The Drivers of Social and Individual Environmental Behaviour

Sidonie Commarmond

May 18, 2022

Abstract

This study looks into the drivers of environmental behaviour. Based on a theoretical model
that takes into account the effect of cognitive biases, and in particular the present bias; we show
that social norms can be a powerful tool to generalise pro-environmental behaviour. From this
model, we look into the determinants of environmental behaviour and social norms using an
experimental approach with an online panel. We conducted a randomised survey experiment
study on a panel from California, from which we will exploit the variation in distance to the
wildfires during the 2020 and 2021 season. In particular, we evaluate how individual behaviour
and the perception of social norms evolve when individuals are exposed to climate change-
related events or to information about environmental consequences of human behaviour. We
further investigate the heterogeneity of the effects using demographic and economic preferences
data from the panel. We find that time-preferences and social norms perception can explain
environmental behaviour.

Master Thesis - Economics

Under the supervision of Prof. Emeric Henry

School of Research, Sciences Po - Paris

2021-2022

1
Acknowledgements

This Master’s degree, and especially this thesis has been an incredible learning experience. I would

like to thank my supervisor, Emeric Henry, for making this research possible and funding the

experiment we conducted for the thesis. I am immensely grateful for his time, enthusiasm and

insightful advice. I would also like to highlight the invaluable help of our partner Respondi, who

hosted the experiment, and in particular Juliette, who was always available to answer my questions.

In addition, I am thankful to the teachers and students of the Economics Master in Sciences

Po, for the engaging classes and projects we have worked on, providing me with the tools to build

this piece of research. In particular, thank you Dima and Lorenzo; as well as Louis-Marie, Olivia,

Leanne, Hugo, Nathaniel, and others for the insights and the stimulating discussions before, during

and after classes.

Last but not least, I thank my friends and family for their unconditional love and support

during these two years.

All remaining errors are mine.

2
Contents

1 Introduction 6

2 Related literature 9

2.1 Climate change and human behaviour . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.2 Individual behaviour and social norms . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2.3 Social Norms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

3 Model 12

3.1 Model Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3.2 Model Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

3.3 Present Bias Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

3.4 Social norms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

3.4.1 Present Bias Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

4 Experimental Protocol 22

4.1 First Round . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

4.2 Second Round . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

4.3 Treatment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

5 Descriptive Statistics and Balance Checks 25

5.1 Computing the distance to fires . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

5.2 Computing the time bias . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

5.3 Descriptives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

5.4 Balance Checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

3
6 Results 34

6.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

6.1.1 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

6.1.2 Explanatory variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

6.2 Study of the outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

6.2.1 Time Preferences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

6.2.2 Exposure to fires and distance . . . . . . . . . . . . . . . . . . . . . . . . . . 40

6.2.3 Beliefs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

6.2.4 Correlation between outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

7 Analysis and Treatment Effects 48

7.1 Travel and Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

7.2 Donation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

7.3 Heterogeneity in Treatment Effects Analysis . . . . . . . . . . . . . . . . . . . . . . . 56

8 Discussion 58

9 Limits 60

10 Conclusion 61

11 Bibliography 63

12 Appendix I 69

12.1 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

12.1.1 Period 3 decision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

12.1.2 Period 2 decision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

4
 12.1.3 Period 1 decision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

12.1.4 Decision rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

12.1.5 Social Norms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

12.2 Experiment Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

12.2.1 IPCC Report . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

12.2.2 Treatment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

12.2.3 Survey Situations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

12.3 Survey Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

12.3.1 Further analysis of the distance to the fires . . . . . . . . . . . . . . . . . . . 78

12.3.2 LASSO-Selected Covariates . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

12.3.3 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

12.3.4 Identification of the Treatment Effects . . . . . . . . . . . . . . . . . . . . . . 84

12.3.5 Treatment Effect on Beliefs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

12.3.6 Treatment Effect Heterogeneity . . . . . . . . . . . . . . . . . . . . . . . . . . 86

12.3.7 Robustness Checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

13 Appendix II 90

13.1 Treatment Slides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

13.1.1 Survey Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

13.2 RIS Approval . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

5
1 Introduction

The latest Intergovernmental Panel on Climate Change’s (IPCC) report highlights the increasing

emergency of the climate crisis and the narrowing window of time for actions to be taken in order

to ensure a liveable future. This is the Sixth Assessment Report of the IPCC since 1990, confirming

an important number of findings. The report highlights unprecedented changes occurring in the

climate system, some of which will be irreversible for thousands of years. The influence of human

activity-related emissions over these changes is established with full certainty. Every ton of CO2 will

contribute to extra warming which will have further consequences on the equilibrium of the climate

system. The international scientific community is therefore categorical: we need an “immediate,

rapid and sustained reduction in GHG emissions” (IPCC (2021)). To this day, the actions taken

by governments around the world are insufficient to limit global warming to liveable temperatures

before mid-century.1

Making sense of the gap between the magnitude of the crisis and the weakness of the global

response is a substantial challenge for social sciences. Why aren’t governments taking action to

protect their populations? Why do environmentally damaging behaviours persist when their con-

sequences are known and documented? Why is collective action so weak/slow to prevent the crisis

and adapt to the consequences?

Cognitive sciences and behavioural economics research has demonstrated that individual are

biased in their decision-making (Tversky and Kahneman (1974)). This is also true for environmental

behaviour: Thunström et al. (2014) show, for example, that we exploit the underlying uncertainty

around the consequences of global warming as a way of justifying anti-environmental action (a

phenomenon called ”strategic ignorance”). Research in psychology and cognitive sciences further

provides evidence of cognitive biases that contribute to delaying climate action and favouring the

status quo. Effective policies to foster pro-environmental action among citizens should therefore

account for and address those biases.

Due to the regenerative and interrelated characteristics of environmental systems, market-based

solutions such as carbon prices or pollution taxes are only marginally efficient in addressing the ex-

1
[Link]

6
ternalities of human activity on the environment and biodiversity (IPCC (2022)). The climate and

environmental crises are typical situations of social and temporal dilemmas where the individual

costs of sustainable behaviours outweigh the collective benefits of such behaviours. Hence, alongside

addressing cognitive biases, the solutions to the climate and environmental crises lie in the transfor-

mation of social and collective behaviour, through structural, cultural and economic changes (IPCC

(2022)). The role of social norms, in situations of social dilemmas where formal institutions fail to

provide adequate incentives, is key to collectively sustaining virtuous behaviours (Ostrom (1990)).

This thesis aims at understanding the drivers of environmental behaviour, subject to cognitive

biases and how social norms can help counter the effect of the biases to improve collective action.

While several articles have studied the effect of social norms on individual behaviour, there is little

research that looks at the drivers of social norms directly. We look at the role of information and

the experience of climate-change related weather events on the evolution of individual and social

norms. This could ultimately inform policy recommendations to engage in large-scale societal

change. We consider the changes in individual behaviour, willingness to enforce social norms and

perceived social norms,that results from being exposed to climate change information, controlling

for underlying socio-economic characteristics and cognitive biases.

In order to do so, we develop a theoretical model that explores the effect of present-bias on

environmental externalities, using a three-period utility framework in which the agents’ choice

creates an externality that accumulate over the periods (like CO2 emissions). We show first that

present biased individuals contribute to increasing aggregate emissions. Then, we see that social

norms can reduce the effect of the present bias and hence contribute to decreasing the overall level

of emissions. Alongside the model, we use an experimental approach to evaluate the change in

individual and social norms for a panel of 800 individuals in California. We conduct a randomised

survey in two rounds, one week apart from each other. In the first round, the participants are

randomly allocated to the treatment or control group. We provide a short summary of the latest

IPCC report to the treated group and evaluate the choices of the respondents in two situations:

whether to take a plane or not for a week-end trip and their preference in between a renewable

energy provider and an non-renewable energy provider for residential electricity. The treatment

identifies the effect of providing information on individual choices and norm perception in different

7
situations. In the second round of the survey (we have a 65% recall rate), we elicit the participants’

beliefs and concern about climate change, and ask about their willingness to donate their earnings

from the experiment to an environmental NGO.2 We measure the effect of social norms in this

round by informing half of the participants that their choices will be observed and judged by other

members of the panel. Finally, we use geographical data to study the impact of the exposition to

the 2020 and 2021 fire seasons in California.3

We draw several findings from the survey answers. In line with the model, we find that time-

preferences are important drivers of individual behaviour, in particular in the case of longer-term

choices such as the choice of an energy provider. Second, we find that the information treatment

has a varying effect depending on the situations. In the case of travel, participants exposed to

climate change-related information are 15% less likely to take the plane. The effect is less clear for

the choice of energy provider. For willingness to donate, we find that treated individuals are more

likely to donate by 12% when they are in the observed group. These results suggest that being

exposed to information about the consequences of human behaviour on climate change contributes

to changing behaviours in cases where the decisions are public and do not involve a long-term

commitment. The effect on the willingness to donate also shows a persistence of the treatment

effect over time when the decision is observed. Regarding the norms, the exposure to information

increases the social inappropriateness of travelling by plane by 0.2 points (on a scale from 1 to 4);

suggesting that treated individuals are more likely to enforce pro-environmental norms after the

treatment. Regarding the norms, we find that the treatment does not significantly change their

perception and that they are generally underestimated across situations and treatment groups (in

terms of pro-environmental choices). Ultimately, we find that stronger beliefs and concern about

climate change significantly explain the observed outcomes.

The study is organised as follows: Section 2 provides an overview of the related literature,

Section 3 sets out the model used to show the effect of time-bias and social norms while Section 4

details the experimental protocol we used for the empirical approach. Sections 5 and 6 show the

descriptive statistics and results of the survey, and Section 7 analyses the treatment effects. Finally,

2
Friends of the Earth: [Link]
3
See news report by LA Times: [Link]
worst-fires-in-california-history-dixie-camp-and-more

8
Sections 8 and 9 provide a discussion of the results as well as some limitations to the study. Section

10 concludes.

2 Related literature

2.1 Climate change and human behaviour

Cognitive science and psychology literature has highlighted the importance of behavioural biases

in explaining the lack of mitigation, adaptation and environmentally sustainable action (Gifford

(2011); Hoffman and Bazerman (2005)). While cognitive drivers of unsustainable behaviour are

manifold (Engler et al. (2018); Shu and Bazerman (2010)), I choose to focus on time discounting in

this thesis. Although some effects are already observed across the globe, most of the high-impact

consequences of climate change lie in the future. Fighting climate change today involves trade-

offs across time, often across generations, which makes the underlying time preferences in decision

making a critical component of environmental behaviour. Milfont (2012) find evidence that time

perspective influences environmental behaviour and engagement. As such, time parameters are key

components of many economic models. This has lead an important body of literature in behavioural

and empirical economics to measure and investigate time discounting.

Traditionally, economic models use a fixed exponential discounting factor, β, implying agents

would treat all future periods equally. This economic assumption has informed inter-temporal

choice models without being grounded in evidence from empirical studies or from the literature

in psychology. In fact, empirical research in economics has found that human behaviour rather

exhibits time-inconsistent preferences (Thaler (1981); Loewenstein and Thaler (1989)). Present-

bias preferences capture this tendency to give a stronger relative weight to earlier moments as they

get closer. This impatience has many implications for behavioural studies, and calls for a new

type of models with “hyperbolic discounting” to account for the different weight assigned to future

periods relative to the present periods (Laibson (1997); O’Donoghue and Rabin (1999)). These

models include a factor βδ where β ∈ [0, 1] is the present-bias factor. A bias factor < 1 means that

individuals are present-biased and discount future periods in an inconsistent way. They can be aware

9
of this bias, in which case they are sophisticates, and put in place strategic self-control measures to

reduce it; or they can be naive about their bias and make their decisions while considering consistent

time discounting (i.e. only accounting for δ). This creates a phenomenon of ”procrastination” where

individuals always believe they will take a costly action in the future but realise, when they get to

that day that it is still too costly for them to take the action.

Weitzman (1998) argues that macroeconomic modelling cannot tackle issues related to climate

change and the environment if the future is discounted at a steady rate, because it does not take

into account the inconsistency in time preferences of agents. Several studies have indeed found that

growth models are closer to the reality of environmental and resource economics when accounting

for the dynamic inconsistency in time preferences (Hepburn (2010); Weitzman (2001)).

2.2 Individual behaviour and social norms

Preventing climate change and biodiversity collapse will require deep structural changes to social and

individual habits (IPCC (2022)). This will predominantly be achieved through strong government

action, leading an integrated and multi-sectoral approach to adapting our societies to sustainable

levels of production and consumption. The COVID-19 crisis provides evidence that this level of

governmental action, in western democracies, is possible and feasible. However, while the strong

COVID response of European governments was largely endorsed by the population, the threat of the

pandemic was also tangible and imminent; making the policies more acceptable. In dealing with

climate change, the threat is physically and geographically diffuse, as well as distributed across

time; which makes the necessary transformations socially and politically challenging to implement.

For this reason, there is an important amount of work required to ensure the social acceptability

of policies needed for the transition. This implies changing the collective imaginary and the social

norms on which our consumption and production systems thrive.

Environmental economics explains environmental issues through the existence of externalities

that are not accounted for in the markets (Pigou (1920); Coase (1960)). These externalities arise

due to the underlying uncertainty and hidden costs. On top of this, the difficulty in quantifying en-

vironmental impact creates asymmetric information and hence limited third-party verifiability. The

10
literature on the management of the commons documents alternative mechanisms that could con-

tribute to protecting the environment and reducing those externalities: Elinor Ostrom showed how

social norms, as an equilibrium in coordination games, play a key role in the collective management

of resources (Ostrom (1990); Ostrom (2000)). Social norms act as informal institutions to overcome

“market imperfections”, and are particularly relevant in fostering pro-environmental behaviour (Ny-

borg (2018)). However, norms can also underpin conservatism or obstruct pro-environmental be-

haviour. Biel and Thøgersen (2007) find that social norms can be identified as a reason for the

departure from the rational choice in social dilemmas and that the ”norm activation” mechanisms

are based on the interactions and common beliefs of individuals in a group.

Recent evidence from Andre et al. (2021) shows that there is a strong correlation between

perceived norms and individual willingness to fight climate change. They elicite norms perception

from survey questions and measure the willingness to fight climate change through offering the

possibility to donate to an NGO. They find that individuals tend to underestimate the norms -

i.e. the average collective behaviour - regarding pro-environmental action; and that correcting

those beliefs increases their willingness to donate. These results therefore highlight the importance

of conditional cooperation as a driver of environmental behaviour. It also further confirms the

importance of understanding the drivers of social norms and the conditions under which they can

change.

Correcting beliefs about sustainable actions can also increase individual willingness to fight

climate change, as measured by Pace and van der Weele (2020) in a randomised survey study.

This may be due to a change in the perception of the ”right thing to do” or the ”commonly

accepted behaviour”, which could suggest a change in the perception of the social norms. Our

study investigates how both the perception an enforcement of social norms changes with information

about climate change.

2.3 Social Norms

In economics, social norms are defined as “social elements that are jointly perceived and commonly

recognised by members of a social group” (Elster (1989)). Elinor Ostrom describes social norms

11
as a “shared understanding about actions that are obligatory, permitted or forbidden” (Ostrom

(2000)). Social norms thus rely not only on an agreement upon a set of behaviours and actions that

are commonly accepted but also on a set of punishment and rewards for deviance from or adherence

to the norms.

Benabou and Tirole (2011) look at how private and public policies are shaped by personal and

societal preferences (values), explicit incentives (laws) and social sanctions and rewards (norms).

