers have remained elusive (Salehyan, 2014, p. 2).

With the Intergovernmental Panel on Climate Change (IPCC) predicting increased climate variability and frequency of extreme-weather events in the coming century that will strongly impact human societies (IPCC, 2014, p. 13) the current knowledge about climate-conflict links is insufficient from both a research and a policy perspective. Therefore, this thesis aims to contribute to the larger issue of achieving more robust knowledge about the relationship between climate change and violent conflict.

1_{According to the IPCC, climate change is defined as ‘a change in the state of the climate that can be identified}

(e.g., by using statistical tests) by changes in the mean and/or the variability of its properties, and that persists for an extended period, typically decades or longer.’ (IPCC, 2014, p. 5). However, these long-term changes are most often not what is studied in climate-conflict research, mainly because of data constraints. Instead, most studies use year-to-year climate variability in temperature, rainfall and/or drought patterns, or extreme weather

events as proxies for the effects of climate change. Throughout the thesis, I will use the term climate variability

while addressing changes in temperature or rainfall between years or shorter, and climate change when talking about changes that persist over decades or longer.

(6)

2

1.1. Research problem

Climate-conflict research is a truly interdisciplinary field. It spans from the natural sciences studying physical changes to the environment and climate variability to political science, economics and sociology when studying how these translate into societal effects in human societies. Within IR and security studies, the climate-conflict field has seen a rapid development since 2007 when the IPCC Fourth Assessment Report stated that warming of the planet was ‘unequivocal’ and that this could increase the risk of future conflicts by exacerbating scarcity of natural resources (Koubi, 2019, p. 344).

So far, little evidence points toward any direct effect of climate change on conflict. The inconclusive state of evidence led the IPCC to write in its Fifth Assessment Report in 2014 that ‘collectively the research does not conclude that there is a strong positive relationship between warming and armed conflict’ (Field, Barros and IPCC, 2014, p. 772). In recent years however, there seems to be an emerging scholarly consensus that climate change can indeed influence the risk of conflict, but that this is an indirect effect dependent on several intermediate and contextual factors. Moreover, the effect that climactic factors have had on conflicts in the last century is thought to be small compared to socio-economic and political factors (Mach et al., 2019, p. 3; Ide, 2020, p. 2).

Out of the possible theoretical mechanisms outlined in current climate conflict-theory the ‘most plausible’ (Mach et al., 2020, p. 2) revolves around adverse economic outcomes caused by weather phenomenon linked to climate change, such as increased temperature and rainfall variability and extreme-weather events (Koubi, 2019, p. 351). Particularly negative agricultural production shocks have been put forward as the one of the most promising avenues of research (Buhaug et al., 2015, p. 1).

According to current theory, reduced crop yields and failed harvests caused by effects of climate change such as increasing temperatures, drought and rising sea levels are theorized to lead to increased competition between societal groups and to lower the opportunity cost of armed rebellion through an income effect. This agricultural production mechanism is thought to be found in countries where agriculture plays a large part in the economy (Mobjörk and van Baalen, 2016, p. 17; Wischnath and Buhaug, 2014b, p. 7).

However, even the empirical results revolving around this ‘most plausible’ mechanism remain preliminary and inconclusive (Koubi, 2019, p. 353). Much like the larger

(7)

climate-3 conflict field, a key problem is that few results have proven to be robust across cases and studies (Buhaug, 2015).

How is it that climate-conflict research has not arrived at more secure knowledge? Key limitations to knowledge-building given in reviews of the field are theoretical underdevelopment, lack of consensus on key concepts and their operationalizations, an over-reliance on quantitative methods and a regional bias toward studying Sub-Saharan Africa. The combined effect is a state of uncertainty on what conclusions can actually be drawn from the current body of literature (Salehyan, 2014; Koubi, 2019; Ide, 2020).

This thesis aims to address some of these limitations and contribute to more robust conclusions on the relationship between climate-change and conflict. It does so by drawing on state-of-the-art findings and methodology to test the link between climate variability, agricultural production and political violence in the Socialist Republic of Vietnam (hereafter Vietnam) in the period 2010-2019. It outlines and critiques current climate-conflict theory with a focus on agricultural production and food security, and tests hypotheses derived from it against the empirical evidence.

1.2. Research question

As outlined above there is a hypothesized causal relationship between climate change and violent conflict through agricultural production shocks in developing countries dependent on agriculture. Whether such a relationship exists is crucial knowledge for successful climate change mitigation and adaptation. So far there is no conclusive evidence however, with a lack of findings that are robust across cases and existing research being heavily focused on Sub-Saharan Africa. This leads to the following research question:

• How do agricultural production shocks influence political violence in contemporary Vietnam?

For the sake of analytical clarity this research question can be further divided into two sub-questions that guide the research design:

1. Does climate variability cause agricultural production shocks in contemporary Vietnam?

2. Do negative agricultural production shocks increase political violence in contemporary Vietnam?

(8)

4

1.3. Contribution to earlier research

This thesis provides three important contributions to existing research. First, it analyses a previously unstudied country in Southeast Asia, a region where several important factors line up that might make climate-conflict link plausible but where little evidence exists so far (Nordqvist and Krampe, 2018, p. 4). Given the previous research focus on Sub-Saharan Africa this thesis will provide important evidence on the theory’s generalizability (Hendrix, 2017; Adams et al., 2018).

With time, space, and data limitations preventing a total study of Southeast Asia, Vietnam was chosen as the country to be studied since viewed from the lens of climate-conflict theory it could be considered a representative case for the larger population of countries in Southeast Asia (Jason Seawright and John Gerring, 2008, p. 299). Vietnam is today a low-middle income country in where economic growth has led to agriculture decreasing in relative importance for the national economy, from 38,7 percent of total GDP in 1990 to 14 percent in 2019 (World Bank, 2020c). This is representative of a larger pattern of economic diversification in the region (Asian Development Bank, 2017, p. iv). A higher level of development means a harder test for the theory compared to much of previous research that has focused low-income countries. This harder test is a second contribution to the field. Vietnam is very much characterized by a kind of developmental dilemma that the Asian Development Bank uses to describe South East Asia with regards to climate change. On the one hand, economic and human development is making people and societies more resilient to environmental hazards. At the same time, local development coupled with global climate change is opening new sources of vulnerability such as sea-level rise, heat waves hitting increasingly urbanised populations, and decreased crop productivity leading to rural poverty and migration caused by climactic factors (Asian Development Bank, 2017, pp. iv–v).

