The educational value of virtual ecologies in Red Dead Redemption 2

1. Playing video games is often perceived as the antithesis of engaging with, and learning about, the natural world. Nevertheless, there is growing recognition that digital media is now a central part of many people's lives. This has led to increased efforts to harness the power and popularity of digital games for both ecological education and conservation advocacy.


| INTRODUC TI ON
Digital technologies have a potentially significant role in the future of education (Barab & Luehmann, 2003;Tewksbury et al., 2014), particularly in communicating with learners via media to which they are accustomed (Annetta, 2009). Non-traditional educational formats, including digital technologies, have many benefits, including novelty, immersion, a rich sensory environment and 'learning through play' (Behrendt & Franklin, 2014;Guerrero & Reiss, 2020;Waite, 2011).
Video games are an underexplored and underused educational tool. As they increase in complexity, they can provide immersive, sensory learning environments (Squire, 2011). Playing video games can have cognitive benefits (Granic et al., 2014), promote active learning (Johnson & Johnson, 2008;Michael, 2006;Prince, 2013) and provide learning autonomy, strengthening the acquisition of knowledge and skills (Mouaheb et al., 2012;Squire, 2008;Turkay et al., 2014). Research with higher education students has found that video games require players to exercise and improve a range of skills including communication, adaptability and resourcefulness (Barr, 2017(Barr, , 2018. Despite this potential, most efforts to use them in educational settings have been limited to teaching about coding and game design (Squire, 2008), rather than exploiting the possibilities of gaming to immerse players in complex learning environments and teach them about a wider range of subjects.
One subject that could benefit from digitally enhanced teaching is natural history. Broadly conceived as the 'description of nature' (Schmidly, 2005), natural history involves the study of organisms, environments, evolution and the changing relations between these (Tewksbury et al., 2014). Understanding of natural history underpins much of biology (Greene, 2005). Without it, we would not have ecology, evolutionary biology or animal behaviour (King & Achiam, 2017).
While education in natural history has declined over recent decades, there has recently been a movement towards reintroducing this more explicitly into curricula: in the UK, for example, OCR (a national examination board) has proposed a Natural History GCSE (OCR, 2020).
Knowledge of natural history, and its emphasis on biological diversity, further acts as a crucial tool for applied scientists working to address challenges in human health, food security, climate change and biodiversity conservation (King & Achiam, 2017;Tewksbury et al., 2014). We are not arguing here that knowledge about an organism automatically causes people to care about it (indeed, there is plenty of evidence to suggest that there is a minimal link between education and behaviour change; Schultz, 2011). Nevertheless, basic awareness of a species' existence is arguably fundamental to an interest in its conservation (King & Achiam, 2017), and ecological knowledge can both inspire people to care and empower those who do (Carlone et al., 2015).

| Video games and conservation education
Employing video games for conservation has potential (Sandbrook et al., 2015). Games have been successfully used for education and training purposes in other settings; they provide engaging, fun educational experiences that reinforce learning through repetition.
However, games risk: distracting players from the real world and its problems; and/or providing 'a hyper-real and cognitively satisfying alternative to real nature that reduces concern about conservation' (Sandbrook et al., 2015). Conservationists designing games to inform about and generate support for a cause must therefore strike a balance between entertainment, education and advocacy. Recent examples include Kakapo Run , designed to entertain while raising awareness of the plight of the kakapo Strigops habroptilus, and the forthcoming Wildchain (http://wildc hain.io), in which players donate funds to digitally adopt virtual representatives of real endangered animals.
Pokémon is a franchise comprising video games, television series, manga comics and a trading card game, in which players collect fictional animal species. Its creator, Satoshi Tajiri, used to collect insects as a child and designed Pokémon to recreate this experience for modern children (Balmford et al., 2002). Almost 20 years ago, Balmford et al. (2002) showed that children were better able to identify species of Pokémon than UK wildlife, arguing that conservationists have been less effective than game developers at teaching children species diversity and identification. In 2016, the augmented reality mobile game Pokémon Go also attracted interest from conservation scientists, who suggested that such games could be harnessed as educational and engagement tools (Dorward et al., 2017). Small-scale versions of this idea using real-world rather than fictional species have been trialled (http://pocke tpals app.com; https://www.safar icent ralga me.com/) but not widely promoted. Edwards and Larson (2020) recommend several strategies to engage digitally oriented young people with nature. These include effectively marketing games to reach and appeal to intended audiences and incorporating gamification techniques into educational programme design. These strategies respond to the challenge that 'nature-oriented' games are most likely to appeal to those already interested in natural history, and that educational games might struggle to reach broader audiences. Most educational video games are perceived as such and are therefore unpopular when compared with entertainment games. Educational games often have much smaller development budgets than those produced by major publishers and can therefore disappoint those who play premium games at home, However, games that are not designed to be educational, and that already reach broad audiences, might nevertheless contribute to ecological education. In short, if educational games are unpopular, can popular games be educational? Recent research has begun investigating the effects of entertainment-focused forms of visual media (e.g. movies and narrative nature documentaries) on people's understanding of, and interactions with, the natural world Fernández-Bellon & Kane, 2020;Siriwat et al., 2020;Veríssimo et al., 2020;see also Silk et al., 2018). Video games are another widespread form of visual media that have not yet received the same research focus. A notable exception (Truong et al., 2018) found a preference for green and natural landscapes among players of World of Warcraft, interpreted in terms of 'virtual biophilia': an attraction to healthy, vegetated landscapes that many found difficult to access in real life. Here we examine whether and how another popular video game (Red Dead Redemption 2: Rockstar Studios, 2018), designed primarily for entertainment, could also have educational value through its realistic simulation of real-world ecosystems. We are particularly interested in the role of 'vicarious experiences' of nature produced through interactions with virtual animals, ecologies and dynamic landscapes.

