Coronavirus Perceptions And Economic Anxiety
Thiemo Fetzer, Lukas Hensel, Johannes Hermle, Christopher Roth
CCoronavirus Perceptions and Economic Anxiety ∗ Thiemo Fetzer Lukas HenselJohannes Hermle Christopher RothFirst version: March 6, 2020This version: July 7, 2020Forthcoming, Review of Economics and Statistics
Abstract
We provide one of the first systematic assessments of the development and determinantsof economic anxiety at the onset of the coronavirus pandemic. Using a global dataseton internet searches and two representative surveys from the US, we document a sub-stantial increase in economic anxiety during and after the arrival of the coronavirus. Wealso document a large dispersion in beliefs about the pandemic risk factors of the coron-avirus, and demonstrate that these beliefs causally affect individuals’ economic anxieties.Finally, we show that individuals’ mental models of infectious disease spread understatenon-linear growth and shape the extent of economic anxiety.
Keywords:
Economic Anxiety, Health beliefs, Mental Models
JEL code:
D12, D83, D84, E32 ∗ Thiemo Fetzer, University of Warwick, CEPR, CESifo, Thiemo [email protected]; Lukas Hensel, University of Oxford,[email protected]; Johannes Hermle, University of California, Berkeley and IZA, [email protected]; Christopher Roth, Uni-versity of Warwick, briq, CESifo, Cage, CEPR, [email protected]. We thank the editor and two anonymous referees fortheir insightful comments and suggestions. We also thank Eric Avis, Joshua Dean, Stefano DellaVigna, Jonathan de Quidt, Armin Falk,James Fenske, Marta Golin, Yuriy Gorodnichenko, Johannes Haushofer, Matt Lowe, Andrew Oswald, Daniel Sgroi, Leah Shiferaw, ThomasFerguson, and Dmitry Taubinsky for very useful comments as well as Anna Lane and Ivan Yotzov for excellent research assistance. Fi-nancial support from a grant by the The Institute of New Economic Thinking (Grant INO20-01) is gratefully acknowledged. Lukas Henselgratefully acknowledges financial support from the Wellspring Philanthropic Fund. Ethical approval was received by the Blavatnik Schoolof Government’s Departmental Research Ethics Committee (BSG_C1A-20-16) of the University of Oxford and the Humanities and SocialSciences Research Ethics Committee at the University of Warwick (protocol HSSREC 76/19-20). a r X i v : . [ ec on . GN ] J u l Introduction
The worldwide spread of the novel coronavirus (SARS-CoV-2) (Li et al., 2020; Wu et al.,2020; Zhu et al., 2020) has led to a substantial disruption of global economic activity.This article provides one of the first systematic assessments of the rapid emergence andcausal determinants of economic anxiety at the onset of the coronavirus pandemic, whenthere was large uncertainty about the extent of its economic impact. We focus on howperceptions of pandemic risk factors shape economic anxieties. Understanding the de-velopment and causes of economic anxiety in the wake of the coronavirus pandemic isessential from both a scientific and practical perspective, particularly given recent em-pirical evidence demonstrating that perceptions and expectations about the macroeco-nomic environment substantially shape households’ economic decisions (Bailey et al.,2019, 2018; Coibion et al., 2019a; D’Acunto et al., 2019a; Kuchler and Zafar, 2019).Predicting the development of economic anxiety and assessing its underlying mech-anisms in the context of a pandemic is difficult when relying on historical accounts.Unlike regular economic downturns which begin with a moderate but accelerating de-cline in economic activity, the arrival and rapid global spread of the coronavirus pose arare, sudden shock (Ma´ckowiak and Wiederholt, 2018). Several aspects of human beliefand expectation formation render the environment of the coronavirus pandemic distinctfrom that experienced during a conventional economic downturn. In particular, indi-viduals have difficulty forming beliefs about the future in the wake of infrequent majorevents (Gallagher, 2014; Rabin, 2002). Moreover, when updating their beliefs, individu-als place a disproportionate weight on the most recent events (Malmendier and Nagel,2011), especially when these events are particularly salient (Bordalo et al., 2013; Tverskyand Kahneman, 1973). As a consequence, belief formation may differ substantially inthe unprecedented environment of the coronavirus pandemic as compared to more con-ventional economic shocks. Thus, relative to relying on historical accounts, employingcontemporaneous data provides a more promising approach to assess the evolution ofcontemporaneous economic anxiety. n this article, we collect contemporaneous data to systematically investigate the de-velopment and determinants of economic anxiety at the onset of the coronavirus pan-demic. We study the underlying psychological mechanisms that shape economic anxietyin the environment of a pandemic by assessing the role of beliefs and information aboutpandemic risk factors as well as individuals’ subjective mental models of infectious dis-ease spread.To set the stage for our analysis, we document a rapid increase in economic anx-iety during and after the coronavirus has reached a country. Employing global dataduring the period of massive global spreading in January and February 2020, we showthat Google search intensity for topics indicative of economic anxiety surged substan-tially after the virus has reached a country. To measure economic anxieties directly andin real-time after the arrival of the coronavirus, we conducted two survey experimentswith representative samples of the US population on March 5, 2020 and March 16, 2020.In this 11 day period, the United States saw massive within-country spreading with a26-fold increase in the number of confirmed cases from 176 to 4576. Moreover, publiccommunication of the crisis’ severity had shifted dramatically once the WHO declaredit a pandemic on March 11. The data indicate a substantial increase in economic anxietyafter the arrival of the coronavirus in the United States.The rapid surge in economic anxiety sets the stage to study the underlying informa-tional and psychological mechanisms that shape economic anxiety in the wake of a pan-demic. First, we study individuals’ beliefs about the mortality and contagiousness of thecoronavirus - two key characteristics relevant for assessing pandemic risks and predict-ing the severity of the coronavirus crisis. We elicited these beliefs in our March 5 surveybefore any lockdown measures had been put in place and the crisis had not yet been de-clared a pandemic. We document substantial dispersion in beliefs about both mortalityand contagiousness. Moreover, the median participant overestimates both the mortalityand contagiousness of the virus relative to the upper bound of estimates currently avail-able in the medical literature. We show that beliefs about mortality and contagiousnessare associated with participants’ economic worries about the aggregate economy and heir personal economic situation.To further understand the precise causal relationships between coronavirus percep-tions and economic anxiety, we embedded two experiments in our March 5 survey thatvaried the framing of coronavirus mortality as well as a treatment that studied the roleof information about contagiousness. These real-time experiments allow us to shed lighton the influence of information and its framing in an environment marked by large un-certainty about the future extent of the crisis.The first component of our experiment focuses on the framing of mortality risk. Par-ticipants were either truthfully informed, based on official estimates at the time of thesurvey, that the death rate from the coronavirus is “20 times higher than for the flu”(high mortality treatment) or “5 times lower than for SARS” (low mortality treatment).The wording was chosen to mirror the way such information is commonly communi-cated in the media. We find that participants in the high mortality treatment report sig-nificantly higher concerns, both in a statistical and economic sense, about the aggregateeconomy and their personal economic situation. These results highlight the influence ofthe framing of news on public perceptions and economic expectations in times of highuncertainty (Chong and Druckman, 2007; Prat and Strömberg, 2013).To investigate the effect of information regarding contagiousness, participants in atreatment group were, based on scientific estimates at the time of the experiment (Li etal., 2020; Wu et al., 2020), informed that “approximately 2 non-infected people will catch thecoronavirus from a person who has the coronavirus’.’
