An experimental game to examine pastoralists' preferences for human– lion coexistence strategies

1. Reconciling conflicts between wildlife conservation and other human activities is a pervasive, multifaceted issue. Large carnivores, such as the African lion Panthera leo are often the focus of such conflicts as they have significant ecological and cultural value but impose severe social and financial costs on the communities that live alongside them. 2. To effectively manage human– lion conflict, it is vital to understand stakeholder decision-making and preferences regarding mitigation techniques and coexistence


| INTRODUC TI ON
Negative interactions between people and wildlife, often termed human-wildlife conflicts (HWCs), emerge when the presence or behaviour of wildlife poses actual or perceived, direct, and recurring threat to human interests or needs, leading to negative impacts on people and wildlife (IUCN, 2020;Nyhus, 2016). Examples include damage to crops and predation on livestock , direct attacks on humans , disease transmission between wildlife and people (Gibb et al., 2020) and retaliatory killing or lethal control of the species involved (Ontiri et al., 2019).
Human-wildlife conflicts exemplify the fundamental challenge of reconciling local concerns for security and development, with international concerns for saving threatened species Treves et al., 2006). HWCs may, therefore, be reframed as human-human conflicts or, more broadly, conservation conflicts (Redpath et al., 2015), that is, conflicts between stakeholders with differing goals and values rooted in economic, socio-political and cultural history (Redpath et al., 2013). Escalating anthropogenic pressures and rapid human encroachment into remaining wilderness areas (Haddad et al., 2015;Watson et al., 2014) increase the risk of contact between humans and wildlife. The management of conservation conflicts to mitigate negative outcomes for biodiversity and wellbeing is, therefore, becoming increasingly important.
Apex predators such as the African lion Panthera leo are often the focus of conservation conflicts. Lions are a conservation flagship species, with important ecological, cultural and economic value (Di Minin et al., 2012;Ripple et al., 2014;Stolton & Dudley, 2019). Yet they pose a severe threat to human life (Packer et al., 2005;Sommers et al., 2010) and can incur considerable social and financial costs on the communities who live alongside them (Kissui, 2008;Packer et al., 2005).
Livestock depredation is widespread, with lions targeting culturally and economically valuable cattle (Muriuki et al., 2017). Attacks on humans, although rare, generate intense hostility and greatly increase levels of fear . Other impacts, including mental health issues and opportunity costs due to movement restriction or the need for guarding are less well understood (Barua et al., 2013;Dickman & Hazzah, 2016). Lion populations have declined by >43% over the last two decades (Bauer et al., 2016), with local extinctions in several regions of the species' range (Riggio et al., 2013). Realised or perceived negative impacts of lions on communities and subsequent retaliatory killing is a key driver of these declines (Bauer et al., 2016;Dickman et al., 2014).
This issue is exemplified in the Ruaha-Rungwa landscape of southcentral Tanzania, a high priority region for carnivore conservation.
This landscape is considered to be one of only four lion 'strongholds' in East Africa (defined as areas with stable populations of >500 individuals based on 2006 IUCN assessments; Riggio et al., 2013).
Previous surveys of communities in this area revealed that nearly two-thirds of respondents had experienced livestock depredation , with lion and spotted hyena being responsible for the majority of attacks. Consequently, lions in this landscape have experienced high levels of human-induced mortality due to retaliatory and preventative killing, with at least 136 lions killed between (Coals et al., 2020Dickman et al., 2014).
Identifying solutions for managing human-lion conflict is, therefore, a priority for people and wildlife in this landscape, as elsewhere.
Conservation advocates tend to assert their interests through legislation and enforcement which renders lethal retaliation illegal and/or socially unacceptable (Carter et al., 2017;Redpath et al., 2017). To reduce livestock losses to lions, technical interventions are often implemented and include physical barriers, improved guarding and nonlethal deterrents such as visual and auditory scaring devices (Lesilau et al., 2018;Lichtenfeld et al., 2014;Miller et al., 2016). Conservation performance payments, providing financial incentives conditional on a specific conservation outcome, have been suggested as an additional strategy to encourage human-wildlife coexistence (Dickman et al., 2011;Zabel & Holm-Müller, 2008). Examples include payments based on the number of carnivore reproductions that occur on village land (Zabel & Holm-Müller, 2008), or incentives to protect habitats and set aside areas of land to be free of human use (Mishra et al., 2003;Nelson et al., 2010). The effectiveness of these management interventions varies widely and is context-dependent (Eklund et al., 2017;Miller et al., 2016) but particularly for performance payments, there are limited operational examples from which to assess effectiveness and acceptability (Nelson et al., 2010;Persson et al., 2015).
Experimental games are emerging as a low risk, low-cost tool in the exploration of conservation conflicts and acceptable mitigation solutions (Rakotonarivo, Bell, Abernethy, et al., 2021;Rakotonarivo, Jones, et al., 2021;Redpath et al., 2018). Games allow affected stakeholders to explore the potential of various mitigation methods, including those that are sensitive, such as lethal control, in a safe and relaxed atmosphere . Experimental games have been used to understand the effects of payments and incentives in a range of situations. These include harvesting of fish from protected areas (Travers et al., 2011), pesticide use and cooperative management (Bell et al., 2016), forest resource use (Andersson et al., 2018) and preferences for conflict interventions in response to crop-raiding .
In this study, we developed an experimental game framed around lions and livestock protection, for a case study of rural, engaging approach that can be used to explore preferences and encourage discussion of conflict management.

