Available via license: CC BY 4.0
Content may be subject to copyright.
1
A Finite Pool of Worry or a Finite Pool of Attention? Evidence and Qualifications
Matthew R. Sisco1, Sara M. Constantino2, Yu Gao3, Massimo Tavoni4, Alicia D. Cooperman5,
Valentina Bosetti4,6, and Elke U. Weber2
1Columbia University
2Princeton University
3Peking University
4European Institute on Economics and the Environment
5Texas A&M University
6Bocconi University
Keywords: Finite Pool of Worry, COVID-19, affect generalization, big data
Acknowledgements: This research was funded by 1) the European Research Council under the European
Community's Programme "Ideas" — Call identifier: ERC-2013-StG/ERC grant agreement no. 336703—
project RISICO "Risk and uncertainty in developing and implementing climate change policies", 2) the
cooperative agreement NSF SES-1463122 awarded to the Center for Research on Environmental
Decisions, and 3) NSF Grant, SES-2030800 "RAPID: Public Responses to Personal and Societal Risk:
Attitudes and Behavior on COVID-19 and Global Change".
2
Abstract
The Finite Pool of Worry (FPW) hypothesis states that humans have finite emotional resources for
worry, so that when we become more worried about one threat, it can decrease worry about other
threats. Despite its relevance, no conclusive empirical evidence for the hypothesis exists. We leverage
the sudden onset of new worries introduced by the COVID-19 pandemic as a natural experiment to test
the FPW. In six metropolitan areas across three countries (USA, Italy, and China) we assessed social
media attention, news attention, self-reported attention, and self-reported worries about various
threats (climate change, terrorism, economy, and unemployment) throughout the pandemic. As
attention to and worry about COVID-19 increased, we find that attention to climate change decreased
but that worry about it did not. Results are confirmed by further analysis with a large, and nationally
representative U.S. sample. We find some perceived similarity between COVID-19 and climate change,
but this does not fully explain the positive relationship in worry we see between them. We also find that
more negative personal experience with COVID-19 is positively associated with climate change worry
even while controlling for relevant covariates. We lastly examine the aggregate effect of COVID-19
worry on support for climate policies and find that greater COVID-19 worry is associated with more
cross-partisan support for climate change policies, even when controlling for political ideology and other
covariates. In summary, our findings suggest that while there appears to be a Finite Pool of Attention to
threats, we do not see evidence of a Finite Pool of Worry.
3
The Finite Pool of Worry (FPW) hypothesis (Weber, 2006) states that since humans have
1
cognitive resource constraints (Simon, 1957), they may also have finite emotional resources. Worrying
2
more about one threat may exhaust our resources for worry and make us proportionately less worried
3
about other threats. A contrasting hypothesis, Affect Generalization (Johnson and Tversky, 1983), draws
4
on the emotions-as-information literature: increased worry about one threat can be (mis)attributed to
5
other threats or, alternatively, transferred via associative networks to other threats. It predicts that
6
worrying more about one threat makes us more worried in general and/or more worried about related
7
threats.
8
Whether or not humans have a finite pool of worry has critical implications for climate change
9
communications and policymaking. When new threats emerge, such as the COVID-19 pandemic, does
10
worry about climate change decrease? If this were the case, those working to organize the public to
11
taking action against climate change would have good reason to slow their efforts until citizens regain
12
the capacity to worry sufficiently. In contrast, if new threats have no adverse effect on climate worries
13
or even a positive effect, we may do well by seizing policy opportunities created by new threats. This
14
could mean the introduction and implementation of longer-lasting changes directed at climate change
15
mitigation, such as green infrastructure programs.
16
Past work has suggested limited support for the FPW hypothesis, but no empirical study to date has
17
demonstrated ample evidence for or against it. The finding by Linville and Fischer (1991) that people
18
express a preference for experiencing multiple negative events not simultaneously but separated in time
19
suggests finite emotional resources. Linville and Fisher's results show that people believe meta-
20
cognitively that they have finite resources for worry, but not that new worries replace existing worries.
21
In an experiment examining the FPW hypothesis more directly, Hansen et al. (2004) found that
22
increased climate change worries by Argentinian farmers after a seminar on the topic were associated
23
with reduced reports of political worries. This result is consistent with the FPW hypothesis but is notably
24
limited by the study's small sample size (N=14). Nakayachi et al. (2015) reanalyzed survey data collected
25
from Japanese citizens and evaluated measurements taken four years before an earthquake and nuclear
26
disaster occurred compared to ten months after. They found that participants were more worried about
27
both earthquakes and nuclear-related threats and less worried about other threats after compared to
28
before.
29
Some scholars have pointed to the drop in societal concern about environmental problems that
30
occurred around the Great Recession of 2008 as a real-world example of the FPW effect (Kohut et al.,
31
2009; Whitmarsh, 2011). It has been proposed that increased worry about economic problems
32
decreased worry about environmental ones, but the results are correlational and have also been
33
attributed to changing rhetoric by elected officials in the Republican party around this time
34
(Mildenberger and Leiserowitz, 2017; Kenny 2018). In sum, the empirical evidence supporting the FPW
35
hypothesis is notably limited due to power, sample, or design issues.
36
New worries may increase the magnitude of existing worries through a kind of spillover effect or affect
37
generalization, as reported by Johnson and Tversky (1983). Their results show that increases in
38
perceived risk about one threat can generalize to other threats. To the best of our knowledge, no
39
research to date has comprehensively examined whether the introduction of worries about a new risk
40
increases or decreases the magnitude of worries about preexisting risks.
