CLIMATE CHANGE
BELIEFS
What fuels motivated beliefs?
Heij, Frans de
6140033 11-7-2016
Abstract
This paper reports on the drivers of the differences in beliefs of climate change. An overview of relevant theories is discussed and a method is developed using literature. Tests are carried out to investigate correlations between the potential impact of a climate change and the beliefs in that area and the production of fossil fuels and climate change beliefs. Analyses are made using three different datasets, respectively data of US states, the EU28 and 126 countries around the world. There is some evidence a negative causal effect of the production of fossil fuels on beliefs in climate change, although this evidence is weak. There is no evidence found that changes in beliefs are influenced by climate risks. Based on this research motivational cognition plays a role in the division in climate change beliefs.
Content
Introduction... 2
Theoretical framework ... 4
The demand for motivated beliefs ... 4
The supply of motivated beliefs ... 7
Predictions on climate change beliefs ... 9
Hypotheses ... 13 Data ... 15 Methodology ... 20 Results ... 21 Discussion ... 26 Data ... 26 Further research ... 28 Conclusions ... 29 Variables ... 30 Theories ... 31 References ... 34
Introduction
Around 43% of the US does not believe in anthropogenic climate change (Yale, 2015), and in Europe this is on average 30% (Gallup, 2008). The beliefs in climate change differ between nations, ranging from 92% in South Korea to 15% in Liberia (Gallup, 2008). Remarkable, as 97% of the scientific community agrees that anthropological climate change is
happening. The scientific consensus that anthropological climate change is happening is not only found by the Intergovernmental Panel on Climate Change or IPPC (2007), but also in experts reports surveys of experts on climate change and comprehensive reviews of literature (Cook et al, 2013).
The gap between what is believed to be true by the public and the opinion of the scientific community holds mitigation efforts back (Leiserowitz, 2012). Meanwhile action to tackle climate change becomes more urgent and therefore it is important to understand how this difference emerges and holds. Solutions and implications need to be addressed publicly and globally and therefore the question is relevant why the scientific evidence is not widely accepted by the public.
Norgaard (2011) argues that most existing explanations emphasize the lack of information or lack of concern. Climate science is too complex, people have not enough access to information or media and climate skeptic programs have misled them. Or people are too self-centered, or simply busy with more immediate problems. These explanations are also referred to as the information deficit model (Bulkeley, 2000). If people knew better they would have acted differently. Kahan (2011) argues that educating people on the scientific consensus will probably not influence their beliefs in climate change. Hence this paper focusses on a different explanation, one concerning motivated cognition.
An explanation for the differences in beliefs between the scientific community and the general population is that the population holds
motivated beliefs. Underlying the distorted views which people may have on their abilities, morality and future fate, are mental processes which psychologists indicate as motivated reasoning and cognition. By such strategies people defend themselves against threatening evidence,
sometimes incurring (and inflicting) very high costs (Benabou, 2015). In case the population holds motivated beliefs, and rationalizes their beliefs by other drivers than science, this would explain the gap between public beliefs and the scientific consensus on climate change. This paper will analyze several ideas around motivated reasoning and provides insight in what is driving these differences in beliefs around climate change. The aim of this study is to get a better view on if motivational cognition is
happening and which variables are driving motivated beliefs on climate changes. The method used is to test several theories which predict beliefs on climate change with real data, provided by professional data providers like the EU, Yale, Gallup, DATASETA and the Energy Information
Administration in the US.
The following ideas will be analyzed in this paper. Benabou (2013) argues that the motivated beliefs are formed as a result of anticipatory utility. Others that the motivation is not necessarily fueled by the problem but by the solutions associated with the problem (Campell et al, 2014). Or that group identity plays a role and that people conform their ideas to the people around them (Kahan, 2012). Besides the reason why people form motivated beliefs the strategy they use to form them is also still open for debate. One might willfully ignore evidence to keep a high self-image (Grossman et al, 2016), or as Tremewan (2009) argues, when people come to a conclusion they are influenced by the desire to come to a particular solution, a directional goal, and a desire for their conclusion to be justified by evidence, an accuracy goal.
The main focus of this paper will be to link two variables to the beliefs around climate change. One is the income of production of fossil fuels.
Based on the self-image protection model and the solution aversion model this variable should negatively influence beliefs in climate change. The other variable is climate impact risk and this paper researches how that influences climate change beliefs. Climate impact risk is a variable that influences anticipatory utility. As you are more at risk, hence have a worse looking future, anticipatory utility is stronger.
The outline of the paper is as follows. First, the ideas mentioned above will be further explored, then the methods and data used will be elaborated on, followed up by the results and discussion. The conclusion elaborates on the findings and their implications. Three different datasets are used in this paper, one about the US states, one containing the EU28 and one containing 126 countries all over the world. The content for these datasets are delivered by different agencies.
Theoretical framework
There are multiple explanations for why and how people form motivated beliefs. It is likely that there is not just one model or theory that can explain this phenomenon; a mixture of processes should be taken into account. In thinking about motivated beliefs it is useful to separate the demand side from the supply side. The demand side is about why people want to hold, or are drawn to distorted beliefs. The supply side focusses on how people manage to hold such beliefs. This chapter is divided in three sections. First, the demand side of motivated beliefs is introduced. The second chapter reviews the supply or the forming of these beliefs. The third chapter lists the implications for the beliefs around climate change, based on the predictions by the presented theories.
The demand for motivated beliefs
The demand for motivated beliefs is the reason why people want to hold alternative beliefs, which may be caused by different sources. This paper explores in particular theories explaining group behavior. For instance
people want to adapt their beliefs to their surroundings (Kahan et al, 2012), or they want to protect a high self-image (Grossman et al, 2016). People might experience anticipatory feelings, as described in the
groupthink model by Benabou (2013). Another theory is the solution aversion model, in which the solutions associated with the problem are aversive and therefore the whole problem becomes aversive (Campbell et al, 2014). Other more individual focused explanations, which will not be analyzed in this paper, consider conspirational mindsets, self-serving biases, backwards rationalization and scientific illiteracy.
Conformism
Kahan et al (2012) argue that the misbeliefs around climate change are not due to a deficit of comprehension, but that conformism plays a role. In their study they found that people with the highest degrees of science literacy are not the most worried about climate change. Instead they are amongst the most divided and in this group cultural polarization is the strongest. This result suggest that the differences in beliefs are not due to a lack of comprehension but are caused by a conflict of interest.