| Vicarious experiences
Playing video games is often perceived as the antithesis of experiencing the natural world. Digital media compete for time with 'nature-based' activities (Larson et al., 2019), and there is widespread concern that urbanisation and risk aversion to unstructured outdoor play for children in post-industrial nations (Edwards & Larson, 2020;Skår & Krogh, 2009) are producing societies that are more 'videophilic' than 'biophilic' (Pergams & Zaradic, 2006). Less time physically spent in green or blue spaces has been linked with consequences for health and well-being, environmental attitudes and participation in pro-environmental behaviours (Bratman et al., 2019;Cervinka et al., 2011;Soga & Gaston, 2016). Further, self-reported 'connectedness to nature' (Mayer & Frantz, 2004) is higher in those who are more likely to take part in outdoor and environmental protection activities (Hartig & Kahn Jr., 2016). Consequently, there has been growing interest in determining how more people, and particularly young people, might be 'reconnected with nature' (Cohen, 1995;Ives et al., 2018;Miller, 2006).
Here, following Clayton et al. (2017), we take a broader view of what constitutes 'experiences of nature'. A person's understanding of the world they inhabit is formed not only through direct interactions with physical living beings and landscapes, but also vicariously (Kellert, 2002), through the representation of wildlife, ecosystems, weather and other natural entities in art, stories and symbolism. These vicarious experiences are not new: artistic depictions, tales and received knowledges of the biophysical world are ancient, and often central to the transmission of ecological knowledge. It is likely, however, that the ratio of direct to vicarious experiences has shifted in post-industrial societies where people can now live with few direct interactions with 'nature'. It is worth noting two things at this juncture. First, this perception of disconnection is just that: a perception. Even individuals who rarely venture outdoors are connected to the rest of the socio-natural world through networks of extended relations, including the food and energy they consume, the products they use, and the politics they endorse. Second, we recognise fundamental challenges with both (a) the continued reference to nature as something external to humans that must therefore be connected with (see Fletcher, 2017) and (b) the reduction of 'experiences of nature' to a selection of benign, often ethnocentric practices such as walking in woodlands and observing wildlife. We broaden this framework by recognising the validity and potential impact of vicarious, virtual experiences, as well as the existence of more violent practices and interactions with wildlife.
With growing urbanisation and use of digital technologies, experiences are not necessarily lost, but transformed (Clayton et al., 2017). As Truong and Clayton (2020) argue, technological or virtual representations of natural entities may increasingly outnumber material experiences with them. While they lack some of the sensory and place-based elements of embodied experiences, virtual experiences are not necessarily less affecting or informative than those taking place in the 'real world' (Pallavicini et al., 2018): indeed, some commentators on video games prefer the term 'actual world' over 'real world' to emphasise that digital experiences are still 'real'.
Consequently, 'rather than dismissing these [virtual] forms as inau-thentic… we have to examine the consequences such a shift could have on people's lives in the long run' (Truong & Clayton, 2020). Calls for reconnecting people, and children especially, with nature often revolve around recreating the experiences of previous generations, assuming these will lead to pro-environmental behaviours (Clayton et al., 2017). However, these calls tend to rely on a set of culturally specific activities and behaviours that often fail to acknowledge class, race and gender politics (Dickinson, 2013). These experiences can also be nostalgically idealised, glossing over the environmental disconnects and challenges that were equally pressing for previous generations (Kahn, 2002). Today's social contexts are incomparable to those of the past, not least due to far greater presence of technologies. Vicarious and virtual experiences are therefore worthy of investigation, including exploration of their dimensions-are they positive or negative, shared or solitary, observational or interactive (Clayton et al., 2017)?-and their effects. We therefore focus on the role of 'vicarious experiences' with simulated wildlife and ecosystems in Red Dead Redemption 2.