They build a model that incorporates a social norms factor, hence enabling norm-based policy

simulations that alter people’s perception of the norms. Building on this framework, Bursztyn et al.

(2020) use election outcomes as information about social norms to analyse how they might influence

individual behaviour. In particular, they demonstrate that social norms can change rapidly with

new information. An additional study by Xie et al. (2011) evaluates the tipping point of a social

norm becoming influential when around 10% of the population changes behaviour.

The literature typically distinguishes in between two types of norms: the descriptive norms,

that reflect the collective adherence to a given behaviour, and the injunctive or prescriptive norms

that capture the perception of how acceptable a given behaviour is. It is an important distinction as

people use information both about what they think others are doing (descriptive norms) and about

what they think others should do (injunctive norms). Our study measures the perception and

the judgement towards others’ behaviour in order to evaluate both types of norms. As mentioned

previously, Andre et al. (2021) find that correcting beliefs about both descriptive and prescriptive

norms increases the individual willingness to fight climate change. Other empirical settings have

found that making both type of norms more prominent can change energy consumption (Schultz

et al. (2007); Allcott and Rogers (2014)) or mobility choices (Raux et al. (2015); Mundaca et al.

(2022)). This research demonstrates that there exists a large scope for social norms to be used as

a lever to foster long-lasting collective change.

3 Model

We want to model dynamic decision-making, that accounts for the accumulation of externalities

over time. Over the course of three periods, agents decide on a level of action a ∈ {0, 1}, from

12
which they obtain a benefit γ > 0 (a = 1 being the harmful behaviour). We assume agents want

to maximise their inter-temporal utility. They choose whether or not to perform the action in

every period t ∈ {1, 2, 3}. Agents can be of different types θi ∈ [θ; θ̄] = Θ that are attributed by

nature. The individual types are only observed by the agents but the distribution of the types in

the population is known (Fθ ). Higher types θi are more considered more ”pro-social”. The action

a = 1 creates an externality e that increases in the aggregate level of actions taken in the previous

periods. An agents’ contribution is captured by the accumulation function of personal emissions:

f (Ait ).

To illustrate this situation, let action a be a high-emitting leisure, such as taking the plane.

Taking the plane allows the individual to travel, which has a direct benefit on their utility γ > 0

and carries an externality through greenhouse gas emissions that come from burning jet fuel, e > 0.

The externality materialises in subsequent periods as a function of the cumulative average decisions

a¯t :
3 n
!
X 1X i
a¯t = at
n
t=1 i=1

We consider that the marginal contribution of an agent to the flights’ overall emissions is negligible,

and hence do not account for it in the maximisation of private utility and take the cumulative

emissions as given at the start of each period. Moreover, depending on their type θi , individuals

are more or less morally affected from having contributed to the externality in the previous periods.

Individuals that are more concerned with the environment or their impact on society (i.e. higher

θi ) are therefore more affected in their utility. We call higher types ”altruistic” and lower types

”selfish”.

3.1 Model Setup

The utility of the agents depends on their type θi and the action chosen ai . We can write:

ui1 = ai1 γ − ea¯1 − θi f (Ai1 )

ui2 = ai2 γ − ea¯2 − θi f (Ai2 )

ui3 = ai3 γ − ea¯3 − θi f (Ai3 )

13
Agents use hyperbolic discounting with β < 1 (i.e. they are present biased). We can write the

inter-temporal utility of the agents in the first period:

X
ui1 + β δ t uit = ui1 + βδui2 + βδ 2 ui3 (1)
t>1

We further define the individual accumulation function:

!2
X
f (Ait ) = I{at−1 < at } + ait−1 (2)
t=1

Where I{at−1 < at } is a cost for inconsistency across periods. That is, agents face a cost when they

choose to pollute in a given period after having abstained from polluting in the previous period.

3.2 Model Solutions

In order to evaluate the aggregate level of emissions in the society for each period, we need to

determine the actions taken by individuals for each type θi in periods 1, 2 and 3.

Proposition 1: There exists values of θ ∈ Θ, labelled θt∗ , defining decision rules for each type

θi in any period t ∈ {1, 2, 3}. Theses values and associated decision rules are illustrated in Figure

1.

Figure 1: Decision Rules

Proof: We define a cutoff θ∗ as the individual type for which the individual is indifferent in

between choosing ait = 1 and ait = 0 in period t. Since the agents maximise their inter-temporal

utility, as defined in equation (1), we use backwards induction and solve first for period 3.4 Intu-

itively, since this is the last period, individuals do not face the moral cost of creating the externality

4
For computational and methodology details, see Appendix Section 12

14
anymore and should rationally choose to pollute in all cases. However, there may be some altruistic

individuals with higher types θi that chose not to pollute in the previous periods and would face a

cost for inconsistency in the last period. The choice of action in the last period therefore depends

on the type of the individuals as well as on the choices made in the previous periods. We find two

cutoff points for the third period under which the individual chooses ai3 = 1 :

γ
θ3∗ (a1 = 1, a2 = 0) = and θ3∗∗ (a1 = a2 = 0) = γ
3

These values are conditional on the actions in previous periods, due to the interdependence in

between the periods. Using the decision rules for the third period, we compute the threshold values

for the second and first periods:

γ(1 + βδ) γ
θ2∗ = and θ1∗ =
3βδ βδ(1 + δ)

From this, we can illustrate the decision rules distribution for all the types θi ∈ Θ, under the

condition that βδ(1 + δ) > 1, as follows:5

1. If θi < θ2∗ , then ai1 = ai2 = ai3 = 1

2. If θ2∗ < θi < γ then if ai1 = 1, ai2 = ai3 = 0 and if ai1 = 0, ai2 = ai3 = 1.

3. If γ < θi < θ1∗ , then ai1 = 1 and ai2 = ai3 = 0

4. If θi > θ1∗ , then ai1 = ai2 = ai3 = 0

Figure 2: Optimal Decision Rules across types θ ∈ Θ

5
Note: in the interval θ2∗ < θi < γ, we know that the agent chooses ai1 = 1, hence the intertemporal choices in
that interval are always ai2 = ai3 = 0

15
3.3 Present Bias Effect

Following O’Donoghue and Rabin (1999), we consider that some individuals are naive about their

present bias, as opposed to sophisticates that are aware of their bias. The naive individuals may

therefore follow different decision rules.

Proposition 2: Under specific assumptions on β and δ, the present-bias increases the level of

aggregate emissions, due to naive individuals following biased decision rules.

Proof: We use the decision rules obtained in the previous section, to determine rules used by

the naive individuals. We know that naiveté about the bias means that they do not take the β

discounting parameter into account when making decisions. We can therefore see that the cutoff

values for θ2∗ and θ1∗ for naive individuals will be:

γ(1 + δ) γ
θ2N = and θ1N
3δ δ(1 + δ)

We notice that:
γ(1 + δ) γ(1 + βδ) γ γ
< and <
3δ 3βδ δ(1 + δ) βδ(1 + δ)

Hence, naive individuals base their decisions on cutoff values that are lower than the true cutoffs:

they consider themselves more altruistic than they actually are.

Furthermore, under certain conditions for δ,6 we have θ1N < γ. Naive individuals for which
h h
γ
θi ∈ δ(1+δ) , γ therefore believe they are in the section to the right of θ3∗ and choose a1 = 0.

However, once in period 2, they realise they are actually in the [θ2∗ , θ3∗ ] decision rule: they end up

choosing ai2 = ai3 = 1, as per the decision rules from Figure 1. The present-bias therefore adds to the

number of individuals polluting the second and third period, which increases aggregate emissions

over the three periods of the model. This is illustrated in Figure 3:

6
If δ(1 + δ) > 1, which is true for realistic values of δ found in the literature (i.e. δ0.8

16
 Figure 3: Decision Rules under Present Bias

Assuming the types are distributed according to Fθ (·), we can compute the subset of potential

additional polluters:
 
γ
Fθ (γ) − Fθ
δ(1 + δ)
γ
where Fθ is the distribution of θ. Assuming that θi is uniformly distributed on (θ, θ̄), with θ < 3

and θ̄ > γ, we can compute the additional emissions g N , with x, the proportion of naive individuals

in the population:
 
N 1 γ
g = γ− ×x
θ̄ − θ δ(1 + δ)

We can see that the amount of additional polluters as an effect of present-biased time preferences

increases in the value of the benefit from the action: the higher the present-value, the more indi-

viduals are present-biased. We can see that as the gap in between extreme types θ and θ̄ increases,

the additional emissions decreases. This is due to the fact that the benefit of the action doesn’t

change depending on the number of individuals taking it, which may be a limitation of the model.

The naive cutoff θ2N does not imply a change in the decision-rule for period 1. However, there

is a range:
 
γ(1 + δ) γ(1 + βδ)
θi ∈ ,
3δ 3βδ

in which naive individuals follow the second decision rule, despite their θi really being in the first.

They choose action a1 = 1 in period 1, thinking they would choose a2 = a3 = 0 in subsequent

periods. Yet, once they arrive in period 2, they realise that θi < θ2∗ (i.e. they are not altruistic

enough) and choose a2 = a3 = 1. This, however, does not change the aggregate level of emissions

over the three periods, because the naive individuals do not change the decision rules (i.e. they end

up following the same decision rule as sophisticates of the same type). However, the present-bias

17
would be costly for naive individuals if the actions in period 2 and 3 were conditioned to some

commitment made in period 1. They would’ve failed to judge their decisions for the following

periods appropriately due to the present-bias; which is a form of procrastination (for discussion, see

O’Donoghue and Rabin (1999)).

3.4 Social norms

As suggested in O’Donoghue and Rabin (1999), some commitment mechanisms can help address the

present-bias for sophisticated individuals that are aware of their biases. In this study, we want to

investigate the role of informal institutions, such as social norms in contributing to fostering more

sustainable behaviour collectively. The hypothesis that stronger social norms could reduce the effect

of cognitive biases such as the present bias seems plausible for several reasons: first, they decrease

the value of choosing a emitting action, as others would judge that behaviour negatively, which

imposes a cost to the agent. Social norms could also, to some extent, act as a commitment over

several periods to maintain a given behaviour, hence directly countering the effect of the present-

bias. Based on the model in Benabou and Tirole (2011), we investigate the effect of adding social

norms in the utility. In this study, for simplicity, we assume social norms only enter the individual’s

decision in the first period, thereby only investigating the first hypothesis.

Social norms enter the utility as follows:

ui1 = ai1 γ − ea¯1 + µEj̸=i [θi |ai1 ]

ui2 = ai2 γ − ea¯2 − θi f (Ai2 )

ui3 = ai3 γ − ea¯3 − θi f (Ai3 )

The last term, µEj̸=i [θi |ai1 ] is the expectation of individual i’s type, θi , from the point of view

of other individuals in the population. They based their expectation of the type upon observing

individual i’s action in the first period ai1 . µ is the weight that individuals put on the others’

inference of their type.

18
 Following Benabou and Tirole (2011), we define the following moment conditions:

M+ (θ) = E[θ̃|θ̃ > θ] = E[θ̃|aθ = 0]

M− (θ) = E[θ̃|θ̃ < θ] = E[θ̃|aθ = 1]

Where M+ captures the expectation of θi given the virtuous behaviour is observed and conversely

for M− . We define the difference in those expectations:

∆(θ) = M+ (θ) − M− (θ)

We can see that ∆(θ) increases if the virtuous behaviour becomes rare (M+ increases) or if the less

virtuous behaviour is marginalised (M− decreases). That is, if only very high θi individuals choose

ai1 = 0 or if only individuals with very low θi will choose ai1 = 1. We notice that these two effects

are opposed, and the magnitude of the change in ∆(θ) will depend on which effect dominates. A

change towards more pro-environmental norms would for example decrease ∆(θ) if it is the effect

of M+ that dominates.7

Proposition 3: There exists values θ11∗ , θ12∗ and θ13∗ that are implicitly defined by ∆(θ11∗ , ∆(θ12∗ )

and ∆(θ13∗ ). These values are the cutoffs that determine the decision rules for each type θi ∈ Θ in

period 1.

Proposition 4: Stronger pro-environmental norms decrease the probability of an individual

with type θ of choosing ai1 = 1.

Proofs: We are solving for the decision rule in period 1, based on the decision rules computed in

section 3.2. We know that, given γ, e and the forecast decisions for the following periods, individuals

choose ai1 = 1 if θi ≤ θ∗ , thereby defining a cutoff rule that is a function of ∆(θ∗ ). The individual

solves:

γ + µE[θi |ai1 = 1] + βδ(γai2 − θi ) + βδ 2 (γ − θi (1 + ai2 + I{ai2 < ai3 })2 ) ≥

µE[θi |ai1 = 0] + βδ(γai2 − θi (I{ai1 < ai2 })2 ) + βδ 2 (γ − θi (ai2 + I{ai2 < ai3 })2 )

7
This is the idea of stigma vs honor discussed in Benabou and Tirole (2011).

19
Using the decision zones defined by θ2∗ and θ3∗ from Figure 1, we evaluate the actions chosen in

period 1 for individual i.

1. For θi < θ2∗ , we know that ai2 = ai3 = 1 in all cases. From this, we find that the individual

will choose ai1 = 1 if and only if:

γ + µ E[θi |ai1 = 1] − E[θi |ai1 = 0]

θi ≤
3βδ 2

Thereby defining a cutoff value that satisfies the following equation:

γ − µ∆(θ11∗ )
θ11∗ =
3βδ 2

2. If θ2∗ < θi < γ, an implicit cutoff is defined, such that the individual will choose ai1 = 1 if:

γ(1 − βδ(1 + δ)) ≥ µ∆(θ12∗ )

That is, if θ12∗ is such that the above inequality holds, individual i will choose ai1 = 1 in all cases. We

note that the LHS of this inequality represents the value of choosing ai1 = 1 minus the discounted

values of choosing ai2 = ai3 = 1. If this value is larger than the benefit from being perceived as

altruistic, then individuals will choose ai1 = 1.

3. Finally, if θi ≥ γ, we know that ai2 = ai3 = 0 in all cases. We find a cutoff value θ13∗ , satis-

fying the following equation:

γ − µ∆(θ13∗ )
θ13∗ =
βδ(1 + βδ)

Stronger pro-environmental norms means that more individuals choose ai1 = 0, which reduces

the values of the θ∗ cutoffs in all three cases and reduces aggregate emissions in the first period.

20
3.4.1 Present Bias Effect

Proposition 5: The effect of the present-bias can be countered by the effect of social norms, if the

norms are pro-environmental.

Proof: Following a similar reasoning to section 3.3, we can quantify the present-bias effect on

the aggregate amount of emissions:

γ − µ∆(θ13∗ )
 
N 1
h = γ− ×x
θ̄ − θ δ(1 + δ)

We can see that the level of additional emissions is increasing in ∆(·) and decreasing in µ. This

equation illustrates the ambiguity in social norms change: as discussed previously, ∆(·) increases as

a result of either behaviour (altruistic or selfish) becoming marginal. However, if altruism (a = 0)

becomes the norm, people expect lower values of θ to take action a = 0, hence decreasing the value

of M+ and decreasing ∆(·) which reduces additional emissions. Similarly, if most individuals choose

to pollute (a = 1), the value of M− increases, which decreases ∆(·), and reduces the additional

amount of emissions. In both cases, ∆(·) change the value of the action depending on the number

of people choosing it, which captures the effect of the social norms and the desire for conformity

and improved social image.

There are therefore several ways that social norms can influence the level of aggregate emissions,

and in particular the additional emissions introduced by the present bias. More importantly, we see

that, as the norms change, making pro-environmental behaviour more prominent, the present value

of the action decreases which reduces the number of naive individuals that change decision rules as

a result. Second, through increasing the value that individuals put on social image, we reduce the

value of choosing the polluting action as individuals know they will be ”assigned” to lower values

of θ which reduces their utility.

