Finding the Utility of Force to Protect Civilians from Violence: Exploring Outcomes of United Nations Military Protection Operations in Africa (1999–2017)

ABSTRACT

For more than two decades, the United Nations Security Council has tasked Blue Helmets to use all necessary means – including lethal force – to protect civilians from violence. Despite policymakers’ emphasis on reducing threats to civilians, we remain ignorant of the conditions leading to successful outcomes when force is used to protect. Introducing four promising causal conditions, this article employs a new dataset capturing military protection operations across 10 UN missions in Africa from 1999 to 2017 to explore combinations of necessary and sufficient conditions for successful protection outcomes, finding promising recipes for success.

Introduction

What determines UN troops’ ability to protect civilians from physical violence? Puzzlingly, while protecting civilians from violence is the priority task for nearly every military United Nations (UN) peacekeeper, we have few insights into the conditions leading to successful outcomes when they use force to protect. This knowledge-gap is both due to the lack of a theory on the utility of force to protect civilians from violence and to the lack of data suitable for analysing outcome variations of military protection operations across time and UN missions (Dewaal 2014, Diehl and Druckman 2015, Clayton 2016, Williams 2016). This article seeks to address both gaps.

I set out to explore promising conditions that may increase UN military troops’ ability to protect civilians from physical violence, providing the first building blocks towards a theory on the utility of force to protect. To identify potential explanations for successful outcomes, I have cast the net wide into various strands of existing literature and UN policies, capturing four promising causal condition candidates that can influence UN troops’ ability to protect civilians from violence: i) deterrent presence, ii) willingness to accept risk, iii) pre-emption and iv) matching the perpetrators of violence.

To test whether these four conditions hold explanatory power, I use Qualitative Comparative Analysis (QCA) on a dataset I have developed (United Nations Protection of Civilians Operations (UNPOCO). UNPOCO captures these causal conditions and other core characteristics of 200 UN military protection operations at the tactical and operational levels across 10 UN missions in Africa from 1999 to 2017 (Kjeksrud 2019a).¹ One hundred and twenty-six cases are used for the analysis in this article.²

I find that tailored operations, matching the perpetrators of violence, emerge as the only necessary condition for successful outcomes. However, while matching occurs in almost every case with a successful outcome, matching also occurs when UN troops fail to protect. Second, I find that pre-emption and matching the perpetrators of violence come together in a causal pathway to explain almost half of the positive outcomes. It is not enough to be in the right place at the right time to intervene before perpetrators attack, the use of force must simultaneously be tailored to the threat civilians are facing. Equally interesting is the finding that troop-to-population ratios do not appear to explain outcomes across cases. Although we know that large uniformed components decrease conflict intensity and civilian targeting across operations, I do not find this effect reflected in the outcomes of protection operations at the tactical and operational levels (Hultman et al. 2013a, 2014, 2019, Hultman 2016). The same seems true for willingness to accept risk. Having risk-willing troops taking part in operations does not necessarily lead to better outcomes across cases.

Despite the need for relevant theory and data, we find many diverse explanations for UN protection successes and failures. Failures often point to rather severe challenges. Some argue that Blue Helmets are unfit to wield military force for any purpose, although they may effectively persuade conflict parties verbally in some situations (Howard and Dayal 2018, Howard 2019). More commonly, UN missions often deploy insufficient numbers of troops on the ground, diminishing prospects for deterring armed groups from targeting civilians (Sundberg 2020). Blue Helmet contingents are often risk averse due to debilitating political caveats, cumbersome command and control systems and inadequate resources (Holt and Berkman 2006, p. 64, Tardy 2011, United Nations Office of Internal Oversight Services 2014, High-Level Independent Panel on Peace Operations 2015, dos Santos Cruz et al. 2017). Despite much training material developed by the UN Secretariat, UN troops are commonly not well trained on how to protect, as this responsibility rests with the troop contributing countries (Holt and Berkman 2006, Integrated Training Service 2008, Cammaert 2016, Rosén et al. 2016). Sometimes, UN troops do not even possess basic military skills (dos Santos Cruz et al. 2017, p. 13). Troublingly, most missions neither possess technologies to provide early warning nor relevant language skills, leading to wanting situational awareness (Dorn 2010, 2016, Willmot 2017). Despite improved formal guidance, there is still great variation in how troop contributors understand and implement their protection mandate in practice (Bode and Karlsrud 2018).

Notwithstanding this rather long list of significant obstacles to effective Blue Helmet performance, UN peacekeeping has been quite effective in the aftermath of civil war (Fortna and Howard 2008, Di Salvatore and Ruggeri 2017). The presence of large UN peace operations reduces the intensity of conflicts, decreases their duration and increases the longevity of peace after conflict (Hegre et al. 2010, 2011, 2015, Hultman et al. 2013a, 2013b, 2019, Haass and Ansorg 2018, Phayal and Prins 2018, Fjelde et al. 2019). Despite being deployed to the most difficult cases, UN peacekeeping largely does work (Gilligan and Stedman 2003, Fortna 2007, Gilligan 2008). Crucially, UN troops do on occasion protect civilians by force. Successes include defeating armed insurgents – such as the M23 in the DRC (2013) – and ending a regime crackdown on civilians in the Ivory Coast (2010–2011), alongside French forces (PKSOI 2013, Novosseloff 2015). These observations indicate that UN troops can provide protection to civilians under the right conditions. However, existing literature provides few systematic ideas of what these conditions are and how they come about.

Understanding the Utility of Force to Protect Civilians from Violence

Although the protection mandate has been a key feature of UN peacekeeping for two decades, policies and guidance for UN troops on how to protect only emerged within the past few years (United Nations 2015a, 2017a, 2019a). From existing literature, there is scant theory development to lean on to understand this phenomenon. In addition, military force is almost never used to protect, limiting the number of relevant cases (United Nations Office of Internal Oversight Services 2014).

The literature we do find is diverse, challenging studies of causal relations and outcome variations across time and UN missions. Historically, peacekeeping studies followed the practice of peacekeeping, which only occasionally described peacekeepers’ efforts to protect (Fortna and Howard 2008). After the infamous protection failures of the mid−1990s, most of the literature addressed the limits of peacekeepers’ ability to protect civilians (see, e.g., Schmidl 1997). Only by the mid−2000s did the literature start exploring variations between success and failure (Fortna and Howard 2008, p. 284). Since using force remains controversial for the UN, much of the literature is still concerned with the UN’s limitations of using force (Tardy 2007, 2011, Howard 2008, Berdal and Ucko 2014, Karlsrud 2015, Willmot et al. 2016, de Coning et al. 2017, Howard and Dayal 2018, Nadin 2018).

Acutely aware of the challenges of wielding force in UN operations, I seek to understand what works when UN troops use force to protect civilians from violence. Leveraging military theory in studies of peacekeeping is rare, but necessary (Williams 2023). My understanding of military force leans heavily on Rupert Smith, who has written a seminal account of how military force can be employed more wisely in intra-state conflicts ‘among the people’ to increase its utility (Smith 2006). It is one of the few works that combine deliberations on the utility of force with concern for civilian life during armed conflict. Smith criticises today’s military interventions for their ‘deep and abiding confusion between deploying a force and employing force’ (Smith 2006, p. 6). To remedy this confusion, Smith demonstrates how to increase the utility of military force by better understanding its four main functions and the contexts in which each function is most relevant: i) amelioration, ii) containment: iii) deterrence/coercion and iv) destruction. Smith’s conceptualisation of force moves far beyond a mere analysis of kinetic measures, as it treats what these military functions are meant to achieve. This understanding of force holds much promise in studying Blue Helmet operations as it captures the employment of force for other purposes than merely destroying things, which is the traditional understanding of force. I will return to Smith when presenting the matching theory, the first attempt to develop a tailored theory on the utility of force to protect civilians from violence.

