Introduction
Since the late 1990s there has been a sharp increase in attacks on aid workers, and the humanitarian community faces new challenges in responding to, and mitigating the risks of, such violence (Fast 2014). In the same period, the aid community has experienced a well-documented process of securitization, whereby aid has become more closely linked to security concerns for myriad reasons (Brown and Grävingholt 2016; Duffield 2001, 2012). The securitization of aid has resulted in development and humanitarian assistance that is explicitly designed to increase the security of both donor and recipient countries, and is delivered in tandem with non-humanitarian and sometimes military actors. This move has seen aid operations extended into conflict-affected areas where they may not have entered previously.
Earlier research has highlighted this shifting context as a contributing factor to the incidence of violence against aid workers and organizations, and case studies and incident reports show how securitized aid appears to make aid workers more vulnerable (Fast 2014). Such literature is geared largely to the policy and development practitioner audience, and the limited academic research on the subject has mostly used descriptive statistics to illustrate recent trends and describe various incidents (Fast 2010, 2014). To date, there has been no macro-level investigation into whether there is indeed a definitive link between the securitization of the global aid agenda and the rising incidence of violence against aid workers. This gap in the literature is what this paper seeks to address. And if securitized aid is a key reason why aid workers are becoming the targets of violent attacks, then understanding this risk can help aid organizations to more accurately develop the tools needed to mitigate risk and better protect the personnel they have working in those environments.
To analyse this relationship, our study considers theoretically how the securitization of aid might contribute to fostering an environment that enables such violence. We then analyse data on securitized foreign aid flows and aid worker violence for the period 1997 to 2019 to give a quantitative answer to our research question. We further hypothesise that there is a connection between the securitization of aid and violence committed against aid workers because this form of aid alters the environment for aid work in three key ways: (1) it introduces aid to a changing and more complex political context; (2) it prompts increasingly risky aid worker behaviour; and (3) it provides opportunity to those groups and individuals who might attempt to benefit from violent acts against aid workers and organizations.
In the sections that follow we first introduce some existing literature on aid securitization and violence against aid workers and establish a framework for understanding the determinants of this form of violence. We then describe our data and analysis and present our findings. We conclude with a discussion on the implication of these findings.
Violence Against Aid Workers and the Securitization Of Aid
Aid Worker Violence on the Rise
Violence against aid workers and loss of life in the line of duty is of serious concern for humanitarian agencies and both donor and recipient governments. This violence targets humanitarian and development workers, expatriate and local, and poses a challenge to the philosophies that motivate humanitarian actions. Current research suggests that there has been an escalation of violence against aid workers over the past two decades (Stoddard et al. 2006, 2009). This trend is a concern to the aid community as it affects many aspects of aid operations (Stoddard et al. 2009).
Research shows that the number of violent incidents against aid workers increased notably in the first decade of the 2000s (Fast 2010). The rate of attacks per aid worker was estimated as 61 per cent higher in 2008 than in any of the 12 previous years (Stoddard et al. 2009). According to the Aid Worker Security Dataset, 2023 established a new high for violence against civilian aid operations, with 240 separate attacks affecting a record 538 aid workers. Of these 538 victims, 240 were killed, 196 were wounded, and 78 were kidnapped (Humanitarian Outcomes 2024). Indeed, the AWSD indicates that more than 450 aid workers have been affected by such incidents annually since 2019, marking a new period of intensification of such violence.
The Securitization of Aid
Violence against aid workers comes in many forms, including kidnapping, physical and sexual abuse, threats and intimidation, emotional and social abuse, and murder. While there have been high profile incidents of violence against aid workers perpetrated by national militaries, in most cases the violence is perpetrated by those who are, or who have been, in positions of powerlessness: members of marginalized groups, criminals, terrorists or other non-state actors (Clapham 2006). Not surprisingly, there are various determinants at the micro- and macro-levels that have been linked to this phenomenon. The securitization of humanitarian aid constitutes the most macro-level contextual factor.
Securitization is a process by which subjects and processes are reframed by relevant actors as matters of security (Wæver 1995). In the case of foreign aid, this securitization process has been occurring since the 1990s, but has advanced significantly in the last two decades (Brown and Grävingholt 2016; Swiss 2011). The securitization of aid can be considered in two ways: first, the use of development assistance and humanitarian aid as a tool to increase the security of both donor and recipient countries; and second, the provision of aid in increasingly insecure environments. Both forms of securitization have been evident in the bilateral aid context in recent years, with aid dollars increasingly allocated to security aims and with larger portions of aid budgets being delivered in insecure contexts (Swiss 2011, 2018).
Petřík (2008) argues that the securitization of aid is perceived in both positive and negative terms. On the one hand, it enables aid organizations to continue to operate in challenging contexts and provides an innovative tool for improving security. On the other, some oppose it as a source of promoting donors’ national interests, viewing securitized aid as stimulating more violence and instability in the recipient country. Monaghan (2017), for instance, argues that the securitization of aid is a means through which wealthy donor countries establish their own security management structures within the Global South. In this way, donor countries can ensure global security compliance amongst otherwise deviant countries.
Substantial evidence suggests that security interests have been sufficient to shape the objectives, policies and practices of aid, which has implications for the aid industry at multiple levels (Duffield 2001; Howell and Lind 2009; Rosser 2009; Swiss 2011, 2016). For instance, aid agencies continue to professionalize, establishing additional security posts at headquarters, and field locations, as well as training staff on an array of security issues (Cooley and Ron 2002; James 2016; Walker and Russ 2010). In 2008, the International NGO Safety and Security Association (INSSA) was founded by InterAction – an umbrella group of major development NGOs in the USA – to facilitate, certify, and standardize security responses by humanitarian and development NGOs and professionals (Walker and Russ 2010). These developments indicate the degree to which the securitization of aid has filtered down to all levels of the aid industry.
