Intervener trustworthiness predicts cooperation with conservation interventions in an elephant conflict public goods game

1. Conservation conflicts exist in complex socio-ecological systems and are damaging to both people and wildlife. There is much interest in designing interventions to manage them more effectively, but the importance of who does the intervening remains underexplored. 2. In particular, conflicts are influenced by perceptions of the trustworthiness of nat-ural resource managers and conservation organizations. However, experimental studies of how the different facets of trustworthiness shape responses to inter - ventions are rare in conflict settings.

Hilbig, 2015; Young, Searle, et al., 2016). However, a widely accepted definition in conservation (Riley, Ford, Triezenberg, & Lederle, 2018;Young, Searle, et al., 2016) describes trust as a product of social relationships whereby actors 'accept vulnerability based upon positive expectations of the intentions or behaviours of others' (Rousseau, Sitkin, Burt, & Camerer, 1998, p. 395). This definition draws specific attention to the role of trustworthiness, which is itself defined in terms of an actor's beliefs about others (Sharp, Thwaites, Curtis, & Millar, 2013). Trustworthiness encompasses perceptions of an actor's ability to carry out an action, their benevolence (i.e. their intention to act in the interest of the trustor) and their integrity (i.e. their adherence to an acceptable set of principles; Mayer, Davis, & Schoorman, 1995). Alongside risk, these perceptions of trustworthiness are thought to shape the levels of trust in both individuals (interpersonal trust) and organizations (organizational trust; Davenport et al., 2007;Pirson & Malhotra, 2010;Riley et al., 2018). Stern and Coleman (2015) identify four forms of trust: dispositional trust (i.e. the general tendency to trust others), rational trust (i.e. trust based on calculated decision-making), affinitive trust (i.e. trust based on relationships between trustor and trustee) and procedural trust (i.e. trust based on interactions and systems governing between trustor and trustee).
Perceptions of a resource manager's ability are considered formative in assessments of rational trust, whereas perceptions of their integrity and benevolence tend to inform assessments of affinitive trust (Stern & Coleman, 2015). Likewise, trustworthiness is considered an important determinant of perceptions of procedural justice, which alongside perceptions of competence, is thought to dictate the levels of organizational trust (Riley et al., 2018;Rudolph & Riley, 2017). In one study, perceptions of the trustworthiness of fishery management (including perceptions of deception) predicted the rates of compliance (Shirley & Gore, 2019), and in another, trust in a state wildlife agency was more strongly predicted by the perceptions of procedural fairness than technical competence (Riley et al., 2018). Hence it appears likely that the different components of trustworthiness influence trust and conflict-related behaviours differentially. However, experimental evidence testing the relationship between trustworthiness and responses to interventions remains rare in wildlife conservation settings. Such studies are important for testing behavioural theory and management assumptions while providing much-needed evidence to inform conservation interventions (Pollard et al., 2019;St. John, 2014;Sutherland & Wordley, 2017).
The purpose of this study is to experimentally test the importance of three components of trustworthiness-ability, benevolence and integrity (Stern & Coleman, 2015)-in shaping stakeholder support for conflict-reducing conservation interventions. To do so, we use a novel experimental public goods game, framed around elephant conflict interventions, in a Tanzanian Community Wildlife Management Area (WMA). Our game draws upon classic games in behavioural economics (Cookson, 2000;Hasson, Löfgren, & Visser, 2010) and recent games in conservation research (Redpath et al., 2018). Specifically, we test whether cooperation with interventions is linked to the identity of the intervening group (organization) and perceptions of their trustworthiness measured in pre-game surveys. These results are then contextualized using post-game individual and group debrief interviews to help to inform conflict intervention best practice.

| Study area
We conducted our study across three villages in Enduimet Wildlife Management Area (WMA), northern Tanzania-a wildlife corridor in the Amboseli-Kilimanjaro ecosystem where conflicts between local communities and elephant Loxodonta africana conservation are rife and damaging (Bluwstein, Moyo, & Kicheleri, 2016;Homewood, 2017). Here, communities derive some material benefits from conservation-related income, and a local conservation organization works with the government to administer the WMA (Wright, 2017). Conflict interventions range from WMA officers using vehicles to deter wildlife, to them distributing torches and firecrackers to local young men of the 'Moran' age-set-who in Maasai culture are the traditional defenders of villages from both people and wildlife. Moran frequently form small groups to guard village crops from elephants and other herbivores at night, but crop-raiding is still common. Compensation payments have been promised for wildlife-related damage or human deaths, but none have been delivered (Homewood, 2017) and there is a history of distrust and resentment towards tourism operators and the WMA among some residents (Wright, 2017).