Given that 81% of respondents over-estimate this statistic, the information treatment should decrease the perceptions of thecontagiousness of the virus. We find that treated respondents report significantly lowerworries about their personal economic situation. These causal results underscore the rolethat information plays in shaping economic anxiety in an environment characterized bylarge uncertainty and highlight the importance of both factual and targeted communica-tion during health crises (Person et al., 2004; Razum et al., 2003). For instance, the New York Times and The Telegraph compared the coronavirus to the flu and SARS( ; , last accessed April 30 th econd, besides taking information into account, forward-looking individuals alsorely on their subjective mental models of the world to make predictions about the fu-ture (Andre et al., 2019). To understand the role of these mental models in shaping crisisbeliefs and economic anxiety, we elicited participants’ predictions of the growth of a fic-titious disease in our March 16 survey. Consistent with exponential growth bias (Levyand Tasoff, 2016; Stango and Zinman, 2009; Wagenaar and Sagaria, 1975), we documentthat the majority of individuals underestimate the non-linear nature of infectious dis-ease spread. Furthermore, we show that mental models of infectious disease spread aresubstantially associated with participants’ beliefs about the severity of the current coron-avirus crisis: respondents who show a better understanding of non-linear disease spreadanticipate a higher severity of the crisis and display higher worries about the aggregateUS economy.We contribute to the literature by documenting the development and underlying de-terminants of economic anxiety in the wake of a global pandemic. In particular, weprovide novel causal evidence on the impact of information about pandemic risk factorson the formation of economic anxiety. Furthermore, we demonstrate the role subjectivemental models of infectious disease spread play in shaping heterogeneity in economicanxiety. Our paper is most closely related to concurrent work by Binder (2020) who con-ducted a survey using a sample from Amazon Mechanical Turk in early March. Binder(2020) documents cross-sectionally that greater concerns about the coronavirus are as-sociated with higher inflation expectations and more pessimistic unemployment expec-tations. She also studies how information provision about the Fed’s interest rate cut inresponse to the coronavirus affects inflation and unemployment expectations. Our pa-per complements Binder (2020) by documenting how the spread of coronavirus affectseconomic anxieties over time, and by providing both descriptive and causal evidence onhow perceptions of the pandemic risk factors affect economic anxiety. We also relate tosubsequent work studying the impact of the coronavirus on the economy (Adams-Prasslet al., 2020; Bartik et al., 2020; Coibion et al., 2020a; Hanspal et al., 2020).More generally, our work is related to a growing literature investigating the forma- ion of economic sentiment and expectations about the macroeconomy among house-holds and firms (Binder and Makridis, 2018; Coibion and Gorodnichenko, 2012, 2015a,b;Coibion et al., 2019b, 2018; Fuster et al., 2012, 2010; Malmendier and Nagel, 2011). Rela-tive to prior work, our evidence is unique in assessing economic sentiment and its driversbefore and during a historic public health crisis in real time. We particularly relate to theliterature studying the role of information in shaping economic sentiment and behavior(Armona et al., 2018; Bailey et al., 2019, 2018; Binder and Rodrigue, 2018; Coibion et al.,2020b; D’Acunto et al., 2019a; Roth and Wohlfart, 2020). We also relate to the literaturestudying the role of cognitive processes in forming economic sentiment and macroe-conomic expectations (Andre et al., 2019; D’Acunto et al., 2019b). We add to this litera-ture by highlighting the importance of subjective mental models about infectious diseasespread for shaping economic anxiety in the wake of a rare, unexpected, and unfamiliarpublic health shock.Finally, we also contribute to the broad literature on the perception of health risks(Carbone et al., 2005; Fortson, 2011; Heimer et al., 2019; Kerwin, 2018; Oster et al., 2013).While existing evidence has primarily focused on individuals’ beliefs about risks to theirown health (Kan and Tsai, 2004; Liu and Hsieh, 1995; Winter and Wuppermann, 2014),we contribute to this literature by providing new evidence on the perception of factorsrelevant to pandemic in addition to individual risks. We begin by documenting the emergence of economic anxieties at the onset of the coro-navirus pandemic. First, we focus on the period of the initial global spread of the coro-navirus during January and February 2020. Leveraging global data on Google searchesindicative of economic anxieties, we study the evolution of economic anxiety during thearrival of the coronavirus in a country. Next, we use survey data to study the develop-ment of economic anxiety within the US after the arrival of the coronavirus. .1 Observational Evidence from Internet Searches during GlobalSpread Data and Empirical Specification:
We leverage data on internet search intensityfrom Google Trends. These data have been used in the past to detect influenza epi-demics (Ginsberg et al., 2009) and to nowcast economic activity (Choi and Varian, 2012). The Google Trends platform provides an interface to query search data, providing foreach query a measure of search intensity scaled from 0 to 100, with 100 representing thehighest proportion among the queried terms within a selected region and time frame.Google Trends queries can be constructed based on individual search terms or searchtopics which encompass groups of related individual search terms. We employ queriesby search topics, an approach that has the advantage of capturing a broader set of searchterms and not requiring any translations across languages.To study the development of economic anxiety, we extracted Google search activityfor the topics "Recession" and "Stock Market Crash" for a total of 194 countries and ter-ritories listed in Online Appendix Table A.1. We also leverage data on the search topics"Survivalism" and "Conspiracy Theory" which capture panic reactions among the public.We collected these data for January and February 2020 to study the developments duringthe initial global spread of the coronavirus when there was still significant uncertaintyover whether a pandemic would emerge. To make effect sizes interpretable, we nor-malize the search intensity at the country level by the mean search intensity prior to thearrival of the coronavirus in each country. To study the impact on search activity we exploit the precise timing of coronavirus Moreover, as shown by prior studies, such internet searches serve as a measure of economic sentimentamong households and thus as a predictor of future economic demand and activity (Choi and Varian, 2012;Vosen and Schmidt, 2011). To qualify this claim, in Online Appendix Table A.2 we use quarterly data from 2015to 2019 and show that real GDP growth and real growth in consumption and imports are significantly lower,in both a statistical and economic sense, in the quarters following increases in "Recession" topic searches. Online Appendix Figure A.1 shows the time series for the four topics of interest at the global level. Specifically, for the normalization we use the mean search intensity between December 1, 2019 and thedate of arrival of the coronavirus in a given country. This normalization makes the coefficient estimates in-terpretable as percentage changes relative to pre-coronavirus levels without having to resort to the mean ofthe dependent variable for interpretation. Results are not affected by this normalization, see Online AppendixTable A.4. rrival in a country. The underlying coronavirus case data are from Dong et al. (2020).Econometrically, we perform the following difference-in-differences regression using dailydata: y c , t = α c + day t + β × C c , t + (cid:101) c , t (1)where y c , t measures the search intensity in country c on day t for a specific topic. C c , t is a dummy variable indicating either having had at least one confirmed case or havinghad at least one human-to-human transmission of the coronavirus in country c at time t . The regressions control for country fixed effects α c , absorbing fixed and time-invariantdifferent levels of search intensities across countries. The time fixed effects day t absorba level shifter for each day, capturing the global trend. We cluster standard errors atthe country level. Intuitively, this analysis captures the impact of the local arrival of thecoronavirus conditional on the global trend. Results:
The data indicate that the arrival of the coronavirus in a country substantiallyincreased search intensity for topics related to economic recessions by 17.8 ( s . e . = s . e . = s . e . = s . e . = Additionally, the response of search intensity to the first human-to-humantransmission of the coronavirus in a country corroborates these results (Figure 1, PanelB and Online Appendix Table A.3). In a placebo test, we find no impact of the arrival ofthe coronavirus on a series of unrelated Google searches such as ’Dog’, ’Horse’, ’Insect’,’Rain’, or ’Rainbow’ (Online Appendix Table A.5). In sum, this evidence indicates thatthe arrival of the novel coronavirus leads to a spike in economic anxieties. Google searchers for prayers also increased sharply during the coronavirus crisis ( ? ). Results are further robust to dropping each country in turn or to dropping all countries pertaining to anyof the 17 subregions globally in turn (see Online Appendix Figures A.2 and A.3).
Notes:
Figure 1 shows the impact of the arrival of the coronavirus (Panel A) and first human-to-human transmission (Panel B) in a country on Google search intensity for the topics "Reces-sion", "Stock Market Crash", "Conspiracy Theory", and "Survivalism" obtained from difference-in-differences regressions conditional on country and day fixed effects. The dependent variablemeasures Google search intensity by topic indicated in column header, normalized by the averagesearch intensity in a country prior to the coronavirus arrival. The Google searches are collectedfor the time span between January 1st and February 29th, 2020. In all panels, error bands indicate95% confidence intervals obtained from standard errors clustered at the country level. .2 Micro-evidence after Arrival of Coronavirus in the UnitedStates Does economic anxiety increase further as the novel coronavirus spreads within a coun-try after the first domestic case occurs? We provide real-time evidence on this questionusing two surveys that measure economic anxiety in the United States. The surveyswere administered to broadly representative samples of the US population on March 5( n = n =
1, 006). Within this 11-day time span the num-ber of confirmed cases within the United States jumped by a factor of approximately 26,from 176 to 4576. Hence, this time frame captures a period of substantial within-countryspread. In both surveys, we investigate participants’ beliefs about the severity of thecrisis for the world and the US as well as their worries about the aggregate economy andtheir personal economic situation. For the precise wording of the questions and responsescales, see Figure 2.
Results:
The evolution of our survey measures over time between March 5 and 16 isvisualized in Figure 2. We document a substantial increase in participants’ beliefs aboutthe severity of the crisis for the world (Figure 2, Panel A) and the US (Figure 2, PanelB) as well as in their worries about the aggregate US economy (Figure 2, Panel C) andtheir personal economic situation (Figure 2, Panel D). Quantitatively, these increases aresizable. For instance, the fraction of respondents who were worried about the impact ontheir personal economic situation increased from 47% to 74% (p < 0.001) (see also OnlineAppendix Table A.9). In addition, in Online Appendix Figure A.4 we investigate heterogeneity across sev- Our sample is representative of the US population in terms of income, region, gender, age, and education(see Online Appendix Tables A.6 and A.7). We collaborated with an online panel provider (Luc.id) which iswidely used in the social sciences. In Online Appendix A we present cross-sectional results from the US in mid-February, during a time inwhich the US reported only 13 cases across the whole country. Respondents from states with any coronaviruscases exhibit significantly more pessimistic expectations (Online Appendix Table A.8). In our March 5 survey, we elicit economic anxieties after the relative mortality framing described in Section3.2. The descriptive patterns of an increase in economic anxiety from March 5 to March 16, however, hold inboth cases: when we focus either on respondents exposed to the high relative mortality framing or respondentsexposed to the low relative mortality framing. ral subgroups, dividing the sample by gender (Panel A), age (Panel B), and political af-filiation (Panel C). We do not find any differences between women and men (Online Ap-pendix Figure A.4, Panel A). Similarly, old and young individuals do not differ stronglyexcept that young people show substantially higher worries regarding their personal eco-nomic situation, potentially due to their higher unemployment risk (Online AppendixFigure A.4, Panel B). Finally, we observe stark partisan differences (Online AppendixFigure A.4, Panel C). Individuals who identify as Democrat hold substantially higherbeliefs about the severity of the crisis and show higher economic concerns. However,independent of the specific demographics, we observe that beliefs about the severity ofthe crisis as well as economic worries increased for all subgroups between March 5 andMarch 16.In sum, the data indicate that over 11 days individuals’ perceptions of the severity ofthe crisis strongly intensified and their economic worries substantially increased. Thisfinding is in line with results obtained using other data sources. Within the same timeframe, aggregate Google search intensity for the "Recession" topic increased by a factorof 10 in the US and by a factor of 5.5 on the global level (Online Appendix Figures A.5,Panels A and B). We also confirm our findings using other nationally representativeopinion polls conducted between March 5 to 8 and March 16 to 19 2020 (Online AppendixFigure A.6). The evolution of the search patterns for the topics "Stock Market Crash", "Conspiracy Theory", and "Sur-vivalism" was qualitatively similar (Online Appendix Figures A.5, Panels C-H).