I have cast the net wide into various strands of existing literature and UN policies, capturing four promising causal condition candidates that can influence UN troops’ ability to protect civilians from violence: i) deterrent presence, ii) willingness to accept risk, iii) pre-emption and iv) matching the perpetrators of violence. The first two conditions – deterrent presence and willingness to accept risk – capture preconditions for military protection operations, i.e., variations in troop numbers and caveats of troop-contributing countries, while the latter two conditions – pre-emption and matching – capture variations in the modus operandi of UN troops in actual operations, i.e., when and how they used force to protect in particular situations. There certainly are other conditions influencing the outcomes of operations, but in lieu of an encompassing theory on the utility of force to protect, these four conditions make out potentially relevant theoretical stepping-stones that could form part of a future theory on the utility of force to protect.

Deterrent Presence

An important strand in the literature traces the deterrent effects of UN troop numbers. The underlying hypothesis is that the presence of enough uniformed personnel will deter perpetrators from attacking civilians. Quantitative studies have provided much-needed methodical rigour to test this proposition, greatly contributing to our understanding of the macro-effects of UN peacekeeping on civilian security (Hegre et al. 2010, 2015, Hultman et al. 2013a, 2014, 2016, 2019, Kathman and Wood 2014, Hultman 2016, Kim et al. 2020). UN peace operations actually score very well, reducing the intensity of conflict and civilian targeting, reducing the duration of armed conflict and reducing the spread of armed conflicts. These effects, however, only seem to appear when the UN deploys large operations, meaning those missions that deploy thousands of uniformed military personnel. Still, none of these studies pinpoints threshold numbers that trigger this deterrence mechanism.

Military studies, however, have identified specific troop-to-population ratios that correlate with successful outcomes. Mostly concerned with counterinsurgencies, this literature implicitly seeks to identify correlations between ‘enough’ troops and successful outcomes, meaning improved civilian security and stability. The most influential and controversial is James Quinlivan’s 1:50 troop-to-population ratio (Quinlivan 1995). Influential because it has been referred to in US counterinsurgency doctrine, controversial because his findings are based on only a handful of cases that actually obtained the 1:50 ratio (HQ Department of the US Army 2014). Steven Goode has challenged Quinlivan by proposing a ratio of 1:357, based on a larger set of cases (Goode 2009). Goode is sceptical of the explanatory power of ratios as ‘having enough forces does not equate to victory’ (Goode 2009, p. 56). He cautions against overreliance on troop-to-population ratios, as success obviously depends on several other conditions.

Most UN peace operations deploy fewer troops to large areas with huge populations, such as the Democratic Republic of the Congo (DRC) and Mali. Still, some UN missions meet favourable troop-to-population ratios, including in Abyei, Sierra Leone and Liberia. As both literatures seem to point to a deterrent effect of deploying ‘enough’ troops, it seems worthwhile to investigate if larger uniformed components can be part of the explanation for successful outcomes when UN troops use force to protect. We should still keep in mind that many perpetrators keep attacking civilians also in the vicinity of large UN deployments (MONUSCO 2011). Deterrent presence does not always work.

Willingness to Accept Risk

Protecting civilians from violence sometimes demands considerable risk-taking. Many troop contributors are seldom willing to take such risks (Berdal and Ucko 2015). Internal UN reviews have found that UN troops mostly avoid using force altogether and that ‘some TCCs impose[d] written and unwritten national caveats on their contingents, effectively ruling out the use of force’ (United Nations Office of Internal Oversight Services 2014, 2017, para. 8). There are even indications of TCCs trying to influence UN staff involved in investigations of underperforming troops (United Nations Office of Internal Oversight Services 2017, paras 43–45). However, none of these documents name and shame particular underperforming countries, which would be highly controversial.

We also know that some troop contributors are willing to take more risks than others are, such as the Chadian contingent deployed to Mali and the Mongolian contingent deployed to South Sudan (Karlsrud 2015, p. 47, Mold 2017). There is a great variation in how troop contributors relate to risk and the use of force (see ‘Country Profiles’ Providing for Peacekeeping 2020). I am interested in exploring variations in outcomes of operations where TCCs willingness to accept risk differs. The underlying hypothesis is that troops coming from countries that are more willing to accept risk systematically perform better than those deployed from more risk-averse countries. I systematically analyse whether national restrictions linked to the will to take risk can form part of outcome variation explanations. The analysis finds that troops coming from countries that are more willing to intervene militarily to protect civilians will systematically perform better than troops deployed by countries that are more hesitant.

Pre-emption

This is about when UN troops intervene to protect. Timing is critical since responding too late may jeopardise any chance of influencing the outcome. Commonly, UN peace operations are quite static, mostly responding to violence against civilians after the fact. Interestingly, the UN policy on the protection of civilians in UN peace operations highlights the need for pre-emptive operations to protect effectively. Furthermore, some strands in the literature are interested in understanding and improving the UN’s early warning systems and intelligence structures, which can facilitate pre-emptive operations. Few, however, have systematically studied whether UN troops are more effective in pre-emptive or reactive modes of operation. My theorising is that we should expect to see more successful outcomes when UN troops seek to pre-empt perpetrators of violence, based on the hypothesis that reactive responses will often make it challenging to influence the outcome, and therefore lead to more harm to civilians. When civilians are under imminent threat of violence, UN troops may need to intervene before attacks materialise. Ideally, pre-emptive efforts will deny perpetrators the opportunity to attack civilians altogether or reduce their ability to inflict harm. While the academic literature overlooks this aspect, the UN POC policy indicates that pre-emption is a critical component of effective protection operations (United Nations 2019a, para. 56). I seek to investigate variations in outcomes of both reactive and pre-emptive operations, expecting successes and failures in both modes of operation.

Matching Perpetrators of Violence

Smith’s deliberations of the utility of force provide a starting point to study how Blue Helmets use force effectively, but he underplays the agency of the perpetrators of violence. The author behind the ‘matching theory’, Alexander W. Beadle, has rather used Smith’s four functions as a point of departure to develop a theory that includes the role of the perpetrators (Beadle 2011, 2014, 2015). Mirroring Smith’s four functions of force, he finds that perpetrators employ four types of violence against civilians:

1. Impairment: Fostering insecurity by threatening civilian life without physically targeting civilians (Beadle 2014, p. 10). Perpetrators may impair civilian security through their threatening presence or by using civilians as human shields.

2. Incitement: Using violence against civilians to spread fear and insecurity, including through improvised explosive devices and suicide bombers (Beadle 2014, p. 10). Perpetrators are not seeking to kill as many as possible, rather to undermine the government’s ability to protect its citizens.

3. Deterrence/coercion: Using violence to change civilian behaviour, often to deter collaboration with the opposition or to coerce populations into compliance (Beadle 2014, p. 10).

4. Destruction: Using violence to directly destroy civilians or civilian installations, such as during genocide and mass killings (Beadle 2014, p. 10).

Beadle’s argument is that to find utility of force to protect, the function of force employed by the protector must match the type of violence applied by the perpetrator (Beadle 2011, pp. 35–36). If a perpetrator aims to ‘destroy’ an ethnic group, the protector will not find utility of force by ‘ameliorating’ the situation by merely supporting the delivery of humanitarian aid. In this situation, greater utility of force is found in matching the perpetrator, by destroying his ability to conduct mass killings. Conversely, if a perpetrator uses ‘incitement’ or ‘impairment’ against civilians to undermine the legitimacy of a government, using coercive or destructive force against them is likely to lead to stronger incentives to scale up attacks against civilians. In addition, if the most violent functions of force are applied, they risk causing more harm during operations than would otherwise occur in these less violent situations. Instead, ‘containment’ and ‘amelioration’ are better suited to protect civilians in such scenarios. Consequently, to maximise the utility of force, protectors must match the perpetrator’s violence against civilians. Table 1 illustrates how military forces ideally can match the four ways perpetrators use violence against civilians in order to protect more effectively.