Stoddard and co-authors (2009) likewise observed these policy transformations within the agencies and programmes of the United Nations, with the new Department of Safety and Security (UNDSS) established to develop programme-led approaches to risk management for all UN aid agencies. While this form of securitization has been concrete in some aid agencies, Fast (2010) describes the reaction of the aid community to the challenge of insecurity as contradictory and uneven, noting that while some agencies have adopted refined security analysis and reporting tools, policies and procedures, others continue, despite risks, to operate without standard or even established protocols or security management strategies. The extent of securitization’s influence on aid organizations varies given the size and nature of each organization.
The securitization of aid, as a growing global trend, has the potential to fuel violence against aid workers as it feeds directly into the three determinants already outlined (1) it results in a changing and complex political context for aid; (2) it prompts increasingly risky aid worker behaviour; and (3) it provides opportunity to those groups and individuals who might attempt to benefit from violent acts against aid workers and organizations.
Context
Due to the securitization of aid, development and humanitarian aid is increasingly being delivered in countries at risk of, embroiled in, or recovering from conflict (Duffield 2001; Stoddard 2020). Aid actors are more likely to be placed in proximity to conflict, actors involved in conflict, and communities made desperate by the effects of poverty and conflict.
In one of the few monographs on this subject, Larissa Fast (2014) examines violence against aid workers and argues convincingly that the cause that receives most attention is the changing nature of the context where aid workers operate, and the power politics contained therein. Aid agencies are not only operating in areas where earlier aid may not have been delivered but are also engaging with other actors and groups with whom they may not previously have had as much contact (state militaries, criminal groups, combatants, other non-state actors) (Stoddard 2020). This is borne out by evidence of the increased prevalence of NGO and aid worker security incidents in provinces of Afghanistan where Provincial Reconstruction Teams (PRTs) were operating (Mitchell 2015) or where aid workers were seen to be most supportive of the Afghan national government (Narang and Stanton 2017).
Moreover, securitization can change the context of aid by altering the actors involved in its delivery, introducing more security, para-military and military personnel into the aid enterprise. Due to this changing context, established notions about the impartiality and neutrality of humanitarian workers are being challenged, and the potential for violence becomes more likely. Aid workers may then become a strategic target in conflict (Narang and Stanton 2017).
Oxfam America (2008) argues that the securitization of aid will ultimately be self-defeating. Where civilian aid responsibilities are handed over to military personnel, project failures become more likely. Soldiers are not sufficiently skilled and equipped to build schools or promote gender equality in aid recipient communities/society. In this way, aid agencies with security responsibilities, or security agencies delivering aid and security sector reform, can become diverted from the ‘neutral’ political ground of delivering humanitarian services and, in so doing, put their staff in danger of armed attacks (Mitchell 2015; Narang and Stanton 2017; Skeppström et al. 2015).
Other evidence points to a more complex relationship between the conflict context and violence directed at aid workers. For instance, Hoelscher et al. (2017) find that such violence increases in association with measures of conflict intensity, number of battle deaths, and the presence/scope of UN peacekeepers but is unaffected by the presence of another foreign military. This suggests that complicated power relations ultimately drive the contextual determinants.
Behaviour
Fast (2014) argues that the change in context and the politicization of development or humanitarian assistance are insufficient on their own to explain the phenomenon of violence against aid workers. If we overlook the role that aid worker and organizational behaviour plays in the process, we resort to a view that aid workers are entirely blameless. This so-called untouchability of humanitarianism leads analysts to disregard various micro-level causes of violence against aid workers in favour of the macro.
Securitization can also affect aid worker behaviours. Putman et al. (2009) share the experience of Guatemala and argue that the macro-level culture of violence in which aid workers live and operate leads to the cycle of violence, as they position themselves at increased risk of exposure to violence through their work. Jenkins (2002) also suggests that those experiencing higher levels of violence in such cultures are usually at much higher risk of developing post-traumatic stress disorder (PTSD) symptoms. Schaefer et al. (2007) make the case that aid workers serving in countries where there are high rates of poverty and crime are more likely to be exposed to violent and traumatic events than those posted in more stable areas. The exposure and trauma associated with these experiences can lead to the normalization or internalization of violence, which may translate into riskier behaviour.
Gassmann (2005) looks for causal explanations in organizational dynamics or the individual actions of aid workers. Both organizational factors, such as organizational culture, trust, and security practices, and individuals’ actions, such as not following operating procedures and standards, may increase the likelihood of becoming a target of violence. Rowley, Crape and Burnham (2008) support this claim, arguing that violence-related mortality and morbidity of aid workers is closely linked to the personal behaviours and security training of aid workers themselves, especially when workers are in vulnerable situations, such as when they are en route to, or from, project sites.
Opportunity
Foreign aid has been long portrayed as an outside influence that can disturb the local dynamics of a country or community by introducing new resources into the mix. In some ways, this is what it is intended to do. De Ree and Nillesen (2009), for example, examine whether there is a negative association between the impact of foreign aid and the risk of civil conflict in sub-Saharan Africa. Collier and Hoeffler (1998) echo this finding and argue that aid can reduce conflict through increasing national incomes and income per capita.
But these disruptions are not always so beneficial. The influx of foreign dollars that accompany aid work make humanitarian operations opportunistic targets. The mere presence of humanitarian aid in a conflict-affected region can prompt a higher incidence of local violence as insurgents and desperate individuals may be drawn to the resources and opportunities available where aid is concentrated (Stoddard 2020; Wood and Molfino 2016; Wood and Sullivan 2015; Zürcher 2017). Thus, securitized aid can contribute to opportunity motives, whether political, military, or criminal, for those who perpetrate violence against aid workers who now are present in communities where they might not have been before (Narang and Stanton 2017). Fast (2014) suggests that the introduction of new aid resources into a community can make aid workers and organizations ‘tempting and valuable targets’ for those who might choose to benefit economically from the opportunity associated with their presence (p. 106). Securitized aid in fragile and failing states, or in warzones, makes them, as well as political and private sector targets, vulnerable to kidnapping for ransom or for political gain like those witnessed in many parts of the world in similar contexts (Forest 2012; Oriola 2016; Stubbert et al. 2015).
