This collection contains three packages of data relating to hunting and law enforcement in Keo Seima Wildlife Sanctuary, Cambodia: (1) a household survey intended to estimate the prevalence of different hunting behaviours and wildlife consumption, local communities’ knowledge of rules, and their perceptions of the ranger patrols responsible for enforcing rules, (2) an experiment designed to measure the ability of ranger patrols to detect snares in a tropical forest environment, and (3) an experiment designed to measure the length of time a snare remains an active threat after it is set. Human-caused environmental destruction is a major challenge to the sustainability of life on earth. For effective solutions, we need to learn about damaging behaviours and discover how best to encourage change. Exciting developments in fields concerned with human behaviour (such as economics and psychology) are helping to explain why people make the decisions they do. In parallel, ecologists have developed sophisticated methods for analysing data collected by ordinary people ("citizen scientists"), aided by new technologies such as smart phones. Up to now, these developments have remained separate, but closer integration would benefit both science and practice. Behavioural scientists would gain from the adoption of powerful new analytical techniques from ecology, which enable them to use data collected in new ways to understand how humans interact with the environment. Ecologists would benefit from being able to include a solid theoretical model of human behaviour into their understanding of how ecological outcomes arise from human actions. Managers and policy-makers will benefit from evidence-based understanding of how to change behaviour in the real world. To illustrate how powerful this combination of approaches can be, we will apply them to a key problem facing global conservation: how to manage protected areas so that they can act as effective refuges for endangered species in the face of illegal poaching and other threats. Learning about illegal behaviour is difficult because those involved are rarely willing to talk openly, so the 'conservation detective' must make deductions from other sources of information. Many conservation organisations now collect reports made by the rangers who patrol parks. This is potentially very informative, but also potentially very misleading. Consider snaring as an example: a ranger seeing a snare is the outcome of several interacting processes (where the poacher decides to lay their snare, where the ranger decides to patrol, and whether the ranger spots it in the undergrowth), and removing that snare may affect the future decisions of the poacher; so the data are the product of a game of cat-and-mouse played out in a dynamic landscape. This makes patrol data very hard to interpret. To tackle this issue we will build two types of computer model to explore how rangers and poachers interact with one another and their environment: i) conceptual models of the underlying processes that lead to the observation of a snare, based on ecological and behavioural theory and our understanding of our system, with simulated patrol records as their outcome; ii) statistical models that start with the snare data, and see which combination of factors best explains it. Building both models means that each can be used to inform the other. We will test the models in two ways; firstly in an abstract system, where we can vary the behaviour of the patrollers and poachers and the environment in which they interact, and see how this affects the resultant patterns of snare observations, and secondly in a real-world system, the Seima Protection Forest in Cambodia. Here we have substantial existing knowledge to help us to build our models, and will collect new information to improve our understanding. Our work will also be able directly to inform their conservation strategy. For the first time it will be possible to paint an accurate picture of illegal behaviour within parks and to give managers scientific advice about how to design their patrols. We will also explore how this novel approach can be used more widely to tackle other environmental issues.

(1) Household surveys: between February and April 2018, we interviewed respondents from 705 households in 18 villages. (2) Snare detection experiment: we adapted a methodology originally piloted by O’Kelly et al. (2018), and established five 3.25km transects around a patrol station. Either side of each transect, we delineated 6 x 0.25 sq km (500m x 500m) quadrats at 50m intervals. Within each quadrat we set between zero and 15 snares (the number was randomly drawn from a Poisson distribution with mean = 7.5), based on estimates of typical snare densities identified by other studies (Dobson et al., 2019). Single foot snares made from black nylon string (5mm), an inexpensive material often used by hunters in this area, were set without a trigger mechanism to prevent harm to wildlife, and all snares were successfully removed at the end of each transect survey. In total, 886 artificial snares were set, 442 in dry season and 444 in wet season. We recruited local guides from surrounding communities, who were instructed to set single snares as a local hunter might, in locations they deemed suitable to catch popular prey species such as wild pig (Sus scrofa), Northern red muntjac (Muntiacus vaginalis) and sambar (Rusa unicolor). Prior to setting snares, teams explored each quadrat for 30 minutes to identify suitable snare locations. (3) Snare persistence experiment: conducted between November 2017 and August 2018. We selected two sites with different forest types in which hunting is thought to occur. Site A was situated in a patch of scrub forest surrounded by chamkar. The habitat mainly consisted of large bamboo and is thought to suffer high levels of disturbance. Site B was situated deeper in the forest around a salt lick in an area known to support relatively high densities of wildlife including elephant. The forest here primarily consisted of small bamboo groves with semi-evergreen patches.