Edinburgh Research Explorer Conservation

Conservation con ﬂ icts are widespread and are damaging for biodiversity, livelihoods and human well-being. Con ﬂ ict management often occurs through interventions targeting human behaviour. Conservation interventions are thought to be made more e ﬀ ective if underpinned by evidence and a Theory of Change – a logical argument outlining the steps required to achieve goals. However, for conservation con ﬂ icts, the evidence and logic sup-porting di ﬀ erent types of interventions has received little attention. Using con ﬂ ict-related keywords, we re- viewed trends in behavioural intervention recommendations across con ﬂ ict contexts globally, as published in peer-reviewed literature. We developed typologies for con ﬂ ict behaviours, intervention recommendations, and con ﬂ ict frames and identi ﬁ ed associations between them and other geographical variables using Pearson's Chi-squared tests of independence. Analysing 100 recent articles, we found that technical interventions (re- commended in 38% of articles) are signi ﬁ cantly associated with con ﬂ icts involving wildlife control and the human-wildlife con ﬂ ict frame. Enforcement-based interventions (54% of articles) are signi ﬁ cantly associated with con ﬂ icts over illegal resource use, while stakeholder-based interventions (37% of articles) are associated with the human-human con ﬂ ict frame and very highly developed countries. Only 10% of articles o ﬀ ered “ strong ” evidence from the published scienti ﬁ c literature justifying recommendations, and only 15% outlined Theories of Change. We suggest that intervention recommendations are likely in ﬂ uenced by authors' perceptions of the social basis of con ﬂ icts, and possibly also by disciplinary silos.


Introduction
Conservation conflicts are some of the most intractable problems facing conservation and are increasing in frequency and intensity globally (Young et al., 2010). These conflicts negatively impinge upon biodiversity, livelihoods and human well-being, and therefore considerable effort is put into their management (Redpath et al., 2015b). Conflicts involve situations where multiple stakeholders with strongly held positions clash over conservation objectives, and when one party imposes their interests over another (Redpath et al., 2013). They are hard to define and are often interpreted differently by authors, managers, and stakeholders involved in the conflict. The language used to describe a given interpretation of a conflict can be considered as a "frame" Fisher, 2016), and in the conservation literature conflicts are framed in many different ways (Table 1). Commonly, authors frame conflicts as primarily occurring between wildlife and humans -"human-wildlife conflict" - (Woodroffe et al., 2005). Others, however, posit that underpinning human-wildlife impacts such as crop-raiding are actually conflicts between different human interests, such as between conservation and agriculture Young et al., 2010). Under this interpretation, the umbrella of conservation conflict extends far beyond wildlife impacts on humans and also involves other conflicts such as those over resource-use, land-use or even animal welfare (Redpath et al., 2015a). For example, in many cases conservation rule-breaking, from illegal wildlife killing to resource use, has been identified as representing political protest or resistance to conservation (De Pourcq et al., 2017;Holmes, 2016).
The ultimate drivers of many conservation conflicts may be rooted in larger societal issues, such as poverty and inequality (Czech, 2008;Vedeld et al., 2012), imbalances of power (Raik et al., 2008) and inappropriate governance processes (Lute et al., 2018) (Table 1). However, the majority of interventions aimed at reducing conservation conflicts focus on the proximate human behaviours which impinge upon conservation interests (Schultz, 2011). These proximate behaviours are often referred to as behavioural "threats" (Salafsky et al., 2008) and interventions commonly target their proximate drivers. For instance, the retaliatory killing of wildlife is often addressed by attempts to reduce wildlife impacts (Nyhus, 2016), deforestation by stronger enforcement  and active opposition to conservation by efforts to improve stakeholder trust (Young et al., 2016) though other social outcomes may also be targeted independently of conservation.