K E Y W O R D S
African lion, conservation conflicts, experimental games, human-wildlife conflict, incentives, stakeholder engagement pastoralist communities in the Ruaha landscape of Tanzania. First, we use the game to explore individual preferences for coexistence strategies and to test the effects of three incentive structures on players' propensity to scare lions using non-lethal deterrents and to provide wildlife habitat. Second, using structured survey data we explore the relationship between within-game behaviour and attitudes/characteristics known to affect stakeholder support for conservation, such as wealth, education, and experience of stock loss (Hazzah et al., 2017;Kideghesho et al., 2006).
The three incentive treatments include (1) support to compensate the cost of non-lethal deterrents, (2) individual subsidies for provision of wildlife habitat and (3) community subsidies for provision of wildlife habitat. We hypothesised that (1) in a scenario where no incentive was offered for any action which benefits lion conservation, stakeholders would prefer the use of non-lethal deterrents over sacrificing land for wildlife habitat and (2) in a scenario where stakeholders received subsidies for providing wildlife habitat, the number of habitat choices would increase, and more so when subsidy levels were higher.

| Study area and population
We conducted experimental games in village land adjacent to the Ruaha National Park (RNP) and Pawaga-Idodi Wildlife Management Area (WMA), in the Iringa Rural District of Tanzania ( Figure 1). The Ruaha landscape is home to globally important populations of several threatened carnivore species, harbouring an estimated population of 3779 lions, at least 200 adult cheetahs, and the third largest population of the endangered African wild dog (Abade et al., 2014;IUCN, 2007;Strampelli et al., 2021). The village land located on the south-east border of RNP is inhabited by more than 60,000 people from at least 30 different ethnic groups (Abade et al., 2019). The dominant ethnic group are the Hehe, who are agriculturalists indigenous to the region. However, in recent decades there has been a significant in-migration of pastoralists and agro-pastoralists including the Bena, Sukuma, Barabaig, and Maasai (Dickman, 2008).
There are no fences separating RNP and the WMA from the village land and, thus, wildlife can move freely. Local communities previously reported highly negative attitudes towards lions and other carnivores due to their predation on livestock .
Although depredation accounts for a small percentage of total stock loss, direct experience of depredation is widespread and generates intense hostility . The importance of Ruaha's carnivore populations and the level of human-carnivore conflict in the area highlight the need for continued research into humancarnivore interactions and potential coexistence strategies in this landscape.
During the day, livestock graze across the village land, typically guarded by a herder and untrained guarding dogs. At night, herds are contained in enclosures made predominantly from thorn bushes.
The use of non-lethal deterrents such as lights, horns or fire is relatively rare in this area (Abade et al., 2014). Cultural lion hunting by some ethnic groups in this landscape has traditionally been used by young men to gain social status. These traditional hunts are now uncommon (though they still occur, with two events in 2019). Levels of where retaliatory hunts serve some traditional purpose, with young men receiving community accolades and sometimes gifts for killing problem lions (Dickman, pers. comm).
The Ruaha Carnivore Project (RCP), a research organisation established in 2009 (now part of Lion Landscapes; www.lionl andsc apes.org), provides several benefits to local communities including assistance for building wire bomas, school scholarships and employing local warriors as 'Lion Defenders' to assist with livestock protection. A camera-trapping initiative, known as 'community camera-trapping' is also used to link community benefits directly with wildlife presence. Camera traps are placed on village lands and each village receives points for any wildlife photos captured. At the end of each quarter, these points are translated into benefits requested by the community such as medical supplies, school supplies and veterinary medicines.