Notes:
Figure 2 compares the levels of economic anxiety in the US in surveys conducted on March5 and March 16. Panel A shows beliefs about the severity of the coronavirus crisis for the world.Panel B shows beliefs about the severity of the coronavirus crisis for the US. Panel C shows wor-ries about the US economy. Panel D shows worries about one’s personal economic situation.Lighter columns indicate data collected on March 5, 2020, while darker columns indicate datacollected on March 16, 2020.
To further quantify the effects of the within-country spread of the coronavirus, weanalyze the local arrival in a difference-in-differences analysis at the state level (describedin section A of the Online Appendix), exploiting the fact that some states saw their firstconfirmed case in the time between our two surveys. The time fixed effects we includeallow us to control for aggregate developments, such as stock market movements. OnlineAppendix Table A.10 shows that having at least one case is associated with significantlymore pessimistic beliefs about the severity of the impact on the world (0.23 standarddeviations, s . e . = s . e . = ssociated with higher worries about the US economy by 0.22 standard deviations ( s . e . = The rapid increase in economic anxieties during and after arrival of the coronavirus setsthe stage to test for the underlying determinants. In this section, we examine the roleof perceptions of pandemic risk factors along two dimensions. First, we conducted twoexperiments that allow us to causally assess the impact of individuals’ perceptions ofcoronavirus mortality and contagiousness on economic anxiety. Importantly, throughexperimental variation we are able to isolate the direct effect of perceptions from otherenvironmental variables that affect all participants symmetrically, such as stock marketconditions. Second, we study respondents’ mental models of infectious disease spreadand the role of these mental models in shaping economic anxiety.
What beliefs did people hold about pandemic risk factors at the onset of the coronaviruscrisis? At the time of our first survey on March 5, there was still substantial uncertaintyand public disagreement about how severely the US economy would be affected by thecoronavirus. In our survey, we measured participants’ beliefs about two key character-istics that are relevant for the pandemic threat of the coronavirus: mortality and conta-giousness (R0), i.e. the expected number of infections directly caused by one infectedperson.We elicited participants’ beliefs about the mortality of the coronavirus using the fol-lowing question: “Out of 100 people who are infected with the coronavirus, how many do youthink will die as a result of catching the virus?” . Beliefs about the contagiousness (R0) of thevirus were elicited using the following question: “Think of a person who has the coronavirus. ow many non-infected people do you think will catch the virus from this person?” .Our data indicate substantial heterogeneity in participants’ beliefs about these char-acteristics of the coronavirus (see Panels A and B of Figure 3). On average, participants’beliefs about both the mortality from the coronavirus as well as its contagiousness weresubstantially higher than official and scientific estimates. The median participant esti-mated a mortality of 5% (mean of 14%) compared to an estimate of 3.4% provided by theWorld Health Organization (WHO) around the time of the surveys. Similarly, the me-dian participant estimated a contagiousness (R0) of 10 (mean of 43) relative to scientificestimates at the time of the survey in the range of R0 ≈ s . e . = s . e . = To understand whether beliefs about the mortality and contagiousness of the coronaviruscausally affect economic anxiety, we administered one framing treatment as well as aninformation treatment. The structure of the experiments was as follows: in the first com-ponent of the experiments, a random subset of respondents was assigned to receive the The positive association is particularly pronounced for individuals who hold lower beliefs about coro-navirus mortality and contagiousness, potentially because increases at low levels induce larger perceivedmarginal effects of the crisis on economic prospects. high relative mortality” treatment, while the remaining respondents were assigned toreceive the “low relative mortality treatment”. Subsequently, we elicited participants’economic worries. In the second component of the experiments, we randomly assignedsome respondents to get truthful information about the contagiousness and then re-elicited participants’ economic worries. Framing of relative mortality:
Our first experimental variation focuses on the fram-ing of mortality risk. In the experiment, participants were either truthfully informed,based on the same scientific estimate of coronavirus mortality at the time of the survey,that the death rate from the coronavirus is “20 times higher than for the flu” (high mortalitytreatment) or “5 times lower than for SARS” (low mortality treatment). The wording waschosen to mirror how information is commonly communicated in the media. We studyhow these different framings of mortality of the coronavirus affect participants’ expec-tations about the severity of the effects of the coronavirus in general, and participants’worries about the effects on the aggregate economy and their personal economic situa-tion. Econometrically, we estimate treatment effects using the following specification: y i = β + β highrelativemortality i + ε i (2)where y i is the z-scored outcome of interest for individual i and highrelativemortality i is adummy variable indicating whether individual i was exposed to the high mortality fram-ing. In additional robustness tests, we also test for the robustness of the results whenincluding demographic and socioeconomic controls, including gender, age bin dummies,log income, log income squared, dummies for having a high school degree and havingsome college education, dummies for being unemployed, currently working, a studentand for self-identifying as Democrat or Republican.Relative to the low mortality treatment, the high mortality treatment causally leadsparticipants’ to hold higher beliefs about the crisis’ severity for the world and the US:respondents in the high mortality treatment display 0.28 ( s . e . = s . e . = Randomization achieved excellent balance (see Online Appendix Table A.13). .066) standard deviations higher beliefs about the crisis’ severity for the world and theUS, respectively (Online Appendix Table A.14).These treatment differences also persist for participants’ economic worries (Figure 3,Panel D and Online Appendix Table A.15, Panel A): relative to the low mortality treat-ment, respondents in the high mortality treatment increase their worries about the effectsof the coronavirus on the US economy by 0.16 ( s . e . = s . e . = Information about contagiousness:
Besides mortality, contagiousness is a key char-acteristic that influences the pandemic risk of an infectious disease. The higher diseasecontagiousness, the larger is the risk of widespread and fast infection of the populationwhich can lead to an overload of the health care system (Massonnaud et al., 2020) andcostly disruption of economic activity (Adda, 2016). To understand the causal effect ofbeliefs about contagiousness on economic anxiety, in the second part of the experimentwe administered an additional information treatment.After eliciting participants’ beliefs about the contagiousness (R0) of the coronavirus,the participants were randomly assigned to be either in a “contagion information group”or a control group, which received no information. Based on scientific estimates (Li etal., 2020; Wu et al., 2020), respondents in the contagion information group were informedthat “approximately 2 non-infected people will catch the coronavirus from a person who has thecoronavirus” . Given that 81.4% of respondents overestimate this statistic, the informationtreatment should decrease the perceptions of the contagiousness.To test for the effect on economic anxieties, we re-elicited participants’ worries aboutthe effects of the coronavirus on the US economy and their household’s economic sit- ation as before. To analyze this treatment, we use an ANCOVA specification of thefollowing form: y i = δ + δ contagiousnessinfo i + δ y i , − + ε i (3)where contagiousnessinfo i is a dummy variable indicating whether individuals wereprovided the treatment information. y i , − is the outcome of interest measured in thesame survey prior to the second experiment. Respondents in the contagiousness information treatment show 0.09 ( s . e . = s . e . = In sum, the experimental evidence indicates that perceptions and information regard-ing coronavirus mortality and contagiousness are significant causal determinants thatshape individuals’ expectations about the aggregate economy and their personal eco-nomic situation at a time of high uncertainty. Randomization achieved excellent balance (see Online Appendix Table A.13). We do not control for y i , − in the first specification because we did not collect any outcome data prior tothe relative mortality treatment. As Online Appendix Table A.17 indicates, there are no significant interaction effects between the treat-ments.
Notes:
Figure 3 displays perceptions of the novel coronavirus and the experimental results. Thedata were collected on March 5. Panel A and B show the distribution of beliefs about mortalityand contagiousness (R0) of the coronavirus. Panel C shows the effect of overestimating mortalityand contagiousness relative to official numbers on worries about the aggregate US economy andrespondents’ personal economic situation. Panel D shows the experimental results on the effect ofinformation about the coronavirus on economic worries. The two leftmost bars in Panel D showthe effect of information suggesting high relative mortality as opposed to low relative mortalityon worries about the aggregate US economy and one’s personal economic situation. The tworightmost bars in Panel D show the effect of information about contagiousness on worries aboutthe aggregate US economy and one’s own personal economic situation. In all panels, error barsindicate 95% confidence intervals.
