From Traditional to Game-Based Learning of Climate Change: A Media Comparison Experiment

Widespread climate change engagement is needed to confront our current environmental crises, but it remains difficult to attain. Methods such as visualizations and experiential learning activities, including games and gamification, have been proposed to engage citizens beyond what generic and one-way information sharing can, but rigorous studies comparing the effects of game-based learning with traditional methods are rare. Therefore, this study investigates the effects of a serious game vs. control on learning outcomes related to climate change concepts. We conducted an experiment involving N=105 participants randomly assigned to two treatment groups (a desktop screen-based video game and an immersive VR version of the same game) and a control (a text with charts) and investigated the differences between pre- and post-intervention measures of knowledge. The results show that all three conditions had a large effect on learning, but there were no significant improvement differences between groups. Therefore, video games, either on desktop or virtual reality, may be as effective as more traditional instructional materials. Based on detailed observations of the questionnaire data, we also provide game design recommendations. Future studies could focus on specific features of learning and cognitive engagement and expand this experimental design to affect and behavior.


INTRODUCTION
The climate crisis threatens both human societies and biodiversity to the extent that urgent action toward reducing greenhouse gas emissions is needed [65].However, current policies are insu cient to achieve the Paris Agreement warming goal [66].One of the reasons why climate change is di cult to mitigate is its wickedness, or the fact that its scale and complexity make it both di cult to understand and to act upon in a manner that would be acceptable by all actors [35].Furthermore, the situation is complicated by the fact that action is urgent, those tasked with its mitigation also contribute to it, and no single actor has enough political power to address the issue e ectively, leading to irrational delay [46].
In parallel to lacking political and economic action, the widespread level of climate change engagement needed to demand and support the necessary changes remains insu cient and di cult to attain [41,49,89].Although public climate change engagement requires a state of connection with the issue that transcends understanding [49], knowledge remains a crucial component as it can provide a meaningful justi cation for action [49].Thus, the importance of public understanding of climate change, coupled with the limited e ectiveness of top-down information delivery, has led experts to propose consciously framed messages, visual media, and dialogic or experiential methods as e ective communication techniques [29,58,60,90].
These recommendations point toward the potential of both immersive virtual reality (VR) as a technology and gami cation as an engagement technique for climate change learning and engagement.On the one hand, the simulational aspects of 3D environments can enhance learning [48], especially in immersive VR where people can act and sense within a space [88].Immersive VR can also allow users to experience abstract and distant concepts such as climate change as tangible and close [12,53], making them more memorable and emotionally impactful [53].In addition, immersive VR is often perceived as engaging [70] and typically leads to experiences of presence, or the feeling of being in the virtual space [19], which can positively a ect attention, memory and reasoning [70].Thus, the existing evidence suggests that immersive VR is e ective in promoting climate change learning [11].On the other hand, digital games support e ective learning by promoting direct experience and guided inquiry within virtual worlds [58,75], providing enjoyment and emotional engagement through elements such as stories and characters [34,75] and motivating through the exercise of competence, autonomy and relatedness [75,79].
However, our current understanding of the potential of games and immersive technologies for climate change engagement is limited.Game designs and their impacts di er in many ways, as do contexts and audiences that can bene t from their use [29].Comparisons of climate change games and other media are rare [29], and the existing ones paint a nuanced picture in which games present speci c advantages such as retention and engagement [73] or are not signi cantly more e ective than other methods [68,81].In particular, immersive VR games for climate change engagement are uncommon [23], as is the evidence of their e ectiveness [11].
To paint a more complete and accurate picture of the potential of gameful climate change engagement, more studies, especially controlled experimental designs, are needed [29].Whereas no single study may be able to provide de nitive answers, the accumulation of rigorous evidence is critical for future reviews and meta-analyses seeking to advance our knowledge beyond the current level (for existing reviews see, e.g., [21,26,29,76]).
This study randomly assigned 105 participants to three conditions with the aim of exploring and comparing the learning e ects of a text with static charts, a screen-based PC game, and the same game in immersive VR.The text and the game, created for the purposes of this study and titled Climate Connected: Outbreak, focus on the connections between climate change causes, its physical manifestations, and its consequences for human and non-human life, particularly the emergence and expansion of infectious diseases and pandemics.Our exploratory null hypotheses include: • H 0.1 : There is no signi cant di erence between pre-and post-intervention performance.• H 0.2 : There is no signi cant di erence in post-intervention performance between text readers, PC players, and immersive VR players.
• H 0.3 : There is no signi cant di erence in post-intervention performance between text readers and game players.
The results suggest that the intervention led to large positive learning outcomes, but there were no signi cant di erences between the three groups.The statistical tests are complemented with descriptive analyses of the data to explore what patterns can be found across questions, conditions, and moments of data collection, with the goal of better understanding the details of the participants' learning.Our discussion of these results suggests aspects of the game and the text that may have supported or hindered learning, and proposes design recommendations.Taken together, these results provide insight into the potential of game-based learning for climate change engagement, especially when the goal is to acquire information through single-player narrative games on desktop screens and immersive VR, in comparison with traditional instruction.In doing so, it complements previous e orts by the CHI Play [13,15,23,36,62,84] and CHI [24,61] communities to bring positive impact through technology and games in the elds of environmental sustainability and climate change.

