Mental models and framing

Psychology of Climate Change Communication · Week 4

Dr. Viktoria Cologna and Dr. Jan Pfänder

Trigger warning to psychologists

A narrow definition of mental models:

A simplified picture, in someone’s head, of how something works (e.g., climate change as a bathtub (Sterman & Sweeney, 2002)).

A broader definition I would like to take today:

General features of human psychology that describe how people think, decide, and act across a range of contexts.

The bathtub is the example for the narrow sense. Sterman & Sweeney 2002, System Dynamics Review, DOI 10.1002/sdr.242: graduate students were given descriptions of the climate system from the IPCC. From the abstract: “Overall performance was poor. Subjects often select trajectories that violate conservation of matter”, and many “believe that stabilizing emissions near current rates would stabilize the climate, when in fact emissions would continue to exceed removal”.

The picture: the atmosphere is a bathtub. Emissions are the tap, removal is the drain, and the CO2 concentration is the water level. As long as more flows in than drains out, the level keeps rising, even if the tap is not opened any further. Sterman’s wording for the mistaken belief: “analogous to arguing a bathtub filled faster than it drains will never overflow” (https://www.mit.edu/~jsterman/Understanding_public.html).

Recap

Week 2 · Public opinion on climate change and the deficit model
Can facts change people’s minds? Maybe, but they are often not quite enough.

Week 3 · A question of trust
Without trust in the messenger, it is hard to change minds.

TODAY

Week 4 · Mental models and framing
How can we communicate more effectively to change minds? By appealing to features of human psychology.

(some of these features were teased in Week 1 already)

Overview

1. A quick experiment
2. Mental models
3. Framing effects in climate change communication
4. Revisiting the deficit model

1. A quick experiment

The next slide has one text for each half of the room. The left half reads the left column, the right half reads the right column.

Read your text, make your choice, and write it down.

Imagine that Switzerland is preparing for the outbreak of an unusual disease, which is expected to kill 600 people. Two alternative programs to combat the disease have been proposed. Assume that the exact scientific estimate of the consequences of the programs are as follows:

Left half

If Program A is adopted, 200 people will be saved.

If Program B is adopted, there is 1/3 probability that 600 people will be saved, and 2/3 probability that no people will be saved.

Right half

If Program C is adopted 400 people will die.

If Program D is adopted there is 1/3 probability that nobody will die, and 2/3 probability that 600 people will die.

Which of the two programs would you favor?

Wording from Tversky & Kahneman 1981, Problems 1 and 2. Two changes, the same ones Many Labs 1 made: “the U.S.” is replaced by the country of data collection, and “Asian” is dropped from “an unusual Asian disease”.

Left half: who chose Program A?

Right half: who chose Program C?

A and C are the same programme: 200 of the 600 people live for sure. B and D are the same gamble.

The original study

When the programs were described in terms of lives saved (left half), 72% of participants chose the sure option (Program A).

When the same programs were described in terms of deaths (right half), only 22% chose the sure option (Program C).

In other words, people were more risk seeking when lives might be lost, than when they could be gained.

(Tversky & Kahneman, 1981)

Tversky & Kahneman 1981, Science, DOI 10.1126/science.7455683. Problem 1: N = 152, Program A 72%, Program B 28%. Problem 2: N = 155, Program C 22%, Program D 78%. Students at Stanford University and the University of British Columbia.

Programs A and C are the same program. So are B and D.

The facts are identical. What differs is the wording, and the wording seems to change what people choose.

This is called a framing effect. It happens because we use mental models to evaluate information.

From homo economicus to homo psychologicus

Much of economics and political science starts from rational choice theory.

The main idea: individuals make rational decisions by analysing the costs and benefits of every available option.

However, some human judgements and behaviors seem hard to reconcile with this view.

In the previous example, the costs and benefits of the two programs were identical. A purely rational actor would not have made a difference.

2. Mental models

A (kind of) definition of mental models

Features of human psychology that describe how people think, decide, and act…

… that often seem to deviate from rational choice.

Many terms for roughly the same idea: heuristics, biases, mental shortcuts…

The mind as an information-processing device

Input → Computation → Output

In cognitive science, people study the mind as an information-processing device.

Information comes in, the mind does something with it, and a perception, a judgement or a decision comes out.

To convince you that our mind is not just a perfect mirror of reality, let’s have a look at some well-known optical illusions.

Which square is darker, A or B?

Checker shadow illusion, Edward H. Adelson

Image: Wikimedia Commons, “Grey square optical illusion”, original by Edward H. Adelson, licence “copyrighted free use”.

They are the same

Edward H. Adelson

What happened here?

Squares A and B have exactly the same shade of grey. Yet we see B as lighter.

The visual system does not simply report what is on the screen. It seems to take the shadow into account, and to correct for it.

This correction is fast and automatic. Knowing about it does not make the illusion go away.

Which monster is bigger?

Roger Shepard, Terror Subterra

Which monster is bigger?

Roger Shepard, Terror Subterra

Which monster is bigger?

Roger Shepard, Terror Subterra

They are the same

Roger Shepard, Terror Subterra

From visual perception to information processing

Visual illusions occur because our brains rely on certain standard operating procedures (e.g., correcting for shadows).

These procedures use simplified assumptions about how the world works.

The idea behind today’s lecture is that this is also true for information processing.

Here, we will call these standard operating procedures mental models.

Why do we use mental models?

The world is too complex to reason through every judgement and decision from scratch.

Most of the time they serve us well. Sometimes they lead to choices that look odd from the outside.

Today

We will look at five mental models that matter for communicating about climate change.

1. Loss aversion (losses weigh more than gains)
2. Temporal discounting (now beats later)
3. Availability bias (experience beats statistics)
4. Social norms (we conform to what others do)
5. Mental accounting (we put money into boxes)

Those who did today’s reading will recognize the first four.

1. Loss aversion

Back to our experiment. Why would “saved” and “die” lead to different choices?

“choices involving gains are often risk averse and choices involving losses are often risk taking”

(Tversky & Kahneman, 1981)

In our experiment: faced with a certain loss (400 deaths), most people (78%) preferred the gamble that offered a chance of losing nothing.

In other words, people try hard to avoid losses (more than they try to increase gains).

Tversky and Kahneman’s answer: people judge outcomes as gains or losses relative to a reference point, and they treat the two differently.

The quote is from Tversky & Kahneman 1981. The reading (insight 4): “people psychologically evaluate gains and losses in fundamentally different ways”. Prospect theory patterns replicated in 19 countries (Ruggeri et al. 2020, n = 4,098: “The results replicated for 94% of items, with some attenuation.”).

Technically, loss aversion and risk taking are two different things. Loss aversion is “the idea that losses loom larger than equivalent magnitude gains”, for example that “the loss of $5 is more painful than the gain of $5 is pleasurable” (Gal & Rucker 2018, Journal of Consumer Psychology, DOI 10.1002/jcpy.1047, p. 497). To show it, a loss has to be compared with a gain of the same size. Risk taking is about the choice between a sure outcome and a gamble. Our experiment shows the second: people take the sure option when the outcomes are described as gains, and they gamble when the same outcomes are described as losses. Gal and Rucker say this about the disease problem directly: it “manifests the reflection effect whereby people tend to be risk seeking in the domain of losses and risk averse in the domain of gains”, which is “distinct from loss aversion” (p. 509).