The public health impacts associated with a pandemic vary as a disease spreads throughspace and time. We already documented that economic anxieties evolve dynamicallywith this spread. However, so far we have not analyzed how the anticipation of such de-velopments shapes economic anxieties. Besides information about risk factors, forward-looking individuals rely on mental models of the world to make predictions about thefuture, and in the context of a pandemic, the future extent of disease spread. To analyzethis question, we investigate participants’ subjective mental models of infectious diseasespread to understand the role of cognitive processes and their limitations in shaping eco- omic anxiety in response to the outbreak of the coronavirus pandemic.As humans are organized in networks, disease spread typically follows a non-linear(e.g. logistic or quasi-exponential) function, at least in the beginning of an outbreak(Keeling and Rohani, 2011; Kermack and McKendrick, 1927). Hence, a small numberof cases can rapidly evolve into a widespread pandemic. Such a trajectory can be vastlyunderestimated if individuals do not take into account the non-linear nature of diseasespread but rather adopt a mental model of linear growth.To systematically investigate this question, we asked participants in our March 16survey to predict the spread of a fictitious infectious disease under simplifying assump-tions. We elicited participants’ predictions about the spread of a fictitious disease ratherthan asking participants for their estimates of the future number of coronavirus cases forthree reasons. First, investigating the role of cognitive processes requires the elicitationof individuals’ abstract mental models rather than their predictions for the specific caseof the coronavirus pandemic. Second, predictions about the future severity of the coron-avirus pandemic will be crucially shaped by individuals’ expectations about the extent ofendogenous containment measures as well as societal reactions which are independentof the general nature of infectious disease spread. Third, no reliable benchmark is avail-able for the future spread of the coronavirus, making it infeasible to assess the ex-anteaccuracy of estimates.Participants were instructed to assume that on a day 1, one person has a fictitious dis-ease and that each day a newly infected person infects two healthy people before stop-ping being contagious. To provide some guidance, participants were further informedthat on day 2, 3 people will be infected as the person who had the disease on day 1 spreadit to two other people on day 2. Participants were then asked to predict the total numberof people infected with the fictitious disease on day 5, 10, and 20.Figure 4, Panel A shows the median participant’s estimates and the correct predictionvalues. The results indicate that the average individual highly underestimates the spreadof the fictitious disease. In contrast to correct prediction values of 31 on day 5, 1023 onday 10, and 1,048,575 on day 20, the median participant estimates a case number of 16 n day 5, 30 on day 10, and 60 on day 20. Inconsistent with non-linear growth, thepredictions of the median participant can be well approximated by a subjective lineargrowth model (as exemplified by the green line in Panel B of Figure 4 for a linear growthrate of 2 per day). A linear mental model, however, is not uniformly present for theentire population. In particular, the 90th percentile prediction in our sample very wellcaptures the correct quasi-exponential growth, indicating heterogeneity in individuals’mental models of infectious disease spread. To understand how contemporaneous economic anxiety is associated with individu-als’ mental models of the spread of infectious diseases, we correlate economic anxietiesdescribed in Section 2.2 with participants’ predicted number of people infected with thefictitious disease on day 5, 10, and 20 (Figure 4, Panel C and Online Appendix TableA.19). To address outliers in participants’ predictions, we use a z-scored transformationof the logarithm of the predicted number of infected people.The data show statistically significant positive associations between participants’ pre-dictions and their beliefs about the crisis’ severity for the world and the US as well astheir worries about the aggregate US economy. For example, a one standard deviationincrease in the estimate of infectious disease spread after 10 days is associated with anincrease of 0.11 ( s . e . = To investigate the sources of heterogeneity, we explore the correlates of mental models in Online AppendixTable A.18. Across several specifications, we find that being older than 65 as well as having higher levelsof education and income are positively associated with a more accurate mental model of infectious diseasespread.
Notes:
Figure 4 visualizes mental models of infectious disease spread and their association witheconomic anxieties. The data were collected on March 16. Panel A shows participants’ medianbelief about the spread of a fictitious disease on a linear scale. Panel B shows participants’ median,75th percentile, and 90th percentile belief about the spread of a fictitious disease on a logarithmicscale. Participants were instructed to predict the number of cases of a fictitious disease on day5, 10, and 20. Participants were informed that on day 1, one person has the disease and thateach day a newly infected person infects two healthy people and then stops being contagious. Inboth panels, the solid line indicates the correct prediction. In panel B the dashed line indicatesan incorrect linear model with a growth rate of 2 per day. Panel C displays the association ofpredicted spread of the fictitious disease with participants’ beliefs about the severity of the impactof the coronavirus pandemic on the world and the US, as well as worries about the aggregate USeconomy and their personal economic situation. In all panels, error bars indicate 95% confidenceintervals. cases on day 5, 10, and 20. Panel A in Online Appendix Figure A.8 indicates that theobtained 3 types can be summarized as a linear, exponential, or an intermediate non-linear model. Panel B reveals that around 64.8% of participants exhibit a roughly linearmodel, while 15.7% display a roughly exponential model, and 19.5% an intermediatenon-linear mental model. Finally, Panel C confirms the previous results that holding a ore accurate (non-linear) mental model of infectious disease spread is associated withhigher beliefs about the crisis’ severity as well as higher worries regarding the aggregateeconomy.In sum, the results indicate that individuals who exhibit a more accurate mentalmodel of non-linear growth of infectious disease spread are more worried about the ag-gregate effects of the coronavirus pandemic, potentially as they foresee a greater poten-tial for a widespread contagion of the population. Combining global data from internet searches and two online experiments with repre-sentative samples of the US, this article documents a rapid emergence of economic anxi-ety at the onset of a major pandemic, and studies perceptions of pandemic risk factors ascorrelational and causal determinants.Our results point to a critical role of subjective beliefs about pandemic risks as wellas mental models of infectious disease spread in shaping public perception of the sever-ity of the contemporaneous health crisis and economic anxiety. For the present case ofthe coronavirus, we find substantial heterogeneity in beliefs about mortality and conta-giousness, two key characteristics relevant for pandemic risk. In real-time experiments,we show that information provision regarding these characteristics causally shapes eco-nomic anxiety among the population. Our experiment also shows that framing of in-formation about the coronavirus matters for the inferences that people make. Specifi-cally, our experiment highlights that even if journalists base their comparisons on thesame mortality statistics, the choice of comparison matters. These results speak to an im-portant debate on how media coverage and public communication of disease outbreaksaffect people’s beliefs (Bursztyn et al., 2020).Moreover, consistent with exponential growth bias (Levy and Tasoff, 2016; Stangoand Zinman, 2009), for the majority of the population subjective mental models under-state the non-linear nature of infectious disease spread. The heterogeneity in individ- als’ mental models crucially shapes their perception of the severity of a major globalpandemic and affects their worries about the impact on the aggregate economy. eferences Adams-Prassl, Abi, Teodora Boneva, Marta Golin, and Christopher Rauh , “Inequal-ity in the Impact of the Coronavirus Shock: Evidence from Real Time Surveys,”
IZADiscussion Paper No. 13183 , 2020.
Adda, Jérôme , “Economic Activity and the Spread of Viral Diseases: Evidence from HighFrequency Data,”
The Quarterly Journal of Economics , 2016, (2), 891–941.
Andre, Peter, Carlo Pizzinelli, Christopher Roth, and Johannes Wohlfart , “SubjectiveModels of the Macroeconomy: Evidence from Experts and a Representative Sample,”
CESifo Working Paper No. 7850 , 2019.
Armona, Luis, Andreas Fuster, and Basit Zafar , “Home Price Expectations and Be-haviour: Evidence from a Randomized Information Experiment,”
The Review of Eco-nomic Studies , 2018, (4), 1371–1410. Bailey, Michael, Eduardo Dávila, Theresa Kuchler, and Johannes Stroebel , “HousePrice Beliefs and Mortgage Leverage Choice,”
Review of Economic Studies , 2019. , Ruiqing Cao, Theresa Kuchler, and Johannes Stroebel , “The Economic Effects ofSocial Networks: Evidence from the Housing Market,”
Journal of Political Economy ,2018, (6), 2224–2276.
Bartik, Alexander W., Marianne Bertrand, Zoë B. Cullen, Edward L. Glaeser, MichaelLuca, and Christopher T. Stanton , “How Are Small Businesses Adjusting to COVID-19? Early Evidence From a Survey,”
NBER Working Paper 26989 , 2020.
Binder, Carola , “Coronavirus Fears and Macroeconomic Expectations,”
Review of Eco-nomics and Statistics , 2020, (forthcoming). and Alex Rodrigue , “Household Informedness and Long-Run Inflation Expectations:Experimental Evidence,”
Southern Economic Journal , 2018, (2), 580–598. nd Christos Makridis , “Stuck in the Seventies: Gas Prices and Macroeconomic Ex-pectations,” Working Paper available at SSRN 3267721 , 2018.
Bord, Richard J, Robert E O’connor, and Ann Fisher , “In What Sense Does the PublicNeed to Understand Global Climate Change?,”
Public Understanding of Science , 2000, (3), 205–218. Bordalo, Pedro, Nicola Gennaioli, and Andrei Shleifer , “Salience and ConsumerChoice,”
Journal of Political Economy , 2013, (5), 803–843.
Bursztyn, Leonardo, Aakaash Rao, Christopher Roth, and David Yanagizawa-Drott ,“Misinformation During a Pandemic,”
Becker Friedman Institute for Economics WorkingPaper 2020-44 , 2020.