Individuals have a personal interest to form beliefs in line with those around them, while the collective interest to promote welfare would be to make use of the best available science. Kahan et al refers to this
alternative explanation as the cultural cognition thesis (CCT). CCT posits that individuals, as a result of a complex psychological mechanism, tend to form beliefs of societal risks that cohere with the characteristics of the groups with whom they identify (Kahan, 2009). Members of groups are motivated to fit their interpretations of scientific evidence to the
philosophy of their group. Anticipatory utility
The term groupthink was first brought up by Janis (1972) when studying policy decisions such as the Cuban missile crisis and the escalation of the Vietnam war. He identified a cluster of symptoms, including collective
denial, which he called groupthink. Benabou (2013) investigated the
collective denial and wilful blindness in groups, organizations and markets. He assumes that agents with anticipatory utility choose how to interpret information signals about future prospects. Anticipatory utility is takes next periods’ utility into account for the utility in the current period. For instance a person might already feel worse because he knows tomorrow is going to be a bad day.
Assume that there are two possible states for the next period, a High and a Low state (H/L). The High state is a state in which everything is fine and business as usual works, and the Low state is when business as usual is harming the group as a whole. Harmful group delusions may happen when the L state is negative, that is when the L state is true it harms everyone. People who do not recognize the reality of the L state and do not change their behaviour make things worse for everyone. Hence the more people deny the L state the worse the L state gets, making it more painful to acknowledge the L state. This is the ‘Mutually Assured Delusion’ (MAD) effect (Benabou, 2013). The stronger the MAD effect the higher the desire to motivate the beliefs.
Self-image protection
Grossman and Van der Weele (2016) argue that an individual cares about his self-image and has the opportunity to learn the social benefits of his actions. Wilful ignorance is a tool that can be used to justify less social behaviour, by staying uninformed about the consequences a person can keep the image that one would have acted differently when consequences would have been known. An individual has two active agents, an observer and a decision maker. The decision maker is aware of the person’s
preferences and acts upon them. The decision maker aims to impress an ‘uniformed observer’ who lacks introspective knowledge of the person’s preferences. An equilibrium can form where selfish people choose to remain ignorant in order to maximize both their self-image and material
rewards. In other words, one might want to look the other way to maintain a high self-image.
Solution aversion
Campbell and Kay (2014) argue that people hold beliefs which are
inconsistent with what science tell us, because people believe the actual problem but do not like the related solutions. The solution aversion model predicts that certain solutions associated with problems are more aversive and more threatening to individuals who hold an ideology that is
incompatible with or even challenged by the solution, and this increases scepticism of the problems’ existence. In an experiment they conducted they would give different solutions for different groups, one which would be more and one less compatible with Republicans ideology. They found that having a more compatible solution has as a result more people
acknowledging the problem. Hence people motivate their beliefs not based on information but on the implications the beliefs have.
The supply of motivated beliefs
Turning now to the supply side, how are desired beliefs achieved and maintained, sometimes against strong evidence? And what is the role of the media and politics in the supply of motivated beliefs? Goal directed beliefs is the collective name for several ways of motivated reasoning. Two main mechanisms are discussed below, one in which one abstains from information and one in which one suppresses information.
In his paper about the beliefs around immigration Tremewan (2009) developed a simple model on belief formation based on the concept of motivated reasoning. Coming to a conclusion people are influenced by the desire to arrive at a particular conclusion, a directional goal, and by the desire for their conclusion to be justified by evidence, an accuracy goal. Motivated reasoning is a well-known psychological phenomenon,
“… People motivated to arrive at a particular conclusion attempt to be rational and to construct a justification of their desired conclusion that would persuade a dispassionate observer… In other words they maintain an ‘illusion of objectivity’. To this end they search memory for those beliefs that could support their desired conclusion”
One way to hold motivated beliefs is by staying uninformed about a
certain topic. Benabou (2013) uses the term wilful blindness which consist in avoiding information sources that might hold negative news. Grossman and van der Weele (2016) explain the term wilful ignorance as deliberate avoidance of evidence about the social impact of one’s decisions. They set up an experiment which shows that wilful ignorance can serve as an
excuse for selfish behaviour. Dana et al (2007) showed something similar in an experiment where participants were reluctant to learn how their choices affect others, even if that information can be obtained without cost. Participants behaved different to hidden information and freely available information. These paradoxical results demonstrate that people cultivate uncertainty about the social outcomes of their actions in order to justify self-interested decisions (Heffernan, 2012). Later in this paper a link is established between online search engine behaviour regarding climate change and climate change beliefs.
When a person is knowledgeable on his actions one can suppress those signals. Self-signalling is about receiving the information and signalling oneself about it. Benabou (2013) states that when self-relevant beliefs are involved, an individual will tend to process good and bad news differently, trying to ignore, discount, rationalize away or put out of mind those he does not like. The subject can find ways of not internalizing the
information, rationalizing it away and convincing himself that the evidence is not true. Surpassing denial is a form of self-signalling, where the
subject manufactures a desired signal himself which is subsequently interpreted as impartial (Quattrone and Tversky, 1984).
As media and politicians are a major source of information to the public, the information they provide is essential for the forming of motivated beliefs. In the US several prominent Republicans have discarded climate change beliefs, which influences the public in their belief formation. The same phenomenon goes for media, one might only seek those media which confirm individual believes, resulting in motivated beliefs. A person may only seek specific media which do not cover certain topics at all. Media and politician might also have an agenda to promote motivated beliefs to the public for their own benefit. For example the lobby groups that are active around the politicians who hold agendas in favour of the fossil fuel industry and try to influence the beliefs people hold.
Predictions on climate change beliefs
In this section first the psychological properties of climate change are briefly discussed. Subsequently, the demand theories as explained above will be analyzed in order to predict about the beliefs in climate change. Properties of climate change
The effects of climate change are surrounded by uncertainties. Climate change is not a hazard per se, but a potential driver for many different hazards. Due to geographic variations and variations in capability to adapt are climate change driven hazards place-specific. Climate change is also accelerating and does not necessarily follow a linear trend. While there are current effects which are urgent and consequential, the most serious
impacts will come in the ‘far future’, beyond the planning horizon of most individuals. Also without climate changes triggered by human behavior, the world is (always) changing, making anticipations even more difficult (Swim, 2009).