Parts of Vietnam that have been pointed out as highly vulnerable to effects of climate change such as rising sea-levels, higher temperatures and increasing frequency of extreme weather events include the fertile Mekong River Delta (Smajgl et al., 2015). This too is representative of a region that has been described as ‘extremely vulnerable’ to climate change (Asian Development Bank, 2017, p. 77) with decreased crop productivity being one of the greatest risks, one that is estimated to have substantial welfare effects on national economies (Zhai and Zhuang, 2012, pp. 10–13).

(9)

5 Like most countries in the region Vietnam is ethnically diverse with 54 officially recognized ethnic communities with substantial regional concentrations (Minority Rights Group International, 2020). Finally, Vietnam is experiencing political violence that seems to exhibit an upward trend (Hutt, 2017; ACLED, 2018).

The third contribution lies in the research design. It is aimed at testing climate-conflict theory revolving a single causal mechanism, agricultural production, using hypotheses derived from theory-informed expectations. This in contrast to testing many possible links between climate variables and conflict outcomes without specified theoretical expectations as has been done previously (Wischnath and Buhaug, 2014a is an example). Rather, the thesis design builds on from the few studies that have been made on the region, namely Crost et al. (2018) that study the impact of rainfall on rice production and civil conflict in the Philippines, and Caruso et al. (2016) that study the impact of temperature on rice production and collective violence in Indonesia.

Since it is testing the theory under plausible conditions in a previously unstudied but influential setting, I would argue that the empirical evidence will be valuable to the research field whether it finds evidence that support the theory or not. If a relationship is found between climate variability, agricultural production and political violence in Vietnam it would mean further evidence for the agricultural production mechanism in Southeast Asia and for the relevance of climate-conflict research for middle-income countries. On the other hand, if no relationship is found between climate variability and violence is found this would provide evidence against the agricultural production mechanism in this setting. At the very least it would be evidence of the current theory needing revision to be more generally applicable beyond Sub-Saharan Africa.

1.4. Ontological and epistemological assumptions

This thesis is written with neopositivist foundational assumptions. While neopositivism is widely commonsensical within political science and its methodology dominant within the climate-security field, it is by no means unquestioned or unquestionable as an approach to scientific enquiry within the Social sciences. Therefore, I think it is important to clearly state the foundational assumptions of the thesis to make them visible and possible to critique (Jackson, 2016, p. 34).

(10)

6 Neopositivism assumes a ‘Cartesian’ separation between the human mind and the world that exists outside and independently of it. This leads to a fundamental epistemological question - how might one arrive at secure knowledge about the world if permanently separated from it? The answer that neopositivism gives is that although absolute certainty can never be achieved, justified beliefs about the world might be arrived at by the combined and systematic use of reason (theory) and observation (empirics). This thesis thus subscribes to the notion that the world can be observed and measured to a reasonable level of accuracy, and ‘that knowledge is constructed through the successive proposing and testing of hypothetical guesses about the character of the world’ (Jackson, 2016, p. 58).

Equipped with theoretical models about a causal relationship between concepts (‘X causes Y’), neopositivist research aims to formulate observable expectations and operationalize concepts as variables that are then tested against the empirical evidence. Study design and multiple investigations have to be made to cross four ‘causal hurdles’: a credible theoretical causal mechanism, the possibility of reverse causality should be eliminated, covariation found, and isolation of the effect being studied achieved by controlling for confounding variables (Kellstedt and Whitten, 2018, p. 56).

While no amount of empirical information can make us certain that a theory is correct, we can succeed or fail to falsify hypotheses derived from it (Jackson, 2016, p. 63). This serves as a basis for evaluating the theory, and observations can thus serve as a base for explanation of phenomena and inference about causality (Kellstedt and Whitten, 2018, p. 64).

The neopositivist methodology faces a specific set of problems in social science. Humans are complex beings, and perfect correlation and deterministic causal relationships are rarely found in human societies. Experimental settings that could allow for perfect control of different variables are often impossible to produce, and their external validity can often be questioned (Risjord, 2014, p. 10). This might mean having to settle for observing causal relationships in human societies in a probabilistic sense, ‘X increases the likelihood of Y’ (Kellstedt and Whitten, 2018, p. 57-58). Social reality is multivariate, and we might have to settle for historical data in specific contexts and a limited number of controls and be conscious of the increased uncertainty this brings to any conclusions (Ibid, p. 57).

(11)

7

1.5. Disposition

The thesis is structured in the following way. In section 2, previous research is reviewed, a theoretical framework is outlined, and the hypotheses formulated. In section 3 the research design is described and motivated. Section 4 provides background information about the Vietnam to provide important context to the study. Section 5 provides the results of the statistical analysis and robustness checks of the results. Section 6 follows this up with a discussion on the results and what they might mean for future research.

2. Theoretical framework

This section provides a summary and subsequent critique of climate-conflict research with a focus on the relationship between climate change, agricultural production and civil conflict. It then presents previous research on climate change and rice cultivation in Southeast Asia. These two combined provides the theoretical framework for the thesis and are drawn on to formulate the hypotheses at the end of the section.

2.1. Previous research

2.1.1. Environmental security - resource scarcity

The idea that environmental conditions could lead to violent conflict is not new. An early example is Thomas Malthus who wrote An essay on the Principles of Population in 1798. Malthus argued that in a world of finite resources, population growth would be checked eventually. If this was not achieved by ‘moral restraint’ then it would happen because of war, famine and disease (Mobjörk and van Baalen, 2016, p. 1).

An influential modern development of similar ideas was made by Thomas Homer-Dixon in several in-depth case studies of modern conflicts (Homer-Dixon, 1994, 1999). In this framework, environmental degradation might lead to increasing resource scarcity of natural resources such as water, arable land and pastures. This would fuel tensions between societal groups, increase existing inequalities through elite resource capture, and lead to an increased risk of violent conflict. The link between environmental scarcity and conflict would be found particularly in countries and regions with a low level of economic development and weak state institutions - factors that might otherwise mitigate natural hazards and their effects (Schleussner et al., 2016, p. 9216, Hsiang, Burke and Miguel, 2013, p. 9).