Carbone, Jared C., Snorre Kverndokk, and Ole Jørgen Røgeberg , “Smoking, Health,Risk, and Perception,”
Journal of Health Economics , 2005, (4), 631 – 653. Choi, Hyunyoung and Hal Varian , “Predicting the present with Google Trends,”
Eco-nomic Record , 2012, , 2–9. Chong, Dennis and James N Druckman , “Framing theory,”
Annual Review of PoliticalScience , 2007, , 103–126. Coibion, Olivier and Yuriy Gorodnichenko , “What Can Survey Forecasts Tell us aboutInformation Rigidities?,”
Journal of Political Economy , 2012, (1), 116–159. and , “Information Rigidity and the Expectations Formation Process: A SimpleFramework and New Facts,”
The American Economic Review , 2015, (8), 2644–2678. and , “Is the Phillips Curve Alive and Well after All? Inflation Expectations and theMissing Disinflation,”
American Economic Journal: Macroeconomics , 2015, (1), 197–232. , Dimitris Georgarakos, Yuriy Gorodnichenko, and Maarten van Rooij , “How DoesConsumption Respond to News about Inflation? Field Evidence from a RandomizedControl Trial,” NBER Working Paper No. 26106 , 2019. Yuriy Gorodnichenko, and Michael Weber , “Monetary Policy Communications andtheir Effects on Household Inflation Expectations,”
NBER Working Paper No. 25482 ,2019. , , and , “Labor Markets during the COVID-19 Crisis: A Preliminary View,”
BeckerFriedman Institute for Economics Working Paper 2020-41 , 2020. , , and Saten Kumar , “How Do Firms Form Their Expectations? New Survey Evi-dence,”
American Economic Review , 2018, (9), 2671–2713. , , and Tiziano Ropele , “Inflation Expectations and Firm Decisions: New CausalEvidence,”
The Quarterly Journal of Economics , 2020, (1), 165–219.
D’Acunto, Francesco, Daniel Hoang, and Michael Weber , “Managing Households’ Ex-pectations with Simple Economic Policies,”
CESifo Working Paper No. 7793 , 2019. , , Maritta Paloviita, and Michael Weber , “IQ, Expectations, and Choice,”
NBERWorking Paper No. 25496 , 2019.
Dong, Ensheng, Hongru Du, and Lauren Gardner , “An Interactive Web-Based Dash-board to Track COVID-19 in Real Time,”
The Lancet Infectious Diseases , 2020.
Fortson, Jane G , “Mortality Risk and Human Capital Investment: The Impact ofHIV/AIDS in Sub-Saharan Africa,”
The Review of Economics and Statistics , 2011, (1),1–15. Fuster, Andreas, Benjamin Hebert, and David Laibson , “Natural Expectations, Macroe-conomic Dynamics, and Asset Pricing,”
NBER Macroeconomics Annual , 2012, (1),1–48. , David Laibson, and Brock Mendel , “Natural Expectations and Macroeconomic Fluc-tuations,” Journal of Economic Perspectives , 2010, (4), 67–84. allagher, Justin , “Learning About an Infrequent Event: Evidence from Flood InsuranceTake-Up in the United States,” American Economic Journal: Applied Economics , 2014,pp. 206–233.
Ginsberg, Jeremy, Matthew H. Mohebbi, Rajan S. Patel, Lynnette Brammer, Mark S.Smolinski, and Larry Brilliant , “Detecting Influenza Epidemics Using Search EngineQuery Data,”
Nature , 2009, (7232), 1012–1014.
Hanspal, Tobin, Annika Weber, and Johannes Wohlfart , “Income and Wealth Shocksand Expectations during the COVID-19 Pandemic,”
CEBI Working Paper No. 13/20 ,2020.
Heimer, Rawley Z, Kristian Ove R Myrseth, and Raphael S Schoenle , “YOLO: Mortal-ity Beliefs and Household Finance Puzzles,”
The Journal of Finance , 2019, (6), 2957–2996. Kan, Kamhon and Wei-Der Tsai , “Obesity and Risk Knowledge,”
Journal of Health Eco-nomics , 2004, (5), 907–934. Keeling, Matt J and Pejman Rohani , Modeling Infectious Diseases in Humans and Animals ,Princeton University Press, 2011.
Kermack, William Ogilvy and Anderson G McKendrick , “A Contribution to the Mathe-matical Theory of Epidemics,”
Proceedings of the Royal Society of London , 1927, (772),700–721.
Kerwin, Jason , “Scared Straight or Scared to Death? The Effect of Risk Beliefs on RiskyBehaviors,”
Working Paper available at SSRN 2797493 , 2018.
Kuchler, Theresa and Basit Zafar , “Personal Experiences and Expectations about Aggre-gate Outcomes,”
The Journal of Finance , 2019, (5), 2491–2542. Levy, Matthew and Joshua Tasoff , “Exponential-Growth Bias and Lifecycle Consump-tion,”
Journal of the European Economic Association , 2016, (3), 545–583. i, Qun, Xuhua Guan, Peng Wu, Xiaoye Wang, Lei Zhou, Yeqing Tong, Ruiqi Ren,Kathy SM Leung, Eric HY Lau, Jessica Y Wong et al. , “Early Transmission Dynamicsin Wuhan, China, of Novel Coronavirus–Infected Pneumonia,” New England Journal ofMedicine , 2020.
Liu, Jin-Tan and Chee-Ruey Hsieh , “Risk Perception and Smoking Behavior: EmpiricalEvidence from Taiwan,”
Journal of Risk and Uncertainty , 1995, (2), 139–157. Ma´ckowiak, Bartosz and Mirko Wiederholt , “Lack of Preparation for Rare Events,”
Journal of Monetary Economics , 2018, , 35–47.
Malmendier, Ulrike and Stefan Nagel , “Depression Babies: Do Macroeconomic Experi-ences Affect Risk Taking?,”
The Quarterly Journal of Economics , 2011, (1), 373–416.
Massonnaud, Clément, Jonathan Roux, and Pascal Crépey , “COVID-19: ForecastingShort Term Hospital Needs in France,” medRxiv , 2020.
Oster, Emily, Ira Shoulson, and E. Ray Dorsey , “Optimal Expectations and LimitedMedical Testing: Evidence from Huntington Disease,”
American Economic Review ,April 2013, (2), 804–30.
Person, Bobbie, Francisco Sy, Kelly Holton, Barbara Govert, Arthur Liang et al. , “Fearand Stigma: the Epidemic Within the SARS Outbreak,”
Emerging Infectious Diseases ,2004, (2), 358. Prat, Andrea and David Strömberg , “The Political Economy of Mass Media,”
Advancesin Economics and Econometrics , 2013, , 135. Rabin, Matthew , “Inference by Believers in the Law of Small Numbers,”
The QuarterlyJournal of Economics , 2002, (3), 775–816.
Razum, Oliver, Heiko Becher, Annette Kapaun, and Thomas Junghanss , “SARS, LayEpidemiology, and Fear,”
The Lancet , 2003, (9370), 1739–1740. oth, Christopher and Johannes Wohlfart , “How Do Expectations about the Macroe-conomy Affect Personal Expectations and Behavior?,” Review of Economics and Statis-tics , 2020, (forthcoming).
Stango, Victor and Jonathan Zinman , “Exponential Growth Bias and Household Fi-nance,”
The Journal of Finance , 2009, (6), 2807–2849. Tversky, Amos and Daniel Kahneman , “Availability: A Heuristic for Judging Frequencyand Probability,”
Cognitive Psychology , 1973, (2), 207–232. Vosen, Simeon and Torsten Schmidt , “Forecasting Private Consumption: Survey-BasedIndicators vs. Google Trends,”
Journal of Forecasting , 2011, (6), 565–578. Wagenaar, William A and Sabato D Sagaria , “Misperception of Exponential Growth,”
Perception & Psychophysics , 1975, (6), 416–422. Weinstein, Neil D , “Optimistic Biases about Personal Risks,”
Science , 1989, (4935),1232–1234.
Winter, Joachim and Amelie Wuppermann , “Do They Know What is at Risk? HealthRisk Perceptions Among The Obese?,”
Health Economics , 2014, (5), 564–585. Wu, Joseph T, Kathy Leung, and Gabriel M Leung , “Nowcasting and Forecasting thePotential Domestic and International Spread of the 2019-nCoV Outbreak Originatingin Wuhan, China: a Modelling Study,”
The Lancet , 2020.
Zhu, Na, Dingyu Zhang, Wenling Wang, Xingwang Li, Bo Yang, Jingdong Song, XiangZhao, Baoying Huang, Weifeng Shi, Roujian Lu et al. , “A Novel Coronavirus fromPatients with Pneumonia in China, 2019,”
New England Journal of Medicine , 2020. nline Appendix Coronavirus Perceptions and Economic Anxiety
Thiemo Fetzer, Lukas Hensel, Johannes Hermle, Christopher RothThis Online Appendix contains supplementary materials mentioned in the main text.Section A provides an overview of the methodology underlying the analysis of the im-pact of state-level coronavirus arrival on outcomes in the US. In Appendix Section B wedisplay additional figures. Online Appendix Figure A.1 displays the global trends insearch intensity for our four main indicators. Online Appendix Figure A.2 displays theimpact of dropping individual countries from the main search intensity analysis. OnlineAppendix Figure A.3 displays the impact of dropping subregions from the main searchintensity analysis. Online Appendix Figure A.4 displays how perceptions of severityand economic worries change over time for different subgroups. Online Appendix Fig-ure A.5 displays search intensity trends in the US and globally for our four main mea-sures. Online Appendix Figure A.6 displays changes in economic expectations from earlyto mid-March using Roper Center polling data. Online Appendix Figure A.7 plots thenon-parametric relationship between mortality and contagiousness perceptions and eco-nomic anxieties. Finally, Online Appendix Figure A.8 displays the results of categorizingindividuals in linear, exponential, and other mental models of disease spread. It alsoshows the correlation between mental models and economic anxieties.Online Appendix Section C displays additional tables. Online Appendix Table A.1contains all countries used for the Google search intensity analysis. Online AppendixTable A.2 displays correlations of Google search intensity with GDP and its components.Online Appendix Table A.3 contains the main Google search intensity results also dis-played in Figure 1. Online Appendix Table A.4 displays the main results without nor-malization of the outcome variables. Online Appendix Table A.5 displays a series ofplacebo difference-in-differences regressions. Online Appendix Table A.6 displays sum-mary statistics for the March 5 survey. Online Appendix Table A.7 displays summary tatistics for the March 16 survey. Online Appendix Table A.8 shows the correlation be-tween having any coronavirus case and coronavirus-related concerns in mid February.Online Appendix Table A.9 shows differences in coronavirus perceptions and economicanxieties over time. Online Appendix Table A.10 show the results of the difference-in-differences regression described in Section A. Online Appendix Tables A.11 and A.12display correlations of coronavirus perceptions and economic anxieties for binary andcontinuous variables, respectively. Online Appendix Table A.13 shows balance testsfor both experiments. Online Appendix Table A.14 shows the impact of informationabout relative mortality on the perceived severity of the crisis. Online Appendix TableA.15 shows the impact of coronavirus-related information on economic worries. On-line Appendix Table A.16 shows the impact of contagiousness information controllingfor treatment assignment in the relative mortality experiment. Online Appendix TableA.17 explores interactions effects between the experiments. Online Appendix Table A.18displays predictors of mental models of disease spread. Finally, Online Appendix TableA.19 displays the correlations between predicted disease spread and severity perceptionsas well as economic worries. The Impact of Coronavirus Arrival in the US
In this section, we describe the methodology for the additional analysis linking con-firmed coronavirus cases in the US to the perceived threat of the pandemic and economicanxieties. For this purpose, we leverage a public opinion poll and our own survey data.