Table 1. Perpetrator’s use of violence vs. protector’s use of military force.

Perpetrator violence against civilians	Protector use of military force to protect
Impairment (e.g., presence of armed actors and constant threat of armed clashes)	Amelioration (e.g., presence of observers reporting human rights violations)
Incitement (e.g., indiscriminate attacks by insurgents in government-held areas)	Containment (e.g., creation of weapon-free zones, counter-IED operations)
Deterrence or coercion of civilians (e.g., threats or retaliatory attacks against civilians associated with the enemy, or demonstrative violence to make people flee) (e.g., threats or retaliatory attacks against civilians associated with the enemy, or demonstrative violence to make people flee)	Deterrence or coercion of the perpetrators (e.g., threats or actual use of force to alter the willingness to target civilians through robust show of force or punishing attacks)
Destruction of civilian life or property (e.g., massacres or scorched earth policies) (e.g., massacres or scorched earth policies)	Destruction of perpetrator capabilities (e.g., neutralisation of rebel forces) (e.g., neutralisation of rebel forces)

In a first empirical test of Beadle’s theory, I map the different types of violence against civilians as well as the functions of force used to protect. This enables an analysis of to what degree UN forces have been able to match perpetrators of violence in each case, and if that appears to systematically influence outcomes across operations.

I should note that I only treat military operations against non-state armed groups or rogue state security forces. I do not study the UN’s failure to protect civilians from host state perpetrators of violence as goes against the grain of one of the bedrock principles of peacekeeping, consent of the host-nation. A case of a UN military protection operation is therefore a social phenomenon fulfilling all the following four criteria:

1. perpetrators physically threatened or harmed civilians;

2. UN military troops−with a mandate to protect civilians−deployed to the location where civilians were threatened or harmed;

3. UN troops used military force to protect civilians, and;

4. the UN Secretary-General’s reporting to the UN Security Council captured the incident.

This definition of a case implies that the data captured in UNPOCO primarily rest on the UN Secretary-General’s (UNSG) reporting to the UN Security Council (UNSC), arguably ‘the most regular and visible reporting on mission operations’, although not without bias and flaws (United Nations Office of Internal Oversight Services 2014, para. 16).

Method

The cases I study occurred in different countries, at different times, involving different troops faced with different perpetrators targeting different civilian populations in different ways – phenomena characterised by causal complexity – which is challenging to capture with statistical analyses. Causal complexity, as understood in this study, refers to social phenomena portraying characteristics such as equifinality (different paths leading to the same outcome), conjunctural causation (causal conditions that only in combination lead to the outcome of interest) and asymmetry (that the occurrence and non-occurrence of any given social phenomenon might require different explanations involving different causal conditions) (Schneider and Wagemann 2012, pp. 5–6). I therefore pursue answers using fuzzy set Qualitative Comparative Analysis (fsQCA), a method developed by Charles Ragin to help social scientists explain causal complexity across a larger number of cases (Ragin 2008, Schneider and Wagemann 2012, p. 8). Rather than isolating independent variables’ effect on a dependent variable, QCA uses Boolean algebra to discover causal pathways, combinations of causal conditions that are either necessary or sufficient for an outcome. To arrive at such pathways, QCA portrays each case as a combination of causal and outcome conditions. These combinations can be compared with each other and then logically simplified (Ragin et al. 2006). The comparison is performed with the help of software. The results of these comparisons are displayed as a ‘truth table’, where each row ‘denotes a qualitatively different combination of conditions, i.e., […] a difference in kind rather than difference in degree’ (Schneider and Wagemann 2012, p. 92). In practical terms, I score each case’s membership in sets. Membership scores come in two forms, either ‘crisp’ (either 0.0 or 1.0) or ‘fuzzy’ (somewhere along the scale between 0.0 and 1.0). The fuzziness does not imply a lack of clarity. It permits membership in the interval between 0 and 1 (Ragin et al. 2006).

The comparison of cases, as mentioned, rests on a sub-set of 126 cases compiled in a dataset termed UNPOCO fsQCA, derived from the more encompassing UNPOCO-dataset (Kjeksrud 2019a). The main criterion for case selection has quite simply been sufficient depth and quality of information. This approach to case selection is not ideal.

Reporting bias is one potential pitfall. Since the UN itself produces these reports, they may evade controversial aspects, including failures to protect (Lynch 2014). It is therefore possible that successful operations feature more prominently than failures. Nevertheless, since many UN member states are highly sensitive to the use of military force, the Secretary-General is likely to report most events where UN military personnel were involved in forceful protection operations. It has also become increasingly difficult to suppress reporting on violent incidents, as news outlets, social media and other UN agencies increasingly monitor and report on these through other channels. It is still likely that many protection operations have not made their way into the official UN reporting chain. Another challenge is the non-systematic way in which these reports present and analyse data on the protection of civilians. Only recently do the Secretary-General’s reports even include paragraphs under the heading ‘Protection of Civilians’. Moreover, the Secretary-General’s reports tend to highlight factors that are easily measurable, such as the number of military patrols, the number and type of illegal arms seized and so forth, which is not enough to measure the effect of military efforts to protect. Consequently, multiple reporting channels, paired with fragmented information, and a lack of analysis of cause and effect, make it challenging to capture relevant information systematically across operations. Whenever possible, I have included other sources, mainly in existing literature and news media. All sources are referenced in the dataset.³ Each case has been calibrated according to standard principles for fsQCA (Schneider and Wagemann 2012).

The outcome estimates degrees of success of UN military protection operations. Building on counterfactual reasoning, I have asked of each case: what is likely to have happened to civilians under threat without the military intervention of the UN troops? Counterfactual reasoning rests on possible worlds that must resemble the actual event as closely as possible (Menzies 2001). When introducing a counterfactual condition, it should make minimal change to the real situation, and must therefore be close in time to where the real world and the possible world branched off. I try to establish a counterfactual baseline of a possible world where the UN did not intervene, to be able to analyse what effect the intervention had in real life. I try to minimise the changes to the possible world, by only removing one counterfactual condition, i.e., the military protection operation. I then compare the possible world with the actual outcome following a UN intervention, analysing whether few or many civilians were protected. I do not attempt to explore long-term effects, which would undermine the value of the counterfactual reasoning by introducing second-order effects.

Calibrated as a fuzzy set, rather than a dichotomous (crisp) set, the outcome includes cases at the extreme ends of the scale as well as two ‘fuzzy’ variations in between. Outcomes are scored as ‘everyone protected’ (1.0) in cases where UN peacekeepers protected all potential victims in a specific area at a certain time. One way to achieve this is to deny a perpetrator access to a contested area, by using military force. For example, the former UN/African Union (AU) hybrid operation in Darfur (UNAMID) has on several occasions successfully intervened to prevent armed militias from entering camps for internally displaced persons (United Nations 2014, 2016, 2017c). In these cases, the militias were deterred, withdrew and no one was harmed. Armed communal militias entering camps where unarmed civilians from the opposing side reside represent a direct and imminent threat to civilians. Based on the modus operandi of the communal militias in this conflict, it is highly likely that this situation could have led to many civilian casualties, if left alone.

Outcomes are coded and scored as ‘many protected’ (0.75) when UN troops used force to protect quite effectively, although some civilians were still killed and/or harmed. For example, on 6 July 2013, more than 30 armed and predatory Mai-Mai elements in civilian clothes attacked M23 elements in Kanyaruchinya, close to Goma in the Democratic Republic of the Congo (DRC), while also firing at the local population, killing one person. In response, MONUSCO engaged the Mai-Mai elements, killing one, injuring two and arresting another (United Nations 2013). This situation, if left alone, could easily have led to more than one civilian casualty, both because of the indirect threat to civilians from a possible firefight between the Mai-Mai and the M23, but also due to the sheer number of predatory armed fighters in the vicinity of civilians, on whom they often prey for profit and survival.

