Da Li Su Pesimisti U Pravu

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/272248616
Economic Conditions, Risk Perception and Climate Policy Preferences
Article in SSRN Electronic Journal · January 2014

DOI: 10.2139/ssrn.2464031
CITATIONS READS
0 229
3 authors, including:
Aya Kachi Thomas Bernauer

University of Basel ETH Zurich
42 PUBLICATIONS 332 CITATIONS 323 PUBLICATIONS 7,327 CITATIONS
SEE PROFILE SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Spatial Methods View project
Political Economy of Food System Transformation View project
All content following this page was uploaded by Aya Kachi on 05 June 2018.
The user has requested enhancement of the downloaded file.

Climate Policy in Hard Times: Are the Pessimists Right?
Accepted for publication at Ecological Economics: Tuesday, March 10,

2015
Aya Kachi (ETH Zurich)*
Thomas Bernauer (ETH Zurich)
Robert Gampfer (European Commission, Joint Research Centre)
Abstract
Conventional wisdom holds that the state of the economy has a strong impact on
citizens’ appetite for environmental policies, including climate policy. Assuming
median voter preferences prevail, periods of economic prosperity are likely to be
conducive, and economic downturns are likely to be detrimental to ambitious climate
policy. Using original surveys in the United States and Germany, we engage in a
critical re-assessment of this claim. The results show that, for the most part,
individuals’ perceptions of their own economic situations have no significant effect on
their policy support. Negative perceptions of the national economic outlook reduce
support for climate policy in the US, but not in Germany. The magnitude of this
national economy effect in the US is small, however. On the other hand, individuals’
climate risk perceptions consistently have a statistically significant and large effect
across various model specifications, and interestingly, this pattern holds for the US,
whose government is among the less ambitious in global climate policy, as well as
Germany, which is among the frontrunners. Our study indicates that the state of the
economy may not trump climate risk considerations as much as conventional wisdom
claims.

*
This author’s affiliation will change to the Faculty of Business and Economics at the
University of Basel, Switzerland, starting on May 1, 2015.
1
1. Introduction
Originally coined by James Carville, the strategist of Bill Clinton’s successful

1992 United States (US) presidential campaign, "It's the economy, stupid" has come to
mean that economic conditions leave a strong imprint on voter preferences and
political choices more generally. In line with this common belief, policy-makers and
interest groups have attributed the lack of ambitious climate policies in recent years to
the slow recovery of national and global economies after the recession during the early
2000’s (e.g. Egenhofer and Alessi 2013; IMF 2009). Are the climate-policy pessimists
right about the negative impact of economic downturns?
Various social polls provide macro-level overviews of this issue. At first

glance, these opinion polls seem to back up the claim that economic downturns reduce
citizens’ appetite for environmental policies. The proportion of respondents in Gallup
surveys who were concerned about global warming “a great deal” increased from 26%
to 41% between 2004 and 2007, but then began to decrease from 2007, reaching 28%
by 2010 (Brulle et al. 2012, p.170; Gallup Poll Social Series)†. One might thus suspect
that the economic downturn in 2008-2009 triggered this decreasing trend in public
concern about global warming.
Some time-series data show indeed negative correlations between the perceived
salience of climate and economic issues. For example, the US Gallup surveys from
1984 through 2012 ask respondents what they think “is the most important problem
facing” the country. In their responses, the frequency of “unemployment” correlates
negatively with that of “environment/pollution” (magnitude: -0.43)‡ and the correlation
is statistically significant (p-value: 0.04). Likewise in Germany, the Eurobarometer
survey asks “[w]hat are the two most important issues facing Germany today.” The
correlation between “unemployment situation” and “environment, climate and energy”
is again negative and statistically significant (magnitude: -0.12, p-value: 0.00).

†
Gallup is a worldwide public opinion poll company. The exact survey wording in the cited Gallup Poll
Social Series on environment is as follows. “I'm going to read you a list of environmental problems. As I
read each one, please tell me if you personally worry about this problem a great deal, a fair amount, only
a little, or not at all. First, how much do you personally worry about global warming?” Information
available at http://www.gallup.com/poll/168236/americans-show-low-levels-concern-global-
warming.aspx (Last accessed on January 4, 2015.)
‡
Pearson’s correlation coefficient.
2
Several academic studies have also addressed these questions and identified
key factors shaping public opinion on climate issues in relation to the effect of
economic conditions (see for example, Brulle, Carmichael and Jenkins 2012; Kahn and
Kotchen 2011; Scruggs and Benegal 2012; Krosnick and McInnis 2012); yet the
empirical findings in the present literature are mixed at best, and their empirical design
suffers from several limitations. First, a large subset of empirical analyses stays at the
macro level, relating the size of economy-related concerns and that of climate-related
concerns. Yet some of them make a theoretical claim about policy preferences held by
individuals, potentially suffering from the ecological fallacy problem. Second, many of
the individual-level studies claim the importance of promoting climate policy
acceptance and implementation; yet they investigate issue salience and climate-risk
perceptions—thereby, conflating potential antecedents of policy support with policy
support itself. Drawing on the literature of policy preference, we argue that perception
of climate risks should be rather be an important predictor of policy support. Another
issue is the common use of unemployment rate as a proxy for economic conditions.
While absolute indicators such as official unemployment rate and gross domestic
product (GDP) capture a certain aspect of a country’s (or a region’s) economic
conditions, we should not automatically equate these indicators with citizens’
individual views on the economy.
To address these issues, we draw on data from our original surveys conducted
in the US and Germany, and reexamine the effect of perceived economic conditions on
individuals’ climate-policy preferences. We select these two countries based on their
contrasting statuses concerning climate politics—the US, whose government is among
the less ambitious in global climate policy, and Germany, which is among the
frontrunners (Weidner and Mez 2008). Using the survey data, we evaluate the extent to
which citizens’ support for climate policies depends on their perceptions of economic
outlook at their household as well as national levels. In so doing, we also compare the
effect of economic conditions with that of another likely determinant, climate risk
perceptions. We find that, for the most part, individuals’ perceptions of their own
economic situations have no significant effect on their policy support in either country.
Negative perceptions of the national economic outlook reduce support for climate
policy in the US, but not in Germany. On the other hand, individuals’ climate risk
3
perceptions have a statistically significant and large effect across various model
specifications, and interestingly, this pattern holds for the US and Germany.
The following section further elaborates upon the limitations of the present
literature and outlines our theoretical arguments. After which we lay out our survey
design and method for creating key variables via factor analysis. We then analyze the
relationship among economic conditions, climate risk perceptions and policy support
using ordinary least square (OLS) and logistic regressions. Before concluding, we
present the results from two sets of robustness tests: one using an alternative (absolute)
measure of economic conditions, and the other relying on an experimental approach to
examine the causal effect of economic conditions. We conclude the paper by
summarizing our findings concerning whether and to what extent economic conditions
affect individual climate policy preferences, compared to another likely determinant,
climate risk perceptions.
2. Effects of economic conditions and risk perceptions on policy

support
To investigate the widespread belief that climate policies could not earn citizens’
approval during economic downturns, several academic studies have addressed related
questions. Brulle, Carmichael and Jenkins (2012) show that aggregated “public-mood”
for climate concern in the US significantly depends on structural economic factors
measured by unemployment rate and gross domestic product (GDP). Kahn and
Kotchen (2011) find that increasing unemployment rates in US states “reduce the
number of Google searches for global warming” (p.263) within the states and “increase
the number of Google searches for unemployment” (p.263). Based on another policy-
preference measure at the individual level, they also find that the US citizens from
states with lower unemployment rates are more likely to find global warming a serious
problem and they are more willing to accept climate policies. Scruggs and Benegal
(2012) address similar questions in both the US and European contexts, and conclude
that individuals in both regions are less likely to consider climate change a risk when
economic conditions are worse. Others have produced conflicting empirical results,
however. Notably, using data from a nationally representative survey for the US,
4
Krosnick and McInnis (2012) report that people in states with struggling economies
did not exhibit a larger decline in support for climate policy.
These studies provide a good starting point; however we identify four

limitations in their arguments and design that prevent us from drawing a firmer
conclusion. The remaining of the section details the four issues and presents why a
new set of empirical analyses is necessary.
2-1. Level of analysis — citizens
First, the ecological inference problem arises in macro-level empirical studies when
they are motivated by theoretical arguments on changes in individuals’ policy salience
and preferences. For example, Kahn and Kotchen (2011) analyze the number of
Google searches for global warming and unemployment as evidence for the effect of
business cycles on the public’s climate concern. The potential inferential problem is
not that macro-level data are used, but that the discussion is tied back to the dynamic
of citizens’ or households’ climate risk perceptions. This trend can be observed also in
other studies such as Del Río and Labandeira (2009) and Pew Research Center (2009).
Consistent with the initial motivation of these studies, we also investigate the
claim that economic downturns might influence individuals’ policy appetite. In order
to overcome potential ecological fallacy; i.e. in order to investigate whether those
individuals greatly concerned about the economy are the ones who are less concerned
about climate, it is necessary to design an empirical analysis where both the economic-
condition and policy-preference variables are measured at the individual level.
2-2. Explaining policy support, rather than salience and concerns
Another limitation of the literature is the common conflation of policy preference and
issue salience. The majority of existing public-opinion surveys on climate issues—
including the Gallup polls, the PEW surveys, and Eurobarometer§—focus primarily on
issue salience and climate risk perceptions (concerns). For instance, some of the most
typical survey items ask respondents whether they think the effects of global warming

§
We refer to Eurobarometer ver.77.3, May 2012, ZA No. 5612 (European Commission 2012).
5
have already begun to happen, what they think are the most important issues facing
their countries (with climate change being one of the possible responses), and to what
extent respondents worry about global warming. Not surprisingly, empirical studies
that use these data focus on explaining the public’s sentiment, belief, risk perceptions
and awareness regarding climate change (e.g., Scruggs and Benegal 2012; Leiserowitz
2005).
However, in line with the broader public opinion literature, we argue that it is
the public’s policy preferences, rather than policy salience or awareness that drives
policy adoption and government’s responsiveness (see, for example, Page and Shapiro
1983; Stimson et al. 1995). The reason is, as Druckman (2013) states succinctly,
elected officials “would not pass laws that are not supported by their constituents —
and on what consumers in the marketplace are willing to accept” (617).
The conceptual difference between preference and salience is important.