(12)

8 Since the effects of climate change are expected to lead to adverse consequences throughout the world and particularly in developing countries in the Global South, many climate-conflict scholars have imported arguments from Homer-Dixon into theory on how climate change might influence the risk of conflict (Meierding, 2013, p. 189). However, the environmental security theory has been criticised both for being ‘untestable’ because of its many intervening variables (Gleditsch, 1998, p. 390) and deterministic since it underestimates human capacity to adapt and cooperate, something that is emphasised by resource-optimistic ‘cornucopian’ scholars. Whether theories are ‘(neo)-Malthusian’ or ‘cornucopian’, and what those labels actually mean continues to form a point of contention in current literature (Ide and Scheffran, 2014, p. 268).

Most contemporary researchers agree that resource scarcity of renewable resources plays little part in causing war between states (Verhoeven, 2011, p. 682). It appears there are more cost-efficient ways for states to deal with scarcity than going to war. In fact, cooperation seems to be more common than conflict over renewable resources such as water sources (Koubi et al., 2014).

Among the most important findings from this strand of the literature is that a) that resource scarcity can contribute to conflict, although the explanatory power is limited compared to factors like economic development and dysfunctional institutions (Theisen, 2008, p. 810) and b) that scarcity is more likely to lead to inter-communal and intra-state conflict than war between states (Hardt and Scheffran, 2019, p. 5).

2.1.2. Civil war literature – adverse economic outcomes

The other main strand of climate-conflict theory comes from the civil war literature. Quantitative research on civil war onset has robustly identified several factors as increasing the risk of civil war including poverty, low economic growth, weak governance and recent history of conflict (Adger, Barnett and Dabelko, 2013). Climate change is theorized to influence the risk of conflict indirectly through economic factors, mainly through adverse economic outcomes - including income, agricultural production, food prices, and migration (Koubi, 2019, p. 351).

With an increased climate variability and increased natural hazards, climate change is thought to lead to macroeconomic shocks that weaken state revenues while increasing incentives to violently change the status quo, leading to an increased risk of a coup d’état or civil conflict.

(13)

9 This would be particularly evident in countries where the national economies are heavily dependent on agriculture. A frequently cited study by Miguel et al. (2004) uses rainfall as a proxy for economic growth in Sub-Saharan Africa 1989-1999 given the regional dependence on rain-fed agriculture and finds a significant relationship between less rainfall and low growth, and from low growth to conflict risk (Koubi, 2019, p. 351).

Although not initially written as part of climate-conflict research the Miguel et al. study has become influential and similar ones have been conducted on other regions and time periods, with mixed results (Miguel and Satyanath, 2011). More recent studies have failed to find a robust link between temperature and/or rainfall as a proxy for economic growth and civil conflict onset globally (Koubi et al., 2012), in Sub-Saharan Africa (Weezel, 2015) and in Asia (Wischnath and Buhaug, 2014a).

2.1.3. Agricultural production and violent conflict

With the connection between weather conditions and agricultural yield, many climate-conflict scholars have pointed specifically to agricultural production as one of the most plausible causal pathways from climate conflict (Mach et al., 2020, p. 2). Integrating several theoretical strands, a comprehensive framework by Buhaug et al. (2015) is illustrated in Figure 1.

(14)

10 Drawing on conflict opportunity theory developed by Collier and Hoeffler (2004) and Fearon and Laitin (2003), climate-conflict theory views economic opportunities as central to whether a civil conflict breaks out as opposed to ‘objective’ political grievances. In this framework negative agricultural production shocks and loss of agricultural income decrease the opportunity cost of violence for rural populations dependent on agriculture for their livelihoods. When harvests are lost, opportunities for legal income and employment decline and economic incentives to use violence and join insurgent movements increase in relative importance (Wischnath and Buhaug, 2014b).

Responses and adaptation strategies are important in how climate variability translates into a society. First, existing vulnerabilities determine the effect of weather on crop yields. For instance, irrigation systems might improve the resilience of local agriculture to drought (Wischnath and Buhaug, 2014a, p. 713). This means that similar weather conditions might have different effects on different communities.

Second, while effective aid or insurance schemes might alleviate a situation when agricultural production has been affected (Wischnath and Buhaug, 2014b, p. 7) weak and/or corrupt institutions might lead to maldistribution of aid. Furthermore, marginalized groups such as ethnic and religious minorities are less likely to receive government aid and compensation in case of disaster or livelihood loss (Uexkull et al., 2016, p. 12391) which might become a further source of grievance. This means that the capacity and willingness of the state and of other institutions such as international organizations to respond to food insecurity and livelihood loss can affect the outcome.

Third, communities experiencing loss of livelihoods and employment opportunities might respond by increased migration to regions with better economic opportunities. In the short term this would primarily be seasonal and temporary migration of single household members, while long-term climactic stress can cause permanent displacement (Brzoska and Fröhlich, 2016, p. 199). Migration furthermore could create economic stress in receiving regions and/or enforce existing ethnic and cultural divides and lead to non-state conflict between groups, although the empirical evidence for this remains mixed (Brzoska and Fröhlich, 2016, p. 194). There is an ambivalence in the literature as to whether migration represents successful or failed adaptation to climate change (Brzoska and Fröhlich, 2016, p. 198). Unfortunately, migration is often left unmeasured in previous research because of data limitations and the complexities with forming theoretical expectations and separating environmental issues from

(15)

11 other factors motivating migration (Piguet, 2010, p. 1; Wischnath and Buhaug, 2014a, p. 714).

Finally, food price shocks are a theorized driver of violence. The pathway from climate to violence runs mainly from increased food insecurity and price spikes because of crop failure. As food takes up a larger share of income, living standards are decreased. This in turn fuels existing grievances between socioeconomic classes and ethnic groups and lead to resentment against local elites (Brinkman and Hendrix, 2011). The most likely violent outcome of this is thought to be urban food riots, with the main actors being the urban population not involved in agriculture. The income effect of food prices is thus felt outside the communities that are directly involved in agriculture (Rudolfsen, 2020, p. 20).