As climate change is nearly impossible to detect from personal experience, it makes sense to leave this task to climate scientists. Personal
experiences will not provide trustworthy data and therefore climate
change is a phenomenon where people have to exert judgment based on reports in the media. Hence, most people experience climate change
almost entirely indirectly and virtually, brought by news coverage of distant events, such as the melting of glaciers on the Artic (Swim, 2009). For most people in the US, perceptions of the risks of a climate change that rely on personal experience will lead to the judgment that the risks are low. Even people working in weather and climate dependent jobs such as farmers or fishers might not get enough feedback from their experience to be alarmed about climate change. However a survey shows that
exposure to climate change raises increased concerns and willingness to take action (Leiserowitz & Broad, 2008).
Another important aspect of people’s reaction to a climate change is that it needs global cooperation to mitigate the problems (IPCC, 2007). An individual or group can barely mitigate any of the effects of climate change and negative spill overs are common. When people believe that they have no control over climate change, it facilitates negative emotions and mechanisms such as denial (Gifford, 2008).
Anticipatory utility
Benabou’s model predicts that when the Low state is negative, is harmful, the MAD principle (page 6) can occur and by denying a climate change an equilibrium is achieved. So when enough people deny a climate change and business goes on as usual, an individual may, for personal interest, deny a climate change as well. When the timeframe in which the
anticipatory feelings are felt is very long, as is the case in climate change, it becomes more attractive, and yet harder to deny the Low state. Also when the majority does not act like the Low state is happening one’s personal preventive measures might be in vain. A person may want to stop taking measures as it decreases his utility and has close to zero effect.
In this paper the variable climate risk represents an indication of the exposure a certain area has to climate change. Following Benabou’s theory two opposite forces are active in the formation of climate change beliefs with respect to climate risk. If climate risk increases, anticipatory
utility decreases and denying climate change becomes more attractive. On the other side being exposed to higher risks makes it more costly to
ignore the signals and people might be more willing to accept a climate change in an attempt to mitigate the effects. Hence, based on anticipatory utility, it is not clear what the effect of climate riskwould be. However as Gifford (2008) argued that the effects of a climate change become more overwhelming and harder to mitigate, the more mechanisms such as
denial will be facilitated. Summarising, a higher exposure to climate risk is expected to lead to more denial.
Conformism
Kahan et al. states that motivated beliefs are formed according to the group people associate themselves with. People form their belief about climate change based on the beliefs held by the group with whom they identify. In the case of Republicans and Democrats that means that if you are an Republican you might motivate your belief into denying the
existence of climate change. Identification may also work for other group stratifications, such as gender, age, income or levels of education. This may lead to assumptions that younger people believe more in a climate change then elderly. People tend to follow the beliefs held by their social environment.
The research presented in this paper includes variables for political affiliation, as well as the variables education and age. Based on Kahan’s theory political affiliation is expected to correlate with climate change beliefs, ranging from acceptance of climate change on the left hand side of the political spectrum to climate change denial on the right hand side. Hornsey et all (2016) showed that political affiliation is indeed a significant predictor of climate change belief.
Self-image protection
As climate change is a global problem its proposed solutions to mitigate climate change often follows the adage ‘the polluter pays’. Reducing their CO2 footprint people have to take personal costs. In case of more ties
with fossil fuels, these personal costs will increase. To avoid such
additional costs, it might be beneficial for an individual to ‘look the other way’ and to stay wilfully ignorant about climate change. Hence this theory predicts, as does the solution aversion model, that in areas with stronger ties to fossil fuels production people will believe less in a climate change. This theory also predicts that those who deny climate change will either avoid information or only pick up information which doubt climate change, in order to keep a high self-image. This last statement will be briefly
reflected in the chapter on discussion. Solution aversion model
The solution aversion model predicts that people’s believe in a climate change is influenced by the possible solutions associated with a climate change. In an experiment Campbell and Kay (2013) changed the
proposed solution for climate change and noticed a significant difference in the acceptance rate of an IPCC-statement. This predicts that among those who dislike governmental interference the beliefs in climate change should be lower, as is the case with the Republicans. Also those producing fossil fuels are more likely to dislike solutions associate with climate change and therefore they will believe less in climate change. This predicts a negative correlation between belief in climate change and the production of fossil fuels.
To conclude these theories predict that denial of climate change will happen more frequent in areas with high production of fossil fuels and in areas that have a lot of exposure to climate change. Besides those
variables it is evident that political affiliation has an impact on climate change beliefs; a more left political view will correlate with believing more in climate change. Hence, next to looking into control variables, this
research will look into the dependent variable of ‘belief in climate change’, and two explanatory variables, respectively climate risk and production of fossil fuels.
Hypotheses
The focus of this study is on several variables and their impact on climate change beliefs. The variables of interest are the production of fossil fuels and the climate impact risk. Theory predicts that the production of fossil fuels correlate negatively with climate change, as does climate impact risk. Three different datasets will be used to calculate these correlations. After that this paper will try to go further and establish causality besides the correlation for the fossil fuel income. This approach leads to three hypotheses, two focusing on the correlation and the other on causality. The correlation hypotheses
The first hypothesis is that there is a negative correlation between production of fossil fuels, and the beliefs in climate change on an aggregate level. This implies that the coefficient of the variable fossil production in both model 1 and 2 will be negative. In the theoretical framework the production of fossil fuel was identified as a possible
explanatory variable for believe in climate change. The solution aversion and self-image protection theories suggest that the more fossil fuels are produced in a person’s area the less such a person believes in a climate change. The following hypothesis is set to test if people do indeed believe less in areas with high fossil fuel production.
H0: There is no correlation between climate change beliefs and local production of fossil fuels.
H1: There is a negative correlation between climate change beliefs and local production of fossil fuels.
The second hypothesis is that there is a negative correlation between climate risk and beliefs in climate change on an aggregate level. This
implies that the coefficient of the variable climate risk in the regressions of model 2 is negative. The anticipatory utility theory predicts that people in those areas are more at risk people and therefore have a larger incentive to deny climate change, and as the potential impact will increase, the
denying will become harder. This means that the anticipatory theory does not give a clear prediction on how the impact of climate risk correlates to climate change beliefs, but it does suggest that a correlation will be there. Following Gifford (2008) a negative correlation is expected between
climate change beliefs and climate impact risk.
H0: There is no correlation between climate change beliefs and climate impact risk.
H1: There is a negative correlation between climate change beliefs and climate impact risk.