Data and Methodology
Little research to date has connected empirically the increase in violence against aid workers and an increasingly securitized aid agenda. We hypothesise that there is a direct link, and in this section lay out the empirical evidence.
In order to analyse effects over time, we compiled a dataset that incorporates the best available data on violence against aid workers and the most comprehensive data available on annual foreign aid flows (Humanitarian Outcomes 2024; Tierney et al. 2011). For the first, we used the data from the Aid Worker Security Database (AWSD), which is an open-access database that tracks serious incidents of violence committed against aid workers. These incidents include killings, kidnappings, detentions, major injuries requiring medical attention and sexual violence. The AWSD is updated daily by a team of researchers who compile incident information from a range of media sources in conjunction with other conflict and security data repositories. The information gathered is vetted for accuracy, including a secondary verification of the data from the affected group. The database does not include accidental incidents, such as vehicle collisions or unintended detonations during mine clearance operations (Humanitarian Outcomes 2025).
We employ the AWSD in our analysis because it is the best available source of data on violence against aid workers globally. The AWSD benefits from both an expansive data collection process and an extensive time period over which daily incidents are captured. Together, these features ensure that this dataset is comprehensive and not overly skewed towards specific countries or regions. For example, if we consider the top five countries in 1999, 2009 and 2019 in terms of total number of incidents (Table 1), we can see that these lists vary considerably. This reflects the diversity of cases captured within the dataset over this extended period of time.
Table 1
Top five countries with highest number of violent incidents committed against aid workers.
| 1999 | 2009 | 2019 | |||
|---|---|---|---|---|---|
| Angola | 13a | Afghanistan | 62 | Syria | 86 |
| Somalia | 10 | Pakistan | 40 | South Sudan | 62 |
| Sudan | 9 | Sudan | 35 | Afghanistan | 50 |
| Indonesia | 7 | Somalia | 33 | Democratic Republic of Congo | 48 |
| Liberia | 6 | Sri Lanka | 26 | Mali | 29 |
[i] a Total number of aid workers killed, kidnapped or injured in that year.
Within our models, we analyse the AWSD data at the country level and create a sample of 136 aid-receiving countries for a total of 2,880 country-year observations in an unbalanced annual panel capturing data from 1997 to 2019. Country-years with missing data on any measure are dropped from the sample if not otherwise indicated.1 We briefly outline the nature of measures used in our analysis in the remainder of this section.
Dependent Variables: Violence against Aid Workers
To account for violent incidents against aid workers we first create annual counts of the incidents tracked in the AWSD (Humanitarian Outcomes 2024). We focus our analysis on three types of incidents: killings, injuries, and kidnappings, as well as a category summing all affected workers. After summing the total workers affected in each category and the overall total for each country in each year, we also create a dichotomous measure for each category that simply indicates the presence of at least one of those events in that country and year. This approach helps to ensure that our analysis captures the range and diversity of incidents across the globe over this time period without giving undue weight to countries that might have a larger proportion of incidents year-on-year.
Figure 1 shows the respective global counts of total aid workers affected in each year for each category, and the overall bar height shows the total workers reportedly affected in each year. There has clearly been a significant overall increase in the number of workers affected by violent incidents recorded in the AWSD from 1997 onward, with fewer than 100 workers reportedly affected in 1997 to more than 500 in 2023.
Figure 1
Aid Workers Affected, 1997–2023.
Source: Aid Worker Security Database.
Some of this increase may be owing to the nature of retrospective vs. real-time reporting: in this respect, the more recent data is likely a fairer representation of actual levels of aid worker violence than the earliest years in the AWSD. Also, some of this change in prevalence may be due to changes in the overall levels of aid provided in this period. Indeed, total ODA climbed from only US$69.5 billion (constant 2021 USD) in 1997 to nearly US$253.3 billion (constant 2021 USD) in 2022. As humanitarian aid grows and organizations expand, the magnitude of the humanitarian footprint also increases (King 2002). More aid may mean more aid workers at risk of violence. Yet, when violent events are measured in contrast to the overall number of aid workers in the field, findings suggest that the number of victims is disproportionate to the expansion of aid worker population (which topped 290,000 in 2008), resulting in a rising rate of attacks per 10,000 workers. Even with these estimates, understanding these rates of violence is not simple, owing to the limited sources and reliability of data available and their inability to capture all acts of violence against aid and humanitarian workers. Fast (2010) cautions that discerning real changes in the rates of violence is a challenge because estimating the ‘denominator’ in the rate equation, that is, the total number of workers – expatriate and local – involved in the aid sector annually in each country, is fraught with difficulty. Without a comprehensive dataset, we cannot reliably establish rates of violence against aid workers, whether globally or at a national level. Thus, we take the AWSD at face value and examine violence against aid workers not as a rate, but instead as a simple count and as a binary variable.
In the models that follow, we focus our analysis on this dichotomous measure of whether a specific type of aid worker violence was present in each country in each year using longitudinal logistic regression. Additional analysis, available on request, repeats this analysis using counts of these incidents via longitudinal negative binomial regression. In our sample, 71 of 136 countries (52 per cent) reported aid workers wounded and 56 of those countries also experienced aid worker deaths. In total, 60 of 136 countries (44 per cent) experienced at least one killing of an aid worker. Kidnappings occurred at least once in 41 of 136 countries (30 per cent). In sum, 77 of our 136 countries (57 per cent) experienced some form of violence affecting aid workers between 1997 and 2019. This suggests that, rather than an isolated phenomenon happening in a select few conflict-affected countries, violence against aid workers is a commonplace occurrence in more than half of our sample of aid-receiving countries.
