The complexity of climate change mitigation: an experiment with large groups

We have studied the problem of climate change mitigation in large groups by means of a series of experiments with 1785 people. Our participants included both young university students and people of relevance in different organizations, in particular, those attending the presentation of the annual report on innovation by Fundación COTEC (Spain). In the experiment, the participants, distributed in groups of more than 100 people, faced a dilemma: to avoid a global catastrophe that destroys any possibility of making profits, a certain collective sacrifice has to be made by contributing to reach a global threshold. When the threshold was low, the students reached the amount of overall contribution necessary to avoid it. But in the case of a high threshold, none of the populations reached the threshold. In fact, they were far from it. In this sense, the collective behavior of the students and of people of relevance was fundamentally the same. The majority of participants in the high-risk case fell into four categories: those who did not contribute (around 10%), those who contribute half of their means (15%) but less than the fair share required to reach the threshold, those who contributed the fair share (10%), and those who contributed everything they had, so that their personal benefit was zero. In the case of students this last percentage was 10%, but in the other sample it reached almost 30%. We also found that individuals could be classified as being optimistic or pessimistic, and in general they behaved accordingly with regard to their contributions. Our results highlight the complexity of mitigating climate change in large groups and specially the difficulty in communicating the issue to foster action in a general population.


Introduction
According to the Intergovernmental Panel on Climate Change (IPCC) (IPCC 2021), each of the last four decades has been successively warmer than any of the decades preceding it since 1850, and the global surface temperature in the period 2001-2020 was about 1 • C higher than that of 1850-1900.Only in recent years has a clear majority, ranging from 83% in the USA to 95% in Germany in the nine countries surveyed by Eichhorn et al (2020), of Europeans and Americans finally become aware that the climate is changing, and also that human activity is a major cause of climate change (from 79% in the USA to 90% in Italy according to Eichhorn et al 2020) 5 .In Spain, 97% of those surveyed by Lázaro Touza et al (2019) agree that climate change exists, and 92% of them agree that it is caused by human activity.At the country level, the 2015 Paris agreement reached at the United Nations Climate Change Conference (COP21) asked for a set of nationally determined contributions, i.e. self-defined national climate pledges by individual countries to commit to a fixed level of emissions reductions.Interestingly, from this point of view, the latest report of the IPCC (IPCC 2021) states that 'collective action and strengthened network collaboration, more inclusive governance, spatial planning and the provision of risk-sensitive infrastructure will contribute to reducing risks' .In other words, there is a growing awareness that social issues are a key consideration in addressing this problem (Ordner 2017).
In this context, the appropriate way to address the current climate crisis is in the framework of global public goods: goods whose impacts are spread indivisibly across the planet (Faunce 2012, Schmid 2015).Clearly, climate is a global public good and climate change is probably the most difficult collective action problem in human history (Barrett 2018).Collective action problems are also known as social dilemmas and specifically, when the definition is applied to common pool resources, such as climate, they are presented as the 'Tragedy of the Commons' (Hardin 1968).Climate change is a social dilemma in which people must choose between their own short-term self-interest and the long-term interest of the entire population, operating at multiple scales (individual, national, international).
In fact, climate change mitigation is a special case of the public goods paradigm, that has been referred to as the collective-risk dilemma (Milinski et al 2008) or public goods game (PGG) with a threshold (Pacheco et al 2009).In this situation, the people involved are faced with a dilemma: to avoid a global catastrophe, which destroys any possibility of enjoyment, a certain collective sacrifice must be made.This sacrifice can be distributed among all, or it can be made by a majority while others avoid the sacrifice.But if total sacrifice does not reach the threshold, there is a large disaster.In other words, avoiding catastrophe is complex and risky since the effort will be in vain if there is no strong social support.This experimental approach is proving to be very useful in addressing issues related to climate change and the ecological transition (Antonioni et al 2022).Indeed, there is a number of works in which this setup has been studied to address different questions (Milinski et al 2011, Santos and Pacheco 2011, Tavoni et al 2011, Vasconcelos et al 2014, Abou Chakra et al 2018, Brañas-Garza et al 2022).
Our research complements the above works by extending them to a more realistic, complex situation, namely that of large groups where it is impossible to keep track of others' actions, or to enforce any previous, proposed agreement to act.This line of research is important in so far other experiments have shown that large groups behave differently from small ones, because of information effects, or because it is not possible to keep track of what everybody else is do.To the best of our knowledge, the collective-risk dilemma has not been studied in large groups, and we are aware only of experimental results of this type on PGG.Interestingly, the few papers currently available in the literature present contradictory results.Indeed, Weimann et al (2019) run PGG experiments with 30, 40 and 100 participants and found that small and large groups behave similarly both in the first round and as the game is repeated.On the other hand, Diederich et al (2016) studied groups of 10, 40 and 100 members and did a positive group size effect, with contributions declining more slowly in the large groups.Free riding was invariant as a function of the group size though.The largest group sizes were studied by Pereda et al (2019), who run sessions with up to a 1000 simultaneous participants, showing again that contributions seemed to decay slowly, but also that the individual behaviors were very different depending on the information provided to the participants, even if the average contributions were similar.It is important to note that the boundary between small and large sizes may in fact be setting-dependent, as shown, e.g. by Cardoso et al (2020) in experiments involving networks, where they observed that the participants could not really grasp what was going on already in groups of 50 people.These works point to the interest of considering large group sizes in the collective-risk dilemma, not only because the results may differ from smaller groups, but also because to be a more realistic description of the climate change problem groups need to be really large.
It is clear that this problem is essential for humanity, but it has another aspect that touches on innovation and development.There are many companies and institutions that could develop innovations that would make a decisive contribution to the fight against climate change (Nylund et al 2021).However, at the corporate level, they are faced with the same dilemma as at the individual level (Nelson and Allwood 2021): should a company devote a considerable effort, spending shareholders' money, into something that is only profitable if there are many others who do it?That is why the attitude of people in decision-making positions is very important to understand how to foster the fight against climate change.Therefore, we run our experiment, in addition to a sample of volunteers from our laboratory, who are typically young university students (Exadaktylos et al 2013), with people of relevance in different organizations in Spain, as we will describe below.This will also allow us to grasp the differences in behavior coming from the diverse roles played by the individuals of the two collectives.We are aware that the fact that the two types of participants were recruited through very different processes might make it hard to compare the results of both experiments.However, it is also clear that the very different demographics of the involved subjects calls for different recruitment methods, as older people with important corporative responsibilities do not typically participate in experimental subject pools.Therefore, we do believe that our results will be relevant in spite of this possible issue.On the other hand, as shown by Nielsen et al (2021), people of high socio-economic status have a disproportionate impact in climate change, and hence it is of the utmost importance to understand their behavior even beyond the comparison with students or other populations.
The rest of the paper is structured as follows.First, the design of our experiment is described in detail.Then the results of both the initial experiment with a population of students and the one conducted with governmental and company representatives are presented.The research itself closes with a discussion of our main findings.Technical details can be found in the appendices.