2.1.4. Studying climate, agriculture and violence

Agricultural production as a pathway from climate to conflict has seen significant scholarly interest in recent years. Common among this quantitative research is to make use of the fact that climate variability, measured in the form of temperature and rainfall fluctuations, can be considered independent from socio-economic and political factors. It is thus simple to rule out reverse causality (that crop failure or violence would cause higher temperatures). Beyond that, two main empirical strategies can be discerned:

Many studies on the agricultural production mechanism use an instrumental variables approach (IV) to address the fact that income could be considered endogenous to conflict. Gawande et al. (2017) and Wischnath and Buhaug (2014b) find correlations between rainfall shocks and increased violence in the Maoist insurgencies in India, while Maystadt and Ecker (2014) connect livestock prices and communal conflict in Somalia. Blakeslee and Fishman (2014) find that drought influence the incidence of crime in India and Hidalgo et al. (2010) that rainfall extremes affect the frequency of land invasions in Brazil.

As for Southeast Asia, few similar studies exist. Caruso et al. (2016) use minimum temperature as an IV for rice production and finds a link from decreased rice yields and inter-communal violence in Indonesia. Wischnath and Buhaug (2014a) and Buhaug et al. (2015) using a similar approach find no relationship between agricultural production and violent conflict in Asia and Sub-Saharan Africa. Taken together, these studies paint a promising but inconclusive picture (Koubi, 2019, p. 353-354).

(16)

12 In Vietnam, while no study has been made linking climate to conflict, Baronchelli and Ricciuti (2018) do find a connection between climate and migration. By using an IV approach inspired by Caruso et al. (2016) on a dataset of 2,088 Vietnamese households in the period 2010-2016 they find that negative shocks to rice production increases the likelihood of at least one household member migrating in the following year (Baronchelli and Ricciuti, 2018).

The other main method employed by researchers is to use within-year climate variability and compare how adverse and beneficial weather conditions for local agriculture affect conflict levels (Koubi, 2019, p. 353). This provides a form of counterfactual evaluation, allowing the researcher to more robustly conclude that the correlation between temperature and/or rainfall and conflict levels is indeed dependent on the effect it has on local agriculture.

By comparing rainfall during the wet season (when the effect on rice cultivation is small) and dry season (when rainfall is important) in the Philippines, Crost et al. (2018) find that negative rainfall shocks in the dry season increase civil conflict incidence in the Philippines in the following year. However, no such relationship with conflict is found between rainfall in the wet season when rainfall levels above average are not beneficial for rice cultivation. A similar conclusion is reached by Jun (2017) for maize in Sub-Saharan Africa. Harari and Ferrara (2018) meanwhile show that rainfall shortages during the growing season for maize has a higher influence on conflict incidence that at other times during the year in 39 African countries.

Summing up this section, there is some empirical support for a link between climate and conflict running through the effects of climate change producing adverse weather conditions for agriculture. This relationship is dependent on local contexts ranging from geographical conditions and the crops being grown in the region to how dependent the local economy is on agriculture, as well as whether the state and other institutions are willing and able to mitigate the loss of agricultural livelihoods. Moreover, these results seem to show that this loss of agricultural production can lead to different forms of violence depending on the social and political context.

2.2. Critique of previous research

Despite intense research during more than two decades, in fact there is still little that can be decisively said about any climate-conflict links (Salehyan, 2014; Buhaug, 2015; Ide, 2020).

(17)

13 While the reason for this could be that no causal relationship exists between climate and conflict, many researchers have pointed to theoretical and methodological limitations in the state of the art that has prevented progress towards more secure knowledge (Koubi et al., 2012; Ide and Scheffran, 2014; Seter, 2016; Adams et al., 2018).

Below follows a critique of previous research structured around these limitations and focused on the link between agriculture and conflict. Finally, it discusses the implications for future research.

2.2.1. Theoretical underdevelopment and operationalizations

A key problem of both theory and methodology of climate-conflict research is the appropriate spatiotemporal dimensions (Buhaug, 2015, p. 272). Put simply, what are the proper time periods and geographical spaces to investigate climate-related weather conditions and where and when to expect specific societal effects?

While many studies have tended to use a one-year lag between climatic condition and social outcome (violent conflict), there is often little theoretical motivation to why this is appropriate. This is a problem especially for the studies using conflict onset as the dependent variable. In the case of Darfur, one of the frequently mentioned examples of ‘climate-related conflict’, current qualitative research suggests that it took decades for drought to translate into conflict. Such long-term effects would not be found using the standard methods of climate-conflict research (Selby, 2014, p. 837).

Interannual effects of climate variability is more plausibly linked to conflict intensity rather than outbreak, with income shocks affecting recruiting opportunities for existing armed groups among the rural populations and increasing resentment towards the government (Wischnath and Buhaug, 2014b, p. 8). This would lead to a more severe conflict in terms of participants and casualties. Koubi argues that evidence so far suggests that climate-related economic shocks ‘do not lead to a higher risk of conflict, but rather they affect dynamics of conflict such as duration, severity, and intensity.’ (2019, p. 354).

There is also an important theoretical distinction to be made between agricultural income and food prices and their links to violence. Although the two could both be related to local climactic conditions, they are differently related to fluctuations on the world market and/or extreme weather events in other parts of the world. For example, Sternberg (2012) argues that

(18)

14 drought in North-eastern China contributed to a bread price spike in Egypt in 2010/2011 which contributed to the Egyptian revolution.

From an actor perspective, food price increases could have the opposite effect for rural producers and urban consumers. A higher food price would increase the agricultural income for farmers. Therefore, these two mechanisms should preferably be distinguished between in the study design (Rudolfsen, 2020, p. 19) but this is not always reflected in previous research (Wischnath and Buhaug, 2014b; Buhaug et al., 2015).

Moreover, as described above many studies have used rainfall as an instrumental variable (IV) for agricultural income. Doing so rests on the assumption that rainfall is only related to conflict through rural livelihoods, but this is often questionable. Notably, Sarsons (2015) uses rainfall as a measure of agricultural income and finds a relationship between income shocks and Hindu-Muslim rioting in India. However, an increase in riots was observed in regions downstream from a dam despite these regions being protected from rainfall shocks thanks to the irrigation systems. The effect remains when controlling for migration and spillover, which suggests that rainfall is related to conflict in other ways than agricultural income. This puts in question the isolation criterion that is central to instrumental variable designs (Sarsons, 2015, p. 29).