Individuals can be highly aware of potential damages caused by global warming; yet
they can still make conscious decisions not to support a new climate policy initiative,
due to, for instance, economic difficulties facing their households. Studies in public
opinion have demonstrated this distinction empirically. Lax and Phillips (2009) show
that the level of policy salience is a mere amplifier of the effect of policy preferences
on policy adoption in the context of US domestic policy-making. Similarly, other
studies find that environmental attitudes, rather than mere awareness or knowledge on
the issue, help explain variation in environmental behavior (e.g. Kollmuss and
Agyeman 2002).
Therefore, it is necessary to measure individuals’ policy preference as a

primary outcome variable, in order to investigate the relationship between economic
conditions and the policy inertia. Following the definition in Leiserowitz (2006), we
propose to operationalize policy preferences as individual support for new regulatory
policies.
6
2-3. Risk perceptions influence policy support
The present literature focusing on the economic effect treats climate concerns as the
main outcome variable.** However, studies explaining policy support—which the
present paper intends to investigate—argue that climate risk perception is one of the
most relevant determinants of climate policy support (e.g. Egan and Mullin 2012,
Leiserowitz 2006, Leiserowitz 2011, Tjernström and Tietenberg 2008). Gardner et al.
(1987) further details the causal mechanism among awareness, risk perception and
policy response by explaining that public awareness of hazards is a potential basis for
risk perceptions, but the variation in risk perceptions is the direct determinant of
variation in policy preferences. Therefore, we take a similar approach to these scholars
and argue that individuals’ climate risk perceptions—how consequential they believe
global warming is—is an important determinant of their support for climate policy,
alongside economic conditions and other socio-demographic variables.
2-4. Effects of economic conditions: “absolute” vs. perceived, and household vs.
national
The extant climate-policy literature that relates economic conditions and public
opinion uses absolute economic indicators (in most cases unemployment rates), as
opposed to economic conditions as perceived by the public (e.g., Kahn and Kotchen
2011; Scruggs and Benegal 2012; Guber 2013). However, studies that examine the
effect of economic conditions on political outcomes—such as election outcomes—
have highlighted the importance of economic conditions perceived by citizens (e.g.
Sanders 2000). Blendon (1997)’s survey-based study explicitly compares the
congruence between public perceptions and the expert views on the past, current and
future economy in the US. The study finds significant discrepancies between the public
and expert views. Drawing on this literature, we argue that if economic conditions
affect individuals’ policy preferences in any particular way, then it is more likely their
perceived economic situations that count.†† Similarly to Blendon (1997)’s approach,

**
As mentioned in Introduction, the empirical findings in the present literature are mixed as to whether
economic conditions affect individuals’ climate risk perceptions.
††
Our additional analyses using actual economic indicators—unemployment rates—are reported in the
robustness-check section after the main empirical results.
7
we ask questions about respondents’ views on both the past and future economic
conditions.
Finally, distinction between household’s and national economic situation is
another noteworthy aspect of economic effects to be considered. For example,
Markus’s (1998) study on the US presidential election finds that individuals’ vote
choices heavily depend on personal financial circumstances, so-called the
“pocketbook” criteria, compared to the national circumstances or “socio-tropic”
criteria. To the best of our knowledge, the present literature on climate politics has not
conducted empirical tests for this classical debate regarding pocketbook-versus-
sociotropic criteria in the public’s policy evaluation. Therefore, we propose to
investigate whether people’s policy preferences depend primarily on economic
conditions of their own or of the society as a whole.
2-5. Theoretical framework
Based off these findings and shortcomings of extant work, we identified following
three key effects to be examined in our study. The first is the effect of perceived
economic circumstances on climate policy support. We separate this effect into
individuals’ perceptions of the household economy (the pocketbook criterion) and of
the national economy (the sociotropic concern). Given the mixed findings of the
literature on climate policy, the presence of the overall economic effect is an empirical
question. However, if findings from the election studies in the US hold more generally,
then we should expect that citizens react to policies based on their personal economic
circumstances more strongly than the national condition.
The second is the significance of the economic effect as a whole compared to

the effect of individuals’ climate risk perceptions. The literature suggests that
individuals who hold greater climate concerns support new regulations more likely,
and its effect is prominent. We investigate how persistent the potential positive effect
of risk perceptions is, in the presence of negative effects that economic downturns
might generate.
Third, as a robustness check and for the purpose of comparing with some
extant studies, we investigate whether bringing back an absolute economic indicator—
namely unemployment rate—changes our main empirical result. We examine this by
8
replacing the perceived national economic condition with the officially reported
regional unemployment rate data.
3. The empirical strategy: Evaluating the effects of perceived

economic conditions and risk perceptions on policy support
Using original survey data from the United States and Germany, we engage in a
critical re-assessment of the popular claim that economic downturns trump public
support for climate policies. Our survey design entails constructing these key outcome
and explanatory variables—policy support measures, perceived states of the household
and national economy, and perceived risks associated with climate change, as well as a
variable indicating respondents’ general interest in the climate issue and their
demographic characteristics as control variables.
Some of these variables are not directly observed, or can be captured by a

single question due to their multi-facetedness. These variables include the level of
citizens’ policy support, interest in the issue and perceptions of climate risk. For these,
we develop a comprehensive battery of survey items that collectively manifest an
underlying latent concept, and use factor analysis to construct a latent measure
representing each of the unobserved concepts. Confirmatory factor analysis (CFA)
allows us to assess statistically whether the survey data fit our measurement model; i.e.
whether the relationship we suspect between multiple observed survey items and their
underlying latent constructs of policy support or climate risks indeed exists (Schreiber
et al. 2006).
Moreover, for the outcome variable measuring climate policy support, we

dissect the concept from three distinct angles: (1) general attitude towards climate
policy (Attitude); (2) behavioral intention concerning climate mitigation efforts
(Behavioral Intention, or more simply Behavior); and (3) willingness to pay (WTP).
The dichotomous indicator of willingness to pay (WTP) is measured by whether a
respondent preferred a cash payment from a lottery (which took place after the survey)
or donating to a US- or Germany-based third-party climate organization investing in
carbon dioxide reduction projects (so-called carbon offsetting). The remainder of the
section describes the survey design and details the construction of our key variables.
9
3-1. Survey and sampling design
We designed and implemented our original nationally representative surveys in

Germany (N=1,651) and the US (N=1,559) between October 16 and November 2,
2013. Sampling and data collection were carried out by the YouGov survey firm
(www.research.yougov.com).‡‡
YouGov constructs nationally representative samples by first interviewing a

larger number of (potential) respondents followed by the targeted number. The
interviewed respondents are matched to a sampling frame on several demographic
characteristics: gender, age, race, education, party identification, ideology, and
political interest. The baseline sample characteristics are constructed by stratified
sampling from the full 2010 American Community Survey (ACS) sample, with
selection within strata by weighted sampling with replacements. YouGov then uses
propensity scores to weight the matched cases to the sampling frame. According to
YouGov, the propensity score function “included age, gender, race/ethnicity, years of
education, ideology, and geographical region.” Our data analysis uses sampling
weights provided by YouGov.
There has been some debate regarding the quality of samples from online
surveys, including YouGov surveys. However, an increasing number of meta-studies
show that the quality of opt-in online surveys is as reliable as the quality of traditional
phone- and mail-based survey. For instance, Ansolabehere and Schaffner (2014)
compare three modes of surveys in the US—an opt-in internet survey, a telephone
survey via Random Digital Dialing (RDD), and a mail-in survey. Many of the
variables within this questionnaire were verifiable with the government data and they
conclude that carefully executed opt-in online surveys are as accurate as a traditional
telephone survey. Thus we are fairly confident about the mode of survey we employ in
this study.
3-2. Outcome variables

As noted above, we measure three facets of support for or opposition to climate policy.
These are: (1) general attitude towards climate policy (Attitude), (2) behavioral

‡‡
We selected YouGov because of its strong reputation as a survey firm serving the scientific
community. In political science, for instance, 47 articles published in either the American Political
Science Review, American Journal of Political Science, or Journal of Politics used survey data collected
by YouGov/Polimetrix between 2006 and 2013.
10
intention concerning climate policy (Behavior, or Behavioral Intention), and (3)
willingness to pay, measured by the willingness to invest lottery winnings into a
carbon offsetting project (i.e. their WTP), rather than cashing in.
Since (1) attitudinal support for climate policies and (2) behavioral intention
concerning climate policy are rather broad and multi-faceted concepts, we used
multiple survey items to construct two composite measures of public support via
confirmatory factor analysis (CFA). CFA seeks to assess statistically whether the
survey data fit our measurement model; i.e. whether the relationship we suspect
between observed variables (survey questions we chose to use) and their underlying
latent constructs of policy support indeed exists (Schreiber,et al. 2006). The survey
items used to construct each concept are listed in Table 1.§§
The third response variable, Willingness to Pay (WTP) is a single-item

measure. As summarized in Table 1, this item simply asks whether respondents would
like to invest lottery winnings into a carbon offsetting project (i.e. their WTP), rather
than cashing in for. We include this item, in order to incorporate a long tradition in
economics to rely more “on revealed preferences rather than on opinion surveys to
explain public behavior” (Blendon 1997, p.107). This type of behavioral measure is
useful when people might have an incentive not to state their policy attitudes correctly
or honestly in opinion surveys. In the context of climate issues, such inaccurate reports
can be caused by social desirability; i.e., respondents might overstate their
environment-policy friendliness due to social norms and pressure. We based our
survey design on the general strategy taken in prior studies that investigate values of
environmental features, whose prices are typically not determined via a market. These