Following Heberlein (2012), human behavioural interventions can be categorised into "technical", "cognitive" and "structural" fixes. Technical fixes attempt to change the external environment and commonly target wildlife impacts such as crop-raiding and livestock depredation. These may include the erection of fences, provision of deterrents, the encouragement of wildlife-friendly products or the diversionary feeding of wildlife (Nyhus, 2016;. These interventions operate under the assumption that retaliatory killing of wildlife, or active opposition to conservation, is directly related to human-wildlife impacts (Pooley et al., 2016). Cognitive fixes instead attempt to change behaviour through information dissemination. Examples include conservation or livelihood education and conservation awareness campaigns (Baruch-Mordo et al., 2011;Holmes, 2003). Structural interventions attempt to change the context itself. Examples include financial instruments (such as incentives, insurance or compensation) or alternative livelihoods to reduce the physical or opportunity costs incurred by wildlife or conservation-related resource restrictions, or to discourage certain resource use (Kremen et al., 2000;Ravenelle and Nyhus, 2017). Likewise, structural fixes include the creation or enforcement of new rules aiming to increase compliance or discourage certain behaviours such as illegal resource use (Agrawal et al., 2014;Arias, 2015). Contrastingly, stakeholder engagement, mediation programmes and conflict transformation efforts are structural fixes which target the social dimensions of conflicts. These operate under a range of rationales, from engendering greater support for conservation, to championing environmental justice (Madden and McQuinn, 2014;Redpath et al., 2017).
Like other types of conservation, conflict interventions are expected to be more effective if they are informed by evidencefrom scientific evidence  to local ecological knowledge (Sterling et al., 2017) and underpinned by a valid Theory of Change (ToC) (Biggs et al., 2017;Margoluis et al., 2013), which describes the logical and ordered sequence of interventions, actions, perturbations and outcomes identified during the planning process (Qiu et al., 2018). However, the evidence underpinning interventions is often lacking (Eklund et al., 2017;Treves et al., 2016), and the extent to which recommended conflict interventions are supported by ToC has not been assessed. Nor has there has been much consideration of the reasons underpinning different conflict interventions.
The purpose of this review is to contribute towards informed conservation conflict management by exploring, across a range of conflict contexts globally, behavioural intervention recommendations as presented in peer-reviewed academic journal articles. We aim to scrutinize how the types of behavioural intervention recommendations differ across these contexts and to inform researchers and decision-makers, particularly those acting at the local scale. To generate a sample of conservation conflict case-studies and intervention recommendations for comparison, we conducted a sampled literature review, and analysed 100 recent articles from the published conservation literature related to conflicts. To identify the prevailing intervention types, we first developed conflict typologies from directed content analysis and then highlighted the most common intervention types recommended by authors in different contexts. To further understand why certain types of intervention are recommended in certain contexts, we explored associations between the recommended interventions, different behavioural threats and conflict frames. We hypothesised that authors who frame conflicts as primarily occurring between humans, would be more likely to recommend stakeholder-based interventions. As some conflict interventions, such as compensation (Ravenelle and Nyhus, 2017) and militarized enforcement , appear to vary regionally, we also considered whether different types of interventions correlate with other geographical factors, such as the development status of nations and the conservation status of species and areas. To identify any possible gaps in the intervention evidence-base, we assessed the extent to which intervention recommendations are supported by scientific evidence and ToC. Lastly, we also estimated the proportion of articles that focus on other forms of evidence (e.g. stakeholder knowledge), and explored whether intervention recommendations and framing could be Table 1 A non-exhaustive and non-mutually exclusive list of different conflict drivers and associated frames presented in the literature, based upon our interpretation.

Conflict drivers
Otherwise framed as

Wildlife impacts
Including livestock depredation or crop-raiding and/or human injury, with associated retaliatory killing or persecution of wildlife and/or active opposition to conservation efforts trying to prevent this. Similar conflicts surround proposed reintroductions, or predator management on recreational hunting estates.