| Red Dead Redemption 2
Red Dead Redemption 2 (RDR2) is a AAA video game, meaning that it was produced and distributed by a major publisher (Rockstar Games), had a long development time (~8 years), and a large development team (~1,600 staff) and budget (reportedly >£100 million).
Released in October 2018, as of February 2021 it has sold more than 36 million copies (Knezevic, 2021). RDR2 is an adult actionadventure game set in the USA c1899. The player assumes the role of Arthur Morgan, an outlaw gang member. The storyline follows Arthur as he questions his lifestyle and place in a changing America, while engaging in shootouts, raids and gang conflicts. The game features an honour system, through which players experience different consequences depending on their actions.
RDR2 has a large, immersive 'open world' setting. RDR2 is unusual among contemporary AAA games in that its fictional world aims to realistically depict (a snapshot of) 19th century America.
The world depicted includes richly detailed simulations of a range of natural environments (Figure 1 RDO takes place in the same game world but is separate from the main story. Rather than a narrative storyline, RDO focusses on freedom of choice and inter-player interactions, both cooperative and competitive. Players can choose to specialise in several 'roles', with distinct tasks, challenges and rewards. Of particular relevance is the 'Naturalist' role which rewards players for studying and protecting wildlife. 'Naturalists' are encouraged to interact non-violently with wildlife through photography, tracking, sample collection and protection from poachers. As it is only available in RDO, the Naturalist role is unrelated to the main game storyline. It is technically available as soon as players begin a game in RDO but requires a specific amount of in-game experience to be accessed.
This study aimed to investigate whether immersion in the virtual environment of RDR2 and RDO teaches players about the real-world wildlife and ecology of their setting, and how. We employed mixed methods to address this central aim, applying different methodological approaches to ask questions about different aspects of learning. We used a wildlife identification quiz to quantify, across a broad sample of gamers, whether playing RDR2 is consistently associated with greater species identification skills. In doing so, we aimed to identify whether there is objective evidence of learning having taken place independently of participants' perceptions or memories of playing the game. We compared players of RDR2 against non-players in a natural experiment, to identify whether playing RDR2 was associated with improved quiz performance. Additionally, to consider the more subjective, experiential dimensions of learning, we conducted a qualitative analysis of open survey questions to explore whether and what participants felt they had learned, and to identify the in-game experiences perceived as most memorable.

| ME THODS
We distributed an online survey using the Qualtrics (Provo, Utah) platform. The survey's main component was designed as a quiz in which participants were asked to identify 15 animal species from photographs. Participants who indicated that they had previously played RDR2 were offered optional questions about their experiences of wildlife and natural environments within the game. This study received ethical approval from the University of Exeter (Ref: eCORN002008). The full survey is included as SI1.

| Survey design
An introductory page informed participants about the survey structure, that participation was voluntary and anonymous, and that they could exit at any time. Participants were required to confirm their consent and that they were over 18 years old before proceeding.

| (i) Species and image selection
The quiz was designed to test whether players of RDR2 are better able to identify real-life wildlife species than non-players. Fifteen species featured in RDR2 (Table 1) were selected for the quiz based on several criteria: (a) they are common enough in-game to be encountered by most players; (b) they represent a range of taxonomic groups and (c) they include species commonly and less commonly used as resources in-game. We excluded domestic animals and minimised 'charismatic megafauna' that may be easily identifiable by all participants (e.g. bears, wolves) which could subsequently have made it more difficult to identify a learning effect of gameplay. Clear photographs of each species were presented for participants to identify. We used photographs of real animals (rather than images from RDR2) but avoided photographs in which a species appeared F I G U R E 1 Screenshot of white-tailed deer Odocoileus virginianus from Red Dead Redemption 2. Image credit: Instacodez via Flickr/Rockstar Games markedly different from its in-game representation (e.g. due to intraspecific variation).