Another example of multifaceted effects of climate on conflict is found in East Africa. Quantitative studies have found that conflict events are more likely to occur during rainy years rather than dry ones. Subsequent qualitative studies have shown that this is due to the tactical considerations of armed groups. Rainfall and the subsequent increased vegetation make it easier to find cover and for cattle to cover long distances, which facilitates cattle raids (Mobjörk and van Baalen, 2016, pp. 27–29). Furthermore, studies on extreme weather events have shown that they might disrupt the logistical networks and damage capabilities of armed groups which would lead to decreased violence at least in the short term (Walch, 2018). While the above examples do not necessarily contradict climate-conflict theory, they show that there might be several different effects of climactic conditions translating into different social effects at the same time.

(19)

15 2.2.2. Interaction with environmental peace theory

The ‘environmental peace perspective’ is comprised of several bodies of literature and intellectual traditions that cannot easily be summarised into one (Ide, 2019, p. 328). This thesis is mainly concerned with those parts of peace and conflict studies and disaster studies that focus on empirical links between environmental and cooperation rather than conflict, and the relevant contextual factors for this to occur.

There are several theoretical mechanisms for an environment-peace link. They include cooperation to solve uncoordinated competition over scarce resources, that might decrease the risk of conflict. Empirical examples on the international level include interstate cooperation over shared rivers such as the Indus Treaty, the Nile Basin Committee and the Mekong River Initiative (Ide, 2019, p. 332). Although such cooperation has not prevented conflict completely, transboundary conservation projects correlated with reduced violence (Barquet, Lujala and Rød, 2014).

On the sub-national level, disaster studies are divided as to whether extreme-weather events increase or decrease the risk of civil conflict (Hollis, 2018, p. 22). The arguments for disaster-conflict links lean on largely the same theoretical foundations as climate-disaster-conflict research. On the other hand, disaster sociology has identified that the shared experience of a disaster can reduce individual and cultural differences across groups, creating a common understanding of the situation and promoting cohesiveness and solidarity (Slettebak, 2012, p. 165; Hollis, 2018, p. 24). Of note is Slettebak (2012) that uses a multivariate analysis and finds that disasters related to extreme-weather events decreased the risk of civil war.

Cooperation can also lead to the formation of shared institutions that are created to address environmental problems but end up facilitating communication between groups on non-environmental matters, decreasing the risk of further conflict. Similarly, existing local institutions can help moderate increased environmental stress and prevent conflict (Adano et al., 2012; Linke et al., 2018).

There is considerable evidence that in certain contexts environmental stress and hazards could lead to peace and cooperation rather than violence, but climate-security interaction with this perspective has been small considering the relevance of the two fields for each other. People all over the world are already responding to climate change in various ways, many of them non-violent and cooperative (Ayeb-Karlsson et al., 2016). By focusing solely on risks

(20)

16 of conflict and not including the potential for cooperation in the climate-security literature risks missing a large part of the picture.

2.2.3. Regional bias

Finally, there is a strong regional bias in current climate-conflict research where a majority of both quantitative and qualitative studies have focused on countries in Sub-Saharan Africa (Adams et al., 2018). To a degree, a vulnerabilities-based approach motivates investigations on climate-conflict links in the Global South since dependence on agriculture, low economic development, weak institutions and recent history of conflict are thought to be necessary contextual conditions for the theory (Scheffran, Ide and Schilling, 2014, p. 372).

However, Hendrix (2017) argues that the focus on Sub-Saharan Africa, and even the sample selection within the region represents a ‘streetlight effect’ where research is based on convenience rather than academic or policy relevance. Former British colonies (meaning better access to English-language sources) and countries with greater civil liberties and political stability receive more scholarly attention as opposed to vulnerability to climate change or adaptive capacity (Hendrix, 2017, p. 1)

Besides criticism that the narrow geographical focus leads to regions becoming ‘stigmatized as inherently violent and unable to cope with climate change peacefully’ (Adams et al., 2018, p. 202; Hardt and Scheffran, 2019, p. 7), there are also serious concerns about what this means for knowledge-building (Wischnath and Buhaug, 2014a, p. 710). For instance, some of the most important findings in climate-conflict concern pastoral and sedentary communities in East Africa (Mobjörk, 2017, p. 294). However, it is questionable how generalizable these results are to other regions of the world that do not have large pastoral populations (Nordqvist and Krampe, 2018, p. 2). This highlights the need to study other parts of the world where plausible conditions for climate-conflict links exist.

Taken together the critique of the climate-conflict field presents some important points to consider for future research. For one, research designs should allow for the possibility that weather conditions might have multiple, and sometimes opposite effects on a society at the same time. For instance, the same rainfall that is beneficial to local agriculture might create disruptions of rural infrastructure. Second, the possibility for non-violent or cooperative outcomes should be incorporated into theory and methods. Finally, climate-conflict theory

(21)

17 needs to broaden its sample and include more countries and regions to arrive at more generalizable findings.

2.3. Climate change and rice cultivation

While a link between climate change and violent conflict remains contentious the research is significantly more conclusive when it comes to the effects on agriculture. Agriculture has been pointed out as one of the economic sectors that are most vulnerable to climate change, with the IPCC projecting that it will affect food security at local, regional and global levels (Zhai and Zhuang, 2012, p. 1). These vulnerabilities include the cultivation of rice, the staple food crop for nearly half the world’s population. About 90 percent of global rice cultivation is done in Asia and rice is essential both for food security and national income in most parts of Southeast Asia (Zhai and Zhuang, 2012, pp. 10–13).

The two main effects of climate change that are most likely to impact rice production are growing concentrations of CO2 in the air and increasing temperatures, respectively. While the

former is projected to have a positive effect on global rice yields, this effect is cancelled out by the major negative impact caused by the latter (FAO, 2018, pp. 11–13). Agronomical research on how climate impacts rice plants stress the importance of minimum (night-time) temperature and maximum (daytime) temperature as opposed to temperature averages. Particularly, Peng et al. (2004) find that each degree increase of the minimum temperature during the critical growing period decreased rice yields by ten percent from 1991 to 2003 (FAO, 2018, p. 15).