Materials and methods
To generate a sample of conservation conflict case-studies we conducted a search of peer-reviewed conservation literature using ISI Web of Knowledge in October 2016. To facilitate reproducibility and transparency, we followed best-practise guidelines (Haddaway et al., 2015) and applied carefully designed keyword search-strings to capture a wide variety of conflict contexts, including those not necessarily identified in the conservation conflict literature (Table 1).
To focus on interventions, in our final search we included wildcard search terms for a series of active verbs. Using the English language only, we searched for the following combination of terms in the titles, abstracts or keywords of all articles in the ISI core collection: "conservation conflict*" OR ("conservation" AND "illegal") OR ("conservation" AND "conflict" AND ("stakeholder*" OR "human-wildlife")) AND either -"prevent*" OR "mitigat*" OR "reduc*" OR "resolv*" OR "resolution*" OR "solv*" OR "solution*" OR "manag*" OR "interven*" OR "improv*". To avoid unconscious bias in the sample selection (Haddaway et al., 2015), we decided the temporal and spatial boundaries before the final search. We excluded publications before 2011 to focus on the most recent interventions. To aid comparison, reviews and book chapters were excluded to focus on primary case-studies of roughly similar length. The final search yielded 897 results.
To produce a representative sample for analysis, we used a random list generator to sort the sample into a randomly ordered list, from which we analysed articles sequentially. We excluded any publications (N = 57) which did not describe contexts falling within the definition of conservation conflicts provided by Redpath et al. (2013), those which we could not access, reviews, and those which did not make any intervention recommendations (Appendix Table A12). We continued analysing articles, following the random sequence until we had a total sample of 100 relevant articles. This total sample size (N = 100) and proportion of articles reviewed (157/897) was comparable to previous similar studies (Estévez et al., 2015;Peterson et al., 2010;Redpath et al., 2015a). Demonstrating representativeness, there was no significance difference in the proportions of key search terms between the analysed sample and non-analysed sample (Appendix Table A1).
To avoid selection bias (Haddaway et al., 2015) we developed our conflict and intervention typologies (Table 2) and our coding system prior to collecting and analysing our final sample. We used directed content analysis (Hsieh and Shannon, 2005), whereby we first derived each typology from previous reviews, before refining each typology through analysing a large sample of conflict case-studies. This preliminary sample of case-studies (N = 150) was drawn from the published literature using a similar search and sampling process described above (Appendix Search 1).
Following Heberlein (2012), we first categorised interventions into "technical", "cognitive" and "structural" types. With reference to previous conservation conflict reviews (Dickman, 2010;Nyhus, 2016) and content analysis of the preliminary sample, we subdivided "structural" further into "economic", "enforcement" and "stakeholder" types. Our typology of human behavioural threats was derived from existing literature (Salafsky et al., 2008) and content analysis of the preliminary sample to include: "wildlife control", "resource-use", "environment change", "indirect damage" and "active opposition". Likewise, from existing reviews we identified two key frames -"human-wildlife conflict" (HWC) and "human-human" conflict (HHC) Redpath et al., 2015a). We then derived an additional frame -'illegal resource use' (IRU)from content analysis of the preliminary sample.
All data analysis was conducted by the lead author, but the typologies were created and refined in consultation with co-authors. In the final sample, each article was analysed at least twice to check for errors, with ambiguous articles marked and returned to. For all variables (besides framing), we used a binary coding system within larger non-mutually exclusive categoriese.g., articles could describe more than one threat or intervention type, but were categorised as one frame. The development status of nations (as designated by the Human Development Index) (UNDP, 2016), protected area presence, the conservation status of species (as designated by the IUCN Red List) (IUCN, 2017) was recorded, as was the identification of stakeholder groups, wildlife impacts and illegal activity.
After categorising each article in our final sample (N = 100), we calculated intervention recommendation proportions across variables, and identified associations between interventions, behavioural threats and frames, using Pearson's Chi-Squared test for independence and a mosaic plot of Pearson's residual values (using the "vcd" package) in R 3.4.1 (R Development Core Team, 2014).