Outcomes are scored as ‘few protected’ (0.25) when UN troops used force to protect, but many civilians were still killed or harmed. For example, in July 2005, to protect civilians, the UN mission in the DRC (MONUC) and the national armed forces (FARDC) ran a series of joint military operations in South Kivu to obstruct the movement of the Democratic Forces for the Liberation of Rwanda (FDLR). After warning the FDLR combatants to leave, MONUC and FARDC destroyed several FDLR camps (United Nations 2005). The rationale was to force FDLR fighters to relocate to areas where they would pose less of a threat to the civilian population. Despite robust joint operations, unidentified armed elements attacked the village of Ntulamamba – west of Bukavu in South Kivu – close to where these joint UN/FARDC operations were being conducted. A verification mission by MONUC found that some 47 persons, mostly women and children, had been killed (United Nations 2005). The FDLR denied involvement. In this case, the joint operations against FDLR bases were likely to have triggered revenge attacks inflicting many casualties, even when compared to the typical modus operandi of FDLR. This case was thus coded as a situation where few were protected.

Outcomes are assessed as ‘no one protected’ (0.0) in cases where UN forces have failed to protect victims in a specific area at a certain time, despite having intervened militarily. In many cases, such failures occur because UN forces are unable to respond in time or use tactics that have no effect on civilian security. For example, in 2010, ‘at least 387 civilians, including 300 women, 23 men, 55 girls and 9 boys, were raped by a coalition of combatants from the Democratic Forces for the Liberation of Rwanda and the Mai-Mai Sheka, as well by residual elements of Lieutenant Colonel Emmanuel Nsengiyumva’ (MONUSCO 2011, p. 1). In addition, almost a thousand houses and several shops were looted, and more than 100 civilians were subjected to forced labour (MONUSCO 2011, p. 1). The UN troops stationed nearby responded only after 2 days of attacks (MONUSCO 2011, p. 1). Even then, they were unable to identify the ongoing attacks, as most of the population had fled to find shelter in the bush. The UN troops did not link the deserted villages to the presence of violent perpetrators. The presence of Blue Helmets in this situation did nothing to improve civilian security.

Although I have applied this ‘possible world’-thinking to every case, my coding comes with inherent weaknesses. Due to almost any number of variables in each case, and the fact that I compare operations with events that did not occur – the possible world – it is only possible to broadly assess the outcomes of operations across time and place. My findings should be read in that light.

Table 2 shows that 21 out of 126 operations were assessed as ‘everyone protected’, 49 operations were assessed as ‘many protected’, 46 operations were scored as ‘few protected’ and 10 operations were scored as ‘no-one protected’. Although there are more positive outcomes (70) than negative (56), I do not suggest that protection operations have been successful more often than not, as the cases are selected based on the quality of information.

Table 2. Outcome calibrations with fuzzy scores, qualitative descriptions, number of cases and case IDs from UNPOCO.

Fuzzy score	Description	No. of cases	Case IDs from UNPOCO
1.0	Everyone protected	21	UNMIL Liberia4	UNISFA Abyei2 Abyei6 Abyei8 Abyei11 Abyei12 Abyei20		UNMISS SouthSudan3		UNAMID Darfur10 Darfur11 Darfur13 Darfur14 Darfur15 Darfur16 Darfur17
			MINUSMA Mali2 Mali5 Mali6	MONUSCO DRC106		MINUSCA CAR1 CAR6
0.75	Many protected	49	UNMIL Liberia1 Liberia3 Liberia5 Liberia6 Liberia8 Liberia15	UNISFA Abyei1 Abyei3 Abyei4 Abyei5 Abyei7 Abyei9 Abyei18		UNMISS SouthSudan1 SouthSudan13		UNAMID Darfur1 Darfur5
			MINUSMA Mali4 Mali9	MONUC DRC1 DRC5 DRC8 DRC12 DRC14 DRC15	DRC16 DRC17 DRC21 DRC22 DRC24 DRC26	MONUSCO DRC28 DRC45 DRC47 DRC51 DRC53 DRC54	DRC55 DRC56 DRC61 DRC64 DRC69 DRC85	MINUSCA CAR7
			UNMIS Sudan1 Sudan2	UNOCI IvoryCoast1 IvoryCoast2 IvoryCoast3
0.25	Few protected	46	UNMIL Liberia7 Liberia9 Liberia10	UNAMSIL SierraLeone1		UNMISS SouthSudan5 SouthSudan6 SouthSudan11 SouthSudan12		UNAMID Darfur2 Darfur4 Darfur6 Darfur7 Darfur9
			MINUSMA Mali1	MONUC DRC2 DRC3 DRC4 DRC10 DRC13	DRC19 DRC23 DRC25	MONUSCO DRC44 DRC49 DRC57 DRC63 DRC66 DRC68 DRC70 DRC71 DRC72 DRC74 DRC76	DRC77 DRC79 DRC81 DRC82 DRC83 DRC88 DRC91 DRC99 DRC102 DRC103	MINUSCA CAR4 CAR8
			UNOCI IvoryCoast6
0.0	No-one protected	10	UNISFA Abyei10 Abyei13 Abyei14 Abyei15 Abyei16 Abyei19 Abyei21	MONUSCO DRC27		UNMISS SouthSudan10 SouthSudan15

Deterrent Presence

To capture variations in the deterrent effect of the number of uniformed peacekeepers, I have developed troop-to-population ratios (see Table 3) for each case, using monthly data on uniformed UN deployments in combination with data on the national population size from the UN and the World Bank (United Nations Department of Economic and Social Affairs 2016, United Nations 2018, World Bank 2018). Actual local deployment numbers and local population numbers would have captured this condition more precisely. However, such data combinations are not available.

Table 3. Troop-to-population ratio calibrations with fuzzy scores, descriptions, number of cases, ratio thresholds and case IDs from UNPOCO.

Fuzzy score	Term	No.	Threshold	Case-IDs from UNPOCO (126)
1.0	Fully in	20	<1:100	UNISFA Abyei116 Abyei18–21
0.75	Mostly in	29	>1:100 <1:500	UNAMSIL SierraLeone1	UNMIL Liberia3–10		UNAMID Darfur1–2 Darfur4–7 Darfur9–11 Darfur13–17	CAR CAR4 CAR6–8
				UNMISS SouthSudan5–6
0.25	Mostly out	10	>1:500 <1:1,000	UNMIL Liberia1 Liberia15	UNMISS SouthSudan3 SouthSudan10–13 SouthSudan15		UNOCI IvoryCoast6	CAR CAR1
0.0	Fully out	67	>1:1,000	MONUC DRC1–5 DRC8 DRC10 DRC12 DRC13–17 DRC19 DRC21–26	MONUSCO DRC27–28 DRC44–45 DRC47 DRC49 DRC51 DRC53–57 DRC61 DRC63–64 DRC66 DRC68–72	DRC74 DRC76–77 DRC79 DRC81–83 DRC85 DRC 88 DRC91 DRC99 DRC102–103 DRC106	UNOCI IvoryCoast1–3	UNMIS Sudan1–2
				UNMISS SouthSudan1	MINUSMA Mali1–2 Mali4–6 Mali9

The calibration leans on theories on troop-to-population ratios adapted to a UN setting. Full membership in this set (‘fully in’ (1.0)) is assigned to cases with a ratio better than 1:100, which is half the number of troops as ascribed by Quinlivan (1:50). The reason for this modification is that UN operations are almost never set up for combat operations; they operate with the consent of host authorities and do so impartially. Consequently, peacekeeping should require fewer troops than counterinsurgencies. Partial membership in this set is assigned to cases that fall between the 1:100 troop-to-population ratio and the cut-off point at 1:500, receiving the ‘mostly in’-score of 0.75. The cut-off point is determined by the ratio suggested by Goode (1:357), although slightly increased to better reflect the fact that UN forces mostly operate in non-combat environments. The ‘mostly out’- score (0.25) is given to cases with a troop ratio between the cut-off point of 1:500 and the ‘fully out’-score (0.0), which has been set at 1:1000.