Whether losses in general weigh more than gains is disputed: Gal & Rucker 2018 conclude that “current evidence does not support that losses, on balance, tend to be any more impactful than gains”. How they reach this conclusion:

• The classic demonstrations (status quo bias, endowment effect) can be explained by inertia, which “requires no differential sensitivity to losses” (p. 503).
• When taking a bet is not the more active option, the asymmetry disappears. In one low-stakes study “48% of participants chose the safe option” and 52% the risky one (p. 506).
• “with low stakes, gains actually appear to loom larger than losses” (p. 507).
• A meta-analysis “of 93 studies involving over 20,000 participants” on message framing “did not find a single context where loss-framed appeals had statistically greater persuasive power than gain-framed appeals” (p. 509).
• They grant that losses may weigh more when stakes are high, and for protected goods “such as a town’s mountain view or lakefront access” (p. 512).

2. Temporal discounting

Would you rather receive CHF 100 today, or CHF 110 in a year?

Hands. The amounts are an illustration. The structure of the choice (10% more after one year) is the first question in Ruggeri et al. 2022, two slides further.

The Marshmallow Test

▶ If the video does not load: watch it on YouTube

The Marshmallow Test (Walter Mischel, Stanford, 1960s) was on the Week 1 slides. It illustrates the choice between a smaller reward now and a larger reward later.

The famous follow-up finding, that waiting predicts later achievement, turned out weaker in a replication: the correlation “was only half the size of those reported in the original studies and was reduced by two thirds in the presence of controls for family background, early cognitive ability, and the home environment” (Watts, Duncan & Quan 2018, Psychological Science, DOI 10.1177/0956797618761661).

Temporal discounting in 61 countries

In 61 countries, 13,629 people were asked this kind of question: an amount of money now, or 10% more in 12 months (Ruggeri et al., 2022).

For participants in the United States, this meant US$500 immediately or US$550 in one year.

The result: “Temporal discounting was present in all countries, with only modest variability in national means”.

Ruggeri et al. 2022, Nature Human Behaviour, DOI 10.1038/s41562-022-01392-w, CC BY 4.0. First choice: “approximately 10% of the national monthly household income average” immediately, “or 110% of that value in 12 months”. Those who chose the immediate option were then offered 120%, and then 150%. Caption of Fig. 3: “all locations indicate some preference for immediate over delayed”. Also: “countries with lower incomes typically had greater temporal discounting levels in the baseline items”. The paper’s main interest is in intertemporal choice anomalies: “54% of participants showed at least one anomaly”.

Temporal discounting in 61 countries

(Fig. 4c from Ruggeri et al., 2022, CC BY 4.0, cropped)

One line per country, ordered from the lowest score (Japan) to the highest (Argentina). Dots are country means with prediction intervals, grey bars are standard deviations. Colours: purple for high-income, red for upper-middle-income and yellow for lower-middle-income countries.

What the temporal discount score is: a count of how often a participant chose the earlier option. The authors “constructed a temporal discounting score that allows us to rank individuals by the number of immediate/earlier choices they made across all questions” (supplementary information).

Why 0 to 19: the score is a sum of points.

• Three questions worked like a ladder and gave 0 to 5 points each: gains (about 10% of a monthly household income), the same amounts as payments, and larger gains (a full monthly income). Example for the United States: $500 now or $550 in a year? Whoever takes the $500 is then offered $600 in a year, and then $750. Whoever waits is offered $510 and then $505. The points say where on this ladder a person switches. “An individual who always chose the later option was assigned a score of 0. An individual who always chose the sooner option was assigned a score of 5.” That makes up to 15 points.
• Four additional single choices with longer delays (12 and 24 months) add one point each when the earlier option is chosen. That makes up to 4 points.
• “Adding this score across questions results in a total score of 0 for individuals always choosing the delayed amount and 19 for individuals always choosing the sooner amount.” The paper does not spell out the split into 15 plus 4. It follows from the three questions scored 0 to 5 and the four additional questions, and it fits a detail on Estonia: two of the additional questions were faulty there, and the scores were first calculated on a 0 to 17 scale.

How to read it: a higher score means more choices for the earlier option, so stronger temporal discounting. A score of 0 would mean always waiting for the larger amount. The score is not a discount rate in percent. The authors note that “the raw scores (0–19) have no standard to compare against”. It is useful for comparing countries with each other.

What to see: no country is near 0. National means range “from Japan (mean = 7.1, s.d. = 3.9) to Argentina (mean = 14.1, s.d. = 3.0)”, with an aggregate mean of 10.3. Switzerland (CHE) is the fifth line from the top.

3. Availability bias

What causes more deaths: tornadoes or asthma?

In the original study, students saw tornadoes as more frequent killers than asthma, “even though the latter is 20 times more likely” (Lichtenstein et al., 1978).

From today’s reading (Van Der Linden et al., 2015):

• “The human brain privileges experience over analysis”

• In this case, hardly lived experience, but rather “disproportionate exposure, memorability, or imaginability” (Lichtenstein et al., 1978), reinforced, e.g., by news coverage

Hands before the answer appears. Lichtenstein et al. 1978: “Tornadoes were seen by the student subjects as more frequent killers than asthma, even though the latter is 20 times more likely”. Their Table 1 gives a death rate of 44 per 100 million for tornadoes and 920 per 100 million for asthma. Another pair from the same paper: “Death by lightning was judged less likely than death by botulism even though it is 52 times more frequent”. The same example was on the Week 1 slides.

4. Social norms

“People are social beings who respond to group norms” (Van Der Linden et al., 2015)

We pay close attention to what others think and do, and we tend to adjust to it.

The quote is the heading of insight 2 in the reading.

Social norms: paying taxes

In the United Kingdom, people who were late with their tax payment received a reminder letter. Some of these letters included one additional sentence, for example: “Nine out of ten people pay their tax on time.”

Across two field experiments with more than 200,000 people, including such messages “increases payment rates for overdue tax” (Hallsworth et al., 2017).

Hallsworth, List, Metcalfe & Vlaev 2017, Journal of Public Economics, DOI 10.1016/j.jpubeco.2017.02.003. Abstract: “two large-scale natural field experiments”, “administrative data from > 200,000 individuals in the United Kingdom”.

Messages about what others do (descriptive norms) seemed to be more effective than messages about what one ought to do (injunctive norms). Abstract: “Descriptive norms appear to be more effective than injunctive norms”.

Social norms: paying taxes

(Own figure, data from Table 4 in Hallsworth et al., 2017)

Experiment 1: 101,471 people in the UK who had not paid their tax by 31 July 2011. Everyone received a reminder letter. In five groups, one sentence was added after the first sentence of the standard letter. Outcome: paid, or started to pay, within 23 days. With the standard letter, 35.8% did. The bars show how many percentage points more did so with each added sentence (Table 4, column I). Jointly, the five messages “increase payments after 23 days by 2.2 percentage points compared to the control message”.