A.1 Cross-sectional Evidence from the US in mid-February
The US reported its first case of coronavirus on January 22, 2020 (Dong et al., 2020). Formost of February, the case count within the US remained fairly flat, increasing from 8cases on February 1st to 24 cases by February 29. We present descriptive evidence doc-umenting that there is an association between the spread of coronavirus within the USand increased anxieties using opinion polling data from individuals across US states.The Kaiser Family Foundation poll was conducted from February 13 to February 18,2020 among a sample of 1207 US residents and included a few questions relating to coro-navirus. During that whole period, there were no reported new cases of coronavirus acrossthe US with the total confirmed case count staying flat at 13 cases. 46 states reportedno case. Three states (Washington, Massachusetts and Arizona) reported a single caseeach, Illinois reported two cases and California reported eight cases. We assess associ-ations between residing in a state with at least any coronavirus case and responses tocoronavirus-related questions. To do so, we estimate the following simple regression: y i , s , t = γ × anycase s + β (cid:48) X i + η t + (cid:101) is where anycase s is a dummy indicating the presence of at least one coronavirus case.As indicated, there was no further spread recorded during the sample period accordingto data from Dong et al. (2020). The underlying micro data are made available through the Roper Center ID 31117209. The fact that the US reported little intracommunity spread during much of February is likely not the resultof no spread occurring, but rather due to the failure of the US to ramp up testing, and the use of a faulty test,see nature.com/articles/d41586-020-01068-3 . he dependent variable y i , s , t measures a survey respondent’s responses to a set ofcoronavirus related questions: whether individuals are (very) concerned or (not at all)concerned about: the family or oneself getting sick, a negative economic impact, or awidespread outbreak of coronavirus in the US.We study whether individuals living in states with any coronavirus cases during thetime period give different responses. We control for interview date fixed effects, η t , alongwith a set of individual-level controls. Results
The results are presented in Table A.8 and suggest that respondents fromstates with any case of coronavirus during that sample period are more concerned about:themselves or family members getting sick; the negative impact on the US economy; andabout a widespread outbreak of coronavirus in the US.
A.2 Difference-in-differences Analysis in March
To go beyond a purely cross-sectional analysis of the relationship between the pres-ence of coronavirus cases and economic anxieties we conduct a state-level difference-in-differences analysis. We use the fact that in the 11-day period between our surveys onMarch 5 and 16, 31 states recorded their first coronavirus case leaving only three stateswithout a case on March 15. This allows us to estimate a regression of the following form: y ist = α × anycase st + δ s + η t + β (cid:48) X i + ε ist (4)where anycase st is a dummy variable indicating whether a given state has a confirmedcoronavirus case by the time of the survey. We also include state fixed effects to accountfor permanent differences in economic anxieties across states. Finally, time fixed effectsaccount for any level differences across time that affect all states in the same way. Weconduct this analysis for the main four outcomes measuring perceptions of the severityof the pandemic and worries about its impact on the economy. esults Online Appendix Table A.10 displays estimation results. We find that indi-viduals in states with at least one coronavirus case exhibit a significantly higher levelof perceived threat by the pandemic and worries about the US economy. The relation-ship impact on worries about personal economic circumstances is also positive but it issmaller and not statistically significant. Online Appendix Figures
Figure A.1: Time Series of Global Internet Searches for January and February 2020
Notes:
Online Appendix Figure A.1 shows the time series of the search intensity for the Googletopics "Recession", "Stock Market Crash", "Conspiracy Theory", and "Survivalism" from January1st to February 29th, 2020 as well as the number of countries with a confirmed coronavirus case.
Notes:
Online Appendix Figure A.2 shows boxplots of the point estimates obtained from estimating equation 1 when dropping each country in turn. The coefficient obtained fromestimation on the full sample is indicated by the horizontal red line. igure A.3: Robustness of results to dropping individual sub-regions Notes:
Online Appendix Figure A.3 shows boxplots of the point estimates obtained from estimating equation 1 when dropping countries belonging to each of the 17 different sub-regionsin turn. The coefficient obtained from estimation on the full sample is indicated by the horizontal red line. igure A.4: Evolution of Beliefs about Severity of Crisis and Economic Worries by SubgroupsA By GenderB By AgeC By Political Affiliation Notes:
Online Appendix Figure A.4 shows for different subgroups the evolution of beliefs about the severity of the crisisfor the world (leftmost panels) and the United States (second leftmost panels) as well as worries about the aggregateeconomy (second rightmost panels) and worries about respondents’ personal economic situation (rightmost panels). Thedata were collected on March 5 and March 16, 2020.
Notes:
Online Appendix Figure A.5 shows time series of the search intensity for Google topics "Recession", "Stock Mar-ket Crash", "Conspiracy Theory", and "Survivalism" from February 19th to March 16th, 2020 for the United States andworldwide.
Notes:
Online Appendix Figure A.6 compares polling results among representative samples of the US population from anopinion poll conducted between March 5 and 8 2020 (American Research Group Poll, Question 4, Roper Center Poll Iden-tifier: 31117199.00003) with results from an opinion poll conducted between March 16 and 19 2020 (Quinnipiac UniversityPoll, Question 19, Roper Center Poll Identifier: 31 31117223.00023). The Figure displays answers to the following questionin both opinion polls: “Do you think the national economy is getting better, staying the same, or getting worse?”
Notes:
Online Appendix Figure A.7 displays the non-parametric relationship between perceptions of the novel coronavirusand economic anxieties. The data were collected on March 5. Panel A and B show non-parametric relationships betweenrespondents’ worries about the aggregate and personal economic situation and their belief about coronavirus mortality.Panel C and D show non-parametric relationships between respondents’ worries about the aggregate and their personaleconomic situation and their beliefs about coronavirus contagiousness (R0). Each plot shows a binscatter plot where x-values correspond to the midpoint of each bin. y-values indicate the mean of the outcome variable for the respectivebin.