Willingness to Accept Risk

I investigate whether official caveats of UN troop contributors systematically influence outcomes of protection operations across time and UN missions. Accordingly, contributors are ascribed different memberships−either ‘in’ (1.0) or ‘out’ (0.0)−in a crisp set that captures their willingness to use force to protect civilians (see Appendix B). Scores in Table 4 and Table 5 are based on existing literature, official national policies, statements in the UN General Assembly, as well as expert opinions on how each country’s troops operate on the ground (Chesterman 2004, United Nations 2010, 2015a, 2015b; Bellamy et al. 2013, Government of Rwanda 2015, Providing for Peacekeeping 2020).

Table 4. Calibration of TCCs' willingness to accept risk.

Crisp score	Description	# TCCs	TCCs
1.0	Willing	26	Ethiopia Benin Burkina Faso Chad Gabon Gambia Guatemala Guinea Ireland	Malawi Mauritania Mongolia Nepal Netherlands Niger Nigeria Portugal Rwanda	Senegal South Africa Sweden Tanzania Togo Ukraine Uruguay
0.0	Hesitant	12	Bangladesh Cambodia China Egypt	Ghana India Indonesia Jordan	Kenya Morocco Pakistan Philippines

Table 5. Calibration of TCC's willingness to accept risk, including constallations of willing/hesitant TCCS, fuzzy scores, description, number of cases, and case IDs.

Fuzzy score	Description	No.	Case IDs from UNPOCO
1.0	Willing	54	MONUC DRC2	MONUSCO DRC54 DRC56–57 DRC61 DRC70–71 DRC81	DRC85 DRC88 DRC91 DRC99	UNAMID Darfur1–2 Darfur4–7 Darfur9–11 Darfur13–17
			UNOCI IvoryCoast2–3	UNISFA Abyei1–16 Abyei18–21		UNMIL Liberia7
			MINUSMA Mali2 Mali4–6	MINUSCA CAR6
0.75	Fairly willing	32	MONUC DRC1 DRC10 DRC19	MONUSCO DRC49 DRC55 DRC64 DRC66 DRC68–69 DRC72	DRC74 DRC76–77 DRC79 DRC82–83 DRC102–103 DRC106	UNMISS SouthSudan5 SouthSudan10 SouthSudan13 SouthSudan15
			UNOCI IvoryCoast6	MINUSCA CAR1 CAR8		UNMIL Liberia1 Liberia6 Liberia15
			MINUSMA Mali1 Mali9	UNAMSIL SierraLeone1
0.0	Hesitant	40	MONUC DRC3–5 DRC8 DRC12–17 DRC21–26	MONUSCO DRC27–28 DRC44–45 DRC47	DRC51 DRC53 DRC63	UNMISS SouthSudan1 SouthSudan3 SouthSudan6 SouthSudan11–12
			UNMIS Sudan1–2	UNOCI IvoryCoast1		UNMIL Liberia3–5 Liberia8–10
			MINUSCA CAR4 CAR7

In many cases, more than one TCC have been involved. In order to reflect this qualitative difference, I have added a third score. If one or more of the TCCs involved in a case come from a country coded as ‘willing’, these cases receive a membership score 0.75 (‘fairly willing’). The underlying hypothesis is that the presence of at least one ‘willing’ troop contributor will have some positive effect on the outcome.

Pre-Emption

I have coded all 126 cases according to the type of operation UN forces have conducted to protect in Table 6, meaning either reactive or pre-emptive. Forty-three of the 126 cases are coded as ‘pre-emptive’ (scored 1.0), while the remaining 83 cases are ‘reactive’ (scoring 0.0). This is also a crisp set, where cases are either ‘in’ or ‘out’. Pre-emptive operations are those cases where UN forces have tried to intervene before attacks against civilians materialised. Conversely, ‘reactive’ operations respond to situations where violent attacks on civilians are already underway.

Table 6. Calibration of the pre-emptive/reactive character of UN military protection operations, including fuzzy scores, description, number of cases and case IDs from UNPOCO.

Crisp score	Description	No.	Case-IDs from UNPOCO
1.0	Pre-emptive	43	UNAMSIL SierraLeone1	UNMIL Liberia4		UNMIS Sudan1
			MONUC DRC5 DRC8 DRC10 DRC12 DRC14–17 DRC19 DRC21	MONUSCO DRC28 DRC45 DRC55 DRC61 DRC64 DRC68–69 DRC88 DRC106		UNOCI IvoryCoast2–3
			UNISFA Abyei2 Abyei4 Abyei6–7 Abyei10 Abyei12 Abyei20	UNMISS SouthSudan3		MINUSMA Mali2 Mali5–6
			UNAMID Darfur13–17	MINUSCA CAR1 CAR6–7
0.0	Reactive	83	UNMIL Liberia1 Liberia3 Liberia5–10 Liberia15	UNAMID Darfur1–2 Darfur4–7 Darfur9–11		UNMIS Sudan2
			MONUC DRC1–4 DRC13 DRC22–27	MONUSCO DRC44 DRC47 DRC49 DRC51 DRC53–54 DRC56–57 DRC63 DRC66 DRC70–72	DRC74 DRC76–77 DRC79 DRC81–83 DRC85 DRC91 DRC99 DRC102–103	UNISFA Abyei1 Abyei3 Abyei5 Abyei8–9 Abyei11 Abyei13–16 Abyei18–19 Abyei21
			MINUSMA Mali1 Mali4 Mali9	UNOCI IvoryCoast1 IvoryCoast6		UNMISS SouthSudan1 SouthSudan5–6 SouthSudan10–13 SouthSudan15
			MINUSCA CAR4 CAR8

Matching the Perpetrators of Violence

According to the only existing theory on how to maximise the utility of force to protect civilians, military protectors must match the perpetrator’s violence against civilians (Beadle 2011, 2014). This condition has been operationalised as shown in Table 7 by first ascribing one or more functions of force to the protector in each case and then assessing the type of violence committed against civilians by the perpetrator before comparing the two in order to evaluate if the use of force is a ‘match’. Through a crisp-set approach, a ‘match’ is scored 1.0 (‘in’ the set) while a ‘mismatch’ scores 0.0 (‘out’ of the set). In 99 cases, the protectors have matched the perpetrators, while the remaining 27 are coded as a mismatch.

Table 7. Calibration of UN troops’ ability to match the perpetrators by force, including fuzzy scores, description, number of cases and case IDs from UNPOCO.

Crisp score	Description	No.	Case IDs from UNPOCO
1.0	Match	99	UNAMSIL SierraLeone1	MONUC DRC1 DRC4–5 DRC8 DRC10 DRC12 DRC14–17 DRC19 DRC21–24 DRC26 DRC28	MONUSCO DRC45 DRC47 DRC49 DRC51 DRC53–57 DRC61 DRC64 DRC68–72	DRC74 DRC76–77 DRC79 DRC81–83 DRC85 DRC88 DRC102–103 DRC106
			UNMIL Liberia1 Liberia3–10 Lieria15	UNOCI IvoryCoast1–3	UNAMID Darfur1 Darfur5 Darfur10–11 Darfur13–17
			UNMIS Sudan1	UNMISS SouthSudan1 SouthSudan3 SouthSudan10	UNISFA Abyei1–9 Abyei11–16 Abyei18–20
			MINUSMA Mali2 Mali4–6 Mali9	MINUSCA CAR1 CAR6–8
0.0	Mismatch	27	UNISFA Abyei10 Abyei21	MONUC DRC2–3 DRC13 DRC25	MONUSCO DRC27 DRC44 DRC63 DRC66 DRC91 DRC99
			UNAMID Darfur2 Darfur4 Darfur6–7 Darfur 9	UNMIS Sudan2	UNMISS SouthSudan5–6 SouthSudan11–13 SouthSudan15
			MINUSMA Mali1	UNOCI IvoryCoast6	MINUSCA CAR4

Necessary and Sufficient Conditions for Successful Outcomes

The QCA-analysis is performed in two steps. The first is an analysis of necessary conditions. The second identifies potential causal pathways, i.e., combinations of necessary and sufficient conditions, producing successful protection outcomes across cases. Both analyses rest on the calibrations presented above. Together, the 126 cases now form a QCA matrix, where all cases’ membership scores in all condition sets and the outcome set are compiled (see Appendix A). I use QCA-software to perform the analysis (Ragin et al. 2006). Although UNPOCO covers the period from 1999 to 2017, there were no reported cases identified from 1999, 2001 and 2002. This is also the case for the QCA-matrix. All 10 UN missions represented in UNPOCO also appear in the QCA matrix.