General context
For our experimental study, we use a one-shot version of the collective risk social dilemma as proposed in the paper by Milinski et al (2008).At the beginning of the game, each participant receives an endowment, and they must decide whether to contribute, up to a predefined amount, to the common good for a fixed number of rounds (in our case it is only one round).If the combined contributions of all participants over these rounds are equal to or above a certain threshold, the catastrophe is avoided and they receive the remainder of the endowment as a reward (hence the dilemma).Conversely, if the target is not reached, there is a probability that a catastrophe will occur, resulting in an economic loss for all participants (they lose the remainder of their endowment).In experiments, people only tend to contribute to avoid disaster if they perceive the risk to be high (Milinski et al 2008, Hagel et al 2016).Clearly, the context of climate change is very well represented through the characteristics of this social dilemma of collective risk.The risk parameters, the threshold, and the need to avoid losses make the game non-linear and the collective benefit uncertain, as it can only be achieved in the future.
Let us now go into the specifics of our experimental setup.First, participants were informed of a lottery for a check for 400 or 4000 euros depending on the type of participants.We stress that these amounts were real and a lottery winner would actually receive that amount of real money.All the participants received 100 tickets for the lottery, but they were told that they can use part or all of those tickets for a common good: buying trees.To that end, they had to decide how many of their 100 tickets go in the 'money' box or in the 'trees' box, which was to be destined to the common good.That is, if one of the participant's tickets was the winning ticket, i.e. she won the lottery, two things can happen: if she had put that ticket in the 'money' box she would take the money, but if she had put it in the 'trees' box, then the trees would be bought.We made it clear to the participants that they could contribute any quantity, from 0 to 100 lottery tickets.A company specialized in carbon dioxide offsetting by tree plantations was contacted to do the job prior to the experiment, and this was also told to the participants.
The key of this experiment is an additional restriction: in order for the lottery to actually take place, the 'tree' box must have at least 70% of all the tickets distributed (in the case of the students, there was another treatment where 40% was needed).If 70% was not reached, there would be no draw, no money or anything else.
In short, as mentioned in section 1, the idea behind the experiment is that the participants have to solve the dilemma of how much to contribute, something that is very risky because only if people are able to coordinate and reach a minimum will the catastrophe be avoided, and if this is not achieved, the effort will be in vain.
The experiment was approved by the Ethics Committee of Universidad Loyola Andalucía and was carried out according to the corresponding guidelines.