Both hypotheses are tested using the three data sets; the correlation is tested on worldwide country level, US state level and EU28 country level. The causality hypothesis
The causality hypothesis formulates that the production of fossil fuels has a negative causal effect on climate change beliefs. If there is a correlation between the production of fossil fuels and people’s beliefs in a climate change, it is relevant to gain insights in the understanding of the causality within in this relation. The solution aversion theory predicts that producing more fossil fuels will lead to less belief in climate change. However it is also possible that believing less in climate change increases the production of fossil fuels. This research will try to find evidence for a negative causal effect of the production of fossil fuels on climate change beliefs.
H0: There is no causal effect of the production of fossil fuels on climate change beliefs.
H1: There is negative causal effect of production of fossil fuels on climate change beliefs.
The causality hypothesis will be tested with one data set, containing data on US state level. To test this the increase in production of fossil fuels due to the fracking revolution will be used as an exogenous variable, which will be further discussed below. In short, some states produced near zero in 2010 and a significant amount in 2015, how did this alter the beliefs compared to states where production remained constant
Fracking revolution
Innovations in hydraulic fracturing and horizontal drilling (often collectively referred to as “fracking”) have produced a technological revolution in natural gas and oil extraction. The United States, the world leader in these technologies’ development and exploitation, has suddenly returned to the role of energy-producing superpower (EIA, 2013). While some states forbid fracking due to local environmental problems, some states have massively increased their fossil fuel output between 2010 and 2014. Especially the fact that some states forbid it and other did not
makes this variable useful for researching if there is an causal effect of production of fossil fuels on climate change beliefs. This paper will test if states where the production of fossil fuels changed significantly have changed their believes differently compared to states where the
production remained constant. One can arguably use this as an exogenous variable to test this causality. However it also has shortcomings, such as reverse causality, which will be further discussed in the discussion.
Data
In this section, first the variables, the data and the data sources will be briefly discussed. The dependent variable, belief in climate change, is
discussed first and subsequently the explanatory variables are reviewed: climate risk and income from fossil fuels. After that this report zooms in on the control variables. The whole dataset and the Stata14 do file is
provided as an attachment to this paper. As this paper uses three different datasets the variables are not exactly the same across the datasets. The US data is collected from 2014, the EU dataset from 2013 due to
availability of data. The World dataset used data combined from different years ranging from 2008 to 2013, due to missing data.
Belief
This paper categorizes people to be believers in anthropologic climate change, when they believe that climate change is happening and that human activity makes a significant contribution. So if individuals believe that climate change is happening but that it is not caused by human activity they are labelled as non-believers or climate change deniers. The dependent variable is ‘belief’, for which different data sources are used. For the US estimated data produced by Yale (2014) are used. For each state the percentage of people is estimated who believe that climate change is at least partly anthropological. For the EU data from the special ‘Eurobarometer reports’ is used. The euro barometer is a survey
consisting a wide range of questions and conducted by the EU in all the countries, specifically edition 409 is used as this edition is dedicated to questions around climate change. In this survey Europeans are asked to rank how serious of a problem climate change is from their point of view on a scale from 1 to 10. For the EU-dataset this paper takes those who believe that climate change is a serious problem as those who believe in anthropological climate change. For the World-dataset the Gallup (2008) survey data is used for belief. Those participants of the survey who said to believe that climate change is anthropologically caused are considered as believing in climate change. The beliefs of 2010 US citizens are extracted from a dataset of DATASETA to fill the US-dataset with 2,006
observations. In the DATASETA survey people answered the question whether they believe a climate change is happening and whether this
change is mainly anthropological driven. States representing less than 25 observations are dropped from the data. The online sources are added as an attachment to this paper.
Production of fossil fuels
Three types of fossil fuels are part of this study, namely gas, oil and coal. The production of each type of fossil fuel is for each country or state
multiplied by the local import price of that fuel type. The local import price is used as it is arguably the best proxy for how much a region would have to pay to import those fossil fuels, and hence gives the best approximate value of the production. For the US-dataset the information was extracted from website of the Energy Information Administration (EIA). For the EU- dataset the data was conducted based on data of Eurostat. The fossil fuel income variable is total production of the year of each fuel, 2014 for US and 2013 for EU, multiplied by the average import price and then put together. For the World-dataset four different prices are used, respectively one for Europe, North America, Africa and Asia. To generate a per capita variable is the result divided by the population of the relevant region. The log of that income is used to reduce the extent of the right tail, in order to achieve more representative results.
Climate risk
The variable of ‘climate risk’ represents the potential impact which climate change can have on a certain area. Based on the anticipatory utility
theory, the variable climate risk is added as well. An increase in climate risk will reduce the anticipatory utility and denial might become more attractive. On the other side, if the future will be worse, it is harder to deny climate change. Due to differences of the available datasets, two different approaches are used for the US and EU/World. This part of the US-dataset is retrieved from States At Risk, in which states get a grade based on their potential impact of and their preparedness for a climate change. The impact is based on climatological threats such as droughts or massive rainfall. Preparedness is about how well the problems can be
mitigated with resources like money or city structures. These grades are used as a numerical variable, so five categories going from A to F,
converted for this research into a 1 to 5 scale, where 1 corresponds with A and 5 with F. Having a lower grade indicates that a citizen is exposed more to risk. Data from the Centre for Global Development is used for the EU- and World-dataset. The Centre for Global Development mapped the potential impact of climate change for almost all countries in the world. The impact index is the quantitative value used in the EU-dataset, which ranges from minus 10 to plus 100; the higher the index the more a country is at risk. Also for this variable, like the others, new data is not created, but data is used from datasets mentioned.
Control variables
Hornsey et al (2016) conducted a meta-analysis of the determinants of belief in climate change. They reported significant correlations between belief in climate change and several demographic variables, as listed in table 1. The results of this study are used as control variables.
Demographics Correlation P-value
Sex (male = 0, female = 1) 0.029 0.001
Age -0.125 0.003 Income 0.057 0.000 Education (years of schooling) 0.117 0.002 Race (white = 1, non-white = 2) 0.032 0.000 Political affiliation
(Higher scores represent more ‘left-wing’)
0.301 0.004
Political ideology
(Higher scores represent more ‘left-wing’)
0.149 0.015
Sex and race have a significant, but low correlation to climate change, and are therefore kept out of the control. Keeping gender out of the control is no problem, as the gender division is close to 50 %, so male and female are both about equal represented. As for race, the amount of immigrants could be included, but this would probably not sufficiently cover the
variable of an aggregated belief. Thus four control variables remain: Age, Income, Education and Political affiliation.