41
4
In the current paper, we present three studies related to the FPW hypothesis. We address the question
42
of whether we have a finite or a growing pool of worry by analyzing various sources of data on worry
43
about climate change and other threats before and during the COVID-19 pandemic. We leverage the
44
unexpected emergence of the COVID-19 pandemic in January 2020 as a natural experiment that permits
45
us to examine the causal effects of a novel threat on attention to and worry about climate change,
46
terrorism, economic problems, unemployment, and immigration. We focus on in-depth analyses on the
47
threat of climate change.
48
Attention and exposure to information about a threat are closely tied to worry about it (Grupe and
49
Nitschke, 2013; King et al. 2017). For example, Mesch et al. (2013) report that participants' attention to
50
the media strongly predicted their concerns about infection during the Swing Flu outbreak of 2009.
51
Thus, we extend our analysis of the effects of COVID-19 on both attention to and worry about climate
52
change and other threats.
53
In Study 1 we analyze how macro-level societal attention to COVID-19 impacts attention to other threats
54
portrayed on social and news media in three countries (US, Italy, China). We then analyze the effects of
55
COVID-19 daily regional case counts on individual-level self-reported attention to and worry about other
56
threats. Next, in Study 2, we utilize a multi-national longitudinal survey (run from December 1st, 2019
57
through August 1st, 2020) to examine the effects of new worries on pre-existing worries about climate
58
change and other threats. Across these analyses, we find that attention to COVID-19 crowds out
59
attention to climate change and other unrelated threats, consistent with a variant of the FPW
60
hypothesis, the Finite Pool of Attention (FPA) hypothesis. As one would expect, for highly related threats
61
(e.g., economic and unemployment threats) attention to COVID-19 is positively correlated. However,
62
when examining self-reported worry, we find that increased COVID-19 worry is associated with
63
increased worry about climate change as well as related threats.
64
In Study 3, we assess these results' robustness and explore these relationships further with a second
65
large survey run in the US (administered in April 2020). In this nationally representative quota sample,
66
we added questions that assessed the perceived association between COVID-19 and climate change
67
risks. We test whether the positive relationship we find between COVID-19 and climate worry could be
68
due to perceived similarities between the two threats, controlling for other factors, including political
69
affiliation. While there is some perceived association in the US public between these threats, this
70
perceived similarity does not fully account for the positive relationship we find between worry about the
71
two risks. We also examine if personal experience with COVID-19 affects climate worry.
72
Increases in worry about climate change are especially important if they translate into a greater
73
willingness to support climate action. We extend our analysis to study the support of climate policy
74
measures using data from both survey datasets. We find that increased COVID-19 worry and exposure
75
to COVID-19 is associated with stronger support for climate policies. This effect is seen among liberals
76
and conservatives but strongest for conservatives, suggesting potential cross-partisan support for green
77
recovery measures.
78
The analyses of the social media data, news data, and survey data to examine the FPW hypothesis
79
reported in this paper were preregistered before data collection. We did not specifically anticipate
80
COVID-19 occurring at the time of preregistration for the multi-national panel study.
1
Still, we designed
81
1
The preregistration can be found here: https://osf.io/fvzx3/ where our plans for testing the FPW hypothesis are show in
research question #10. Our preregistered power analysis can be found here: https://osf.io/4b3t7/
5
data collection to evaluate the FPW hypothesis by leveraging natural changes in worries introduced by
82
new threats occurring over time.
83
84
Study 1: Do new threats crowd out attention to other threats on social media and/or news media?
85
Methods.
86
Social media data. We collected social media data from the Twitter API by connecting to the real-time
87
streaming endpoint to ingest samples of Tweets from our target locations continuously (every three
88
minutes). Our social media data range from December 1st, 2019, through August 1st, 2020. We identified
89
Tweets from six target cities (New York, Dallas, Milan, Rome, Beijing, Shanghai) by specifying boundary
90
boxes encompassing each city's inner areas and metropolitan areas. We analyze a large sample of 17.6
91
million Tweets across all target areas.
92
In China, it is common to use the social media platform Weibo instead of Twitter. To address this, we
93
extracted daily counts of Weibo messages based on our keyword lists used to analyze the Twitter data
94
and verified that the Weibo records correlate highly with the Chinese Twitter data we analyze (r = .87,
95
95%CI=[.82, .9], p < .001 for Beijing and r = .94, 95%CI=[.91, .95], p < .001 for Shanghai COVID-19
96
message counts).
97
We measure the attention to different threats in social media messages by quantifying the percent of all
98
messages that contained a keyword indicating that the message discussed each threat, t, in a week w,
99
and area a, (see Supplemental Note 1 for the keywords and methodology we used).
2
100
101
News data. We analyze news data relevant to the US, Italy, and China that we collected on a daily basis
102
from December 1st, 2019 through August 1st, 2020. We extracted the most popular news articles for the
103
languages most commonly spoken in these countries (English, Italian, and Mandarin Chinese,
104
respectively) from the Google News API four times per day at evenly spaced intervals. We collected
105
news data based on language rather than country of origin. It as it is common for citizens to read news
106
from news agencies based in different countries (such as Americans reading BBC) as long as the
107
information is delivered in a common language. In total, we analyze 82,976 news articles across
108
countries.