BACKGROUND 2.1 Climate Change Engagement
Climate change engagement as a psychological concept can be de ned as "a personal state of connection with the issue of climate change ... concurrently comprising cognitive, a ective and behavioural aspects" [49, p. 446].These three dimensions form a complex and nonlinear system where knowledge is an important component but not necessarily a precursor to action or emotion [41,49,89].Furthermore, education and environmental knowledge have been deemed important, albeit not su cient, for pro-environmental decision-making [31], and even given a secondary role in pro-environmental behavior [41].However, behavior based on conscious understanding can provide a deeper reason for mitigation measures beyond non-environmental motivations such as nancial gain [49].It must not be forgotten that climate change information interacts with personal and contextual elements, from values and attitudes to economic, social and cultural factors [31,49,89].When faced with a message about climate change, people are subject to biases in attention and perception [50].In fact, knowledge itself can be seen as a complex construct comprising di erent degrees-from isolated facts to true comprehension of systems, and from causes to consequences to mitigation and adaptation measures and their level of e ectiveness [37,41].
This complexity, which mirrors that of climate change itself, has sparked decades of research examining the social and psychological barriers preventing action [31,41,49,89].Accordingly, various strands of research have emerged aiming to understand the keys to promote climate change engagement via various media and message strategies, including frames that make the issue salient for speci c audiences (see, e.g., [1]).

Communicating, Visualizing, and Interacting With Climate Change
In climate change communication, making issues and solutions visual has been hailed as an e ective method to promote engagement [80].Consequently, the role of visuals in this area has garnered research attention, as has the potential of technologies that allow for their e ective use [60], in particular information and communications technology (ICT) for interactive visualization [91].This includes virtual environments such as those enabled by immersive VR technology, e.g., headmounted displays (HMD).Some cognitive advantages of interactive visualizations are a more e cient use of cognitive capacity, ease of information search and pattern recognition, simpli ed complexity, and manipulability [17].
3D virtual environments, and immersive VR in particular, present potential advantages for learning through environmental simulation [48].According to the theory of grounded cognition, cognition can be based on modal simulations, or the mental reenactment of past experiences, among other mechanisms (bodily states, situated action) [5].This theory and adjacent concepts, such as embodied cognition, posit that our thinking is based on mental representations that our memory captures in a situated way, that is, via physical sensation and environmental perception [5].According to these theoretical perspectives, sensory-motor impressions remain perceptual in our minds, rather than being transformed into an amodal language (e.g., a list of an object's features) [6].The involvement of multiple neural networks associated with multiple sensory-motor aspects enriches a concept's meaning and makes it readily applicable in more contexts [33].
In connection with the former, perceptual simulations help promote understanding, and tasks involving direct manipulation tend to be more memorable [8].For example, past experiences with historical video games have been found to aid symbolic learning (e.g., from reading) of new information [8].However, not all stimuli related to a concept are equally powerful for learning.For example, good educational visual representations should activate perceptual symbols at the core of the concept represented [14], e.g., showing ongoing changes as a result of stocks, ows, and feedback loops when teaching about systems thinking.
While desktop-based 3D environments can o er multimodal stimuli and interaction, immersive VR typically presents more vivid or realistic sensory information and interactions that involve the body more fully [48] through features such as head and hand tracking.These features allow users to see the 3D environment all around them, move, and manipulate objects using their arms and hands, for example.Thus, more complete and explicit mental representations can be built from more complex body movements and environments o ering a variety of modal perceptions (e.g., tactile, in addition to auditory and visual), supporting memorization [48].As said, rich interactive visualizations can be especially relevant for phenomena that are often perceived as distant and abstract, such as climate change [58,80].
In addition to the above considerations, engagement can also lead to memorable learning experiences.Comparative studies between immersive VR and other forms of instruction in the elds of science, technology, engineering, and mathematics (STEM) show that VR is typically perceived as more engaging and can support learning by inducing emotions and a sense of presence, which promotes not only satisfaction but also cognitive factors such as attention, memory, and reasoning [70].The excitement of immersive VR may be at least partially related to the fact that it o ers both novel and complex experiences, which tend to elicit curiosity [7].
However, some comparative studies have found limitations.Immersive VR was less e ective at teaching science than a slideshow [69] and less [51] or as e ective as desktop computer versions of the same content [59], which may be attributed to a higher cognitive load and potential for distraction in VR [51].Similarly, watching 360º video through an HMD rather than a screen can negatively a ect attention and recall because users feel more compelled to explore, which detracts attention from voiceover narration [4].When information in immersive VR is presented as text, readability can also be poor due to low resolution [38], which can result in tiredness.
In the eld of climate change communication, the fact that immersive VR tends to be more enjoyable and immersive than other media also has speci c implications for climate change engagement, since such states of cognitive absorption may counteract biases based on previous beliefs and motivations [50].Previous research on the potential of VR for environmental communication shows mostly positive ndings when attempting to impact attitudes and behavior [11].HMDs, computers and mobiles can simulate experiences such as immersive VR eld trips and 360º spatial explorations, which can promote learning about climate change [53,54] and other environmental topics, even to a higher degree than conventional teaching materials [67].Immersive media such as 360º videos can increase the salience of distant issues [12], which is especially relevant for climate change.A sense of presence, which more immersive technologies tend to elicit [19], can support learning about climate change and increase environmental concern [54], although a direct e ect of presence on learning was not observed in a study comparing 360º video-based climate change news on an HMD, a screen, and a text with pictures [3].
In summary, visual and interactive media, and immersive VR in particular, present potential advantages for learning about climate change, although the evidence and learning e ect comparisons with other media are limited [11].Given the multiplicity of contexts and audiences relevant to climate change learning and the disparity of outcomes, more comparisons with other instruction and communication methods are needed.