| Game design
We developed an experimental game to be played by four participants using tablet computers linked by a mobile hotspot. The game was created using NetLogo (Wilensky, 1999), a multi-agent modelling environment, and is part of a suite of NetLogo-based games developed to examine conflicts between people, wildlife and resources (Bell et al., 2016;Rakotonarivo, Jones, et al., 2021).
Each of the four players makes management decisions on nine cells arranged in a 3 × 3 contiguous section of the game board ( Figure 2). In each of these cells a player can either (1) graze livestock, (2) graze livestock and scare lions off cells using non-lethal methods (horns, lights), (3) graze livestock and attempt to spear lions in the cell (lethal control) and (4) leave the cell free from livestock to provide habitat for lions and wild prey (lion habitat). Each option has different costs, benefits and parameter settings ( Table 1). The parameter values and game settings were pre-tested by ~50 players, made up of community members and RCP staff, prior to data collection.
In each game round, there are a number of lions in the landscape which select cells based on the 'attractiveness' of the landuse options. Each management choice has a 'weight' assigned to it, with bigger weights meaning higher probability of attracting lions (Table 1). Wildlife habitat is the most attractive option, given that lions will avoid people and hunt wild prey where possible (Patterson et al., 2004;Valeix et al., 2012). Lion habitats have a 'neighbourhood effect' of adding to the weight of any cell around them (Table 1) natural habitats (Loveridge et al., 2010). Lions on grazed cells generate livestock losses ( Table 1). The participant's overall score on their set of n = 9 squares is calculated as: Each game starts with 16 lions 'roaming' the landscape. Lions move independently and stochastically among the landscape cells and decisions of players in the current round can affect payoffs in future rounds. 'Scaring' displaces lions from cells that experience it with a probability of 80% (Table 1). In any given round, if a player chooses to scare on any of their cells, lions in those cells will move to another location probabilistically based on cell weights. If a player chooses the lethal option, the lion will be removed from the landscape with a probability of 30% ( Table 1). The number of lions in subsequent rounds therefore decreases based on the frequency of killing choices. 'Failed' attempts at scaring or spearing result in lions remaining in the cell on which the decision has been made. Once all four players have confirmed their decisions in the current round, the position of lions in the landscape will reorient for the following round.

| Experimental design and data collection
Each session started with a short practice game of three rounds, to enable players to become familiar with the tablets and the game setup. This was followed by four randomly ordered treatments of six to eight rounds each ( Table 2). The number of rounds was randomised to prevent participants from anticipating the conclusion of a game. Treatments were designed to emulate common conservation interventions. For example, the Support for Deterrents treatment represents free provision of equipment such as lights or horns.
The bonuses for provision of wildlife habitat are similar to conservation performance payments, offering financial or other incentives for particular conservation actions. Incentives may be provided to the individual actor (Individual Subsidy treatment) or given to the community (Community Subsidy treatment), for example via donation of funding/supplies to schools or clinics.
Communication between participants was permitted at the start of each round to mirror real-world decision-making. Players were informed of the changes in parameters before beginning each new treatment ( Table 2). Following the game, we administered a questionnaire survey with each player (Table S1). This enabled us to collect detailed information on participants, including household demographics and socio-economic characteristics, experience with carnivore conflict and attitudes towards wildlife, trust and equity.
The games were facilitated by a three-person team (R.S. and two local field assistants) between May and July 2019. The game settings, instruction protocol and questionnaires were piloted prior to data collection in April 2019. Games were implemented in Swahili, with instruction also provided in two local languages, Maa and Barabaig.
The study was approved by the Newcastle University Faculty of Science, Agriculture and Engineering Ethics Committee (19-SAR-010) and the Tanzania Commission for Science and Technology (2019-95-NA-2018-348). We received oral or written consent from participants (dependent on literacy) following provision of information sheets (written in Swahili) and detailed spoken translations.
Participants were informed that all data would be anonymised, and only aggregated results would be published. Fieldwork was hosted by RCP, who provided logistic support, field assistants and local contacts. We ensured that before each session all participants were aware that the lead author was independent of RCP, that only the aggregated results would be shared with RCP and emphasised the neutrality of the research.
Grazing score n + Subsidy n − Lion damage n − Costs n . In total, we conducted 43 game sessions across eight villages context (Bell et al., 2015;Meinzen-Dick et al., 2016;Rakotonarivo, Bell, Abernethy, et al., 2021). Our aim was to ensure that players used the game as a tool to express their preferences and think about how they would behave in reality, rather than fixate on rewards (Hur & Nordgren, 2016) and aim to win.
To gain additional details on participants' rationale for their decisions and attitudes towards management options, at the end of study period we invited 20 random participants to attend a debriefing. These informal discussions were conducted in groups of 5 and were not audio recorded. During these 30-min debriefings, notes and direct quotes were taken to further understand participants' reasoning and motivations.