We recorded articles as demonstrating reasoning akin to a ToC if they identified the steps required for interventions to achieve a desired outcome. We assessed the level of published scientific evidence supporting recommendations using three categories. "Strong" evidence included articles in which all, or nearly all, recommendations were supported either by reference to previous studies, and/or by experimental, correlative or comparative evidence from the study itself. "Partial" evidence included articles in which over half of recommendations were supported by references or within-study evidence. "Weak" evidence included articles in which less than half of recommendations were supported by references or within-study evidence. Following Estévez et al. (2015), we also explored author affiliations (region) and journal geographical scope, and attempted to categorise institution and journal types by disciplinary focus. However, during analysis we found that the interdisciplinary nature of many conservation-related journals and departments meant such a categorisation approach was ultimately unsatisfactory (Appendix 'Journals and Affiliations'). Lastly, following our initial analysisin which we (unintentionally) overlooked non-scientific forms of knowledgewe later attempted to overcome this by estimating the proportion of articles in the whole sample which focused on stakeholder-based knowledge specifically. To do so, we conducted a keyword search (in article titles, abstracts and keywords) of the entire sample (N = 897) for: "local knowledge", "traditional knowledge", "ecological knowledge", "stakeholder knowledge" or "indigenous knowledge".

Results
Across the final sample (N = 100), we categorised 30 articles as using the frame "human-wildlife conflict" (HWC), 41 as 'illegal resource use' (IRU), and 29 as "human-human conflict" (HHC). Of these, we recorded 32 articles describing wildlife control, 59 resource use, 26 environment change, 34 indirect damage and 33 active opposition. 48 articles included IUCN Red Listed species, 40 articles focused on very high development countries, 20 high development, 31 medium development, and 9 low development. 61 articles described protected areas, and 66 reported illegal behaviours (Appendix Table A2). 88% of articles were published in journals with a global scope (Appendix Table A11) and both study locations and author affiliations were spread across the worlds regions (Appendix Fig. A1).
Technical interventions (such as fences, diversionary feeding or guarding tools) were over 2.5 times more likely to be recommended (Odds ratio (OR) > 2.5) when authors reported behaviours related to wildlife control (such as retaliatory killing) (OR: 2.63, P < 0.001) (Fig. 2) and when they used the HWC frame (OR: 2.59, P < 0.001) (Appendix Table A3).
Cognitive interventionssuch as livelihood training and education awareness programmesshowed no clear associations with any conflict variables. Economic interventionssuch as compensation payments or alternative livelihoodsdid not associate with any threat, but were positively associated with high, mid and low development countries (OR, 1.94, P = 0.005), and were negatively associated with very high development countries (OR: 0.51, P = 0.005) (Appendix Table A3).
Enforcement interventionssuch as anti-poaching patrols and new regulationsare positively associated with the threats of resource use (OR: 1.99 P < 0.001), and indirect damage (such as wildlife collisions or pollution) (OR: 1.67, P = 0.005), the illegal resource use frame (OR: 2.09, P < 0.001), and the reporting of illegal behaviours (OR: 2.96, P < 0.001). Enforcement is negatively associated with active opposition (OR: 0.35, P < 0.001) and the human-human conflict frame (OR: 0.56, P = 0.012). Enforcement is also negatively positively associated with high, mid and low development countries (OR: 1.73, P = 0.006)  (Gilman, 2011;Lin et al., 2013) Active opposition Protest Civic protest, lobbying, campaigns against conservation efforts Resistance Sabotage, hostility, non-participation with conservation efforts (Holmes, 2007;Stern, 2008) Frame Human-wildlife conflict (HWC) Authors describe conflict as primarily occurring between humans and other animals. Often involves crop/livestock loss and associated retaliatory killing of wildlife (Nyhus, 2016;Woodroffe et al., 2005) Illegal resource use (IRU) Authors describe rule-breaking natural resource use (such as illegal wildlife trade, logging, bush meat, fisheries, encroachment), without reference to underlying relationships between different stakeholders. These behaviours are usually considered illegitimate (Nijman, 2010;Solomon et al., 2015) Human-human conflict (HHC) Authors describe human disagreements between particular actors over conservation actions or decisions Conservation-related rule-breaking may be considered as acts of protest or resistance (De Pourcq et al., 2017;Redpath et al., 2015a) Z. Baynham-Herd et al. Biological Conservation 222 (2018) 180-188 and negatively associated with very high development countries (OR: 0.58, P = 0.006).