A link to loss aversion: the gain-framed and the loss-framed sentence about public services did equally well. Abstract: “We find no evidence that loss framing is more effective than gain framing.”

Does it replicate? The record is mixed. A meta-analysis of tax nudges (Antinyan & Asatryan 2024, Economic Journal, DOI 10.1093/ej/ueae088): “simple reminders increase the probability of compliance by 2.7 percentage points, while tax morale and deterrence nudges increase compliance by an additional 1.4 and 3.2 percentage points”. A study with all income taxpayers in Belgium (De Neve et al. 2021, Journal of Political Economy, DOI 10.1086/713096): “invoking tax morale is not effective and often backfires”.

5. Mental accounting

Imagine you are about to buy a jacket for $125 and a calculator for $15. The calculator is on sale for $10 at another branch of the store, 20 minutes’ drive away. Would you make the trip?

Now imagine the calculator costs $125 and is on sale for $120 at the other branch. Would you make the trip?

Hands after each version. Adapted from Tversky & Kahneman 1981, Problem 10. In the original the two versions went to two different groups (N = 93 and N = 88) and the jacket price was swapped as well ($15 in the second version), so that the total was the same.

5. Mental accounting

In the original study, 68% were willing to make the trip to save $5 on a $15 calculator. Only 29% were willing to make the same trip to save $5 on a $125 calculator (Tversky & Kahneman, 1981).

People do not seem to treat the $5 as simply $5. They set it against the price of the item, as if each purchase had its own account.

This is called mental accounting: people seem to keep separate mental budgets, and money in one budget is not treated like money in another.

A second example in the same paper, Problems 8 and 9: after losing a $10 bill, 88% would still buy a $10 theatre ticket. After losing the ticket itself, only 46% would buy another one. Mental accounting effects replicated across 21 countries (Priolo et al. 2026).

3. Framing effects in climate change communication

How does framing relate to mental models?

They are two sides of the same coin.

A framing effect is the “odd observation”, where the same information only packaged slightly differently causes different reactions.

If similar framing effects are regularly appearing across contexts, we call it a mental model.

A (kind of) practical definition of framing

How to package the essentially same information slightly differently, so that it appeals to mental models.

What’s the evidence on framing effects in climate change communication?

1. Loss aversion and climate change

“Much of the media, scientific, and policy discourse around climate change has consistently invoked the idea of ‘losses.’ For example, climate solutions are often framed as an immediate loss for society (e.g., higher taxes, reducing energy consumption).” (Van Der Linden et al., 2015)

The quote is from the reading (insight 4, “Framing the Big Picture: Nobody Likes Losing (but Everyone Likes Gaining)”).

Gain frame or loss frame?

In one experiment, 161 participants read the same information about climate change impacts, framed either in terms of gains or in terms of losses (Spence & Pidgeon, 2010).

Gain frame

“By mitigating climate change, we can prevent further increases in winter floods in maritime regions and flash floods throughout Europe.”

“By preventing further sea-level rises, we can prevent the inland migration of beaches and save up to 20% of coastal wetlands …”

Loss frame

“Without mitigating climate change, we will see further increases in winter floods in maritime regions and flash floods throughout Europe.”

“With further sea-level rises, beaches will migrate inland and threaten up to 20% of coastal wetlands …”

Gain frames led to more positive attitudes towards climate change mitigation than loss frames.

Spence & Pidgeon 2010, Global Environmental Change, DOI 10.1016/j.gloenvcha.2010.07.002. The extracts are from the paper’s Appendix A. The text was adapted from the 2007 IPCC report, about 900 words in total, “adjusted to either describe the losses that will occur from climate change or the gains that will occur from climate change mitigation”. Participants lived in or near Cardiff. A second manipulation varied whether the impacts were described for the UK or for continental Europe.

Worth pointing out: the gain frame here is about losses that can be prevented. Both versions talk about floods. What differs is whether the floods are presented as avoidable through action or as coming without action.

Abstract: “gain frames were superior to loss frames in increasing positive attitudes towards climate change mitigation, and also increased the perceived severity of climate change impacts.” The loss frame had effects of its own: fear was higher (M = 3.45 against 3.10) and people recalled more statements (M = 6.93 against 4.96). The abstract says the advantage of the gain frame “was partially suppressed by lower fear responses and poorer information recall”.

How are these findings (gain frames beat loss frames) compatible with loss aversion (losses weigh more than gains)?

Go read the section in the paper again and discuss with your neighbor (5 minutes).

(Insight 4 in the reading, “Framing the Big Picture: Nobody Likes Losing (but Everyone Likes Gaining)”)

Pointer for students: insight 4 in the reading, “Framing the Big Picture: Nobody Likes Losing (but Everyone Likes Gaining)”.

Why gain frames, if people would like to avoid losses?

They are not compatible.

Important take-away: mental models are no universal rules that always apply in different contexts.

Sometimes, other factors are more important. In this case, maybe providing hope and optimism.

The major take-away of this exercise: a mental model describes a regularity. It does not predict what happens in every context.

On hope and optimism: the authors of the experiment suggest something close to it. Presenting mitigation in a gain frame “might cue a sense of personal or societal efficacy (in that participants then believe there is something which can be done about the problem)” (Spence & Pidgeon 2010, discussion).

One explanation, which Spence & Pidgeon 2010 give themselves: what matters is whether the behaviour is seen as safe or as risky. Faced with losses, people take risks. Faced with gains, they prefer the safe option. Mitigating climate change is the safe option, because it prevents possible future damage, so a gain frame fits it better. They start from loss aversion, “the idea that individuals tend to dislike losses more than equivalent gains”, and note that “Empirical evidence supporting the stronger influence of loss frames is inconsistent”. The pattern in the literature: “loss frames are more effective in changing behaviours considered risky whilst gain frames are more effective within behaviours considered safe”. The reason from prospect theory: “people are less inclined to take risks when considering gains”, while “people will take risks to avoid losses”.

Their example from health: sunscreen use is a prevention behaviour, seen as low in risk, and is promoted better with gain frames. Testing for HIV is a detection behaviour, seen as risky in the short term because one may get bad news, and is promoted better with loss frames. Their argument for climate change: “climate change mitigation actions are most similar to prevention behaviour in that mitigation behaviour is undertaken in order to prevent potential future negative impacts of climate change.”

The link to our experiment: described as gains, most people took the sure option (72% chose Program A). Described as losses, most people gambled (78% chose Program D).

On the puzzle itself: a meta-analysis “of 93 studies involving over 20,000 participants” on message framing “did not find a single context where loss-framed appeals had statistically greater persuasive power than gain-framed appeals” (as reported in Gal & Rucker 2018, p. 509).