| Game design
To ease game and participant organization in the rural field setting and maximize statistical power, our experimental public goods game employed a within-subject design (N = 212) with four players assembled around a physical board. Following classic public goods games (Cookson, 2000), players were instructed that the total amount of tokens they each amassed during each game would determine their earnings. In each of five rounds, participants were endowed with five crop-tokens, and one additional token that they could choose to a) contribute towards elephant guarding (at a personal cost in income, but to a group benefit in reduced crop loss) or, to b) keep for themselves (personal benefit, group cost). The two treatments differed only in the description of the group providing the elephant guarding: government-led WMA or community-led Moran (Supporting Information, Game design). Although per-round there was a 50% chance each player incurred crop-raiding, the damage incurred (i.e. number of tokens lost) decreased in proportion to the total elephant-guarding contribution (Equation 1).
where C is the expected loss from crop-raiding, P is the probability of crop-raid (0.5) and N is the number of cooperators (represented as number of guarders).
For any one player, the expected personal benefit from cooperating-public marginal per capita return (MPCR public ; Hasson et al., 2010)-was half a token, which was less than the expected personal benefit of not cooperating (MPCR private ), which was one token. Both were less than the total group benefit of any one player's cooperation (n × MPCR public ) which were two tokens. Hence, the game satisfies the conditions for a social dilemma (Equation 2), since for rational individuals maximizing short-term earnings, it pays less to cooperate (Table 1).
where n is the number of players.
Each group played the game twice (with the order of the treatment rotated). Players were able make comments to each other within the group throughout the game (such as commiserating losses or wishing for luck), but they were asked not to discuss their previous or intended cooperation decisions, and this rule was observed without problem throughout the games. Cooperation decisions were made anonymously, and were never disclosed (Aswani, Gurney, Mulville, Matera, (1) C = P(5 − N), (2) MPCR public < MPCR private < n × MPCR public , TA B L E 1 Game pay-off table showing that each player (in groups of four) were given five tokens representing crops, and one token representing resources which they could choose to (a) contribute towards elephant guarding (at a personal cost in income but group benefit in reduced crop loss), or to (b) keep for themselves (personal benefit, group cost). For any one player, regardless of what others do, the individual pay-off from cooperation was half a token (realized by reduced crop loss risk), but the cost of cooperation was one token (MPCR = 0.5). Accordingly a Nash equilibrium is formed at no cooperation, which is less than the Pareto Optimum (full cooperation) No. other co-operators

| Data collection
Between April and June 2017, male participants from three villages in Enduimet WMA were recruited from randomly ordered lists of inhabitants known to be present in each village, created in consultation with village chairpersons (Supporting Information, Participant recruitment). As cultural norms prevented gender mixing (Smith, 2015), rather than split the experiment across male-and female-only samples (e.g. Keane et al., 2016), we opted to maximize statistical power and the sample size within one group-males-who were also more dominant in wildlife guarding (Homewood, Kristjanson, & Trench, 2009) and more easily recruited. However, as recruitment was achieved via mobile phones, which have high but not universal penetration in the local area (Soares, 2018), the sample was likely biased towards more literate and potentially wealthier males.  This study was approved by the University of Edinburgh School of GeoSciences Ethics Committee. We received oral informed consent from participants to participate in this research. This was obtained orally following the provision of information sheets and appropriate spoken translations, and participants were told that they did not have to continue with any interview or game if they did not feel comfortable (Tindana, Kass, & Akweongo, 2006). We did not seek written consent due to high levels of illiteracy in the study population. We anonymized both raw game, survey and debrief data to ensure that no personal information could be traced back to individuals (John et al., 2016). These data were stored digitally in the field on a password-protected laptop and was not shared with any other parties.

| Data analysis
We conducted our analyses in r (version 3.4.4; R Development Core Team, 2016) using the statistical package lme4 (Bates, Sarkar, Bates, & Matrix, 2007). As individual levels of cooperation did not show a statistically significant trend over rounds ( Figure S3), we summed the cooperation score for each player over the five rounds of each treatment. To test whether cooperation differed between intervening groups, a priori predictors of cooperation and socio-demographic variables, we ran a series of generalized linear mixed effect models with binomial errors (Table 2). Unique identifiers for groups (N = 53) and participants (N = 212) were modelled with independent random intercepts reflecting the grouping structure within the data. These groups did not account for participants' village, which was instead