Focal Independent Variables: Securitized Aid Flows
In the same period that witnessed a sharp increase in the reported violence against aid workers globally, there were significant increases in donor spending on aid linked to conflict, peace, and security outcomes. This has been a feature of specific donors’ aid programmes, as well as a broader shift in the global normative agenda around the security-development nexus (Brown and Grävingholt 2016; Swiss 2011, 2012, 2014). Taking even just the bilateral aid funds dedicated to conflict and security issues, the growth in securitized aid has been overwhelming.
Figure 2 shows the annual totals for aid programming in three areas related to security and conflict: aid to security-sector reform; aid to civilian peacebuilding; and all other conflict-related aid using figures derived from AidData (Tierney et al. 2011). In 1997, at the start of our study period, the total amount in these categories was less than US$100 million. By 2019, this amount had grown to more than US$1.7 billion, and spending on securitized aid had exceeded US$1.5 billion annually for more than a decade. Overall aid also grew in this same period, but not at this same rate. As a barometer for securitization, these spending figures and the sharp growth trend they show are perhaps our best quantitative indicator of the securitization of aid. Although Figure 2 and our multivariate regression results show only three categories of securitized aid spending, our full dataset includes seven separate conflict-related categories based on the OECD DAC classification coding scheme, as well as an eighth aggregate category. Descriptive statistics for each of these categories are reflected below in Table 2. In the analyses that follow, we standardize each of the aid measures so that the model results reflect the effect of a one standard deviation (SD) change on the odds of violence.
Figure 2
Annual Total Aid Related to Conflict, Peace and Security, 1997–2019.
Source: AidData.org.
Table 2
Summary Statistics.
| MEAN | MIN | MEDIAN | MAX | SD | |
|---|---|---|---|---|---|
| Total aid committed to conflict & security (millions constant USD) | 12.1 | 0 | 0.8 | 916.0 | 46.8 |
| Aid committed to conflict prevention and resolution, peace and security (millions constant USD) | 12.1 | 0 | 0.8 | 916.0 | 46.8 |
| Aid committed to security system management and reform (millions constant USD) | 2.0 | 0 | 0 | 451.2 | 12.4 |
| Aid committed to civilian peacebuilding, conflict prevention and resolution (millions constant USD) | 6.0 | 0 | 0.004 | 659.2 | 29.3 |
| Aid committed to post-conflict peacebuilding (UN) (millions constant USD) | 2.0 | 0 | 0 | 473.3 | 12.5 |
| Aid committed to reintegration and SALW control (millions constant USD) | 0.6 | 0 | 0 | 44.2 | 2.8 |
| Aid committed to land mine clearance (millions constant USD) | 1.6 | 0 | 0 | 344.4 | 9.2 |
| Aid committed to child soldiers (prevention and demobilization) (millions constant USD) | 0.05 | 0 | 0 | 12.3 | 0.4 |
| Aid per capita (constant USD) | 53.2 | 0.06 | 27.1 | 837.6 | 82.8 |
| World Bank income category | 1.9 | 1 | 2 | 3 | 0.8 |
| Governance Index | 35.1 | 0.4 | 34.2 | 88.5 | 18.4 |
| Total natural resources rents (% of GDP) | 8.8 | 0 | 4.0 | 87.6 | 11.7 |
| Total unemployment, male (% of male labor force) | 7.7 | 0.1 | 5.6 | 37.0 | 6.1 |
| Battle-related deaths (1000s of people) | 0.4 | 0 | 0 | 71.4 | 2.9 |
Control Variables
As very few quantitative studies of violence against aid workers have been conducted previously, we derive our control variables from other plausible explanations for the incidence of violence in each country over time. Unless otherwise indicated, control variables are collected from the World Bank’s World Development Indicators dataset (World Bank 2023). Descriptive statistics for each control variable are shown in Table 2. We include controls which address both the contextual and opportunity determinants of violence against aid workers.
Contextual
Socio-Economic Development: Our models include two measures of socio-economic development in each country: aid per capita (in constant 2005 USD) and World Bank country income group category. Three income groups are reflected: low-income (reference category), lower-middle income, and upper-middle income. We control for overall aid levels and national income because we expect that wealthier countries and those receiving less aid overall are likely to experience lesser odds of violence against aid workers.
Governance: We expect that countries with governance weakened by conflict and insecurity will be more likely to experience violence against aid workers. We account for governance by creating a simple index of the percentile ranks of six governance indicators from the World Bank’s World Governance Indicators dataset. These include: control of corruption; government effectiveness; political stability and absence of violence/terrorism; rule of law; regulatory quality; and voice and accountability. We calculate the mean percentile rank for these six measures annually for each country and use the score as our indicator of average governance.
Conflict: The final contextual control included in our analysis is the number of battle-related deaths reported during conflict in each country-year. It is likely that aid workers would be at a greater risk of violence in countries where a higher intensity conflict is ongoing (Fast 2010, 2014).
Opportunity
Natural Resources: Collier and Hoeffler argue that countries with a greater amount of natural resource extraction in their economies are at higher risk of internal conflict (Collier and Hoeffler 1998). We introduce natural resource rents as a means of controlling for this indicator. Where there are opportunities to control the rents from natural resources we predict higher risks of violence against aid workers (Edmark 2005).