Treatments
Before making a decision on how many tickets to keep (or donate), participants had to watch a 30 s video (without sound and designed to be clearly visible even on cell phones).The video appeared in the same manner on all participants' devices, but there were three types of videos (which are assigned to each participant with probability 1/3; links to the videos are provided in appendix A).
• T1 : scientific authority (science): the scientific evidence on global warming is explained to them.• T2 : moral authority (role models): quotes from public figures on global warming are collected: Pope Francis, Angela Merkel and Antonio Guterres.• T3 : neutral video without any environmental content (baseline).
To check whether the videos might partially explain the variation (if any) in the decisions, before the video and after the decision the participants were asked to rate from 1 to 10 the question 'On a scale of 1 to 10 how bad do you think global warming is right now?' .A full description of our experimental setup including screenshots of the different phases of the experiment, their translation and links to the videos with a transcription of their texts is provided in appendix A.

Experiment with students
The experiment was conducted first online on Tuesday, 8 March 2022 with university students from Loyola, Granada, Pablo de Olavide and Málaga (Spain).The experiment was programmed in LimeSurvey and hosted on the Loyola University server.The experiments were organized in samples of 400 to 500 participants (as stated in the instructions).In each of the three experiments there was a prize of 400 euros.A total of 1246 students participated.Each participant was randomly assigned to one of the following treatments with probability 1/6: • T1: provision 40% -baseline • T2: provision 40% -role models • T3: provision 40% -science • T4: provision 70% -baseline • T5: provision 70% -role models • T6: provision 70% -science.
The first 1000 participants (based on their treatment) were assigned to two different groups: one that had to reach a 40% of the tickets on the 'tree box' , and another one that had to reach 70%.The first group reached the provision point, and correspondingly the lottery was made and had a winner.The second did not reach the provision point and, therefore, there was no lottery.With the rest of the remaining participants (the last ones who had entered the experiment, about 250 subjects) a third group was formed also with a provision point of 70%.As in the larger group, the participants did not reach the provision point and there was no lottery.
Figure 1 shows the most important result, given by the average of the contributions made by the participants in the different treatments.As can be seen from the plot and has been already mentioned, in the low-risk treatment (40%) the contributions were sufficient to achieve the objective of combating climate change, i.e. the provisioning point.In the treatments with high risk (70%), the amount devoted in total to planting trees were very far from reaching the provision point in all cases.Even taking into account the deviation from the mean, the participants did not come close to the target.For both risk values, contributions were larger when moving from the baseline video to the science video, and increased even further when going from the science video to the role models video.From a statistical point of view, this is only significant in the low-risk case, although the high-risk case shows the same trend.The statistical analysis from figures 1 and 2 are taken from tables B1 and B2 in the appendix.In there we estimate two types of models (model 1 without controls and model 2 with controls) within the low risk (40%) and the high risk (70%) condition samples: Model 1: Model 2: Another important remark is that in the treatment of 70%, the contribution is significantly higher6 .This implies that at higher severity there is a higher contribution.While this is not sufficient, it is at least a positive message about the willingness to contribute to climate change mitigation.
We need to briefly discuss statistical power, before we proceed.Taking into account the more restrictive sample size of respondents-La Gala experiment (n = 531)-and the baseline empirical standard deviations (sd = 31.92for the overall contribution), the minimum detectable effect size with our statistical power is 6.1 for the contribution metric, following the methodology of Aberson (2019).Given this, we are adequately powered to detect all the effects for which we find statistical significance.These are the differences in contributions between information treatments (role models and science vs. baseline) in the low-risk (40%) condition, as well as the difference in contributions between the low-risk (40%) and high-risk (70%) conditions.
In the study we also analyzed the effect of the videos on their perception of climate change 'On a scale of 1 to 10 how serious do you think global warming is right now?' .This question is asked twice: at the beginning of the experiment and then after the decision (and having watched the video).As can be seen in the figure 2, the changes between the first and the second answer are not very large (between −0.20 and +0.25 taking the most extreme values) but there are interesting treatment effects.The science video generates the greatest variation with respect to the control, while the role models video of the personalities has a positive effect (compared to the control).
As a general conclusion of this first experiment, it can be said that if the risk of catastrophe is perceived to be very great, people act, but insufficiently, so it seems important to communicate the risks adequately.On the other hand, in order to promote the fight against climate change, it seems that personalities are more influential than scientific data.In fact, the latter have a decided effect on the perception of events (and greater than that of personalities), but this effect is not strong enough to change their decisions sufficiently to avoid the catastrophe.