Income
For income this paper used the GDP per capita of the appropriate year. The data on income of the US-dataset was retrieved from the US Census Bureau and for the EU- and World-dataset data provided by the World Bank is used, all expressed in US dollars. And as is the case with the fossil fuel income the natural log is taken to reduce the right tail.
Education & Age
In the case of education different proxies had to be used within the different sets. In the US-dataset education is the percentage of the population finishing a bachelor degree, data obtained from US Census Bureau. For the EU-dataset the percentage of the population that has finished secondary schooling programme is used, data retrieved from Eurostat. In case of the World-dataset the average years of education is used, data collected from the World Factbook. A similar proxy is applied in all three sets for the variable age, the median age of the state or country. Political Affiliation
In political affiliation this paper makes a distinction for the US in five categories, namely 1) Solid republican, 2) Lean republican, 3)
Competitive, 4) Lean democratic and 5) Solid democratic. Based on these categories each state received a value ranging from 1 for solid republican to 5 for solid democratic. The data is collected from Gallup. Based on survey results considering questions about political affiliation, states were rated from solid republican to solid democratic. In the EU a government
usually represents a more pluralistic political landscape, making it more difficult to estimate how left or right a government. Therefore, in case of the EU, three categories are used: left, liberal and right. The attributed regression values are respectively 1, 2 and 3. The data is retrieved from Guardian Research (2013). The Gaurdian mapped the political affiliations of the EU28 based on the head of the country’s government, the
representatives in the EU Parliament and the European Council. In the World-dataset the variable political affiliation is not used. It is too complex and out of the scope of this study to label governments around the world in a simple left, centre or right category.
Methodology
To test the correlation hypotheses multiple regressions are run for all three datasets. Testing solely the correlation between explanatory variables and the climate change beliefs without using control variables would give biased results. Hence, regressions are presented using all control variables introduced. The regressions are conducted in Stata14. The input data for the variables has been described at the data section. Two correlation hypotheses have to be tested and therefore two models are used, where the second model is an extension of the first model. Since it is not evident that climate risk has indeed an impact on climate change beliefs, climate risk is absent in the first model and added in the second one. With the two models the effects of the two explanatory variables, fossil income and climate risk, are both tested, together and separately. Using two models also reduces the possibility of overfitting; the discussion section will provide more information on possible overfitting.
Model 1
𝑩𝒆𝒍𝒊𝒆𝒇 = 𝜷𝟎+ 𝜷𝟏𝑳𝒐𝒈(𝑭𝒐𝒔𝒔𝒊𝒍𝑰𝒏𝒄𝒐𝒎𝒆) + 𝜷𝟐𝑳𝒐𝒈(𝑰𝒏𝒄𝒐𝒎𝒆) + 𝜷𝟑𝑬𝒅𝒖𝒄𝒂𝒕𝒊𝒐𝒏
FossilIncome is in Model 1 the predictor variable, which is the income gained from the production of fossil fuels in that area. The other variables are control variables, which are taken from the work of Hornsey et al (2016), reporting on significant correlations between climate belief and several demographic variables. The underlying data for the variables are described in more detail in the data section earlier in this report.
Model 2
𝑩𝒆𝒍𝒊𝒆𝒇 = 𝜷𝟎+ 𝜷𝟏𝑳𝒐𝒈(𝑭𝒐𝒔𝒔𝒊𝒍𝑰𝒏𝒄𝒐𝒎𝒆) + 𝜷𝟐𝑳𝒐𝒈(𝑰𝒏𝒄𝒐𝒎𝒆) + 𝜷𝟑𝑬𝒅𝒖𝒄𝒂𝒕𝒊𝒐𝒏 + 𝜷𝟒𝑨𝒈𝒆 + 𝜷𝟓𝑷𝒐𝒍𝒊𝒕𝒊𝒄𝒂𝒍𝑨𝒇𝒇𝒊𝒍𝒊𝒂𝒕𝒊𝒐𝒏𝑱+ 𝜷𝟔𝑪𝒍𝒊𝒎𝒂𝒕𝒆𝑹𝒊𝒔𝒌 + 𝜺
In model 2 the same regression is run as in model one with one add-on which is the variable climate risk. In the EU- and World-dataset this is a quantitative value, in the US-dataset it is a numerical variable.
A different approach is used for the second hypothesis, the causality hypothesis. In this case the dataset includes beliefs in climate change in the US on state level in the years 2010 and 2014. However, due to a lack of available information the data includes 36 states. Within this data two groups are created, one for those states that lowered or kept their
production of fossil fuels constant between 2010 and 2014, and in the other group the states increased their production in that same period. This led to two groups each consisting of 18 states. For these groups an average of the shift in climate change beliefs between 2010 and 2014 is conducted. These averages are compared with each other in order to verify whether there is a significant difference when using an unpaired T-test. The causality hypothesis predicts that the group containing those states that increased their production of fossil fuels will have lowered their beliefs compared to the states that lowered their production.
Results
In this section the results from the regressions and T-test will be given. Both models are regressed with the data of all three sets. First the
characteristics of the datasets will be given, including the correlations of the variables with climate change beliefs, followed by the results of the regressions for model 1 and 2. This chapter ends with the result of the unpaired T-test for comparing means.
Characteristics
This section reports for each dataset the basic characteristics of the variables, including the correlation of the four variables with the variable belief is given. The characteristics of the US-dataset are reported in Table 2, the EU-dataset in Table 3 and the World-dataset in Table 4. Each table presents a correlation between the variables and a climate change belief. This correlation provides a first idea of the influence of the variable on climate change belief, the control variables are not yet applied. The correlations are calculated with Stata14.