Games and Gamification for Climate Change Engagement
Beyond climate change visualization, experts have proposed a shift from one-way communication towards dialogic processes [60,90] and other methods where people are given a chance to build knowledge on their own terms [58].These recommendations imply that experiential methods such as games and gami cation harbor a potential for climate change engagement.
Game-based learning describes the use of games, whether designed as educational artifacts or not, to engage players in learning activities that incorporate the topic as part of a game system.Players typically engage with the system through a role, making choices and confronting the consequences [72].Meanwhile, the term "gami cation" has often been used to describe the application of game design elements in non-game contexts, as opposed to full-edged games [20].However, gami cation is also used as an umbrella term for any transformation of activities or systems to a ord game-like experiences to support change, including serious games and game-based learning [32].While this study uses terms such as game or game-based learning when precision is needed, it adopts the conceptualization of gami cation as a term referring to any form of gameful engagement, as has been done in climate change engagement literature [21,29,76] and more broadly [40,42].
The use of digital game-based learning includes the advantages of exploring visual climate change through virtual worlds and the possibility of learning through direct experience and inquiry [58].Beyond this, games can also adapt to player performance, not only maintaining an adequate challenge level but also supporting learning [75]; elicit strong emotional engagement with events, stories, and characters [34]; and motivate players through intrinsic and extrinsic means [75] to continue engaging with the game and even to change their behavior.
Desktop 3D game learning environments have been found to be more e ective than technically similar virtual worlds or simulations [56] and traditional instruction, especially when they span multiple sessions or are combined with other methods [92].Various digital games have successfully promoted learning about climate change [29].In the eld of human-computer interaction, several scholars have presented their gami ed pro-environmental contributions in venues such as CHI and CHI Play, including gameful science communication [61], a climate action simulation game [13], and educational games about climate-related biodiversity issues [10,62], energy [2], mobility [27], sustainable production [77] and other topics, sometimes involving methods such as solar-powered wearables [15], playful artifacts [36], and social learning environments [43,52,57].The creation of games as part of climate science education has also been explored [84], as well as the analysis of existing ones in search for climate change engagement potential [23,24].
However, the eld of applied games lacks media comparison studies involving games and control conditions with similar informational content [81].This includes the area of gami cation for climate change engagement [29].When comparisons exist, games' advantage may be either non-signi cant [68,81] or a ect speci c features of learning such as retention, in addition to being more engaging [73].Furthermore, immersive VR is rarely used in game-based climate change engagement [23], despite having received attention as a visual climate communication tool.1), as was their gender distribution (control: Female=20, Male=14, Other=1; PC: F=20, M=14, O=1; VR: F=21, M=13, O=1).

Materials
All materials were designed for the needs of this study and provide an understanding of the connections between climate change causes; physical manifestations; impacts on biodiversity and human societies, in particular infectious diseases and pandemics; and mitigation actions.The materials include a questionnaire, answered by all participants before and after the intervention; Climate Connected: Outbreak, a digital game for desktop PC and VR created by the rst author using the Unity engine (version 2020.3.19f1); and a text document adapting the game's informational material and owcharts.All the materials were in English and are described in more detail next.

Pre-and Post-test.
The pre-and post-test instrument is a 14-question questionnaire (available in [28]) created to address the learning objectives and the content of the game and the text.Three answer options were given: true, false, or "I don't know."All three groups answered the same questions, which were presented in a di erent order pre-and post-intervention.In accordance with the game and text contents, the 14 questions focus on the causes of climate change, its physical impacts such as sea level rise and droughts, its connection with pandemics, the origin of most human diseases, the e ects of climate change on infectious disease vectors and other animals and life forms, the use of cereals in the world, possible forms of climate action, and the implications of using air conditioning for climate adaptation.

3.2.2
Game System and Control.The game's design considers recommendations and gaps identi ed in the games for climate change engagement literature [23,24,29].Thus, it frames climate change as a health and wellbeing issue, which is an underexplored [85] and potentially engaging [1,86] approach; frames the player as a citizen with many capabilities beyond consumer behavior [82,90]; and promotes real-world action.Climate Connected: Outbreak underwent multiple testing iterations on PC and VR before arriving at the one used in this study (for studies based on a previous PC version, see [22,25]).
Content description.The game's content seeks to reinforce three forms of knowledge: declarative, including awareness of climate change facts and understanding of systems; procedural, which refers to actions that can be taken; and e ectiveness, which gauges the mitigation potential of di erent actions [37].Information about climate change and infectious diseases from dozens of scienti c and educational sources (e.g., [39,65,74]) was included.In addition, three experts were consulted about the accuracy of the content during the development process and their suggestions were implemented.They were a doctor in atmospheric sciences with expertise in global change and science communication; a professor of aerosol physics in charge of teaching a course about climate science; and a sustainability and environmental policy researcher and lecturer.
The information is presented as part of an interactive story that, following climate change communication best practices [80,90], puts the player and the ctional local space at the center, visually framing climate change as critical to human and environmental well-being and connecting day-to-day habits and elements to its causes, consequences, and mitigation.Before proposing any actions to the player, the game asks them about their position towards climate change and o ers adapted motivational support [78] mainly based on self-determination theory [18] and environmental amotivation-countering practices [71].Then, players can choose to act in the real world in multiple ways (action sources include, e.g., [82,90,93]) according to their preferences.
PC and VR gameplay design.The game features simple mechanics: select dialog options (i.e., self-paced progress through textual information with occasional choices resulting in small feedback variations), navigate 3D spaces using a teleporting system, and interact with objects (select, grab and throw).On PC, the left mouse button is used for all actions; in VR, a trigger in the controller is pressed.However, VR allows for more complex and physically involving forms of interaction: for example, in the various minigames that require manipulating and throwing objects, VR players can grab them by pointing at them and holding the trigger, as well as throw them with an arm motion while releasing the trigger.The one-button interaction, as well as the natural gestures in the immersive VR version, were chosen to provide comfort to those unfamiliar with digital games and/or immersive VR.The fact that players could use any of the two VR controllers provided aimed to accommodate both right-and left-handed users, and teleportation was chosen as a locomotion method that minimizes the possibility of motion sickness.
The player engages with the game's linear story across four sequential chapters: 1.A Dreadful Future.The introductory chapter presents the game's setting and goal.The game begins in the year 2050, with the player looking at a sustainable city from the balcony of their apartment, located in a distant town.However, the positive picture soon turns into a much more negative one where the air is polluted, and the land is barren soil and concrete.Then, a spirit of nature appears, tells the player that a pandemic rages on in this future world, and proposes to go on a journey around the apartment to nd its origin; that is, the connections between climate change and infectious diseases.As a tutorial, the player is tasked with nding two easily identi able elements, hand sanitizer and a face mask; playing a minigame associated with each, i.e., applying sanitizer to their hands and masking people around them; and seeing the relevant concept added to a owchart.These three steps ( nding an object, playing a minigame, and completing the owchart) constitute also the second chapter's basic gameplay loop, as can be seen in Figure 1.
2. Connections.The spirit of nature declares that it is time to start tracing back the connections from consequences to causes.In this chapter, the player navigates the apartment nding 12 more items representing nodes that connect the causes of climate change with pandemics.Once an item has been found, the player completes a corresponding minigame (e.g., escaping environmental degradation as a bird or nding places where mosquitoes can breed after a ood).Then, the node corresponding to the issue just seen is added to the owchart, which keeps growing as the game progresses.Hence, the textual information is provided by a character, and the main concepts are reinforced with interactions and visual elements and environments.Additional descriptions of the minigames and their relationship to the knowledge tested can be found in the results section, where they contextualize some of our ndings.
3. The Full Picture.Once all the elements are in place, the guide quizzes the player about them, from the causes of climate change to its various impacts on human and ecosystem wellbeing.