The reading adds a second argument about public debate in general. Its passage (insight 4): “when climate change impacts are framed as potential (i.e., uncertain) losses in the distant future, whereas climate change solutions are framed as certain losses for society at present, it encourages people to conclude that maintaining the status quo may be ‘worth the gamble.’” Just before: “prospect theory (Kahneman & Tversky, 1979) demonstrates that people are more risk-seeking in loss domains than they are in gain domains. In particular, people are more reluctant to take action when losses are paired with uncertainty (Tversky & Shafir, 1992).” This argument is about how climate policy is usually talked about. It does not explain the result of the experiment, because the texts in the experiment never mentioned the costs of acting.

One could link this second argument to temporal discounting: the losses from not acting lie in the distant future, where they may count for less. This link is our reading, the reading itself stresses the uncertainty.

A limitation: do future losses fade like future gains?

The temporal discounting argument assumes that a loss in the distant future counts for less than a loss today. In the 61-country study, 40.1% of participants did not treat losses that way. They preferred to receive money now, but they also preferred to pay now over paying more in 12 months (Ruggeri et al. 2022). For these people, a future loss does not seem to fade like a future gain.

What we would expect. Someone who takes $500 now over $550 in a year tells us that money in a year is worth clearly less to them than money today. If now beats later in the same way for losses, the same person should prefer to pay $550 in a year over paying $500 today, because the later payment should also feel smaller. The temporal discounting argument uses exactly this logic: the climate loss is far away, so it shrinks, and the immediate cost of acting looms larger.

What Ruggeri et al. 2022 find. Many people do the opposite for losses. The authors’ example: people “prefer to receive $500 now over $550 in 12 months, but also prefer to pay $500 now over paying $550 in 12 months”. They take the money now, and they also pay now, although paying later would be the consistent choice. The paper calls this gain–loss asymmetry: “Gains are discounted more than losses”. It was the most common of the anomalies they tested: “from 13.8% for absolute magnitude to 40.1% for gain–loss asymmetry”, and “Gain–loss rates were the most common anomaly in 80.3% (49) of the countries”.

Why this matters for us. These people accept a smaller loss today to avoid a larger loss later. This is the kind of choice climate action asks for. So a delayed loss does not simply fade for them, and “the distant loss counts for less” is weaker than it sounds. Note that this does not go against loss aversion itself. It fits the idea that people take losses seriously, also when they lie in the future. What it goes against is the combination of loss aversion and temporal discounting.

What it does not show. 40.1% is a minority of participants. The choices are about money, with delays of 12 months. And risk was not part of the task: “we do not explicitly test risk in the temporal measures”. Climate impacts are uncertain and decades away. The part of the reading’s argument that rests on uncertainty (“worth the gamble”) is untouched, and this is the part the reading itself stresses.

2. Temporal discounting and climate change

For many people, climate change is distant in all three ways at once: “a nonurgent and psychologically distant risk—spatially, temporally, and socially” (Van Der Linden et al., 2015).

The reading’s advice: “Emphasize the present and make climate change impacts and solutions locally relevant.”

The quote is from the reading (insight 3, “Out of Sight, Out of Mind: The Nature of Psychological Distance”), the advice from its Table 1.

Discounting in a climate game

In a group experiment framed around climate change, participants could contribute money to avoid a collective loss.

The rewards of cooperating were paid out after one day, after seven weeks, or several decades later.

With a delay of several decades, cooperation decreased markedly: “all groups failed to reach the collective target”. The seven-week delay had a weaker effect.

(Jacquet et al., 2013)

Jacquet, Hagel, Hauert, Marotzke, Röhl & Milinski 2013, Nature Climate Change, DOI 10.1038/nclimate2024.

The setup: a “collective-risk group experiment framed around climate change”. Participants played in groups of 6 over 10 rounds. In each round, every player could invest €0 (selfish), €2 (the fair share) or €4 (altruistic) towards a collective target. If the group missed the target, the players risked “losing an additional endowment with a high probability”. So cooperating means giving up money now for a benefit that only comes if the group as a whole succeeds.

What was varied: when the reward for cooperation was paid. “The rewards of defection were immediate, whereas the rewards of cooperation were delayed by one day, delayed by seven weeks (intragenerational discounting), or delayed by several decades and spread over a much larger number of potential beneficiaries (intergenerational discounting).”

The result: “intergenerational discounting leads to a marked decrease in cooperation; all groups failed to reach the collective target. Intragenerational discounting was weaker by comparison.”

The authors’ conclusion: “international negotiations to mitigate climate change are unlikely to succeed if individual countries’ short-term gains can arise only from defection”.

3. Availability bias and climate change

“Statistical descriptions of the risk of climate change often fail to elicit action because statistical information, by itself, means very little to (most) people. Experience, on the other hand, can be a powerful teacher.” (Van Der Linden et al., 2015)

The quote is from the reading (insight 1, “The Human Brain Privileges Experience Over Analysis”). It continues with an example of availability: “although the odds of death or injury from a terrorist attack in the United States are very low, terrorism is ranked as a top national priority, whereas the reality of climate change is not. The difference lies in the fact that for terrorism, vivid, memorable experiences readily come to mind”. And on why climate change is different: it “is a slow-moving, ‘invisible’ process that cannot easily be experienced directly”.

Today’s weather and beliefs about climate change

582 Americans and 290 Australians were asked how convinced they are that global warming is happening, and whether the day was colder or warmer than usual (Li et al., 2011).

Those who found the day much warmer than usual were about one point more convinced, on a 4-point scale, than those who found it much colder.

In a further study, they also donated more to a climate charity: $2.04 on average, against $0.48.

People who found today warmer than usual also reported more unusually warm days over the past year (N = 270, r = 0.41) (Zaval et al., 2014), suggesting availibiltiy bias in memory.

Li, Johnson & Zaval 2011, Psychological Science, 22(4), 454–459, DOI 10.1177/0956797611400913.

Studies 1a and 1b: a short web survey with 582 people in the United States (in winter) and 290 in Australia (in summer). Belief (“how convinced are you that global warming is happening”) and concern were measured from 0 (not at all) to 3 (completely / a great deal). Perceived temperature went from −2 (much colder than usual for the time of year) to 2 (much warmer). “For the two temperature extremes, the difference in perceived deviation resulted in a one-point difference (on a 4-point scale) in both belief and concern.” The order of the questions made little difference, which suggests that people were not aware of the influence.

Study 2: 251 participants, with 18 minutes of unrelated tasks between the two questions. They could donate part of their payment to the charity Clean Air-Cool Planet. “Participants who thought the temperature was much colder than usual donated an average of only $0.48 (SD = $1.74; 24% of participants donated), whereas those who thought the temperature was much warmer than usual donated an average of $2.04 (SD = $2.43; 63% of participants donated).”

The data are correlational. To address reverse causation (believers may perceive the day as warmer), the authors used the actual temperature deviation as an instrument for the perceived one. The effect on belief and concern remained.

Their interpretation is attribute substitution, “in which an easily accessible judgment (the current day’s local temperature) is used in place of a more complex and less accessible one (global temperature trends)”.