§§
Among the four survey items used to construct the Behavioral Intention construct (see Table 1), the
wording of the second item “personal_2” can potentially be interpreted as a knowledge question rather
than a behavioral intention question. Our original intention was to gauge the perceived likelihood that a
respondent “would avoid activities that emit carbon dioxide” because it would “contribute to solving the
problem of global warming.” Our intention of using the milder wording was to alleviate the potential
social desirability effect that might have been caused by a direct question, such as “do you intend to
avoid activities that emit…” However, to take extra precautions, we have conducted an additional set of
empirical analyses by dropping this item “personal_2”. Appendix 2 fully reports the results.
11
Table 1: List of survey items used to construct climate policy support measures
Variable Survey Survey item wording Response scale

item
Attitude climate_po People hold different views about whether 1 Should give priority
licy_1 policy-makers should give priority to measures 2 Should not give priority
(Composite against global warming, even if such measures 3 Don’t know
variable have a negative effect on the economy. What is
consisting of 5 your view?
survey items) climate_po To deal with global warming, do you think the 1 Too much
licy_2 government of the United States is doing … 2 About the right amount
3 Not enough
4 Don’t know
climate_po Do you favor or oppose preserving or 1 Favor strongly
licy_3 expanding forested areas, even if this means 2 Favor somewhat
less land for agriculture or construction? 3 Oppose somewhat
4 Oppose strongly
5 Don't know
climate_po Do you favor or oppose increasing the 1 Favor strongly
licy_4 requirements for fuel efficiency of automobiles, 2 Favor somewhat
even if this raises the cost of cars and bus fares? 3 Oppose somewhat
4 Oppose strongly
5 Don't know
climate_po Imagine that taking effective steps against 1 Would be willing
licy_5 global warming would increase energy costs to 2 Would not be willing
the average household in the United States by 3 Don’t know
20 dollars per month. Would you be willing or
not be willing to pay this additional cost as part
of taking steps against global warming?
Behavioral al personal_1 If I had to reduce my energy consumption and 1 Strongly agree
Intention carbon dioxide emissions, this would reduce 2 Mostly agree
my quality of life too much. 3 Mostly disagree
(Composite 4 Strongly disagree
variable personal_2 If I avoid activities that emit carbon dioxide, I 1 Strongly agree
consisting of 4 contribute to solving the problem of global 2 Mostly agree
survey items) warming. 3 Mostly disagree
4 Strongly disagree
personal_3 I prefer to enjoy life without having to worry 1 Strongly agree
about how much energy I consume and how 2 Mostly agree
much carbon dioxide I emit. 3 Mostly disagree
4 Strongly disagree
cartype Imagine you are buying a new car and you have 1 Larger, more powerful car
to choose between a larger, more powerful car consuming more fuel
that consumes more fuel, and a smaller and less 2 Smaller, less powerful car
powerful car that consumes less fuel. Assume consuming less fuel
that this will be the only car you own, and that
both cars cost exactly the same. Which car
would you buy?
Willingness to lottery We now ask you to participate in a lottery. - A cash prize of $25.
pay (WTP) Once this survey is completed, 20 participants If you choose this option and
will be randomly selected to receive a prize of win, $25 will be added as a
(A single-item $25 each. You will be notified by e-mail if you bonus to your regular
variable) are among the winners. payment.
- A so-called carbon offset of
At this point, please choose between two types $25.
of prizes you can receive if you win: If you choose this option and
win, we will donate your prize
of $25 on your behalf to a
project that reduces carbon
dioxide emissions. Within the
next few days, we will send
you, by email, a confirmation
from Climate Trust, a well-
established and trustworthy
organization based in the
United States that manages the
project, so that you can verify
your contribution to reducing
carbon dioxide emissions.
Note: Items in Table 1 are worded for the US survey. The German version was translated accordingly.
12
studies quantify such values by looking at respondents’ willingness to pay (WTP), for
instance, to green power (Champ and Bishop 2001, Poe et al. 2002), as well as their
willingness to donate to other environmental causes (e.g., Carlsson and Marinsson
2001; Brown et al. 1996).
By constructing broad policy-attitude measures with these two approaches—

aggregating multiple survey items and incorporating the revealed-preference
approach—we believe our measure of policy attitude is significantly more robust and
improved compared to the past research.
3-3. Key explanatory variables

The main explanatory variables in our analysis, besides standard socio-economic
determinants, are two types of perceived economic conditions at household and
national levels (“Household” and “National”), and climate risk perceptions (“Risk”).
For economic variables, as summarized in Table 2, we asked participants how they
expected their economic situation to change in the following three years. Each of these
variables captures the “pocketbook” (household) and “sociotoropic” (national) aspect
of perceived economic outlook respectively (Markus 1998). Each of them is measured
by a single survey item. However, we should not assume away the possibility that
citizens’ policy preferences depend more on the change they have already experienced
in the past rather than what they anticipate in the near future. As a robustness check,
we repeated our empirical analyses using perceived economic conditions over the past
three years (“Household past” and “National past”).
The second key determinant, climate risk perceptions can be measured in

various ways, for instance in terms of how serious a person thinks the climate change
problem is, whether she/he believes that the principal cause of global warming is
anthropogenic emissions, or by how she/he perceives the level of consensus in the
scientific community. Again, we used CFA to aggregate several survey items into a
broader concept of perceived climate change risks.
Survey items used to construct the climate risk variable are listed in Table 2.
The reasoning for including these three items is the following. The first part captures
respondents’ own perceptions about how consequential climate change (or global
13
warming) is. *** The second item shows their views on how consequential human
activities are. The last item is an indirect measure of risk perceptions—perceptions of
risks via their instinct about the expert sentiment concerning the issue. This item
reflects findings that disagreement among scientists can significantly lower policy
support for pro-climate actions (Aklin and Urpelainen 2014) .We theorize that each of
these items reflects the scale, depth and scope of individuals’ climate concerns,
constituting a comprehensive measure of perceived risks.
Table 2: List of survey items used to construct measures of risk perceptions and perceived
economic conditions (main explanatory variables)
Variable Survey Item wording Response scale
item
Risk gwarm_1 In your view, is global warming, also known 1 Very serious problem
(Composite as climate change, a very serious problem, a 2 Somewhat serious
variable somewhat serious problem, a not too serious problem
consisting of problem, or not a problem? 3 Not too serious
3 survey problem
items) 4 Not a problem
5 Don’t know
gwarm_2 When coal, oil, natural gas, or other fossil 1 Causing global
fuels are burned for energy, this releases warming
carbon dioxide. Some say that these carbon 2 Contributing a little
dioxide emissions are causing global warming bit to global warming
and more extreme weather events. Others say 3 Not causing global
that changes in global temperatures and warming
weather are happening for other reasons and 4 Don’t know
have little to do with emissions of carbon
dioxide by people. What is your view? Do you
think that emissions of carbon dioxide are …
gwarm_3 On the subject of global warming, is it your 1 Most scientists think
impression that among the scientists of the the problem is urgent
world … and enough is known to
take action
2 Most scientists think
the problem is not
urgent, and not enough
is known to take action
3 Views are evenly
divided among
scientists
4 Don’t know
Household econ_2 Do you think that your household income in Increase a lot
(A single- the next three years will… Increase somewhat
item Stay the same
variable) Decrease somewhat
Decrease a lot
National econ_10 How do you expect the economic situation in Become a lot worse
(A single- the United States as a whole to develop over Become worse
item the next three years? It will … Become better
variable) Become a lot better
Note: Items in this table are worded for the US survey. The German version was translated accordingly.

***
The first item “gwarm_1” is an adjusted version of a Eurobarometer question categorized as climate
concern (European Commission 2012).
14
3-4. Control variables
In estimating the effects of three key explanatory variables on policy support for
environmental policy, we control for several socio-demographic characteristics of our
survey respondents. We specify our regression models controlling for gender, age,
education, party affiliation (only for the US)††† , income, and religion, following the
convention of the existing climate-attitude research that investigate the effects of
economic conditions (e.g., Egan and Mullin 2010; Scruggs and Benegal 2012; Kahn
and Kotchen 2011).‡‡‡ Although the extant work mainly looks at individuals’ climate
concerns rather than policy support, we nevertheless expect similar effects of these
covariates. That is to claim that higher policy support comes from female, older, more
highly educated, and Democrat (for the US case) (Egan and Mullin 2010; Scruggs and
Benegal 2012). We have mixed expectations regarding Income. Surveys in the US tend
to reveal high correlations between the income level and the Republican partisanship;
hence one might expect a negative effect of income on climate policy support. At the
same time, resourceful respondents might be more supportive of costly climate actions
compared to lower income citizens (e.g., Dietz 2007).
Another common control variable is religion, in particular for studies on the US
citizens. Some studies control for the frequency of church attendance (e.g., Scruggs
and Benegal 2012) and others control for various Christian denominations (e.g, Barker
and Bearce 2013). Based on Barker and Bearce’s findings, we control for the effect of
most influential religious ideology—the belief in “Second Coming” or “born again”. It
is widely known that this group has a tendency to be skeptical about climate change
and climate policies (e.g., Greeley 1993; Guth et al. 1995).§§§

†††
We control for party ID for the US case, because the climate debate—or for that matter also many
political and social issues—has become extremely politicized in the US, and the partisan divide on this
issue is stark. On the other hand, we maintain that, in German, citizens are much less polarized by party
line (Bartels 2013), and that partisanship is neither fixed over time nor is a good predictor for their
climate policy attitudes (Zuckerman and Kroh 2006 ). Therefore in order to maintain the parsimony of
our regression specification (whose variable sets are already large), we do not include Party ID in our
models for Germany.
‡‡‡
However, note that the main outcome variables in these existing studies are the issue salience and
interest, which differ from our interest, policy support. Therefore our interest is not to make direct
comparisons of our coefficient estimates with findings from existing studies.
§§§
The theoretical mechanism behind this relationship seems to be indeterminate in the literature. Barker
and Bearce (2013) argue that this group of citizens tend to have shorter policy time horizons in mind due
to the Second Coming (i.e., the end-times) belief, and hence do not support climate policies, which are
designed to affect the society in the long run.
15
In addition to the conventional set of demographic characteristics, we also
control for the general level of self-reported interest (and partially knowledge) in issues
related to global warming. Our expectation is that a person with higher level of interest
and knowledge of climate issues shows a higher level of support for climate mitigation
policies. We expect the effect would be minor, however, compared to the effect of risk
perceptions, because it is most likely the believer of climate “risks” who supports
costly policies rather than somebody who is merely “interested” or “knowledgeable” of
the issue.**** We refer to this variable as “Interest”. It is a composite measure
constructed from three survey items via CFA (see Table 3). We deigned the three items
such that they capture the level of both familiarity and interest in the topic.
Table 3: List of survey items used to construct measures of interest/knowledge levels

“Interest”
Variable Survey Item wording Response scale
item
Interest Q1 How much had you thought about global 1 A lot
(Composite warming before today? 2 Some
variable 3 A little
consisting of 4 Not at all
3 survey
items) Q2 Do you think you would be able to describe in 1 Yes
very simple terms to another person what the 2 To some extent
problem of global warming is? 3 No
Q3 How often do you watch, listen to, or read 1 Never
news media reporting on global warming or 2 Rarely
discuss the issue with colleagues, friends, or 3 Sometimes
family? 4 Often
5 Very often
Note: Items in this table are worded for the US survey. The German version was translated accordingly.
4. Descriptive statistics of the policy-attitude measures and aggregated

explanatory variables
The first step of our empirical analysis entails constructing latent measures for multi-
faceted concept via confirmatory factor analysis (CFA) that were laid out in the
previous section. This section reports the results of our CFA and summarizes
descriptive statistics of the resulting composite variables.