Welch et al. (2010) arrive at similar conclusions studying 277 rice fields managed by farmers in six major producer countries in the region. In their study, maximum temperature has a weaker but positive effect during the early growing stages of the rice plant, which is called the vegetative phase and typically lasts around 60 days. The effect of temperature on the rice plant is related to the optimal temperature for the plant, and so these results hold for countries in tropical and subtropical Southeast Asia while the effect was the opposite in North and Northeast China because of the lower average temperatures in northern latitudes (Tao et al., 2008).

The effects of climactic factors on rice yields in Vietnam have been studied by Baronchelli and Ricciuti (2018) and Chung, Jintrawet and Promburom (2015) among others. Both studies are conducted with attention to the growing seasons for rice. Vietnamese rice is grown

(22)

18 primarily in the Mekong River Delta and to a lesser extent the Red river Delta and the Northeast and North Central Coast. There are three planting seasons in the Mekong Delta: Mua (monsoon) which is the main season, He-Thu (or Summer-Autumn) and Dong-Xuan (or Winter-Spring). Most of the rice in the country is grown during the two former periods, with Winter-Spring rice making up a minority of the yield (Baronchelli and Ricciuti, 2018, p. 3). Table 1 shows the rice production seasons in Vietnam.

Table 1. Rice production seasons in Vietnam

Planting Harvesting

Main May-Aug Sep-Dec

Winter-Spring Dec-Feb Apr-Jun

Summer-Autumn Apr-Jun Aug-Sep

Source: GRiSP (2013); Baronchelli and Ricciuti (2018); FAO, (2020)

Baronchelli and Ricciuti conduct a large N-study of 2,088 households distributed around the country, measuring the effects of minimum temperature in June since this coincides with the two main growing seasons. The study finds a significant negative effect of minimum temperature in June on yearly rice production per household (Baronchelli and Ricciuti, 2018, p. 14).

However, the case study conducted by Chung et al. in the Central Highlands of Vietnam finds no significant relationship between minimum temperature and rice yield, while higher minimum temperature during the Winter-Spring season increased rice yields. These seemingly opposite results are explained by the fact that the measured months in the Winter-Spring growth season occur during the coldest time of the year, where cold rather than heat is problematic for growing rice (Chung, Jintrawet and Promburom, 2015, p. 86). This mirrors the fact that increased minimum temperatures had no negative effect on rice yields in temperate climate areas (Welch et al., 2010).

The above means that the combined effect of climate change is estimated to be a significant decrease in rice yields in Vietnam (Thế et al., 2015). Crucially though variations in temperature seem to have different results depending on not only on when they occur during the year. Specifically, minimum and maximum temperatures in the main growing season (May-August) seem to have opposite effects on rice yields.

(23)

19

2.4. Hypotheses

Combining previous research on climate-conflict links and how rice cultivation is affected by climate change provides a framework from which to formulate the hypotheses. The hypotheses with respective null hypotheses are summarised in Table 2.

With temperature in the growing season shown to have a significant effect on rice yields in farms throughout Southeast Asia (see Welch et al., 2010) and on a household level in Vietnam (Baronchelli and Ricciuti, 2018) the same effect could be expected to be found when aggregated to the province level (H1a). Specifically, rice production in Vietnam can be expected to be negatively associated with minimum temperatures in the growing season (H1b) while maximum temperatures have no or a weak positive effect (H1c).

Furthermore, all other things being equal, negative chocks to rice production are expected to increase the likelihood of political violence through affecting agricultural income and/or food prices, lowering the opportunity cost for violence (H2a). The income effect would take some time to translate, which is why the expectation is to see the effect in the following year. Importantly, given the opposite effects of minimum and maximum temperature on rice, if there is indeed a causal relationship between temperature and violence running through rice production one would expect to see a corresponding difference in their effects on violence. Higher minimum temperatures in the growing season for rice should have a positive relationship with violence (H2b), while maximum temperatures should not (H2c).

Table 2. Hypotheses

1. Temperature and rice production

H1a. Climate variability affects rice production.

H1a0. Climate variability has no effect on rice production.

H1b. Higher than average minimum temperatures in the growing season decrease rice production. H1b0. Minimum temperatures in the growing season do not affect rice production.

(24)

20

H1c. Higher than average maximum temperatures in the growing season do not decrease rice production. H1c0. There is no difference in effect between minimum and maximum temperature on rice production.

2. Temperature and political violence

H2a. Negative shocks to rice production increase the likelihood of political violence. H2a0. Rice production has no effect on political violence.

H2b. Higher minimum temperatures in the rice growing season increase the risk of political violence in the

following year.

H2b0. Minimum temperature is not related to political violence in the following year.

H2c. Higher maximum temperature in the rice growing season does not increase the risk of political violence in

the following year.

H2c0. There is no difference in the effect of minimum and maximum temperature on political violence.

3. Methods

3.1. Research design

To test the hypotheses outlined in the previous section and answer the research question outlined in section 1.2., the thesis uses statistical analysis to conduct a panel study on the 63 provinces of Vietnam in the period 2010-2019. This kind of study where the same subjects (the provinces) are observed repeatedly over multiple points in time is also known as a panel design study and is frequently used in political science and econometrics to make causal inference with observational data (Allison, 2005).

To allow for inference on the empirical relationship between climate variability, agricultural production and political violence, a two-step research design is used. In the first step, the validity of the effects of minimum temperature, maximum temperature and rainfall in June

(25)

21 (which is in the mid-point of the main growing season) on rice production will be analysed using a linear mixed model. In the second step, political violence in the following year will be regressed on these climate variables in a mixed effects negative binomial regression. The expectation is that the design will allow for comparison of the effect on political violence of climate variables that are negative to rice production and those that are not.

The statistical models used are mixed effect linear and negative binomial regressions, respectively. Mixed effect models are appropriate for studying data that is structurally correlated, such as where the same subjects are repeatedly measured over time which is the case here (IBM, 2014). They allow for modelling unmeasured differences between subjects as random intercepts and slopes while preserving statistical power to measure the effect of fixed effect coefficients (the explanatory variables of interest) on the dependent variable. Mixed effect models can also fit multiple relationships, such as when subjects belong to different sub-groups in the dataset that are correlated (Love, 2018). This is particularly useful in this case where provinces in the same region might share common characteristics.

