The impact of uncertainty on cooperation intent in a conservation conflict

1. Stakeholder cooperation can be vital in managing conservation conflicts. Laboratory experiments show cooperation is less likely in the presence of uncertainty. Much less is known about how stakeholders in real-life conservation conflicts respond to different types of uncertainty. 2. We tested the effects of different sources of uncertainty on cooperative behaviour using a framed field experiment and interviews. The experiment compared a baseline scenario of perfect certainty with scenarios including either: (a) scientific uncertainty about the effectiveness of a conflict-reduction intervention; (b) administrative uncertainty about intervention funding; or (c) political uncertainty about the extent of community support. We applied these scenarios to a conservation conflict in the Outer Hebrides, Scotland, involving the management of geese to simultaneously meet both conservation and farming objectives. We asked 149 crofters (small-scale farmers) if they would commit to cooperate with others by helping fund a goose management plan given the three sources of uncertainty.


| INTRODUC TI ON
Conflicts in conservation are ubiquitous around the globe and are damaging to both conservation efforts and people's lives (Redpath et al., 2013). Fostering cooperation between stakeholders with conflicting values is a priority of conservation conflict management as it builds trust and reduces conflict, both under experimental conditions and in real-life (Yamagishi, 2005;Young et al., 2016b). One important factor that reduces the chances of achieving cooperation in conflict is uncertainty, which will generally decrease the tendency to trust and cooperate (Rapoport, Sundali, & Seale, 1996). Rittel and Webber (1973) describe three broad sources of uncertainty in social ecological systems (SES): scientific uncertainty from incomplete knowledge of the research system; political uncertainty regarding power relationships and values; and administrative uncertainty surrounding cost and responsibilities.
Experimental economics methods have been used to test cooperation in collective-action problems (Cárdenas & Ostrom, 2004), including in the presence or absence of uncertainty. For example, Barrett and Dannenberg (2012) used laboratory experiments with volunteers to investigate decision-making in the context of climate change negotiations, showing that uncertainty of the position of an emission threshold resulted in lower cooperation than uncertainty surrounding the impacts of exceeding that threshold. However, volunteers in a laboratory setting will act differently to stakeholders in a real-world situation (Levitt & List, 2007). Working with stakeholders involved in a conservation conflict (rather than with volunteers) and framing the experiment in a way which reflects a real collective-action problem, allows real-life aspects of the conflict such as knowledge of the system, underlying values and perceptions of others, to be taken into account. Here, we use an experimental economics method to explore how three types of uncertainty (scientific uncertainty, administrative uncertainty and political uncertainty) influence the intention to cooperate, of people in a real-life conservation conflict.
Conservation conflicts involving the damage of crops by wildlife are widespread globally (Treves, Wallace, Naughton-Treves, & Morales, 2006). In Northern Europe, reduction of agricultural yield due to grazing of wild geese is a well-documented problem (Cusack et al., 2018;Simonsen, Tombre, & Madsen, 2017). Methods for reducing goose damage to crops include regulating population (e.g. shooting), non-lethal scaring or providing sacrificial feeding areas (Fox, Elmberg, Tombre, & Hessel, 2016). Stakeholders involved in a goose conflict can include those who: suffer directly from goose damage; wish to maintain the conservation status of the geese and their habitat; are responsible for scientific support of management; are required to fulfil practical management activities; and, provide funding or practical support. Mapping the specific stakeholders and uncertainties has been identified an important step in understanding the context for conservation conflict management (Redpath et al., 2013); however, less is known regarding how cooperative behaviour of stakeholders in a conflict is affected by different sources of uncertainty.
In this paper, we test how scientific, administrative and political uncertainties impact on stakeholders' willingness to cooperate on goose management in the Outer Hebrides, Scotland. Resident greylag goose (Anser anser) numbers have been increasing steadily from historic low points in the mid-twentieth century, to record highs. While this is seen by many as a conservation success story, the geese are responsible for damage to arable crops and to pasture intended for livestock. (Bainbridge, 2017;Mitchell, Griffin, Trinder, & Newth, 2010). The majority of agricultural activity in the Outer Hebrides takes place on crofts; small-scale farms of typically 5 ha, culturally unique to the more remote and less productive areas of the Highlands and Islands of Scotland. Crofting is regarded historically and legally as a distinct category of farming in Scotland and is recognised by the Scottish Government as being vital in maintaining the population of remote areas, supporting local businesses and managing important natural habitats (Scottish Government, 2016). Crofters (farmers of croft land) impacted by geese essentially take part in a form of public goods game, where they each choose whether to voluntarily contribute to the maintenance of a non-excludable, non-rivalrous public good (cooperate with goose management by contributing to scaring actions) or not (defect).
Defection is less costly in the short term where benefits of the public good can be obtained without contribution (elsewhere called free-riding) but runs the risk of losing the benefits should enough others do the same.
Presenting crofters with a set of four public goods scenarios for goose management-a baseline with no uncertainty and three treatments with differing sources of uncertainty-we aimed to: 1. examine how crofters' intention to cooperate was influenced by different types of uncertainty 2. determine which variables (e.g. crofting location, time spent as a crofter, experiences of goose damage) were most important for describing cooperative behaviour. will impact intention of stakeholders to cooperate; and (b) take steps (such as uncertainty reduction, communication or acceptance) to reduce the negative impact of uncertainty on cooperation.