| RE SULTS
In the framed public goods game, the proportion of participants cooperating was consistently higher in the local group treatment ('Moran') than in the WMA treatment. In the absence of other covariates, treatment was a significant predictor of cooperation, and the odds that participants cooperate with the Moran were 60% greater than for the WMA (p < 0.01, odds ratio 0.41, 0.95 CI: 0.32-0.54; Model 1). The time of day in which the game was played (morning or afternoon) was not significant, but cooperation was significantly lower in games in which the WMA treatment came first (p < 0.01, odds ratio 0.32, 0.95 CI: 0.14-0.69; Model 1).
Levels of cooperation corresponded closely to perceptions of the trustworthiness of each group. When our aggregate trustworthiness score was included alongside treatment (Model 2), the effect size of treatment was no longer distinguishable from zero-suggesting that the effect of treatment is mediated by perceived levels of trustworthiness ( Figures S3 and S4). When aggregate trustworthiness was replaced as a predictor in the model by the three-component measures (Model 3), benevolence and integrity were found to be significant predictors of cooperation, but ability was not. Integrity was a stronger predictor of cooperation (odds ratio 1.88, 0.95 CI: 1.55-2.28), with an odds ratio 25% greater than for benevolence (odds ratio 1.50, 0.95 CI: 1.20-1.87).
To explore the robustness of this finding, we ran further models which included a range of a priori predictors of cooperation and socio-demographic variables ( Table 2) and integrity-based reasoning (6% more people) was biased towards the Moran (Figure 2).