3.2. Data structure and sources

The units of analysis are the 58 provinces and 5 municipalities that make up the first-level administrative divisions of contemporary Vietnam (see Figure 2). With a total of 63 administrative divisions (I) studied over 10 years (T) this means a grand total of 630 observations (N) in the main study (step 2). While the full dataset runs from 2010-2019, data on temperature, rainfall and rice production per province runs from 1995-2019 which allows step 1 of the analysis to include 25 years and 1548 observations.

Data on agricultural production, population, infant mortality and other socioeconomic factors, and migration has been retrieved from publicly available information at the General Statistics Office (GSO) in Vietnam (GSO, 2020). The GSO is regarded a competent statistical agency, with an overall statistical capability score of 78.9 as calculated by the World Bank, which is above the average value for countries in Southeast Asia at 76.25 (World Bank, 2020a). As the GSO provides macroeconomic data to FAO and the World Bank, country aggregates have been cross-referenced to see that they match these databases. Data on province-level irrigation systems was retrieved from AQUASTAT, FAO’s global information system on water management systems (FAO, 2016).

(26)

22 Climate data is retrieved from the Global Historical Climatology Network (GHCN) at the National Oceanic and Atmospheric Administration, under the United States Department of Commerce (NOAA, 2020). The GHCN-Daily database integrates information on daily measurements on temperature and rainfall collected from thousands of land-based weather stations distributed around the globe and is frequently used by researchers.

Monthly averages in June and January for the stations are constructed from the daily measurements of average temperature (Tavg), maximum temperature (Tmax), minimum

temperature (Tmin) and rainfall (Ravg). The monthly average in the period 1995-2019 is

subtracted from the observation which create a centered variable that captures how the specific month year t deviated from the mean of the period. The variables are centered around the mean to allow for modelling interaction effects since the variables are correlated.

There are a total of 14 land-based stations located in Vietnam 1995-2019 in the GHCN-Daily dataset and each province has been paired to the closest weather station available at the time of measurement using their coordinates and publicly available coordinates for the provinces (Google, no date). If more than two weather-variables are missing in a specific year, the second closest weather station has been used instead. The pairings are shown in Table 3 and illustrated in Figure 2.

As the response variable, a variable is constructed measuring the number of violent events per province and year has been created using data from the Armed Conflict Location & Event Data Project (ACLED) dataset (Raleigh et al., 2010). ACLED monitors ‘political violence’, defined as ‘the use of force by a group with a political purpose or motivation’ (ACLED, 2019, p. 6). Data is recorded in events divided into six main types and with 25 subcategories including battles, explosions, violence against civilians, and riots. However, in some cases nonviolent demonstrations and strategic events are included ‘to capture the potential pre-cursors or critical junctures of a violent conflict’ (Ibid, p. 6).

ACLED has been chosen for its wider definition of political violence that is more in line with the social consequences expected by the theoretical framework as opposed to the main alternative, the Uppsala Conflict Data Program (UCDP). Moreover, the narrow definition of conflict used by the UCDP-GED has been criticised for not accurately capturing violence in Southeast Asia when compared to local violence monitoring systems in Indonesia, Thailand and the Philippines (Barron, Engvall and Morel, 2016). The difference between the datasets in Vietnam is quite illustrative: while the UCDP-GED records no events in in Vietnam the

(27)

23 period 2010-2019, ACLED records 509 events and 132 fatalities. Of these fatalities, a majority occurred during two peaks; Hmong ethnic riots in May 2011 and anti-China protests in May 2014 (ACLED, 2018).

Table 3. Weather stations in Vietnam 1995-2019

Station Associated provinces Active Map ref.

Cao Bang Bac Kan, Cao Bang, Ha Giang 1959- A

Lang Son Lang Son, Quang Ninh 1974- B

Son La Dien Bien, Lai Chau, Lao Cai, Son La, Yen Bai 1997- C Noi Bai Int. Bac Giang, Bac Ninh, Phu Tho, Thai Nguyen, Tuyen Quang, Vinh

Phuc

1958- D

Ha Dong Ha Nam, Ha Noi, Hung Yen 2012- E

Phu Lien Hai Duong, Hai Phong, Nam Dinh, Thai Binh 1974- F Thanh Hoa Hoa Binh, Ninh Binh, Thanh Hoa, 1974- G

Vinh Ha Tinh, Nghe An 1974- H

Dong Hoi Quang Binh, Quang Tri, 1974- I

Danang Int. Binh Dinh, Da Nang, Gia Lai, Kon Tum, Quang Nam, Quang Ngai, Thua Thien-He

1957- J

Phan Thiet Binh Thuan, Dak Lak, Dak Nong, Khanh Hoa, Lam Dong, Ninh Tuan, Phu Yen

1956- K

Tan Son Hoa

Ba Ria – Vung Tao, Ben Tre, Binh Duong, Binh Phuoc, Dong Nai, Dong Thap, Ho Chi Minh city, Long An, Tay Ninh, Tien Giang, Tra Vinh, Vinh Long

1951- L

Ca Mau Bac Lieu, Ca Mau, Can Tho, Hau Giang, Kien Giang, Soc Trang 1957- M

(28)

24

Figure 2. Map of provinces (1-63) and weather stations (A-O) in Vietnam

(29)

25

3.3. Empirical strategy

3.3.1 Climate variability and rice production

In the first step, a linear mixed model is used to examine the effects of variations in temperature and rainfall on agricultural production. Given the singular importance of rice both as a food crop and export commodity in the Vietnamese economy, log rice production per province Rice (ln) is used as the response variable. Considering previous research on climate and agriculture in the region, climate variability is measured as deviation of minimum temperature and rainfall in June. The resulting model can be illustrated by the following equation:

Rice (ln)it-1 = β0 + β1Tminit-1 + β2Tmaxit-1 + β3Rainit-1+β4Time + εit-1 + u1i + u2iTime

(1) With production in province i at time t being predicted by climactic factors during the growing season that same year. The model controls for unobserved differences between provinces using random intercepts for each province (u1i) and correlations between provinces

in the same region, and a random slope (u2i) that accounts for unobserved differences over

time for each province. To capture the general time trend such as technological development and the introduction of new crop variants, a time fixed effect (β3) is used. Finally, εit is a normally distributed error term clustered at the province level. The covariance structure chosen is first-order autoregressive (AR1) because of the assumption that production levels will be more correlated the closer they are in time.