Necessary Conditions

A condition is necessary ‘if, whenever the outcome Y is present, the condition is also present. In other words, Y cannot be achieved without X” (Schneider and Wagemann 2012, p. 69). A necessary condition is as such a super-set of the outcome. Table 8 portrays the results of the analysis.

Table 8. Analysis of necessary conditions for the presence of positive outcomes.

Condition	Consistency	Coverage
Deterrent presence (deter)	0.411552	0.644068
Willingness to accept risk (risk)	0.657040	0.583333
Pre-emption (pre-empt)	0.480144	0.773256
Matching (match)	0.902527	0.631313

The software performs two analyses to determine whether a particular condition is necessary. First, a consistency analysis assessing ‘how far the outcome can be considered a subset of the condition’, and second, a coverage analysis measuring ‘the relevance of a necessary condition’ (Schneider and Wagemann 2012, pp. 143, 147). According to common QCA-standards, the consistency threshold should be at least 0.9 (Ragin 2006, Schneider and Wagemann 2012, p. 143). Only one of my conditions – matching the perpetrators of violence (match) – portrays a value that fulfils the threshold for a necessary condition, just breaching the 0.90 threshold (0.902). The QCA-matrix shows that 70 cases portray either fully successful outcomes (21) or partially successful outcomes (49). Matching occurred in 68 of these 70 cases. This explains the high consistency score, although the condition is not always present when a fully successful or partially successful outcome is present. Deterrent presence, willingness to accept risk and pre-emption all score well below the 0.90 threshold. It follows that neither good troop-to-population ratios, willingness to accept risk or pre-emption is necessary to achieve successful outcomes.

The high score of matching require further examination to determine whether this is a trivial or non-trivial relationship of necessity (Schneider and Wagemann 2012, pp. 139–50). The difference between trivial and non-trivial relationships can be portrayed with the help of Venn-diagrams, as shown in Figure 1.

Figure 1. Venn-diagrams portraying the logic of a trivial (1) and non-trivial (2) necessary condition.

Both Venn-diagrams portray the logic of a necessary condition, in that the conditions are super-sets of the outcome. However, they also show different degrees in this relationship. Diagram 1 depicts a trivial relationship, while diagram 2 depicts a relevant, or non-trivial, relationship.

Does the coverage score of 0.631 indicate a non-trivial set relationship? No standard thresholds are provided in the literature. According to one example, a coverage score of 0.65 indicates a non-trivial relationship (Schneider and Rohlfing 2013, p. 565). The 0.631 result from my coverage analysis seemingly does support the claim that matching is indeed relevant for positive outcomes. However, some doubts remain, as trivialness can occur in two ways (Schneider and Wagemann 2012, p. 146). One, the condition set is much larger than the outcome set (ref. Venn diagram 1), and two, both the condition and the outcome are large sets and roughly equal in size, i.e., close to being constant. It follows that because of their size, both the outcome and the condition cover almost the entire universe of cases (Schneider and Wagemann 2012, p. 146). The first phenomenon does not appear in my case, while the latter could be relevant. The software used for this analysis does not fully capture the second type of necessity trivialness (Schneider and Wagemann 2012, pp. 233–237).⁴

I return to the QCA matrix to shed some additional light on this aspect. Although matching is usually present alongside successful outcomes, I also find that UN troops matched the perpetrators of violence in 31 out of 56 cases where few or no civilians were protected. Negative outcomes are not part of the necessity analysis and do not influence the coverage score. However, it does indicate that although matching seems relevant for almost all positive outcomes, it also appears quite often alongside failures to protect.

Causal Pathways for Successful Outcomes

The second analytical step in QCA is to search for causal pathways, or combinations of necessary and sufficient conditions leading to the outcome. QCA now introduces a truth table, which sorts all cases into combinations of sufficient conditions leading towards the outcome in different rows. Before using the analytical tools provided by the software, the researcher must decide the consistency cut-off point for relevant solutions, which determines which rows of combinations will be part of the analysis. This is critical, as the cut-off point will influence the causal pathways’ consistency and coverage scores. From the truth table, it follows that I have had to decide between a cut-off point between 0.79 and 0.81, marked in bold in Table 9.

Table 9. Relevant rows from the truth table derived from the analysis of the QCA matrix.

deter	risk	pre-empt	match	number	raw consist.	PRI consist.	SYM consist.
1	0	1	1	2	1.000000	1.000000	1.000000
1	1	1	1	13	0.913043	.906977	0.951220
0	0	1	1	12	0.810345	.717949	1.000000
0	1	1	1	15	0.796610	.750000	0.818182

The consistency score refers to the percentage of configurations of similar composition which result in the same outcome. Coverage refers to the number of cases for which a configuration of conditions is valid.

A cut-off point at 0.81 only captures 27 cases – possibly increasing solution consistency scores, but certainly decreasing their coverage. A cut-off point at 0.79 will add 15 more cases to the analysis − risking a lower solution consistency score but increasing the chances for a higher coverage score. I opted for a lower cut-off point at 0.79, which yielded the following results:

As shown in Table 10, the analysis only proposes one causal pathway based on these choices and calibrations, a combination of pre-emptive operations that also match the perpetrators of violence. Although this combination scores reasonably well on solution consistency (0.79), it covers less than half of the outcome set (0.48). I also performed a robustness check, analysing the truth table with a higher cut-off point (0.81). Now, a slightly more consistent solution appeared. In addition, matching and pre-emption were now joined by deterrent presence, i.e., good troop-to-population ratios (match pre-empt deter), yielding a consistency score of 0.85. However, as expected, this solution only covered about a third of the outcome set (0.33).

Table 10. Intermediate solution analysis of the truth table derived from the QCA matrix.

Combinations	Raw coverage	Unique coverage	Consistency	Solution coverage	Solution consistency
Match* pre-empt	0.480144	0.480144	0.791667	0.480144	0.791667

The * only indicates the logical combination of match and pre-empt.

What can we derive from this analysis? The most interesting is perhaps the results that do not appear. First, favourable troop-to-population ratios are not alone or together with other conditions able to explain positive protection outcomes across operations at the tactical and operational levels. Being present in large enough numbers is just not enough to protect civilians from violence from imminent threats. Troop-to-population ratios do seem to be part of the explanation in about one-third of positive outcomes, but that also indicates that this condition demands other explanatory factors to become relevant. This provides important nuance to our existing knowledge about the overall conflict reducing effect of large, uniformed components. Although the presence of thousands of troops reduces the intensity of conflict, it does not necessarily explain how UN troops fare in protecting civilians from perpetrators, who often continue to attack civilians even in the presence of Blue Helmets.

Second, Blue Helmets’ willingness to accept risk does not appear to be part of the answer to how they fare in protecting civilians across conflicts and time. Keep in mind that I do not capture the cases where civilians were under threat without a military UN intervention. Hence, a pool of more risk willing troop contributors might have improved the UN’s results overall. However, what we can gather from openly accessible reporting, troops’ country of origin is not able to explain outcomes across cases. Combined with the first insight, it also underlines another main point: It seems to matter more what UN troops do rather than where they are from.