109
As with the social media data, we quantify the attention given to different threats each week as the
110
percent of articles that contained a keyword for that threat over all articles published that week in each
111
country. The keywords we used for each threat are the same as those used with the social media data.
112
Results. In Figure 1, panel A, we show the relationships between attention to COVID-19 and attention to
113
other threats (climate change, terrorism, economy, and unemployment) on social media. The evident
114
pattern is that COVID-19 attention is associated with increased societal attention to the threats directly
115
tied to it (problems with the economy and unemployment) and decreased attention to threats largely
116
unassociated with it (climate change and terrorism). This pattern of results is seen across all three
117
countries. We conducted a regression analysis to statistically evaluate these patterns and find that the
118
2
We excluded outliers, defined as data points greater than four standard deviations from the mean value for each threat in
both the social media and news media analyses.
6
effects of COVID-19 attention on climate change and terrorism attention are significantly negative
119
( . In contrast, the
120
effects of COVID-19 attention on economic problems and unemployment are significantly positive
121
( . The full regression
122
table is shown in Supplemental Table 4.
123
124
125
Figure 1. COVID-19 attention on social media and news media crowds out attention to climate and
126
terrorism. Units are percentages of all social media messages / news articles aggregated weekly. Data
127
points in panel A are aggregated by city and in panel B aggregated by language. The lines show the
128
bivariate regression fits and the shaded regions show the 95% CIs.
129
130
Figure 1, panel B shows the relationships between attention to COVID-19 and attention to climate
131
change and other threats covered by news media. Interestingly, the pattern is essentially the same as
132
found with the social media data. COVID-19 attention in news articles is associated with increased
133
societal attention to the threats it is directly tied to (problems with the economy and unemployment),
134
and decreased attention to threats not directly associated with it (climate change and terrorism). This
135
pattern of results is found across the three countries. We conducted a regression analysis to statistically
136
evaluate these patterns and find that the effects of COVID-19 attention on climate change and terrorism
137
attention are significantly negative (
138
. The effects of COVID-19 attention on economic problems and unemployment are
139
A
B
7
significantly positive (
140
. The full regression table is shown in Supplemental Table 5.
141
142
Study 2: Examining the FPW and FPA with multi-national survey data.
143
Methods.
144
We surveyed residents from the New York City, Dallas, Milan, Rome, and Beijing and Shanghai
145
metropolitan areas. The data were collected daily from December 1st, 2019 (pre-COVID) through August
146
1st, 2020. This is the same time window as used for examining social media and news media data in
147
Study 1. We sampled 15,480 unique participants, with 2,302 participants taking the survey a second
148
time, one month after their first responses. Sample sizes were evenly distributed across cities. In the
149
analyses below, we analyze different subsets of this full sample, and we note the sample sizes used in
150
each analysis in the regression tables. Participants were recruited through an online-survey research
151
firm, which ensured that the sample's demographic distributions were balanced on age, gender, and
152
political ideology.
153
Our dependent variables are self-reported worries, thoughts, and discussions about climate change,
154
terrorism, economic problems, and unemployment. To measure worries, we showed participants a short
155
description of each threat. We then asked, "How worried are you about [threat]?" with response
156
options including Extremely worried, Very worried, Somewhat worried, Not very worried, and Not at all
157
worried. To measure thoughts, we asked, "How often did you think about [threat] over the past few
158
days?". To measure discussions, we asked, "How often did you discuss [threat] with your friends and
159
family over the past few days?" Response options for quantifying thoughts and discussions both
160
included Several times, Once, and Not at all.
161
We measured age, income bracket, education, and ideology and treat them as continuous variables
162
included as controls in the regressions. Ideology is coded so that higher values indicate more liberal
163
identification. Gender is modeled as a dummy variable, with female participants' values set to 1. The
164
exact wording of these questions can be found in Supplemental Table 6.
165
COVID-19 cases data. We link our survey data with daily regional counts of COVID-19 cases as a proxy
166
for objective COVID-19 risk. Our COVID-19 cases data for the US and China come from the Johns Hopkins
167
University Center for Systems Science and Engineering (CSSE) public repository (Dong et al., 2020). Our
168
cases data for Italy come from the Italian government’s Civil Protection Department (Morettini et al.
169
2020). These datasets have been actively maintained and widely used during the pandemic. Thus, even
170
though some of the case counts may be adjusted in the future as records are further checked for
171
accuracy, the data we use reflect the information available to journalists and the public at the time.
172
The earliest case record in these data is January 22nd, 2020 so we backfilled case counts on days before
173
this with zero values. The CSSE data provide each region's cumulative count of cases each day, which we
174
transformed to obtain the number of new cases recorded each day. Sometimes there is a delay in cases
175
being recorded (such as weekend cases being recorded on Monday) so we transform each raw value
176
with a moving average with a span of two days before and after. The mean number of cases per day in
177
our data varied substantially across regions (
178
179
) so we
180
8
log transformed these values to make them more comparable (adding 1 first to avoid log(0)). The log
181
transformation also addresses that the distributions of raw case counts are highly skewed. We use raw
182
case counts rather than population-weighted case counts in our main analyses. As a robustness check,
183
we also ran the analyses with population-weighted case counts and obtain virtually the same results.
184
Results.