Experiment at La Gala
The experiment was conducted prior to the start of La Gala on 14 October 2022.La Gala is an event to present the annual report on innovation in Spain by Fundación COTEC.In this occasion it was held at La Nave, in the Madrid district of Villaverde, with nearly 800 guests in attendance and broadcast live online.Just to give an idea of the type of participants we recruited there, the event was presided by HM Felipe VI.Among other personalities in attendance were the Minister of Science and Innovation, Diana Morant; the Secretary of State for Digitalization and Artificial Intelligence, Carme Artigas; the Secretary of State for Telecommunications and Digital Infrastructure, María González Veracruz; and the High Commissioner for Spain as an Entrepreneurial Nation, Francisco Polo.Representing COTEC as vice presidents of the Foundation were the mayor of Málaga, Francisco de la Torre; and the president of CaixaBank, José Ignacio Goirigolzarri; as well as the president of COTEC, Cristina Garmendia.The experiment took place in the waiting time between the closing of the doors and the arrival of HM Felipe VI, who did not take part in the experiment.Cards were placed in the seats of the audience so that attendees could easily access the experiment through the Internet from their cell phones, and the presenter of La Gala invited the attendees to participate.In the end, 539 people participated in the experiment carried out in the run-up to La Gala, 494 of whom completed all the questions up to the end.The comparison between this population and the one studied in our first experiment is shown in table 1 (Online 70 refers to the students who participated in the   70% threshold treatment, the only one carried out at La Gala).A visual summary of the experiment can be found in www.youtube.com/watch?v = JFnA6MX3I4s.
A first summary of the results is shown in table 1, where the average contributions and the degree of awareness about climate change for both experiments are shown.As can be seen, the main result of the La Gala experiment is that once again the provision point or threshold was not reached so that the lottery could take place, despite the fact that contributions were higher than in the case of students (62% versus 55%).
The histogram in figure 3 shows the distribution of contributions in the two experiments, with students and at La Gala.It can be seen that in both there are about 10% of the participants who do not contribute any tickets, another 15% or so who contribute half of their tickets, and another 10% who contribute what would be the fair part of the effort, 70% of their tickets.The difference between both experiments is the percentage of participants who contribute all their tickets (giving up all their chances to obtain the prize): if in the case of the students this percentage is 10%, in La Gala it reaches almost 30%.It is the option most frequently chosen by the participants and the reason why the total contribution is much higher.Once this difference is taken into account, the degree of similarity between both histograms is surprisingly high, pointing to the generality of the observed behaviors.We note specially the peak of people contributing a little bit below the fair amount, around 50%, an issue toward which we will return later in view of its relevance for fostering climate change mitigation.
The results of contributions disaggregated by treatment are presented in figure 4 for a more detailed comparison.In the case of the La Gala experiment, we found no significant differences between treatments using the two-sample Kolmogorov-Smirnov test for equality of distributions or the Kruskal-Wallis test for equality of populations.Keeping in mind the lack of significant differences7 , it is interesting to note the fact that for the La Gala participants the video that to some extent increased their contributions the most was the one about science, while in the Online 70 experiment it was the role models one.Figure 5 shows also the mean variation by treatments of awareness of the global warming problem in the La Gala experiment.As can be seen in the graph, the most notable difference is that for the students the scientific evidence video increases their awareness (so does the personalities video) but in the case of La Gala the effect of the scientific video is, interestingly, negative.An interesting question is whether participants who were members of COTEC, the foundation organizing the event and responsible of the annual report, behaved differently from other attendees.This is a very interesting collective because, typically, members of COTEC have higher corporative responsibilities and consequently higher salaries, and at the same time they are well informed people particularly about the impact that climate change may have on their businesses.Otherwise, they are high officers of the Spanish or the regional governments and are in principle people more inclined to serve the community, which could make them more likely to contribute more.It is possible to have information in this regard because the results were stored by location in the audience.For reasons of respect to the rules of the game (anonymity) the data are not discussed here individually (by seat) but aggregated: all those who were in the VIP area (COTEC members) are aggregated and on the other hand the rest.If we exclude the first two rows (who entered the hall with HM Felipe VI and did not even receive a participation card), there were 80 people in the VIP area.