The follow-up: Zaval, Keenan, Johnson & Weber 2014, Nature Climate Change, DOI 10.1038/nclimate2093, five studies with US participants. People “use less relevant but available information (for example, today’s temperature) in place of more diagnostic but less accessible information (for example, global climate change patterns)”.

• Study 1 (N = 686): the effect appeared whether the question said “global warming” or “climate change”.
• Study 2 (N = 330): reading a text on the difference between weather fluctuations and climate change did not remove it. “increased knowledge does not eliminate the local warming effect”.
• Study 3a (N = 300): words related to heat, used as a prime, increased belief. This is the study that did not replicate in Many Labs 2 (original d = 0.31, replication d = −0.03, N = 4,204).
• Study 3b (N = 251): asking about yesterday’s temperature had no effect. It seems to be the immediate experience that counts.
• Study 4 (N = 270), on the slide: participants were asked “over the past year, what percentage of days seemed to be warmer than usual for that time of year, compared with the historical average?” This estimate correlated with how warm today seemed (r = 0.41), and with belief in global warming (r = 0.35). The authors conclude that present temperature abnormalities “lead to an overestimation of the frequency of similar past events”.

4. Social norms and climate change

“the more people hear social referents (e.g., family and friends) talk about the risk of climate change, and the more climate change is viewed within one’s social network as a risk that requires action, the more it amplifies an individual’s own risk perception and intention to act” (Van Der Linden et al., 2015)

The quote is from the reading (insight 2, “People Are Social Beings Who Respond to Group Norms”). The reading also names the two kinds of evidence that come later in this block. On consensus: “highlighting descriptive norms such as the high degree of scientific agreement (97%) on human-caused climate change can lead to greater science acceptance and support for public action”. On energy use: “when people are informed about the average energy consumption of their neighbors, they tend to adjust their own energy use to conform to the group norm”. And a condition: social norms are an “underemployed” lever, but “to be leveraged, they must first be in place”.

Would you be willing to give 1% of your income to fight climate change?

What share of people in your country do you think would say yes?

Hands for the first question. For the second: more than half, or less than half? The question wording here is simplified, it is not the survey’s exact item.

Most people say yes, and most underestimate how many others do

(Fig. 3b from Andre et al., 2024, CC BY 4.0, cropped)

Andre et al. 2024: representative survey in 125 countries, nearly 130,000 people. “69% of the global population expresses a willingness to contribute 1% of their personal income, 86% endorse pro-climate social norms and 89% demand intensified political action.” Left: the world average, actual willingness 69%, average perceived willingness of others 43%. Right: one line per country. Circles are actual proportions, triangles are perceived proportions.

Telling people what others think

If people underestimate how much others care about climate change, the norm they respond to is weaker than the real one. This could hold climate action back.

Our mental model 4 suggests a simple fix: tell people how many others care.

However, in a study across 11 countries, giving people the actual numbers was “largely ineffective” (Geiger et al., 2025).

Andre et al. 2024 make the first point: “This perception gap, combined with individuals showing conditionally cooperative behaviour, poses challenges to further climate action.”

Geiger et al. 2025, Psychological Science, n = 3,653 in 11 countries. People in all 11 countries underestimated the share of others who think that climate change is mainly or partly human-caused. Perceived minus actual share, in percentage points (Fig. 2): Brazil −20.8, Poland −19.7, Mexico −17.1, Germany −17.0, Thailand −16.4, Canada −15.7, Italy −14.7, Japan −14.1, China −12.6, India −12.0, Indonesia −7.5.

The intervention told participants the actual share from a recent survey in their country. Providing this information “was largely ineffective, except for a slight increase in willingness to express one’s proclimate opinion”. It “did not change personal willingness to change one’s lifestyle” and “did not influence personal support for government action”.

Social norms that change behaviour: energy use at home

In the United States, households received letters “comparing their electricity use to that of their neighbors”. In randomised field experiments with 600,000 households, these reports reduced energy consumption by 2.0% on average (Allcott, 2011).

This equals the effect of a short-run electricity price increase of 11 to 20%.

And it lasts: when the reports stopped after two years, the effect decayed by only 10 to 20 percent per year (Allcott & Rogers, 2014).

A synthesis of 430 field studies on climate-friendly behaviour found that “interventions based on social comparisons or financial incentives were the most effective” (Bergquist et al., 2023).

Allcott 2011, Journal of Public Economics, DOI 10.1016/j.jpubeco.2011.03.003, on the Home Energy Reports of the company Opower: “Using data from randomized natural field experiments at 600,000 treatment and control households across the United States, I estimate that the average program reduces energy consumption by 2.0%.” The effect “is equivalent to that of a short-run electricity price increase of 11 to 20%”. It differs by household: “households in the highest decile of pre-treatment consumption decrease usage by 6.3%, while consumption by the lowest decile decreases by only 0.3%”.

Allcott & Rogers 2014, American Economic Review, DOI 10.1257/aer.104.10.3003, on the same programme: “social comparison-based home energy reports are repeatedly mailed to more than six million households nationwide”. “if reports are discontinued after two years, effects are relatively persistent, decaying at 10–20 percent per year”, and consumers “continue to respond to repeated treatment even after two years”.

Bergquist et al. 2023, PNAS, DOI 10.1073/pnas.2214851120: a second-order meta-analysis of 10 meta-analyses with 430 primary studies in field settings. “interventions based on social comparisons or financial incentives were the most effective, while education or feedback was the least effective.” How big? Overall, adjusted for publication bias, d = 0.18, which “translates into a mean treatment effect of 7 percentage points”. In large-scale interventions (n > 9,000) the effect “was adjusted downward to approximately 2 percentage points”, and home energy reports are their example of such a large-scale, less direct technique.

A caution on the wording of norm messages (Schultz et al. 2007, Psychological Science, DOI 10.1111/j.1467-9280.2007.01917.x): a message about average neighbourhood energy use “produced either desirable energy savings or the undesirable boomerang effect, depending on whether households were already consuming at a low or high rate”. Adding a message of social approval or disapproval “eliminated the boomerang effect”.

Social norms: what scientists think

“97% of climate scientists agree that human-caused climate change is happening.”

Across 27 countries (n = 10,527), the consensus message substantially reduced misperceptions and slightly increased climate change beliefs.

It did, however, not directly increase support for public action (Većkalov et al., 2024).

The scientific consensus can be read as a norm among experts. van der Linden 2021 (Current Opinion in Psychology, DOI 10.1016/j.copsyc.2021.01.005) describes the gateway belief model as an “account of how attitude change happens in response to normative cues about scientific agreement”.

van der Linden, Leiserowitz, Feinberg & Maibach 2015, PLoS ONE, DOI 10.1371/journal.pone.0118489: “increasing public perceptions of the scientific consensus is significantly and causally associated with an increase in the belief that climate change is happening, human-caused and a worrisome threat. In turn, changes in these key beliefs are predictive of increased support for public action.”