185
First, we use a series of regression models to evaluate the effects of objective COVID-19 risk on self-
186
reported worries, thoughts, and discussions about two largely unrelated threats (climate change and
187
terrorism) and two directly related threats (problems with the economy and unemployment). We use
188
self-reported thoughts and discussions as measures of individual-level attention to these threats. We
189
take each outcome variable, for example, climate worry, and regress it on (log) COVID-19 daily regional
190
cases, demographic control variables (income, age, gender, education, and ideology), geographic fixed
191
effects, and time (day) fixed effects. We only include participants' first responses (no-repeat responses)
192
in this analysis. The sample sizes per regression vary (ranging from 3,771 to 15,271) as some outcome
193
variables were only asked to subsets of participants (climate change items were measured for all
194
participants). The precise sample sizes for each regression can be found in Supplemental Table 1. The
195
coefficient estimates for (log) COVID-19 cases are visualized in Figure 2. The full regression table with
196
estimates for all explanatory variables and outcome variables is shown in Supplemental Table 1.
197
198
199
Figure 2. Worries, discussions, and thoughts about four threats regressed on (log) COVID-19 regional
200
cases. Each value corresponds to a beta coefficient estimate for (log) COVID-19 cases predicting one of
201
the twelve outcome variables shown in each row. The full regression table with estimates for all
202
explanatory and outcome variables is shown in Supplemental Table 1. Error bars depict 95% CIs.
203
204
9
Where COVID-19 risk significantly predicts one of the dependent variables, we interpret this as a causal
205
effect based on the assumption that daily COVID-19 risk is quasi-randomly assigned to participants on
206
any given day. Participants surveyed each day were randomly selected from pre-built lists of survey
207
participants; therefore, the day each participant is included in the study is arguably orthogonal to the
208
counts of daily COVID-19 cases. By having geographic fixed effects, the analysis focuses on within-region
209
variation in COVID-19 cases.
210
Figure 2 shows that objective COVID-19 risk decreased thoughts and discussions about climate change
211
and terrorism, but contrary to the FPW hypothesis COVID-19 risk did not decrease worry about climate
212
change or terrorism, and even increased worry about climate change. In contrast, objective COVID-19
213
risk increased thoughts, discussions, and worries about the related threats of problems with the
214
economy and unemployment. Thoughts and discussions about threats occupy individuals' attention to
215
them, so this pattern of results is consistent with the findings of Study 1. Worries are a different
216
construct and represent inner feelings of concern toward threats, which we find for climate change and
217
terrorism to be either increased or unaffected by introducing a new threat.
218
We also ran this analysis on the samples from each country individually, shown in Supplemental Figure
219
1. We find the patterns of within-country results for Italy and the USA to be highly similar to the pooled
220
pattern shown in Figure 2. In contrast, the Chinese results show mostly nonsignificant differences. This
221
could be due to the risk of COVID-19 being the lowest in China during the time window of our data
222
compared to the US and Italy. Central to our examination of the FPW hypothesis, we find no evidence in
223
any country of COVID-19 cases significantly reducing worry about climate change or terrorism even
224
though they reduce attention to these threats.
225
COVID-19 worry vs. other worries. Using regional daily case counts as our measure of COVID-19 risk has
226
the limitation that participants may perceive different levels of risk given their individual media
227
consumption habits and preconceived beliefs about pandemics. We conduct further analyses to address
228
these limitations, first examining cross-sectional survey data and then analyzing our repeat participants.
229
We conduct a cross-sectional analysis that examines the effects of self-reported worry about COVID-19
230
on worries about other threats. Our sample size for this analysis is N=10,155 as we began fielding the
231
COVID-19 worry question only in March 2020. We regress worry about each of our four target threats
232
on COVID-19 worry, demographic control variables, city (with Shanghai as the intercept), and time fixed
233
effects:
234
235
236
237
The inclusion of ideology as a covariate is essential to our interpretation of the model results. We know
238
a priori that political ideology has a strong association with public opinion on COVID-19 risk perceptions
239
(Calvillo et al., 2020) and the other threats such as climate change (Dunlap and McCright, 2008;
240
McCright and Dunlap, 2016). We include political ideology in the cross-sectional regression models to
241
ensure that this third variable does not account for the positive associations between COVID-19 worry
242
and other worries.
243
10
The results of these analyses are presented in Table 1. We see that COVID-19 worry positively and highly
244
significantly predicts worries about both unrelated (climate change and terrorism) and related
245
(economic problems and unemployment) threats.
246
247
Table 1. COVID-19 Effects on Other Worries
DVs:
Climate
worry
Terrorism
worry
Economy
worry
Unemploy.
worry
(M1)
(M2)
(M3)
(M4)
COVID-19 worry
.36***
.41***
.30***
.33***
Income bracket
.02***
.02***
0.004
-.01*
Age
-.01***
-.01***
-0.00**
-.01***
Female
0.00
.10***
.04*
.03
Education
-0.004
-.01
.01
-.01
Ideology
.13***
-.11***
-.01*
-.01
Week
0.001***
0.001***
0.00
0.001*
Constant
.19**
2.30***
2.34***
2.46***
City fixed effects
✓
✓
✓
✓
Observations
10,155
10,155
10,155
10,155
Adjusted R2
.18
.18
.16
.21
Note: *P<0.05; **P<0.01; ***P<0.001
248
COVID-19 worry vs changes in other worries. The models presented in Table 1 make use of our cross-
249
sectional survey observations from only the first time each participant completed the survey. Next, we
250
analyze a subset of participants that completed the survey a second time, one month after their first
251
responses. All repeat participants analyzed here reported their COVID-19 worries in both their first and
252
second responses, with which we calculate a within-person change in COVID-19 worry for each
253
participant. All repeat participants were asked in both first and second responses about their climate
254
worries and worries about one (randomly selected) non-COVID threat. We analyze a sample size
255
(N=1,009) of repeat observations of climate worry and sample sizes of 238 through 274 for worries
256
about the other non-COVID threats.