Of those 80 people, 42 completed the experiment.We estimated the impact of being in the VIP zone on contributions and the estimated effect is 4 (p = 0.474), which means that they are not significantly different.
In other words, COTEC members contribute to the common good just like the rest of the La Gala attendees, although the small number of people in this sample does not allow us to make a strong claim about this.
In relation to this result, it is worth mentioning that although the difference in contributions is not significant, the contributions themselves are different: in fact, those in the VIP area of COTEC contribute 69.6 on average (practically what is necessary to hold the draw, but still below) and the rest of the participants of La Gala contribute 61.4 (the average of La Gala is 62).However, the result quoted above, that 'positive and non-significant 4%' comes out of a model where the contributions are estimated controlling for several basic variables.The result of the model is that the factor that explains the largest contribution is age: the regression shows an estimated coefficient of 0.6 for each year (p < 0.01) in contributions.This explains the difference, as people in the VIP zone tend to be older than general attendees, although we cannot verify this hypothesis due to the impossibility of using individual personal data in this regard.
This result about participant age effects suggests that a deeper look into this factor can be relevant.Table 1 collects more details on the demographics of the participants of both experiments.Note that in parentheses the standard error is indicated and in brackets the min/max.Several observations emerge from the table that may be relevant to the comparison.First, La Gala participants are indeed older, and there are also fewer women than in the Online 70 experiments.Secondly, La Gala participants are more committed to the environment (8.66 vs. 8.08).On the other hand, it is also observed that the percentage of participants in each treatment is perfectly balanced.Turning to the contributions in both experiments, the data from both experiments can be accumulated, and in this case a regression shows us the effect of doing the experiment with students.The result, as can be seen in table 2, is a drop of 7 points for an average of 62.
Let us now explore the results on the participants' expectations, both in terms of what they expected to happen and the degree to which these expectations were accurate.The predictions are presented in figure 6.As can be seen, the Gala participants expected much higher contributions than the students.Turning now to the question of whether they got their predictions right, it is interesting to note that the La Gala participants missed, while the students were more accurate.The contribution was 62.0 at La Gala and the prediction was 66.5, while the students' contribution was 55.0 and their prediction was 55.8.Therefore, in La Gala the expectations were too optimistic.
Combining the data on expectations with the contributions we can analyze how consistent the individual behaviors of all participants were.To do this, we divided the participants into four groups based on their decisions.Thus, we separated the people who predicted that the average contribution would be more than 70% (optimists) from those who predicted less than 70% (pessimists), and at the same time we separated those whose contribution was equal to or greater than 70% (cooperators) from those who contributed less than 70% (non-cooperators).The result is that 39% of participants were optimistic and cooperative, 14% of participants were optimistic and uncooperative, 11% of participants were pessimistic and cooperative, and 34% of participants were pessimistic and uncooperative.It is interesting to note that non-cooperative behaviors can be understood as rational.This is obvious in the case of pessimistic and non-cooperative participants, who understand that there is no point in contributing anything because the threshold will not be reached, but it is also rational to be optimistic and non-cooperative.If a participant believes that the threshold will be comfortably met, he or she can relax and not contribute.
Finally, we have analyzed the effect of the videos in more detail, and we have found that, as we mentioned above, the type of video shown does not seem to influence the contributions.However, other variables do have an effect, such as the starting position on global warming or age.The time spent reading the instructions is also relevant in La Gala at 5%, while the time spent online is relevant at 10%.More information on these analysis is contained in appendix B.