Većkalov et al. 2024, Nature Human Behaviour, DOI 10.1038/s41562-024-01928-2, online convenience samples from 27 countries: “the classic message substantially reduces misperceptions (d = 0.47, 95% CI (0.41, 0.52)) and slightly increases climate change beliefs (from d = 0.06, 95% CI (0.01, 0.11) to d = 0.10, 95% CI (0.04, 0.15)) and worry (d = 0.05, 95% CI (−0.01, 0.10)) but not support for public action directly.” Both messages were “more effective for audiences with lower message familiarity and higher misperceptions, including those with lower trust in climate scientists and right-leaning ideologies”.

The same pattern as with the public consensus on the previous slides: the misperception is corrected more easily than attitudes or behaviour are changed.

5. Mental accounting and climate change

Carbon taxes meet strong public resistance. How acceptable they are depends on what happens with the revenue.

Across six experiments in the United Kingdom and France (N = 7,100), a carbon tax was more acceptable when its revenue went to environmental projects (Mus et al., 2023).

The authors’ explanation: people expect money raised in one domain to be spent in that same domain. The same pattern appeared for tobacco taxes and health spending, and for inheritance taxes and social spending.

Mus et al. 2023: “people create mental budgets where the origin of revenues is matched thematically with their domain of use”. Most support for a mixed scheme: most revenue to green projects, the rest to low-income households.

Mental accounting and carbon taxes

(Fig. 2 from Mus et al., 2023, CC BY 4.0)

Mus, Mercier & Chevallier 2023. Support for a carbon tax increase on a ten-point scale, with revenues either not earmarked or earmarked for environmental protection. UK: N = 706, d = 0.23. France: N = 429, d = 0.71.

Other framing effects

Not every frame builds on one of our five mental models.

Some frames seem to work through the words that are chosen, or through the values and identities that a message appeals to.

Other framing effects: words

Is it “global warming” or “climate change”?

In a survey experiment with 2,267 Americans, Republicans were less likely to say the phenomenon is real when it was called “global warming” (44.0%) than when it was called “climate change” (60.2%).

Democrats were unaffected by the wording (86.9% and 86.4%).

(Schuldt et al., 2011)

Schuldt, Konrath & Schwarz 2011, Public Opinion Quarterly, DOI 10.1093/poq/nfq073. “the partisan divide on the issue dropped from 42.9 percentage points under a ‘global warming’ frame to 26.2 percentage points under a ‘climate change’ frame”.

Other framing effects: labels

898 Americans chose between otherwise identical products. One of them included a surcharge for emitted carbon dioxide.

The surcharge was labelled either as a “tax” or as an “offset”.

The label changed the preferences of Republicans and Independents. It did not affect the preferences of Democrats.

(Hardisty et al., 2009)

Hardisty, Johnson & Weber 2010, Psychological Science, “A Dirty Word or a Dirty World?”, DOI 10.1177/0956797609355572. The abstract speaks of “an earmarked tax or an offset”, a link to mental accounting.

Other framing effects: values

Environmental messages are mostly phrased in terms of harm and care, moral concerns that are more deeply held by liberals than by conservatives.

When pro-environmental messages were reframed in terms of purity, a value that resonates more among conservatives, the difference in environmental attitudes between liberals and conservatives was “largely eliminated”.

(Feinberg & Willer, 2012)

Feinberg & Willer 2013, Psychological Science, “The Moral Roots of Environmental Attitudes”, DOI 10.1177/0956797612449177 (published online 2012). United States.

Other framing effects: patriotism

In the United States, pro-environmental initiatives were framed “as patriotic and necessary to maintain the American ‘way of life’”.

This framing increased belief in climate change and support for climate policies (Mason et al., 2024).

Mason, Vlasceanu & Jost 2024, PNAS. The same paper also tested the frame across 63 countries, see “How big are the effects?”. The words and labels examples show a similar thing within the United States: who hears a frame matters. More on identity and motivated reasoning in week 9.

Other framing effects: data

The same climate trend can be shown as continuous data (mean temperature) or as binary data (did the lake freeze this winter, yes or no).

People who saw the binary data perceived the impact of climate change as larger (d = 0.40, N = 799) (Liu et al., 2025).

The authors link this to the “boiling frog” effect: a slow, gradual shift is easy to overlook.

Liu, Snell, Griffiths & Dubey 2025, Nature Human Behaviour. “our approach does not involve selective data presentation but rather compares different datasets that reflect equivalent trends in climate change over time”. A link to mental model 3.

How big are the effects?

The ten most-cited climate messages

A megastudy tested the ten most-cited climate messaging strategies on 13,544 Americans (Voelkel et al., 2026).

Six of the ten messages changed attitudes, by 1 to 4 percentage points.

No message increased donations to environmental causes.

A megastudy tests many messages at once, in one large sample, on the same outcomes. Voelkel et al. 2026: “Six messages significantly affected multiple preregistered attitudes, with effects ranging from 1 to 4 percentage points.” “No message increased pro-environmental donations, suggesting costly behaviours are difficult to influence with messaging alone.” “Persuasiveness varied little across party lines”. Five replications of three highly cited strategies (N = 3,216) found “limited evidence of persuasive effects”.

Eleven interventions in 63 countries

59,440 participants were assigned to one of 11 interventions designed by experts (Vlasceanu et al., 2024).

Effects were small, and largely limited to people who were not sceptical about climate change to begin with.

Reducing psychological distance, our mental model 2, strengthened climate beliefs by 2.3%.

No intervention increased an effortful behaviour, a task that led to trees being planted. Several interventions reduced it.

Vlasceanu et al. 2024. “the interventions’ effectiveness was small, largely limited to nonclimate skeptics, and differed across outcomes: Beliefs were strengthened mostly by decreasing psychological distance (by 2.3%), policy support by writing a letter to a future-generation member (2.6%), information sharing by negative emotion induction (12.1%), and no intervention increased the more effortful behavior—several interventions even reduced tree planting.”

One frame in 63 countries

The patriotic frame from the United States was also tested across 63 countries. Similar messages had small effects on average (Mason et al., 2024).

In some countries the frame was moderately successful (Brazil, France, Israel). In others it backfired (Germany, Belgium, Russia).

Mason, Vlasceanu & Jost 2024. The country-level analyses were exploratory.

Effects of messaging interventions, even if framed by psychological researchers who are experts, are small.

They typically shift attitudes by only a few percentage points.

We will discuss these studies in more detail in session 11, “From communication to behavior change”.

4. Revisiting the deficit model

In week 2, part of the conclusion was that facts alone rarely change minds.

A recent line of research complicates this picture.

This is often the starting point when arguing that psychology matters.

Conversations with an AI

2,190 people who believed in a conspiracy theory discussed it with an AI model, which responded with evidence tailored to their own arguments (Costello et al., 2024).

Belief in the conspiracy theory fell by about 20%.

The effect was still there two months later, even among participants with deeply entrenched beliefs.

Costello, Pennycook & Rand 2024. “personalized evidence-based dialogues with GPT-4 Turbo. The intervention reduced conspiracy belief by ~20%. The effect remained 2 months later, generalized across a wide range of conspiracy theories, and occurred even among participants with deeply entrenched beliefs.” Their reading: “previous failures in correcting conspiracy beliefs may be due to counterevidence being insufficiently compelling and tailored”.