K E Y W O R D S
conflict, conservation management, decision-making, experimental economics, goose, public goods game, uncertainty 2 | MATERIAL S AND ME THODS

| Study area
North Uist, Benbecula and South Uist (hereafter, the Uists), are part of the Outer Hebrides; an island chain off the northwest coast of Scotland, UK. The Uists provide year-round habitat for greylag geese which damage both arable crop and pasture (Bainbridge, 2017).

| Crofter recruitment and data collection
In August 2016, a list of all crofts in Uist (N = 1,579) was obtained (Registers of Scotland, 2016). Potential interviewees were sequentially approached down a randomised copy of the list, until the end of the data collection campaign in November 2016. This resulted in 149 crofters agreeing to be interviewed. We used face-to-face interviews to ensure crofters' understanding of the questions and to capture qualitative responses accurately. Information from crofters on themselves, their crofting and their experiences of goose impact was collected using a structured questionnaire, to allow statistical analyses on the data collected (Newing, 2011

| Willingness to pay
Crofters were asked if they would be willing to pay (WTP) an annual fee along with other crofters, for a project which would completely mitigate all the negative impacts of the geese, using a contingent valuation technique (Pearce, Ece, & Özedemiroglu, 2002). Those who were unwilling to pay were asked to give reasons. The responses were then coded post-hoc using theoretical thematic analysis (Braun & Clarke, 2006). Those who responded that they would be WTP were then asked to indicate how much they would pay annually into a fund with other crofters for 100% mitigation of the negative goose impacts (hereafter, the WTP amount or C wtp ). The primary aim was to identify a WTP amount for each individual which could then be used in the subsequent cooperation scenario. This was done to account for individual differences in value placed on goose impact reduction.
The stated WTP amount was then repeatedly used in each cooperation scenario (see below). Where crofters were WTP but could not specify an amount, the modal WTP amount identified during piloting (£50 per year) was used as C wtp .

