Abstract
Despite alarming predications about the Covid 19 pandemic that appear to fit the literature on the impact of natural disasters on civil wars, there are reasons to be suspicious that a rise in militant violence would likely occur quickly or uniformly. Although the COVID-19 pandemic is most definitely a disaster that caught the world by surprise, this “slow-rolling” shock differs in important ways from the more commonly studied acute onset natural disasters such as earthquakes, hurricanes, and tsunamis that often increase violent competition among groups for scarce resources. Instead, the effects of slow-rolling disasters unfold in phases that, at least in the short run, are likely to encourage a period of relative decline in violence, as actors try and assess the effects of COVID-19 on their organization and their opponents. Both statistical and qualitative evidence from the initial months of the COVID-19 pandemic supports the initial phases of our theory.
Section 1: Introduction
How would the COVID-19 pandemic affect militant violence? Alarming predictions by numerous scholars and policy analysts suggested that the virus would lead to growing opportunities for terrorism, insurgency, and armed political violence, especially exacerbating armed conflicts that do not require militants to overcome restrictions on international travel related to the pandemic (Ackerman and Peterson 2020; Byman and Amunson 2020; United Nations Counter-Terrorism Committee 2020). These predictions fit, at first glance, the larger literature on the effects of natural disasters on militant violence, which generally finds that the shock of major disasters such as earthquakes, hurricanes, and tsunamis increases competition among groups for scarce resources and so provokes greater levels of militant violence and civil war (Brancati 2007; Nel and Righarts 2008; Wood and Wright 2016).
Yet, there are reasons to be suspicious that a rise in militant violence would occur quickly or uniformly. Although the COVID-19 pandemic is most definitely a disaster that caught the world by surprise, this shock differs in important ways from the acute onset natural disasters that are the central focus of the existing literature on natural disasters and political violence. In acute onset disasters, the loss of human life and economic damage are concentrated in the initial days or weeks following the shock. The current pandemic is a different kind of shock, and hence likely to produce effects on political violence differently than acute onset disasters.
A key defining feature of the COVID-19 pandemic is that the death and economic damage unfold over many months and perhaps several years. Compared to acute onset disasters where losses are high, immediate, and largely known, the current pandemic is a slow-rolling disaster, where the losses start low, mount over time at an uncertain rate, and so ultimate losses are highly uncertain at the inception and initial months of the shock. Accordingly, the causal logic of an acute shock leading almost immediately to an abrupt competition for scarce resources and so increased political violence among rival groups cannot be assumed to hold at least in the early stages, if at all.
The purpose of this study is to clarify the logic of how slow-rolling disasters are likely to impact political violence by militant non-state actors and examine the available evidence through the end of December 2021 to see how well patterns of militant violence during this period of the COVID-19 shock fit our theoretical expectations. Apart from the difference in the speed of onset, COVID-19 shares the key elements of the definition of disasters used elsewhere, “a situation or event that overwhelms local capacity, necessitating a request at the national or international level for external assistance; an unforeseen and often sudden event that causes great damage, destruction and human suffering” (CRED 2018, 7). 1
Contrary to acute onset models, our theory of slow-rolling disasters suggests that the relevant political actors – existing militant groups as well as states – are likely to experience the human effects of the shock differently at different stages of the shock, with corresponding changing implications for political violence. Specifically, in the early phase following initial onset, our theory expects militant groups as well as states to react to uncertainty about the protracted consequences of the shock by focusing on information gathering, assessment, and cautious husbanding and protection of capacities. Accordingly, militant violence – even in the face of opportunities due to the weakness and distraction of rivals – is likely to decline immediately after onset, as militant groups may continue on-going operations but suspend the initiation of new plots, schemes, and campaigns. The second phase, where the shock plateaus, involves adjustment to a persistent abnormal and a period of relatively steady state with modest change in violence compared to the prior period. This is followed by a third period of a steady decline of the shock and so ultimately a return to pre-shock levels of violence or higher through multiple pathways by surviving or new groups that emerge during the shock.
Overall, we believe that at least in the short run, there is likely to be a period of relative decline in violence, as actors try and assess the effects of COVID-19 on their organization and their opponents, and only after this period will we see the effects predicted in existing theories of natural disaster and militancy. The long-term effects of COVID-19, such as overwhelming existing local capacity, decreasing economic activity, and raising social and psychological stress, fit with the definition of disasters used elsewhere in the literature. While the effects of the COVID-19 pandemic will likely take longer to be felt than in a sharp-onset disaster such as an earthquake, they will be felt nonetheless.
Both statistical and qualitative evidence from the initial 2 years of the COVID-19 pandemic supports our theory. Given concerns about confounding variables and data quality, we use a discontinuity design that compares incidents of political violence across equivalent periods of weeks immediately before and after a country’s first COVID experience. For the initial phase, we compare violence 3 weeks before and after the date of the first reported COVID-19 fatality to estimate the relationship of COVID-19 to violence, for the second phase we compare violence 52 weeks before and after the date of the first reported Covid-19 fatality. We do not provide evidence for phase 3, as we believe that we are only now entering this phase with the waning of the omicron variant. For each of the first two phases, we find evidence consistent with our predictions on incidents of political violence, and this correlation is robust to different modeling assumptions and measures of the dependent variable.
Further, we test the plausibility of the mechanisms and alternative explanations using a qualitative case study on ISIS in Iraq, a critical case. Based on analysis of ISIS print and video propaganda from 2020, it appears that the group’s calls to action have not increased as a result of COVID-19, and there are clear instances of the group focusing on maintaining the health of their fighters in the face of the pandemic. Further, comparison of 3 years of ISIS attacks from 2019 to 2021 support our expectations for phase one and phase two, a reduction in violence during the immediate onset of COVID in Iraq and a steady state afterwards.
A common challenge in dealing with observational data is the danger of fitting theory to an existing set of data, rather than testing the theory against data. However, the majority of our paper, including the theory and empirical test of the initial phase, was submitted for peer-review to this journal early in the pandemic and are the same as when submitted, 2 while the fifth and seventh sections incorporate new data gathered after submission, allowing for comparison against the original hypotheses with no risk of curve fitting.