Less surprising is the finding that pre-emptive protection operations tailored to particular threats−matching the perpetrators of violence−are important parts of the causal pathways towards successful outcomes. The analysis of necessary conditions further strengthened the relevance of matching, which is present in almost all successful outcomes. We now know that these two combined provide the most consistent solution across almost half of the cases. Nevertheless, although the QCA analysis does provide interesting insights, the results remain inconclusive, as other conditions must be present to explain the majority of outcomes.

Conclusion

I set out to explore causal conditions explaining UN military protection successes at the tactical and operational levels across time and UN missions in African armed conflicts. I identified four promising causal condition candidates from existing literature and UN policies: i) deterrent presence, ii) willingness to accept risk, iii) pre-emption and iv) matching the perpetrators of violence. With the help of fuzzy-set Qualitative Comparative Analysis of 126 cases derived from a new and unique dataset, I found that tailored operations, matching the perpetrators of violence, emerged as the only necessary condition for successful outcomes. However, while matching occurs in 68 out of 70 cases with successful outcomes, matching also occurs when UN troops fail to protect. Second, I found that pre-emption and matching the perpetrators of violence came together in a causal pathway to explain almost half of the positive outcomes. It is not enough to be in the right place at the right time to intervene before perpetrators attack, the use of force must simultaneously be tailored to the particular threat civilians are facing. Equally interesting is the finding that troop-to-population ratios do not appear to explain outcomes across cases. Although we know that large uniformed components decrease conflict intensity and civilian targeting across operations, I did not find this effect reflected in the outcomes of protection operations at the tactical and operational levels (Hultman et al. 2013a, 2014, 2019, Hultman 2016). The same seems true for willingness to accept risk. Having risk-willing troops taking part in operations does not necessarily lead to better outcomes across cases. However, as many cases remain unexplained, future studies would benefit from applying comparative qualitative case study designs systematically exploring proximate causal conditions at the micro-levels of analysis. This would enable a more holistic understanding of what works when UN troops use force to protect.

These findings may be relevant for the policy and practice of UN military protection efforts. First, in order to match and pre-empt the perpetrators of violence, it is essential to understand how, why and with what perpetrators attack civilians. Although the UN system can collect a lot of relevant information about the conflict dynamics in areas they deploy, the organisation has thus far not been able to develop useful POC-specific threat-assessment methods for those militaries set to protect by force. Some attempts exist (see, e.g., UN Integrated Training UN Integrated Training and Service 2018). Unfortunately, these remain rather generic, failing to take into account the specific motivations of different perpetrators, the wealth of information and research we already have about the modus operandi of armed groups that target civilians, as well as lessons learned from military protection operations.

Second, in order to match and pre-empt perpetrators of violence, UN troops need better pre-deployment scenario training that both rest on systematic knowledge about different perpetrators and what types of military efforts have worked in the past. This knowledge is equally important in order to know when force is likely to have a marginal impact on civilian security, or even increase the threat to civilians.

Third, in order for Blue Helmets to become more effective, they need a better working UN intelligence system. In 2017, the UN published its first comprehensive intelligence policy (United Nations 2017b). While this is a significant step forward, much work remains before UN troops on the ground are provided with actionable intelligence to facilitate more effective protection operations.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data that support the findings of this study are openly available in DataverseNO at https://doi.org/10.18710/FZAVCN, United Nations Protection of Civilians Operations dataset (UNPOCO), V1.

Additional information

Notes on contributors

Stian Kjeksrud

Stian Kjeksrud is an Associate Professor at the Norwegian Defence University College/Command and Staff College, heading a research and development programme on United Nations peace operations. His research interests include the role, limitations, and utility of military force in improving security for civilians in full-spectrum conflict. He also develops educational technology tools using Extended Reality tailored to higher military education. Kjeksrud worked as a senior researcher at the Norwegian Defence Research Establishment from 2007 to 2019. In 2014, he was a visiting researcher at the University of Cape Town, South Africa. He holds a Master's in International Relations from Columbia University in New York (2007) and a PhD from the University of Oslo (2019). He has served as a soldier and officer in Lebanon, Kosovo, North Macedonia, and Afghanistan and as a police officer in Oslo Police District. His latest publication is Using Force to Protect Civilians: Successes and Failures of United Nations Peace Operations in Africa (Oxford University Press, 2023).

Notes

1. UNPOCO captures reported military protection operations from: i) United Nations Multidimensional Integrated Stabilization Mission in the Central African Republic (MINUSCA), ii) United Nations Multidimensional Integrated Stabilization Mission in Mali (MINUSMA), iii) United Nations Mission in the Republic of South Sudan (UNMISS), iv) United Nations Interim Security Force for Abyei (UNISFA), v) United Nations Organization Mission in the Democratic Republic of the Congo (MONUC)/United Nations Organization Stabilization Mission in the Democratic Republic of the Congo (MONUSCO) (together counted as one mission), vi) African Union/United Nations Hybrid operation in Darfur (UNAMID), vii) United Nations Mission in the Sudan (UNMIS), viii) United Nations Operation in Côte d’Ivoire (UNOCI), viiii) United Nations Mission in Liberia (UNMIL) and x) United Nations Mission in Sierra Leone (UNAMSIL).

2. In a recently published monograph, I use a variety of methods on all 200 cases to assess to what degree UN troops have been able to protect civilians across time and UN missions (Kjeksrud 2023).

3. The article builds on data and findings underpinning my PhD-thesis (Kjeksrud 2019b).

4. I have not been able to conduct the QCA-analysis with the updated software for this article.

Appendix A

– QCA-matrix

Categories from left to right: No. (case number in the matrix), case ID (corresponding to case ID in UNPOCO), deterrent presence (deter), troop contributors’ willingness to accept (risk), pre-emptive/reactive operations (pre-empt), the ability to match perpetrators by force (match) and the outcome variable (outcome). This matrix combines fuzzy and crisp scores.