Discussion
In this work we have dealt with the problem of individual action in an experimental context that is very similar to the climate emergency we are experiencing: it is necessary to act individually knowing that the success of the effort depends on the cooperation of others, and that it is necessary to collaborate a minimum or else the objective will not be achieved even among all of us.Complexity enters the picture by considering large groups, formed by more than a hundred people each (Pereda et al 2019, Cardoso et al 2020).The results we have obtained do not allow us to be too optimistic.When the effort required is not too high, the objective is achieved, but without too much slack; in general, the participants make just enough effort.When the effort is high, the threshold is not reached, although the effort certainly increases significantly.In our experiment it was not reached in any of the groups we worked with, not by the students population not by the high officers and executives.Of course, when scaling this results to a population of the size of the Earth's, we can only expect the problem to be even more difficult, more so because of the different interests at the different levels of agency.
We would like to offer two directions as possible explanations for this result.The first one relates to the fact that the population seems to be split into two more or less equal halves: the optimists, who believe that the goal will be achieved, and the pessimists, who believe the opposite.An alternative view arises from another division into two halves, this time in terms of contribution to the collective effort: half contribute more than their fair share, and the other half contribute less.Obviously, most pessimists contribute less, but not all; most optimists contribute more, but not all, so expectations do not completely control whether the goal is reached or not.What this division does show is that it seems necessary to get some of the pessimists (possibly those who contribute more) to change their expectations to more optimistic ones, and to get the optimists who, due to irresponsibility or selfishness, do not contribute to do so.In other words, it is necessary to send two types of messages: it is possible to achieve the objective and on the other hand we have to avoid falling into complacency.In terms of policies, our results make it clear that trying to get everybody on board with the mitigation effort is an enormously difficult task, and it will be more efficient and effective to target specific collectives with tailored messaging and campaigns, based on their idiosyncrasy.Guidance about campaigns and messages that could be effective is provided by Tverskoi et al (2023), who showed that a direct recommendation can backfire if part of the population reacts against said recommendation, and observed that it was the people who are less reliant on personal norms and less prosocial who caused this effect.As discussed by Nielsen et al (2021), it is probably better to resort to specifically designed taxation schemes than to simply ask people to behave properly.While this is certainly not an optimistic result, there is some room for hope in the fact that when people are informed of a higher risk, their contributions increase to a large extent as well (see also Szekely et al 2021).In a context in which it is being increasingly realized that there will be no global tipping point (Rietkerk et al 2021), such larger contributions to the mitigation effort may prove very relevant to prevent specific local tipping points when adequately focused.
A word is in order about the possible limitations of this work.A first one arises from the fact that participants at La Gala are, to some extent, self-selected: more people are invited than actually attend the event, and it might be that those who attend are more self-oriented.In fact, La Gala offers a rare opportunity for networking and it is in the people's own interests to attend.A second limitation is the scope of the comparison: we are comparing a very special class of people with students, and it would of course be necessary to study other groups of the society.Another caveat about our study relates to the fact that we have just studied two risk values due to budget and subject availability limitations, and it would be important to expand this research to other values to better assess the influence of risk on the decisions.