Conceptualizing slow-rolling disasters is indispensable for scholars and policy makers to assess the effects of the COVID-19 pandemic on political violence, but also promises more. As we explain in our conclusion, clarifying the theory of a slow-rolling disaster on political violence can help sharpen the boundaries between acute onset catastrophes at one extreme and generational calamities on the other, helping to elucidate the dynamics of violence related to a category of natural shocks that includes draught, famine, and insect infestations, as well as epidemics and pandemics.
Section 2: Theory
Although the impact of major disasters on political violence has long been studied, we lack easily applied theories to help answer questions about the effects, dynamics, and escalatory patterns of militant violence following slow-rolling catastrophes like COVID-19 whose human costs start as a surprise and then mount incrementally over months or several years. Important scholarship has produced studies that earthquakes and other natural disasters increase both monthly violent events and civil war onset and replicated these findings across a variety of specifications. In general, however, these studies are either limited to the study of acute onset disasters where the human effects occur extremely rapidly over the course of a day or days (Brancati 2007; Wood and Wright 2016) or find that the effect of increased violence is sharpest in the case of rapid onset calamities (Nels and Righarts 2008). In addition, there is heated academic debate about the extremely long run, decades long impact of climate change on violent conflict, although the consensus is that other non-environmental factors dominate (Barnett and Adger 2007; Bergholt and Lujala 2012; Devitt and Tol 2012). However, slow-rolling disasters differ in important ways from other types of calamities, especially related to expectations of costs which are central to many theories of political violence.
Conceptualizing Slow-Rolling Shocks
Conceptualizing “slow-rolling” natural disasters such as pandemics, epidemics, droughts, famines, and insect infestations as distinct from both acute onset and generational catastrophes is important. Slow-rolling disasters are shocks, appearing suddenly and with little warning, just as acute onset disasters like earthquakes, volcanic eruptions, floods and hurricanes and volatile environmental consequences from generational catastrophes such as climate change. However, slow-rolling disasters have costs that start low and mount incrementally at an uncertain rate over months or several years, while acute onset disasters have costs that are front-loaded to the days or weeks near the onset and generational disasters have recurring acute onset events repeated annually or seasonally at changing rates over decades.
Slow-rolling natural disasters have analytic features distinct from both acute onset or generational ones and these closely match the known features of the COVID-19 pandemic. With respect to human costs, the defining feature of a slow-rolling disaster is its bounded uncertainty. When it starts, a slow-rolling disaster produces some deaths and economic losses immediately, but the degree of the ultimate losses is unknown due to uncertainty across a range of factors – such as the lethality of the shock, the political, social and economic capacity for meaningful mitigation, and the possibility that the shock itself will become less lethal over time. A slow-rolling disaster must produce sufficient initial harm that its continuation threatens to overwhelm existing public health and safety capacity. It qualifies as a shock in the sense that its onset is a surprise and portends mounting costs that can spread geographically or increase in severity over the coming months or years. However, the actuality of the rate, boundaries of impact, and total costs are highly uncertain at the beginning, since it takes time to gather information and experience about the shock and the capacity for the impacted society and government to adapt and respond in order to clarify uncertainties about the magnitude and trajectory of losses and behavioral impacts. Accordingly, the onset of a slow-rolling disaster is sudden and a surprise and characterized by uncertainty that wanes over time as impacted societies and governments gain information and experience.
The characteristics of the COVID-19 pandemic fit the conditions of a slow-rolling shock with bounded uncertainty surrounding its consequences. Specifically, the pandemic was characterized by surprise onset, wide uncertainty about human effects based on little hard evidence in the initial 3–4 months after first deaths, declining uncertainty about the spread of the pandemic and human consequences as experience and information accumulated both about the virus and about state mitigation policies. Although far from perfect, this improving information about the consequences of the pandemic was also widely and easily available on a variety of media and Internet sources, including improving projections by reputable sources such as the World Health Organization (WHO).
Studying slow-rolling disasters offers an opportunity to improve our understanding of the dynamics of the impact of catastrophes on political violence where ultimate losses are initially uncertain and the trajectory of losses may not be uniform across societies. In abrupt onset disasters, where the vast bulk of the losses are front-loaded to the period of the immediate onset of the shock, the demand for high levels of resources is immediate and may not be possible for the society and/or government to provide an adequate measure to fully satisfy the demand. Hence, acute onset disasters have been found to predictably lead to intense, zero-sum competition among affected groups for resources that can led to violence. In slow-rolling disasters, by contrast, losses are initially uncertain and spread over time. There are also opportunities for societies and governments to intervene with policies that mitigate the harm, potentially reduce zero-sum competition over resources, and so ultimately avoid triggering competition-induced violence.
A Theory of Impact of Slow-Rolling Natural Disasters on Political Violence
Clarifying the analytic features of slow-rolling disasters allows us to create theoretical expectations about the impact of this type of disaster on political violence and generate empirical expectations about the COVID-19 pandemic.
Like many theories of political violence, our theory makes assumptions about the nature of militant groups, their ability to absorb costs, and the strategic logic of their decision making. Specifically, we share the assumptions that militant groups can be socially embedded in or elite networks largely independent from their surrounding societies, face varying constraints depending on different starting endowments, and are engaged in a complex interplay among armed organizations, rival states, and local civilians where their goal is typically to gain strategic advantages for control over territory (Arjona 2016; Kalyvas 2006; Leites and Wolf 1970; Pape 2005; Parkinson and Zaks 2018; Staniland 2017).
We also assume that the slow-rolling shock comes as a surprise to all the political actors, at roughly the same time, and produces significant uncertainty about the magnitude and trajectory of losses in the initial period that reduces as the actors gain information and experience with the shock. In particular, the uncertainty facing the political actors extends beyond simple casualty counts, rates of lethality, and the effectiveness and adoption of possible mitigation measures. The initial uncertainty also includes how the shock will alter the relative strength of these actors, both in the aggregate and in locations within the zone of conflict.