****
One might think that these two variables are highly correlated. However, the correlation between the
two variables is 0.40 for the US and 0.37 for Germany.
16
4-1. Outcome variables
The results of the CFA support our composite measures of public support. As can be
seen in Table 4, all factor loadings are in the acceptable range of 0.6-0.9, and the
Cronbach’s alpha values, which indicates the degree of internal consistency among the
survey items, are above the conventional acceptance level of 0.6 (George and Mallery
2006), suggesting that the survey items are valid to construct a single latent variable.
After carrying out the CFAs, the latent Attitude and Behavior Intention variables were
constructed by regression methods, using the weights suggested by the CFAs for each
survey items.
Regarding the three outcome variables, we find that, both in terms of general
attitudes and behavioral intentions, support for climate policy is higher in Germany
than in the US (Figure 1). Specifically, attitudinal support is on average 14 percentage
points higher in Germany than in the United States, and behavioral support is 10
percentage points higher. This finding corresponds by and large to earlier surveys (e.g.,
PEW 2013), but is based on a broader measurement approach that puts more emphasis
on support for climate policy, rather than measuring primarily public concern over
climate risks. Our results also reveal that the share of respondents with very low
attitudinal support is higher and attitudinal support is much more polarized in the
United States than in Germany. Surprisingly, however, the willingness-to-pay measure
does not differ much between the two countries; it is even two percentage points higher
in the United States than in Germany.
As can be seen in Figure 1, for both countries, the number of observation (N) is
lower for the composite Attitude measure, compared to the other two outcome
measures. This is due to the higher number of “don’t know” (DK) responses in survey
items used for the Attitude variable. We allowed for “don’t know” in most of our
survey questions, hoping to obtain more meaningful responses also for those
respondents who cannot form opinions on certain issues.††††

††††
To our surprise, survey questions concerning what the German or US government is/should be doing
to combat climate change induced quite large numbers of DKs. These unobserved cases are dropped
from the empirical analyses that follow. Since the number of missing observations is large for Attitude,
we suspect that there might be slight differences between the demographic characteristics between the
DKs and the observed. Therefore we have conducted additional balance tests between the missing and
observed groups in Attitude along several demographic variables that later appear in our regression
analyses. (The detailed results of the balance tests are reported in Appendix 3.) In sum, the DKs and the
17
Table 4: Results of confirmatory factor analysis (CFA) for the outcome variables
GERMANY Survey item N Cronbach’s Eigenvalue* Factor Uniqueness
alpha loadings
Variable
Attitude 1087 0.84 2.56
climate_policy_1 0.87 0.21
Behavioral 1500 0.67 1.40
Intention
personal_1 0.59 0.65
cartype 0.58 0.66
US Survey item N Cronbach’s Eigenvalue* Factor Uniqueness
alpha loadings
Variable
Attitude 944 0.96 4.07
Intention
cartype 0.66 0.57
* For all the latent constructs, there was only one eigenvalue that is above 1. Therefore the table reports
the single eigenvalue that was above 1.

observed indeed show slightly different demographics characteristics, implying that the missingness in
Attitude is strictly speaking non-ignorable. The analysis of the missing observations in Attitude—for
instance, who they are and what exactly prevents them from stating clear opinions—is beyond the scope
of this paper; however we should take precautions in interpreting our regression results for the first
outcome variable, Attitude.
18
Figure 1: Climate Policy Support in Germany and the US
Germany USA
4-2. Aggregated explanatory variables: Risk and Interest
The results of the CFAs for Risk and Interest are reported in Table 5 and Table 6
respectively. Again, the Cronbach’s alpha scores are higher than the conventional
acceptance threshold of 0.6, and factor loadings are within the conventionally accepted
range, suggesting that our choice of survey items to construct the latent measure of
Risk is appropriate.
19
Table 5: Results of confirmatory factor analysis (CFA) for the risk perceptions variable
“Risk”
alpha loadings
Variable
Risk 1387 0.69 1.36
gwarm_1 0.73 0.46
gwarm_2 0.77 0.40
gwarm_3 0.48 0.77
alpha loadings
Variable
Risk 1252 0.91 2.35
gwarm_1 0.94 0.12
gwarm_2 0.94 0.12
gwarm_3 0.77 0.40
* For all latent constructs, there was only one eigenvalue that is above 1. Therefore the table reports the
single eigenvalue that was above 1.
Table 6: Results of confirmatory factor analysis (CFA) for the “Interest” variable
alpha loadings
Variable
Interest 1500 0.80 1.77
Q1 0.82 0.32
Q2 0.61 0.63
Q3 0.85 0.28
alpha loadings
Variable
Interest 1500 0.76 1.55
Q1 0.76 0.42
Q2 0.61 0.62
Q3 0.77 0.41
* For all latent constructs, there was only one eigenvalue that is above 1. Therefore the table reports the
single eigenvalue that was above 1.
5. Regression analysis and the results
With respect to the main question of interest here, we focus on comparing the effect of
two types of economic conditions on climate policy support—respondents’ perceived
household and national economic conditions, and on comparing this effect with the
effect of climate risk perceptions. As noted above already, this implies that, contrary to
some earlier studies, we regard climate risk perceptions as an explanatory factor for
climate policy support, rather than a proxy for the latter. To assess the impacts on the
first two outcomes variables—Attitude and Behavioral Intention, OLS regressions are
20
used, and for the dichotomous Willingness To Pay variables, logistic regressions are
used.‡‡‡‡
Figure 2 summarizes the regression results focusing on the main explanatory

variables – household and national economic conditions, and climate risk perceptions.
Table 7 shows detailed regression results, including the estimated effects of the other
covariates. In the table, each column is dedicated to each of the three policy-support
measures. Before discussing the results, it should be noted that the first dependent
variable—Attitude—has a significantly lower number of observations (Germany: 842,
US: 699) compared to the other two variables. As noted in Section 4-1, this is due to
the unexpectedly high numbers of Don't Know’s (DKs) we received for the attitude-
related questions in both countries. For this reason, we are careful not to be
overconfident in the precision of coefficient estimates for the Attitude regressions.
However, patterns in the statistical significance and magnitude are very similar
between Behavior and WTP specifications for the most part, which lends support for
our coefficient estimates for Attitude.
Controlling for a host of other potential factors that may also influence climate
policy support—namely interest in climate issues, gender, education and income for
Germany, and interest in climate issues, gender, partisanship, education, whether or not
a born-again Christian, and income for the US—we find that perceived future personal
economic conditions have almost no statistically significant effects on individuals’
policy support. Only in Germany and only on Attitude (in which we have less
confidence due to many DK’s), it has a significant effect. However, even in this one
case, the magnitude of the coefficient is a small degree of 0.03.
What stands out is that, in the US sample (but not in the German sample),
positive views on future national economic conditions have a positive and statistically
significant effect on all three measures of policy support. The magnitude of effects is,
again, very small. This means that negative perceptions of the future national economy
slightly reduce support for climate policy. For instance, changing one’s judgment of
the national economy from “become a lot better” (the highest category) to “become a
lot worse” (the lowest category) would reduce his or her policy support level by 9%.

‡‡‡‡
Since our outcome variables are ranged between 0 and 1, the more appropriate method would be
GLM with a logit link function. We have run GLM for each specification and the substantive results
remain the same. For simplicity of coefficient interpretation, we will report the OLS results in the paper.
21
The same counterfactual change in her economic view would reduce the probability of
donating to a climate organization by 15%.§§§§
Figure 2. Regression results for the three key explanatory variables
Attitude, (Germany) Attitude, (USA)
Household
National
Risks
Behavior, (Germany) Behavior, (USA)

Perception of...
Household
National
Risks
Willingness To Pay, (Germany) Willingness To Pay, (USA)
Household
National
Risks
0 .2 .4 .6 .8 0 .2 .4 .6 .8
Estimated effect size with the 95% C.I.
Not significant 90% 95% 99%
Note: (1) Dots indicate the estimated effects of the explanatory variables. (2) Whiskers indicate the 95%
confidence intervals around the means. (3) The first two outcome variables (Attitude and Behavior)
range from 0 to 1. These two models are estimated using an ordinary least square (OLS) estimator. (4)
The third outcome variable (WTP) is binary (0 or 1). The model was estimated using a maximum
likelihood estimator (MLE). The estimated effects for this model are the marginal effects computed after
the estimation of the logit model.

§§§§
Our survey also included items measuring perceptions of personal and national economic conditions
in the past few years. Using this data instead of perceived conditions in the coming years produces
essentially the same results. (See Table A1 in Appendix 1.)