| Cooperation scenario
We presented crofters with four scenarios, each detailing a hypothetical goose management plan, using summary cards (see Figure   S1). Crofters could choose to either support the plan (intention to cooperate), or not (intention to defect). Both choices incurred a cost to the crofter, a resulting reduction in goose impact, and a threshold number of crofters that would be required for the management plan to be enacted. This choice is akin to a public good game, where the crofter's payoff (a utility function made up of the sum of the level of goose impact and cost of joining a goose management plan) is dependent on their own course of action as well as the actions of others (to meet the threshold number required) ( Table 1). The goose management plan outlined in the baseline scenario resulted in a decrease of negative goose impact (C d ) down to half the current impact levels. The WTP amount (C wtp ) previously stated by the crofter was for 100% reduction in negative goose impact. Therefore, the cost to each crofter (C mp ) of a management plan which achieved half that reduction as is the case in the baseline scenario, was 0.5C wtp .
The management plan was presented as receiving partial payment from government funds equal to 0.25C wtp , so a cooperating crofter would receive a 50% reduction in goose impact for a C mp = 0.25C wtp .
This resulted in a total payoff to the crofter of C d + C mp = 0.5C wtp + 0.25C wtp = 0.75C wtp . However, the hypothetical management plan needed the number of crofters signing up (N c ) to be at least half of all the crofters in the Uists (N). If this threshold (N c /N) was not reached, crofters did not pay anything (C mp = 0) but there was no goose impact reduction (C d = C wtp ), so total payoff is C d + C mp = C wtp + 0 = C wtp . Choosing to defect always set C mp = 0. The crofter then suffered the full negative impact if the threshold was not reached (as above), or if the threshold was reached the crofter received the benefit of impact reduction without paying for the cost.
Three other scenarios were the same as the baseline, but each contained a single type of uncertainty (Table 1): • The 'Scientific' scenario was described to crofters as representing managers' incomplete knowledge of goose ecology resulting in uncertainty to impact reduction, C d .
• The 'Administrative' scenario was described as representing managers' incomplete knowledge of public funding for the management plan resulting in uncertainty to the cost of the plan to the crofter, C mp .
• The 'Political' scenario was described as representing managers' incomplete knowledge of how much support would be needed for the plan to be initiated, resulting in uncertainty to the threshold of cooperation required from crofters, Th uc .
The baseline was always presented to crofters first, and the following three treatments were randomised. The fixed annual costs remain the same so as time increases, the average payoffs for all four scenarios become equivalent (Table S1). To evaluate the crofters' beliefs about how others would behave in the same scenario, we used a wager method. After each decision, crofters were asked to estimate what percentage of all the crofters in the Uists would cooperate, by splitting a hypothetical £20 wager between 20 equal cells each representing a 5% block of the population. For example, if the crofter thought that between 46% and 55% of others would cooperate, they would write '10' in each of the '46%-50%' and '51%-55%' cells. If the crofter felt they could not estimate or they felt there was an equal chance of all outcomes, they would write '1' in each of the 20 cells. A fixed wager allowed crofters to express confidence in their prediction, responding with the wager spread over a large or small range.

| Statistical analyses
To examine how uncertainty affects the intention to cooperate, as well as which background and impact experience characteristics most strongly predict intent to cooperate, we ran four linear mixed effects models. Analyses were focused on how intention to cooperate and WTP for goose management were influenced by three groups of variables. Firstly, the value a crofter places on cooperation may depend on their current situation including size of their croft, the extent of their crofting experience or their existing access to goose management support via the croft owner or LGMG. Secondly, intention to cooperate may stem from wanting to mitigate personal impacts of geese such as time and money costs. We also include variables to capture crofters wishing to mitigate goose impacts on their community or on natural habitats. Finally, crofters who are aware of existing goose management through formal organisations may support cooperation with other crofters, or conversely believe that responsibility lies elsewhere. The individual variables for each of the groups are shown Table 2.
Firstly, for each analysis a 'global' model was built containing the predictor and random variables thought relevant to that analysis.
The function 'dredge' (r package MuMin) was then used on the global models to build and rank models by finite-sample corrected Akaike information criteria values (AIC c ) calculated using maximum likelihood. No interactions between variables resulted in a better fitted model, according to AIC c . Best-fitting models (ΔAIC c < 2) were retained and were then standardised by dividing the continuous fixed variables by two standard deviations allowing direct comparison of coefficients between continuous and binary variables (Gelman, 2008).
The area under the curve (AUC) of receiver operating characteristic (ROC) plots was calculated for all models with a binary output TA B L E 1 Crofter payoff (per year) matrices under four treatments of varied uncertainty. Here, payoffs are costs to the crofter, so rational behaviour seeks to minimise total costs under each treatment. Total cost to the crofter in bold, is the sum of the respective cost of management plan (C mp ) and the cost of the negative goose impacts (C d ).   (Bolker, Skaug, Magnusson, & Nielsen, 2012) and pROC (Robin et al., 2011).