What Can We Do?
The spirit of nature presents climate action as a good way to address the causes of many pandemics rather than adopting only a reactive stance.Then, the player is invited to express their feelings towards the game's content, including how much they believe in and care about it.If they choose to share their doubts or issues preventing them from acting, the spirit of nature provides some additional considerations and arguments.Then, if players agree to engage in one form of climate action, they are shown di erent possibilities, including information sources from which to learn more and ways of acting individually and collectively.If they choose one, they are sent an email after the experiment with further ideas.
The game's design incorporates some game-based learning principles as described by Gee [30].These include the player taking an identity, in this case a possible future self; interacting with the game world, i.e., acting and receiving feedback; producing the story through interaction (i.e., progressively building the owchart by nding objects and playing minigames); linking concepts to experiences, i.e., o ering situated meanings; promoting lateral thinking by presenting common elements in a new light, that is, as symbols of climate-related phenomena; in the same way, focusing on systems thinking through the owchart; attempting to be well-balanced in terms of di culty, for which playtesting was conducted; providing relevant information at strategic points; and presenting the quiz as a challenge for knowledge consolidation only after the player has had the opportunity to learn all the connections.However, as a linear story-based single-player design, the game does not focus on o ering other principles such as customization, character knowledge that the player would not have on their own, progressive di culty focused on skill development, learning from failure, multiplayer features, and ample autonomy, since the game o ers one learning path even though players can choose their desired action type at the end [30].
In addition, some principles of multimedia learning have been taken into account [55].The game makes use of both text and visual environments; relevant words and graphics are presented simultaneously and closely; we avoid complicated language as much as possible, in part thanks to the playtesting process; we signal the game's chapter structure to players and the owchart serves as a progress map; and the content is segmented based on discrete concepts.Although the addition of background music is not recommended due to it being potentially distracting from the content [55], the game does not have audio narration, so there is no direct competition within the auditory channel; furthermore, playtesting of earlier game versions suggested that players expect games to have music, even if, as it is the case, it is rather soft and non-intrusive.
Text-based control.To compare the learning e ects of playing the game to reading a text, we created a PDF document containing the textual information in the game, the owcharts, and descriptions of the ctional world matching what can be seen in the game (e.g., "Your town in the year 2050 is part of a clean, healthy, green world.From your balcony, you see a grassy road below where electric trams pass, bicycles stationed under trees, an urban garden, wind turbines moving.").Thus, the 4,600-word text narrates a story in which the reader is addressed directly, but no interaction exists.Any information that could have been acquired by playing the minigames is described instead, and the author's voice in addressing the reader, while present as in the game (e.g., "I wanted to share a simple idea with you"), lacks a de ned character.The quiz in chapter 3 has been replaced by a written account of the right answers, and the questions about the player's feelings in chapter 4 are just described in sequence, accounting for every option that the participant may be interested in (e.g., "Maybe you feel none of this has much to do with you," followed by the corresponding response).The only part that had to remain interactive, choosing an action at the end and receiving an email about it, has been o set to a questionnaire form provided to the participants after reading, and is clearly separated from the document reading activity.

Procedure and Experimental Design
The lab experiment's aim was advertised as immersing participants in di erent forms of climate change information to gather insights on their experience and outcomes so that the communication methods' e ects could be understood.The study was advertised through digital and physical (i.e., lea ets and posters) channels related to the university and the city.Participants were informed in advance that VR may be used so they should not wear glasses if possible.At the end of the session, participants were also encouraged to ask their acquaintances to participate without disclosing their experience.The study was open to any adult who provided their informed consent, complied with national ethical guidelines, and was approved by the university's data protection o cer.
Participants were randomly assigned to one of the three groups before arriving at the site.The researchers only balanced the groups towards the end of the data collection to ensure that the gender distribution was similar.This was done because gender di erences have been observed in environmental attitude, concern and behavior, whether due to personality tendencies, social norms and practices, identity, or other factors [9,31].While gender is but one of the personal and social variables that may a ect an individual's climate change engagement [31], its classi cation as just three values in this study made it simple enough for it to be considered in the experimental design, and the added complexity of balancing multiple additional variables would have exceeded our available resources.As said above, age and educational level, which are also potentially relevant personal factors [31], were similar across groups as well.
The experiment included the questionnaire and two other tasks before and after playing the game or reading the text.The other tasks did not provide information that could interfere with the questionnaire's outcomes.The questionnaire was answered by all participants before and after reading or playing.The questions' order was presented di erently before and after.Participants were given a maximum of 10 minutes to answer the questionnaire, but all were nished earlier.
After this, the respective activity was described to the participants along with the goal-to "get as good an understanding as possible of the content of the game/text" because they would be asked some questions about it afterwards.Participants who read the text or played the PC game stayed in the same computer, while VR participants moved to a di erent space where they would use Oculus Quest 2 with two standard controllers.Participants were informed that they had between 30 and 60 minutes to complete the game.
They were told that if 60 minutes were reached, they would be invited to nish as soon as possible.During the intervention, participants were noti ed when 30, 45, 55, and 60 minutes had passed.Text readers were told that, if they nished before 30 minutes had passed, they would be free to re-read any part of the text until the minimum time had been reached.PC and VR players were told about the game controls.Text readers took an average of 32 minutes to nish; PC players completed the game in 40 minutes on average; VR players nished in 43 minutes.