| D ISCUSS I ON
This study affirms that the perceived trustworthiness of the group delivering a conservation intervention predicts the levels of stakeholder cooperation. It also finds that different components of trustworthiness-perceived ability, benevolence and integrity-differ in their influence.
Our result that trustworthiness predicts cooperation was unsurprising given previous findings and the nature of public goods games.
Indeed, in both public goods games (Bouma, Bulte, & Van Soest, 2008) and natural resource management (Davenport et al., 2007), cooperation is known to vary with the levels of trust held between participants. Trust is also known to heavily shape stakeholder responses to wildlife management efforts, including where these efforts are contested (Riley et al., 2018;Schroeder et al., 2017;Young, Searle, et al., 2016). However, in some cases, higher trust actually leads to reduced engagement with interveners, as individuals have confidence that the interveners will act competently, and in their interest, without their involvement (Smith et al., 2013;Terpstra, 2011 What was unexpected, however, was our finding that perceptions of integrity and benevolence were stronger predictors of cooperation than ability. This finding also appears to contradict sentiments expressed in our post-game debrief interviews, in which intervener competence was most frequently emphasized. However, we cannot ascertain to what extent post-game justifications reflect post hoc rationalizations or actual drivers of game behaviour. From previous studies (Riley et al., 2018;Rudolph & Riley, 2017) it is clear that beyond rational outcome-based assessments, perceptions of intervener integrity and benevolence are also important in dictating responses to wildlife interventions, but the relative importance of each construct is less clear. For instance, perceptions of managers' trustworthiness (including perceived levels of deception) have been shown to influence stakeholder compliance (Shirley & Gore, 2019) and cooperation or support for interventions (Hamm, 2017;Ordoñez-Gauger, Richmond, Hackett, & Chen, 2018).
In some quantitative (Hamm et al., 2016), and qualitative (Wald, Nelson, Gawel, & Rogers, 2018) (N = 132), who justified in-game cooperation with reasoning related to: the ability, benevolence or integrity of themselves or the intervening groups (bar colour), the benefits of cooperation generally or concerns related to wildlife conservation trustworthiness constructs on responses to conflict interventions may be context-dependent.
From our interviews, and from previous studies in the region, there appear to be numerous factors which are likely to shape the (often negative) perceptions of the trustworthiness of the local conservation managers studied here. In this study location (Homewood, 2017), and across Tanzania, trust in conservation has been depleted by community displacement, resource restrictions and broken promises (Bluwstein et al., 2016;Moyo, Ijumba, & Lund, 2016;Wright, 2017). Elsewhere, interpersonal trust in natural resource managers has been found to be shaped by the perceptions of their responsiveness, honesty and dedication (Davenport et al., 2007). In our interviews, respondents commonly identified a lack of transparency, compassion and accountability within the WMA and previous conservation programs. Such experiences might explain the greater importance placed on perceptions of benevolence and integrity, which inform affinitive trust-based assessments (Stern & Coleman, 2015). Nonetheless, conceivably our analysis failed to capture the effect of ability-perhaps due to the relatively lower variation observed for this component. Indeed, in debrief interviews interveners' ability was by far the justification most commonly used by participants to explain their game behaviours. Likewise, it is likely that the relative importance of each component of trustworthiness is context-dependent. For instance, intervention preferences (Keane et al., 2016), risk perceptions (Gore & Kahler, 2012) and trust (Shirley & Gore, 2019;Xiao & McCright, 2015) have been shown to differ across different groups and geographies. Hence, rather than identifying which trustworthiness constructs generally matter more, this study highlights that they matter differentially.
Our approach further demonstrates the value of using experimental games to study conservation interventions (Andersson et al., 2018).
Previous studies (Hamm et al., 2016;Smith et al., 2013;Wald et al., 2018) have gained rich insights into stakeholder cooperation or support using surveys or interviews. By contrast, games enable experimental manipulation, but unrealistic incentives can lead to poor external validity (i.e. low correspondence to real-world behaviour ;Redpath et al., 2018). Here, this pitfall is minimized as our conclusions rest on relative, not absolute, differences in cooperation between groups.
Likewise, although within-game behaviour and consistency can be influenced by other factors (e.g. group dynamics, or game-order, communication, or game-understanding; Andersson et al., 2018;Aswani et al., 2013;Cookson, 2000), here the lack of variation in cooperation across rounds demonstrates apparent consistency in decision-making throughout the game, and our mixed effects model accounted for between-group variation. However, we did find that having WMA in the first round significantly reduces cooperation in the subsequent round with Moran, which demonstrates that game decisions were at least somewhat influenced by participants' previous experiences within the games (in this instance, lower levels of previous group cooperation).
Furthermore, one constraint on the external validity of the study stems from the lack of disclosure and discussions of in-game decisions. This is because in reality, wildlife guarding and related cooperation decisions would likely be publically known and subject to discussion and influence between individuals (Lowassa, Tadie, & Fischer, 2012;Travers, Clements, Keane, & Milner-Gulland, 2011). For instance, on the basis of previous research (Andersson et al., 2018), we suggest that more open communication in our game would have increased cooperation, particularly where perceptions of intervener trustworthiness were lower, and that such possible communication effects warrant further study.
When using experimental games to study behaviour, it is also good practice to draw upon qualitative data to validate and contextualize the experimental results (Redpath et al., 2018). Triangulation between pre-game interviews, game behaviour and post-game debriefs also gives us confidence that players were making their decisions based upon their own experiences with each group and were interpreting the target concepts in the intended manner. Moreover, the negative association of the rich wealth category, and positive association of primary education, with game cooperation should be interpreted carefully due to the wide confidence intervals of their estimated effects ( Figure 1); however, similar demographic effects have been found elsewhere to shape trust-based responses to resource managers (Shirley & Gore, 2019 (Davenport et al., 2007;Riley et al., 2018;Smith et al., 2013).
Those designing conservation interventions should therefore closely consider stakeholder relations and messenger effects (Dolan et al., 2012;Veríssimo, Tully, & Douglas, 2019), and explore whether particular individuals, third-parties or local institutions might be more effective in delivering interventions than others if they are more highly trusted or trained in stakeholder engagement (Riley et al., 2018;Sommerville, Jones, Rahajaharison, & Milner-Gulland, 2010;Young, Searle, et al., 2016). Secondly, our findings that beliefs about an intervener's integrity and benevolence were stronger predictors of cooperation than beliefs about their ability suggest that technical interventions or enforcement (such as efforts to reduce crop-raiding or illegal killing) might benefit from accompanying efforts to improve the perceptions of trustworthiness and build greater affinitive-based trust between resource users and resource managers (Rudolph & Riley, 2017;Stern & Coleman, 2015).
Thirdly, this study further suggests that improving the perceptions of trustworthiness and building affinitive-based trust is likely important in improving the success of community-based conservation programs Shirley & Gore, 2019). Although challenging, improving greater stakeholder trust in such community-based settings can be realized in several ways, ranging from improving communication and transparency (Rudolph & Riley, 2017), to active mediation efforts (Madden & McQuinn, 2015) and collaborative decision-making (Mishra et al., 2017;Young, Searle, et al., 2016), to avoiding making unrealistic promises (Mabele, 2017).
While these findings need to be tested across a range of contexts, elsewhere, the levels of protest, illegal harvest and active opposition to conservation have been found to be associated with perceptions of managers' honesty and fairness (linked to integrity; Shirley & Gore, 2019;Stern, 2008a), perceptions of care and community-mindedness (linked to benevolence; Hamm et al., 2016) and general levels of affinitive trust (Stern & Baird, 2015). Nonetheless, wildlife impacts (Cusack et al., 2018) and trust (Stern, 2008b) can shift over time, and that continued engagement and responsive approaches may be required for long-term conservation success (Butler et al., 2015). Future work could explore the relative importance of other types of trust, such as negative trust and systems-based trust (Stern & Baird, 2015), uncertainty (Pollard et al., 2019), how perceptions of trustworthiness are shaped by demographic background, cultural affiliations and norms (Rizzolo et al., 2017), and how insights from experimental games correspond to those derived from other methods.

CO N FLI C T O F I NTE R E S T
Steve Redpath was an Associate Editor for People and Nature at the time this manuscript was being reviewed, but was not involved in the peer review and decision-making process. We declare no other conflict of interest.

AUTH O R S ' CO NTR I B UTI O N S
All the authors helped in designing the research approach and wrote the manuscript. Z.B.-H. carried out the data collection and analysis.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data for this study have been anonymized and filtered to remove any sensitive or identifying information in accordance with ethical