| Intention to cooperate
Two global models were built to investigate intention to cooperate.
The first willingness to pay (WTP) global model included all crofters, whereas the second only included those advancing to the cooperation scenario. Both models have a binary response variable (Cooperate/ Defect), so we used generalised linear mixed effects models (GLMMs) with binomial error structure and a logit link. For predictor variables included in the WTP global model (see Table 2

| Willingness to pay -amount
The WTP amount global model used the same predictor and random variables as the WTP global model above.

| Perception of others' intention to cooperate
Predictor variables included in this global model were the same as for the cooperation scenario, with the addition of the measure of how crofters compared their own goose damage with that of others (

| RE SULTS
All best fitted models had ΔAIC c ≤ 2 (Tables S4-S7 for the output of all best fitted models). Results from the simplest (lowest number of TA B L E 2 Variables measured for modelling intention to cooperate and willingness to pay for goose management. Not all predictor and random variables were included in all models predictor variables) of each best fitted model and predicted effect sizes are described below and in Tables 3 and 4. Population level data for each predictor variable used in the models can be seen in Supporting Information.

| Intention to cooperate
Most of the crofters who were interviewed (76.5%; 95% CI = 69.1%-82.6%) were WTP for goose management. Reasons for crofters being UTC are shown in Table 5. The most common reason under no uncertainty was that geese did not affect them enough. In the presence of each type of uncertainty, the most common reason given for UTC behaviour was the unsatisfactory risk of a worse outcome compared to the baseline scenario.
Crofters' concern for others and their time as crofters were the two significant predictor variables (Figure 1, Table 3). The longer an individual had been a crofter the lower the predicted probability of cooperation (e.g. 10 years of crofting P(coop) = 0.75; 50 years of crofting P(coop) = 0.51) and crofters who showed concern for others had a higher predicted probability of cooperation than those who did not (at mean time crofting (32 years), showing concern for others P(coop) = 0.86, no concern for others P(coop) = 0.63) ( Table 4). Fixed effects accounted for 13% of total variation in the model but there was essentially no variation between locations (Table S4). There was no significant difference (assessed by AIC c ) between models with and without the random variable. The AUC of the ROC was 0.72.
Under all treatments of the cooperation scenario, most crofters were WTP for goose management. Under the uncertainty scenarios, type of uncertainty was the only significant predictor variable for intention to cooperate (Figure 1, Table 3). In the absence of uncertainty (baseline), predicted probability of cooperation was >0.98 (Table 4). The presence of each of the three types (scientific, administrative and political) significantly decreased the predicted probability of cooperation compared to the baseline. The greatest effect was seen in the administrative scenario (P(coop) = 0.77), followed by small but significant effects with scientific (P(coop) = 0.93) and political (P(coop) = 0.98) (

| Willingness to pay -amount
The modal WTP amount was £50 per year and the mean £59.81 per year. Cost of goose scaring (time) and concern for others suffering damage were the two significant predictor variables for WTP amount (Figure 1, and those indicating concern for others would pay £52.27. The model variance attributable to crofter location (random variable) was 0.13 (Table S4).