It should be highlighted that the above model rests on the assumption that the relationship between climate and rice yield can be accurately captured by a linear model. This assumption should be evaluated by examining the data but corresponds with the findings of Welch et al., (2010).

3.3.2. Climate variability and political violence

In the second step, to estimate the effect of climate variability on violence working through agricultural production, lagged rainfall, minimum and maximum temperature are used as the explanatory variables to predict violence in the following year. According to the framework,

(30)

26 temperature is expected to affect political violence through agriculture, and the year lag is introduced to allow for plausible time for this effect to play out.

The response variable in this case (violent events) is a count variable that can take on any integer from 0 upwards. Upon examination the data was found exhibit overdispersion (the variance being much larger than the mean) which means that a mixed effects Negative Binomial (NB) regression is the most appropriate model for the data rather than a linear regression, which assumes a continuous dependent variable, or a Poisson regression that assumes a variance equal to the mean (Date, 2020).

Unlike a linear regression, NB regression is calculated using a log link function that exponentiates the linear explanatory variables (UCLA: Statistical Consulting Group, no date). The results of the fixed effects can be returned either as log likelihood or transformed back to a linear coefficient, in which case it is interpreted in much the same way as a linear regression. The resulting model can be illustrated by the following equation:

Violenceit = β0 + β1Tminit-1 + β2Tmaxit-1 + β3Rainit-1+ βQQit-1 + εit + + u1i + u2iTime

(2) Where conflict intensity in country i at time t is predicted by lagged temperature in the previous year. The model will rely on a set of covariates (Q) selected from previous literature and available data. These control variables are meant to capture relevant socioeconomic and political differences between countries and over time. The error term is represented by εand standard errors are clustered at the provincial level. Random intercepts (u1i ) and slopes in the form of a random time effect (u2i) for provinces based on region-groupings will be used to control for unobserved differences between subjects in the same way as in step 1. As in the previous model covariance structure chosen is first-order autoregressive (AR1) because of the assumption that violence levels will tend to be more correlated the closer they are in time.

The selection criteria for the control variables are that they could be thought to influence both the explanatory variable, that is rice production level and frequency of political violence. Apart from previous conflict, these are population (ln)t-1 for population size and infant

mortality ratet-1 as a measure of economic development level in the province. To decrease

sensitivity of the results to model specifications, alternative specifications are tested and compared, and further covariates added in robustness tests.

(31)

27

3.4. Discussion of the research design

The chosen research design allows for an in-depth examination of interactions between climate variability and political violence in Vietnam. It uses an empirical strategy that is similar to what has been employed in recent single country studies. Exploiting time-variations that has previously been employed in previous studies in Sub-Saharan Africa (Jun, 2017; Harari and Ferrara, 2018) as well as in the Philippines (Crost et al., 2018). In this case, Tmin and Tmax are measures collected at the same time of the year but represent different times

of the day, with the former being reached at night-time and the latter during daytime. As they are expected to have different effects on rice production which allows for a counterfactual comparison.

Given the lack of results robust between cases and the lack of regional diversity in climate-conflict research, Southeast Asia as a region, the design should be able to answer the research question and thus provide important knowledge to climate-conflict research. Still, there are important design choices that should be highlighted. The chosen research design addresses in changes in weather patterns within and between years (climate variability) and not climate change per se that takes place over decades. To a degree this is offset by the first step dataset spanning 24 years, but the fact remains that the main dataset only covers a single decade. I argue that climate variability is a valid measure. Most scholars within the field agree that yearly variations are valid, although imperfect, proxies for climate change (Mobjörk and van Baalen, 2016, p. 41) and temperature increases are among the most well-documented consequences of global climate change. Also, given the available data on inter-communal and political violence, an analysis spanning several decades is currently not possible in Southeast Asia (let alone Vietnam). Disaggregated datasets of political violence spanning the region over multiple decades similar to the Social Conflicts in Africa Database (used in Salehyan et al., 2012) do not currently exist.

Perhaps more importantly, an investigation into the effects of agricultural income on conflict severity is motivated irrespectively of if the source of income fluctuations can be decisively linked to climate change. By analysing the societal outcome of agricultural production shocks in general the study connects with the wider civil war literature. Therefore, the investigation is a relevant contribution to existing research even if it cannot be decisively shown that climate change is the ultimate cause of adverse weather conditions (Wischnath and Buhaug, 2014b, p. 11).

(32)

28 A crucial part of climate-conflict research designs is the choice of the appropriate time and spatial units of analysis. Using years as the time unit is motivated primarily by parsimony and the nature of the dependent variable. A much more complex research design would be needed to address long-term effects of climate change, and still would inevitably involve several arbitrary methodological choices if done as part of a master’s thesis. In comparison, I argue that the selected design is simple enough to be performed within the thesis format while conforming to the methodological rigour that can be demanded of it.

As for the choice of provinces as spatial units, it follows from the framework that inter-communal and civil violence is the expected outcome of negative shocks to agricultural production. The framework does suggest that migration could lead to the effect being transferred to other regions, particularly richer regions and cities. Because of space constraints and a lack of developed theoretical expectations an analysis of the role of migration is not included in the analysis.

It should also be clearly spelled out why a research design exploiting intra-year climate variability was chosen in favour of the instrumental variables approach and Two-Stage Least Squares Regression (2SLS). Using an IV approach would mean that temperature and rainfall are used as exogenous predictors agricultural yield without being included directly in the model. While in theory this would be a way to address the problem of reverse causality and arrive at more conclusive evidence of a causal relationship, true instrumental variables are hard to find.

In addition to relevance in the form of substantive effect on the endogenous independent variable (agricultural production) IV:s must also fulfil the exclusion criteria to only affect the response variable via the endogenous explanatory variable, agricultural production (Sovey and Green, 2011). As was noted in the in section 2, the assumption that climate variability affects conflict levels only through agricultural production is hard to sustain given the findings that suggest the contrary, particularly Sarsons (2015).

Finally, it could be argued that the over-reliance on quantitative studies in climate-conflict research makes this study irrelevant. However, as was shown in section 2, I argue that the fundamental problems within the field have been not so much the use of quantitative methods per se. Instead, it is the lack of theory informed design choices, sampling bias and lack of interaction with qualitative studies that constitute flaws in current climate conflict research.

Growing crops or growing conflicts? : Climate variability, rice production and political violence in Vietnam