21
3.5 Summary

This model illustrates a few key insights from the literature. In section 3.3 we found that the

present-bias increases the number of individuals emitting in the second and third period, hence

increasing aggregate emissions. Section 3.4.1 showed that pro-environmental norms could reduce

that effect, through reducing the value of the present-day action and hence reducing the ”gap”

for naive individuals in between the real and biased values of the decision rules. Finally, the

more individuals care about the others’ perception of their type and care about being perceived

as altruistic, the less likely they are to choose the emitting action, which captures the desire for

conformity and beneficial social image (Cialdini (2005)).

4 Experimental Protocol

The experiment was conducted with an online platform, Respondi,8 which provided the panel and

hosted the survey. We interviewed 800 individuals from California, selected to be a representative

sample of the state’s population in terms of gender (50% male - 50% female) and age. The age

distribution was chosen based on the 2020 Census Data9 .

The first round started on the 11th of April, 2022 and we obtained the 800 responses by the

20th of April. The second round started on the 25th of April, approximately 1 week after the first

round. Participants were recontacted and asked about their willingness to donate their earnings

from the experiment to an environmental NGO, Friends of the Earth.

4.1 First Round

We collected pre-treatment socio-economic and geographical data from the participants, as well as

variables indicating time and risk preferences. The participants were then randomly allocated to

the treatment and control group. The treated participants were asked to read through 7 slides10

8
[Link]
9
[Link] 13% 18 to 24 years old; 20% 25 to 34 y.o.; 17% 35 to 44 y.o.; 17% 45 to
54 y.o.; 16% 55 to 64 y.o.; 19% 65+ y.o.
10
See slides in the Appendix - Section 12. We added a timestamp on the slides, to ensure that participants spent
at least 20 to 30 seconds on each slide.

22
with information about the causes and consequences of climate change, with a focus on the impacts

in California. The treatment also included information about the main activities contributing

to climate change worldwide (based on GHG11 emissions). The control group did not see any

information. Both groups were then presented two hypothetical choice situations with more or less

CO2-intensive options. In each situation, we asked the participants to state three opinions which

constitute the main outcomes: their personal choice (labelled decision), their belief about the

average answer among the panel (labelled norm); and how socially appropriate they judged the

most emission-intensive option when chosen by someone else (labelled judgement).

The first question allowed us to measure the changes in the individual behaviour, while the

second and third question captured the social norms. Specifically, the judgement question captures

the way the individual enforces the norm and the third question captures the perception of the

norm. This is an application of the Krupka and Weber (2008) method of norm elicitation, in a

non-incentivised version.12

In the first situation the participants chose in between travelling by plane for a weekend to

visit friends or renouncing to travel. We provided information about the monetary cost and CO2

emissions of the trip.13 In the second situation, we asked the participants to choose an energy

subscription among two plans from the Clean Power Alliance, a Californian electricity distributor.14

The first option was a subscription for electricity from 100% renewable energy sources and the

other a subscription with electricity generated from a mix of 60% from coal and gas and 40% from

renewables. We provided information about the financial cost and CO2-emission per kWh for each

contract.

At the end of the survey, we asked the participants to report how exposed they had been to

the 2020 and 2021 fires in California. As the answers were collected post-treatment and may have

been influenced by it, we treat these as outcome variables in the first round.

11
Greenhouse Gases, of which CO2 is the main contributor to global warming IPCC (2021)
12
Research on experimental elicitation reports that, in the case of norm elicitation, non-incentivised experimental
results do not substantially differ from incentivised experiment’s results (see Vesely (2015)).
13
See full text in the Appendix - Section 12
14
[Link]

23
4.2 Second Round

In this round, participants answered questions about their beliefs regarding the likelihood of the 2020

and 2021 fires in California being related to climate change.15 Beliefs can be an important driver of

decisions, in particular in the context of environmental behaviour (see e.g., Pace and van der Weele

(2020)). Additionally, we surveyed the participants about their level of concern regarding climate

change in California and globally. We also asked which party they voted for in the latest election

for the state governor, in 2018. 16

Each participant was then informed about the number of credits17 they would obtain from

participating in the experiment (136 credits). We asked them whether or not they would be willing

to donate the amount of credit earned in the survey (from both rounds) to an environmental NGO,

Friends of the Earth.18 At the start of the second round, we randomly allocated the participants in

one of two categories in between observed and unobserved. The ”observed” group was told, before

they made their decision to donate, that their action was going to be showed to other participants,

who would provide a judgement on the decision (without any consequences for the donor). Finally,

we asked each group whether they believed that other participants in the panel donated or not, on

average. These two answers constitute the last two main outcomes of the survey (i.e. the donation

decision and the donation norm). Aims for this round were threefold. First, the delay after the

first round allows us to measure the persistence in time of the treatment. Second, the donation

situation was a way to materialise the action taken, with respect to the hypothetical situations of the

first round. It is more likely that the respondents reveal their true preferences in a concrete choice.

However, we couldn’t implement a direct donation through the Respondi website. Instead, the

participants were provided with a link to the donation page of the NGO and reminded about their

commitment at the end of the experiment. Finally, the observation treatment creates an influential

effect of ”enforcement” of social norms on individual decisions (see e.g., Sutter et al. (2009)). We can

therefore evaluate how the prominence of norms, materialised through the observation, influences

behaviour.
15
Note: this was explicitly stated as a fact in the treatment
16
The elected governor was Gavin Newsom, from the Democrat party.
17
These credits are provided by the company in charge of providing the panel (Respondi’s partner in the United-
States). Once the participants have accumulated enough credits, they can convert them into cash or into vouchers.
18
[Link]

24
4.3 Treatment

The treatment is based on the latest IPCC report,19 providing a summary of the facts stated across

the three parts of the report (see Appendix for more details). The text was written and designed

using insights from climate change communication science. The main challenges in communicating

about climate change are the uncertainty, the complexity and the temporality of the issue. The

objective of climate change communication is to strike a balance in between exposing the scientific

facts and navigating the political debate around the policy instruments used to respond to the issue.

The IPCC report was chosen as a reference for the treatment in order to also convey a message

about the scientific consensus around climate change, which has been shown to have a significant

effect in shifting beliefs and perceptions about climate change, across all political affiliations (van

der Linden (2015)). Regarding the communication of plain ”facts”, some studies find that it may

increase political attitudes and polarisation of the debate (Kahan et al. (2012)) while others find no

evidence of such polarisation (Cook et al. (2014); Guess and Coppock (2015)). For this experiment,

we choose to state the facts about “physical” changes happening in the climate system, which we

related to direct impacts, such as potential changes in weather events frequency and intensity, in

order to make the issue more tangible to the public (Akerlof et al. (2013)). There is a tension in

communicating information about impacts and risks; as it may create a feeling of powerlessness

and hence decrease the chances of shifting beliefs or expressing a willingness to invest resources

into political action (Levine and Kline (2017)). This is why we also add some information about

potential solutions put forward by the scientists to can reduce the pace of climate deregulation.

5 Descriptive Statistics and Balance Checks

The survey data contains many socio-economic and demographic variables which we use to analyse

the treatment effect of the experiment, as they may also influence the outcomes. Before providing

descriptive statistics on the panel profile, we compute two variables that we use for the analysis:

the distance to the fires and the time bias.

19
Sixth Assessment Report, IPCC, 2021-2022

25
5.1 Computing the distance to fires

From the zip codes provided by the participants in the survey, we are able to compute, for each

respondent, the distance to the fires that occurred in California in 2020 and 2021, using QGIS.20 To

map the location of the 2020 and 2021 fire events in California, we used data from the Californian

Fire Agency, CalFire21 and the Wildland Fire Locations Full History dataset compiled by the

Wild Fire Interagency Geospatial Service (WFIGS - a service of the National Interagency Fire

Centre22 ). The dataset contains the 22,901 wildfire23 events reported for the years 2020 and 2021

in California (this does not include the prescribed fires, supposedly contained and unlikely to affect

local residents). For each wildfire event, we observe the initial latitudes and longitudes of the

fire. That is, ”the latitude/longitude location of the initial reported point of origin specified in

decimal degrees”.24 As the 22,901 observations contain many small fires that are unlikely to have

had a tangible impact, we filter the wildfires by size, using the information provided by CalFire

Redbooks25 for each year, which provide a list of fires that are larger than 300 acres (which is

equivalent to 121 hectares - that is, roughly 172 football fields). Using the unique fire identifiers, we

can match those fires from CalFire to the WFIGS database, and hence compile a list of 169 wildfire

events, with corresponding coordinates, for wildfires larger than 300 acres over 2020 and 2021.

Using this information, we locate the wildfires in QGIS and add the zip codes of the survey

participants; for which we recover the coordinates from the Census Bureau.26 The ”Distance to

Nearest Hub” package from the Vector Analysis tools then allows us to compute the distance, in

kilometers, from each zip code to the nearest wildfire event. Figure 4 below is a map of the fires (in

red), the zip codes reported by the participants (in blue) and the distances (in yellow).

20
[Link]
21
[Link]
22
[Link]
23
A wildfire is defined by the National Wildfire Coordinating Group as a ”wildland fire originating from an
unplanned ignition, such as lightning, volcanos, unauthorized and accidental human caused fires, and prescribed fires
that are declared wildfires”. Source: 2009 Guidance for Implementation Federal Wildland Fire Management Policy
and Fire Management Board Memorandum 19-004a
24
[Link]
about
25
[Link]
26
[Link]

26
 Figure 4: Participant and Fire Location Map (2020 & 2021 fires > 300 acres)

We obtain the following measures:

Table 5.1: Descriptive Statistics - Distance to the fires

Mean SD Min Max Count

Distance to nearest fire (km) 33.2 24.9 1.32 132.6 796
Log of distance 3.19 0.85 0.28 4.89 796
Fire less than 10 km away 0.15 0.36 0 1 796
Fire less than 15 km away 0.28 0.45 0 1 796
Fire less than 25 km away 0.47 0.50 0 1 796

5.2 Computing the time bias

As we saw in the model, time preferences are an important factor in explaining pro-environmental

behaviour. Behavioural scientists and economists do not provide a clear consensus on the best

elicitation techniques to measure time preference (Frederick et al. (2002)). While the main technique

is the Multiple Price List (MPL) approach, it is heavily criticised for yielding very high estimates

of the discount rates which is inconsistent with models of aggregate discounting. Andreoni and

27
Sprenger (2012) suggest that the bias in the discount rate may be due to the underlying assumption

on the linearity of the utility function, which would bias the discount rate upwards if the utility

is actually concave. In order to reconcile the intertemporal optimisation and the discontinuity in

budget which constrains the choices in the MPL approach, they propose a Convex Time Budget

approach to identify the time preference parameters. They measure significantly lower discount

rates with respect to those measured using MPL. Since we do not need to estimate precisely the

discount factors, but rather measure the existence of a potential bias and how it is distributed among

the participants, we use a non-incentivised Multiple Price List approach to elicit time preferences,

following Andersen et al. (2009).27 The participants were presented with this table and asked to

state their preference in between the two payment options for each decision row:

Figure 5: Multiple Price List Table - Round 1

That is, for each row, we recover an observation indicating whether they prefer the earlier

(option A) or the later (option B) payment. The answers given by the respondents therefore

provide indications on the discount rate they apply to future payment occurrences. The lower the

discount rate, the more the individual values the present-day payments relative to future payments.

In general, we assume that individuals have similar discount rates across the population. Following

the definition from O’Donoghue and Rabin (1999), an individual is present-biased when they give

stronger relative weight to a moment as it gets closer. If we assume that the participants in the

panel have the same discount rates, any differences in the answers to the MPL could be attributed to

some form of present-bias. Based on these assumptions, we consider the point at which individuals

switch to preferring later payments as a proxy for their present bias.

27
See Brañas-Garza et al. (2021) for discussion on the non-incentivised version of the MPL.

28
 Before computing the level of the present bias for each participant, we use the MPL answers

to evaluate the coherence of the answers provided by the participants: if they switch back to the

sooner payment (option A) after choosing the later payment (option B) in one row, we consider

that they report incoherent preferences. We find that there are 7.25% of incoherent responses in the

panel (58 observations), which are dropped for further analysis. Among the coherent respondents,

65.5% are unbiased: they choose the later payment in the first alternative and throughout the price

list. For the biased individuals, the median switch point (i.e. the point at which they indicate a

preference for the later payment) is the third alternative. Figure 6 below provides the distribution

among biased individuals, each bar representing the decision row at which they switch:

Figure 6: Distribution of the present-bias among biased individuals in the panel

To account for the present-bias in the analysis, I define a Present Bias variable measuring the

extent to which the participants are biased as an index of the switch points (unbiased = 0, switch

1 = 1, switch 2 = 2, ... etc).

5.3 Descriptives

Table 5.2 provides descriptive statistics on the demographic and socio-economic profile of the panel

for each round. In total, we obtained 800 responses in the first round, from which we removed

58 observations that were providing incoherent responses and 4 observations that did not answer

all the questions. This brings the number of observations to 738 for the first round. Of the 800

29
 respondents of the first round, 698 (87,25%) accepted to be recontacted for the second round of the

study. Out of those, 478 responded to the second survey. Overall, we have a recall rate of 59.75% of

participants that answered both surveys. Additionally, of the 478 participants in the second round,

235 (49.16%) were from the control group and 243 from the treatment group (50.84%). Removing

individuals that provided incoherent answers, we end up with 467 observations for the second round

(63.28% final recall rate). The descriptive statistics for each round are provided in the table below:

Table 5.2: Descriptive Statistics

Round 1 Round 2
Mean SD Min Max Count Mean SD Min Max Count
Age 46.9 16.9 18 93 738 50.3 15.3 19 83 467
Female 0.52 0.50 0 1 738 0.53 0.50 0 1 467
Male 0.48 0.50 0 1 738 0.47 0.50 0 1 467
Level of Education (1 to 6) 3.50 1.15 1 6 738 3.68 1.07 1 6 467
Student 0.065 0.25 0 1 738 0.036 0.19 0 1 467
Employed full-time 0.46 0.50 0 1 738 0.48 0.50 0 1 467
Employed part-time 0.12 0.32 0 1 738 0.12 0.33 0 1 467
Unemployed 0.17 0.37 0 1 738 0.15 0.36 0 1 467
Retiree 0.20 0.40 0 1 738 0.21 0.41 0 1 467
Household (1 to 6) 2.66 1.37 1 6 738 2.57 1.31 1 6 467
Distance to nearest fire (km) 33.4 25.0 1.32 132.6 738 33.1 24.6 1.32 124.4 467
Risk Preferences (1 to 10) 5.36 2.30 1 10 738 5.34 2.21 1 10 467
Present Bias 0.80 1.47 0 5 738 0.70 1.43 0 5 467
Fire emotional exposure (1 to 5) 3.03 1.14 1 5 467
Fire physical exposure (1 to 5) 2.09 1.12 1 5 467

The panel is rather educated, with a majority of full-time workers. We group ”full-time”

and ”part-time” workers as ”employed” individuals to simplify the analysis. We note that the

participants report a small level of present bias and are mostly risk-neutral. These statistics are

mostly preserved in the second round, except for students that did not respond to the callback as

much as other categories. This should not significantly change the results as the students represent

6% and 3% of the samples.

Table 5.3 presents the descriptive statistics for variables that are specific to the round 2. We

find no evidence that treated participants are more likely to participate in the second round relative

to non-treated participants. Similarly, there is no indication that more environmentally-friendly

30
individuals (i.e. those that did not travel and chose the clean mix) self-select into the second round.