| Data analysis
We examined two game outcomes measured at the individual participant level: (1) decisions to provide wildlife habitat and (2) decisions to scare lions. We modelled these outcomes as the number of scare or habitat provisioning decisions using linear mixed-effects (LME) models in the nlme package (Pinheiro et al., 2021). Both outcomes were measured as counts and were, therefore, log transformed to normalise the data. Player ID was included as a random effect (RE) nested within Game ID, to account for unmeasured individual and inter-group variation. As the data consisted of repeated measures of the same individuals, we also controlled for learning by including round in the game as an explanatory variable, and by including an autocorrelation structure to adjust for serial correlation between rounds.
To examine the lagged effect of one round on the next, we ran For each game group we also calculated the mean of several sociodemographic variables and categorised gender and ethnic group as either mixed or uniform. We then compared these variables with the RE effect of game ID to examine group-level variation. Because the data were not normally distributed (they contained categorical and ordinal data) we used the heterogenous correlation function of the polycor package (Fox, 2019) to estimate polychoric and polyserial correlation coefficients. We used this post-hoc analysis approach to avoid over-parameterisation of the LME due to the large number of player characteristics measured and the mixture of numeric, categorical, and ordinal data.
We used a bootstrapping procedure to resample the observed data and constructed mean correlation coefficients (r) and con-   (Table S3).
Participants reported highly positive attitudes regarding trust and equity ( Table S3) Individuals were asked to rate their familiarity with each of the other three players on a scale of 1-4 (from '1: I hardly know this person' to '4: I know them well'), to obtain an overall familiarity with the group ranging between 3% and 12. 87% of players had total scores of ≥9 indicating that, unsurprisingly for a small rural community, the majority of players knew each other. When participants were asked what their main goal was when playing the game, 48% said it was to win and 43% said it was to behave as they would in real life (Table S3).

| The effect of game variables on players' willingness to provide wildlife habitat
In the Baseline (no intervention) treatment, on average, the proportion of cells over which players chose to graze livestock without using any deterrents ranged between 59% and 64% across rounds, whilst mostly choosing non-lethal scaring in the remaining cells <4% of choices ( Figure 4). We, thus, focussed our analysis on the provision of wildlife habitat and use of non-lethal deterrents.
Individual Subsidies generated an increase in the amount of wildlife habitat provided in comparison to the Baseline, although the effect was strongest for higher subsidies (i.e. 8 and 12; Table 3; Figure 4). For the Community Subsidies, a subsidy of 4 points had no effect on number of habitat choices, while 8 and 12 points did increase habitat provision in comparison with the Baseline, but to a much lower extent than the Individual Subsidies (Table 3;  (Table 3).
There was a moderate autocorrelation (ϕ = 0.43, 95% CI [0.40, 0.46]), indicating a positive correlation between choices of individuals across rounds. There were also substantial REs for Player ID and Game ID, indicating variation between subjects and groups. 15% of the residual variation in the model could be explained by Player ID and 24% by Game ID (Table 3).
An additional model, excluding round 1 data, allowed us to examine the lagged effect of one round on the next. Round number and lion damage suffered in the previous round did not improve the model and were removed (see Table S4 for full summary of