257
The repeated observations allow us to examine the effects of changes in COVID-19 worry on other
258
worries in a within-person analysis. For each repeat participant, we created a change in COVID-19 worry
259
score by subtracting each first observation from the second observation for each participant. This score
260
quantifies the direction of change over time in COVID-19 worry for each participant.
261
Figure 3 shows the positive relationship between within-person change in COVID-19 worry and reported
262
climate worry. We regressed worries about non-COVID threats reported in subjects' second responses
263
on change in COVID-19 worries, while controlling for participants' worries reported in their first
264
response and key demographic variables. In Table 2 we show the regression results.
265
11
266
Figure 3. Climate worry and within-person change in COVID-19 worry. Climate worry shown here is
267
measured on participants' second survey response (approx. one month after their first responses). The
268
line shown is a bi-variate regression fit and the shaded region shows the 95% confidence interval.
269
270
271
Table 2. Non-COVID worries regressed on within-person change in COVID-19 worry
DVs:
Climate
Worry
Terrorism
Worry
Economy
Worry
Unemploy.
Worry
1
2
3
4
Change in COVID-19 worry
.10***
-.06
.15**
.17*
Climate worry (first response)
.71***
Terrorism worry (first response)
.51***
Economy worry (first response)
.60***
Unemploy. worry (first response)
.60***
Female
-.02
.10
.07
.03
Education
-.03*
.04
.01
-.04
Ideology
.07***
-.03
.06
.02
Constant
.81***
1.44***
1.10***
1.06**
City fixed effects
✓
✓
✓
✓
Observations
1,009
238
274
248
Adjusted R2
.60
.30
.36
.39
Notes: *P < .05; **P < .01; ***P < .001
12
We find that increased COVID-19 worry is associated with significantly higher worries about climate
272
change, the economy, and unemployment. As we would expect, the effects of increased COVID-19
273
worry on economic and unemployment worries are the greatest. We do not see a significant effect on
274
worries about terrorism. The result that positive change in COVID-19 worry is associated with higher
275
climate worry while controlling for prior climate worry is strong evidence that COVID-19 worry does not
276
crowd out other unrelated worries, but can increase them.
277
Our main results from Studies 1-2 are conceptually synthesized in Figure 4. We find support for the FPA
278
hypothesis at societal and individual levels, but do not find support for the FPW hypothesis. Rather we
279
find that worries generalize from new threats to preexisting ones.
280
281
282
Figure 4. Conceptual diagram summarizing the main results of the studies presented here. Plus or minus
283
signs indicate a positive or negative relationship between two constructs.
284
285
Study 3: Examining moderators of COVID-19's effect on climate change worry.
286
To evaluate these results' robustness and explore them further, we analyze a second survey with a
287
nationally representative quota sample matched to US Census data (N=5,059). We begin by replicating
288
the results above showing a positive correlation between worry about COVID-19 and worry about
289
climate change and other threats (the economy, immigration, and illegal immigration). We also
290
introduce two new variables: perceived similarity between the COVID-19 crisis and the climate crisis;
291
and personal experience with COVID-19, which quantifies the self-reported negative exposure
292
participants had to COVID-19. We use the perceived similarity variable to examine the extent to which
293
13
the relationship between COVID-19 worry and climate worry is due to a perceived association between
294
the two threats.
295
Before we conclude that the threat of COVID-19 does not crowd-out worry about other threats, we
296
examine the effects of direct personal experience with COVID-19 on other worries. It is plausible that, as
297
the FPW hypothesis would predict, the threat of COVID-19 decreases climate worry when personal
298
experience with COVID-19 is severe. Thus, we test if personal experience with COVID-19 affects climate
299
worry.
300
301
Methods.
302
Survey data and analyses. We administered this survey between April 16th and 24th using a survey panel
303
provider. The quota sample is nationally representative of age, region, race, ethnicity, gender,
304
education, income, and political affiliation. Quota values are based on the 2018 American Communities
305
Survey 5-Year Estimates Data Profiles, except for political affiliation based on the American National
306
Election Studies 2016 poll online sample.
307
We measure worry about COVID-19 and other issues (climate change, the economy, immigration, and
308
illegal immigration) using a 4-pt Likert-style scale (Very worried, Somewhat worried, Not very worried,
309
Not at all worried). We construct a negative personal experience with COVID-19 index that is increasing
310
in severity and includes whether an individual contracted coronavirus, lost a job due to coronavirus, had
311
their pay or hours cut due to coronavirus, or has close friends or family members who have contracted,
312
been hospitalized or died from coronavirus. The index is the natural logarithm of the sum of the
313
variables (plus 1); it varies from 0 to 2.3 with a mean of 0.77.
314
We accounted for perceived similarity or difference between COVID-19 and climate change by averaging
315
the following two questions into an index ranging from different (0) to similar (4), with a mean of 2.1.