Table 5.3: Descriptive Statistics - Round 2

Mean SD Min Max Count

Democrat Vote 0.52 0.50 0 1 467
Republican Vote 0.24 0.43 0 1 467
Other Vote 0.24 0.42 0 1 467
Treatment 0.51 0.50 0 1 467
Travel decision 0.74 0.44 0 1 433
Energy decision 0.22 0.42 0 1 467
Click to get further information about FoE 0.0064 0.080 0 1 467

5.4 Balance Checks

In both rounds, the participants were split in between two groups: in the first round, we have a

treatment and a control group; which are split in between observed and unobserved in the second

round. The following tables provide balance checks across the different sub-groups. Columns (1) to

(3) in Table 5.4 describe the characteristics of the participants in the whole panel (column (1)) and

by treatment group (columns (2) & (3)). Column (4) reports the estimates and p-values for the test

of equality in between the means of the treatment and control groups for each variable. We can see

that there are no significant differences in between the treatment and control groups in the panel

across the variables, which demonstrates the effectiveness of the randomisation into treatment and

control groups. For the second round (Tables 5.5 and 5.6), we check that the 4 groups are different

from the reference group: participants in the control group that were not observed (column (4)). We

can see that, although most variables are not significantly different across the groups, the ”control

& unobserved” group has less participants that are retirees than the other groups. However, this

should not significantly affect the results.

31
 Table 5.4: Balance Checks - Round 1

(1) (2) (3) (4)

Full sample Treatment Control Difference (3)-(2)
mean sd mean sd mean sd b p
Age 46.86 16.94 47.18 17.13 46.55 16.77 -0.63 (0.61)
Female 0.52 0.50 0.52 0.50 0.52 0.50 0.00 (0.95)
Male 0.48 0.50 0.48 0.50 0.48 0.50 -0.00 (0.95)
Level of Education (1 to 6) 3.50 1.15 3.51 1.15 3.49 1.14 -0.02 (0.77)
Student 0.07 0.25 0.06 0.24 0.07 0.25 0.00 (0.83)
Employed full-time 0.46 0.50 0.46 0.50 0.45 0.50 -0.01 (0.84)
Employed part-time 0.12 0.32 0.12 0.33 0.11 0.32 -0.01 (0.74)
Unemployed 0.17 0.37 0.15 0.36 0.18 0.39 0.03 (0.21)
Retiree 0.20 0.40 0.21 0.41 0.18 0.39 -0.02 (0.43)
Household (1 to 6) 2.66 1.37 2.64 1.35 2.68 1.40 0.05 (0.65)
Distance to nearest fire (km) 33.38 24.98 34.06 25.42 32.71 24.56 -1.34 (0.47)
Risk Preferences (1 to 10) 5.36 2.30 5.36 2.38 5.37 2.23 0.01 (0.94)
Present Bias 0.80 1.47 0.83 1.53 0.77 1.41 -0.06 (0.59)
Democrat Vote 0.52 0.50 0.53 0.50 0.52 0.50 -0.01 (0.90)
Republican Vote 0.24 0.43 0.24 0.43 0.24 0.43 -0.00 (0.97)
Other Vote 0.24 0.42 0.23 0.42 0.24 0.43 0.01 (0.86)
Observations 738 365 373 738

32
 Table 5.5: Balance Checks - Round 2 - Panel A : Descriptives by group

(1) (2) (3) (4)

Treat & Obs Control & Obs Treat & Unobs Control & Unobs
mean sd mean sd mean sd mean sd
Age 50.24 15.93 51.41 14.98 51.00 15.29 48.56 14.89
Female 0.50 0.50 0.57 0.50 0.53 0.50 0.53 0.50
Male 0.50 0.50 0.43 0.50 0.47 0.50 0.47 0.50
Level of Education (1 to 6) 3.75 1.04 3.53 1.05 3.66 1.09 3.76 1.10
Student 0.05 0.22 0.02 0.13 0.03 0.16 0.05 0.22
Employed full-time 0.50 0.50 0.41 0.49 0.47 0.50 0.53 0.50
Employed part-time 0.14 0.35 0.12 0.32 0.10 0.31 0.13 0.33
Unemployed 0.08 0.28 0.20 0.40 0.16 0.37 0.15 0.36
Retiree 0.23 0.42 0.25 0.44 0.23 0.42 0.14 0.35
Household (1 to 6) 2.64 1.28 2.46 1.20 2.53 1.31 2.63 1.44
Distance to nearest fire (km) 32.56 24.40 31.97 22.79 32.96 25.17 34.98 26.01
Risk Preferences (1 to 10) 5.22 2.26 5.40 2.08 5.38 2.32 5.36 2.17
Present Bias 0.73 1.51 0.78 1.40 0.65 1.41 0.66 1.40
Democrat Vote 0.54 0.50 0.48 0.50 0.52 0.50 0.56 0.50
Republican Vote 0.31 0.47 0.28 0.45 0.16 0.37 0.20 0.40
Other Vote 0.15 0.36 0.24 0.43 0.32 0.47 0.24 0.43
Treatment 1.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00
Travel decision 0.66 0.48 0.80 0.40 0.69 0.46 0.83 0.38
Energy decision 0.16 0.37 0.28 0.45 0.26 0.44 0.21 0.41
Observations 121 111 116 119

Table 5.6: Balance Checks - Round 2 - Panel B : T-Tests across groups

Difference (4)-(1) Difference (4)-(2) Difference (4)-(3)

b p b p b p
Age -1.677 0.401 2.851 0.149 -2.437 0.217
Female 0.034 0.605 0.038 0.563 0.004 0.957
Male -0.034 0.605 -0.038 0.563 -0.004 0.957
Level of Education (1 to 6) 0.004 0.975 -0.225 0.115 0.101 0.479
Student 0.001 0.976 -0.032 0.182 0.025 0.329
Employed full-time 0.034 0.605 -0.115 0.082 0.055 0.399
Employed part-time -0.014 0.743 -0.009 0.837 0.023 0.589
Unemployed 0.069 0.099 0.047 0.350 -0.013 0.793
Retiree -0.089 0.079 0.109 0.037 -0.090 0.078
Household (1 to 6) -0.014 0.935 -0.171 0.333 0.104 0.562
Distance to nearest fire (km) 2.420 0.458 -3.019 0.352 2.024 0.545
Risk Preferences (1 to 10) 0.138 0.630 0.035 0.901 -0.018 0.951
Present Bias -0.063 0.736 0.120 0.516 0.017 0.925
Democrat Vote 0.026 0.689 -0.086 0.196 0.046 0.483
Republican Vote -0.112 0.047 0.078 0.169 0.038 0.455
Other Vote 0.087 0.089 0.008 0.888 -0.084 0.153
Observations 240 230 235

33
6 Results

6.1 Overview

This section provides summary statistics for the survey outcomes and looks at the explanatory

power of the covariates described in section 5.3, that is: socio-economic variables, time and risk

preferences and the distance to the fires.

6.1.1 Summary Statistics

Table 6.1: Summary Statistics

Mean SD Min Max Count

Panel A: Summary Statistics - Main Outcomes
Travel decision 0.72 0.45 0 1 690
Travel norm 0.89 0.32 0 1 690
Travel judgement 2.02 0.63 1 4 738
Energy decision 0.24 0.43 0 1 738
Energy norm 0.46 0.50 0 1 738
Energy judgement 2.36 0.79 1 4 738
Willingness to donate 0.24 0.43 0 1 467
Donation Norm 0.22 0.41 0 1 467
Panel B: Summary Statistics - Fires
Fire physical exposure (1 to 5) 2.13 1.18 1 5 738
Fire emotional exposure (1 to 5) 2.99 1.17 1 5 738
Panel C: Summary Statistics - Beliefs & Concerns
Wildfire Beliefs 3.87 1.24 1 5 467
Concern: personal harm (1 to 5) 3.53 1.24 1 5 467
harm to people in the US (1 to 5) 3.75 1.22 1 5 467
harm to people in other countries (1 to 5) 3.69 1.26 1 5 467
harm to future generations (1 to 5) 4.12 1.14 1 5 467
harm to plants and animal species (1 to 5) 4.14 1.12 1 5 467

Table 6.1 provides summary statistics for the different outcomes variables.28 We can see that

72% of the participants choose to travel while only 24% choose the non-renewable energy mix.

Several reasons can be put forward to explain this. It is highly probable that travelling is a much

more ”embedded” behaviour in the lifestyle of Californian residents, given the distances in between

28
Note: for the travel decision, we drop 51 observations due to a change of questions after the first 51 responses to
the survey.

34
cities in the state. This makes travelling a much harder behaviour to change. It can also be related

to the framing of the situations, with a clear incitement to travel in the first situation. The energy

situation was more presented as a choice in that respect. The exact wording for the travel situation

was:”Suppose you are planning a week-end trip to visit your friends [...]. The plane to go there cost

$200 round-trip [...]”, which pre-supposes that the trip is happening, making renouncement a much

more important behavioural ”deviation”. Finally, only 24% of the second-round participants (15%

over the whole sample) are willing to donate their experiment earnings to the NGO Friends of the

Earth.

Regarding the fires, the participants report a lower physical exposure than emotional exposure,

with most participants reporting a low physical exposure to fires.29 This makes sense as a majority

of the respondents live in San Francisco or Los Angeles areas, which were less affected by fire

casualties. We note however that both measures are relatively high on average (2.2 for physical

exposure and 3 for emotional exposure, on a scale from 1 to 5). To simplify the analysis, we define

an exposure dummy variable that is equal to 1 for ”exposed” participants that report fire exposure

levels above 2 for both emotional exposure and physical exposure (240 individuals).

Participants report relatively high levels of concern and strong beliefs that the wildfires events

in California are related to climate change. We can see that participants are mostly concerned for

future generations and biodiversity regarding climate change impacts. For the subsequent analysis,

in order to facilitate the interpretation of this measure, we use the average level of concern reported

for each participant as the concern outcome (the mean average concern across the sample is 3.84

out of 5).

6.1.2 Explanatory variables

The following tables show the effect of covariates described in section 5.3 on the main outcomes

(Table 6.2) and on beliefs (Table 6.3).

For the individual decisions, we can see that the employment status and the level of education

play a role in explaining the results. The level of concern also seems to affect the choices towards the

29
See the distribution of the responses to fire exposure in the Appendix - Section 12

35
less emitting options. The norms do not seem to be strongly affected by either of the explanatory

variables. Regarding the beliefs and level of concern, we can see that they are highly driven by the

voting behaviour and exposure to fires. Voting for a democrat candidate increases the beliefs by

1 to 2 points out of 5 with respect to voting for a republican candidate (i.e. the person increases

their belief in the likelihood of the relationship in between the fires and climate change by 1 to 2

points). Additionally, the level of exposure to fires - in particular the emotional exposure - explains

a significant part of the beliefs. Finally, we notice that more present-biased individuals report lower

levels of concern, which is consistent with the bias, as climate change is perceived as something that

will occur in the future.30

30
[Link]

36
 Table 6.2: Explanatory variables on main outcomes

(1) (2) (3) (4) (5) (6)

Travel Decision Energy Decision Donation Decision Travel Norm Energy Norm Donation Norm
Age -0.000311 -0.000163 -0.000836 0.000739 0.00121 -0.00283
(0.871) (0.924) (0.645) (0.544) (0.569) (0.106)
Gender -0.0691 -0.0510 0.0275 -0.00342 -0.0633 0.00655
(0.104) (0.180) (0.497) (0.899) (0.182) (0.866)
Level of Education (1 to 6) 0.00279 -0.0530∗∗ 0.00217 -0.00303 -0.00221 0.0144
(0.898) (0.006) (0.914) (0.825) (0.926) (0.461)
Student 0.142 -0.284∗ -0.0374 0.0419 0.0218 -0.0492
(0.308) (0.027) (0.783) (0.636) (0.891) (0.707)
Employed -0.00645 -0.129∗ -0.0201 0.0193 -0.105 -0.0884
(0.922) (0.028) (0.746) (0.643) (0.149) (0.140)
Unemployed -0.175∗ -0.118 0.0506 -0.0526 -0.0574 0.0540
(0.036) (0.115) (0.524) (0.317) (0.538) (0.481)
37

Household (1 to 6) 0.00537 0.0184 0.00813 0.00211 -0.0110 0.0193

(0.750) (0.231) (0.619) (0.844) (0.565) (0.221)
Distance to nearest fire (km) 0.000947 0.000310 -0.000604 0.000674 -0.00120 0.000488
(0.268) (0.689) (0.463) (0.214) (0.214) (0.538)
Risk Preferences (1 to 10) 0.0151 0.0110 0.000992 0.000179 -0.00202 0.00603
(0.129) (0.218) (0.917) (0.977) (0.856) (0.511)
Present Bias 0.00283 0.0359∗∗ -0.00530 -0.00957 0.00324 -0.0188
(0.846) (0.008) (0.710) (0.300) (0.846) (0.172)
Democrat Vote -0.0968 -0.180∗∗∗ 0.148∗∗ -0.0361 0.00103 0.0415
(0.068) (0.000) (0.003) (0.281) (0.986) (0.388)
Other Vote -0.0419 -0.173∗∗ 0.110 -0.0326 0.0478 0.00845
(0.511) (0.003) (0.069) (0.419) (0.501) (0.885)
Observations 433 467 467 433 467 467
R2 0.048 0.100 0.027 0.022 0.024 0.034
Note: This table presents coefficients estimated with a LPM for the main outcomes
 Table 6.3: Explanatory variables on beliefs & concerns

(1) (2) (3) (4) (5)

Wildfire Beliefs Concern Personal Concern Future Concern Biodiversity Average Level of Concern
Age -0.00430 0.000840 -0.000167 -0.00301 -0.000166
(0.335) (0.846) (0.967) (0.439) (0.963)
Gender 0.172 0.0665 0.257∗∗ 0.0439 0.172∗
(0.084) (0.492) (0.005) (0.614) (0.032)
Level of Education (1 to 6) 0.124∗ 0.0493 0.123∗∗ 0.0634 0.0851∗
(0.012) (0.304) (0.006) (0.142) (0.033)
Student 0.118 0.338 0.126 0.0650 0.234
(0.723) (0.297) (0.677) (0.823) (0.382)
Employed 0.0371 0.128 0.0828 0.0983 0.0976
(0.806) (0.385) (0.550) (0.458) (0.424)
Unemployed 0.0675 0.135 0.143 0.102 0.0948
(0.728) (0.473) (0.420) (0.549) (0.544)
Household (1 to 6) -0.0399 0.0335 0.0381 0.00566 0.0382
38

(0.317) (0.388) (0.296) (0.871) (0.234)

Distance to nearest fire (km) -0.000393 -0.000799 0.000523 -0.000985 -0.000392
(0.844) (0.682) (0.775) (0.574) (0.808)
Risk Preferences (1 to 10) -0.0117 0.0105 -0.0161 0.00804 -0.00954
(0.614) (0.642) (0.449) (0.693) (0.611)
Present Bias -0.0521 -0.0377 -0.0471 -0.0282 -0.0471
(0.135) (0.265) (0.139) (0.354) (0.094)
Fire emotional exposure (1 to 5) 0.103∗ 0.280∗∗∗ 0.183∗∗∗ 0.222∗∗∗ 0.234∗∗∗
(0.038) (0.000) (0.000) (0.000) (0.000)
Fire physical exposure (1 to 5) 0.0824 0.127∗∗ -0.0128 -0.0459 -0.00256
(0.096) (0.008) (0.776) (0.288) (0.949)
Democrat Vote 1.518∗∗∗ 1.311∗∗∗ 1.343∗∗∗ 1.389∗∗∗ 1.382∗∗∗
(0.000) (0.000) (0.000) (0.000) (0.000)
Other Vote 1.146∗∗∗ 1.134∗∗∗ 1.045∗∗∗ 1.057∗∗∗ 1.103∗∗∗
(0.000) (0.000) (0.000) (0.000) (0.000)
Observations 467 467 467 467 467
R2 0.323 0.355 0.329 0.361 0.408
Note: This table presents coefficients estimated with a LPM for the beliefs & concerns outcomes
6.2 Study of the outcomes

This section looks at the results obtained for the different outcomes evaluated. As a reminder, we

are looking to measure how cognitive biases, in particular the present bias, can explain the low level

of environmental behaviour; how social norms influence individual behaviour and how those norms

evolve with the exposure to information about climate change.