Based on these assumptions, we conceptualize slow-rolling disasters as having three phases of interest – initial onset, persistent abnormal, and stable decline – which roughly correspond with the period in which actors learn they are in crisis, adjust their operations to deal with shock-induced constraints and opportunities while the losses continue to mount, and the eventual point where they adapt to a new status quo as the losses steadily recede and the shock dissipates. These phases of a shock are illustrated in order below, showing the different progress of COVID-19 in the United States from January to the end of July 2020; while these phases are general across shocks, the length of each phase will differ, so the exact dates given should not be taken as a constant across areas suffering from COVID-19
Our theory suggests different expectations for violence in each phase.
Initial Onset
During the onset phase, actors are rapidly trying to understand the nature of shock they face. While they may have warning or pre-established plans, this period of time is best thought to represent a phase of assessment, as actors have to rapidly determine basic facts. Moreover, this is a high tempo phase, in which demands and information flow at high speed through an organization, albeit often with a high degree of uncertainty.
Initial Onset in terms of the COVID-19 Shock is the period of time where COVID-19 cases increase at an increasing rate. This is shown in Figure 1 below of the weekly fatality count of COVID-19 across time in the US, where we see a dramatic increase starting in March 2020 (Weeks 11–15). Weekly reported COVID-19 fatalities in the US, January 1 to 31 July 2020 (WHO data).
3
During this period, it was not clear to government officials that COVID-19 had established itself, nor the appropriate guidance to give. While much of this fault rests with the federal government, neither city nor state officials had a better handle on the situation. 4 During this initial period, some hot spots such as the Northeast States experienced high infections and losses and adopted serious mitigation programs, while others such as Southern states did not, and so over time important information could be gained about both the virus and mitigation efforts.
Although the curve in Figure 1 above shows a decline at the end of the period, it is probably best to conceptualize the entire period from February when the first confirmed US COVID-19 death occurs through July as the relevant time-frame for phase one. According to our theory, what matters in phase one is not the exact short-term rise and fall of deaths. Rather, phase one is about gaining sufficient information and experience with the shock to enable meaningful assessments about the likely trajectory of the shock and impact and likelihood of potential mitigation measures. In the case of the US, the period of February through July provided substantial information for public health officials in countries around the world to establish firmer understandings of the lethality of the virus, its ease of transmission, and other factors important in making more reliable projections about its trajectory in subsequent months. 5
It does not take much imagination to see that these dynamics are similar for other actors facing the onset of a major, but slow rolling crisis. In many cases, despite what may feel like a time of frantic and decisive action, the main actions taken by an organization during this initial time period is simply identifying the presence and scope of the crisis. We believe that during this period the assessment of disruption on current operations and existing plans take precedence to continuing operations. While decisions made during this period are likely to be critical junctures in determining what happens later in a shock, during this phase our expectation would be that most armed organizations will focus on assessing the situation, rather than being in a position to exploit its impact on their opponents.
Accordingly, phase one – Initial Onset – is likely to be a period of relative decline in violence, as actors try to assess the effects of the slow-rolling shock on their organization and their opponents. Although the exact degree of decline will depend on their starting endowments, in the face of uncertainty about the immediate future, militant groups will aim to maintain some already on-going operations but hesitate to initiate new plots, schemes, or campaigns. Instead, they will take steps to husband resources, protect capacities, and assess the new situation by gathering information both about the nature of the shock and its impact on their rivals and civilians. Hence, in phase one we expect declines of militant violence even in the face of apparent opportunities due to short-run weakness in rival capacities or opportunities to gain support from the civilian population.
Persistent Abnormal
During the Persistent Abnormal phase, actors are forced to adjust their activities to the constraints imposed by the crisis. While the organizations in question are what Weber would term ‘rational-bureaucratic’, in that they have and are capable of adjusting their operations, during this period they will be forced to adjust their old ways of operating with the constraint imposed by the shock.
In terms of the COVID-19 Shock, “persistent abnormal” refers to the period of time when there is still an elevated number of cases, yet they are increasing at a decreasing rate or stabilizing at broadly predictable rates. Importantly, there may be changes in the pattern of virus cases, with small rises and declines, but in this phase, these changes are now occurring as predictable patterns related to the virus and adoption (or not) of mitigation policies. For COVID-19, this phase began around July 2020, when public health officials had sufficient evidence and experience to reliably project that the virus would not disappear on its own, might diminish in warmer weather, but would likely return with more virulence in the fall and continue to spread widely until effective vaccines would become available in 2021.
This phase is similar to longer running shocks, such as a prolonged drought or famine, where actors have more time to make decisions and adapt, yet have less flexibility as the effects of the shock hit home. For example, organizations may struggle to mobilize individuals to replace those lost to illness, to maintain spending as sources of income dry up, and individuals may struggle with economic and social dislocation caused in the first phase. While we believe during this phase that organizations will re-organize their operations, the expectation is that they will still remain wedded to their pre-shock goals, and that the process of adapting their operations to the new environment will be challenging.
In an ongoing conflict, we would expect that precise levels of violence are a function of the degree of arming and recruitment, impacted locally by the shock on local wages, resources available to rivals, nationalist sentiments, and organizational ties (Blattman and Miguel 2010; Kalyvas 2006; Pape 2005; Petersen 2001; Staniland 2014). These common drivers of local violence would be bounded by constrained expectations and more informed calculations about the potential that violent operations could increase the consequences of the slow-rolling disaster for a group and its supporters.
Thus, phase 2 – Persistent Abnormal – is likely a period of a relative steady state with modest increases or decreases in violence compared to the prior period. Actors make planned adjustments to their operations based on assessments of the effects of the slow-rolling shock on their organization and their opponents. As uncertainty about the immediate future is replaced with growing certainty about the patterns in how the shock is spreading, its concrete human effects, and actual propensity of mitigation to be adopted and prove effective, militant groups will continue operations that have proven successful or low risk, but also continue to adopt restraint based on their growing knowledge about the firmness of constraints, even in the face of apparent opportunities.