Unemployment: There is also a known link between conflict and a lack of opportunity in society, in particular for young men. Research shows an association between youth unemployment and crime rates in many contexts. Criminological theories back this up, suggesting that the lack of legitimate opportunities in a community may make deviance and crime in the name of opportunity more likely (Stubbert et al. 2015). With this in mind, we introduce male unemployment rates as a control variable in our models. We expect that countries with higher rates of male unemployment will have a higher incidence of violence against aid workers.
Methods
This paper employs random effects panel logistic regression to assess the association between securitized aid and the odds of violence against aid workers. This modelling strategy accounts for unobserved correlation within countries and heterogeneity between countries over time while evaluating the effects of aid on the incidence of violence against aid workers. Importantly, this modelling approach is not able to draw a distinct causal relationship between the explanatory and dependent variables. It can, however, uncover whether these variables are significantly correlated and also express the magnitude and direction of any relationships that exist.
First, for each variety of security-related aid, we run bivariate models to assess the baseline effect of aid securitization in its many forms on the different measures of violence (See Figures 3, 4, 5, 6 for bivariate results). Then, focusing on three categories of security-related aid — total aid to conflict and security; aid to security system management and reform; and aid to civilian peacebuilding and conflict prevention/resolution) — we show the bivariate model and full model for each of the four dependent variables in Tables 3, 4, 5, 6.
Figure 3
Bivariate Effect of a One Standard Deviation Aid Increase on Killing of Aid Workers, 1997–2019.
Figure 4
Bivariate Effect of a One Standard Deviation Aid Increase on Injury to Aid Workers, 1997–2019.
Figure 5
Bivariate Effect of a One Standard Deviation Aid Increase on Kidnapping of Aid Workers, 1997–2019.
Figure 6
Bivariate Effect of a One Standard Deviation Aid Increase on Any Violence Against Aid Workers, 1997–2019.
Table 3
Random Effects Logistic Regression of Any Aid Workers Killed, 1997–2019.
| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | |
|---|---|---|---|---|---|---|---|---|
| Aid Measures | ||||||||
| Total aid to conflict and security | 1.59*** | 1.31** | ||||||
| (0.18) | (0.14) | |||||||
| Aid to security system management and reform | 1.27* | 1.14 | ||||||
| (0.15) | (0.15) | |||||||
| Aid to civilian peacebuilding, conflict prevention and resolution | 1.41*** | 1.21* | ||||||
| (0.13) | (0.10) | |||||||
| Aid to participation in international peacekeeping operations | 1.30* | 1.23 | ||||||
| (0.16) | (0.17) | |||||||
| Contextual Factors | ||||||||
| Aid per capita (constant USD) | 1.00 | 1.00 | 1.00 | 1.00* | ||||
| (0.00) | (0.00) | (0.00) | (0.00) | |||||
| World Bank Income Group (ref: low-income) | ||||||||
| Lower-middle income | 0.79 | 0.77 | 0.80 | 0.78 | ||||
| (0.23) | (0.22) | (0.23) | (0.23) | |||||
| Upper-middle income | 0.09*** | 0.11*** | 0.09*** | 0.10*** | ||||
| (0.06) | (0.07) | (0.06) | (0.06) | |||||
| Governance Index | 0.91*** | 0.91*** | 0.91*** | 0.91*** | ||||
| (0.01) | (0.01) | (0.01) | (0.01) | |||||
| Battle-related deaths (1000s of people) | 1.31*** | 1.32*** | 1.32*** | 1.33*** | ||||
| (0.09) | (0.09) | (0.09) | (0.09) | |||||
| Opportunity Factors | ||||||||
| Total natural resources rents (% of GDP) | 0.99 | 0.99 | 0.99 | 1.00 | ||||
| (0.01) | (0.01) | (0.01) | (0.01) | |||||
| Unemployment, male (% of male labor force) (modeled ILO estimate) | 1.03 | 1.03 | 1.03 | 1.03 | ||||
| (0.03) | (0.03) | (0.03) | (0.03) | |||||
| Constant | 0.02*** | 0.38 | 0.01*** | 0.37 | 0.02*** | 0.38 | 0.01*** | 0.33 |
| (0.01) | (0.24) | (0.01) | (0.24) | (0.01) | (0.25) | (0.01) | (0.22) | |
| Log Likelihood | –646.68 | –570.01 | –654.79 | –573.01 | –649.78 | –571.04 | –656.81 | –572.61 |
| Observations | 2880.00 | 2880.00 | 2880.00 | 2880.00 | 2880.00 | 2880.00 | 2880.00 | 2880.00 |
| Countries | 136.00 | 136.00 | 136.00 | 136.00 | 136.00 | 136.00 | 136.00 | 136.00 |
| Intraclass Correlation (rho) | 0.63 | 0.49 | 0.67 | 0.51 | 0.65 | 0.51 | 0.68 | 0.51 |
| AIC | 1343.36 | 1204.02 | 1359.58 | 1210.01 | 1349.56 | 1206.07 | 1363.62 | 1209.21 |
[i] Exponentiated coefficients; Standard errors in parentheses.
* p < 0.05, ** p < 0.01, *** p < 0.001.