| (ii) Quiz design
We gave participants two opportunities to identify each species. First, they were offered the photograph with a 'text entry' box to determine whether they could identify the species without prompting.
Participants unable to identify the species could leave this box blank.
Participants were then shown the same photograph again and presented five multiple-choice options. For the four incorrect options, we selected species that were either taxonomically similar to, or had species names that could reasonably refer to, the correct answer (e.g. incorrect options for the blue jay Cyanocitta cristata included 'blue finch'

Porphyrospiza caerulescens and 'mountain bluebird' Sialia currucoides).
Participants could not return to the text entry box once they had seen the multiple-choice options. This design assumes that participants who correctly named a species in the text entry box demonstrated a higher level of identification skill than those who only picked the correct option out of five. All 15 species were presented in random order, but the multiple-choice question always immediately followed its associated text entry question. Multiple-choice options were presented in a random order for each participant.

| (iii) Gameplay and demographic questions
Following the quiz, we asked participants whether they played video games and whether they had ever played RDR2 (from hereon, those who reported ever having played RDR2 are referred to as 'RDR2 players'). We further asked RDR2 players: (a) whether they had completed the main storyline (indicating an average of 40-60 hr in-game); (b) an estimated number of hours played; (c) an estimated number of species encountered and (d) whether they had played the 'Naturalist' role in RDO. We also asked participants' age, gender and country of residence.

| (iv) Follow-up questions for RDR2 players
Participants who had not played RDR2 ('non-players') were then shown debriefing information and their quiz results, which concluded the survey. To better understand RDR2 players' subjective experiences of learning, and explore factors that could influence what they learned, RDR2 players were offered additional questions about their time playing the game. We used multiple-choice questions to ask in what ways participants regularly interact with wildlife, both in RDR2 and real life.
We also asked two open-ended questions: (

| Survey distribution
We piloted the survey with 16 participants and subsequently made minor adjustments to wording and survey logic. We distributed the final survey using social media platforms Reddit, Facebook and Twitter. We targeted self-identified 'gamers' including both those who had played RDR2 (using dedicated groups and threads) and those who had not (using more general gaming groups and threads).
The survey was advertised as a 'Wildlife Quiz for Gamers'; RDR2 was not mentioned by name. The survey was open from 25 October to 15 November 2020.

| Statistical analysis
For each of the free text and multiple-choice answers to the quiz we tested (i) whether playing RDR2 was associated with more correct answers overall; (ii) what gameplay characteristics of RDR2 players were associated with more correct answers overall; (iii) whether playing RDR2 was associated with answering correctly on a species-byspecies basis and (iv) what gameplay characteristics of RDR2 players were associated with answering correctly on a species-by-species

basis.
For all analyses, we excluded poorly represented groups from the final dataset to improve statistical inference. Therefore, we did not analyse people aged 65 or older (n = 2), people who did not identify as either female or male (n = 38), or (for analyses ii and iv) those who were unsure whether they had completed the game (n = 1).
For free text answers we used a set of acceptable answers (provided in Table S1). We used fuzzy text matching with a maximum allowable distance of 10% (default) to allow for any spelling errors. To test what aspects of playing RDR2 were associated with more correct answers in the quiz we reanalysed the subset of participants who were RDR2 players. We again fitted either binomial (multiple choice answers) or zero-inflated binomial (free text answers) GLMMs with the number of correct and incorrect answers as a response variable and country of residence as a random intercept. For each answer type, we fitted seven competing models ( Table 2)  and 1 month ago, and Past = more than 1 month ago). We included age group and gender as additional fixed effects. We determined the most parsimonious explanation of our results by selecting the model with the lowest AIC from the candidate set.

| (iii)
Are there species differences in the association between playing RDR2 and answering correctly?
To test for species differences in the association between having played RDR2 and answering correctly we analysed the answer to each question separately using a logistic regression approach. We used binomial family GLMMs and included random intercepts of both participant ID and participant country of residence alongside fixed effects of species (question), whether a participant had played RDR2 or not, their interaction (species × played game), age group and gender. We then assessed the importance of whether a participant had played RDR2 or not by comparing the AIC of a model containing the interaction effect to a null model without the interaction.
2.3.4 | (iv) Are there species differences in the association between RDR2 players' gameplay characteristics and answering correctly?
We used a similar logistic regression approach to test the importance of how and when people had played RDR2 for the probability of correct answers. Similar to analysis (ii), we compared seven competing models with different combinations of whether a participant had completed the main storyline of RDR2 or not, whether a participant had played RDR2 online in the Naturalist role or not, and when a participant had last played RDR2 fitted as interactions with species (Table 2).
Age group and gender were also included as fixed effects, and participant ID and participant country of residence as random effects.