And for climate change?

Here the picture is more mixed. In two studies, climate sceptics had short conversations with ChatGPT (Hornsey et al., 2025).

The conversations led to small increases in intentions to act, and to less confidence in sceptical views.

The effects on scepticism itself were inconsistent across the two studies, and they tended to fade over time.

Hornsey et al. 2025. Study 1 n = 949, study 2 n = 333. “the effects on reducing scepticism among sceptics are inconsistent across two studies and susceptible to decay over time. Increasing the conversation length from three to six rounds yielded no additional benefit.”

Why do facts seem to work in conversations with chatbots?

A reasoning problem

Pierre looks at Linda. Linda looks at Henry.

Pierre is married. Henry is not married.

Does someone who is married look at someone who is not married?

Yes · No · We cannot know

First alone: hands for each of the three answers. Then two minutes with the neighbour, and hands again.

Solution: yes. We do not know whether Linda is married, but it does not matter. If she is married, she looks at Henry, who is not. If she is not married, Pierre, who is married, looks at her.

What tends to happen with problems like this one: reasoning alone, many people settle on a wrong answer (“We cannot know”). In discussion, the correct argument tends to convince them. The next slide has numbers for a similar task.

Reasoning as a social process

“The main function of reasoning is argumentative” (Mercier & Sperber, 2011).

On this view, reasoning “should produce its best results” in group discussions. In a classic logic task, about 10% of people find the correct answer alone. In groups, 80% do.

Alone, people mostly look for “arguments supporting their views”. In a discussion, others evaluate these arguments.

Mercier & Sperber 2011, Behavioral and Brain Sciences, DOI 10.1017/S0140525X10000968. “The main function of reasoning is argumentative: Reasoning has evolved and persisted mainly because it makes human communication more effective and advantageous.” (p. 60) “Accordingly, reasoning should produce its best results when used in argumentative contexts, most notably in group discussions” (p. 61).

The logic task is the Wason selection task: “the performance of participants is generally very poor, hovering around 10% of correct answers”, but “when participants had to solve the task in groups, they reached the level of 80% of correct answers” (pp. 62–63). And: “group discussions provide a much more efficient way of holding the confirmation bias in check” (p. 65).

Mass persuasion is difficult

A review of the evidence on religious proselytizing, propaganda, political campaigns and advertising concludes that “communication is much less influential than often believed” (Mercier, 2017).

People seem to check messages against what they already believe, and against how much they trust the source.

“Beliefs that lead to costly behavior are even less likely to be accepted.” This fits what the large climate studies found.

Mercier 2017, Review of General Psychology, DOI 10.1037/gpr0000111, abstract. People “check the plausibility of messages against their background beliefs, calibrate their trust as a function of the source’s competence and benevolence, and critically evaluate arguments offered to them.” “Converging evidence reveals that communication is much less influential than often believed—that religious proselytizing, propaganda, advertising, and so forth are generally not very effective at changing people’s minds.” On political campaigns and the media: “In most cases, the influence of the media and political campaigns is small.” (p. 113)

The link back to “How big are the effects?”: one-way messages moved attitudes by a few percentage points and did not increase costly behaviour.

Chatbots as scaled-up dialogue

A single message cannot answer back. In a discussion, a person’s counterarguments can be addressed one by one.

This may also explain the conversations with an AI. Earlier attempts may have failed because the counterevidence was “insufficiently compelling and tailored” (Costello et al., 2024).

Altay, Schwartz, Hacquin, Allard, Blancke & Mercier 2022, Nature Human Behaviour, DOI 10.1038/s41562-021-01271-w. Their starting point: “Discussion is more convincing than standard, unidirectional messaging, but its interactive nature makes it difficult to scale up. We created a chatbot to emulate the most important traits of discussion.” The result: “Providing participants with good arguments rebutting the most common counterarguments against GMOs led to much more positive attitudes towards GMOs, whether the participants could immediately see all the arguments or could select the most relevant arguments in a chatbot.” And: “Participants holding the most negative attitudes displayed more attitude change in favour of GMOs.”

A caveat from the same study: it may be the good arguments more than the interactivity. “Participants updated their beliefs when presented with good arguments, but we found no evidence that an interactive chatbot proves more persuasive than a list of arguments and counterarguments.”

A second example: in France, 701 participants could ask a chatbot their questions about COVID-19 vaccines. After a few minutes, their attitudes towards vaccination were more positive (ß = 0.23) and their intentions to get vaccinated were higher (ß = 0.12) (Altay, Hacquin, Chevallier & Mercier, Journal of Experimental Psychology: Applied, DOI 10.1037/xap0000400). Their argument: if people “do not have an interlocutor who can address these counter-arguments (e.g., if they read a leaflet), they are less likely to change their minds”.

Costello et al. 2024 on their own result: the AI can “tailor counterarguments to specific conspiracies, reasoning, and evidence the believer brings to bear”. They also write that “the causal mechanisms underpinning our results remain unformalized”.

To take home

Mental models are general features of human psychology that describe how people think, decide, and act. They often seem to deviate from rational choice.

Framing means packaging essentially the same information differently, by appealing to these mental models.

Mental models are no universal rules that always apply in different contexts. Sometimes, other factors are more important.

To take home

For climate change, large studies suggest that framing helps modestly: a few percentage points on attitudes, and little on costly behaviour.

Facts still matter: they seem to work better in a dialogue, where arguments can be exchanged, than in one-way messages.

References

Allcott, H. (2011). Social norms and energy conservation. Journal of Public Economics, 95(9–10), 1082–1095. https://doi.org/10.1016/j.jpubeco.2011.03.003

Allcott, H., & Rogers, T. (2014). The Short-Run and Long-Run Effects of Behavioral Interventions: Experimental Evidence from Energy Conservation. American Economic Review, 104(10), 3003–3037. https://doi.org/10.1257/aer.104.10.3003

Andre, P., Boneva, T., Chopra, F., & Falk, A. (2024). Globally representative evidence on the actual and perceived support for climate action. Nature Climate Change, 14(3), 253–259. https://doi.org/10.1038/s41558-024-01925-3

Bergquist, M., Thiel, M., Goldberg, M. H., & van der Linden, S. (2023). Field interventions for climate change mitigation behaviors: A second-order meta-analysis. Proceedings of the National Academy of Sciences, 120(13), e2214851120. https://doi.org/10.1073/pnas.2214851120

Costello, T. H., Pennycook, G., & Rand, D. G. (2024). Durably reducing conspiracy beliefs through dialogues with AI. Science, 385(6714), eadq1814. https://doi.org/10.1126/science.adq1814

Feinberg, M., & Willer, R. (2012). The Moral Roots of Environmental Attitudes. Psychological Science, 24(1), 56–62. https://doi.org/10.1177/0956797612449177