Only 22% of articles recommended just one intervention type, and on average authors recommended 2.16 intervention types. No authors recommended interventions pertaining to all five of our intervention categories, and only enforcement and stakeholder types showed a significant (negative) association (P = 0.004) (Appendix Table A7).
Many of the conflict variables associated with different intervention types were also strongly associated with each other (Appendix Table  A6). The HWC frame was positively associated with articles describing wildlife control, wildlife impacts and IUCN Red-Listed species. The IRU frame was positively associated with articles describing resource use, indirect damage, illegal activity and high, mid and low development countries. In contrast, the HHC frame was positively associated with articles describing active opposition, environment change, stakeholder groups and very high development countries.
15% of articles outlined the steps required for an intervention to reach a goal, but none of these were explicitly referred to as ToC. 10% of articles offered "strong" scientific evidence to justify recommendations, 65% offered "partial" scientific evidence and 25% offered "weak" scientific evidence. Articles offering "weak" evidence tended to recommended less interventions, but this relationship is not significant (Appendix Table A9). Economic recommendations were positively associated with ToC (OR: 1.94, P = 0.006) and strong evidence (OR: 2.13, P = 0.004) and enforcement was positively associated with weak evidence (OR: 1.58, 0.037). Only 16 (1.8%) articles out of the entire search sample (N = 897) made explicit reference to stakeholder-based forms of knowledge in their titles, abstracts or keywords. 68% of first-author affiliations corresponded to same geographical region as the study conflict (Fig. A1). Of those that studied a conflict in a different region, 88% of first-author affiliations were based in Europe or North America. Fig. 1. Chord diagram showing the relationship between behavioural threats (top) and recommended intervention types (bottom). The width of each outer rim depicts the proportion of total articles describing each threat and intervention type. The direction and width of inner flows show the proportion of articles within each behavioural threat category that recommend each intervention type. "Env" = Environment.

Discussion
Globally, many different actors, from scientists, to practitioners to governments, design and implement interventions to tackle conservation conflicts, and these conflicts take many forms. From reviewing the published academic literature, we compare together for the first time a wider range of conservation conflict contexts and show that conflict intervention recommendations vary with regards to the behaviours they target, the way conflicts are framed, and the evidence and reasoning underpinning them.
In contexts where there are human-wildlife impacts (e.g. crop or livestock loss) and often the subsequent retaliatory killing of wildlife, we find that authors tend to recommend technical interventions. Such technical interventions (including wildlife fences and diversionary feeding) aim to alter human behaviour by changing the external environment (Heberlein, 2012). Like others (e.g., Pooley et al., 2016), we find that that those who recommended these interventions typically reason that the retaliatory killing of wildlife will reduce as the damage exerted by wildlife reduces. In contexts where there is illegal natural resource use, or indirect environmental damage, and in countries with lower levels of human development, we find enforcement-based interventions are favoured. As elsewhere (Keane et al., 2008) we identify that enforcement-based interventions are often recommended under the logic that the greater policing of natural resources and stricter regulations will reduce over-harvesting and illegal behaviour directly. Where there is undesirable environment changesuch as agriculture or recreation expansionor active opposition to conservationsuch as protests, hostility or objectionsand in more highly developed countries, we find that stakeholder-based interventions are favoured. These authors often perceive that social, sometimes non-material factors, sustain the conflict and hence stakeholder interventions commonly target emotions and aim to increase dialogue and trust, with the idea that shared, and agreed-upon problems and solutions can be met Young et al., 2016). However, as documented elsewhere (Peterson et al., 2005;Reed, 2008) in our sample, stakeholder-based interventions vary considerably in style and motivation. Some advocates for collaborative decision-making or more devolved governance (Dandy et al., 2014), whereas others focus on increasing decision-making transparency or on conducting stakeholder consultations (Elston et al., 2014).