22
Table 7: Regression Results: Effects of Perceived Future Economic Conditions
GERMANY Attitude Behavior Willingness

Dependent to pay
variable (lottery) Marginal
(OLS) (OLS) (Logit) effects++
Household 0.03*** 0.002 0.16 0.03
(0.01) (0.01) (0.12) (0.02)
National -0.01 -0.02* -0.004 -0.001
(0.01) (0.01) (0.10) (0.02)
Interest 0.06 0.04 0.93* 0.17*
(0.05) (0.04) (0.56) (0.10)
Risk 0.68*** 0.41*** 2.64*** 0.48***
(0.04) (0.04) (0.54) (0.09)
Female 0.02 0.04** 0.44** 0.08**
(0.02) (0.01) (0.20) (0.04)
Education 0.02 0.03** 0.50*** 0.09***
(0.02) (0.01) (0.20) (0.03)
Income 0.002 -0.002 0.06 0.01
(0.003) (0.002) (0.04) (0.01)
Age 0.001 0.003*** 0.02*** 0.004***
(0.001) (0.0005) (0.007) (0.001)
Intercept 0.04 0.18*** -6.03*** 0.004***
(0.06) (0.06) (0.91) (0.001)
Log-likelihood -566.3
N 842 1083 1083
R2 0.4 0.3 0.1+
AIC -425.5 -725.6 1130.6
BIC -382.9 -680.7 1175.5
USA Attitude Behavior Willingness to
Dependent pay (lottery)
variable Marginal
(OLS) (OLS) (Logit) effects
Household 0.01 -0.004 -0.014 -0.02
(0.01) (0.01) (0.12) (0.02)
National 0.03*** 0.03*** 0.33*** 0.05***
(0.01) (0.01) (0.11) (0.02)
Interest 0.09*** 0.02 0.84 0.14
(0.03) (0.04) (0.54) (0.09)
Risk 0.75*** 0.44*** 3.92*** 0.64***
(0.03) (0.03) (0.45) (0.07)
Female 0.02 0.05*** 0.88*** 0.14
(0.02) (0.02) (0.21) (0.03)***
Democrat 0.05*** -0.003 -0.13 -0.02
(0.01) (0.01) (0.15) (0.02)
Education 0.01 0.03 0.23** 0.04**
(0.01) (0.02) (0.10) (0.02)
Bornagain -0.01 0.03 0.09 0.01
(0.02) (0.02) (0.24) (0.04)
Income 0.0007 0.002 0.04 0.01
(0.002) (0.003) (0.03) (0.01)
Age -0.0004 0.001*** 0.02*** 0.003***
(0.0005) (0.0005) (0.01) (0.01)
Intercept -0.11** 0.09** -7.06***
(0.05) (0.05) (0.01)
N 699 973 973
R2 0.8 0.5 0.3+
AIC -464.1 -581.2 892.9
BIC -414.0 -527.6 941.7
+
Note: the R-squared values for the logit regressions reported above are McFadden’s pseudo R-
squared scores. Note that this score is computed as 1- (model likelihood)/(null likelihood), and it
is different from the R-squared in the OLS context, which indicates the percentage of the
outcome-variable variation explained by the fitted linear model. ++The 5th column reports
23
marginal effects for the logit regressions, indicating how much the probability for a person to
choose donation changes when we increase the level of a regressor by one unit, holding all other
regressors at their mean values.
It is easier to see how small these economic effects are when one compares
them with the effects of climate risk perceptions, which are much stronger. For the
United States, changing individual climate risk perception from the minimum level (0)
to the maximum level (1) increases the behavioral measure of policy support by 44
percentage points, whereas changing the perception of future national economic
conditions from the minimum (most pessimistic) to the maximum (most optimistic)
increases behavioral-intention measure by 9 percentage points. Likewise, the same
counterfactual shock to Risk would boost the WTP probability by 64 percentage
points, while a maximum positive shock to National would only improve Behavioral
Intention by 15 percentage points. The same pattern is observed for Germany.
As for demographic characteristics, if we look at Behavior and WTP

regressions, an older, female, and more highly educated person emerges as a
respondent who exhibits higher degree of policy support. This is common across both
countries. The interest level in the global warming problem does not have a consistent
effect on policy support. This could, however, be due to the artifact that those with
higher risk perceptions tend to be the one with higher interest levels in general.
Finally, one might find it interesting that income is not statistically significant
in any specifications and in neither country. From this result, it is indicative that, in
Germany, individual citizens’ climate policy preference are not driven by their
economic concerns as much as by their risk perceptions. In the US, an economic effect
associated with the national economic outlook seems to exist, but not to a person’s
current or anticipated economic situation.
6. Robustness Checks
We took two additional steps to corroborate our main results. First, we disaggregated
our survey data geographically and then analyzed the effect of local economic
conditions on climate policy support. To this end, we identified the geographic location
of respondents and regressed our climate policy support measures on officially
24
reported unemployment rates of US counties and German Bundesländer.***** ††††† This
step produces results that are somewhat comparable with results of studies by Kahn
and Kotchen (2011) and Scruggs and Benegal (2012), though the latter study uses
climate risk perceptions rather than climate policy support as the dependent variable.
It turns out that unemployment rates are not significant predictors of any of the
three measures of climate policy support (see Table 8). This finding suggests that
economic downturns measured by a typical macro-economic indicator such as
unemployment might reduce concern over climate risks, as shown in previous studies,
but do not seem to reduce support for climate policy. Furthermore, this result suggests
that, as we suspected, there might be indeed a difference between what perceived
economic conditions and absolute economic conditions predict in terms of the public’s
sentiment regarding climate policies.
Table 8: Regression Results: Effects of Perceived Future Economic Conditions
GERMANY Attitude Behavior Willingness

Dependent to pay
variable (lottery)
(OLS) (OLS) (Logit) Marginal
effects++
Household 0.019 -0.001 0.125 0.023
(0.01) (0.01) (0.13) (0.02)
Unemployment 0.003 0.004 0.015 0.003
(0.003) (0.002) (0.04) (0.01)
Interest 0.054 0.026 0.51 0.091
(0.06) (0.05) (0.61) (0.11)
Risk 0.661*** 0.390*** 2.800*** 0.51***
(0.05) (0.04) (0.59) (0.10)
Female 0.024 0.040** 0.415 0.08
(0.02) (0.02) (0.21) (0.04)
Education 0.023 0.029* 0.611** 0.11**
(0.02) (0.01) (0.21) (0.04)
Income 0.001 -0.003 0.039 0.007
(0.004) (0.003) (0.04) (0.01)

*****
Ideally, “Landkreis” is the German equivalent of “county” in the US; however, we used the
unemployment data at the “Bundesländer” level due to availability of the geographical information in
the YouGov survey.
†††††
One could probably think of various economic indicators other than unemployment rates, such as
the sizes of economic crises and growth rates. However in order to be somewhat consistent with the
existing literature and existing empirical studies, we use unemployment rates in our robustness check. In
existing surveys (e.g., PEW, Gallup, Eurobarometer), unlemployment/unemployment rate is the
economic issue used for comparisons with the salience of climate issues. Likewise, existing empirical
studies that motivated our analysis (e.g., Scruggs and Benegal 2012; Kahn and Kotchen 2011) use
unemployment rate as the measure of economic conditions.
25
Age 0.0004 0.003*** 0.026** 0.005**
(0.007) (0.0006) (0.01) (0.001)
Intercept 0.047 0.142* -5.930*** 0.004***
(0.07) (0.06) (0.90) (0.002)
Log-likelihood 274.5 300.7 -466.5
N 694 904 1083
R2 0.4 0.2 0.1+
AIC -337.6 -583.4 951.00
BIC -296.6847 -4540.2 994.3
USA Attitude Behavior Willingness
Dependent to pay
variable (lottery)
(OLS) (OLS) (Logit) Marginal
effects
Household 0.024** 0.009 0.131 0.021
(0.01) (0.01) (0.13) (0.02)
Unemployment -0.002 -0.002 -0.083 -0.013
(0.004) (0.003) (0.05) (0.01)
Interest 0.064 0.002 0.692 0.111
(0.04) (0.04) (0.53) (0.08)
Risk 0.776*** 0.483*** 4.393*** 0.71***
(0.03) (0.03) (0.47) (0.07)
Female 0.016 0.043** 0.726*** 0.117***
(0.02) (0.02) (0.22) (0.03)
Democrat 0.066*** 0.001 -0.116 -0.019
(0.01) (0.01) (0.15) (0.02)
Education 0.005 0.04 0.292** 0.047**
(0.01) (0.01) (0.10) (0.015)
Bornagain -0.012 0.034 0.030 0.005
(0.02) (0.02) (0.25) (0.04)
Income 0.0003 0.002 0.049 0.007
(0.002) (0.003) (0.03) (0.01)
Age -0.0004 0.001* 0.022** 0.003***
(0.0005) (0.0005) (0.01) (0.01)
Intercept 0.039 0.132* -6.327***
(0.06) (0.06) (0.88)
Log-likelihood -387.8 280.0 -387.8

N 644 902 902
R2 0.8 0.5 0.3+
AIC -416.5 -537.9 797.6
BIC -367.4 -485.1 850.4
Note: (1) +the R-squared values for the logit regressions reported above are McFadden’s
pseudo R-squared scores. Note that this score is computed as 1- (model likelihood)/(null
likelihood), and it is different from the R-squared in the OLS context, which indicates
the percentage of the outcome-variable variation explained by the fitted linear model.
++
The 5th column reports marginal effects for the logit regressions, indicating how much
the probability for a person to choose donation changes when we increase the level of a
regressor by one unit, holding all other regressors at their mean values. (2) the sources
of unemployment rates are the Bundesagentur für Arbeit
(http://statistik.arbeitsagentur.de/Navigation/Statistik/Statistik-nach-Themen/Statistik-
nach-Themen-Nav.html) for Germany and the US Bureau of Labor Statistics
(http://data.bls.gov/cgi-bin/dsrv?la) for the US (3) The sample sizes here are slightly
smaller than those in Table 7, where perceived measures of economic conditions were
used. This is due to the fact that some respondents preferred not to report their county of
residence to the YouGov survey company, preventing us from using these observations.
26
Second, strictly speaking, observational studies do not allow for identification
of causal effects. This means that the estimated association between the variables of
interest might not reflect the causal relationship we theorized. For instance, our
regression analysis shows that individuals’ perceived national economic conditions are,
to a small degree, statistically significantly correlated with their support for climate
policies. This does not automatically mean that perceived economic conditions are
causing higher or lower policy support levels. To cross-examine our correlational
findings, we implemented an additional survey experiment that investigates the causal
effect of economic conditions on climate policy support.
While the empirical methodology used in existing climate research that focuses
on public opinion is still mostly limited to descriptive or correlational approaches, the
use of experiments or survey experiments, which aim to test for causal relationships,
has been definitely increasing during the recent years. Bernauer et al. (2014), for
instance, employ survey experiments to examine whether different types of costs
associated with potential climate mitigation policies have direct causal effects on the
level of public support for such policies. Framing experiments are another kind of
survey experiments that has drawn increasing attention recently. This strand of
research is characterized by its emphasis on the effect of policy communications,
“frames”, on citizen support for climate policies (e.g., Aklin and Urpelainen 2013;
Bain et al. 2012; Myers et al. 2012).
What makes our survey experiment different from the existing research and
potentially extremely difficult is that economic conditions that face respondents are, in
a sense, not easily manipulable. It is likely that respondents are already pre-treated by
real and present economic conditions that surround them, either positively or
negatively. Unlike the mode of policy communications, the randomly assigned
treatment conditions are more difficult to influence their opinion formation.
Nevertheless, we believe that an attempt of causal studies via survey experiments is
useful.
The experiment was fielded in the April 2 – 30, 2013, period using a
convenience sample 1080 US respondents recruited via the crowd-sourcing platform
Amazon Mechanical Turks (AMT, www.mturk.com). After recruiting participants
27
through AMT, the online survey was implemented using a survey platform called
Unipark (Unipark, 2012).
After an introduction to the topic of global warming, participants were