Model name Full model for analysis (ii)
Full model for analysis (iv) separately for subsets of the data that (a) included only participants residing in USA and Canada (n = 196) and (b) from all other countries excluding those where no country information was provided (n = 323). These analyses were repeated without country as a random intercept, hence models for quiz scores were re-fitted as generalised linear models (GLMs). We re-fitted only the best-fitting model as identified by AIC in the full analysis where relevant (RDR2 players only data).
All analyses were conducted in R 3.6.1 (R Core Team, 2020).
We used the package lme4 (Bates et al., 2015) to fit most statistical models described and the package glmmTMB (Brooks et al., 2017) to fit zero-inflated binomial models for free text answers for analyses (i) and (ii). Full R code and model results are provided in SI2 and SI3.

| Qualitative analysis
The two open-ended questions about RDR2 players' experiences were qualitatively analysed in NVivo (v1.3.1) by a single coder. All responses to these questions (no participants or responses were excluded from the qualitative analysis) were coded through an inductive process of close reading, labelling responses in relation to thematic categories, and then refining the groupings. For question (i), which asked what players had learned, responses were coded into broad 'subject areas' of knowledge such as species appearance, animal behaviour and interspecies interactions. For question (ii), which asked about players' memorable experiences, responses were coded twice, according to both type of experience (e.g. an interaction with wildlife) and by explanation as to why it was memorable (e.g. an emotional response). Further details of the coding categories are provided in Section 3.4. 17) with 55-64-year-olds (n = 4) and those 65+ (n = 2) poorly represented. This is broadly consistent with demographic data from the USA which indicate that the majority of console action game players are males aged between 18 and 54 (ESA, 2020). The age structure of participants who had played RDR2 was similar to that of participants overall. Participants resided in 55 countries (see

| Quiz performance
Among participants in the dataset analysed (younger than 65, and either identifying as men or women), the number of species correctly identified in the free text answers was low (mean = 1.92, median = 1, interquartile range = 0-3) and 215 participants (39.5%) failed to identify any photos correctly. Only 28 participants (5.1%) answered more than half of these questions correctly. The number of species correctly identified was much higher in the multiple-choice answers (mean = 9.14, median = 9, interquartile range = 7-11).

| Quantitative findings
3.3.1 | (i) Is playing RDR2 associated with more correct answers overall?
RDR2 players identified more species correctly, both in the multiplechoice ( Figure 2) and free-text questions ( Figure S1). Non-players had a median score of 7 (IQR: 6-9) when answering multiple choice questions and a median of 0 (IQR: 0-1.5) answers correct in the free text version. RDR2 players had median scores of 10 (IQR: 8-12) and