| Perception of others' intention to cooperate
Individual cooperation, type of uncertainty, membership of SCF and perceived relative level of goose damage (Figure 1,

| How uncertainty affects crofters' intention to cooperate
When faced with a choice of discrete courses of action, people generally select those with lower uncertainty (Kahneman & Tversky, 1979;Lundhede, Jacobsen, Hanley, Strange, & Thorsen., 2015). This expectation is supported by our findings, with the presence of scientific uncertainty (from incomplete knowledge of the research system), administrative uncertainty (surrounding cost and responsibilities) and political uncertainty (regarding power relationships and values) each significantly decreasing the predicted probability of cooperation compared to a baseline scenario with no uncertainty.
Administrative uncertainty causes the largest decrease in terms of probability of cooperation. The administrative treatment was presented as uncertainty about whether public funding would be able to either pay all the cost of the management plan (thus, free for the crofter) or pay nothing towards the plan (doubling the cost to the crofter compared with other treatments). A view of shared responsibility was evident under the scenario of administrative uncertainty as the second most given reason for defecting was that others should contribute to goose management (Table 5). In this case, administrative uncertainty caused crofters to question the commitment of another stakeholder group, causing defection.
The negative effect of scientific uncertainty on probability of cooperation was small but statistically significant. Scientific uncertainty was framed as full enactment of management actions but with ecological uncertainty of how actions would affect the geese and the resulting level of damage caused. Here, defecting crofters did not mention a 32 years crofting − concern for others 0.63 (0.45-0.78) a 32 years crofting + concern for others 0.86 (0.78-0.92) 10 years crofting − concern for others 0.75 (0.45-0.92) 50 years crofting − concern for others 0.51 (0.10-0.91) (c) Willingness to pay − Amount Willingness to pay amount (£) −cost of goose scaring − concern for others 34. 16 (24.22-48.18) −cost of goose scaring + concern for others 52.27 (31.60-86.48) +cost of goose scaring − concern for others 73.98 (  TA B L E 5 Reasons given by crofters for choosing not to cooperate in the willingness to pay (WTP) and in the three scenarios with uncertainty. n = 138 for WTP and 97 for the other three scenarios.
Crofters were asked if they were WTP for goose management and if they indicated they would, then they were given four further choices (cooperation scenarios). The baseline treatment is not included in the table as there were no noncooperation responses. Sum of percentages may be greater than 100% as crofters could give more than one reason F I G U R E 1 Standardised effect size (±95% confidence intervals) of predictor variables on: intention to cooperate with other crofters on a cooperative goose management plan under different types of uncertainty (a) or with no uncertainty (b); amount willing to pay into a cooperative goose management plan (c); and crofters' prediction of others to cooperate (d). Outputs are from the simplest, best-fitting models. Effect sizes have been standardised *(p < 0.05); **(p < 0.01); ***(p < 0.001). Full model outputs in Tables S3-S6, for  other stakeholders (as with administrative uncertainty), so seemed to be reacting to uncertainty directly (Table 3). In this scenario, general aversion to uncertainty may be contributing to much of the decrease in intention to cooperate (Lundhede et al., 2015).
Compared to the baseline scenario, the decrease in effect size under political uncertainty was very small but significant. The uncertainty in this scenario affected how many other people crofters thought might need to get involved, but also changed the conditions for accessing benefits without contribution. The small effect size means we cannot separate decreased probability of cooperation under political uncertainty from the general negative utility experienced from any type of uncertainty (Lundhede et al., 2015).

| Describing crofters' cooperative behaviour
Financial loss via goose damage was not a significant predictor variable for any model. Crofters were more likely to cooperate on a goose management plan and would pay more into such a plan when they indicated concern for others suffering from goose impacts. This pattern of cooperation would be expected if goose management payments were seen more as a charitable donation than self-serving (Park & Lee, 2015). The probability of cooperation decreased with increased time as a crofter. This result may be driven by crofters approaching retirement as 20% of crofters who chose to defect gave the reason that they were exiting crofting soon.
Many crofters chose to defect but not one crofter indicated that they were aiming to gain benefits without contributing. Crofters may not want to gain benefits this way because they see it as unfair, or they would not want to be seen as being unfair by their community.
Small agricultural communities have strong reciprocal relationships between individuals (Sutherland & Burton, 2011), which can decrease behaviour perceived as unfair (Ostrom, 2010

| Predicting others behaviour
The largest predictor of whether crofters thought others would cooperate with each management scheme was their own preference to cooperate or defect. All types of uncertainty were also significant in the same direction and in the same rank order as with crofters' own choices. Individual crofters believed other crofters in the Uists would act similarly to themselves and did not indicate they thought others would attempt to gain benefits without contributing. Both these crofter predictions are consistent with the false consensus effect, where people project their own behaviour onto others (Ross, Greene, & House, 1977).