316
“The coronavirus epidemic and global warming/climate change both show that if we don’t act before
317
widespread effects become evident, the situation may escalate out of control,” and “Media reports
318
linking global warming/climate change and coronavirus are misleading because these two situations are
319
very different.” Responses were on a 5-pt Likert scale from Strongly Agree to Strongly Disagree.
320
Results. All regressions below control for age, gender, income, education, party affiliation, ideology, and
321
region (to account for different degrees of spread across the United States). Consistent with the results
322
from Study 2, we find that worry about COVID-19 is significantly positively correlated with worry about
323
climate change and all other issues save for illegal immigration (
324
325
.
326
Additionally, we find that negative personal experience with COVID-19 is positively correlated with
327
worry about all issues (save for illegal immigration), albeit with lower effect sizes (
328
329
). The full regression results are shown in Supplemental Table 2.
330
Next, we look at whether the extent to which people perceive COVID-19 and climate change to be
331
different or similar moderates the relationships between COVID-19 worry (experience) and climate
332
worry. We regressed climate worry on COVID-19 worry (experience) interacted with perceived similarity
333
(see Table 3). We find no interaction between perceived similarity and COVID-19 worry on climate
334
worry, and we find a negative and significant interaction between the effect of perceived similarity and
335
experience with COVID-19 on climate worry. The main impacts of COVID-19 worry and of COVID-19
336
14
experience remain positive and significant when including perceived similarity as a moderator. Among
337
those who perceive them as different issues, higher COVID-19 experience is associated with higher
338
worry. Those who perceive the issues as similar tend to have high climate worry, regardless of personal
339
experience with COVID-19. These results indicate that perceived similarity between the issues of COVID-
340
19 and climate change is not requisite for the positive relationship of COVID-19 worry on climate worry.
341
A visualization of these findings is shown in Supplemental Figure 4.
342
343
Table 3. The role of perceived similarity between COVID-19 and Climate Change
344
DV: Climate Worry
COVID-19 Experience
.21***
COVID-19 Worry
.40***
Perceived Similarity
.50***
.38***
COVID-19 Experience * Perceived Similarity
-.05**
COVID-19 Worry * Perceived Similarity
-0.00
Party-Ind
-.21***
-.14***
Party-Rep
-.38***
-.29***
Ideology
.11***
.09***
Age
-0.00***
-0.00***
Income
-.01
-.01
Education
.05***
.04***
Gender-Male
.04*
.08***
Gender-Other
.12
.33
Constant
1.49***
.62***
Region Fixed Effects
✓
✓
Observations
5,059
5,059
Adjusted R2
.35
.44
Notes: *P < .05 **P < .01 ***P < .001
345
The influence of new COVID-19 worries on climate policy support. Given our results that COVID-19
346
decreases personal and societal attention to climate change but increases personal worry about it, we
347
next examine the aggregate effects of new threats such as COVID-19 on support for public climate
348
policies. We analyze climate policy support measures from both survey datasets to address this
349
question.
350
In the next analysis, we regress a composite climate policy support variable on COVID-19 worry,
351
demographic covariates, and city fixed effects using the survey data from the multi-national panel
352
survey data. This policy variable is the mean response to four support/oppose questions about the
353
15
carbon tax, regulating CO2 as a pollutant, fossil fuel car phase-out, and funding renewable energy
354
research policies. The full text of these questions can be found in Supplemental Note 2. The results are
355
shown in Table 4 in Model 1. We find a positive association between COVID-19 worry and climate policy
356
support, while controlling for ideology and other covariates.
357
We also examined whether this effect is moderated by political ideology. We find a negative interaction
358
between political ideology and COVID-19 worry on policy support, suggesting a stronger positive
359
association between COVID-19 worry and climate policy support for conservatives. This result is shown
360
in Model 2 in Table 4 and shown visually in Supplemental Figure 3. A replication of this result with the
361
nationally representative American survey can be seen in Supplemental Table 7.
362
363
Table 4. The relationship between COVID-19 worry and climate policy support.
DV: Climate Policy Support
M1
M2
COVID-19 worry
.64***
1.43***
Income
.02
.02
Age
-.01***
-.01***
Female
-.10
-.09
Ideology
.61***
1.42***
Education
-.04*
-.04*
COVID-19 worry * Ideology
-.20***
Constant
12.51***
9.32***
City fixed effects
✓
✓
Observations
9,932
9,932
Adjusted R2
.20
.21
Notes: *P < .05; **P < .01; ***P < .001
Next, we analyze three distinct climate policy support measures from the nationally representative
364
American survey introduced in Study 3 and an index of self-reported behavior, and regress these on
365
COVID-19 worry, demographic covariates, and region fixed effects. The policy support and behavior
366
measures include: support for a green infrastructure plan (Green Stimulus; asked of half of our
367
respondents), a recognition of the need for individual support of environmental policies (Act), support
368
for tracking of individual carbon emissions (Tracking), and an index of individual behaviors taken to
369
mitigate the risks of climate change (see Supplemental Note 2 for exact wording). The results are shown
370
in Table 5.
371
We see across both datasets that there is a highly significant positive effect of COVID-19 worry on
372
climate policy support, as well as on self-reported pro-environmental behaviors. In Supplemental Table
373
3 we provide results using data from the first survey broken down by the three constituent countries.
374
Across all three countries, we see a significant positive effect of COVID-19 worry on climate policy
375
support.