Table 4
Random Effects Logistic Regression of Any Aid Workers Wounded, 1997–2019.
| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | |
|---|---|---|---|---|---|---|---|---|
| Aid Measures | ||||||||
| Total aid to conflict and security | 2.25*** | 1.90*** | ||||||
| (0.35) | (0.31) | |||||||
| Aid to security system management and reform | 1.08 | 1.04 | ||||||
| (0.08) | (0.07) | |||||||
| Aid to civilian peacebuilding, conflict prevention and resolution | 1.99*** | 1.75*** | ||||||
| (0.29) | (0.26) | |||||||
| Aid to participation in international peacekeeping operations | 1.51** | 1.39* | ||||||
| (0.21) | (0.19) | |||||||
| Contextual Factors | ||||||||
| Aid per capita (constant USD) | 1.00 | 1.00 | 1.00 | 1.00 | ||||
| (0.00) | (0.00) | (0.00) | (0.00) | |||||
| World Bank Income Group (ref: low-income) | ||||||||
| Lower-middle income | 0.59 | 0.62 | 0.61 | 0.61 | ||||
| (0.16) | (0.17) | (0.17) | (0.17) | |||||
| Upper-middle income | 0.29** | 0.33* | 0.29** | 0.31** | ||||
| (0.12) | (0.14) | (0.13) | (0.13) | |||||
| Governance Index | 0.93*** | 0.92*** | 0.93*** | 0.93*** | ||||
| (0.01) | (0.01) | (0.01) | (0.01) | |||||
| Battle-related deaths (1000s of people) | 1.33*** | 1.37*** | 1.35*** | 1.38*** | ||||
| (0.11) | (0.11) | (0.11) | (0.11) | |||||
| Opportunity Factors | ||||||||
| Total natural resources rents (% of GDP) | 0.99 | 0.99 | 0.99 | 1.00 | ||||
| (0.01) | (0.01) | (0.01) | (0.01) | |||||
| Unemployment, male (% of male labor force) (modeled ILO estimate) | 0.99 | 0.99 | 0.98 | 0.99 | ||||
| (0.03) | (0.03) | (0.03) | (0.03) | |||||
| Constant | 0.01*** | 0.08** | 0.00*** | 0.07** | 0.00*** | 0.08** | 0.00*** | 0.06*** |
| (0.00) | (0.06) | (0.00) | (0.06) | (0.00) | (0.06) | (0.00) | (0.05) | |
| Log Likelihood | –680.77 | –627.42 | –700.40 | –637.61 | –682.76 | –628.49 | –696.76 | –634.53 |
| Observations | 2880.00 | 2880.00 | 2880.00 | 2880.00 | 2880.00 | 2880.00 | 2880.00 | 2880.00 |
| Countries | 136.00 | 136.00 | 136.00 | 136.00 | 136.00 | 136.00 | 136.00 | 136.00 |
| Intraclass Correlation (rho) | 0.57 | 0.40 | 0.65 | 0.46 | 0.59 | 0.41 | 0.64 | 0.44 |
| AIC | 1411.54 | 1318.84 | 1450.80 | 1339.22 | 1415.52 | 1320.98 | 1443.52 | 1333.06 |
[i] Exponentiated coefficients; Standard errors in parentheses.
* p < 0.05, ** p < 0.01, *** p < 0.001.
Table 5
Random Effects Logistic Regression of Any Aid Workers Kidnapped, 1997–2019.
| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | |
|---|---|---|---|---|---|---|---|---|
| Aid Measures | ||||||||
| Total aid to conflict and security | 1.42*** | 1.21 | ||||||
| (0.14) | (0.13) | |||||||
| Aid to security system management and reform | 1.13* | 1.00 | ||||||
| (0.07) | (0.06) | |||||||
| Aid to civilian peacebuilding, conflict prevention and resolution | 1.47*** | 1.32** | ||||||
| (0.14) | (0.12) | |||||||
| Aid to participation in international peacekeeping operations | 0.94 | 0.93 | ||||||
| (0.06) | (0.08) | |||||||
| Contextual Factors | ||||||||
| Aid per capita (constant USD) | 1.00 | 1.01* | 1.00 | 1.01** | ||||
| (0.00) | (0.00) | (0.00) | (0.00) | |||||
| World Bank Income Group (ref: low-income) | ||||||||
| Lower-middle income | 0.78 | 0.79 | 0.83 | 0.81 | ||||
| (0.30) | (0.31) | (0.32) | (0.31) | |||||
| Upper-middle income | 0.05** | 0.06** | 0.03** | 0.06** | ||||
| (0.05) | (0.06) | (0.04) | (0.07) | |||||
| Governance Index | 0.87*** | 0.86*** | 0.87*** | 0.86*** | ||||
| (0.02) | (0.02) | (0.02) | (0.02) | |||||
| Battle-related deaths (1000s of people) | 1.06* | 1.07* | 1.06* | 1.06* | ||||
| (0.03) | (0.03) | (0.03) | (0.03) | |||||
| Opportunity Factors | ||||||||
| Total natural resources rents (% of GDP) | 0.99 | 1.00 | 1.00 | 1.00 | ||||
| (0.02) | (0.02) | (0.02) | (0.02) | |||||
| Unemployment, male (% of male labor force) (modeled ILO estimate) | 1.07 | 1.07 | 1.06 | 1.07 | ||||
| (0.05) | (0.05) | (0.05) | (0.05) | |||||
| Constant | 0.00*** | 0.03** | 0.00*** | 0.03** | 0.00*** | 0.03** | 0.00*** | 0.03** |
| (0.00) | (0.04) | (0.00) | (0.03) | (0.00) | (0.04) | (0.00) | (0.04) | |
| Log Likelihood | –403.54 | –339.40 | –408.48 | –341.08 | –400.55 | –336.24 | –410.43 | –340.60 |
| Observations | 2880.00 | 2880.00 | 2880.00 | 2880.00 | 2880.00 | 2880.00 | 2880.00 | 2880.00 |
| Countries | 136.00 | 136.00 | 136.00 | 136.00 | 136.00 | 136.00 | 136.00 | 136.00 |
| Intraclass Correlation (rho) | 0.67 | 0.56 | 0.71 | 0.59 | 0.67 | 0.56 | 0.73 | 0.59 |
| AIC | 857.08 | 742.80 | 866.96 | 746.17 | 851.09 | 736.48 | 870.86 | 745.20 |
[i] Exponentiated coefficients; Standard errors in parentheses.