In terms of behavioural threats, we find that economic interventions are recommended less selectively, but they are more common in less developed countries. This result contrasts with that found for wildlife impact compensation (Ravenelle and Nyhus, 2017), but this might be because we also considered other economic mechanisms (like alternative livelihoods), and other contexts such as natural resource where economic interventions are common (Agrawal et al., 2014). Economic interventions were generally best supported by evidence and reasoning, but no article considered whether it mattered which group or institution was conducting the recommended intervention, despite indications that perceptions of trust can play a key role in responses to conservation interventions (Stern and Coleman, 2015). Cognitive interventions associated with no variables, suggesting they may be deemed suitable across contexts. However, we found many cognitive interventions to be undeveloped in reasoning and unsupported by evidence. Given critiques of the information deficit model underpinning information-based interventions (Heberlein, 2012;Schultz, 2011), we suggest they would benefit from further testing.
Like similar reviews (Estévez et al., 2015;Peterson et al., 2010;Ravenelle and Nyhus, 2017), we were unable to include non-Englishlanguage articles or grey literature, which would likely have provided further insight. Our conclusions are also limited to recommendations about interventions which are unlikely to be accurate reflections of actually implemented interventionsas recommendations are likely less limited by resources or other constraints. Hence, comparing our findings with implemented interventions, including in regions such as South America which are underrepresented in our sample, would be useful future work. The rigour of the analysis could also have been improved by training multiple coders (e.g., Peterson et al., 2010), increasing the sample size and checking the quality of references used as evidence. Experiments could also be designed to test our findings; for example, a choice experiment with conflict mangers or researchers could test the effect of framing on intervention preferences.
Our finding that framing seems to influence whether socially-focused interventions are recommended is significant because all conservation conflicts are ultimately rooted in social conflicts (Redpath Fig. 2. A mosaic plot depicting the association between intervention recommendations and behavioural threats, colour-coded by Pearson's residual values, with blue cells indicating significantly more observations than would be expected under independence (positive association), red cells indicating fewer observations than would be expected (negative association). Box size is proportional to the observed frequencies of each cross-classification. 'Control = Wildlife control, 'Use' = Resource Use, "Env" = Environment change, "Indirect" = Indirect damage, "Opposition" = Active opposition, and "Enforce" = Enforcement. et al., 2013). For instance, beyond wildlife impacts, cultural factors such as religion, or levels of opposition to conservation can determine levels of the retaliatory killing of wildlife (Dickman and Hazzah, 2015;Mariki et al., 2015). Likewise, illegal activities such as poaching or protected area encroachment often reflect protest, opposition or resistance to conservation (Holmes, 2007;Stern, 2008). Reframing conflicts to better reflect their root cause is therefore crucial for successful conflict management Young et al., 2010). Our attempts at exploring the possible influence of disciplinary silos on both framing and intervention recommendations proved unfruitful. However, others have identified disciplinary silos in conservation (Margles et al., 2010), and that interventions recommended by conservation researchers may reflect their disciplinary training (Sandbrook et al., 2013). Hence, given these findings and the importance of framing identified here, we suggest it would be beneficial for researchers to think more broadly about conflicts in conservation, and look beyond the literature specifically related to their study context. Future work should examine the extent to which authors' disciplinary background, beliefs, expertise or the nature of the conflict itself influence their intervention recommendations. For instance, does variation in ethical positions or rationales for conservation (Holmes et al., 2017) influence the types of intervention recommended? Do those that perceive illegal behaviour as being more or less legitimate (e.g., Sheil et al., 2016) differ in the extent to which they advocate enforcement over participatory approaches? Likewise, the reasons why enforcement and stakeholder-based interventions appear to differ depending upon the development status of countries needs to be explored. Does this trend just reflect the increased presence of threatened species or protected areas, or does it represent perceptions of the strength of governance, or more problematic biases revolving around top-down conservation that prevail where conservationists have relatively more power (Duffy, 2014;Kashwan, 2017;Sandbrook, 2017)? Future work could also look at factors such as the broader socio-economic, cultural or governance context, as well as the involvement of particularly marginalised or minority communities in conflicts.