Variable #obs = 48
Mean Min Max Correlation
with Belief
Belief 47.96 42 58
Income fossil fuels (log) 3.915 0 10.82 -.4749*** Income (log) 10.25 9.954 10.58 0.5507*** Education 27.17 17.3 38.2 0.6659*** Age 38.06 30.5 44.1 0.2550** Climate risk 1 = A, … , 5 = F 3 1 5 -0.416*** Political affiliation 1 = solid republic 5 = solid democratic 2.86 1 5 0.8705***
Table 2, characteristics of the US-dataset
Variable #obs = 26
Mean Min Max Correlation
with Belief
Belief 91.31 81 97
Income fossil fuels (log) 2.999 0 7.464 -.3173 Income (log) 10.40 9.660 11.42 -.2847 Education 77.33 42.2 93.3 -.4011** Age 41.36 35.7 46.1 0.2725 Climate Risk 2.595 -6.417 7.146 0.4307** Political affiliation 1 = right, …, 3 = left 1.923 1 3 0.3809*
Variable #obs = 123
Mean Min Max Correlation
with Belief
Belief 55.68 15 92
Income fossil fuels (log) 4.004 0 11.44 0.3230 Income (log) 8.652 5.489 11.62 0.2930***
Education 7.885 1.3 12.9 0.2069**
Age 29.30 15.1 51.1 0.2636***
Climate Risk 19.58 -9.417 100 -.1789**
Table 4, characteristics of the World-dataset
As seen in the tables with the characteristics of the dataset the beliefs in climate change are higher in the Europe dataset. Also the correlation with beliefs are all significant in the US dataset and only education, in contrary to what Hornsey (2016) found, negative, and climate risk are significant. In both the EU and the US dataset political affiliation significantly
correlates with beliefs, in the US dataset the correlation is very strong with 0.87. The World dataset has some remarkable properties. Income from fossil fuels increases beliefs contrary to theory and the other
datasets, but is not significant. GDP and education behave as expected, and are significant. Age is positive, while Hornsey showed that younger people are more likely to belief in climate change. Climate risk has a negative correlation with beliefs, contrary to the EU dataset. The US dataset fits the predictions the best.
US-dataset
Table 5 presents the regression output of the US-dataset for Model 1 and Model 2. Political affiliation has a significant impact on the beliefs of
climate change. The more democratic a state, the more their citizens believe in a climate change. Fossil fuel income has a negative coefficient, as predicted by the hypothesis, but insignificant. The other variables push belief in the direction as expected, but not significantly. The same is true for income (increasing beliefs), education (increasing beliefs) and age (decreasing beliefs). Adding the variable for climate risk the model as a performs the same, hence the adjusted R² remained constant. Climate risk is very insignificant and does not add any value to the model.
Variable Model 1 Model 2
Constant 37.59
(25.09)
38.03 (26.62)
Income fossil fuels (log) -0.076
(0.068) -0.076 (0.069) Income (log) 0.3533 (2.507) 0.320 (2.635) Education 0.164 (0.103) 0.164 (0.103) Age -0.062 (0.094) 0.062 (0.095) Political affiliation
Ranging from 1 (right) to 5 (left) 1.606*** (0.232) 1.602*** (0.232) Climate risk
Ranging from 1 (low) to 5 (high) -0.016 (0.243)
Adjusted R-squared 0.827 0.827
# observations 48
Table 5, regression output US-dataset
EU-dataset
Table 6 presents the regression output of the EU-dataset for Model 1 and Model 2. In the EU-dataset model one does not have a lot of explanatory value, with only the constant and political affiliation having a significant effect. Also the other variables do not push belief in the direction expected from theory. For instance education seems to have a negative impact on beliefs. The political affiliation variable is significant, indicating that left orientated countries have a stronger belief in climate change. When
adding climate risk for model 2 the explanatory value increased. In model 2 the coefficient of income from fossil fuels is negative and significant at a 10% level, as predicted by the hypothesis. Also the variable climate risk is significant and increase beliefs, contrary to the predictions and
hypothesis.
Variable Model 1 Model 2
Constant 117.6***
(21.32) 89.65*** (17.27) Income fossil fuels (log) -0.416
(0.311) -0.546* (0.270) Income (log) -3.251* (1.665) -1.131 (1.433) Education -0.086 (0.051) -0.058 (0.041) Age 0.306 (0.251) 0.361 (0.217) Political affiliation
From 1(right) to 3(left)
1.456** (0.660)
1.513** (0.615)
(0.274)
Adjusted R-squared 0.454 0.614
# of observations 26
Table 6, regression output EU-dataset
World-dataset
Table 7 presents the regression output of the World-dataset for Model 1 and Model 2. As reported before, political affiliation is missing (page 20). The R-squared is very low in both models and therefore explanatory power is lacking. Adding the variable climate risk did not help the model. Only the constant and the GDP per capita are significant. Income from fossil fuels has a negative coefficient, in line with theory and as predicted by the correlation hypothesis.
Variable Model 1 Model 2
Constant 25.12**
(11.09) 25.86** (12.77) Income fossil fuels (log) -0.323
(0.498) -0.324 (0.500) Income (log) 3.656* (2.120) 3.623* (2.183) Education -0.794 (1.025) -0.820 (1.083) Age 0.222 (0.322) 0.222 (0.324) Climate Risk -0.010 (0.076) Adjusted R-squared 0.094 0.095 # of observations 123
Table 7, regression output World-dataset
Causality
The beliefs difference in climate change in the US between 2010 and 2014 have a mean of -0.029, a minimum of -15.4 and a maximum 23. Hence, the average believe in climate change decreased by 0.029% between 2010 and 2014. The data contained 36 states, the remaining states did not have enough data available. Means are set up between those states that increased their production and those that remained constant or lowered their income from production of fossil fuels. The mean of those that increased production is -3.37, 18 observations, and of those that did not is 3.31, again 18 observations. To clarify this, the states in which
production of fossil fuels decreased between 2010 and 2014 have
increased their belief in climate change on average by 3.31%. While the states that increased their fossil fuel production between 2010 and 2014 have on average decreased their beliefs by 3.37%. Comparing means with a unpaired T-test gives a t-score of 2.426. With 32 DF that gives an one-sided P-value of 0.01, hence a significant result that shows that an
increase in production of fossil fuels leads to a relative decrease in climate change belief.
A possible problem in this analysis is that there is a selection bias based on the data available. It is possible that for instance those states that where dropped where mainly republicans states or vice versa. To mitigate this problem the differences in means for the two most significant
variables are tested, namely the political affiliation and the income from fossil fuels. The states dropped have a mean in political affiliation of 2.5 compared to 3.0 in the remaining states. A T-test for comparing means shows that this difference is not significant (P = 0.36). The fossil fuel mean is 5.2 of the dropped states compared to 3.36 in the remaining states. Again a T-test is conducted to check if this difference is significant, and the results show that it is not significant (P = 0.14). Hence there has not been a selection bias.