376
16
Table 5. Policy support and self-reported pro-environmental behaviors regressed on COVID-19 worry
377
DVs:
Behavior Index
Tracking
Act
Green Stimulus
COVID-19 Worry
.27***
.33***
.42***
.40***
Party-Ind
-.15***
-.28***
-.20***
-.25***
Party-Rep
-.20***
-.27***
-.39***
-.40***
Ideology
.07***
0.00
.18***
.17***
Age
-.01***
-.01***
-0.00***
-0.00*
Income
.02
.01
0.00
0.00
Education
.14***
.03*
.02*
.01
Gender-Male
.16***
.21***
-.14***
-.06
Gender-Other
.13
.18
-.51*
-.07
Constant
-.19
1.74***
2.13***
2.06***
Region Fixed Effects
✓
✓
✓
✓
Observations
5,059
5,059
5,059
2,517
Adjusted R2
.11
.08
.23
.20
Notes: *P < .05 **P < .01 ***P < .001
378
379
Discussion
380
This paper is the first, to our knowledge, to thoroughly examine the FPW hypothesis. We test it at
381
multiple levels of analyses, across several national contexts, and with a robust ensemble of
382
methodological approaches. We first analyzed how societal attention to COVID-19 influenced attention
383
to other threats in social and news media, extending the FPW to a FPA hypothesis. The FPA hypothesis
384
predicts that attention to climate change decreases as attention to a new and unrelated threat such as
385
COVID-19 increases. We find this pattern of results in both social media and news media across three
386
cultural contexts.
387
We then analyzed a large-scale multi-national survey we conducted to examine the FPW and FPA
388
hypotheses at the individual level. Similar to the findings with social media and news media, we find that
389
increased COVID-19 risk crowds out individuals' attention to climate change, which we operationally
390
defined as self-reported discussions and thoughts about the threats. Surprisingly, when we examine self-
391
reported worries about threats, we find the opposite effect: higher COVID-19 risk and worry is
392
associated with increased worry about climate change and other threats.
393
The finding that worry from a new threat can increase worries about preexisting threats is inconsistent
394
with the FPW hypothesis. Rather, it suggests a spillover from one worry to other worries, as predicted by
395
the theory of Affect Generalization (Johnson and Tversky, 1983) — greater worry about one threat will
396
generalize to worry about other ones.
397
To evaluate the robustness of our findings and understand the relationships in more depth, we ran
398
similar analyses on a large, nationally representative American sample. First, we replicated the results of
399
17
Study 2. We find a positive association between COVID-19 worry and climate change worry while
400
controlling for demographic covariates. Additionally, we show that self-reported COVID-19 experience—
401
an index that includes material and physical impacts to self and close friends and family—is also
402
positively related to worry about climate change and other threats.
403
With this sample, we further ask whether the positive relationship between COVID-19 worry and climate
404
worry we find is due to a perceived association between the two threats. We find a positive relationship
405
between the perceived association of climate change and COVID-19 and climate worry. However, the
406
positive relationship between COVID-19 worry and climate worry exists strongly for those who see no
407
similarity between the two threats.
408
Given that we find COVID-19 decreases attention to climate change but increases worry about it, we
409
also examine the important question of how COVID-19 worry affects support for climate policies and
410
pro-environmental behaviors. We analyze climate policy support measures from both surveys and
411
analyze self-reported behaviors from the American survey to address this question. We find that
412
increased COVID-19 worry is associated with increased support for climate policies and self-reported
413
behaviors, while controlling for demographic characteristics such as ideology and political affiliation.
414
Since public views on climate change can be highly associated with political ideology, we tested if
415
ideology moderates the positive association between COVID-19 worry and climate policy support. We
416
find that this positive relationship is cross-partisan, and interestingly is the strongest for conservatives.
417
This pattern of results could be due to conservatives' tendency to typically report lower worry about
418
climate change and therefore have more room for increases to occur. The partisan divide in attitudes
419
about climate change is a key obstacle for society to collectively mitigate climate change (Mccright &
420
Dunlap 2011). Thus, our finding of worry generalization increasing worry about climate change in
421
conservatives is notable.
422
Across three studies, we shed light on the complex interrelationships of climate change
423
worries/attention with worries/attention to new threats such as COVID-19. At first glance, it may be
424
perplexing that attention toward a threat can decrease while worry about it increases. Attention is
425
necessarily a finite resource at the individual level (Shapiro, 2001). Humans can only attend to a fixed
426
number of stimuli at a time. There necessarily is a limit then on how many threats we can dwell on or
427
address at a time. However, this does not mean that unattended worries have reduced intensity if they
428
are brought to our attention again. Rather, our results suggest that when some worries (such as about
429
climate change) are not attended to due to another new threat dominating our attention, they simply
430
lay dormant. But when they are brought to mind, they show the same or even greater intensity than
431
they did before the new threat was introduced. When a new threat competes for attention with
432
preexisting ones, the capacity for feeling worry is not depleted. The relationship between threats is not
433
competitive: we see positive spillovers between worries.
434
435
Conclusion
436
There are critical implications of our central finding that worries can generalize. This result suggests
437
highly contrasting courses of action compared to the belief that worries crowd out other worries as has
438
been widely discussed and anticipated. If worry about one threat crowds out worry about others, then
439
organizations working to organize the public to mitigate climate change would have good reason to slow
440
18
their efforts when new threats such as COVID-19 are introduced. However, our findings recommend the
441
opposite strategy. Our result that worries generalize, or spillover, to other worries implies that
442
communications and calls to action about climate change can achieve the same or greater success in the
443
context of a new threat, even one that dominates public attention.