randomly assigned either to a control or to one of two treatment groups: one treatment
emphasizes that the US economy is still experiencing serious problems, and the other
emphasizes strong recovery of the economy. To remain credible, neither treatment
contained information counterfactual to the actual economic situation in the US, the
information presented was merely framed differently. After exposure to one of these
treatments (unless in the control group), respondents were asked to respond to the same
items on climate policy support as used in the representative national surveys. Those
assigned to the control group proceeded directly to survey items measuring their policy
support levels, after reading the introductory paragraph about global warming. Table 9
summarizes the wording of the introduction and the information treatments that
participants received during the experiment.
It turns out that neither the positive nor the negative economic news treatment
has a significant effect (compared to the control group) on any of the three measures of
public support for climate policy. Admittedly, it is probably difficult to effectively treat
citizens with negative or positive assessments of real economic conditions. The reason
may be that most citizens pay considerable attention to economic conditions and have
rather stable views on what those conditions are. Hence, these views are difficult to
manipulate by means of an experimental information treatment.
Table 9: Informational treatments in the survey-embedded experiment
Introduction (for all respondents)
In many countries around the world, policies for dealing with global warming, also known as climate
change, are being considered. People hold different views about whether global warming is a problem
and what, if anything, should be done about it. We are interested in your views about this issue.
Control group
assign-
Rando
ment
No further information treatment.

m
28
Treatment group 1: bad economic news
Most experts and policy-makers, both Republican and Democrat, agree that the US
economy has, since 2007, been facing a major economic crisis, and it remains unclear
when this crisis will end. The biggest problems of the US economy are:
• The unemployment rate is still nearly 8 percent, and the number of long-term
unemployed Americans is almost 5 million.
• The percentage of working age Americans without a regular job has been
around 35 percent for more than three years in a row.
• Since 2008, the US economy has lost 8.7 million jobs.
• Average household income has declined by more than 4000 dollars in the
past few years.
• US home values have on average declined by 12 percent in the past few
years.
• The number of Americans on food stamps has grown from 32 million to 46
million.
• More than half of all Americans now have to rely at least partially on
financial aid from the government.
• The US government has run a budget deficit of well over a trillion dollars
almost every single year since 1970.
• The US national debt has increased by 50 percent which is an average of
more than $64,000 per taxpayer.
Treatment group 2: good economic news
Most experts and policy-makers, both Republican and Democrat, agree that 2013
should be a good year for the US economy. The main reasons are:
• Since mid-2011, the unemployment rate has dropped from 9.1 percent to 7.7
percent.
• In February 2013 alone, the economy created a net 236,000 new jobs.
• Since the start of 2010, economic growth was on average 2.2 percent, and in
2013 growth of 3.5 percent is likely.
• State and local governments are probably done with their process of financial
consolidation. Budgets will not have to be cut any more, and tax revenues
are growing again. Because of that, state and local government employment
seems to have bottomed out, and is on a gentle upswing.
• The number of housing units being built is up 22 percent from a year earlier
and up a whopping 58 percent from two years earlier. The upswing has
continued into 2013. This strong growth in the number of housing units
started is still well below many estimates of the longer-term rate of
household formation. This means there is plenty more room for housing
construction activity to grow, which is likely to give a boost to the economy
and create new jobs.
• The ratio of household debt to national income has fallen by 17 percent since
2009.
• The cost of servicing that debt is down even more because of low interest
rates; the ratio of household debt servicing costs to personal income is down
to 10 percent, near its lowest levels on records. Interest rates on credit cards
and auto loans are now very low as well, and more and more people are able
to refinance their mortgages as home prices rise.
• Americans’ household balance sheets are looking as good as they have in a
decade, which means that from this year on, debt overhang no longer looks
like a dark cloud over the economy
29
7. Discussion and conclusion
The main findings reported in this paper quantify and compare the effects of
individuals’ perceived economic conditions and the effect of perceived risks in
explaining individuals’ climate policy support in two large industrialized countries that
are key players in global climate politics, the United States and Germany. The results
challenge the pessimistic notion that negative economic conditions—either real,
perceived, prospective or retrospective—reduce public support for climate policy and,
by implication, trump climate risk perceptions. In our everyday life, we are constantly
exposed to news about macro-economic indicators such as unemployment, growth,
inflation, etc., which does not necessarily overlap with our personal economic
situations; yet our results indicate that, regardless of how we measure, the effect of
economic conditions is none to weak at best. Moreover, the results from our robustness
test using survey experiments are consistent with the correlational results, in that we
are unable to identify a statistically significant causal effect of economic conditions on
climate policy support. Therefore, we consider this a robust empirical result.
"It's the economy, stupid" might be relevant to various areas of public policy-
making. But one should not jump to the conclusion that science-based risk arguments
and related ambitious climate policies are bound to fail when the economy is in
trouble. The main policy-implication is that, at least from the perspective of electoral
support directly concerning individual voters, which is crucial to climate-policy
making in democratic countries, ambitious climate policies seem possible even when
economic conditions are problematic.
In this study, we found one economic variable in one context that has
significant effects on the public’s policy support. That is, the perceived national
economic condition in the US. Even then, magnitudes of estimated effects are rather
small and its counterfactual impact on policy support is limited.
Why do risk perceptions appear to matter much more for climate policy support
by citizens than perceived or real economic conditions? Arguably, the most plausible
candidate explanation is that the average citizen knows rather little about climate
change issues and related specific policies. General views on environmental protection
and politics are thus likely to shape both climate risk perceptions and attitudes towards
30
climate policy. Our statistical model specifications do take care of such potentially
confounding effects, so that we think the estimates do not suffer from estimation bias.
How do our results compare to earlier results by Kahn and Kochen (2011) and
Scruggs and Benegal (2012)? The main difference between their and our study is that
they explain climate risk (problem) perceptions, whereas we explain climate policy
support. Whereas they find at least some evidence that economic conditions have a
significant effect on climate risk perceptions (worse economic conditions lead to lower
perceived climate risk), we find only very weak to no evidence for an effect of
economic conditions on climate policy support. The three studies focus on different
levels of analysis, rely on different measurement concepts for their dependent variable,
and use different data for different time-periods or points in time. Nonetheless, on the
presumption that the findings of all three studies are robust, what could we learn about
the effect of economic conditions in climate governance? One way to look at it is that,
even though negative economic conditions can reduce the public’s appetite to climate
policies through decreased risk perceptions and issue salience, such policies could still
garner public support if policy-makers communicate with voters on climate risks
through multiple channels.
A useful future investigation would be to assess which explanatory model

structure is more plausible both theoretically and empirically. For instance, an
experimental approach could help find out whether economic conditions influence
climate policy support only via an effect on climate risk perceptions, or whether
economic conditions and climate risk perceptions have independent, separate effects
on climate policy support. Our experimental results, which uphold the main finding of
the correlational analysis, suggest that the estimated effect of economic conditions on
climate policy support, alongside climate risk perceptions, is unlikely to be biased by
omitted common influencing factors affecting both perceived climate risk and support
for climate policy.
Our research also contributes to the existing literature by using a more

comprehensive measurement approach for key concepts, such as climate policy
support and climate risk perceptions, and by examining the effects of various types of
measures for perceived and real economic conditions. The downside of our approach
is, obviously, that the analysis is limited to two countries at a given point in time.
31
Past surveys did not include the items that we would need for an in-depth
analysis of the effects of economic conditions on climate policy support along the lines
of this paper. And historical reconstruction of survey data (measuring individual
attitudes and preferences backwards in time) is impossible. This makes it virtually
impossible to implement a multi-country times-series analysis of this kind
retroactively. Nonetheless, it would be very useful if future research could explore the
extent to which our findings for the United States and Germany are relevant for other
countries, e.g. countries at different points in the business-cycle.
In summary, our research injects a dose of skepticism into the conventional