| The effect of game variables on players' willingness to use non-lethal scaring
Support for Deterrents significantly increased the number of scare choices in comparison to the Baseline (Table 5; Figure 4). Offering Individual Subsidies reduced the amount of scaring as players began to switch to habitat provision ( Table 5; Figure 4), with the higher subsidies of 8 and 12 having the greatest effect. When offering Community Subsidies, only subsidy level 12 had an effect and resulted in reduced scaring ( Table 5).
For the Baseline treatment, round number did not influence the number of scare choices. However, as participants progressed through the Support for Deterrents game, the effect of this treatment on number of scare decisions increased (  (Table 5).

| Correlation of participant characteristics with random effects
The RE of Player ID explained ~15% of the residual variation in the models for both habitat choices and scare choices (Tables 3 and 5).

Most of the players' demographic and attitudinal characteristics
were not correlated with the REs (see Table S5   Likert scale from −2 (not at all) to 2 (very much) and the correlation suggests that those who responded more positively to these questions made more scare choices.
Several of the socio-economic and attitudinal characteristics were intercorrelated (Table S5) (Table S6).

| DISCUSS ION
Large carnivores can have profound impacts on the wellbeing of communities in rural landscapes. We provide evidence that experimental games can be used to evaluate the acceptability of mitigation interventions to encourage human-carnivore coexistence. Our results support our hypothesis that stakeholders in the Ruaha landscape, Tanzania, prefer the use of non-lethal deterrents in the absence of incentives, and this preference increased when the cost associated with this intervention was reduced to zero. Non-lethal deterrents include anything you might use to scare an animal, but most participants interpreted this option as a torch. Monetary payments based on individual habitat choices incentivised players to provide wildlife habitat and this effect strengthened with increasing subsidy levels.
However, when the subsidy was shared based on community-level habitat choices only the mid to high subsidy levels increased habitat provision and to a much lesser extent than the individual-based subsidies, with a decreasing effect across rounds. Finally, gender and wealth (here indicated by size of livestock herd) and the participants' social status appear to affect, if weakly, decision-making on interventions.
Increasing payment levels has been found to improve engagement in pro-conservation behaviour in several studies (Handberg & Angelsen, 2019;Rakotonarivo, Jones, et al., 2021;Tuanmu et al., 2016). However, our results suggest that these incentives are more effective when targeted towards individuals rather than being shared by the group, concurring with several other studies (Gatiso et al., 2018;Midler et al., 2015;Ngoma et al., 2020). In Africa, due to the history of community-based resource management and ill-defined land tenure and land-use rights (Dickman et al., 2011;Goldman, 2011), incentive schemes have typically targeted groups rather than individuals (Gatiso et al., 2018) despite mixed evidence for the success of such approaches (Hayes et al., 2019). Communitybased payments require functional systems of collective action (Dickman et al., 2011) and create a social dilemma with the potential for free-riding, where non-complying individuals can benefit from payments while not bearing the cost . Our data provides evidence of this dilemma as habitat provision in the community treatment declined across rounds, suggesting that individuals may have been adapting their strategy in response to other players' cooperativeness (Narloch et al., 2012).
We highlight, however, that games may not accurately reflect complex real-world social interactions, and we emphasise the importance of capturing narratives around game results. Informal debriefings provided anecdotal evidence that most participants approached the games 'to win' but, in reality, benefits for the community were considered important. For example, one player stated that "Because it's only a game, people are looking at points. In real life you cannot live without your community, you have to help each other". Thus, although people may respond more strongly to individual benefits, the importance of community action and incentives is likely to be more nuanced when relating to real-world management strategies. The debriefing discussions also revealed that the preferred mitigation options among participants were torches/lights and improved livestock enclosures.
The evidence for demographic and socio-economic variables such as age, gender and education being predictors of attitude and tolerance to wildlife is mixed (Kansky & Knight, 2014;Kideghesho et al., 2006;Kimmig et al., 2020), and thus our finding of the limited explanatory power of these variables is perhaps not surprising.
We found positive correlations between number of scare decisions and perceptions of being individually respected in the community and feeling that current wildlife management schemes respect local culture. Several studies have found that perceptions of trust and equity affect uptake of management strategies and cooperation (Baynham-Herd et al., 2020;Rakotonarivo, Bell, Abernethy, et al., 2021;Rakotonarivo, Jones, et al., 2021). This highlights the need for interventions to be developed in an inclusive manner, with fair representation of stakeholder interests and genuine participation in decision-making processes (Armitage et al., 2020;Redpath et al., 2017;Sjölander-Lindqvist et al., 2015).
Furthermore, we found some evidence that women made fewer habitat choices and that mixed-gender groups made fewer scare choices than men-only groups. Women may be less likely to engage in pro-conservation behaviour due to higher levels of fear towards wildlife , differences in values, priorities and time constraints (Meinzen-Dick et al., 2014;Yang et al., 2018) or feeling excluded from conservation initiatives (Homewood et al., 2020;Ogra, 2009). Globally, participation in conservation and natural resource management is still largely dominated by men (Coleman & Mwangi, 2013;James et al., 2021).
Cultural barriers in traditionally patriarchal societies, such as the Maasai and Barabaig, mean that women are often excluded from decision-making and political participation (Homewood et al., 2020;James et al., 2021).
The inclusion of women in community conservation initiatives allows for more diverse perspectives and is likely to lead to more socially optimal outcomes (Masuda et al., 2022;Sun et al., 2011).
However, this relies on women's voices being heard and respected.
Gaining a better understanding of the dynamics of mixed-gender groups, including variation in cooperation and leadership is, therefore, important. In particular, the number of women in a group may affect their ability to overcome social norms and personal reticence to actively participate in discussions (Agarwal, 2015). In our study, the correlations between gender and decision-making were weak (r < 0.3) and, given that we targeted heads of households, our sample size of women was small (n = 19). Future work could, therefore, examine these issues further by recruiting a more gender-diverse set of participants and including women-only groups. Future work could consider ways to reduce response bias and possible bystander effects, for example, by developing games which can be played anonymously by single players. However, similar work on human-elephant conflict in rural Gabon, where lethal control is also illegal, found that players were willing to shoot elephants in a game context to both protect crops and express discontent with current policies (Rakotonarivo, Bell, Abernethy, et al., 2021). This suggests that players are prepared to engage in illegal behaviour when playing experimental games (Redpath et al., 2018;Travers et al., 2011).