Statistical Test Results
To test our hypotheses, we rst statistically examined whether the intervention had improved knowledge.Then, we compared the learning outcomes between the three groups.As a descriptive summary, Table 2 includes the participant performance per group before and after the intervention.
To analyze the data, participant performance for each test was calculated as the sum of correct answers, as previous studies of climate change knowledge have done [44,45].Figure 2 depicts the number of correct answers before and after the intervention for all three groups, and Figure 3 shows the di erence, i.e., post minus pre.A one-way ANOVA test (Fisher's) was conducted to ensure that no signi cant di erences existed in the initial number of correct, incorrect, and NA answers between the three groups.Having veri ed that the data for correct answers were normally distributed and the variances homogeneous, the test indicated that no signi cant di erences seem to exist between the groups before the treatment, F(2,102) = 1.127, p =.328.Shapiro-Wilk tests and visual examination of Q-Q plots suggested that the data on incorrect and NA answers is not normally distributed.Therefore, Kruskal-Wallis tests were conducted.No signi cant di erences seem to exist between the groups in terms of incorrect answers before the treatment, H(2)=4.51,p=.105, nor for NA answers, H(2)=1.33,p=.515.
A homogeneity of variances test (Levene's) and Q-Q plot examination suggested that the variances across groups are equal and the standardized residuals of the model are approximately normally distributed.Thus, a repeated measures ANOVA was conducted to determine the e ect of the intervention on learning and possible di erences between groups.The treatment signi cantly increased the number of correct answers from before (M = 9.22, SE = .21)to after the treatment (M = 11.5, SE = .16),F (1, 102) = 156.04,p < .001, 2 = .269, 2 = .266(large e ect [16]), 2 = 0.605.There was not a statistically signi cant di erence in knowledge acquisition between treatment groups (F(2, 102) = 0.654, p = 0.522).
As a complement to the between-subjects part of the repeated measures ANOVA, a nonparametric analysis of covariance (ANCOVA) was conducted using the fANCOVA package (version 0.6-1) [87] in R (version 4.2.2),controlling for the participants' initial level of knowledge.The test used the T.aov function to compare three non-parametric regression curves calculated based on  polynomial regression with automatic smoothing parameter selection via AICC for curve tting.There was not a statistically signi cant di erence in post-test correct answers between the groups, T = 0.025, p = 0.970.
Therefore, based on the within-subjects part of the repeated measures ANOVA and the Wilcoxon rank test, we reject the null hypothesis "H 0.1 : There is no signi cant di erence between pre-and postintervention performance." The intervention had a large positive e ect on the learning outcomes.However, based on the between-subjects part of the repeated measures ANOVA and the nonparametric ANCOVA, we cannot reject "H 0.2 : There is no signi cant di erence in post-intervention performance between text readers, PC players, and immersive VR players." We also classi ed the data in two major groups, control (n=35) and game (n=70, PC and VR together) to examine the overall impact of the game treatment.A repeated measures ANOVA did not nd any between-subjects e ects, F(1, 103) = 1.27, p = 0.262.This test was also complemented with a non-parametric ANCOVA controlling for the participants' initial level of knowledge.There was not a statistically signi cant di erence in post-test correct answers between the groups, T = 1.195, p = 0.587.Thus, the null hypothesis "H 0.3 : There is no signi cant di erence in post-intervention performance between text readers and game players" cannot be rejected.
To test whether other forms of scoring would have led to signi cantly di erent results, we conducted the same statistical tests using two additional scoring systems-subtracting 0.5 points and 1 point for each incorrect answer.This procedure, known as sensitivity analysis [83], yielded no meaningful di erences in statistical signi cance-changes between before and after were signi cant across scoring methods, whereas di erences between groups remained non-signi cant.