wisdom, which is reiterated in many statements by policy-makers and interest group
leaders, that voters will not reward, or will object to, ambitious climate policies as long
as their respective country’s economy is performing poorly. Rather, it suggests that
opposition to climate policy among parts of the electorate is due to specific worldviews
and political preferences that are at work independently of the business cycle – and that
might sometimes be aligned with particular economic interests of industry lobbies
References
Aklin, Michaël and Johannes Urpelainen. (2013) Debating clean energy: Frames,
counter frames, and audience. Global Environmental Change. 23(5);1225-1232.
Aklin, Michaël, and Johannes Urpelainen. (2014) "Perceptions of scientific dissent
undermine public support for environmental policy." Environmental Science &
Policy 38: 173-177.
Ansolabehere, Stephen, and Brian F. Schaffner. (2014) "Does survey mode still
matter? Findings from a 2010 multi-mode comparison." Political Analysis . doi:
10.1093/pan/mpt025
Bain Paul G. et al. 2012. Promoting Pro-Environmental Action in Climate Change
Deniers. Nature Climate Change, 17 June 2012.
Bartels, Larry M. (2013) "Party systems and political change in Europe." Annual
Meeting of the American Political Science Association: Chicago, IL, August.
Bernauer, Thomas, Robert Gampfer and Aya Kachi. (2014) European unilateralism
and involuntary burden sharing in global climate politics: A public opinion
perspective from the other side. European Union Politics 15(1):132-151.
Blendon, Robert J., et al. (1997) "Bridging the Gap between the Public's and
Economists' Views of the Economy." The Journal of Economic Perspectives :
105-118.
32
Bolsen T. & Cook F (2008) The polls-trends: public opinion on energy policy: 1974 –
2006. Public Opinion Quartery 72(2):364-388.
Brown, Thomas C., et al. (1996) "Which response format reveals the truth about
donations to a public good?." Land Economics: 152-166.
Brulle, Robert J., Jason Carmichael, and J. Craig Jenkins. "Shifting public opinion on
climate change: an empirical assessment of factors influencing concern over
climate change in the US, 2002–2010." Climatic change 114.2 (2012): 169-188.
Carlsson, Fredrik, and Peter Martinsson. (2001) "Do hypothetical and actual marginal
willingness to pay differ in choice experiments?: Application to the valuation of
the environment." Journal of Environmental Economics and Management 41.2:
179-192.
Champ, Patricia A., and Richard C. Bishop. (2001) "Donation payment mechanisms
and contingent valuation: an empirical study of hypothetical bias." Environmental
and Resource Economics 19.4: 383-402.
Del Río, Pablo, and Xavier Labandeira. (2009) "Climate change at times of economic
crisis." Colección Estudios Económicos 05-09:1-16.
Dietz, Thomas, Amy Dan, and Rachael Shwom. (2007) "Support for Climate Change
Policy: Social Psychological and Social Structural Influences*." Rural Sociology
72.2: 185-214.
Druckman, James N. (2013) "Public opinion: Stunted policy support." Nature Climate
Change 3.7: 617-617.
Egan PJ, Mullin M (2012) Turning Personal Experience into Political Attitudes: The
Effect of Local Weather on Americans’ Perceptions about Global Warming. The
Journal of Politics 74 (3):796-809.
Egenhofer, C. & Alessi, M. EU Policy on Climate Change Mitigation since
Copenhagen and the Economic Crisis CEPS Working Document 380 (2013).
European Commission. Eurobarometer 77.3: Europe 2020, the Financial and Economic
Crisis, European Citizenship and European Values, May 2012. ICPSR34676-v1.
Cologne, Germany: GESIS/Ann Arbor, MI: Inter-university Consortium for
Political and Social Research [distributors], 2013-06-25.
http://doi.org/10.3886/ICPSR34676.v1
George, D., & Mallery, P. (2003). SPSS for Windows step by step: A simple guide and
reference. 11.0 update 4th ed. Boston: Allyn & Bacon.
Gallup Poll Social Series: March Survey (Environment) 2004 – 2010. Available at
http://www.gallup.com/poll/168236/americans-show-low-levels-concern-global-
warming.aspx (Last accessed on January 3, 2015).
Gowdy, John M. (2008) "Behavioral economics and climate change policy" Journal of
Economic Behavior & Organization 68.3: 632-644.
Greeley, Andrew. (1993) "Religion and attitudes toward the environment." Journal for
the Scientific Study of Religion: 19-28.
Guber, D. (2003) Grassroots of the Green Revolution: Polling America on the
Environment MIT Press, Cambridge.
Guth, James L., et al. (1995) "Faith and the environment: Religious beliefs and
attitudes on environmental policy." American Journal of Political Science
39(2):364-382.
Jones, B. & Keen, M. Climate Policy and the Recovery. IMF Staff Position Note
SPN/09/28 ( 2009).
33
Kahn, M. E. & Kotchen, M. J. Business Cycle Effects on Concern about Climate
Change: The Chilling Effect of Recession. Climate Change Economics 2, 257-273
(2011).
Kollmuss, Anja. & Julian Agyeman (2002) “Mind the Gap: why do people act
environmentally and what are the barriers to pro-environmental behavior?.”
Environmental Education Research 8(3):239—260.
Krosnick, Jon A., and Bo MacInnis. "Trends in american public opinion on global
warming policies between 2010 and 2012." Accessed April 10 (2012): 2012.
Myers, Teresa A., et al. 2012. A Public Health Frame Arouses Hopeful
Emotions About Climate Change. Climatic Change 113: 1105 – 1112.
Lax, Jeffrey R., and Justin H. Phillips (2009) "Gay rights in the states: Public opinion
and policy responsiveness." American Political Science Review 103.03: 367-
386.
Leiserowitz A (2006) Climate change risk perception and policy preferences: the role
of affect, imagery, and values. Climatic Change 77 (1):45-72.
Markus, Gregory B (1988) “The Impact of Personal and National Economic
Conditions on the Presidential Vote: A Pooled Cross-Sectional Analysis.”
American Journal of Political Science 32(1): 137-154.
O'Connor, Robert E., Richard J. Bord, and Ann Fisher (1999) "Risk perceptions,
general environmental beliefs, and willingness to address climate change." Risk
analysis 19.3: 461-471.
Page, Benjamin I., and Robert Y. Shapiro. (1983) "Effects of public opinion on
policy." American political science review 77(1): 175-190.
Pew Research Center. (2009) “Modest support for “cap and trade” policy. Fewer
Americans see solid evidence of global warming.” Washington, DC: Pew
Research Center.
Pew Research Center. Climate change: Key data points from Pew Research Available
at http://www.pewresearch.org/key-data-points/climate-change-key-data-
points-from-pew-research/ (Last acessed on January 3, 2015).
Poe, Gregory L., et al. (2002) "Provision point mechanisms and field validity tests of
contingent valuation." Environmental and Resource Economics 23.1: 105-131.
Sanders, David. (2000) "The real economy and the perceived economy in popularity
functions: how much do voters need to know?: A study of British data, 1974–
97." Electoral Studies 19(2): 275-294.
Schreiber, James B., et al. "Reporting structural equation modeling and confirmatory
factor analysis results: A review." The Journal of Educational Research 99.6
(2006): 323-338.
Scruggs, L. & Benegal, S. Declining public concern about climate change: Can we
blame the great recession? . Global Environmental Change 22, 505-515 (2012).
Steenbergern M, Brewer P (2004) The Not-So-Ambivalent Public: Policy Attitudes in
the Political Culture of Ambivalence. In: Saris W, Sniderman P (eds) Studies in
Public Opinion. Princeton University Press, Princeton.
Stimson, James A., Michael B. MacKuen, and Robert S. Erikson (1995) "Dynamic
representation." American Political Science Review 89.03: 543-565.
34
Tjernström, Emilia, and Thomas Tietenberg (2008) "Do differences in attitudes explain
differences in national climate change policies?." Ecological Economics 65.2:
315-324.
Unipark. Unipark-online Survey Software for Universities, <http://www.unipark.com>
(2012).
Weidner, Helmut, and Lutz Mez (2008) "German Climate Change Policy A Success
Story With Some Flaws." The Journal of Environment & Development 17.4: 356-
378.
Zuckerman, Alan S., and Martin Kroh. (2006) "The social logic of bounded
partisanship in Germany: A comparison of West Germans, East Germans, and
immigrants." Comparative European Politics 4(1): 65-93.
35
Appendix 1: Using perceived personal and national economic conditions in the past years
As mentioned in the Results section, our survey also included items measuring perceptions of personal
and national economic conditions in the past few years. Using these variables instead of perceived
conditions in the coming years produces essentially the same results. The following regression table
reports the additional results using the past perceived economic conditions. All the other covariates
included in these models are the same as our main empirical results.
Table A1 Regression results: effects of perceived past economic conditions

GERMANY Policy Behavior Willingness
Dependent support to pay
variable (lottery)
(OLS) (OLS) (Logit) Marg
Household past 0.01 -0.01 0.11 0.02
(0.01) (0.01) (0.11) (0.02)
National past -0.003 -0.0003 0.11 0.02
(0.01) (0.01) (0.10) (0.02)
Interest 0.07 0.07 1.20 0.22
(0.05) (0.04) (0.56)** (0.10)**
Risk 0.68 0.41 2.59 0.47
(0.04)*** (0.04)*** (0.53)*** (0.09)***
Female 0.02 0.04 0.49 0.09
(0.02) (0.02)** (0.21)** (0.04)**
Education 0.02 0.03 0.52 0.10
(0.02) (0.01)** (0.20)*** (0.03)***
Income 0.001 -0.002 0.04 0.01
(0.003) (0.003) (0.04) (0.01)
Intercept 0.09 0.29 -5.18
(0.06) (0.05)*** (0.83)***
N 842 1083 1083
R2 0.4 0.2
AIC -418.3 -658.1 1145.6
BIC -380.5 -618.2 1185.5
USA Policy Behavior Willingness
Dependent support to pay
variable (lottery)
(OLS) (OLS) (Logit) Marg
Household past 0.01 -0.002 0.05 0.01
(0.01) (0.01) (0.10) (0.02)
National past 0.03 0.02 0.18 0.03
(0.01)*** (0.01)** (0.10)* (0.02)*
Interest 0.08 0.02 0.84 0.14
(0.03)** (0.04) (0.53) (0.09)
Risk 0.75 0.45 3.85 0.64
(0.03)*** (0.03)*** (0.45)*** (0.07)***
Female 0.02 0.06 0.88 0.15
(0.02) (0.02)*** (0.20)*** (0.03)***
Democrat 0.06 0.001 -0.08 -0.01
(0.01)*** (0.01) (0.15) (0.02)
Education 0.004 0.003 0.22 0.04
(0.01) (0.008) (0.09)** (0.02)**
Bornagain -0.01 0.03 0.13 0.02
(0.02) (0.02) (0.24) (0.04)
Income 0.0002 0.002 0.04 0.01
(0.002) (0.003) (0.03) (0.01)
36
Intercept -0.11 0.17 -5.79
(0.03)*** (0.03)*** (0.58)***
N 699 973 973
R2 0.8 0.4
AIC -464.4 -557.1 901.3
BIC -418.9 -508.3 950.1
Appendix 2:Robustness checks using an alternative measure for Behavioral Intention
One of our dependent variables, Behavioral Intention, is a composite variable created from four survey
items via CFA. As mentioned in Footnote 7, however, the wording of the second item “personal_2” (see
Table 1) can potentially be interpreted as a knowledge question rather than a behavioral intention
question.
Our original intention was to gauge the perceived likelihood that a respondent “would avoid
activities that emit carbon dioxide” because it would “contribute to solving the problem of global
warming.” The reason we chose what we thought was a milder expression was to alleviate the potential
social desirability effect. We suspected that operationalizing this item by a direct question such as “do
you intend to avoid activities that emit…” might induce affirmative responses too easily because
responding this question negatively would simply be too strong.
That said, to take extra precautions, we have conducted an additional set of robustness analyses by
dropping this item “personal_2” from the Behavioral Intention variable. More specifically, the following
are the additional analyses we have conducted for each country;
(1) A new set of confirmatory factor analysis (CFA) to create the revised Behavioral Intention
variable using 3 (instead of the original 4) survey items.
(2) The descriptive statistics of the revised Behavioral Intention.
(3) OLS regression analyses using the revised Behavioral Intention as a dependent variable.
Long story short, the reanalysis shows that dropping “personal_2” does not lead us to qualitatively
different conclusions from our original analyses. The CFA (Table A2) and the descriptive statistics
(Table A3) are extremely similar between the two versions of Behavioral Intention. In the regression
analysis (Table A4), there were two slight differences between the two versions. One is that, for
Germany, a borderline significant variable, “Interest”, in the original version has now become
statistically “insignificant” by the conventional 90% significance threshold. The other difference is that,
for the US, the perceived future household economy has now become statistically significant at the 90%
confidence level.
And yet, these differences seem to confirm the robustness of our initial empirical results; they do
not affect our main conclusions that, if anything, (a) the perceived negative national economic
conditions tend to reduce public support for governments’ climate mitigation efforts (which is especially
prominent in the US), and that (b) the said negative effects are miniscule both in Germany and the US
compared to the positive effect of higher climate risk perceptions on policy support.
Table A2. The results of CFA for the alternative Behavioral Intention variable