One implication of our results goes in the direction of innovation and development.As stated in section 1, the problem of individual action moves to a higher level, that of companies and institutions that could develop innovations or implement agreements essential to overcome the climate emergency.In this case, the dilemma remains the same: for example, a company has reasons to be reluctant to tackle something of dubious profitability.In this sense, our results are also a call to action, given that the population of people with high responsibilities present at the COTEC Gala behaved in a very similar way to that of students, with two fundamental differences: they are somewhat more optimistic (but not enough) and there were 30% of people who contributed everything they could (compared to 10% of students).Therefore, thinking about innovation from the companies, the message is the same as for individuals, but who should transmit it, perhaps with incentives, are the authorities in charge of promoting innovation.
(ix) Control question (x) Code creation screen (in case of winning).
Personal data: nothing is collected beyond the code.The code generated by the app is not traceable and therefore the data is anonymous.The winning code will be announced publicly and the owner of the code will be told where he/she can go to collect his/her prize in case he/she actually wins the money.
Links to the videos related to the different treatments.
• T1 : scientific authority (science) Carbon dioxide, the notorious gas that keeps the heat in our atmosphere, has increased by 30% since the Industrial Revolution.Because of that, global temperature has risen by more than 1 • .We are liberating an amount of heat equivalent to four atomic bombs every second.That is why the poles will melt before midcentury.In summer there will be no ice in the Arctic and with this melting, every year the sea will rise by another millimeter.It has risen already 17 cm in 100 years.This is more or less [the size of a spoon shown in the video].• T2 : moral authority (role models) 'Stop treating nature like a WC, humankind is heading towards the abyss.' (Antonio Guterres) 'Everyone of us can play a role in changing our collective response to the threat of climate change.'(Pope Francis) 'In the interest of future generations we must act resolutively to stop climate change.'(Angela Merkel) • T3 : neutral video without any environmental content (baseline) Thank you for participating.After this video you will have to make a decision.We are thankful for your participation.
Translations of the screenshots.Figure A1.Thank you for taking part in this experiment.Among all the participants in this show we are going to raffle a prize of 4000 euros.To this end we are going to give 100 tickets for this lottery to each participant.After taking part in the experiment (seven screens after this one, not more than 5 min) all the tickets from all the participants will enter the lottery.In those screens we will ask you six questions and in one you will have to make a decision.If you do not want to participate click on 'Exit, I do not take part' , otherwise click on 'Next' .Figure A2.Are you a man, a woman, or prefer no to answer?Possible answers: man/woman/I'd rather not answer/other (free text).In which year were you born?How many years have you been working?What is your position in the company?In which sector do you work?In a scale of 1 to 10, how grave is global warming right now? (1 is very little grave and 10 is extremely grave).
Figure A3.The decision.We believe that you may not want a money prize, but rather that this money is used toward a common good, such as buying trees.Trees will be bought through www.nature.org.As you have 100 tickets, we will let you choose how to use the money: in tickets to get money or to buy trees or part of them for one thing and part for the other.When the draw is made, if the winner ticket is one of those allocated to money, the winner will receive privately a check; if the winner ticket has been allocated to buying trees, then trees will be bought and there will be no check for anybody.However, there is a constraint: a minimum of 70% of all the distributed tickets must be allocated to buying trees.If this threshold is not reached all the tickets will be destroyed.There will be no lottery, no trees and no check.
Figure A5.How many of the 100 tickets do you want to allocate to buying trees?Figure A6.In a scale of 1 to 10, how grave is global warming right now? (1 is very little grave and 10 is extremely grave).
Figure A7.Thank you for your participation.You placed 50 tickets for buying trees and 50 tickets towards receiving money.We wish you good luck.
Figure A8.In your opinion, what is the percentage of tickets allocated to buying trees?How sure you are of this prediction?(0% totally unsure, 100% totally sure).
Figure A9.In the instructions it is written that there is a constraint, a minimum percentage of tickets allocated to buying trees.What was this percentage?If you do not remember write −9.
Figure A10.During the day we will announce the winner.We will now generate a code in case you are the winner, please complete the following data: first two letters of your name, last two digits of your year of birth, last two digits of your phone number, and last two digits of your ID card.