Notes:
Online Appendix Figure A.8 shows a classification of mental models of infectious disease spread obtained fromk-means clustering using 3 clusters in the space of predicted log-number of cases at day 5, 10, and 20. Panel A shows themedian number of predicted cases in the three clusters (dots) as well as the cluster centers (shaded areas) that contain 50%of the mass of each cluster. The blue and green lines represent the correct prediction values as well as an incorrect linearmodel with a slope of 2. The three types obtained from the cluster approach can be described as a correct ’exponentialmental model’, an incorrect ’linear mental model’, or an incorrect ’other mental model’. Panel B visualizes the typedistribution in the data. Panel C shows estimates of OLS regressions with an indicator for exhibiting a non-linear mentalmodel as the independent variable and using as the dependent variables participants’ beliefs about the crisis’ severityfor the world and the United States as well as their worries about the aggregate economy and their personal economicsituation. Error bars indicate 95% confidence intervals. Online Appendix Tables
Table A.1: Countries and territories included in the analysis of Google searches
Andorra Dominica Korea, Republic of abc
Palestine, State ofUnited Arab Emirates b Dominican Republic Kuwait b Portugal abc
Afghanistan Algeria b Cayman Islands ParaguayAntigua and Barbuda Ecuador ac Kazakhstan ab Qatar ab Albania Estonia ab Laos ReunionArmenia Egypt ac Lebanon Romania abc
Angola Spain abc
Saint Lucia Serbia abc
Argentina abc
Ethiopia Sri Lanka b Russian Federation abc
Austria abc
Finland abc
Liberia RwandaAustralia abc
Fiji Lesotho Saudi Arabia a Aruba Faroe Islands Lithuania abc
SudanAzerbaijan ab France abc
Luxembourg Sweden abc
Bosnia and Herzegovina Gabon Latvia abc
Singapore abc
Barbados United Kingdom abc
Libya Saint HelenaBangladesh b Grenada Morocco b Slovenia abc
Belgium abc
Georgia Moldova, Republic of Slovakia abc
Burkina Faso French Guiana Montenegro Sierra LeoneBulgaria abc
Guernsey Madagascar SenegalBahrain b Ghana North Macedonia SomaliaBurundi Gibraltar Mali SurinameBenin Greenland Myanmar El Salvador ab Bermuda Guinea Mongolia Sint Maarten (Dutch part)Brunei Darussalam Guadeloupe Macao Syrian Arab RepublicBolivia (Plurinational State of) Greece abc
Martinique EswatiniBonaire, Sint Eustatius and Saba Guatemala Mauritania TogoBrazil abc
Guam Malta Thailand abc
Bahamas Guyana Mauritius TajikistanBhutan Hong Kong abc
Maldives TurkmenistanBotswana Honduras Malawi Tunisia ab Belarus Croatia ab Mexico abc
TongaBelize Haiti Malaysia abc
Turkey abc
Canada abc
Hungary abc
Mozambique Trinidad and TobagoCongo, Democratic Republic of the Indonesia abc
New Caledonia Taiwan, Province of China abc
Congo Ireland abc
Niger Tanzania, United Republic ofSwitzerland abc
Israel abc
Nigeria abc
Ukraine abc
Côte d’Ivoire Isle of Man Nicaragua UgandaChile abc
India abc
Netherlands abc
United States of America abc
Cameroon Iraq Norway abc
UruguayChina ab Iran (Islamic Republic of) abc
Nepal UzbekistanColombia abc
Iceland New Zealand abc
Saint Vincent and the GrenadinesCosta Rica ac Italy abc
Oman Venezuela (Bolivarian Republic of) ac Cuba Jersey Panama Virgin Islands (U.S.)Cabo Verde Jamaica Peru abc
Viet Nam ab Curacao Jordan ab French Polynesia YemenCyprus abc
Japan abc
Papua New Guinea South Africa abc
Czechia abc
Kenya a Philippines abc
ZambiaGermany abc
Kyrgyzstan Pakistan b ZimbabweDjibouti Cambodia Poland abc
Denmark abc
Saint Kitts and Nevis Puerto Rico
Notes:
Appendix Table A.1 lists all countries included in the analysis of the impact of the coronavirus on internet searches on Google. Thesuperscripts a , b , c refer to data availability regarding aggregate demand components for the analysis in Online Appendix Table A.2: a dataavailable on real GDP; b data available on industrial production; c data available on demand factors. (1) (2) (3) (4) (5) (6) (7)Demand factorsReal GDP Industrial production C I G X ML.Recession topic Google searches -1.009*** -1.231* -1.564*** -1.847 -1.345 1.109* -5.063***(0.311) (0.661) (0.506) (1.798) (0.888) (0.657) (1.352)R .716 .446 .627 .27 .282 .236 .314Countries 70 72 58 58 58 58 58Observations 1350 1218 1087 1087 1087 1087 1087Country FE X X X X X X XYear x Quarter FE X X X X X X X Notes:
Appendix Table A.2 displays the relationship between year-on-year growth rates in GDP, industrial production and demand com-ponents and ‘Recession’ topic Google searches. Econometrically, we perform country-level regressions controlling for country and year-by-quarter fixed effects in all specifications. The results show that increases in Google search activity for recession-related topics areassociated with lower growth rates in GDP, consumption spending and imports in the subsequent quarter. The level of analysis is countryand quarter. Data were collected by the Economist Intelligence Unit from 2015 to 2019. The dependent variable in column (1) measuresGDP growth. The dependent variable in column (2) measures growth of industrial production. Columns (3) to (6) measure different com-ponents of aggregate demand. Column (3) shows the association with aggregate consumption. Column (4) shows the association withinvestments. Column (5) shows the association with government spending. Column (6) shows the association with exports. Column (7)shows the association with imports. The independent variable measures Google search intensity for the topic “recession”. For countriesincluded in each regression, see Online Appendix Table A.1. Standard errors clustered at the country level are presented in parentheses. ∗ p < ∗∗ p < ∗∗∗ p < Impact on Goolge search trends(1) (2) (3) (4)Recession Stock Market Crash Conspiracy Theory Survivalism
Panel A: Any Covid-19 case
Post any Covid-19 case 0.178 ∗∗ ∗∗∗ ∗∗∗ ∗∗∗ (0.073) (0.124) (0.091) (0.073)R Panel B: Any human-to-human transmission
Post any human-to-human transmission 0.351 ∗∗ ∗ ∗∗ ∗∗ (0.141) (0.163) (0.164) (0.140)R Notes:
Online Appendix Table A.3 displays the impact of coronavirus arrival on Google searches for search terms related to economicanxiety. The results show that coronavirus arrival is a predictor of Google searches related to economic anxiety. Column 1 shows resultsfor Google searches related to recessions. Column 2 shows results for Google searches related to stock market crashes. Column 3 showsresults for Google searches related to conspiracy topics. Column 4 shows results for Google searches related to survivalism. The dependentvariable measures Google search intensity for the indicated topics normalized by the average search intensity in a country prior to thecoronavirus arrival. The data on Google searches were downloaded from the Google API on March 3 rd . In panel A, we show the impactof a dummy variable indicating at least one coronavirus case. In Panel B, we show the impact of having at least one human-to-humantransmission of coronavirus. The data on first cases stem from Dong et al. (2020). The data on human-to-human transmissions are basedon official reports by the WHO and national authorities. The level of analysis is country-day. Dates included range from January 1 st th ∗ p < ∗∗ p < ∗∗∗ p < Impact on Goolge search trends(1) (2) (3) (4)Recession Stock Market Crash Conspiracy Theory Survivalism
Panel A: Any Covid-19 case
Post any Covid-19 case 2.522 ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ (0.677) (0.538) (0.580) (0.443)R Panel B: Any human-to-human transmission
Post any human-to-human transmission 5.644 ∗∗∗ ∗ ∗ ∗∗ (1.927) (1.069) (1.999) (0.978)R Notes:
The dependent variable measure country-specific search intensity for a topic indicated in the column head from January 2020 toFebruary 29 2020. Column 1 shows results for Google searches related to recessions. Column 2 shows results for Google searches relatedto stock market crashes. Column 3 shows results for Google searches related to conspiracy topics. Column 4 shows results for Googlesearches related to survivalism. The dependent variable measures Google search intensity for the indicated topics. The data on Googlesearches were downloaded from the Google API on March 3 rd . In panel A, we show the impact of a dummy variable indicating at leastone coronavirus case. In Panel B, we show the impact of having at least one human-to-human transmission of coronavirus. The dataon first cases stem from Dong et al. (2020). The data on human-to-human transmissions are based on official reports by the WHO andnational authorities. The level of analysis is country-day. Dates included range from January 1 st th ∗ p < ∗∗ p < ∗∗∗ p < Impact on google searches for(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)Dog Horse Insect DaVinci Nelson Mandela Rain Rainbow Stars Mars (planet) Menstrual Cycle
Panel A: First confirmed case
First confirmed case -0.018 -0.036 -0.007 -0.044 -0.033 0.044 -0.042 -0.012 -0.001 0.033(0.018) (0.027) (0.042) (0.038) (0.058) (0.048) (0.045) (0.033) (0.103) (0.033)R Panel B: First human-to-human transmission
First human-to-human transmission -0.021 -0.085 0.035 -0.030 0.027 -0.041 -0.047 -0.024 0.102 0.039(0.038) (0.064) (0.040) (0.039) (0.059) (0.078) (0.055) (0.048) (0.324) (0.028)R Notes:
Online Appendix Table A.5 displays the impact of coronavirus arrival on placebo Google searches that should not be affected bythe arrival of the coronavirus. The results document that coronavirus arrival does not systematically predict Google searches unrelatedto economic anxiety. The dependent variable measures Google search intensity for the indicated topics normalized by the average searchintensity in a country prior to the coronavirus arrival. The data on Google searches were downloaded from the Google API on March3 rd . In panel A, we show the impact of a dummy variable indicating at least one coronavirus case. In Panel B, we show the impact ofhaving at least one human-to-human transmission of coronavirus. The data stem from Dong et al. (2020). The data on human-to-humantransmissions are based on official reports by the WHO and national authorities. The level of analysis is country-day. Dates includedrange from January 1 st th ∗ p < ∗∗ p < ∗∗∗ p < able A.6: Summary statistics: Experimental sample March 5 (1) (2) (3) (4)Mean SD Median Obs. Panel A: Demographics % Male 49.02 50.02 914% Age < 35 0.23 0.42 914% Highschool education 17.61 38.12 914% College eductation 80.53 39.62 914% Currently working 55.03 49.77 914% Democrat 40.04 49.03 914% Republican 33.15 47.10 914% High trust in science 1.09 10.41 914
Panel B: Economic Anxieties % agree: world severely affected by coronavirus 67.61 46.82 914% agree: US severely affected by coronavirus 55.14 49.76 914% worried about US economy 68.05 46.65 914% worried about personal econ. situation 47.16 49.95 914
Panel C: Coronavirus perceptions
Infectiousness (R0) 43.23 146.17 10 914Predicted mortality rate 13.70 20.84 5 914
Notes:
Online Appendix Table A.6 displays summary statistics for the experimental sample. These data were collected on March 5. Panel Ashows shares of respondents with indicated characteristics. Panel B shows shares of respondents with indicated beliefs about the severityof the crisis and economic anxieties. Panel C shows variables measuring perceptions of the coronavirus. (1) (2) (3) (4)Mean SD Median Obs.