Further, the potential for resource-competition-induced violence remains muted during this period, because during this phase of a slow-rolling disaster, costs are continuing to mount and spread and so policies that might normally be expected to increase the relative power of a militant group – such as seizure and control of local populations – in fact risk over-stretching the capacities of the group to govern any area. Regarding the pandemic, just as states and societies are learning to “live with the virus,” so too are militant groups learning to “fight and kill with the virus.” Hence, phase two expects roughly steady states of militant violence, possibly including modest changes over time, but at rates lower than pre-shock periods.
Stable Decline
The final phase of interest in thinking about the effects of shocks is Stable Decline. While organizations have adjusted their behaviors during the steady state period, this period represents the organizations permanent adaptation to the effects of the shock and recognition that the main consequences of the shock are receding in the foreseeable future. Of course, as with businesses and government agencies, some militant organizations will not have been able to survive to reach phase three even with the best assessments and adjustments in the previous phases due to limited initial endowments. However, surviving groups will now face the prospect of recovery – not only their own recovery, but the recovery of their state rivals and local populations. In many cases, this may involve an organization altering goals to match with the new realities of their changed organizational capacity or the growth of new organizations, a process akin to Schumpeter’s ‘creative destruction’ or theories of punctuated equilibrium in evolution biology, where shocks to a system create room for new growth (Eldredge and Gould 1972; Schumpeter and Swedberg 2005).
While in the future, a period of stable decline for the COVID-19 pandemic would likely occur following effective widespread vaccination campaigns that provide long-lasting immunity to large portions of impacted populations. 6 In this period, likely lasting many months, most societal functions would return incrementally over time, and in many cases organizations are adapting not only their immediate organizational routines, but making longer term decisions about what they and their rivals would look like based on these changes. This phase is similar to a rebound in economies following a severe depression or the rebuilding stage after a natural disaster, when actors are likely to make major adjustments to their goals, and we would expect to find both the death and birth of armed organizations.
Phase three of the shock – Stable Decline – is likely a period of increasing violence, through multiple possible pathways. Surviving militant groups who pre-existed the shock and have now successfully managed the assessment and adjustment phases may face rival states who are still severely weakened and so less capable of intelligence and counter-measures, reducing constraints on militant violence. Further, surviving groups in this situation have shown themselves to be resilient and agile, and thus likely to take advantage of windows of opportunity to make gains versus possibly still weakened state rivals and incentivized to act before these windows close. In addition, as immediate dangers subside, competition over resources for recovery and rebuilding may increase comparable to the aftermath of acute onset disasters, triggering competition-induced violence. Finally, new militant groups may also emerge, in response to disruptions to society, and now face opportunities and incentives akin to surviving groups that may increase violence, either against state or militant group rivals. Given the multiple possible pathways for violence to increase, phase three may experience rising levels of violence surpassing pre-shock periods.
Following these three phases would come a return to what could be considered normal or a new normal, but our theory does not include this period.
Section 3: Research Design and Data
Available data through the end of December 2021 provides an opportunity to test the first and second stages of our theory – initial onset, the period of time when COVID-19 appeared as a sudden surprise, spreading to many countries in the world and when infections and deaths increase at an increasing rate and persistent abnormal, the period when there is still an elevated number of cases, but they are changing or stable at broadly predictable rates.
During the first period of initial onset, our theory of slow-rolling disaster expects a decline in militant violence rather than an increase of violence expected by acute onset theories, since militant groups facing this type of shock will emphasize force conservation, rather than exploiting new opportunities for violence, until they gain more information and understanding about the actual impact of the shock. During the second period of persistent abnormal, our theory expects a relative steady state with modest increases or decreases in violence compared to the prior period as militant organizations re-organize their operations under still challenging circumstances.
Employing both statistical analysis on the relationship between COVID-19 and militant violence around the world as well as a single case study of militant activities in Iraq, we find credible empirical support for our theory. Specifically, the quantitative evidence shows significant correlations consistent with our theory and inconsistent with acute onset theories, using variation in the onset of Covid-19 in countries, and fixed effects to control other potential confounders that might otherwise account for the relationships. The Iraq study provides evidence in an important case about the validity of our assumptions and proposed causal mechanism, and rules out alternative causal pathways. Of course, each component of our mixed method approach has inherent limitations, raising challenges to inference. Taken together, we believe that these limitations offset, and show the value of thinking about COVID-19 as a slow rolling natural disaster.
The quantitative evidence estimates the relationship between incidents tracked in the ACLED dataset and COVID-19 cases and fatalities reported by the WHO, from 1 January 2020 to 31 December 2021 7 . While ACLED data does not cover all countries, it has been used in other cross-national studies on militancy, and includes a wide geographic mix of countries (ACLED 2017; Raleigh et al. 2010). ACLED data provides systematic insight into many of the most lethal cases, including Afghanistan, Colombia, the Democratic Republic of Congo, Iraq, Ukraine, and Yemen. While COVID-19 reporting is undoubtedly incomplete – an issue we will address below--WHO is the paramount global public health organization, providing the best available global reporting on the virus (World Health Organization 2020).
Since our data is observational, an immediate concern for inference is the quality of COVID-19 data and the t non-random assignment of this shock to countries. To address these concerns, our statistical analysis looks at the difference in incidents in similar time periods across relevant points of disjuncture: for phase one, 3 weeks before and 3 weeks after a country’s first COVID-19 fatality, and for phase 2, 52 weeks before and after a country’s first Covid-19 reported fatality. As was noted above, we do not believe that we have entered phase 3, the continuous decline of COVID as of the end of 2021 (When this was written), and our predictions on phase 3 remain untested. Although we agree that reporting of COVID-19 data is assumed as generally unreliable, reporting on the timing of the first death reduces this uncertainty, due to its inherent salience and interest by a variety of actors in updating the exact timing after these events. Further, although public policy and state capacity have important effects on the spread and lethality of COVID-19, and so COVID-19 deaths in general are not randomly assigned, we believe that the timing of the initial fatality carries with it an element of chance. This sets up a natural discontinuity, allowing for an estimation of the initial effects of COVID-19 and militant violence that minimizes concerns about selection effects. Combined with fixed effects which minimize the impact of country and seasonal specific trends on violence, this produces a plausible estimate of the relationship between the onset of COVID-19 and violence.