| (ii) What gameplay characteristics of RDR2
players are associated with more correct answers overall?
Among RDR2 players there was evidence that all three variables considered (how recently they had last played, whether they had completed the storyline and whether they had played in the Naturalist role) affected the likelihood of answering more questions correctly (Table 3).  For both the multiple-choice and free text answers there were statistically significant differences between species in how having played RDR2 was associated with the probability of answering correctly ( Figure 3). For the free text answers, playing RDR2 was associated with a striking increase in the probability of answering correctly for both pronghorn Antilocapra americana and white-tailed deer Odocoileus virginianus. Playing RDR2 was also associated with a substantial increase in the probability of these two species being identified correctly in multiple-choice questions. There were also notable effects for the two fish species (steelhead trout Oncorhynchus mykiss and lake sturgeon Acipenser fulvescens) and smaller positive effects for a range of other species, especially reptiles and amphibians (excluding the green iguana Iguana iguana). Results were qualitatively similar for participants in North America and elsewhere (Table S7).
Participants in North America were more likely to identify some species overall (e.g. black-tailed jackrabbit, blue jay and white-tailed deer). Correspondingly, participants outside North America were more likely to correctly identify these same species only if they had played RDR2. Full model results are provided in Tables S7 and S8. 3.3.4 | (iv) Are there species differences in the association between RDR2 players' gameplay characteristics and answering correctly?
For both multiple-choice and free text answers there were statistically significant differences between species in how participants' gameplay characteristics were associated with the probability of answering correctly. We found strong support for a recency effect, with when a participant had last played RDR2 in both best-fitting models. The recency effect was generally strongest for species for which having played RDR2 was associated with larger increases in the likelihood of answering correctly. For free text answers, the recency effect was especially clear for pronghorn, white-tailed deer and alligator snapping turtle Macrochelys temminckii. For multiplechoice answers it was strongest for white-tailed deer, steelhead trout, alligator snapping turtle and black-tailed jack rabbit Lepus californicus ( Figure S1). For multiple-choice answers, the recency effect was much less apparent for pronghorn, with all RDR2 players having a very high success rate with this question ( Figure S2). We also found some evidence that the association between the likelihood of a correct answer and both completing the main storyline (free text answers only) and playing in the Naturalist role online (multiple-choice answers only) differed between species. Completing the storyline was associated with a slightly higher probability of correct free text answers in several species (including representatives of all classes; Figure S3). Playing in the Naturalist role was most strongly associated with a higher likelihood of answering correctly for the blacktailed jackrabbit, steelhead trout, pronghorn, nine-banded armadillo Dasypus novemcinctus and roseate spoonbill Platalea ajaja ( Figure S4).
Results were generally qualitatively similar for multiple-choice answers between players in North America and those elsewhere in the world (Table S7), with a small number of exceptions (e.g. black-tailed jackrabbit). Full model results are in Tables S9 and S10.

(c) Natural history, behaviour and interactions
Participants reported learning about species morphology, physiology, behaviour and ecological interactions through their gameplay.
Several noted appreciating the 'real' size of some species, notably moose Alces alces, and learning about physiological features such as muskrat Ondatra zibethicus scent glands and the neck length of alligator snapping turtles. Many more, however, recounted learning about animal behaviours and interactions through in-game observations. Comments referred to generally improved understanding of behavioural ecology (e.g. activity times, scavenging behaviour, predation strategies and group living), as well as learning about specific behaviours such as opossums Didelphis virginiana feigning death, foxes Vulpes fulva pouncing into snow, and horses rolling in mud.

(d) Human-animal interactions
Hunting and fishing are key components of gameplay in RDR2. In addition to learning about wildlife, participants also learned about

(b) Observing wildlife
Participants recalled memorable observations of animal behaviour and interspecies interactions. You can go outside right now and see all of this.
Participants reported feeling both impressed by this attention to detail and that it contributed to the immersive experience: I was riding by a lake and suddenly saw an eagle dive… to catch a snake in its talons. It was memorable because until then I hadn't realized that animals could interact in that kind of player-independent manner. The game continually blew me away with its realistic approach to the world it's depicting.