* p < 0.05, ** p < 0.01, *** p < 0.001.
Table 6
Random Effects Logistic Regression of Any Aid Workers Affected, 1997–2019.
| (1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | |
|---|---|---|---|---|---|---|---|---|
| Aid Measures | ||||||||
| Total aid to conflict and security | 2.03*** | 1.79*** | ||||||
| (0.30) | (0.28) | |||||||
| Aid to security system management and reform | 1.18 | 1.09 | ||||||
| (0.14) | (0.13) | |||||||
| Aid to civilian peacebuilding, conflict prevention and resolution | 2.13*** | 1.86*** | ||||||
| (0.35) | (0.30) | |||||||
| Aid to participation in international peacekeeping operations | 1.17 | 1.20 | ||||||
| (0.13) | (0.14) | |||||||
| Contextual Factors | ||||||||
| Aid per capita (constant USD) | 1.00 | 1.00 | 1.00 | 1.00 | ||||
| (0.00) | (0.00) | (0.00) | (0.00) | |||||
| World Bank Income Group (ref: low-income) | ||||||||
| Lower-middle income | 0.59* | 0.59* | 0.59* | 0.59* | ||||
| (0.15) | (0.15) | (0.15) | (0.15) | |||||
| Upper-middle income | 0.16*** | 0.19*** | 0.16*** | 0.18*** | ||||
| (0.07) | (0.08) | (0.07) | (0.08) | |||||
| Governance Index | 0.94*** | 0.93*** | 0.94*** | 0.93*** | ||||
| (0.01) | (0.01) | (0.01) | (0.01) | |||||
| Battle-related deaths (1000s of people) | 2.04*** | 2.06*** | 2.04*** | 2.11*** | ||||
| (0.29) | (0.29) | (0.29) | (0.30) | |||||
| Opportunity Factors | ||||||||
| Total natural resources rents (% of GDP) | 1.00 | 1.00 | 1.00 | 1.00 | ||||
| (0.01) | (0.01) | (0.01) | (0.01) | |||||
| Unemployment, male (% of male labor force) (modeled ILO estimate) | 1.02 | 1.03 | 1.02 | 1.03 | ||||
| (0.03) | (0.03) | (0.03) | (0.03) | |||||
| Constant | 0.02*** | 0.16** | 0.01*** | 0.16** | 0.02*** | 0.17** | 0.01*** | 0.15** |
| (0.01) | (0.10) | (0.01) | (0.10) | (0.01) | (0.11) | (0.01) | (0.09) | |
| Log Likelihood | –851.78 | –765.06 | –867.39 | –773.28 | –849.49 | –763.63 | –867.95 | –772.06 |
| Observations | 2880.00 | 2880.00 | 2880.00 | 2880.00 | 2880.00 | 2880.00 | 2880.00 | 2880.00 |
| Countries | 136.00 | 136.00 | 136.00 | 136.00 | 136.00 | 136.00 | 136.00 | 136.00 |
| Intraclass Correlation (rho) | 0.61 | 0.43 | 0.66 | 0.48 | 0.61 | 0.43 | 0.66 | 0.47 |
| AIC | 1753.57 | 1594.13 | 1784.78 | 1610.57 | 1748.99 | 1591.26 | 1785.89 | 1608.13 |
[i] Exponentiated coefficients; Standard errors in parentheses.
* p < 0.05, ** p < 0.01, *** p < 0.001.
Results
Bivariate Results
Figures 3, 4, 5, 6 show the bivariate odds ratios for the seven aid flow measures included in our analysis for each of the measures of violence against aid workers. In each figure, the odds ratios are depicted with their 95 per cent confidence intervals. Those estimates above 1, and whose confidence interval does not overlap the dashed line at 1 on the X axis, are positively correlated with the incidence of violence against aid workers. As the explanatory variables are standardized, each estimate in these figures shows the effect of a one SD increase in the given type of aid.
Figure 3 reveals that four of the eight measures predict a higher likelihood of the killing of an aid worker. A one SD increase in total aid to conflict and security predicts 70 per cent higher odds of an aid worker being killed. Likewise, a similar increase in aid to civilian peacebuilding, conflict prevention, and conflict resolution predicts a 50 per cent increase in the odds of an aid worker being killed in that country, while increases in aid to international peacekeeping operations and security system management and reform predict smaller increases in odds of approximately 30 per cent each.
In Figures 4, 5, 6, the estimated bivariate odds ratios for securitized aid flows on injury, kidnapping, and any violence against aid workers are shown. As with aid worker killings, the various securitized aid flows are correlated with increased odds of violence in several cases. We will note only a few examples. Across all four dependent variables, the category of ‘civilian peacebuilding, conflict prevention, and resolution’ is strongly positively correlated with increased odds of violence. Total aid to conflict and security more than doubles the odds of injury to any aid workers and a comparable increase in the odds of any form of violence. Although not all categories are statistically significant in all cases, there is significant support for our hypothesis, particularly among the total security aid and civilian peacebuilding categories.
Multivariate Results
Deaths: In Table 3, we include the bivariate and multivariate results of logistic panel regression models for four categories of securitized aid and their effect on the odds of any aid workers being killed in any year. In the bivariate models, we see that total aid to conflict and security, aid to security system management and reform, aid to civilian peacebuilding, and aid to participation in international peacekeeping are each associated with increases in the odds of an aid worker being killed. In the full multivariate models, the strongest effect (in Model 2) is that of a one SD increase in total aid to conflict and security (a more than 30 per cent increase in odds). Aid to civilian peacebuilding and conflict prevention/resolution predicts a 21 per cent increase in the odds of an aid worker killed in a given year (Model 6). The controls show that only contextual variables are correlated with aid worker killings. Increased national income and better governance both predict reduced odds of a killing, while a one SD increase in battle-related deaths predicts about a 30 per cent increase in the odds of aid workers being killed. Interestingly, we find no correlations between male unemployment or total natural resource rents and aid worker killings.