The Iraqi case was selected due to its importance, given ISIS’ central role in organizing trans-national militancy and the large multi-national investment of time and effort to its defeat. We use the case study to show observations of the causal process of ISIS activities after the onset of COVID-19 which are consistent with those proposed by our theory, as well as comparing these observations against alternatives.
The Iraq case study is based both on available secondary sources such as ACLED, as well as analysis of Arabic language propaganda materials from ISIS. ISIS materials were gathered and analyzed by Arabic speaking analysts as part of ongoing research into militant propaganda by the Chicago Project on Security and Threats (CPOST), which pre-dates the COVID-19 epidemic. The analysis includes all publicly available print and video products from ISIS, covering the period of January to 1 August 2020, although comparison is also made to ISIS propaganda from the same period in 2019. Secondary sources are taken from a mix of major English language organizations and analysts.
Our expectations are that in both the quantitative and qualitative evidence we should see a reduction in militant activity in the onset phase of COVID-19, as militant organizations pause their activities to assess impacts. In contrast, predictions from theories of sudden-onset disasters are that militant activities should increase as they face greater competition for scarce resources or seek to take advantage of windows of opportunity against a weakened opponent.
Section 4: Phase One Quantitative Evidence
Figure 2 below shows a monthly comparison between violent events catalogued in the ACLED dataset in 2019 and 2020. On a global level, event counts are similar through January and February of both years, decrease from March through May in 2020 compared to 2019, before somewhat rebounding in June and July. As the trend lines in Figure 2 show, there is a puzzling drop in violence as measured by ACLED during the onset of COVID-19, a finding at odds with the expectations of works predicting more violence as expected by sudden-onset models (United Nations Counter-Terrorism Committee 2020). Comparison, global count of ACLED events by month, 2019 and 2020.
While this is a surprising pattern, there are challenges to inference. First, given that a small number of countries account for most violent activity, it is possible that changes in single cases, such as the drop in violence as a result of the Afghan Peace Process, account for the observed global decrease. Second, it seems more than possible that the types of problems which lead a state to a higher level of COVID-19 cases and fatalities -- inability to conduct epidemiological surveillance, to enforce public health edicts, or unwillingness to ensure equal treatment across groups -- are those which are most likely associated with a high level of militancy. Third, there are serious worries about the accuracy of available data on COVID-19 cases and fatalities, and errors in measurement are unlikely to be randomly distributed, either across cases or time, potentially biasing estimated effects. 8
To address these concerns, we use differences in the onset of COVID-19 across cases as a discontinuity. We strongly believe that public policy and state capacity have important effects on the spread and lethality of COVID-19, and thus any claim to random assignment of COVID-19 fatalities or cases is untenable. However, even with the best policies, we believe that the timing of the initial fatality carries with it an element of chance. This sets up a natural discontinuity, allowing for an estimation of the initial effects of COVID-19 and militant violence that minimizes these concerns.
We find it hard to think of any particular incentive for national leaders to push the announcement of the first fatality from 1 week to the next, a sorting process which might confound our results. Moreover, given the novelty of the virus, we believe that the first reported fatality seems less likely to be misreported than subsequent counts. Figure 3 below provides the distribution of total weekly ACLED Events, by country, for the 3 weeks before and after the reporting of the first COVID-19 fatality (Week 0), as well as the linear trend in violence before the onset of COVID (Dashed Line), and a smoothed trendline for the average number of weekly events (Solid line). While we cannot prove that leaders did not manipulate the reporting of a country’s first COVID fatality, we do not see obvious patterns of sorting around the cut-point which would bias estimation and do see a drop in violence after the onset of COVID-19. Count of ACLED events, by country, 3 weeks prior to 3 weeks post Country’s first COVID Fatality.
We leverage this discontinuity to estimate the relationship of the onset of COVID-19 to militant violence by assigning an indicator variable for the weeks falling after a country’s first reported fatality. Using a narrow bandwidth, we estimate the relationship of this indicator variable to the number of events reported in ACLED, without having to rely on potentially incomplete or biased reporting of COVID-19 cases which could affect our results. In all estimates, we include fixed effects for country, to control for country-specific variables such as income or regime type which might affect the level of violence, but vary little in the short term. We also include fixed effects for calendar week, to control for cross-national time trends which might affect the number of reported incidents, such as religious holidays or seasonal effects which might raise or lower violence at a particular point in time. Taken together, these provide an estimate of the change in reported incidents after the onset of COVID-19 which minimizes potential confounders.
While Figure 3 shows a negative relationship between violent events with the onset of COVID-19, one objection would be that the relationship is driven by ACLED’s inclusive coding of incidents, rather than a true decline in militant violence. For example, Mexico and Brazil account for a substantial number of incidents in the ACLED data, and violence in these countries may be driven by criminal, rather than political concerns (Lessing 2015). The data above also captures numerous protests, a type of contentious politics that may follow different patterns than violent activities by militant organizations (Chenoweth and Stephan 2013).
To address these concerns, we run our analyses using a subset of data which contains only countries coded as civil wars by UCDP-PRIO and only incidents reported as ‘Explosions/Remote Violence’ in ACLED. 9 While potentially missing some acts of militant violence, the estimated relationship in this narrow subset seems unlikely to include non-militant activities.
Estimated difference in ACLED explosive events in UCDP civil war cases in 3 weeks following a country’s first COVID fatality.
+ p < 0.1; *p < 0.05; **p < 0.01. Robust Standard Errors (HC1) Clustered at Country, OLS models.
As can be seen in Table 1, the relationship of COVID-19 onset and violence is consistently negative, and in six of the eight models, statistically significant. Models LM1, LM2, NB3, and NB4 look at the difference between weeks before and after a country has suffered its first COVID fatality. While the OLS model is not statistically significant, the sign is in the correct direction, and the negative binomial models show a negative and statistically significant change in explosive events.
Models LM3, LM4, NB1, and NB2 look at the impact of suffering a COVID fatality in one of the 3 weeks following when a country first suffered its first COVID fatality. We believe this is the better measure, as it captures the consistent appearance of COVID fatalities associated with phase 1, onset of a natural disaster. In all of these models, the relationship between COVID and explosive events are negative, and statistically significant, even when including country fixed effects and controls for the prior week’s number of events, to capture time trends. While the substantive effects vary between models, they are not trivial given that the average ACLED explosive events leads to 11 fatalities.