6.2.1 Time Preferences

We first look at the effect of the Present Bias variable that we elicite from the survey. As pointed

out in Table 6.2, we can see that the time preferences of individuals influences their choices. In

particular, we can see a significant effect for the energy choice: more present-biased individuals are

more likely to choose the non-renewable energy mix (top panel, middle graph):

Figure 7: Outcome choices by time preferences

Regressing the energy decision on the time bias, controlling for characteristics, we can see that

39
the bias increases the probability of choosing the non-renewable mix by 3.2%. This effect may be

explained by the fact that the time perspective involved in the choice of an energy provider is longer

than the choice to travel, which happens at one point in time. People that are more present-bias

therefore prefer short-term private gains (in terms of $ paid) rather than long-term social benefits

of clean energy. The time trade-off may be less prominent in the travel situation, which does not

involve a long-term commitment. Furthermore, the bias does not seem to significantly affect the

norms which is consistent with the hypothesis that the present-bias primarily affects individual

decisions.
Table 6.4: Present-Bias Effect on Energy Choices

(1) (2) (3) (4)

Energy Decision Energy Norm Energy Judgement Travel Decision
Present Bias 0.0329∗ 0.00366 -0.0311 0.00277
(0.027) (0.823) (0.157) (0.844)
Observations 467 467 738 433
R2 0.176 0.034 0.030 0.062
Note: Coefficients estimated with a LPM, including all Lasso-selected covariates.

6.2.2 Exposure to fires and distance

Regarding the distance to the fires, we first look at the correlation in between the participants’

closeness to the fires and the reported exposure in order to evaluate the credibility of the measure.

Closeness to the fire is defined as a binary variable: the individual is considered to live in a zip code

close to the fires (i.e. closeD = 1) if the distance to the nearest fire is less than D = {25, 15 or 10}

km.31 Figure 8 illustrates the closeness to the fires by reported exposure:

31
The average distance is 33 km

40
 Figure 8: Exposure by closeness to the fires (for 25km and 20km away)

There is very little correlation in between the reported exposure to fires and the distance from

the nearest fire. This may be due to the fact that individuals living closer to fires are more likely

to live in high-risk regions, which are more prepared and equipped for fires nearby; which would

result in a lower reported exposure. However, we can see that for smaller distances (less than 10

km32 ), participants living closer report somewhat higher levels of exposure. We use this measure

(close10) to plot the outcomes by closeness to the fires, which does not seem to be a strong driver

of the outcome decisions (see Figure 9). We notice a small effect on beliefs: people living closer to

the fires are more likely to believe that they are related to climate change.

32
This is the case for 116 participants

41
 Figure 9: Outcome decisions by distance to the fires (10km threshold)

The small effect of the distance to the fires on the reported exposure and on the outcomes is

a puzzling result for our analysis. A hypothesis for this could be that individuals in California are

used to fires and that the phenomenon does not enter their decision making as a key variable since

these weather events have historically been a part of Californian climate (although to a much less

extreme extent33 ). Further analysis in relation to the fire history is provided in the Appendix, in

section ??.

6.2.3 Beliefs

As we saw in Table 6.3, beliefs and concern are highly driven by socio-economic preferences, such

as voting behaviour, or exogenous factors like the exposure to fire:

33
[Link]

42
 Figure 10: Beliefs by exposure to the fires

Additionally, we observe no treatment effect for the reported beliefs and concern levels.34 Given

this result, we can (with caution) treat the reported beliefs and level of concern as potential ex-

planatory variables. From Figure 11, it seems that stronger beliefs about the relationship of the

fires to climate change result in a higher probability to donate and to choose the renewable energy

mix (top panel, middle graph). This is consistent, again, with the findings of Pace and van der

Weele (2020), where beliefs are important drivers of environmental behaviour. The travel decision

seems to be less impacted by the beliefs than the other decisions, which may be due to the fact

that travelling is a more mainstream behaviour, hence less subject to environmental beliefs35 and

concerns, or more costly to change based on those.

34
See Appendix - Section 12.3.5
35
Beliefs that the Californian fires are related to climate change

43
 Figure 11: Outcomes by levels of Belief

6.2.4 Correlation between outcomes

We look at the correlation in between outcomes, we can see that participants are more likely to

choose the energy mix if they chose to travel in the first round, irrespective of treatment. Similarly,

they are more likely to donate if they have renounced travel or chosen the non-renewable energy

mix. Figure 12 shows that participants are generally consistent across their individual choices:

44
 Figure 12: Outcome decisions across choices

Turning to the relationship in between norms and individual behaviour, we note that the in-

dividual decision is strongly related to the norm perception of the participants, regardless of their

treatment group, as illustrated in Figure 13:

45
 Figure 13: Outcome norms by respective individual decisions

While the effect is not as strong for the travel situation, looking at the energy and donation

choices, we see that individuals are more likely to perceive a norm that is close to their behaviour.

We notice, in particular that participants choosing the more polluting options are more likely to

perceive a norm that is consistent with their behaviour. We can check this by plotting the average

difference in between behaviour and norm perception across individuals in each treatment group:36

36
A difference that is closer 1 means the individual is more likely to choose to renounce travel, yet believe that
most people wouldn’t make that choice

46
 Figure 14: Mean differences in between the norms and decisions reported by the participants

Overall, we can see that the difference in between the norm and the individual decision is larger

for pro-environmental participants across all three situations. In the case of travel, we can see

that individuals that renounce travel are very pessimistic about the norm: they think most people

take the plane, even though themselves chose to renounce travel. This might be due to the fact

that renouncing to travelling is a stronger divergence from the ”mainstream” behaviour than using

renewable energy. The participants may be aware of this and hence report a higher pessimism

towards the average behaviour.

Finally, the judgement towards others can be interesting to evaluate in comparison to individual

choices, as a key component of social norms: are individuals willing to implement new social norms

when they change behaviour? Or alternatively, are they more lenient on the judgement of others

when themselves chose the more polluting option? In other words, we investigate whether partici-

pants were consistent in choosing a behaviour that they rate as socially appropriate. To evaluate

this, we plot the individual decisions by treatment group and appropriateness rating (inappropriate

= judgement < 2). We obtain:

47
 Figure 15: Individual Choices by Social Appropriateness Ratings

We can see that when individuals choose to renounce travelling, they are also more likely to

rate the action to take the plane as socially inappropriate. They are also more likely to think that

less people would take the plane (i.e. think the norm is to renounce travel). This suggests that

individuals are somewhat coherent in their judgement and individual actions. A similar pattern is

observed for the energy choice.

7 Analysis and Treatment Effects

As described in section 5.3, we collected a large set of explanatory variables that can be used as

controls to isolate the treatment effect. Using the adaptive LASSO (Zou, 2006), we select the

most important determinants of the outcome variables among demographic and socio-economic

characteristics in the data. The results from the adaptive LASSO are presented in Tables 12.1 and

12.2 in the Appendix. We create a set of controls from the selected covariates which we use to

identify the treatment effects in the following regressions using a linear probability model (LPM).37

Figure 16 plots the average answers of the participants, for each situation in the first round, by

treatment groups.38

37
See Appendix - Section 12.3.4 for specification and discussion of identification strategy
38
The outcomes are evaluated as binary variables where the higher-emission choice is logged as 1.

48
 Figure 16: Average Treatment Effects - Round 1 Outcomes

We can see that individuals in the treatment group are less likely to take the plane (66.04%)

compared to individuals in the control group (78.84%). They are also more likely to believe that less

people take the plane as a norm. Finally, they are more likely to judge the action of taking the plane

as socially inappropriate. The difference in between the treatment and control group outcomes for

the energy choice are less clear. There are almost no differences in the individual decisions and

the social appropriateness ratings. We can see a slight difference in the norm perception, where

individuals exposed to the treatment seem to believe that more people choose the renewable energy

option on average. It is interesting to note that in both cases, the norm perception is significantly

higher than the actual outcomes, irrespective of the treatment group.

For the second round, Figure 17 below shows the unconditional means of the donation outcomes

(individual decision and norms perception) across the four different groups:

49
 Figure 17: Average Treatment Effects - Donation Outcome

We notice a difference in the individual decision to donate among observed participants from the

treatment group, while all other groups show similar levels of donation. This suggests first that there

is a limited persistence of the treatment effect on individual behaviour or on norms perceptions.

Second, there is a combined effect of treatment and observation: when observed, treated individuals

seem more likely to donate. This could be related to the norm change after being exposed to the

treatment, however this is not consistent with the first round results, where the treatment did not

significantly influence the norm. It is also at odds with the donation norm perception that does not

seem to change with the treatment group. Alternatively, treated individuals may be more sensitive

to the norms and hence more willing to donate if they are observed.

The observation also has an effect on the norm perception: observed individuals are more likely

to believe that more people have donated than unobserved individuals, regardless of their treatment

group. This is consistent with the fact that telling people that they are observed contributes to

making the norm more salient and hence they are more likely to perceive more virtuous behaviours

as the norm.

50
7.1 Travel and Energy

For both situations, we run treatment effect regressions using the controls selected by LASSO.

The coefficients are, overall, of the expected sign. Being exposed to information on climate change

decreases the probability of choosing a more polluting alternative, strengthens the norm against

such choices and increases their social inappropriateness.

Table 7.1: Treatment Effect on Travel Choices

(1) (2) (3)

Individual Choice Perceived Norm Social Inappropriateness
Treatment -0.147∗∗∗ -0.0258 0.184∗∗
(0.000) (0.322) (0.001)

Age -0.000851 0.000708 -0.00280

(0.649) (0.580) (0.272)

Gender -0.0502 0.000245 0.0120

(0.225) (0.993) (0.840)

Level of Education (1 to 6) 0.0160 0.00216 0.00262

(0.427) (0.883) (0.925)

Employed -0.172∗ -0.0373 0.250

(0.020) (0.598) (0.093)

Unemployed -0.362∗∗∗ -0.117 0.386∗

(0.000) (0.142) (0.012)

Retiree -0.175 -0.0572 0.359∗

(0.082) (0.496) (0.037)

Household (1 to 6) 0.0123 0.00403 -0.00588

(0.463) (0.743) (0.816)

Present Bias -0.0000766 -0.0108 -0.0245

(0.996) (0.280) (0.252)

Average level of concern -0.0802∗∗∗ -0.0326∗∗ 0.155∗∗∗

(0.000) (0.004) (0.000)
Observations 433 433 467
R2 0.097 0.034 0.103
Controls Yes Yes Yes
Note: Coefficients estimated with a LPM including LASSO-selected covariates for the main outcomes. Further details
about the covariates selected for each outcome can be found in the Appendix. We include use a single set of covariates
to facilitate comparison across outcomes.

51
 Table 7.2: Treatment Effect on Energy Choices

(1) (2) (3)

Individual Choice Perceived Norm Social Inappropriateness
Treatment -0.0370 -0.0515 0.0575
(0.302) (0.267) (0.427)

Age -0.000192 0.00108 -0.00344

(0.900) (0.601) (0.293)

Gender -0.0320 -0.0621 -0.0125

(0.379) (0.186) (0.867)

Level of Education (1 to 6) -0.0350∗ -0.00355 -0.0474

(0.048) (0.880) (0.193)

Employed 0.136 -0.126 0.143

(0.088) (0.352) (0.430)

Unemployed 0.129 -0.0658 0.203

(0.149) (0.645) (0.293)

Retiree 0.243∗ -0.0260 0.138

(0.019) (0.870) (0.526)

Household (1 to 6) 0.0249 -0.0109 -0.0278

(0.113) (0.564) (0.341)

Present Bias 0.0307∗ 0.00323 -0.0384

(0.043) (0.846) (0.154)

Average level of concern -0.125∗∗∗ -0.0207 0.201∗∗∗

(0.000) (0.347) (0.000)
Observations 467 467 467
R2 0.165 0.024 0.083
Controls Yes Yes Yes
Note: Coefficients estimated with a LPM including LASSO-selected covariates for the main outcomes. Further details
about the covariates selected for each outcome can be found in the Appendix. We include use a single set of covariates
to facilitate comparison across outcomes.

52
 More precisely, we see that the treatment reduces the probability of taking the plane by 15%,

while the effect is small (3.5%) and not significant for the energy choices. We can put several hy-

potheses forward to explain this result. First, it is important to remember that a lot more people

choose to travel compared to the number of individuals choosing the energy mix in the second

situation. This implies that choosing a renewable energy provider is a much more mainstream

behaviour than renouncing travel: we are less likely to observe a treatment effect on a behaviour

that is already normalised. Second, the choice in between taking the plane or not and the choice

of an energy provider subscription are inherently different in several aspects. Travelling by plane

is a situation of which it more likely that individuals know that it directly contributes to climate

change. This does not suffice in changing behaviours, however, as individuals may operate some

cognitive dissonance, or strategic ignorance in choosing to ignore the fact or exploit the uncertainty

of the emissions entailed by a plane trip. The provision of information makes it harder for treated

individuals to operate this dissonance as they are directly reminded of the facts (see Thunström

et al. (2014)), which would explain why the treatment has a stronger effect in this situation. Ad-

ditionally, travelling is more likely to be observed by other people: the social image component is

stronger than for the energy choice, and social norms may have a greater influence in this situation,

pushing individuals towards more virtuous choices, as seen in the model. The treatment effect on

the judgement towards others’ behaviour could be interpreted as confirming this hypothesis. We

indeed observe that treated individuals judge the action of taking the plane almost 2 points more

inappropriate than non-treated respondents. Similarly, although it is less significant, the treatment

does have an effect in changing the norm perception towards a more pro-environmental behaviour

(2%). Finally, the time perspective of the two choices is also different: the decision to travel is

a one-time decision while the choice of energy provider is a long-term choice, which includes a

commitment in the future (to pay the bills). This hypothesis is reinforced by the effect that we

observed for the present bias on the energy choice: more present-biased individuals are 3.3% more

likely to choose the non-renewable mix. We note that the average level of concern decreases the

likelihood to take the plane and choose the non-renewable energy mix by 8.02% and 12.5% re-

spectively. Similarly, it increases the social inappropriateness judgement of these actions by 1.6

and 2 points respectively. This implies that more concerned individuals are more likely to adopt

53
pro-environmental behaviours. Again, this is in line with the findings in the literature and suggests

that raising individual levels of concern can be a driver of environmental action.

Ultimately,the employment situation also seems to have an effect on the travel choice, specif-

ically, unemployed individuals choosing to travel less than others. This may be explained by the

potentially lower income or lower propensity to consume ”leisure goods” of unemployed individu-

als. Likewise, more educated individuals seem to be more likely to choose the renewable option,

suggesting that sustainable energy choices may be correlated with socio-economic classes.