Descriptive Examination Results
Next, in addition to the statistical inferences above, we describe the data in more detail to see how participant responses changed after the treatment.The data for all participants and the breakdown by conditions can be found in [28].We classi ed pre-post answer pairings as positive changes (wrong-right, NA-right, wrong-NA, assuming that uncertainty is preferable to inaccuracy), neutral (same before and after) or negative (right-wrong, right-NA, NA-wrong).We observe that, of 1470 pairs of responses (14 questions times 105 participants), 19.3% represented an improvement, 74.3% stayed the same, and 6.4% changed negatively.The data within each condition were similar in all three directions (control: 16.5%, 78%, 5.5%; PC: 21.2%, 72.7%, 6.1%; VR: 20.2%, 72.2%, 7.6%).Although the improvement for PC and VR players was larger than that of readers, they had less accurate answers before the treatment.As can be seen in Table 2, the nal number of correct answers was almost identical in all conditions.The largest source of positive change in the participants' answers was from NA to correct (74.5% of positive change) and the largest source of negative change was from NA to wrong (59.6% of negative change).This was the case in all three conditions.Thus, the change was in most cases positive but not always, as there were in total 203 changes NA-correct versus 56 from NA to incorrect.
The accompanying artifact also contains the aggregate participant answers by question, before and after [28].Four of the fourteen questions were answered correctly by over 90% of the participants already before the intervention: "Q5: Climate change can cause animals to become stressed, sick, displaced, and extinct, " "Q9: Life under water is protected from climate change, which only a ects the atmosphere, " "Q10: Climate change is increased by human activities emitting greenhouse gases such as CO2 and methane, which trap heat in the atmosphere," and "Q12: Some forms of energy generation do not involve directly burning fossil fuels."This occurred across all conditions, the exception being 88.6% of correct answers for Q12 in the VR group.Nonetheless, the data suggests that almost all participants had at least a basic understanding of the cause of anthropogenic climate change, its impacts on biodiversity, and one of the main forms of mitigation, i.e., the use of low-carbon energy sources.
Six other questions were answered correctly by over half of the participants before the intervention.These are "Q1: There is no link between climate change and infectious disease outbreaks" (74.3%), "Q2: Regular people can only act on climate change by consuming di erently in areas such as food and energy use" (78.1%), "Q4: Despite climate change, mosquitoes that carry diseases will stay exclusively in tropical areas" (65.7%), "Q11: Human technology is the only way to take greenhouse gases out of the atmosphere once they have been released" (69.5%), "Q13: Air conditioning helps us survive climate change, but it also releases greenhouse gases that make it worse" (64.7%) and "Q14: Most cereals grown in the world are for direct human consumption" (50.5%).The results across groups were similar, except for Q14 (control: 68.6%; PC: 48.57%; VR: 34.3%).When observing the post-treatment answers, some di erences appear between conditions, particularly in Q2, in which VR participants did comparably worse (control: 91.4%; PC: 94.3%; VR: 77.1%); Q11, where less PC players answered correctly (control: 94.4%; PC: 74.3%; VR: 85.7%); and Q13 and Q14, in which fewer text readers answered correctly when compared to the game groups (control: 88.6%, 80%; PC: 97.1%, 88.6%; VR: 97.1%, 97.1%).
The answers to these questions were given in di erent ways in the game.Demonstrating Q1 was the essence of the game, as the whole process aimed to show the links between the two concepts; Q2 appeared mostly within the nal chapter, in which players could select forms of action that were di erent from individual consumption.Q4 was explained and exempli ed through two minigames in which players had to place a mosquito net and nd and remove mosquito breeding grounds.The concept that makes Q11 false, natural carbon sinks such as oceans and forests, was explained through text after a minigame in which players watered crops that were eaten by a cow, which was used to explain Q14.Finally, Q13 was explained and shown through a simple minigame in which players had to turn on and o an air conditioning unit.
The remaining four questions were correctly answered by less than half of the participants before the intervention: "Q3: Most human diseases begin when a person's DNA gets corrupted and develops a new virus, which is then transmitted to other people." (45.7%), "Q6: The global sea level is expected to rise at the same rate over the 21st century as it did during the 20th century, but climate change mitigation can slow it down before 2050."(34.3%), "Q7: Since mosquitoes prefer warmer temperatures, storms and oods severely destroy their habitats." (47.6%), and "Q8: Droughts, or an extended lack of rain, have become more common everywhere in the world due to climate change." (11.4%).Of these, Q3, Q6, and Q7 saw noticeable improvement (20-25% more correct answers), although VR players did comparatively worse in Q6 after playing (control: 65.7%; PC: 60%; VR: 48.6%).Meanwhile, the number of correct answers to Q8 grew only marginally in general (1.9%).This suggests that the intervention raised awareness of the origins of infectious diseases, the dynamics of sea level rise, and the conditions in which mosquitoes thrive, but it did not clearly communicate the idea that climate change will a ect di erent places on Earth di erently, not necessarily leading to droughts everywhere (as can be seen in [65] and is explained in the questionnaire le [28]).Game players' answers to Q8 improved less than text readers', and in fact fewer VR players answered correctly the second time (control: 22.9%; PC: 11.4%; VR: 5.7%).
In the game, Q3 was explained through text and exempli ed via various minigames in which players embodied a bird escaping environmental degradation and arriving to a city; protected from mosquitoes by placing a net over a door; and destroyed the habitat of a river mammal to build human infrastructure.Q6 was shown through a minigame in which the sea level rose in front of the player for di erent periods of time between 1901 and 2100.Q7 was explained and shown through a minigame in which players had to nd and remove mosquito breeding grounds.Finally, Q8 was explained only textually before playing a forest re-extinguishing minigame that was intended to show that, in contrast with the game's main city where it was raining, other places of the world su er more droughts and wild res than before.

DISCUSSION
This study has compared the learning e ects of three di erent stimuli: a game experienced in immersive VR, the same game on PC, and a text with equivalent informational content.Our observations indicate that all three groups signi cantly improved their knowledge as measured through a 14-question test.However, according to repeated measures ANOVA and non-parametric ANCOVA tests, these knowledge gains were not signi cantly di erent.