In summary, the evidence above shows a negative correlation between the onset of COVID-19 fatalities and militant violence. These results fit with our prediction that there is likely to be a lull in violence during the onset of a slow-rolling disaster like COVID-19, rather than the predictions of those focused on sharp-onset disasters. While the data is observational, we leverage variation in the onset of COVID-19, as well as fixed effects to control for potential confounders which might otherwise explain this relationship.
Section 5: Phase Two -- Re-visiting the Argument with New Data
As was noted in the introduction, the previous tests were performed using available data up to August of 2020. While a common concern with observational data is that analysts fit their theory to explain available data, the continued pandemic provides us the opportunity to test our theory against new data that was unavailable when this article was first drafted. This approach helps deal with concerns of curve fitting or publication bias that arise from observational studies. Below, we extend the period of analysis through the first year of a country’s COVID experience (Phase 2), providing an overview of further trends in the data.
To reiterate, our original theoretical predictions for phase 2 are roughly steady states of militant violence, possibly including modest changes over time, but at rates lower than pre-shock periods. We use the 52 weeks after a country’s first COVID fatality as a measure of phase 2, as we believe in the case of COVID this captures a period during which vaccinations were not widespread, and COVID cases and fatalities remained elevated. Figure 4 below shows the weekly number of explosive events reported by ACLED for each country coded by UCDP as being in Civil War for the year before and after the country reported its first COVID fatality. While the variation in levels of violence between countries creates wide bounds in the confidence intervals, we do see a generalized decline in the global average number of these events in the year after a country reported its first covid fatality. Moreover, we see a drop in the highest level of violence, as shown by the decrease of observations in the top right of Figure 4. Figure 4 does not see a quick return to pre-covid patterns of violence, nor a generalized escalation of attacks after COVID, as predicted by pessimistic theories on the effects of COVID on militancy. Weekly explosive incidents as recorded in ACLED by country, along with global trend (black line) and 95% confidence interval (dashed lines) for 52 weeks before and after a country reports its first COVID fatality.
In general, this pattern of violence fits with our prediction, that the initial onset of COVID is likely to lead to a lower level of activity, and that this lower level will be sustained in a second, medium term period of persistent abnormal, taken here to be the first year of the pandemic, where vaccines were not available and there were limits to groups’ ability to adapt.
Section 6: ISIS Response to COVID-19 Onset in Iraq
While the statistical evidence above shows a drop in militant violence during the onset of COVID-19, it relies like all statistical evidence on a number of assumptions. To strengthen these findings, we provide evidence on the Islamic State of Iraq’s activities across the onset of COVID-19 in Iraq. The case is of critical interest, given the group’s lethality, central role in organizing trans-national violence, and the resources and number of countries committed to combatting them.
We first present an overview of the spread of COVID-19 in Iraq, its effects on the Iraqi government, and the public knowledge of these effects, the starting point in the causal process proposed by both sudden-onset disaster theories and our own. The second section, using ISIS propaganda as a window into their decision making and priorities, shows that the group has not ramped up activities as would be expected by sudden onset disaster theories, but instead largely focused on protecting their own fighters, as predicted by our theory. The final section examines the possibility that COVID-19 may have reduced reporting on ISIS, but not their activities, and finds little evidence of this alternative explanation.
COVID-19 in Iraq
The spread of COVID-19 in Iraq started in March of 2020, with deaths in Baghdad and the Iraqi Kurdistan Region. 10 These prompted quick government intervention, in the form of border closings, restrictions on public gatherings, and lockdowns of business activity. 11
While there is some uncertainty as to exact figures, Iraq has certainly seen wide-spread community transmission of the COVID-19 virus. Based on WHO data, Iraq had reported approximately 300,000 cases of COVID-19, of whom over 8000 have died by fall 2020 (World Health Organization 2020). While there is reason to believe that these numbers do not account for all COVID-19 cases or fatalities, they show the virus is widespread, with approximately 0.5-1% of Iraq’s estimated population having had the disease. Moreover, Iraq’s COVID-19 case fatality rate of ∼2.6%, is not out of line with other countries, implying that while there may be underreporting, Iraq is not an outlier. 12
Available evidence is that the Iraqi medical system has been largely overwhelmed by the challenges of COVID-19. In the words of one Iraqi government official, “This is a war against the coronavirus and we have lost the war.” 13 Across Iraq, there was looting and violence as individuals sought oxygen for treatment, and these incidents were not isolated. 14 Economic damage, largely from the collapse of the global oil market as a result of the pandemic led the new Prime Minister to state on a public relations visit to Mosul that “We are witnessing the worst economic situation since the formation of the Iraqi state.” 15
Public cynicism and disenchantment with the government’s response is not without cause. Prior to the onset of COVID-19, major demonstrations in Iraq continued over corruption and poor governance, despite the promise of major reform and the resignation of the prime minister in November 2019. 16 It took 6 months, until May of 2020, for the Iraqi parliament to settle on a replacement for their caretaker government, and even after the announcement of a new prime minister, key positions remain unfilled. 17 As with most post-2003 Iraqi governments, appointments appear as much a matter of appeasing powerful insiders as they are about providing effective government services and representation. 18
The spread of COVID-19 in the Iraqi case would appear to meet the conditions of both theories. COVID-19 quickly overwhelmed Iraqi local medical capacity, and is a major and unexpected non-human source of morbidity. As a result of both local and global actions to combat COVID-19, there were lockdowns, a decrease in economic activity, and a subsequent weakening of the Iraqi state, all of which were public knowledge. Given ISIS’ history of exploiting weakness, this would appear to be an easy case for theories which expect COVID-19 to lead to increased militant violence (Ackerman and Peterson 2020; United Nations Counter-Terrorism Committee 2020; Wood and Wright 2020).