Table

No.	case ID	deter	risk	pre-empt	match	outcome
1	SierraLeone1	0.75	0.75	1.0	1.0	0.25
2	DRC1 (MONUC)	0.0	0.75	.0	1.0	0.75
3	Liberia1	0.25	0.75	.0	1.0	0.75
4	DRC2 (MONUC)	0.0	1.0	.0	0.0	0.25
5	DRC3 (MONUC)	0.0	0.0	.0	0.0	0.25
6	Liberia3	0.75	0.0	.0	1.0	0.75
7	Liberia4	0.75	0.0	1.0	1.0	1.0
8	Liberia5	0.75	0.0	.0	1.0	0.75
9	Liberia6	0.75	0.75	.0	1.0	0.75
10	DRC4 (MONUC)	0.0	0.0	.0	1.0	0.25
11	DRC5 (MONUC)	0.0	0.0	1.0	1.0	0.75
12	DRC8 (MONUC)	0.0	0.0	1.0	1.0	0.75
13	DRC10 (MONUC)	0.0	0.75	1.0	1.0	0.25
14	DRC12 (MONUC)	0.0	0.0	1.0	1.0	0.75
15	DRC13 (MONUC)	0.0	0.0	.0	0.0	0.25
16	DRC14 (MONUC)	0.0	0.0	1.0	1.0	0.75
17	DRC15 (MONUC)	0.0	0.0	1.0	1.0	0.75
18	DRC16 (MONUC)	0.0	0.0	1.0	1.0	0.75
19	IvoryCoast1	0.0	0.0	.0	1.0	0.75
20	Liberia7	0.75	1.0	.0	1.0	0.25
21	DRC17 (MONUC)	0.0	0.0	1.0	1.0	0.75
22	DRC19 (MONUC)	0.0	0.75	1.0	1.0	0.25
23	Liberia8	0.75	0.0	.0	1.0	0.75
24	DRC21 (MONUC)	0.0	0.0	1.0	1.0	0.75
25	DRC22 (MONUC)	0.0	0.0	.0	1.0	0.75
26	DRC23 (MONUC)	0.0	0.0	.0	1.0	0.25
27	Darfur1	0.75	1.0	.0	1.0	0.75
28	DRC24 (MONUC)	0.0	0.0	.0	1.0	0.75
29	DRC25 (MONUC)	0.0	0.0	.0	0.0	0.25
30	Sudan1	0.0	0.0	1.0	1.0	0.75
31	Sudan2	0.0	0.0	.0	0.0	0.75
32	Liberia9	0.75	0.0	.0	1.0	0.25
33	Liberia10	0.75	0.0	.0	1.0	0.25
34	DRC26 (MONUC)	0.0	0.0	.0	1.0	0.75
35	DRC27 (MONUSCO)	0.0	0.0	.0	0.0	0.0
36	DRC28 (MONUSCO)	0.0	0.0	1.0	1.0	0.75
37	Darfur2	0.75	1.0	.0	0.0	0.25
38	SouthSudan1	0.0	0.0	.0	1.0	0.75
39	IvoryCoast2	0.0	1.0	1.0	1.0	0.75
40	IvoryCoast3	0.0	1.0	1.0	1.0	0.75
41	Abyei1	1.0	1.0	.0	1.0	0.75
42	Abyei2	1.0	1.0	1.0	1.0	1.0
43	Abyei3	1.0	1.0	.0	1.0	0.75
44	Abyei4	1.0	1.0	1.0	1.0	0.75
45	Abyei5	1.0	1.0	.0	1.0	0.75
46	Abyei6	1.0	1.0	1.0	1.0	1.0
47	DRC44 (MONUSCO)	0.0	0.0	.0	0.0	0.25
48	DRC45 (MONUSCO)	0.0	0.0	1.0	1.0	0.75
49	DRC47 (MONUSCO)	0.0	0.0	.0	1.0	0.75
50	DRC49 (MONUSCO)	0.0	0.75	.0	1.0	0.25
51	Darfur4	0.75	1.0	.0	0.0	0.25
52	Darfur5	0.75	1.0	.0	1.0	0.75
53	SouthSudan3	0.25	0.0	1.0	1.0	1.0
54	SouthSudan5	0.75	0.75	.0	0.0	0.25
55	SouthSudan6	0.75	0.0	.0	0.0	0.25
56	Abyei7	1.0	1.0	1.0	1.0	0.75
57	Abyei8	1.0	1.0	.0	1.0	1.0
58	Abyei9	1.0	1.0	.0	1.0	0.75
59	Abyei10	1.0	1.0	1.0	0.0	0.0
60	Abyei11	1.0	1.0	.0	1.0	1.0
61	DRC51 (MONUSCO)	0.0	0.0	.0	1.0	0.75
62	DRC53 (MONUSCO)	0.0	0.0	.0	1.0	0.75
63	DRC54 (MONUSCO)	0.0	1.0	.0	1.0	0.75
64	DRC55 (MONUSCO)	0.0	0.75	1.0	1.0	0.75
65	DRC56 (MONUSCO)	0.0	1.0	.0	1.0	0.75
66	DRC57 (MONUSCO)	0.0	1.0	.0	1.0	0.25
67	Darfur6	0.75	1.0	.0	0.0	0.25
68	Darfur7	0.75	1.0	.0	0.0	0.25
69	Darfur9	0.75	1.0	.0	0.0	0.25
70	Mali1	0.0	0.75	.0	0.0	0.25
71	SouthSudan10	0.25	0.75	.0	1.0	0.0
72	SouthSudan11	0.25	0.0	.0	0.0	0.25
73	SouthSudan12	0.25	0.0	.0	0.0	0.25
74	SouthSudan13	0.25	0.75	.0	0.0	0.75
75	Abyei12	1.0	1.0	1.0	1.0	1.0
76	Abyei13	1.0	1.0	.0	1.0	0.0
77	Abyei14	1.0	1.0	.0	1.0	0.0
78	Abyei15	1.0	1.0	.0	1.0	0.0
79	DRC61 (MONUSCO)	0.0	1.0	1.0	1.0	0.75
80	DRC63 (MONUSCO)	0.0	0.0	.0	0.0	0.25
81	DRC64 (MONUSCO)	0.0	0.75	1.0	1.0	0.75
82	DRC66 (MONUSCO)	0.0	0.75	.0	0.0	0.25
83	Darfur10	0.75	1.0	.0	1.0	1.0
84	Darfur11	0.75	1.0	.0	1.0	1.0
85	Mali2	0.0	1.0	1.0	1.0	1.0
86	Mali4	0.0	1.0	.0	1.0	0.75
87	Mali5	0.0	1.0	1.0	1.0	1.0
88	Mali6	0.0	1.0	1.0	1.0	1.0
89	DRC68 (MONUSCO)	0.0	0.75	1.0	1.0	0.25
90	DRC69 (MONUSCO)	0.0	0.75	1.0	1.0	0.75
91	DRC70 (MONUSCO)	0.0	1.0	.0	1.0	0.25
92	DRC71 (MONUSCO)	0.0	1.0	.0	1.0	0.25
93	DRC72 (MONUSCO)	0.0	0.75	.0	1.0	0.25
94	DRC74 (MONUSCO)	0.0	0.75	.0	1.0	0.25
95	DRC76 (MONUSCO)	0.0	0.75	.0	1.0	0.25
96	DRC77 (MONUSCO)	0.0	0.75	.0	1.0	0.25
97	DRC79 (MONUSCO)	0.0	0.75	.0	1.0	0.25
98	DRC81 (MONUSCO)	0.0	1.0	.0	1.0	0.25
99	MALI9	0.0	0.75	.0	1.0	0.75
100	DRC82 (MONUSCO)	0.0	0.75	.0	1.0	0.25
101	DRC83 (MONUSCO)	0.0	0.75	.0	1.0	0.25
102	DRC85 (MONUSCO)	0.0	1.0	.0	1.0	0.75
103	DRC88 (MONUSCO)	0.0	1.0	1.0	1.0	0.25
104	DRC91 (MONUSCO)	0.0	1.0	.0	0.0	0.25
105	DRC99 (MONUSCO)	0.0	1.0	.0	0.0	0.25
106	Darfur13	0.75	1.0	1.0	1.0	1.0
107	Darfur14	0.75	1.0	1.0	1.0	1.0
108	Darfur15	0.75	1.0	1.0	1.0	1.0
109	DRC102 (MONUSCO)	0.0	0.75	.0	1.0	0.25
110	DRC103 (MONUSCO)	0.0	0.75	.0	1.0	0.25
111	DRC106 (MONUSCO)	0.0	0.75	1.0	1.0	1.0
112	Darfur16	0.75	1.0	1.0	1.0	1.0
113	Darfur17	0.75	1.0	1.0	1.0	1.0
114	Abyei16	1.0	1.0	.0	1.0	0.0
115	Abyei18	1.0	1.0	.0	1.0	0.75
116	Abyei19	1.0	1.0	.0	1.0	0.0
117	Abyei20	1.0	1.0	1.0	1.0	1.0
118	Abyei21	1.0	1.0	.0	0.0	0.0
119	Liberia15	0.25	0.75	.0	1.0	0.75
120	IvoryCoast6	0.25	0.75	.0	0.0	0.25
121	SouthSudan15	0.25	0.75	.0	0.0	0.0
122	CAR1	0.25	0.75	1.0	1.0	1.0
123	CAR4	0.75	0.0	.0	0.0	0.25
124	CAR6	0.75	1.0	1.0	1.0	1.0
125	CAR7	0.75	0.0	1.0	1.0	0.75
126	CAR8	0.75	0.75	.0	1.0	0.25

Appendix B