| D ISCUSS I ON
Our findings indicate that, despite RDR2 not being designed as an educational game, it can nevertheless improve players' ability to identify the North American species it features, for both players in North American and those elsewhere in the world. This effect is most evident among those who have played more recently, spent more time playing and played in the Naturalist role online. Our qualitative findings, though not statistically generalisable in the same way as the quantitative results, provide insights into players' experiences of the game, and suggest that beyond simple visual identification skills, the virtual ecology of RDR2 has also indirectly taught players about animal sounds, behaviours and habitats.
The effect of playing RDR2 on correctly identifying species was not consistent across all 15 animals, which provides some indication of the mechanisms by which players learn from the game.
RDR2 players were markedly better at identifying both the ungulate and fish species than non-players. These species would be regularly encountered by players undertaking hunting and fishing activities in-game. This suggests that players learn to recognise species through the regular use and 'added value' of quarry species. Playing in the Naturalist role was associated with being able to identify a broader range of species correctly, especially smaller and frequently encountered species that are not commonly considered as quarry. These findings are reinforced by evidence that players who had completed the storyline (a proxy for at least 40-60 hr ingame) scored more highly than those who had not, as more time playing will increase number of wildlife encounters. In contrast, RDR2 players were no more successful at identifying species rarely encountered in-game than non-players. Examples are the scarlet macaw Ara macao (featured only briefly, in a special location), and the golden eagle Aquila chrysaetos which is uncommonly encountered in-game and often only observed from distance. There is also a clear recency effect, suggesting the knowledge gained through playing RDR2 may be gradually lost over time, mirroring how effective retention of basic science knowledge requires frequent use or testing (Custers, 2010). In general, findings were consistent regardless of country of residence, although participants in North America had higher baseline likelihoods of answering correctly for some species, which at times corresponded with a smaller difference in the probability of answering correctly between those who had and had not played RDR2. While it may be that some of the improved performance is driven through simply knowing the names of relevant species through gameplay, our qualitative results support the notion that many players are indeed learning to identify species using distinctive characteristics. Additionally, as the survey was entitled a 'Wildlife quiz for gamers' there is a possibility that, to some degree, we attracted gamers with a greater than average interest in wildlife. Nevertheless, our results still indicate that some learning has taken place. and cougars, which do not accurately reflect the contemporary risk to humans these animals pose. As the game is a work of fiction, many players will understand this poetic licence. However, as the developers also pride themselves on their realistic renderings of landscapes and wildlife, and many of our respondents assumed these renderings were accurate, such behavioural anomalies might be misleading. That the player is vulnerable to this sometimesantagonistic wildlife is also interesting, however; the natural world players experience in RDR2 is always lively, risky and to be negotiated with, rather than mute, benign and there to be observed.
As Clayton et al. (2017)  ing some players to feel guilty), and skinning is brief but graphic.
The game also includes features (brought to our attention by participants) that are informative and thought-provoking on environmental issues without being overtly moralistic. These include the player's ability to hunt Carolina parakeets to extinction (reflecting real events), for which they receive no reward, and the pollution of a water body due to toxic mining run-off, resulting in unhealthy flora and fauna. A handful of participants felt inspired: to learn more about wildlife, take up photography or just to appreciate real-world biodiversity around them. Our sample size did not enable us to profile the characteristics of respondents who reported this sort of transformative experience. Nevertheless, scaled up to 36 million players worldwide, many of whom have little access to the diverse ecosystems featured in RDR2, it is possible that this immersive game has had no small impact on thousands of players' ecological understanding and appreciation. This would be a valuable direction for future research. This kind of game is also able to reach 'digitally oriented' (Edwards & Larson, 2020) and urban audiences who might otherwise have little enticement or opportunity to engage with wildlife and non-urban landscapes.
Several participants commented, unprompted, that playing RDR2 provided a means of (vicariously) experiencing natures to which they have little physical access. Immersive virtual experiences of wildlife and natural landscapes could even be proffered as an alternative to their real-world visitation, thereby both increasing the accessibility of, and reducing anthropogenic pressure on, protected areas (for example); this idea was included in E. O. Wilson's ambitious 'Half-Earth' proposal (Wilson, 2016). Such a move would carry the risk, however, that experiences of wildlife and wilder places become 'flattened': standardised and sensationalised to meet public demand and increasingly removed from the actual world (Truong & Clayton, 2020;Verma et al., 2015). Experiences tionists might also adopt and adapt ideas from these games: digital media and a big budget are not required to 'gamify' natural history teaching through immersive role-play (e.g. as explorers) and the regular application of knowledge (e.g. teaching foraging, growing produce and creating bird lists).

| CON CLUS ION
For many people, most experiences of wildlife and natural landscapes will be vicarious, mediated by their depictions in books, images and, increasingly, digital media. While vicarious experiences have different qualities from material interactions, they are nevertheless influential in shaping people's knowledge and understanding of the living world. We found that an immersive video game with simulated wildlife and ecosystems can effectively teach players to identify real-world species, both visually and, potentially, by sound.
We also present initial evidence to suggest that simulated ecosystems can teach players about animal behaviour and ecology. Our qualitative findings indicate that immersive gameplay and active learning are key mechanisms for learning vicariously. We therefore conclude that video games, both those designed for education and those designed for entertainment, are worth taking seriously as providers of both environmental education and vicarious experiences of the natural world.

ACK N OWLED G EM ENTS
The authors would like to thank all participants of the survey for their contribution to this research, and two reviewers for their constructive feedback on the original manuscript.

DATA AVA I L A B I L I T Y S TAT E M E N T
Raw data and R code can be accessed via Zenodo at the following URL: https://doi.org/10.5281/zenodo.4899586 . Qualitative coding structure and details are available on request.