B.2. The behavior of COTEC participants
In order to compare the behavior of COTEC participants with that of the rest of the La Gala participants, a model was analyzed in which the contributions are estimated according to several variables, obtaining the results shown in the following table: ( As can be seen, age is behind contributions, noting that it has an estimated coefficient of 0.6 for each year (p < 0.01) on contributions, which is actually a very high value.It is true that being in the VIP zone, as a proxy for belonging to COTEC, is positive (not significant), while the age × VIP interaction is negative, the result quoted in the main text being in fact 3% (not significant).In other words, COTEC participants contribute more but do not give more because they are from COTEC but because they are older.

Figure 1 .
Figure 1.Average of the contributions made by the participants in the three groups organized from students.Colors denote the different videos shown during the experiment.Error bars represent the 95% confidence interval.Asterisks indicate the statistical significance of differences between groups in OLS models: * * * for p < 0.01, * * for p < 0.05, and * for p < 0.1.For detailed data analysis, see 4.1 in appendix B.

Figure 2 .
Figure 2. Differences between the perceived severity of the climate change issue before the experiment and after the experiment, in the three groups organized from students.Colors denote the different videos they were shown during the experiment.Error bars represent the 95% confidence interval.Asterisks indicate the statistical significance of differences between groups in OLS models: * * * for p < 0.01, * * for p < 0.05, and * for p < 0.1.For detailed data analysis, see 4.2 in appendix B.

Figure 3 .
Figure 3. Histogram of the number of participants in the experiment that contributed a given amount of money.Left: La Gala.Right: Online 70.

Figure 4 .
Figure 4. Average of the contributions made by the participants in the three groups organized.Colors denote the different videos they were shown during the experiment.Error bars represent the 95% confidence interval.Asterisks indicate the statistical significance of differences between groups in OLS models: * * * for p < 0.01, * * for p < 0.05, and * for p < 0.1.For detailed data analysis, see 4.1 in appendix B.

Figure 5 .
Figure 5. Differences between the perceived severity of the climate change issue before the experiment and after the experiment, in the three groups.Colors denote the different video they were shown during the experiment.Error bars represent the 95% confidence interval.Asterisks indicate the statistical significance of differences between groups in OLS models: * * * for p < 0.01, * * for p < 0.05, and * for p < 0.1.For detailed data analysis, see 4.2 in appendix B.

Figure 6 .
Figure 6.Histogram of the expectations of participants in the experiment that contributed a given amount of money.Left: La Gala.Right: Online 70.

Figure A1 .
Figure A1.Welcome page explaining the experiment content and duration.

Figure A2 .
Figure A2.Six questions (gender, age, years of experience, sector, position and global warming).

Figure A3 .
Figure A3.Explanatory screen of the decision.

Figure
Figure Global warming question again.

Figure A7 .
Figure A7.Information on your decision.

Figure A10 .
Figure A10.Code creation screen (in case of winning).

B. 1 .
Robustness check, only people who spend at least 10 s watching videoThe time they spend watching media videos is similar.

Figure A11 .
Figure A11.Explanatory screen of the decision.

Figure A12 .
Figure A12.Average of the contributions made by the participants, only people who spend at least 10 s watching video, in the three groups organized from students.Colors denote the different video they were shown during the experiment.Error bars represent the 95% confidence interval.Asterisks indicate the statistical significance of differences between groups in OLS models: * * * for p < 0.01, * * for p < 0.05, and * for p < 0.1.

Figure A13 .
Figure A13.Differences between the perceived severity of the climate change issue before the experiment and after the experiment, in the three groups organized from students, only people who spend at least 10 s watching video.Colors denote the different video they were shown during the experiment.Error bars represent the 95% confidence interval.Asterisks indicate the statistical significance of differences between groups in OLS models: * * * for p < 0.01, * * for p < 0.05, and * for p < 0.1.

a
OLS regression coefficients with p-values in parentheses and 95% confidence intervals (CI) in brackets.

Table 1 .
Detailed demographics of the participants and summary of the key numerical results of the experiments in the La Gala, and in the student sessions with threshold 40% or 70%.

Table 2 .
Linear regression model for the aggregated data of the Online 70 and La Gala experiments.P-values in parentheses and 95% confidence intervals (CI) in brackets.

Table B2 .
] Differences between the perceived severity of the climate change issue.
a OLS regression coefficients with p-values in parentheses, Cohen's d effect size in braces, and 95% confidence intervals (CI) in square brackets.b'RoleM' versus 'Science' , provides point estimates for this specific contrast.aOLS regression coefficients with p-values in parentheses, Cohen's d effect size in braces, and 95% confidence intervals (CI) in square brackets.b 'RoleM' versus 'Science' , provides point estimates for this specific contrast.

Table B4 .
] Table4cont.Linear regression model for the aggregated data of the Online 70 and La Gala experiments.
a OLS regression coefficients with p-values in parentheses, Cohen's d effect size in braces, and 95% confidence intervals (CI) in square brackets.b'RoleM' versus 'Science' , provides point estimates for this specific contrast.aOLS regression coefficients with p-values in parentheses and 95% confidence intervals (CI) in brackets.