| Limitations of the method
The use of contingent valuation methods to accurately value goods and services has been criticised. For example, WTP suffers from hypothetical biases, differences between willingness to pay and willingness to accept values for similar goods, and assumptions about how goods may be embedded in one another (Hausman, 2012).
Hypothetical bias can be reduced by offering payments based on decisions made in the experiment, but it can unrealistically incentivise individualistic behaviour (Vohs, Mead, & Goode, 2008). Using the WTP amount from this study would not be appropriate for costing of a Uist goose management funding scheme. Where good, independent, data are available for goose management costs, a discrete choice experiment between alternative management actions could elicit a more accurate value than our contingent valuation (Johnston et al., 2017). The WTP variables in our modelling did not include a measure of personal wealth or income, which may be expected to have a significant influence on WTP amount (Pearce et al., 2002).
The aim of identifying an individual WTP amount for each crofter to use in the cooperation scenario was achieved with our method.
We focused on the predictor variables that significantly affect cooperation and on the difference between the treatments. However, people also tend to overestimate WTP amount when responding to scenario questions compared to real-life situations (Murphy, Allen, Stevens, & Weatherhead, 2005) and without social interaction people overestimate theirs and others' propensity to cooperate (Vlaev, 2012). Steps were taken to minimise biases of methodological origin, such as by discussing the scenarios in a neutral way. Crofters predicted that others would make very similar choices to themselves, which suggests any bias towards wanting to appear in a good light extended beyond themselves to promoting the community as a whole. Separating bias from the social norms which we are trying to study is an ongoing challenge in field studies such as this.

| Management implications of multiple system uncertainties
The three sources of uncertainty affected crofters' intention to cooperate in different ways. In the presence of administrative uncertainty, defecting crofters indicated that other groups should shoulder some of the burden caused by uncertainty. In the presence of scientific uncertainty, no actions by any other group were mentioned as being involved in crofter cooperation. In the presence of political uncertainty (and in general), cooperating crofters were confident that others would act like them and not try to gain benefits without contributing. Prior to management actions being developed, an important step is for managers to understand the societal dimensions of a conflict, including stakeholder roles and actions (Young et al., 2016a). Our study shows that managers should also include an assessment of how stakeholders' actions may change under different sources of uncertainty, especially if sources are associated with particular stakeholder groups.
Once relationships are better understood, steps can then be taken to cope with uncertainty. Firstly, uncertainty could be reduced by filling scientific research gaps such as the relative efficacy of scaring techniques or goose crop selectivity (Fox et al., 2016  .
Cooperation in the Uists over goose management has been established through formation of the multi-stakeholder LGMG and previous commitment to the 5-year adaptive management pilot.
The current level of cooperation between stakeholders may be at risk if future goose management plans cannot reduce administrative uncertainty (for example, by securing funding) nor demonstrate commitment to the project (for example, by enshrining another multi-year plan).

| CON CLUS IONS
Our work illustrates the potential differences in stakeholders' response to uncertainty in the form of cooperation. Reducing scientific uncertainty, at which conservation practitioners are likely to be most skilled, may not be the most important gap to fill. Variation in behavioural response to uncertainty can be taken into account throughout the conflict management process to target the most effective ways to either preferentially reduce uncertainty itself or increase the acceptance of uncertainty amongst stakeholders. Both tactics mark a way forward to reducing the impacts that uncertainty can cause.