| CON CLUS IONS
Our findings suggest that incentive-based instruments are conducive to pro-conservation behaviour in this conservation conflict setting but that the amount and the level at which these incentives are provided is important. Future work could focus on identifying realistic thresholds for such payments. Non-lethal deterrents appeared to be the preferred mitigation strategy and assisting with the cost of provisioning and upkeep would likely increase their use. We highlight the importance of engaging with the entire community and ensuring that all stakeholders' opinions and traditions are respected when developing conservation initiatives. We also echo previous calls for a shift in focus from human attitudes to behaviour to better guide conservation management and assess program effectiveness (Nilsson et al., 2020).
This study provides some of the first evidence that games could play a role in investigating human-carnivore conflict management.
Generalisation of these results for the wider human-lion conflict context is challenging, given that our participants had significant exposure to a wildlife research organisation and external management interventions. However, we found that the games were well received by the community, with players rapidly understanding and engaging with the NetLogo interface. Our study thus adds to previous work demonstrating the value of experimental games as a tool to investigate stakeholder perspectives and decision-making (Baynham-Herd et al., 2020;Janssen et al., 2014;Redpath et al., 2018). This approach is highly adaptable and applicable across a wide range of conflict contexts (Bell et al., 2016;Rakotonarivo, Bell, Abernethy, et al., 2021;Rakotonarivo, Jones, et al., 2021). Ultimately, to achieve success in managing conservation conflicts, it is necessary to go beyond understanding the ecological system, to develop inclusive approaches that lead to genuine stakeholder participation and improved social outcomes (Redpath et al., 2017). gistics and data collection. We are incredibly grateful to all the community members who assisted us and to all of our study participants.

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data for this study are available at https://doi.org/10.6084/ m9.figsh are.20326188