Lack of Di erences Between Groups
The non-signi cant di erence between the three groups contrasts with the advantages typically attributed to game-based learning in comparison to traditional instruction methods.However, various aspects of the text, the game, the measurement method, and the participants may be related to the non-signi cant result.
First, the control condition combined textual information with schematics; that is, static images that provided an overview of the system being described.Therefore, it contained a ordances that aimed to represent what one may nd in written media as best as possible.In addition, the information was presented as a narration involving the participant in the second person.Thus, the text, while not interactive and lacking the visual 3D elements of the games, could be perceived as reasonably clear and engaging.
Second, and since it conveyed the same explicit information as the control condition, the game was largely text-based.The 3D environments, object-seeking tasks and minigames were intended to represent the concepts in a clearer and more memorable way through direct experience (i.e., via observation and manipulation), and to provide enjoyment and rest in between bits of textual information.However, to keep the experimental design fair, they provided no additional information relevant to the knowledge test.Therefore, a large part of the cognitive engagement that players may have experienced could also be experienced with the control.Thus, while immersive VR may have promoted presence to a larger extent than the PC game, and PC than text, as is to be expected from their a ordances [19], the di erence may not have been su cient to result in a detectable learning di erence.In fact, increased presence does not always lead to increased learning [3].Furthermore, the existence of extraneous material, or elements that divert attention from the learning content, can have a detrimental e ect [69].Although font sizes were tested and adjusted for legibility during development, the use of an HMD may have also limited some participants' performance since reading can be tiring or di cult due to blurriness [38] and they had to carry the device's weight while standing up for 30-60 minutes.
Third, the learning measurement method was a relatively short list of questions aimed at capturing the participants' knowledge of multiple aspects related to climate change and pandemics, which may have been insu cient to assess deeper or more complete forms of understanding, including complex systemic relationships and transfer from the text/game to the participants' experience, usual behaviors, and future intentions.Following grounded cognition [5], it can be argued that the immersive VR stimulus may have been a more complete and realistic source for mental conceptualization, and the PC game than the text.However, the assessment of fact-based knowledge immediately after the stimulus may not have captured nuances that would manifest later, or in other forms of cognitive engagement.Thus, all three media were similarly e ective according to the kind of assessment conducted, but there could be di erences in other forms of learning.
Finally, the participants took part voluntarily in an experiment advertised as about climate change information.While a small compensation was o ered (a movie ticket), many were likely motivated to learn about the topic, and thus probably interested in reading about it.The sample was generally highly educated, which is another particularity that may be expected from a study conducted in a university and advertised mostly around it but is not necessarily representative of other contexts where climate change engagement is also desirable.The enjoyment and curiosityrelated advantages of using a game and immersive VR [7,34,75] could be more salient for audiences less interested in reading about climate change.
These considerations open the door for exploring e ects other than learning outcomes.While the e ects on quiz performance may have been similar between the conditions, participants in the game-based treatments could show heightened constructive attitude towards climate change issues as well as future intention to act.Moreover, it is possible that the learning experience might have been more enjoyable in the game-based treatment conditions which, ceteris paribus, would make the game-based versions of the instructional material preferable.

Descriptive Data Examination
Our observation of the data as a complement to the statistical analysis revealed aspects of interest in the participants' response patterns.These observations are taken as a starting point for the following discussion, which suggests, in turn, ideas to test in future designs and to validate through qualitative and/or quantitative inquiry.
The analysis revealed that despite the lack of statistical di erences between groups, the intervention greatly increased the number of correct answers while reducing uncertainty, as shown in Table 2.All three conditions similarly succeeded in this.However, the overall number of wrong answers remained similar.Given that sustainability issues are embedded in our daily life, successfully managing them may require a process of unlearning old ways before adopting new schemata and behaviors [64], which is challenging even at the most basic level of trying to correct misinformation [47].This means that the di erence between not knowing, on the one hand, and being wrong but thinking that one knows, on the other, can be important.
In this respect, and while the intervention decreased uncertainty, the fact that part of the change was negative suggests that neither the text nor the game were interpreted unequivocally by participants.This indicates that climate change communicators should emphasize clarity when presenting scienti c facts, perhaps repeating them, providing salient examples, and making sure that visualizations and interactions are in line with the takeaway message [14].This may be especially true for facts that the general public is unlikely to have encountered before or tends to misconceive, as there was a large di erence in the rate of correctness between questions about basic climate change facts and other issues.
Although the sample size within each condition is limited, some questions saw noticeable di erences across groups.Q2, which asked about forms of climate action besides personal consumption, was answered correctly by fewer VR players than participants in the other two conditions (control: 91.4%; PC: 94.3%; VR: 77.1%).This may be related to the fact that the climate action segment of the game came at the end, when some participants may have been tired or rushed and thus less likely to pay as much attention to the game.In addition, this part was chie y communicated via text.Although players explicitly chose from a menu of six actions, only one of which was related to individual consumption, there was little in the way of interactive and visual examples to reinforce the message that all of them were valid forms of action, besides a transformation of the 3D environment from negative to positive after choosing.Thus, designers are encouraged to consider the e ects of gameplay over time, and to explicitly reinforce their messages through visualizations and interactions in which a message's implications are clear and explicit.
Another question where some di erences were observed is Q6, which asked about the rate of sea level rise in the 21st century and the possibility to slow it down before 2050.Here, VR players did comparatively worse than the rest after playing (control: 65.7%; PC: 60%; VR: 48.6%).Although the question is complex due to it containing two statements, the most obvious fact rendering it false is that sea level rise is forecasted to be faster in the 21st century than in the 20th.Since both PC and VR players experienced a minigame in which the sea level rose in front of and around them for various periods of time, VR players may have been distracted to a larger extent by their ooded surroundings, paying less attention to the numerical data.This possible explanation would be in line with observations that VR can distract learners from the content and result in a higher cognitive load [4,51].In addition, players just needed to press continue to see the next phase, so in contrast to other minigames, no explicit attention or e ort was required to continue.Therefore, we suggest that designers integrate data within visual and interactive experiences in a way that paying attention is required to progress, ideally making their apparent comprehension part of the gameplay loop.
Question 8 asked whether climate change has already made droughts more common everywhere in the world.A minority of the participants (13.3%) answered in line with the game's message and with the latest evidence [65] even after playing.This may be related to the common idea that global warming results in a reduction in precipitation, no matter where in the world.Although few participants answered correctly in all conditions, game players improved less than text readers, and in fact fewer VR players answered correctly the second time (control: 22.9%; PC: 11.4%; VR: 5.7%).This may be due to the fact that the explanation that some places have not necessarily seen a rise in droughts was followed by a minigame in which players had to extinguish a forest re.Between the text explanation and the interactive, visual, and perhaps even stressful experience, it is likely that players remembered better the second.Therefore, and especially when considering potential deeply rooted misconceptions or very speci c scienti c evidence, as is the case, we recommend that designers align their most important message with what players do, since the most stimulating and action-oriented experience will likely be the one that will be remembered according to the theories of embodied and grounded cognition [5,6,48].In this case, the inclusion of an action minigame may have succeeded at reminding players, in a more embodied way, of what most of them probably knew already; that is, that climate change impacts can be dire.However, the more nuanced idea, which is that these dire impacts are not the same everywhere, may have been lost to many.
Regarding Q11, which asked about forms of greenhouse gas sequestration that do not depend on human technology (e.g., forests, oceans), less PC players answered it correctly than other participants, and text readers outperformed game players (control: 94.4%; PC: 74.3%; VR: 85.7%).This question was mostly answered in the game through a text-based explanation after a minigame in which a cow ate crops grown by the player.Although there was a connection between deforestation for agriculture and stockbreeding in the minigame, the interaction focused on the fact that most cereals are grown for industrial and animal feed purposes rather than for human consumption, whereas the issue of deforestation was only textually explained and implied by the presence of a farm.This suggests that even when a concept is reinforced with visual and interactive stimuli, the implications should be made obvious through environmental cues and player actions, rather than remaining in the conceptual vicinity.
In contrast, questions 13 and 14, which ask about the environmental impact of air conditioning and the rate of cereal consumption in the world, respectively, were more accurately answered by game players than text readers (control: 88.6%, 80%; PC: 97.1%, 88.6%; VR: 97.1, 97.1%).Since these were associated directly with player actions and game feedback in minigames (in one, players had to turn o an air conditioner; in the other, described above, a cow kept eating the player's crops), we recommend once again that designers try to encapsulate what is essential to the message in their gameplay, no matter how simple the mechanic or feedback, to create a memorable connection with learned concepts through these actions and stimuli, in line with embodied and grounded cognition [5,6,48].