ISIS’ Reaction to COVID-19 in Iraq
While there are limits to what evidence can be gathered contemporaneously from a clandestine organize like ISIS, we can compare their public facing statements and propaganda against past patterns to look for changes. Existing theories on sharp onset disasters would expect an increase in ISIS activity in Iraq after the onset of COVID-19, taking advantage of governmental weakness, public dissatisfaction, and the social effects of lockdowns. In contrast, our expectations are that during the onset period of COVID-19, ISIS activity would be subdued as they assess the impacts to their organization.
Figure 5 below gives the number of weekly incidents reported by ACLED from January through July 2020 in which ISIS was a party. There is a spike in violence coinciding with Ramadan in May 2020 (Weeks 18–21), but in the majority of weeks ISIS violence is lower in 2020 than 2019, and there is a decreasing trend line of reported activities from this point. Weekly count of ACLED incidents reporting ISIS as an actor, 1 January - 31 July 2019 and 2020, by Calendar Week.
Figure 6 below shows the trend in reported ISIS incidents and COVID-19 fatalities in Iraq by week, to highlight this trend. While the pattern in the previous chart was mixed, there is a clear drop in ISIS incidents as weekly COVID-19 fatalities spike. Weekly ACLED Events involving ISIS, Weekly WHO Reported COVID-19 Fatalities in 100s for Iraq, By Calendar Week, 1 January - 31 July.
Another approach to understanding ISIS reaction to COVID-19 is through the volume and content of their public facing propaganda. While the statements and themes made through propaganda should not be taken uncritically, it would seem reasonable to expect that bias in these statements to be in a direction which overrepresents ISIS strength and capability, and these biases would have existed before the onset of COVID-19.
Figure 7 below shows there is a drop in the combined number of videos released by ISIS’ Amaq and Wilyat propaganda arms in 2020 compared to 2019. While one interpretation might be that this is due to setbacks suffered by the group, particularly the October 2019 death of ISIS Emir Abu Omar Al-Baghdadi, the central media arm produced either the same or more videos during the first 4 months of 2020, and the drop in video activity occurs in April and continues through July, as COVID-19 continued to spread in Iraq, and slightly ahead of the continued drop in ISIS incidents. Monthly ISIS video releases, January - August 2019 and 2020.
In addition to examining the quantity of ISIS propaganda, we can also look at how the content of their propaganda changes with the onset of COVID-19. The data presented below is taken from an analysis of all issues of ISIS’ Al-Naba propaganda magazine in 2020.
In initial content released at the start of 2020, ISIS propaganda only mentioned the virus was a form of divine retribution, as the initial outbreak in China was linked to State repression against minority Muslim Uighers. 19 In both their print and video propaganda before the onset of COVID-19, ISIS emphasized their operational ability to carry out attacks, and reiterated their strategic commitment to using violence to gain their political goals.
However, in mid-March, after the first COVID-19 death in Iraq, they published ‘Shariah-based Instructions for Dealing with Pandemics’. 20 Presented in its original color version in the supplementary materials, the one-page article provides advice on how to avoid pandemics which is not dissimilar to advice from other public health groups. While couched in the terms of the Hadith, the sayings of the Prophet Mohammed, the article includes common suggestions such as ‘For the healthy not to enter places hit by the pandemic and for those already infected not to leave’ and ‘To cover the mouth when yawning and sneezing’. 21 This change in tone regarding COVID-19, from one emphasizing divine retribution to one providing basic public health advice to followers seems to indicate that the group is taking the possibility of infection of cadres seriously, and that the initial phase of COVID-19 has motivated them to focus on protecting their force, rather than expanding operations.
Looking at their materials, whether video or print, shows a limited call to take advantage of the COVID-19 pandemic. Of the 22 articles ISIS released during the initial pandemic period, only one calls for ISIS to take advantage of the pandemic to carry out attacks. This release occurred on March 19, at the same time that they give instructions to their followers on how to avoid the pandemic. 22 In it, they note the weak security of an Iraqi prison and the decrease in US Forces in the region, and that this represents a prime opportunity to release imprisoned members. Despite their clear articulation that COVID-19 has given them a window of opportunity, as of this writing we are unaware of any ISIS efforts to carry out this plan through November 2021.
While the group’s propaganda still emphasizes their capabilities and dedication to continued violence, their most recent major call to action asks for possible western recruits to stay home. While the group blames the actions of Western states for making travel harder to justify this call, there are no major changes that we are aware of in terms of travel restrictions aimed at militants, only the reduction in flights and imposition of bans and quarantines as a result of COVID-19. 23 Other dedicated news releases on COVID-19 since March have emphasized themes of divine retribution, rather than a theme which calls for attacks given the weakness of outsiders.
Our interpretation of the drop in frequency of propaganda, as well as the change in tone, is that the group is not focused on expanding their operations as a result of COVID-19 during the initial onset of COVID in Iraq. Their public health advice shows concern about the potential impact of the pandemic on their own capacity, and the drop in both the amount of propaganda and in specific calls to action implies they are limiting their operations. This interpretation is also consistent with commentary from others who argue that the onset of COVID-19 may reduce, rather than raise, trans-national violence in the Middle-East (Pollock 2020).
Alternative Explanation: Drop in Reporting
While we believe that the slow-rolling onset of the COVID-19 pandemic explains why ISIS has not expanded its activities, there is one obvious alternative explanation which might explain the lull in ISIS activities without invalidating the predictions of sharp onset theories. Given that the ACLED dataset is based on public reporting, it could be the case that COVID-19 has led to reduced reporting of attacks, rather than a reduction in attacks. This could be due to an increased focus by news organizations on COVID-19 at the expense of other news, or by a reduced ability by journalists to meet with sources and cover stories. While major news organizations observably continue to report from Iraq despite COVID-19, this is a concern worth taking seriously. 24
While we cannot show there is not a growth in unobserved violence, the sources underlying the 2020 incidents in the ACLED dataset continue to be a mix of international, national, regional, and local sources. In supplementary materials, we provide the results of testing three implications of the argument that COVID-19 will lead to a decrease in reporting, rather than violence, and find the opposite relationship in one case, and no support for the argument in two others. Taken together with existing articles showing the presence of reporters, we believe that the available evidence does not support an argument that the observed drop in violence is due to a drop in reporting.