We find that few authors provide ToC, authors rarely justify all intervention recommendations with published scientific evidence, and the adaptive approach was largely overlooked, despite the effectiveness of decision-making frameworks and adaptive management having been regularly advocated (e.g. Bunnefeld et al., 2017). The lack of causalreasoning and scientific evidence is problematic as it suggests conservation interventions often borne out of intuition, group-think or convention rather than evidence (Eklund et al., 2017;Sutherland and Wordley, 2017), which might prevent otherwise successful interventions from being considered. One reason for the lack of ToC might be that only recently has a framework been developed to bridge different methodologies and guide their development for conservation (Qiu et al., 2018).
Step-wise reasoning (ideally underpinned by behavioural theory) and the outlining of clear goals would also make it easier to assess the effectiveness of interventions (Agrawal et al., 2014), thus contributing to the possible evidence-gap that we have highlighted. However, other forms of knowledge, including local ecological knowledge (LEK), or expert/stakeholder experience can also inform interventions (Sterling et al., 2017). We identify that such knowledge forms may be underrepresented in the published literature, and argue that future work could explore this trend further, and identify how best to incorporate multiple knowledge forms in conflict management.

Conclusions
Individuals or groups who actively participate in conservation-related rule-breaking, such as protected area infringement, may as much be in conflict with conservation as those who poison livestock-raiding predators, or those who lobby against conservation regulations in parliament. Behavioural interventions recommended to tackle such conflicts vary with the types of behaviours targeted, the conflict frames adopted by authors, and by the evidence and reasoning underpinning them. Technical intervention recommendations are associated most with conflicts involving wildlife control (such as retaliatory killing) and those framed as "human-wildlife conflict". Enforcement-based recommendations are associated most with conflicts involving (often illegal) natural resource use, and those in less developed countries. In contrast, stakeholder-based intervention recommendations are associated most with conflicts framed as "human-human conflicts" and more highly developed countries. We suggest that effective interventions should be informed by robust and appropriate evidence, and underpinned by carefully considered ToC. We highlight that other factors appear to influence intervention recommendations which might potentially lead to poor decisions being made. Lastly, we recommend that future studies should make the theoretical and evidential basis of their recommendations clearer and research should study why certain conflict frames arise and their impact.

Recommendations
-Researchers should seek to recognise and transcend the arbitrary barriers which categorise different conflicts, so that any entrenched silos do not lead to potentially successful solutions being overlooked. -Researchers should further explore how the framing of conservation conflicts is generated and how it influences intervention suggestions.
-Those recommending conflict interventions should more clearly outline the social and environmental goals targeted, and the steps required to reach these goals.
-Those recommending conflict interventions should justify recommendations with greater evidence, including scientific and stakeholder-based knowledge. -Researchers should aim to contribute to this evidence-base by testing the assumptions underpinning how particular interventions are intended to influence behaviours.

Conflict of interest
The authors have no competing interests to declare.

Policy and ethics
All appropriate ethics and other approvals were obtained for the research.
ZBH was supported by a studentship from the Natural Environment Research Council (NERC) E3 Doctoral training partnership [grant number NERC NE/L002558/1] NERC played no role in study design, data collection, analysis or interpretation.