Injury: Results for the presence of attacks on aid workers where someone was injured are shown in Table 4. These largely mirror the results related to the killing of aid workers, though the effect on odds of aid worker injury is larger in magnitude. Three of the four forms of securitized aid are associated with increased odds of an aid worker being wounded in the full model, ranging from a 39 per cent increase in odds (participation in international peacekeeping) to a 90 per cent increase in odds (total conflict and security category). Turning to our controls, it is the contextual variables related to national income, governance, and battle-related deaths that affect the odds of aid worker injury. For instance, upper-middle income countries have about 67–71 per cent lower odds of having injured aid workers compared to a similar low-income country. Again, neither opportunity measure – male unemployment rate nor natural resource rents – is statistically related to a change in the odds of aid worker injury.
Kidnapping: Table 5 shows the results of models predicting the odds of aid workers being kidnapped. In contrast to the first two dependent variables, only aid to civilian peacebuilding and conflict prevention predicts higher odds of kidnapping. A one-SD increase in aid in this area is estimated to increase the odds of aid worker kidnapping by 32 per cent. The odds of kidnappings having occurred in a given year in upper-middle income countries are 94–97 per cent lower than in low-income countries, all else equal. The odds of kidnapping also increase slightly when battle-related deaths are higher, while the risk of kidnapping decreases in countries that are better governed. Similar to our findings on other forms of violence against aid workers – and in contrast to existing literature – we find that neither of the opportunity measures are correlated with aid worker kidnapping at normal levels of statistical significance.
Finally, Table 6 shows the results for models predicting the occurrence of aid workers affected by any of the three categories of violence (killing, injury, or kidnapping). Here, only two categories of securitized aid flow are associated with increased odds of aid workers being affected by violence. Using the full models, predicted increases in the odds of violence with a one-SD increase in aid are 79 per cent for all conflict/security aid and 86 per cent for aid to civilian peacebuilding and conflict prevention.
In keeping with our other results related to our controls, it appears that contextual variables matter more for predicting the occurrence of any violence than those associated with greater opportunity. The odds of violence are reduced in wealthier and better governed countries, but increase substantially in those with a higher number of battle-related deaths in a given year.
Discussion
Violence against humanitarian aid workers is one of the most distressing and disruptive challenges facing the aid industry. Risks exist in all lines of work, but those experienced by front-line aid workers – both expatriate and local – have grown significantly in recent years. The results of this study provide macro-level evidence of a clear association – if not a definitive causal relationship – between increased flows of securitized aid and escalating violence against aid workers and organizations. Specifically, we find that aid to civilian peacebuilding, conflict prevention and resolution increases the odds of aid workers being killed, wounded, kidnapped or otherwise affected by violence. Similarly, aid associated with participation in international peacekeeping operations increases the odds of aid workers being wounded in that country of operation.
While contextual or country-level factors, including governance, battle-related deaths, aid per capita and income grouping, may impact various forms of violence committed against aid workers, opportunity-level factors do not appear to play a significant role. More specifically, male unemployment and the presence of natural resources in a given country do not appear to be correlated with incidence of violence against aid workers. One explanation may be that aggregate national-level statistics are not well-suited to represent the concept of opportunity in this case. Kidnappings or violent robberies motivated by financial gain may be more evident in case studies or individual anecdotal data than in national statistics.
Our analysis suggests that the securitization of aid is a key element that is currently facilitating an environment in which attacks on aid workers can thrive, regardless of other contextual variables related to poverty, governance, and ongoing conflict. Looking at this evidence, could one argue that security-related aid is only provided in those countries that are already insecure and experiencing or recovering from conflict, which would suggest a spurious relationship between the securitization of aid and violence against aid workers? We feel confident that this is not the case, as our data clearly reveals that security-related aid is not confined only to those countries that are deemed most fragile or conflict-ridden. Since 2005, aid related to conflict and security has been provided to an average 110 countries annually in our sample (min: 95; max 121). This shows how widespread the securitization of aid is, affecting the majority of aid recipient countries. In the cases where this aid has increased sharply, we see heightened risks to aid workers and organizations.
Securitized aid is different from other forms of aid because it is often delivered by or in conjunction with more security, para-military and military personnel, and in fragile and at-risk states, or during ongoing conflicts. This results in the lines becoming blurred between military and humanitarian actors and encourages the targeting of humanitarian aid workers in ways that earlier forms of aid delivery may not have. The collapse of perceived notions of humanitarian neutrality and impartiality no doubt contribute to the strong correlation observed in our data.
Conclusion
While our findings should give pause to aid policy-makers and practitioners alike, this should not be interpreted as a call for donors to simply ‘cut and run’ when violent conflict interferes with the delivery of humanitarian or development assistance. For instance, Autesserre (2006) analyses the changing nature of violence during the transition from war to peace and democracy in the Democratic Republic of Congo between 2003 and 2006. She raises concerns that many deaths were precipitated by disease and malnutrition and could have been prevented if eruptions of violence had not delayed access to humanitarian aid. In instances like this, securitized aid is viewed as desirable, even in the face of potential violence.
Nonetheless, the violent consequences of the securitization of aid are of global concern. As securitized aid increases, we expect to see attacks on aid workers increase. This should be taken very seriously among policy-makers and practitioners working in the field of humanitarian relief and global development. The findings in our study could help humanitarian workers make informed decisions about the risks inherent in their decision to work in regions in which securitized aid is present. They should also help aid organizations institute the policies needed to protect their workers, while continuing to engage in their efforts to deliver aid to those most in need.
Notes
Competing Interests
The authors have no competing interests to declare.