We believe that the Iraqi case provides support for our argument that during the initial stages of a slow-onset natural disaster that militants reduce activities as part of a process of evaluation and husbanding their resources. The Iraqi government was weakened by the initial onset of COVID-19, with a decrease in available revenue and public discontent over their handling of the crisis. Based on their propaganda, ISIS recognized this weakness, and based on outside assessment, were capable of exploiting it. 25 However, since early March, their messaging on COVID-19 has provided information for self-protection, 26 and a July publication asked potential recruits to stay abroad. 27 The frequency with which they produce propaganda and carry out attacks is not significantly different than in 2019, and both decreased on a monthly basis with the skyrocketing of Iraqi COVID-19 caseloads. Even in a case where they identified a now inviting target of key strategic importance in their propaganda, they have failed to act. Based on this evidence, we believe that the predictions of sharp-onset disaster theories are less supported by available evidence, increasing the plausibility of our theory.
Section 7: ISIS Violence in Iraq in Phase 2 of COVID
While the two sections above were written using data available as of August of 2020, Figure 8 below extends this comparison of weekly ISIS violent activity to the end of 2021. While there is a trend towards increased ISIS activity in 2020 over 2019 for the last 3 months of 2020 (weeks 37–50), this pattern occurs nearly 6 months after Iraq’s first COVID fatality (Week 10 of 2020), and comparable to the period just after the death of the group’s leader in 2019 to 2020. Weekly ISIS attacks reported by ACLED, 2019-2021.
The average weekly number of ISIS incidents in Iraq reported in ACLED was 20 in 2019, 20.8 in 2020, and 21.23 in 2021, with a slight up-tick across time. While not wanting to downplay this increase, the standard error of weekly attacks in 2019 was approximately 1 attack per week, and 2019 attack totals are higher or tied with weekly event counts in 2020 and 2021 for 20 of the 53 weeks compared.
While not as strongly supportive of our predictions as the data originally analyzed, we still believe that the comparison across 3 years of ISIS attacks in Iraq supports our interpretation of a drop in activity during the immediate onset of COVID in Iraq and a steady state afterwards, as ISIS and the Iraqi state continued to adjust to a world in which COVID spread and fatalities were consistently high.
Section 8: Addendum
Summary of predictions.
While this paper has focused on both introducing and providing a tentative explanation for what we regard as a puzzling initial drop in militant violence during the initial onset of COVID, there are areas for further research. First, the exact mechanisms of the initial drop in violence may be different from what we suggest above, as perhaps militants focus on winning over local populations with civic action during the initial onset. Second, while we present qualitative evidence, it is only from one case, and mechanisms in this case may not be present with other militant groups or in other countries. As such, we provide further information on the change in violence in four other cases in the supplementary materials to show how these trends unfolded elsewhere, but feel that further qualitative investigation would help show the limits of our theory. Lastly, while we differentiate COVID as a slow-rolling crisis, there are reasons to believe that the phases of assessment and adjustment operate in sudden onset crises, although are often after most costs already exist and possibly in compressed time periods. Further investigation of this would help to bridge extant findings on the relation of natural disasters to violence in the political science literature with sociological findings on disaster, which find limited evidence of crime or violence in the immediate aftermath (Quarantelli and Dynes 1977).
Section 9: Conclusion
We have argued that the nature of the COVID-19 pandemic is different from those of other types of natural disasters. COVID-19 is a slow rolling disaster, as its effects unfold across months, rather than days or weeks associated with other natural disasters such as floods or earthquakes. We posit that given this difference in the speed of onset, armed actors focus their activities during the onset of the disaster in husbanding resources and assessing the magnitude of damage, rather than increasing their competition for scarce resources.
This prediction fits with what we saw as a puzzling trend in militant activities reported in prominent militant violence datasets. Using the inherent unpredictability in the timing of a nation’s first COVID-19 fatality, we find that the correlation between COVID-19 and militant violence is at least initially negative. Looking deeper at the critical case of Iraq, we show that despite the presence of causal pathways which should increase violence in sharp-onset theories of disaster, ISIS appears to initially limit their activities, in line with our predictions. While the evidence we present is not definitive, it explains both drops in observed activity during the initial onset of COVID-19, as well as why other studies find increases in violence after sharp-onset disasters.
While this paper focuses on the drop in violence during the onset period, it is important to stress that once armed groups have fully adapted to the post-COVID-19 world, our expectation is that violence will increase. While we hope that we will not observe this rise, we believe that the decline in militant violence represents one of the few silver linings to the public health crisis of the last 2 years. While there will be many demands on the international community as the world returns to normal with the end of COVID, we believe that one of them may well be a rise in the level of militant violence.
Policy makers should take advantage of the temporary lull in violence, by doing all in their power to slow the spread of COVID-19. Indeed, the West African Ebola crisis of 2014-2016 shows that with cooperation and aggressive efforts to ensure public health, it is possible that pandemics need not lead to a long-term increase in violence. However, the longer the pandemic drags on and the higher the human and economic toll, the more likely that we will see a lengthy period of increased violence as armed actors adapt to the ‘new normal’, an outcome we hope policy makers will avoid.
Supplemental Material
Supplemental Material - “A Slow-Rolling Disaster: Assessing the Impact of the Covid-19 Pandemic on Militant Violence”
Supplemental Material for “A Slow-Rolling Disaster: Assessing the Impact of the Covid-19 Pandemic on Militant Violence” by Robert Pape and Christopher Price in Journal of Conflict Resolution
Supplemental Material
Supplemental Material - “A Slow-Rolling Disaster: Assessing the Impact of the Covid-19 Pandemic on Militant Violence”
Supplemental Material for “A Slow-Rolling Disaster: Assessing the Impact of the Covid-19 Pandemic on Militant Violence” by Robert Pape and Christopher Price in Journal of Conflict Resolution
Footnotes
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
We gratefully acknowledge the research support from Dr Ali Al Nuami and the University of Chicago's College Research Fellows Program (CRFP), the Institute of Politics, and Division of the Social Sciences.
Supplemental Material
Supplemental material for this article is available online.
Notes
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