Limitations
Limitations of this study may have a ected its results.First, participants did not only answer these questions, but also engaged in other tasks as part of the pre-and post-treatment data collection process.While all participants completed these in the same order, they could have a ected their performance depending on their tiredness, as well as prompting thoughts relevant to the test.
Second, to conduct statistical tests a choice of how to count correct, incorrect and NA answers had to be made, which introduces a degree of arbitrariness.However, the choice to simplify scoring as the sum of correct answers mimics pre-existing large-scale survey research [44,45]; we have separately provided the number of right, wrong and NA answers in Table 2; shared the original dataset [28]; and tested other grading systems to con rm that they would lead to similar statistical results.
Third, the questions presented before and after the treatment were the same.Although they were reordered and participants were not informed of their performance at any point, the fact that they had seen them before may have had undesired e ects.Future studies may use di erent but comparable questions, or add an additional post-test only questionnaire to avoid familiarity.However, these methods may also have limitations-the comparability of di erent questions should be convincingly justi ed, while post-test only questionnaires allow to infer di erences between groups but not improvement.
Fourth, it should be remembered that the game used has characteristics (e.g., linear, story-based, text-dependent, simple to interact with) which make it di cult to compare to other genres, such as those involving multiple players, strategic, or with open-ended outcomes.Thus, our knowledge of gami ed climate change engagement will continue to advance through the accumulation of complementary evidence.
Fifth, the experiment was designed using only a non-game condition and two 3D environmentbased versions of the same game.In the future additional conditions, such as a text-based game closer to the control condition, could be added as intermediate steps of gami ed learning.
Finally, and based on the above discussion points, our sample largely consisted of highly educated young and middle-aged adults, many of which may have found the knowledge test relatively easy, or the content of the intervention already known to an extent.This may have limited the potential for knowledge acquisition.Thus, we propose that future studies use similar applications either with more di cult content or involving audiences with a lesser degree of formal education, children, and elderly people, in addition to those who may be skeptical about climate change in some way, from its causes to the severity of its consequences and its proposed mitigating measures.Some audiences, such as the elderly, may also nd usability issues with immersive VR.

CONCLUSION
This study has examined and compared the learning potential of text-and game-based communication of climate change information.The study consisted of an experiment involving N=105 participants randomly assigned to three groups, where they were exposed to a text illustrated with charts, a screen-based PC game, and an immersive VR version of the same game.The results of a pre-and post-test questionnaire suggest that all three media resulted in learning, although there were no signi cant di erences in improvement between the three groups.A close examination of the data also revealed how aspects of the game may have led to various forms of learning, and we derived recommendations for future design and research.These results provide insights into the potential of game-based learning for climate change engagement, especially when the goal is to acquire basic knowledge using single-player narrative games, and its comparison with more traditional forms of instruction.Given the potential for more questions to be answered, this study is a rst exploratory comparison of the potential of text and digital games, both on traditional screen and in immersive VR, for climate change engagement.In the future, more data can be leveraged to continue investigating this space, including the examination of possible moderators such as demographic factors and worldviews; mediators such as enjoyment and gameful experience; and outcomes including systems knowledge, attitudes, intentions, and behavior, both quantitative and qualitatively.

DATA AVAILABILITY STATEMENT
The questionnaire used in the experiment and the anonymized answer data are available as an artifact [28].

Fig. 1 .
Fig. 1.Three important moments in the gameplay loop: finding an object, playing the corresponding minigame, and seeing how the new node fits in the flowchart.

Fig. 2 .
Fig. 2. Correct answers before and a er the intervention per group.All three conditions' median values increased a er the intervention (12, up from 9 and 10), with the lower quartiles taking the value of the former higher quartiles.

Fig. 3 .
Fig. 3. Di erence in the number of correct answers before and a er the intervention per group.A median improvement of 2 points can be seen in all three.None of the lower quartiles are negative.

Table 1 .
Participants' distribution per age group Age group n n control n PC n VR An in-person experiment was conducted at the authors' university.The nal sample was N=105 participants divided into three groups: text with static charts (hereafter control, n=35), screen-based PC game (hereafter PC, n=35), and the same game in immersive VR (hereafter VR, n=35).Their average age was similar in all three groups (control: 29.29; PC: 28.57; VR: 30.63; see Table

Table 2 .
Participants' performance per group, before and a er.For each measure, the table includes the correct answers' mean (M) and standard deviation (SD); total correct (Cor.), incorrect (Inc.), and "don't know" (NA) answers; and pre-vs.post-test change