Geiger, S. J., Köhler, J. K., Delabrida, Z. N. C., Garduño-Realivazquez, K. A., Haugestad, C. A. P., Imada, H., Iyer, A., Maharja, C., Mann, D. C., Marczak, M., Melville, O., Nijssen, S. R. R., Powdthavee, N., Praptiwi, R. A., Ranade, G., Rosa, C. D., Vitale, V., Winkowska, M., Zhang, L., & White, M. P. (2025). What We Think Others Think and Do About Climate Change: A Multicountry Test of Pluralistic Ignorance and Public-Consensus Messaging. Psychological Science, 36(6), 421–442. https://doi.org/10.1177/09567976251335585

Hallsworth, M., List, J. A., Metcalfe, R. D., & Vlaev, I. (2017). The behavioralist as tax collector: Using natural field experiments to enhance tax compliance. Journal of Public Economics, 148, 14–31. https://doi.org/10.1016/j.jpubeco.2017.02.003

Hardisty, D. J., Johnson, E. J., & Weber, E. U. (2009). A Dirty Word or a Dirty World? Psychological Science, 21(1), 86–92. https://doi.org/10.1177/0956797609355572

Hornsey, M. J., Pearson, S., Bretter, C., MacInnes, S., Nylund, J. L., & Rekker, S. (2025). The promise and limitations of using GenAI to reduce climate scepticism. Nature Climate Change, 1–7. https://doi.org/10.1038/s41558-025-02425-8

Jacquet, J., Hagel, K., Hauert, C., Marotzke, J., Röhl, T., & Milinski, M. (2013). Intra- and intergenerational discounting in the climate game. Nature Climate Change, 3(12), 1025–1028. https://doi.org/10.1038/nclimate2024

Li, Y., Johnson, E. J., & Zaval, L. (2011). Local Warming: Daily Temperature Change Influences Belief in Global Warming. Psychological Science, 22(4), 454–459. https://doi.org/10.1177/0956797611400913

Lichtenstein, S., Slovic, P., Fischhoff, B., Layman, M., & Combs, B. (1978). Judged frequency of lethal events. Journal of Experimental Psychology: Human Learning and Memory, 4(6), 551–578. https://doi.org/10.1037/0278-7393.4.6.551

Liu, G., Snell, J. C., Griffiths, T. L., & Dubey, R. (2025). Binary climate data visuals amplify perceived impact of climate change. Nature Human Behaviour, 1–10. https://doi.org/10.1038/s41562-025-02183-9

Mason, K. A., Vlasceanu, M., & Jost, J. T. (2024). Effects of system-sanctioned framing on climate awareness and environmental action in the United States and beyond. Proceedings of the National Academy of Sciences, 121(38), e2405973121. https://doi.org/10.1073/pnas.2405973121

Mercier, H. (2017). How Gullible are We? A Review of the Evidence from Psychology and Social Science. Review of General Psychology, 21(2), 103–122. https://doi.org/10.1037/gpr0000111

Mercier, H., & Sperber, D. (2011). Why do humans reason? Arguments for an argumentative theory. Behavioral and Brain Sciences, 34(2), 57–74. https://doi.org/10.1017/S0140525X10000968

Mus, M., Mercier, H., & Chevallier, C. (2023). Designing an acceptable and fair carbon tax: The role of mental accounting. PLOS Climate, 2(10), e0000227. https://doi.org/10.1371/journal.pclm.0000227

Ruggeri, K., Panin, A., Vdovic, M., Većkalov, B., Abdul-Salaam, N., Achterberg, J., Akil, C., Amatya, J., Amatya, K., Andersen, T. L., Aquino, S. D., Arunasalam, A., Ashcroft-Jones, S., Askelund, A. D., Ayacaxli, N., Sheshdeh, A. B., Bailey, A., Barea Arroyo, P., Mejía, G. B., … García-Garzon, E. (2022). The globalizability of temporal discounting. Nature Human Behaviour, 6(10), 1386–1397. https://doi.org/10.1038/s41562-022-01392-w

Schuldt, J. P., Konrath, S. H., & Schwarz, N. (2011). "Global warming" or "climate change"?: Whether the planet is warming depends on question wording. Public Opinion Quarterly, 75(1), 115–124. https://doi.org/10.1093/poq/nfq073

Spence, A., & Pidgeon, N. (2010). Framing and communicating climate change: The effects of distance and outcome frame manipulations. Global Environmental Change, 20(4), 656–667. https://doi.org/10.1016/j.gloenvcha.2010.07.002

Sterman, J. D., & Sweeney, L. B. (2002). Cloudy skies: Assessing public understanding of global warming. System Dynamics Review, 18(2), 207–240. https://doi.org/10.1002/sdr.242

Tversky, A., & Kahneman, D. (1981). The Framing of Decisions and the Psychology of Choice. Science, 211(4481), 453–458. https://doi.org/10.1126/science.7455683

Van Der Linden, S., Maibach, E., & Leiserowitz, A. (2015). Improving Public Engagement With Climate Change: Five “Best Practice” Insights From Psychological Science. Perspectives on Psychological Science, 10(6), 758–763. https://doi.org/10.1177/1745691615598516

Većkalov, B., Geiger, S. J., Bartoš, F., White, M. P., Rutjens, B. T., van Harreveld, F., Stablum, F., Akın, B., Aldoh, A., Bai, J., Berglund, F., Bratina Zimic, A., Broyles, M., Catania, A., Chen, A., Chorzępa, M., Farahat, E., Götz, J., Hoter-Ishay, B., … van der Linden, S. (2024). A 27-country test of communicating the scientific consensus on climate change. Nature Human Behaviour, 1–14. https://doi.org/10.1038/s41562-024-01928-2

Vlasceanu, M., Doell, K. C., Bak-Coleman, J. B., Todorova, B., Berkebile-Weinberg, M. M., Grayson, S. J., Patel, Y., Goldwert, D., Pei, Y., Chakroff, A., Pronizius, E., van den Broek, K. L., Vlasceanu, D., Constantino, S., Morais, M. J., Schumann, P., Rathje, S., Fang, K., Aglioti, S. M., … Van Bavel, J. J. (2024). Addressing climate change with behavioral science: A global intervention tournament in 63 countries. Science Advances, 10(6), eadj5778. https://doi.org/10.1126/sciadv.adj5778

Voelkel, J. G., Ashokkumar, A., Abeles, A. T., Crawford, J. T., Fuller, K., Redekopp, C., Bongiorno, R., Campbell, T. H., Ecker, U. K. H., Feinberg, M., Hart, P. S., Hornsey, M. J., Jost, J. T., Kay, A. C., Leiserowitz, A., Lewandowsky, S., Maibach, E., Nisbet, E. C., Pidgeon, N. F., … Willer, R. (2026). A registered report megastudy on the persuasiveness of the most-cited climate messages. Nature Climate Change, 1–12. https://doi.org/10.1038/s41558-025-02536-2

Zaval, L., Keenan, E. A., Johnson, E. J., & Weber, E. U. (2014). How warm days increase belief in global warming. Nature Climate Change, 4(2), 143–147. https://doi.org/10.1038/nclimate2093