alpha loadings
37
Variable
Intention
(Original)
cartype 0.58 0.66
Intention
(Alternative)
cartype 0.56 0.69

alpha loadings
Variable
Intention
(Original)
cartype 0.66 0.57
Intention
(Alternative)
cartype 0.65 0.58
Table A3. Descriptive statistics of the original and alternative Behavioral Intention
Germany
Behavioral Intention (Original): Mean = 0.63, sd = 0.20
Behavioral Intention (Original): mean = 0.63, sd= 0.22
USA
Behavioral Intention (Original): Mean = 0.53, sd = 0.25
Behavioral Intention (Original): mean = 0.53, sd= 0.27
38
Table A4. The regression results
GERMANY Behavior Behavior

Dependent (original) (Alternative)
variable
(OLS) (OLS)
Household -0.01 -0.01
(0.01) (0.01)
National -0.01 -0.02
(0.01)* (0.01)**
Interest 0.07 0.07
(0.04)* (0.05)
Risk 0.42 0.38
(0.04)*** (0.04)***
Female 0.04 0.04
(0.02)** (0.02)**
Education 0.03 0.05
(0.01)** (0.02)***
Income -0.001 -0.003
(0.003) (0.004)
Intercept 0.32 0.37
(0.05)*** (0.06)***
Log-likelihood
N 1083 1083
R2 0.2 0.2
AIC -667.0 -358.4799
BIC -627.1 -318.5799
USA Behavior Behavior
Dependent (original) (Alternative)
variable
(OLS) (OLS)
Household -0.009 -0.02*
(0.01) (0.01)
National 0.03 0.04
(0.01)*** (0.11)***
Interest 0.03 0.03
(0.04) (0.05)
Risk 0.44 0.39
(0.03)*** (0.04)***
Female 0.06 0.06
(0.02)*** (0.02)***
Democrat -0.004 -0.01
(0.01) (0.01)
Education 0.003 0.008
(0.008) (0.01)
Bornagain 0.03 0.02
(0.02) (0.02)
Income 0.002 0.004
(0.003) (0.003)
Intercept 0.16 0.18
(0.04)*** (0.05)***
Log-likelihood
N 973 973
R2 0.5
AIC -572.8 -196.1577
BIC -524.0 -147.3539
39
Appendix 3: The demographics differences between the DKs and the observed in the Attitude
outcome variable
As noted in the main text (p.17), the number of observation (N) is significantly lower for the composite
Attitude measure, compared to the other two outcome measures. This is due to the higher number of
“don’t know” (DK) responses in survey items used for the Attitude variable. We allowed for “don’t
know” in most of our survey questions, in high hopes that respondents would not randomly choose
certain categories when they cannot (or feel like they cannot) form a specific opinion on the given topic.
To our surprise, survey questions concerning what the German or US government is/should be doing to
combat climate change generated quite large numbers of DKs.
These unobserved cases are dropped from the rest of the empirical analyses. Strictly
speaking, this procedure is justifiable if these observations are “missing at random” (MAR) or “missing
completely at random” (MCAR) (Rubin 1976; Heitjan and Basu 1996). However, since the number of
missing observations is large for Attitude, we suspected that there might be systematic differences in the
demographic characteristics between the DKs and the observed. In this section, we provide the results
from additional balance tests between the missing and observed groups in Attitude, along several
demographic variables that later appear in our regression analyses.
In sum, our sample balance test show that the DKs and the observed are indeed slightly
different demographic characteristics, implying that the missingness in Attitude is strictly speaking non-
ignorable. Although the surveys are already done, we hope that this section provides a sufficient
explanation for the potential methodological issues associated with the missing observations. As a result
of these analyses, in our interpretations of the regression results, we have shifted weights more toward
the results for the other two outcome variables (Behavioral Intention and Willingness To Pay) than for
the Attitude variable.
The following table (Table A1) summarizes the results of the two-sample t-tests. On four of the
five main demographic characteristics—gender, education, partisanship, and income, the null hypothesis
that the mean scores of the two samples are the same cannot be rejected at conventional significance
levels. This means that the two samples are very similar on these three dimensions. However, in terms of
age, respondents of Experiment 1 turn out to be about 2 years older on average.
Table A5. Balance statistics: comparison of the mean demographics between the missing (“Don’t
Know”) and observed cases in Attitude
Difference = Two-sample t-
DKs - test*
Observed
USA N(DK) = 556, N(Observed) = 944
Household 0.13 0.01 Perceived future household economic

conditions are better among DKs.
National 0.24 0.00 Perceived future national economic

conditions are better among DKs.
Interest -0.15 0.00 Interest/knowledge regarding global
warming is lower among DKs.
Risk 0.09 0.00 Climate risk perceptions are higher among
DKs.
Female 0.20 0.00 DKs include more female respondents.
Democrat 0.07 0.03 DKs include more democrats.
Education -0.34 0.00 The education level among DKs is lower.
40
Bornagain 0.005 0.85 Difference not statistically significant
Income -0.97 0.00 Income among DKs is lower.
Germany N(DK) = 413 , N(Observed) = 1007
Household 0.03 0.63 Difference not statistically significant

National 0.005 0.95 Difference not statistically significant
Interest -0.12 0.00 Interest/knowledge regarding global
warming is lower among DKs.
Risk -0.02 0.20 Difference not statistically significant
Female 0.19 0.00 DKs include more female respondents.
Education -1.59 0.06 The education level among DKs is lower.
Income -0.77 0.00 Income among DKs is lower.
*Probability that the F score exceeds the critical value at the 95% confidence level.
References (Appendix)
Heitjan, Daniel F., and Srabashi Basu. "Distinguishing “missing at random” and “missing completely at
random”." The American Statistician 50.3 (1996): 207-213.
Rubin, Donald B. "Inference and missing data." Biometrika 63.3 (1976): 581-592.
41
View publication stats

Da Li Su Pesimisti U Pravu

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Da Li Su Pesimisti U Pravu

Uploaded by

Copyright:

Available Formats

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

Economic Conditions, Risk Perception and Climate Policy Preferences

Article in SSRN Electronic Journal · January 2014

Aya Kachi Thomas Bernauer

SEE PROFILE SEE PROFILE

Spatial Methods View project

Political Economy of Food System Transformation View project

The user has requested enhancement of the downloaded file.

Accepted for publication at Ecological Economics: Tuesday, March 10,

Aya Kachi (ETH Zurich)*

Thomas Bernauer (ETH Zurich)

Robert Gampfer (European Commission, Joint Research Centre)

Originally coined by James Carville, the strategist of Bill Clinton’s successful

Various social polls provide macro-level overviews of this issue. At first

2. Effects of economic conditions and risk perceptions on policy

These studies provide a good starting point; however we identify four

2-1. Level of analysis — citizens

2-2. Explaining policy support, rather than salience and concerns

The conceptual difference between preference and salience is important.

Therefore, it is necessary to measure individuals’ policy preference as a

2-5. Theoretical framework

The second is the significance of the economic effect as a whole compared to

3. The empirical strategy: Evaluating the effects of perceived

Some of these variables are not directly observed, or can be captured by a

Moreover, for the outcome variable measuring climate policy support, we

We designed and implemented our original nationally representative surveys in

YouGov constructs nationally representative samples by first interviewing a

3-2. Outcome variables

The third response variable, Willingness to Pay (WTP) is a single-item

Variable Survey Survey item wording Response scale

By constructing broad policy-attitude measures with these two approaches—

3-3. Key explanatory variables

The second key determinant, climate risk perceptions can be measured in

Table 3: List of survey items used to construct measures of interest/knowledge levels

4. Descriptive statistics of the policy-attitude measures and aggregated

4-2. Aggregated explanatory variables: Risk and Interest

5. Regression analysis and the results

Figure 2 summarizes the regression results focusing on the main explanatory

Figure 2. Regression results for the three key explanatory variables

Attitude, (Germany) Attitude, (USA)

Behavior, (Germany) Behavior, (USA)

Willingness To Pay, (Germany) Willingness To Pay, (USA)

GERMANY Attitude Behavior Willingness

As for demographic characteristics, if we look at Behavior and WTP

Table 8: Regression Results: Effects of Perceived Future Economic Conditions

GERMANY Attitude Behavior Willingness

Log-likelihood -387.8 280.0 -387.8

After an introduction to the topic of global warming, participants were

Table 9: Informational treatments in the survey-embedded experiment

Introduction (for all respondents)

No further information treatment.

Treatment group 2: good economic news

A useful future investigation would be to assess which explanatory model

Our research also contributes to the existing literature by using a more

In summary, our research injects a dose of skepticism into the conventional

Table A1 Regression results: effects of perceived past economic conditions

Appendix 2:Robustness checks using an alternative measure for Behavioral Intention

GERMANY Survey item N Cronbach’s Eigenvalue* Factor Uniqueness

US Survey item N Cronbach’s Eigenvalue* Factor Uniqueness

Behavioral Intention (Original): Mean = 0.63, sd = 0.20

Behavioral Intention (Original): mean = 0.63, sd= 0.22

Behavioral Intention (Original): Mean = 0.53, sd = 0.25

Behavioral Intention (Original): mean = 0.53, sd= 0.27