7.2 Donation

The following tables provide the treatment effect on the donation decision, controlling for the

covariates selected by LASSO. Table 7.3 displays the treatment and observation effects separately

while Table 7.4 shows the coefficients for the interaction of both effects.

From Table 7.3, we can see that the treatment and observation have marginally significant

effects, increasing the probability to donate respectively by 6.6% and 6.5%. Observed individuals

are also more likely to believe that most people in the panel chose to donate. The observation

increases the norm perception by 9.4%, which confirms the higher salience of the norm under

observation. This also contributes to reinforcing the hypothesis that decisions that can be publicly

observed (such as travelling by plane) are more sensitive to the pressure of the norms than private

decisions (such as the choice of energy provider).

In Table 7.4, we see that having been exposed to the treatment in the first round and subse-

quently being observed increases the willingness to donate by 13% with respect to being in another

group. This result suggests that the treatment effect may increase the sensitivity to the norm, push-

ing observed individuals to donate. In particular, the observation might ”reactivate” the treatment

effect, which is observed to decrease over the long term. This is also in line with the hypothesis that

the treatment may increase the cost of ”strategic ignorance”, which is reinforced by the norm. This

should however be verified by the joint effect on the norm, which, in our case is only marginally

significant.

54
 Table 7.3: Treatment Effects on Donation Choices

(1) (2)
Donation Decision Donation Norm
Treatment 0.0581 -0.00977
(0.137) (0.795)
Observed 0.0610 0.0900∗
(0.120) (0.018)
Age -0.00136 -0.00306
(0.450) (0.060)
Gender 0.0314 0.0144
(0.430) (0.709)
Level of Education (1 to 6) -0.00658 0.0135
(0.727) (0.496)
Employed 0.0571 -0.0267
(0.622) (0.835)
Unemployed 0.137 0.108
(0.265) (0.424)
Retiree 0.0854 0.0664
(0.527) (0.640)
Household (1 to 6) 0.00529 0.0201
(0.727) (0.199)
Present Bias -0.00553 -0.0211
(0.691) (0.066)
Average level of concern 0.0600∗∗ 0.0250
(0.001) (0.194)
Fire emotional exposure (1 to 5) 0.0429∗ 0.0236
(0.038) (0.251)
Fire physical exposure (1 to 5) -0.000924 0.00894
(0.964) (0.644)
Observations 467 467
R2 0.064 0.057
Controls Yes Yes
Note: Coefficients estimated with a LPM including LASSO-selected covariates for the main outcomes. Further details
about the covariates selected for each outcome can be found in the Appendix. We include use a single set of covariates
to facilitate comparison across outcomes.

55
 Table 7.4: Treatment & Observation Effects on Donation Choices

(1) (2)
Individual Donation Donation Norm
Observed & Treated 0.119∗ 0.0802
(0.039) (0.130)
Observed & Control 0.0179 0.0989
(0.742) (0.078)
Unobserved & Treated 0.0165 -0.00111
(0.753) (0.982)
Observations 467 467
R2 0.066 0.057
Controls Yes Yes
Note: This table presents the LPM estimated coefficients for the 4 different groups in the second round. All LASSO-
selected covariates were included.

7.3 Heterogeneity in Treatment Effects Analysis

The treatment effects that we measure from the survey answers in the previous section are the

average effects of the treatments across the panel. As we saw in the description of the data, the

panel presents heterogeneous characteristics, which may drive individual responses to the treatment

in different ways. In this section we propose to examine underlying heterogeneity of the treatment

effects that we measure. In particular, we further investigate the drivers of the treatment × obser-

vation. We evaluate the donation choice among different groups, for each observation and treatment

group:

56
 Figure 18: Donation Choice by treatments and explanatory variables categories

In line with the previous results, we can see that participants that choose to renounce travel

and/or the renewable energy option are more likely to donate. We notice that the treatment ×

observation effect is different in between the two groups: for individuals that renounced travelling by

plane, we see a combined effect of treatment × observation; while we observe the same effect among

individuals that chose the non-renewable energy mix. Additionally, the treatment × observation

effect seems to be substantial for male and unemployed individuals, as well as those that report

high levels of concern.

It is difficult to make sense of those results, and the above comments are subject to cautious

interpretation as the number of observations in each category is very small, leading to weak statistical

significance (as we can see from the 95% confidence intervals). We provide additional figures on

heterogeneity for the other outcomes (travel and energy) in the appendix, section 12.3.6. The

57
relatively low treatment effects and the small number of participants significantly alters our capacity

to provide meaningful analysis of the heterogeneity in the drivers of the observed treatment effects.

8 Discussion

We saw in the model that naive present bias in the first period, subsequently increased the number

of individuals polluting in the second and third period, hence increasing the level of aggregate

emissions. This change in the level of aggregate emissions increases with the present bias: the

higher the bias (lower β), the larger the increase in emissions. Intuitively, the present-bias effect

also increases in the value of the emitting action. The higher the present-value of the benefit from

the action, the stronger the effect of the bias. When adding social norms to the agent’s utility, we

see that they can contribute to reducing the present bias effect through making the less virtuous

action more costly to the individual. Without changing the number of naive individuals in the

population, the sensitivity to social norms in the individual utility reduces the difference in between

the perceived and actual cutoff values.

The survey provides substantial information on the effect of the present-bias and the norms in

the context of environmental behaviour choices. In line with the findings from the model, we measure

an effect of time preferences on pro-environmental behaviour. Individuals that are more susceptible

to be present-biased are less likely to choose the lower-emissions options, in particular in the energy

case, a situation that implies a long-term behaviour change and commitment. We do not measure

whether biased individuals are naive or sophisticates about their present-bias, however, our results

suggest that they do not account for the bias in their choices. Drawing from the model, social norms

could be an efficient lever to correct this bias. It is indeed what is found in the literature: for energy

choices in particular, several studies have found that informing individuals about the local average

consumption, making the descriptive norms more explicit and hence increasing the sensitivity to the

norm, µ, changes individual behaviour to adjust towards the norm (Schultz et al. (2007); Allcott

and Rogers (2014)). Policies that make social norms more salient, both descriptive and injunctive

norms, can therefore foster more environmentally-friendly behaviour. Besides energy choices, this

is relevant for several situations in which behavioural shifts are necessary to fight climate change,

58
such as long-term mobility choices (i.e. buying an electric car vs. a combustion engine car; or a new

car vs. a used car). Furthermore, in order to directly address the present-bias, one could think of

policies that account for time inconsistent preferences, such as progressive price increases for energy

subscriptions, making decisions that involve change in the long-term more appealing.

The analysis of the survey answers shows a strong relationship between individual behaviour and

the norms perceived and enforced by the individual. Furthermore, we find that individuals generally

underestimate the norm, in line with the results from Andre et al. (2021). We find, however, that

the correlation in between decisions, judgement and norm perception varies in the individual type.

Participants choosing less sustainable options are more likely to perceive a norm that is consistent

with their behaviour and rate that behaviour as socially appropriate. On the other hand, pro-

environmental participants believe that most people do not behave like them on average. Yet, we

find that they are more likely to judge polluting behaviours as socially inappropriate, thus enforcing

a more pro-environmental norm. We find furthermore that both the norm perception difference and

stronger inappropriateness judgement is driven by treated individuals. These findings suggests that

individuals dissociate their perception of the average choice from the way they would judge someone

else’s behaviour. The norms enforced by individuals (injunctive norms) seem to change faster than

the descriptive norms, or are more likely to change, when the individual changes their individual

behaviour.

The survey results further show that the treatment has a significant effect on reducing the

probability of choosing to travel by plane. The provision of information on the consequences of

climate change can therefore contribute to changing individual choices. As discussed previously,

this effect is only measured in the case of travel, suggesting that some types of situations may be

more sensitive to this kind of information treatment than others. The hypothesis that it may be

related to social image is also reinforced by the effect of the combined treatment and observation on

the donation choice. This result shows that participants are sensitive to external judgement on their

decision to donate, without changing their perception of the average behaviour. In the first round, we

noticed that individuals that choose to renounce travelling are more likely to judge plane travelling

as a socially inappropriate behaviour, highlighting the enforcement of a pro-environmental norm,

after treatment. Additionally, since we find a very limited treatment effect on norm perception,

59
we cannot relate the behavioural change to a change in the perception of the norms. By analysing

the results of the first and second round jointly, we could infer that the treatment affects the

injunctive norms at first. In the case of the donation, this injunctive norm is materialised through

the observation and activates the initial treatment effect. We can also highlight that the outcome

for observed and treated participants, is the only instance in the second round where we observe

a persistent treatment effect across time. This is in line with the claim that social norms changes

have long-lasting effects that can sustain individual behavioural changes (Kinzig et al. (2013)).

Ultimately, we do not find that beliefs shift as a result of the treatment, which may in part

be due to the delay in between the two rounds. We find, however that the beliefs are driven by

the experiences of the individuals such as the exposure to the fires or the level of education. In

turn, we observe that those beliefs and concern levels are important drivers of pro-environmental

choices. This is consistent with the findings in the literature, as in Pace and van der Weele (2020)

for example. Further research could explore the underlying mechanisms in explaining the main

outcome observations in relation to variations in the beliefs.

Finally, we find very limited evidence that the distance to the fires influences the outcomes.

However, this result should be interpreted with caution, as we also find that the distance we measure

does not correlate strongly with the reported exposure of the participants.39 A more granular

approach to measuring this experience may provide more accurate results, in line with those from

research on the influence of experiencing extreme weather events on the willingness to fight climate

change (Spence et al. (2011)).

9 Limits

There are several limits to this study. On the model, first, we choose not to take into account

the concern about social norms in future periods when the individual decides in period 1. For our

purposes, this is sufficient to show that social norms can counter the effect of the present bias,

however it would be more realistic for individuals to account for future norms concern as well.

Regarding the survey, an obvious limit to this study is that respondents did not face the real
39
See further discussion in the Appendix - section ??

60
consequences of their choices, in the sense that their choices were not actually implemented in the

context of the experiment. Even in the case of the donation, we had no means to evaluate if those

agreeing to donate actually implemented the donation. This creates a ”hypothetical bias” (see

Harrison (2014); Harrison and Rutstrm (2000)) in which individuals may declare a behaviour that

they would not comply to in real-life. This would bias the treatment effects upwards.

In addition to this, the situations in which we evaluate the participants’ environmental be-

haviour may present some intrinsic biases that can influence the outcomes. Travelling by plane is

the kind of practice that is very linked to local cultural contexts, shaped by the distances in between

cities and the mobility infrastructure that is developed to accommodate the inhabitants of the area.

In Europe, travelling by plane is a lot more of a debated issue than in the United-States, because

distances in between cities are relatively shorter and alternative mode of transport are cleaner and

more efficient. In the United-States, the plane still often is the best travel mode in terms of price,

time and comfort. It is obvious that we should take this into consideration when interpreting the

findings from out study.

Finally, the number of respondents to the survey limits the statistical power of determining the

heterogeneity of treatment effects. Due to the small number of individuals in each category, the

study doesn’t provide a clear picture of the characteristics that could explain the treatment effects

for the different outcomes.

10 Conclusion

Both the model and the survey in our study confirm the intuition that time preferences influence

environmental behaviour and that social norms play an important role in shaping this behaviour.

Once again, these results put forward the case that information about the consequences of climate

change can change environmental behaviour and that this effect is reinforced when the decisions

are publicly observed. It also increases individual willingness to enforce pro-environmental norms,

which could ultimately change the collective behaviour of a group. Furthermore, it highlights

that individual environmental choices are clearly driven by cognitive biases and firmly-implemented

beliefs, which means that market-based solutions will never fully be able to compensate for the

61
environmental consequences of our current lifestyles. Instead, it is crucial to use levers like social

norms that can implement long-lasting changes. An interesting case for further research would

be to look at the underlying heterogeneity in the behavioural changes observed as a result of the

treatments. This could help understand the mechanisms to which different individuals respond to

in order to design policies that take these specificities into account. This is particularly important

in the case of climate and environmental policy because of the heterogeneous impacts of climate

change in the population.

62
11 Bibliography

References

Karen Akerlof, Edward W. Maibach, Dennis Fitzgerald, Andrew Y. Cedeno, and Amanda Neu-

man. Do people “personally experience” global warming, and if so how, and does it matter?

Global Environmental Change, 23(1):81–91, 2013. ISSN 0959-3780. doi: [Link]

1016/[Link].2012.07.006. URL [Link]

S0959378012000908.

Hunt Allcott and Todd Rogers. The short-run and long-run effects of behavioral interventions:

Experimental evidence from energy conservation. American Economic Review, 104(10):3003–37,

October 2014. doi: 10.1257/aer.104.10.3003. URL [Link]

1257/aer.104.10.3003.

Steffen Andersen, Glenn Harrison, Morten Lau, and Elisabet Rutström. Elicitation using multiple

price list formats. Experimental Economics, 12:365–366, 09 2009. doi: 10.1007/s10683-008-9204-6.

Peter Andre, Teodora Boneva, Felix Chopra, and Armin Falk. Fighting Climate Change: The

Role of Norms, Preferences, and Moral Values. Working Papers 2021-036, Human Capital and

Economic Opportunity Working Group, July 2021. URL [Link]

wpaper/[Link].

James Andreoni and Charles Sprenger. Estimating time preferences from convex budgets. American

Economic Review, 102(7):3333–56, December 2012. doi: 10.1257/aer.102.7.3333. URL https:

//[Link]/articles?id=10.1257/aer.102.7.3333.

Roland Benabou and Jean Tirole. Laws and norms. Working Paper 17579, National Bureau of

Economic Research, November 2011. URL [Link]

Anders Biel and John Thøgersen. Activation of social norms in social dilemmas: A review of the

evidence and reflections on the implications for environmental behaviour. Journal of Economic

Psychology, 28:93–112, 01 2007. doi: 10.1016/[Link].2006.03.003.

63
Pablo Brañas-Garza, Diego Jorrat, Antonio Espı́n, and Angel Sánchez. Paid and hypothetical time

preferences are the same: Lab, field and online evidence. Working Papers 54, Red Nacional

de Investigadores en Economı́a (RedNIE), April 2021. URL [Link]

wpaper/[Link].

Leonardo Bursztyn, Georgy Egorov, and Stefano Fiorin. From extreme to mainstream: The erosion

of social norms. American Economic Review, 110(11):3522–48, November 2020. doi: 10.1257/

aer.20171175. URL [Link]

Robert Cialdini. Basic social influence is underestimated. Psychological Inquiry, 16:158–161, 10

2005. doi: 10.1207/s15327965pli1604 03.

R. H. Coase. The problem of social cost. The Journal of Law Economics, 3:1–44, 1960. ISSN

00222186, 15375285. URL [Link]

John Cook, Daniel Bedford, and Scott Mandia. Raising climate literacy through addressing misin-

formation: Case studies in agnotology-based learning. Journal of Geoscience Education, 62(3):

296–306, 2014. doi: 10.5408/13-071.1. URL [Link]

Jon Elster. Social norms and economic theory. Journal of Economic Perspectives, 3(4):99–117,

December 1989. doi: 10.1257/jep.3.4.99. URL [Link]

1257/jep.3.4.99.

John-Oliver Engler, David Abson, and Henrik von Wehrden. Navigating cognition biases in the

search of sustainability. AMBIO A Journal of the Human Environment, 48:605–618, 09 2018.

doi: 10.1007/s13280-018-1100-5.

Shane Frederick, George Loewenstein, and Ted O’Donoghue. Time discounting and time preference:

A critical review. Journal of Economic Literature, 40(2):351–401, June 2002. URL https:

//[Link]/articles?id=10.1257/002205102320161311.