Discussion
This section on discussion will elaborate on several points of the paper, introduce other possible explanations and start with suggestions for further research.
Data
One of the major problems faced in this study had to do with the collection of reliable and representative data. For the US-dataset this paper used the estimations of the Yale climate research group. These estimations have been tested in the real world (Yale, 2014). However, one of the problems is that the estimation is based on for instance age,
political affiliation, education levels and other characteristics. Prior to
doing the research one may foresee that those variables used to make the prediction will have a correlation with the estimated beliefs. However as the estimations have been tested in the real world extensively by Yale I believe that it does not undermine this study. Future studies, it seems to me, will be more conclusive when reliable and representative survey data will be used.
Another problem relates to the EU-dataset, where there was missing a solid database on the beliefs on climate change in the EU. For several countries or groups of countries data was available but not for the whole of the EU28. The Eurobarometer used in this research was the only available conclusive dataset with an approximation of beliefs in climate change. However the question answered in the survey was to rank how serious of a problem climate change was in their perception. This is obviously not exactly the same as asking a question like ‘do you believe that anthropologic climate change is happening?’ However I believe that it gives a valid approximation of the climate change beliefs in the specific country.
With the World-dataset there were multiple serious problems. One was a problem of including every nation. With using the Gallup database from 2008 only 128 nations were included. Another problem was with the
pricing of the production of fossil fuels. Where the production is pretty well recorded it was almost impossible to find for every country the specific local import price. I used four prices and tried to give every country a price that best fitted their location. However there can be a huge
difference in prices between neighbouring countries. Yet another problem with the world dataset was that it is out of scope of this research to label every government into right/centre/left. Yet this variable, political
affiliation, was found the most significant in both the US and EU dataset. So for further research the labelling of the governments would be a good next move.
Another problems is the limited sizes of the datasets. Regressions done with a small number of observations can give biased results. One of those is overfitting, in which you try to fit too many different variables on a small sample. As this paper had to use several control variables and the datasets where limited in observations it is possible this occurred.
However using the rule of thumb stating that you should at least have 10 observation for every included variable the US and World-dataset are large enough. However overfitting might be a serious problem in the EU-dataset.
Further research
This research searched to find evidence that climate change beliefs are not due to lack of information but due to motivated cognition. However it is possible that there are different explanations to explain the difference in climate change beliefs. The information deficit model by Bulkely (2000), or more individual orientated explanations such as the self-serving bias, backwards rationalization or scientific illiteracy. However the information deficit model, and scientific illiteracy, has been criticized for instance by Kahan (2011). Self-serving bias and backwards rationalization probably happens but this does not focus on the source and why people form those beliefs. Another reason might be that as climate change becomes more urgent people take it more seriously and will update their beliefs more truthfully. This theory suggest that motivated cognition happens more when the problem is distant enough so that the distorting of ideas does not affect your utility to much yet. It is true that awareness about climate change is increasing over the world (Gallup), however awareness is not the same as believing in climate change. In the US people seem to be more divided on the topic. Also in this research 0.03% more people believed in climate change in 2010 than in 2014 in the US.
This paper has looked at the correlation between belief in climate change and fossil fuel income, climate risk and control variables. However this correlation does not say anything about causality. It is both possible that
those countries or states that believe less in climate change choose to take less safety measures, or that those countries that are more safe from climate change impact believe less in climate change. One could try to find an exogenous change in climate change risk, as tried in this paper with the fracking revolution. To analyse this a more comprehensive dataset is necessary.
Another part that is important to look into is to distinguish more sharply between the solution aversion model and the self-image protection theory. While doing this research I also looked into this and used data on search engine (Google.com) behaviour for the US states. There is a very strong correlation, 0.84 (P-value = 0.000) between average searches per capita and the beliefs in a state. The search-terms I researched where ‘climate change’ and ‘global warming’ in the period May 2015 to May 2016. This makes an argument for the self-image protection model, as people protect their self-image by staying uninformed about climate change. However causality is not proven and it is also possible that those who search
information about climate change update their belief. To distinguish even better between the solution aversion model and the self-image theory one might also want to look into possible solution opportunities that differ between states. For instance areas with a lot of potential for renewable energy could have less solution aversion than others.
This study treated every fossil fuel, oil, gas and coal, as equal and only regarded the productions and prices of these. However gas is less
pollution than coal and oil it might be interesting to take that into account. For instance, one might find stronger results when elimination gas, or when one accounts for the CO2 emissions of the fossil fuels.
Conclusions
As seen in the results section the regressions have several significant outputs. First the important variables will be analysed. Finally the theories
on which this paper is based will be tested with the results to see how their predictions fit with the data.
Variables
The most important variables and their results are briefly discussed in this section.
Fossil fuel income
As this paper focussed mainly on the correlation between belief in climate change and the income of fossil fuels that will be analysed first. In all three datasets there is a negative coefficient for the variable fossil fuel income, however this coefficient is only significant at 5% in model 2 of the EU-dataset. Hence it is not very evident that there is a negative
correlation between fossil fuels income and belief in climate change. So the first hypothesis cannot be fully accepted. However as all datasets show negative coefficients, it is likely that there is a negative correlation. Hence the first correlation 0-hypothesis cannot be fully rejected, nor fully accepted. More research is needed to come to a conclusive result.
For the causality part a major problem is that the correlation has not been proven fully. However the difference in means between states that
increased their fossil fuel production and those that did not is significant. However this did not control for other variables, but it gives an indication that increasing the fossil fuel production does have a significant negative impact on climate change beliefs. Hence the causality 0-hypothesis cannot be rejected fully, however there is a weak evidence indicating that
production of fossil fuels has a causal negative effect on climate change beliefs.
Climate risk
Two models were tested as mentioned in the method section. One that did not include climate risk as an explanatory variable and one that did. In the EU- and World-datasets the R-squared increased, hence increasing the explanatory power of the model. In the US-dataset the climate risk
the model. In the EU-dataset climate risk was a quantitative variable and significant at 5%. The adding of climate risk made several other
coefficients more significant and increased the explanatory power of the model. In the world dataset the adding of climate risk did not help. All the variables remained insignificant and the r-squared barely increased. In the EU dataset climate risk was significant but the coefficient was positive while the hypothesis predicted a negative correlation. Based on the insignificant results in the US- and World-dataset and the significant but positive coefficient of the EU-dataset we cannot reject the 0-hypothesis and cannot find any evidence for a negative correlation between climate change beliefs and climate risk.