444
In summary, our findings suggest that while there appears to be a Finite Pool of Attention to threats,
445
worry begets worry. We find that worry demands and supports action, if not by oneself because of finite
446
attention, then by policymakers. Events such as COVID-19 are unprecedented and tragic periods for
447
many lives and livelihoods. As our results suggest, they may also present moments to introduce new
448
policies to mitigate the even more disastrous looming global crisis of climate change.
449
Ethics Statement
Studies 1-2 were approved by the Columbia University IRB (IRB-AAAS3097), Princeton University IRB
(#11220), Bocconi University Ethics Scientific Office and Peking University IRB (#2019-02). Study 3 was
approved by the Princeton IRB. We obtained informed consent from all participants.
Data Availability Statement
The Twitter data analyzed in the current paper are publicly available through the Twitter API. As per
Twitter’s terms of use, we cannot make available our full dataset of Tweets but can make available the
Tweet IDs upon reasonable request. The news data analyzed in the current paper are publicly available
through the (formerly Google) News API. The survey data analyzed in the current paper are available
from the corresponding author upon reasonable request. The COVID-19 cases data are publicly available
from JHU (https://github.com/CSSEGISandData/COVID-19) and from the Italian Civil Protection
Department (https://github.com/pcm-dpc/COVID-19).
Code Availability Statement
All code used to implement the analyses presented in this manuscript will be added to the
supplementary materials before final publication.
19
References
Calvillo, D. P., Ross, B. J., Garcia, R. J., Smelter, T. J., & Rutchick, A. M. (2020). Political Ideology Predicts
Perceptions of the Threat of COVID-19 (and Susceptibility to Fake News About It). Social
Psychological and Personality Science, 1948550620940539.
Dong, E., Du, H., & Gardner, L. (2020). An interactive web-based dashboard to track COVID-19 in real
time. The Lancet infectious diseases, 20(5), 533-534.
Dunlap, R. E., & McCright, A. M. (2008). A widening gap: Republican and Democratic views on climate
change. Environment: Science and Policy for Sustainable Development, 50(5), 26-35.
Grupe, D. W., & Nitschke, J. B. (2013). Uncertainty and anticipation in anxiety: an integrated
neurobiological and psychological perspective. Nature Reviews Neuroscience, 14(7), 488-501.
Hansen, J. W., Marx, S. M., & Weber, E. U. (2004). The role of climate perceptions, expectations, and
forecasts in farmer decision making: the Argentine Pampas and South Florida: Final Report of an
IRI Seed Grant Project.
Johnson, E. J., & Tversky, A. (1983). Affect, generalization, and the perception of risk. Journal of
personality and social psychology, 45(1), 20.
Kenny, J. (2018). Environmental protection preferences under strain: an analysis of the impact of
changing individual perceptions of economic and financial conditions on environmental public
opinion during economic crisis. Journal of Elections, Public Opinion and Parties, 28(1), 105-124.
King, G., Schneer, B., & White, A. (2017). How the news media activate public expression and influence
national agendas. Science, 358(6364), 776-780.
Kohut, A., Doherty, D. C., Dimock, M., & Keeter, S. (2009). Fewer Americans see solid evidence of global
warming. Pew Research Center for the People & the Press, Washington DC.
Linville, P. W., & Fischer, G. W. (1991). Preferences for separating or combining events. Journal of
personality and social psychology, 60(1), 5.
Mccright, A. M. and Dunlap, R. E. (2011). The Politicization Of Climate Change And Polarization In The
American Public'sPublic’s Views Of Global Warming, 2001-2010. Sociol. Q., vol. 52, no. 2, pp.
155–194, doi: 10.1111/j.1533-8525.2011.01198.x.
McCright, A. M., Dunlap, R. E., & Marquart-Pyatt, S. T. (2016). Political ideology and views about climate
change in the European Union. Environmental Politics, 25(2), 338-358.
Mesch, G. S., Schwirian, K. P., & Kolobov, T. (2013). Attention to the media and worry over becoming
infected: the case of the Swine Flu (H1N1) Epidemic of 2009. Sociology of health & illness, 35(2),
325-331.
Mildenberger, M. & Leiserowitz, A. (2017) Public opinion on climate change: Is there an economy–
environment tradeoff?, Environmental Politics, 26:5, 801-824, DOI:
10.1080/09644016.2017.1322275
Morettini, M., Sbrollini, A., Marcantoni, I., & Burattini, L. (2020). COVID-19 in Italy: Dataset of the Italian
Civil Protection Department. Data in Brief, 105526.
20
Nakayachi, K., Yokoyama, H. M., & Oki, S. (2015). Public anxiety after the 2011 Tohoku earthquake:
Fluctuations in hazard perception after catastrophe. Journal of Risk Research, 18(2), 156-169.
Shapiro, K. (Ed.). (2001). The limits of attention: Temporal constraints in human information processing.
Oxford University Press. https://doi.org/10.1093/acprof:oso/9780198505150.001.0001
Whitmarsh, L. (2011). Scepticism and uncertainty about climate change: Dimensions, determinants and
change over time. Global environmental change, 21(2), 690-700.