Robert Gifford. The dragons of inaction: Psychological barriers that limit climate change mitigation

and adaptation. American Psychologist - AMER PSYCHOL, 66:290–302, 05 2011. doi: 10.1037/

a0023566.

64
Andrew Markus Guess and Alexander Coppock. Back to bayes: Confronting the evidence on

attitude polarization. 2015.

Glenn Harrison. Real choices and hypothetical choices. pages 236–254, 2014. URL https://

[Link]/RePEc:elg:eechap:14820_10.

W. Harrison and E. Rutstrm. Experimental evidence on the existence of hypothetical bias in value

elicitation methods. 05 2000.

Cameron Hepburn. Environmental policy, government, and the market. Oxford Review of Economic

Policy, 26(4):734–734, 2010. URL [Link]

y:2010:i:4:p:734-734.

Andrew Hoffman and Max Bazerman. Changing environmental practice: Understanding and over-

coming the organizational and psychological barriers. 01 2005.

IPCC. Summary for policymakers, 2021.

IPCC. Climate change 2022: Mitigation of climate change. contribution of working group iii to the

sixth assessment report of the intergovernmental panel on climate change, 2022.

Dan M. Kahan, Ellen Peters, Maggie Wittlin, Paul Slovic, Lisa Larrimore Ouellette, Donald

Braman, and Gregory Mandel. The polarizing impact of science literacy and numeracy on

perceived climate change risks. Nature Climate Change, 2(10):732–735, October 2012. URL

[Link]

Ann Kinzig, Paul Ehrlich, Lee Alston, Kenneth Arrow, Scott Barrett, Timothy Buchman, Gretchen

Daily, Bruce Levin, Simon Levin, and Michael Oppenheimer. Social norms and global environ-

mental challenges: The complex interaction of behaviors, values, and policy. BioScience, 63:

164–175, 03 2013. doi: 10.1525/bio.2013.63.3.5.

Erin Krupka and Roberto Weber. Identifying social norms using coordination games: Why does

dictator game sharing vary? Journal of the European Economic Association, 11, 12 2008. doi:

10.1111/jeea.12006.

65
David Laibson. Golden eggs and hyperbolic discounting. The Quarterly Journal of Economics, 112:

443–77, 02 1997. doi: 10.1162/003355397555253.

Adam Levine and Reuben Kline. A new approach for evaluating climate change communication.

Climatic Change, 142, 05 2017. doi: 10.1007/s10584-017-1952-x.

George Loewenstein and Richard H. Thaler. Anomalies: Intertemporal choice. Journal of Eco-

nomic Perspectives, 3(4):181–193, December 1989. doi: 10.1257/jep.3.4.181. URL https:

//[Link]/articles?id=10.1257/jep.3.4.181.

Taciano L. Milfont. The interplay between knowledge, perceived efficacy, and concern about global

warming and climate change: A one-year longitudinal study. Risk Analysis, 32(6):1003–1020,

2012. doi: [Link] URL [Link]

[Link]/doi/abs/10.1111/j.1539-6924.2012.01800.x.

Luis Mundaca, Rocio Román-Collado, and José Cansino. Assessing the impacts of social norms

on low-carbon mobility options. Energy Policy, 162:112814, 03 2022. doi: 10.1016/[Link].2022.

112814.

Karine Nyborg. Social norms and the environment. Annual Review of Resource Economics, 10

(1):405–423, 2018. doi: 10.1146/annurev-resource-100517-023232. URL [Link]

1146/annurev-resource-100517-023232.

Ted O’Donoghue and Matthew Rabin. Doing it now or later. The American Economic Review, 89

(1):103–124, 1999. ISSN 00028282. URL [Link]

Elinor Ostrom. Governing the Commons: The Evolution of Institutions for Collective Action.

Political Economy of Institutions and Decisions. Cambridge University Press, 1990. doi: 10.

1017/CBO9780511807763.

Elinor Ostrom. Collective action and the evolution of social norms. Journal of Economic Perspec-

tives, 14(3):137–158, September 2000. doi: 10.1257/jep.14.3.137. URL [Link]

org/articles?id=10.1257/jep.14.3.137.

66
Davide Pace and Joël van der Weele. Curbing carbon: An experiment on uncertainty and informa-

tion about co2 emissions. Tinbergen Institute Discussion Papers 20-059/I, Tinbergen Institute,

2020. URL [Link]

A. C. Pigou. The economics of welfare / by A. C. Pigou. Macmillan London, 1920.

Charles Raux, Amandine Chevalier, Emmanuel Bougna, and Denis Hilton. Mobility choices and

climate change: Assessing the effects of social norms and economic incentives through discrete

choice experiments. 01 2015.

Paul Schultz, Jessica Nolan, Robert Cialdini, Noah Goldstein, and Vladas Griskevicius. The con-

structive, destructive, and reconstructive power of social norms. Psychological science, 18:429–34,

06 2007. doi: 10.1111/j.1467-9280.2007.01917.x.

Lisa Shu and Max H. Bazerman. Cognitive barriers to environmental action: Problems and solu-

tions. Harvard Business School NOM Unit, Working Paper, 11 2010.

Alexa Spence, Wouter Poortinga, Catherine Butler, and N.F. Pidgeon. Perceptions of climate

change and willingness to save energy related to flood experience. Nature Climate Change, 1, 04

2011. doi: 10.1038/nclimate1059.

Matthias Sutter, Peter Lindner, and Daniela Platsch. Social norms, third-party observation and

third-party reward. Working Papers in Economics and Statistics 2009-08, Innsbruck, 2009. URL

[Link]

Richard Thaler. Some empirical evidence on dynamic inconsistency. Economics Letters, 8(3):201–

207, 1981. ISSN 0165-1765. doi: [Link] URL https:

//[Link]/science/article/pii/0165176581900677.

Linda Thunström, Klaas van ’t Veld, Jason F. Shogren, and Jonas Nordström. On strategic ig-

norance of environmental harm and social norms. Revue d’économie politique, 124(2):195–214,

2014. URL [Link]

Amos Tversky and Daniel Kahneman. Judgment under uncertainty: Heuristics and biases. Science,

67
 185(4157):1124–1131, 1974. doi: 10.1126/science.185.4157.1124. URL [Link]

org/doi/abs/10.1126/science.185.4157.1124.

Sander van der Linden. The social-psychological determinants of climate change risk percep-

tions: Towards a comprehensive model. Journal of Environmental Psychology, 41:112–124,

2015. ISSN 0272-4944. doi: [Link] URL [Link]

[Link]/science/article/pii/S0272494414001170.

Stepan Vesely. Elicitation of normative and fairness judgments: Do incentives matter? Judgment

and decision making, 10:191–197, 03 2015.

Martin Weitzman. Why the far-distant future should be discounted at its lowest possible rate.

Journal of Environmental Economics and Management, 36(3):201–208, 1998. URL https://

[Link]/RePEc:eee:jeeman:v:36:y:1998:i:3:p:201-208.

Martin L. Weitzman. Gamma discounting. American Economic Review, 91(1):260–271, March

2001. doi: 10.1257/aer.91.1.260. URL [Link]

91.1.260.

J Xie, Sameet Sreenivasan, Gyorgy Korniss, Weituo Zhang, Chjan Lim, and Boleslaw Szymanski.

Social consensus through the influence of committed minorities. Physical Review E, 84:011130,

07 2011. doi: 10.1103/PhysRevE.84.011130.

68
12 Appendix I

12.1 Model

This section provides further details on the threshold computations for the model. We want to find

the cutoff points (i.e. θi∗ ) such that individual i is indifferent in between choosing ai = 1 and ai = 0.

Since there are three interdependent periods in the model, we use backwards induction.

12.1.1 Period 3 decision

In period 3, the individual solves the following maximisation problem:

max a3 γ − ea¯3 − θi (ai1 + a2 + I{a2 < a3 })2

a3 ∈{0,1}

Since ai3 is binary, we can consider that the agent will choose ai3 = 1 if and only if:40

γ − ea¯3 − θi (ai1 + ai2 + I{a2 < a3 })2 ≥ −ea¯3 − θi (ai1 + ai2 )2

The choice of action in the last period depends on the type of the individuals as well as on the

choices made in the previous periods. We consider each possible combination of choices for ai1 and

ai2 separately and find that:

1. If ai1 = ai2 = 1

γ − ea¯3 − θi (ai1 + ai2 + I{a2 < a3 })2 > −ea¯3 − θi (ai1 + ai2 )2

⇐⇒ γ − 4θi > −4θi

⇐⇒ γ > 0

Which is always true: individuals always choose ai3 = 1, for all θi ∈ Θ in this case.

40
When the agent is indifferent, we consider that they choose ait = 0

69
 2. If ai1 = 0 and ai2 = 1

γ − ea¯3 − θi > −θi

⇐⇒ γ > 0

Which is always true: individuals always choose ai3 = 1, for all θi ∈ Θ in this case.

3. If ai1 = ai2 = 0

γ − θi > 0

⇐⇒ θi < γ

Individuals choose ai3 = 1 if θi < γ.

4. If ai1 = 1 and ai2 = 0

γ − 4θi > −θi

γ
⇐⇒ θi <
3

Individuals choose ai3 = 1 if θi < γ3 .

γ
We therefore obtain two cutoff types for the third period: θ3∗ = γ and θ3∗∗ = 3 which determine the

decision rules for period 3.

12.1.2 Period 2 decision

Once we know the decision rules for period 3, we can solve the maximisation for period 2:

max a2 γ − ea¯2 − θi (ai1 + I{a1 < a2 })2 + βδ a3 γ − θi (a1 + a2 + I{a2 < a3 })2


a2 ∈{0,1}

Again, the decision here depends on the decisions made in other periods. We study each situation

separately and find that:

1. If a1 = 1

70
Choose a2 = 1 if
γ γ(1 + βδ)
< θi <
3 3βδ

And
γ(1 + βδ) 1
<γ if βδ >
3βδ 2

We define
γ(1 + βδ)
θ2∗ (a1 = 1) =
3βδ

2. If a1 = 0

Choose a2 = 1 if θi < γ.

Therefore the cutoff for period 2 is θ2∗ (a1 = 1).

12.1.3 Period 1 decision

Solve:

max a1 γ + βδ a2 γ − θi (ai1 + I{a1 < a2 })2 − ea¯2 + βδ 2 a3 γ − θi (a1 + a2 + I{a2 < a3 })2 − ea¯3



a1

γ
Let θi < 3

Here, a2 = a3 = 1 in all cases → individual i will always choose a1 = 1

γ γ(1+βδ)
Let 3 < θi < 3βδ

Similarly, a2 = a3 = 1 in all cases → individual i will always choose a1 = 1

γ(1+βδ)
Let 3βδ < θi < γ

When a1 = 1, then a2 = a3 = 0 & if a1 = 0, then a2 = a3 = 1. We have a1 = 1 if:

γ + βδ(−θi ) + βδ 2 (−θi ) > βδ(γ − θi ) + βδ 2 (γ − θi )

⇐⇒ γ > βδγ(1 + δ)

⇐⇒ βδ(1 + δ) > 1

71
If we assume that the last inequality holds → individual i will always choose a1 = 1

Let θi > γ

Here, a2 = a3 = 0 in all cases. The agent will choose ai1 = 1 if:

γ + βδ(−θi ) + βδ 2 (−θi ) > 0

γ
⇐⇒ θi <
βδ(1 + δ)

And
γ
>γ
βδ(1 + δ)
γ
We define a new cutoff θ1∗ = βδ(1+δ) .

12.1.4 Decision rules

From the thresholds for θi computed above, and if βδ(1 + δ) > 1 we have:

1. If θi < θ2∗ , then ai1 = ai2 = ai3 = 1

2. If θ2∗ < θi < γ then if a1 = 1, a2 = a3 = 0 and if a1 = 0, a2 = a3 = 1.

3. If γ < θi < θ1∗ , then ai1 = 1 and ai2 = ai3 = 0

4. If θi > θ1∗ , then ai1 = ai2 = ai3 = 0

12.1.5 Social Norms

Let θi < θ2∗

We know that a2 = a3 = 1, independently of the choice of ai1 here. Therefore, ai1 = 1 if:

γ + µE[θi |ai1 = 1] + βδ(γ − θi ) + βδ 2 (γ − 4θi ) ≥ µE[θi |ai1 = 0] + βδ(γ − θi ) + βδ 2 (γ − θi )

⇐⇒ γ + µ E[θi |ai1 = 1] − E[θi |ai1 = 0] + βδ(γ − θi ) + βδ 2 (γ − θi ) ≥ 0

γ + µ E[θi |ai1 = 1] − E[θi |ai1 = 0]

⇐⇒ θi ≤
3βδ 2

72
This defines a first cutoff satisfying the following equation:

γ − µ∆(θ11∗ )
θ11∗ =
3βδ 2

Let θ2∗ < θi < γ

We know that a2 = a3 = 1 if ai1 = 0 and a2 = a3 = 0 if ai1 = 1.

Therefore, ai1 = 1 if:

γ + µE[θi |ai1 = 1] + βδ(−θi ) + βδ 2 (−θi ) ≥ µE[θi |ai1 = 0] + βδ(γ − θi ) + βδ 2 (γ − θi )

⇐⇒ γ(1 − βδ(1 + δ)) + µ E[θi |ai1 = 1] − E[θi |ai1 = 0] ≥ 0

⇐⇒ γ(1 − βδ(1 + δ)) ≥ µ E[θi |ai1 = 0] − E[θi |ai1 = 1]

There is an implicit cutoff defined here:

γ(1 − βδ(1 + δ)) ≥ µ∆(θ12∗ )

Therefore, if ∆(θ12∗ ) is such that the above inequality holds, individual i will choose ai1 = 1 in all

cases.

Let θi > γ

We know that a2 = a3 = 0 in all cases.

Therefore, ai1 = 1 if:

γ + µE[θi |ai1 = 1] + βδ(−θi ) + βδ 2 (−θi ) ≥ µE[θi |ai1 = 0]

⇐⇒ γ + µ E[θi |ai1 = 1] − E[θi |ai1 = 0] − βδθi (1 + δ) ≥ 0

γ + µ E[θi |ai1 = 1] − E[θi |ai1 = 0]

⇐⇒ θi ≤
βδ(1 + βδ)

Which defines a last cutoff satisfying the following equation:

γ − µ∆(θ13∗ )
θ13∗ =
βδ(1 + βδ)

73
12.2 Experiment Design

12.2.1 IPCC Report

The IPCC is a body of scientists in charge of surveying and summarising the findings in climate

change research. The scientists authoring the Sixth Assessment Report (AR6), 721 experts from 90

countries, reviewed approximately 50,000 scientific papers for the report. It is important to note

that, given the amount of work that is invested in those reports (over 4 years of voluntary work from

the scientists); this is likely to be the last report from the IPCC before 2030. The report consists

of three parts:

• Working Group I report on the Physical Science of Climate Change (released in August

2021): this part reviews the latest scientific findings on the state of the climate, using publi-

cations across a large range of disciplines in order to understand the evolution and magnitude

of the changes occurring.

• Working Group II report on Climate Change Impacts, Adaptation and Vulnerability (re-

leased in February 2022): specifically looks at the impacts of climate change and measures the

vulnerability of ecosystems and societies to them. It highlights the main adaptation solutions

to those impacts, and the risks of mal adaptation or delayed adaptation.

• Working Group III report on Climate Change Mitigation (released in April 2022): assesses

the feasibility and accessibility of solutions to mitigate the changes in the climate system,

using both a technical and socio-economic approach. That is, in addition to reporting tech-

nical/technological mitigation solutions (for example, renewable energy and low-carbon man-

ufacturing), there is also a strong emphasis on the policy and socio-cultural norms approach,

which is inevitably a part of the solution given the importance of the societal and economic

change that is needed to meet the targets and staying below 1.5°C of warming.