Political affiliation
Using all the variables in the models provided, political affiliation is found the most significant predictor of climate change belief. This is in line with existing literature, especially for the US. Being a republican state or having a rightwing government decreases beliefs in climate change compared to center or left orientated states or countries. Hence conformism seems to play a big part, however it is very hard to say anything about causality here.
Theories
In this section the different theories with which this paper started will be hold up against the results to see how their predictions turned out.
Groupthink
Benabou’s model predicted that there are two equilibria, one in which the majority denies climate change and one in which the majority accepts climate change. The more negative the potential impact of climate change the stronger this equilibria should be. Based on the results we see that a higher climate risk increases beliefs in climate change. Especially in the EU dataset climate risk plays an important role. However in the EU the beliefs in climate change are relatively high, with a minimum of only 81%. In the
data and results we do not see two clear equilibria arising, hence the groupthink model does not explain climate change beliefs fully.
Conformism
Kahan et al. (2012) argues that motivated beliefs are formed according to the group one associates himself with. In the results the variable for
political affiliation was by far the strongest predictor for the US-dataset, and also significant for the EU-dataset. This paper could not test this for the world dataset but based on the results in the US- and EU-dataset we can conclude that conformism plays a major role in the beliefs of climate change. The more left orientated a local government is the more likely the people are to believe in climate change. However this is just a correlation and no causality. It is very well possible that people first form climate change beliefs and then elect a government based on that.
Self-image protection
The self-image protection theory predicted that those who would have to put more effort to mitigate climate change are more likely to deny climate change. Based on the results we see that in all the dataset the income from fossil fuels has a negative impact on the beliefs of climate change. However this effect is only significant in model 2 in the EU dataset. Hence there is no strong evidence that that the production of fossil fuels reduces beliefs in climate change, however there is an indication pointing that direction. Also an increase in local fossil fuel production reduce the beliefs in climate as predicted by this theory. Hence the self-image protection theory can partly explain the climate change beliefs. Another argument in favour of the self-image protection theory is that online search behaviour correlates very strongly with climate change beliefs.
Solution aversion model
The solution aversion model predicts almost the same as the self-image protection theory. Those who have strong ties with the fossil fuel industry are more likely to deny climate change. As mentioned above there is no strong evidence for this but an indication that there is indeed a negative
correlation. Hence as the self-image protection theory this model can partly explain climate change beliefs. Especially since there seems to be a causal effect of local fossil fuel production
The explanation of climate change beliefs is too complex to fully explain with one of these models. However this research has found some weak evidence that income from fossil fuel has a negative causal effect on climate change beliefs. Based on the results and the data it is a
combination of all the theories mentioned above, and possibly of other mechanisms besides motivational cognition. However, based on this study, motivational cognition does play a role in climate change beliefs.
References
Benabou, R. (2013). Groupthink: Collective delusions in organizations and markets. Review of economics studies, Vol. 80, pp. 429-462.
Benabou, R. (2015). The economics of motivated beliefs. Princeton University.
Bulkeley, H. (2000). Common knowledge? Public understanding of climate change in Newcastle, Australia. Public understand. Sci. 9, 313-333.
Campbell, T. &, Aaron, K. (2014). Solution Aversion: On the Relation Between Ideology and Motivated Disbelief. Journal of Personality and Social Psychology, 2014, Vol.107(5), pp.809-824.
Cook, J., Nuccitelli, D., Green, S. A., Richardson, M., Winkler, B., Painting, R., Skuce, A. (2013). Quantifying the consensus on anthropogenic global warming in the scientific literature. Environmental Research Letters, 8, Article 024024.
Dana, Jason, Roberto Weber, and Jason Xi Kuang (2007). “Exploiting
Moral Wiggle Room: Experiments Demonstrating an Illusory Preference for Fairness.” Economic Theory, 33(1), 67–80.
ENERGY INFO. ADMIN. (2013). U.S. Expected to Be Largest Producer of Petroleum and Natural Gas Hydrocarbons in 2013.
Gifford, R. (2008). Psychology's essential role in alleviating the impacts of climate change. Canadian Psychology/Psychologie Canadienne, 49(4), 273-280.
Grossman, Z. & Van der Weele, J. (2016). Self-image and wilful ignorance in social decisions. Journal of the European Economic Association.
Heffernan, Margaret (2012). Willful Blindness: Why we ignore the obvious at our peril. Walker Books.
Hornsey, Matthew J. ; Harris, Emily A. ; Bain, Paul G. ; Fielding, Kelly S. (2016). Meta-analyses of the determinants and outcomes of belief in climate change. Nature Climate Change, Vol.6(6), pp.622-626.
Intergovernmental Panel on Climate Change (2007). Climate change 2007: The physical science basis. Contribution of Working Group I to the fourth assessment report of the Intergovernmental Panel on Climate Change. New York, NY: Cambridge University Press.
Janis, I. (1972) Victims of Groupthink: Psychological Studies of Policy Decisions and Fiascoes. Boston, MA: Houghton Mifflin Company.
Kahan, D. M., Jenkins-Smith, H. & Braman, D. (2011). Cultural cognition of scientific consensus. J. Risk Res. 14, 147–174.
Kahan, D. M. et al. (2012) The polarizing impact of science literacy and numeracy on perceived climate change risks. Nature Clim. Change 2, 732–735.
Leiserowitz, A. & Broad, K. (2008). Florida: Public opinion on climate change. A Yale University/University of Miami/Columbia University Poll. New Haven, CT: Yale Project on Climate Change.
Norgaard, K. (2011). Living in denial: climate change, emotions and everyday life. The MIT press, London.
Quattrone, George A.; Tversky, Amos. (1984) Journal of Personality and Social Psychology, Vol 46(2), 237-248.
Swim, Janet, et al. "Psychology and global climate change: Addressing a multi-faceted phenomenon and set of challenges. A report by the
American Psychological Association’s task force on the interface between psychology and global climate change." American Psychological
Association, Washington (2009).
Treweman, J. (2009). Beliefs about the economic impact of immigration. Toulouse school of Economics.
Yale project on climate change (2014). Howe, P., Mildenberger, M., Marlon, J.R., and Leiserowitz, A., “Geographic variation in opinions on climate change at state and local scales in the USA,” Nature Climate Change.