12.2.2 Treatment

All the information in the treatment slides was taken from the AR6 (IPCC, 2021-2022), the En-

vironmental Protection Agency (EPA) and the Government of California documents on climate

74
change impact analysis for the state. The physical science basis was taken from the Working Group

I Report (WGI). Projected impacts in California were combined from the WGII Report on regional

impacts in North-America (AR6, WGII Report - Chp 14, IPCC, 2022), as well as reports from the

EPA (EPA, 2017) and the California State Government Agencies (California Energy Commission,

2017; CalFire, 2021).41 .

The first slide of the treatment presents the IPCC work for the AR6. Slides 2 & 3 report the

conclusions about the physical reality of climate change, and changes in the climate system that

have occurred since pre-industrial times. Then, the impacts and consequences from climate change

are explained, in the Californian context (slide 4). Slide 5 presents the impacts and projected

changes in the Californian climate and weather, while the sixth slide provides information on the

most GHG-intensive sectors put forward by the scientists in the report. The slides are available in

the Second Appendix (section 13).

41
California’s Fourth Climate Change Assessment, California Energy Commission, 2018; Fire Outlook 2021, Cal-
Fire, 2021

75
12.2.3 Survey Situations

Travel

Figure 19: Travel Situation - Survey Screens

76
Energy

Figure 20: Energy Situation - Survey Screens

77
Donation

Figure 21: Donation Situation - Survey Screens

12.3 Survey Results

12.3.1 Further analysis of the distance to the fires

We investigate further the absence of effect from the distance to the fires on the outcome decisions

and beliefs. A hypothesis for this could be that inviduals in California are used to fires and that

the phenomenon does not enter their decision making as a key variable since these weather events

have historically been a part of Californian climate (although to a much less extreme extent42 ).

We know that a particularity of the 2020 and 2021 fires was the fact that some counties that were

usually spared by wildfires had been impacted. Using the dataset that records fires up to 2014, we

identify the counties in the 2020 and 2021 wildfires selected (> 300 acres) that had not been subject

to a major wildfire in the years 2014-2020. To define ”major wildfire”, we filter the observations

to keep the ones recorded as ”fire complex” which refers to the large, combined wildfires. We find

that 39 zip codes, out of the 169 selected fire zones, had previously been exposed to a fire complex,

before 2020. This means that 130 zip code areas had not been exposed to major fires prior to this

fire season. We find that individuals in areas that were previously exposed to fire complexes are

more likely to report higher levels of exposure:

42
[Link]

78
 Figure 22: Fire exposure by history

Furthermore, we can see that this effect increases as individuals experience closer fires in 2020

and 2021.

79
 Figure 23: Fire exposure by history and distance to 2020-2021 fires

We find very little effect, however on the outcome variables:

80
 This puzzling result should be further examined in subsequent research. Due to the time

constraints, this will not be a part of the scope for this thesis.

12.3.2 LASSO-Selected Covariates

The tables below provide the covariates selected by LASSO:

81
 Table 12.1: Individual Choices - LASSO selected covariates

(1) (2) (3) (4) (5)

Travel Energy Donation Beliefs Level of Concern
Level of Education (1 to 6) 0.0199 -0.0338∗ 0.0517 0.0795
(0.192) (0.044) (0.112) (0.054)
Unemployed -0.111∗
(0.025)
Fire physical exposure (1 to 5) -0.0267 0.0846∗
(0.066) (0.035)
Retiree 0.105∗ -0.0975
(0.034) (0.355)
Household (1 to 6) 0.0219 -0.0739∗∗ 0.0391
(0.153) (0.009) (0.198)
Present Bias 0.0304∗ -0.0461
(0.044) (0.127)
Average level of concern -0.126∗∗∗ 0.0597∗∗∗ 0.840∗∗∗
(0.000) (0.001) (0.000)
Fire emotional exposure (1 to 5) 0.0414∗ -0.0982∗ 0.233∗∗∗
(0.021) (0.034) (0.000)
Age -0.00321
(0.206)
Democrat Vote 0.354∗∗ 1.382∗∗∗
(0.002) (0.000)
Other Vote 0.218 1.107∗∗∗
(0.088) (0.000)
Gender 0.167∗
(0.033)
Constant 0.723∗∗∗ 0.733∗∗∗ -0.115 0.685∗ 1.735∗∗∗
(0.000) (0.000) (0.103) (0.018) (0.000)
Observations 690 467 467 467 467
R2 0.017 0.158 0.046 0.632 0.407
LPM

82
 Table 12.2: Norms & Judgement - LASSO-selected covariates

(1) (2) (3) (4)

Travel Norm Energy Norm Travel Judgement Energy Judgement
Age 0.00164∗ -0.00352∗
(0.042) (0.048)
Level of Education (1 to 6) 0.0187
(0.106)
Household (1 to 6) -0.0130
(0.219)
Fire physical exposure (1 to 5) -0.0345∗∗ 0.0319 -0.112∗∗∗
(0.004) (0.170) (0.000)
Employed -0.0705
(0.058)
Student -0.175
(0.053)
Unemployed 0.0887 0.0833
(0.145) (0.265)
Fire emotional exposure (1 to 5) 0.0385 0.105∗∗∗
(0.125) (0.000)
Present Bias -0.0362
(0.106)
Constant 0.852∗∗∗ 0.498∗∗∗ 1.831∗∗∗ 2.464∗∗∗
(0.000) (0.000) (0.000) (0.000)
Observations 690 738 738 738
R2 0.040 0.005 0.021 0.036
LPM

83
12.3.3 Summary Statistics

Round 1 outcomes:

Figure 24 below graphically represents the average individual decisions (right panel), as well as the

distribution of exposure to fire reports (left panels).

Figure 24: Additional Summary Statistics - Main outcomes and fire exposure distributions

12.3.4 Identification of the Treatment Effects

Since the outcomes are binary, we use probabilistic models to evaluate the treatment effect (TE),

such as Linear Probability Model (LPM), Logit or Probit models. Each models have their pros and

cons. Mainly, the LPM assumes a linear relationship in between the regressors and hence can predict

outcomes that are outside the [0, 1] interval, which does not make sense in the case of probabilities.

However, if it fits the data well, it is easier to interpret that non-linear models, such as Logit or

Probit. I will use a LPM going forward, providing robustness checks with a Logit model.

The LPM specification can be written as such:

P r(choice = Yi ) = α + β1 Ti + β2 Xi + εi

84
Where Yi is the outcome variable, Ti the treatment dummy, Xi a set of controls and ε the error

term that is individual specific.43 The treatment effect is identified by the parameter β1 as the

mean difference in between the control group outcome and the treatment group outcome.

12.3.5 Treatment Effect on Beliefs

We can see that the treatment has no long term effect on beliefs, which is confirmed by the following

treatment effect regression, controlling for the LASSO-selected covariates:

Table 12.3: TE on beliefs & concerns

(1) (2)
Wildfire Beliefs Average Level of Concern
Treatment 0.104 -0.0331
(0.143) (0.671)
Average level of concern 0.840∗∗∗
(0.000)
Age -0.00401 0.000484
(0.214) (0.891)
Gender 0.0205 0.171∗
(0.778) (0.032)
Level of Education (1 to 6) 0.0502 0.0916∗
(0.154) (0.017)
Student -0.0477 0.136
(0.816) (0.544)
Retiree 0.0327 -0.0942
(0.765) (0.434)
Household (1 to 6) -0.0745∗ 0.0333
(0.011) (0.297)
Fire emotional exposure (1 to 5) -0.0967∗ 0.231∗∗∗
(0.010) (0.000)
Fire physical exposure (1 to 5) 0.0819∗ -0.00536
(0.024) (0.893)
Democrat Vote 0.355∗∗ 1.377∗∗∗
(0.001) (0.000)
Other Vote 0.221 1.093∗∗∗
(0.057) (0.000)
Observations 467 467
R2 0.633 0.404
Controls Yes Yes
Note: This table presents the treatment effect on beliefs and levels of concern in round 2. All LASSO-selected
covariates were included as controls.

43
Need to assume a distribution?

85
12.3.6 Treatment Effect Heterogeneity

Common sources of heterogeneity underlying beliefs on climate change are age and gender. Figures

25 and 26 look at the treatment effects for each group on the different outcomes. We do not find a

significant pattern of heterogeneity, mostly due to very large confidence intervals for the ATEs.

Figure 25: TE Heterogeneity on individual choices by Age and Gender

86
 Figure 26: TE Heterogeneity on judgement by Age and Gender

12.3.7 Robustness Checks

We provide Logit estimated coefficients for the main outcomes treatment effects as robustness

checks for our results.44 We find that the effects are in the same direction and have the same level

of significance as the effects we find with the LPM. In comparing the marginal effects, we find the

same magnitude as LPM coefficients.

44
The controls were selected with a LASSO Logit

87
 Table 12.4: Treatment Effect on Round 1 Choices - Logit

(1) (2) (3) (4)

Travel Decision Energy Decision Travel Norm Energy Norm
Treatment -0.865∗∗∗ -0.246 -0.564∗ -0.188
(0.000) (0.312) (0.024) (0.210)
Level of Education (1 to 6) 0.00732 -0.267∗
(0.948) (0.022)
Unemployed -1.061∗∗∗ 0.141
(0.001) (0.683)
Retiree -0.245 0.495
(0.412) (0.089)
Fire physical exposure (1 to 5) 0.00636 0.0382 -0.287∗∗ 0.00115
(0.956) (0.757) (0.003) (0.986)
Fire emotional exposure (1 to 5) 0.0701 0.0689
(0.553) (0.583)
Average level of conern -0.534∗∗∗ -0.706∗∗∗
(0.000) (0.000)
Age 0.0262∗∗∗ 0.0106∗
(0.001) (0.016)
Gender -0.0164 -0.380∗
(0.947) (0.011)
Observations 433 467 690 738
Controls Yes Yes Yes Yes
Logit

88
 Table 12.5: Treatment Effect on Round 2 Choice - Logit

(1) (2)
Donation Decision Donation Norm
Treatment 0.343 -0.0645
(0.130) (0.777)
Observed 0.340 0.531∗
(0.133) (0.021)
Level of Education (1 to 6) -0.0242
(0.825)
Unemployed 0.438
(0.166)
Retiree -0.0134
(0.964)
Fire physical exposure (1 to 5) 0.0150 0.117
(0.890) (0.239)
Fire emotional exposure (1 to 5) 0.209
(0.078)
Average level of conern 0.423∗∗
(0.001)
Age -0.0155∗
(0.038)
Gender 0.0481
(0.833)
Observations 467 467
Controls Lasso Lasso
Logit

89
13 Appendix II

13.1 Treatment Slides

90
91
13.1.1 Survey Questions

ROUND 1

Screen 1: ”Hello - Thank you for participating in this survey. The estimated time to complete the

questionnaire is about 20 minutes.”

Screen 2: ”There may be some instances where the button “NEXT” is displayed with a timer:

this is not a mistake. It should work after a short period of time (5 to 30 seconds). In general,

we ask you to carefully read the instructions and pay attention to the questions asked, in order to

92
ensure the quality of the answers.”

Screen 3: We will now ask you some general questions in order to know more about you. Please

answer the following: Screens 4 - 11:

• How old are you?

• What is your gender?
• What is your county of residence?
• What is the name of your city of residence?
• What is the ZIP Code for your city of residence? [Choice list with California ZIP Codes]
• What is your highest level of education?
• What is your current employment status?
• How many people are there in your household?

Screen 12: ”We will now ask you to make some choices. Please not there are no right or wrong

answers: we want to know the choices you would really make.”

Screen 13: ”Suppose you have a chance of winning a money. If you are selected to receive it, you

will have the choice in between two payment options (A and B): option A means you would receive

the money immediately while option B is a delayed payment (5 weeks). There are 6 alternative

scenarios , and you are asked to indicate a choice in between A and B for each alternative. That is,

you need to choose in between A and B for all decision rows.”

Screen 14: ”How do you see yourself: are you generally a person who is willing to take risks or do

you try to avoid taking risks? [1 to 10]

Treated participants see the slides here

Screen 13: ”In this stage you will be asked to state your choices in various hypothetical situations.

We ask you to make a point of answering the questions in a manner consistent with the choices you

93
would’ve made in a real situation.”

Screen 14 - 16: Travel Choices - See Figure 19

Screen 17 - 19: Energy Choices - See Figure 20

Screen 20: Thank you for taking the time to participate in our survey. We have a couple additional

questions. As you may know, California has experienced recent catastrophic wildfires, the most

recent occurrences being in summer 2020 and 2021.

1 - How much would you say you were emotionally affected by the recent wildfire events in California?

[1 to 5]

2 - How much would you say you were physically/financially affected by the recent wildfire events

in California? [1 to 5]

Screen 21: Thank you for your participation, we hope you found it interesting. The aim of this

survey is to study behavioural components of environmental decisions. In particular, we want to

measure the strength of social norms related to environmental issues, and the way they evolve. You

can find additional information about climate change here if you wish to know more.

ROUND 2

Screen 1: ”Hello – Thank you for coming back to participate in the second (and last) round of

this survey. The estimated time to complete the questionnaire is about 5 minutes.”

Screen 2: ”How likely do you think the 2020 and 2021 wildfires in California are related to climate

change? [1 to 5]

Screen 3: This sections aims at understanding your level of concern about climate change.

1 - How concerned are you that climate change might cause harm to [1 to 5]:

• You personally
• People in the United-States
• Future generations
• Plants and animal species

2 - Who did you vote for in the 2018 gubernatorial election in California ? [Democrat, Republican,

Libertarian, Green, Independent, Peace and Freedom, Did not vote, Prefer not to say]

Screen 4: Donation Choice - See Figure 21

94
Screen 5: Thank you for your participation, we hope you found it interesting.

For those that answered “YES” to the donation question: You indicated you would be willing to

donate your earnings from this experiment to the NGO Friends of the Earth. You will find all the

information to proceed with the donation on this webpage. You can also find more information

about climate change here if you wish to know more.

Additional Note:

At the end of the survey, the participants were provided with a short note to explain the aim of

the study and underline the importance of collective action and structural changes; alongside the

implied individual behavioural change in the experiment.

Finally, it is important to highlight the role of individual behaviour change versus structural

societal change. In fact, the IPCC reports that lifestyle changes are a necessary condition for

mitigation: demand-side action can reduce global emissions by 50% to 80% across all sectors (IPCC

(2022) - Chapter 5). However, this number includes also a top-down approach to changing the

architecture of production and choices. The IPCC indeed highlight that “individual behaviour

change is insufficient for climate mitigation unless embedded in structural change” (IPCC (2022) -

Chapter 5, p.4). That is, individual action does not discharge large corporations, companies and

policymakers of their responsibility in operating the transition. In fact, both efforts are related

and will enhance each other: collective behaviour change and norms-shifts would provide strong

ground for policymakers and companies to change their model and adapt the structure of the

society which in turn would help behavioural change. The IPCC measures that “if 10-30% of the

population were to demonstrate commitment to low-carbon technologies, behaviours and lifestyles,

new social norms would be established” and that “the coordination of policies, choice architectures,

physical infrastructures, new technologies and related business models could realise rapid system-

level change” (IPCC (2022) - Chapter 17, p.34). This lifestyle shift is inevitable given the magnitude

of change in the social and economic structure of society that is needed, and all levers (behavioural,

economic, political) should be used.

95
13.2 RIS Approval

96