Panel A: Demographics % Male 52.09 49.98 1006% Age < 35 22.66 41.89 1006% Highschool education 19.98 40.00 1006% College eductation 76.64 42.33 1006% Currently working 52.19 49.98 1006% Democrat 38.57 48.70 1006% Republican 32.11 46.71 1006% High trust in science 1.79 13.26 1006
Panel B: Economic Anxieties % agree: world severely affected by coronavirus 80.12 39.93 1006% agree: US severely affected by coronavirus 77.83 41.56 1006% worried about US economy 87.57 33.00 1006% worried about personal econ. situation 73.76 44.02 1006
Panel C: Coronavirus perceptions
Infectiousness (R0) 49.81 175.13 5 1006Number of cases after 5 days (w) 20.02 20.72 11 1006Number of cases after 10 days (w) 340.29 678.64 30 1006Number of cases after 20 days (w) 122218.17 311256.66 60 1006Predicted mortality rate 15.60 21.47 5 1006
Notes:
Online Appendix Table A.7 displays summary statistics for the experimental sample. These data were collected on March 16.Panel A shows shares of respondents with indicated characteristics. Panel B shows shares of respondents with indicated beliefs aboutthe severity of the crisis and economic anxieties. Panel C shows variables measuring coronavirus perceptions and predictions of fictitiousinfectious disease spread. (1) (2) (3)Strength of concern thatyou/your family getting sick from coronavirus coronavirus will have negative impact on US economy widespread outbreak of coronavirus in USAny Covid-19 case 0.111* 0.124** 0.123**(0.063) (0.057) (0.056)Mean of DV .00278 .00213 .000396R .179 .139 .186States 50 50 50Observations 1197 1192 1197Individual Controls X X XInterview Date FE X X X Notes:
Online Appendix Table A.8 presents results from studying the impact of the coronavirus on perceptions and awareness during theearly period of the spread. Data come from an opinion poll conducted for the Kaiser Family Foundation Poll (Roper Center ID 31117209)from February 13 - February 18. During this time there were only 13 reported coronavirus cases in all of the US which were concentratedin 5 states. Outcomes are measured on a 4-point scale (Not at all concerned; Not too concerned; Somewhat concerned; Very concerned)and standardized to have mean zero and standard deviation one. Individual controls include age, gender, education, income and politicalparty affiliation (5 point scale). Observation counts vary due to “don’t knows” or non-response. Standard errors clustered at the state levelare presented in parentheses with stars indicating ∗ p < ∗∗ p < ∗∗∗ p < March 5 March 16 Comparison of means(1) (2) (3) (4) (5) (6) (7) (8)Mean SD Obs. Mean SD Obs. ∆ p(early = late) Panel A: Economic Anxieties % agree: world severely affected by coronavirus 80.12 39.93 1006 67.61 46.82 914 -12.50 0.000% agree: US severely affected by coronavirus 77.83 41.56 1006 55.14 49.76 914 -22.69 0.000% worried about US economy 87.57 33.00 1006 68.05 46.65 914 -19.52 0.000% worried about personal econ. situation 73.76 44.02 1006 47.16 49.95 914 -26.60 0.000
Notes:
Online Appendix Table A.9 displays summary statistics for economic anxieties (Panel A) and coron-avirus perceptions (Panel B). Columns (1) - (3) display descriptives for Experiment 1 conducted on March 5,while Columns (4) to (6) display the descriptives for Experiment 2 conducted on March 16. able A.10: Impact of coronavirus arrival in US states on economic anxieties Predicted impact on (standardized) Worry about (standardized)(1) (2) (3) (4)World US US Economy Pers. Economic Sit.Any case 0.2318 ∗∗ ∗∗∗ ∗ Notes:
Online Appendix Table A.10 displays the effect of having at least one confirmed coronavirus case oneconomic anxieties using a difference-in-differences estimation with survey data collected on March 5 and 16.All regressions include state and day fixed effects and control for gender, age bin dummies, log income, logincome squared, dummies for having a high school degree and having some college education, dummies forbeing unemployed, currently working, a student and dummies for self-identifying as Democrat or Republi-can. The dependent variables in columns (1) to (2) are agreement on a five-point Likert-scale (from “stronglydisagree ” to “strongly agree”) with the statements “The world will be severely affected by the coronavirus.”(column (1)), and “The US will be severely affected by the coronavirus.” (column (2)). The dependent variablesin columns (3) and (4) are answers on a four-point Likert-scale (from “not at all worried” to “very worried”)to the questions “Are you worried about the effects of the coronavirus on the US economy?” (column (3)) and“Are you worried about the effects of the coronavirus on your household’s economic situation?” (column (4)).All outcomes are standardized to have mean 0 and standard deviation 1 within each survey wave. Standarderrors clustered at the state level are presented in parentheses. ∗ p < ∗∗ p < ∗∗∗ p < able A.11: The association of misperceptions and economic anxieties Predicted impact on (standardized) Worry about (standardized)(1) (2) (3) (4)World US US Economy Pers. Economic Sit.
Panel A: No control variables
Overestimate mortality 0.3655 ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ (0.0647) (0.0637) (0.0655) (0.0635)Overestimate contagiousness 0.5263 ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ (0.0899) (0.0843) (0.0879) (0.0825)R Panel B: Including control variables
Overestimate mortality 0.3830 ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ (0.0654) (0.0642) (0.0668) (0.0645)Overestimate contagiousness 0.5077 ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ (0.0885) (0.0832) (0.0849) (0.0810)R Notes:
Online Appendix Table A.11 displays the raw effect of overestimating mortality and contagiousness of coronavirus (relative toofficial estimates) on the perceived severity of the effects of the coronavirus and economic worries. The data were collected on March 5.Panel A shows the results without control variables. Panel B shows the results controlling for gender, age bin dummies, log income, logincome squared, dummies for having a high school degree and having some college education, dummies for being unemployed, currentlyworking, a student and dummies for self-identifying as Democrat or Republican. The table shows coefficients estimated using linearregressions that include indicators for respondents whose beliefs about coronavirus mortality were higher relative to official estimatesand for respondents whose beliefs about coronavirus contagiousness were higher relative to scientific estimates. The dependent variablesin columns (1) to (2) are agreement on a five-point Likert-scale (from “strongly disagree ” to “strongly agree”) with the statements “Theworld will be severely affected by the coronavirus.” (column (1)), and “The US will be severely affected by the coronavirus.” (column(2)). The dependent variables in columns (3) and (4) are answers on a four-point Likert-scale (from “not at all worried” to “very worried”)to the questions “Are you worried about the effects of the coronavirus on the US economy?” (column (3)) and “Are you worried aboutthe effects of the coronavirus on your household’s economic situation?” (column (4)). All outcomes are standardized to have mean 0 andstandard deviation 1. Heteroskedasticity robust standard errors are presented in parentheses. ∗ p < ∗∗ p < ∗∗∗ p < Predicted impact on (standardized) Worry about (standardized)(1) (2) (3) (4)World US US Economy Pers. Economic Sit.
Panel A: No control variables
Perceived mortality 0.8220 ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ (0.1887) (0.1730) (0.2013) (0.1841)Perceived contagiousness 0.2714 ∗∗ ∗∗∗ ∗∗∗ ∗∗∗ (0.1243) (0.1071) (0.1200) (0.1149)R Panel B: Including control variables
Perceived mortality 0.8841 ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ (0.1932) (0.1804) (0.2098) (0.1919)Perceived contagiousness 0.2728 ∗∗ ∗∗∗ ∗∗∗ ∗∗∗ (0.1239) (0.1070) (0.1226) (0.1187)R Notes:
Online Appendix Table A.12 displays the correlation between mortality and contagiousness of coronavirus on the perceived severityof the effects of the coronavirus and economic worries. The data were collected on March 5. Panel A shows the results without controlvariables. Panel B shows the results controlling for gender, age bin dummies, log income, log income squared, dummies for having ahigh school degree and having some college education, dummies for being unemployed, currently working, a student and dummies forself-identifying as Democrat or Republican. The table shows coefficients for beliefs about coronavirus mortality and about coronaviruscontagiousness estimated using linear regressions. Beliefs about mortality are rescaled to range from 0 to 1. Beliefs about contagiousnessare rescaled to display the association with believing that 100 extra people get infected. Beliefs about mortality and contagiousness arewinsorized at the 95th percentile to account for outliers. The dependent variables in columns (1) to (2) are agreement on a five-pointLikert-scale (from “strongly disagree ” to “strongly agree”) with the statements “The world will be severely affected by the coronavirus.”(column (1)), and “The US will be severely affected by the coronavirus.” (column (2)). The dependent variables in columns (3) and (4)are answers on a four-point Likert-scale (from “not at all worried” to “very worried”) to the questions “Are you worried about the effectsof the coronavirus on the US economy?” (column (3)) and “Are you worried about the effects of the coronavirus on your household’seconomic situation?” (column (4)). All outcomes are standardized to have mean 0 and standard deviation 1. Heteroskedasticity robuststandard errors are presented in parentheses. ∗ p < ∗∗ p < ∗∗∗ p < Table A.13: Experimental integrity: Balance table
Mortality information experiment Contagion information experiment(1) (2) (3) (4) (5) (6)Mean low rel. mortality Mean high rel. mortality p(low rel. mort. = high rel. mort) Mean no contagion info Mean contagion info p(no info = info)% Male 50.55 47.49 0.36 50.11 47.93 0.51% Age < 35 23.74 24.62 0.76 23.96 24.40 0.88% Highschool education 18.90 16.34 0.31 16.70 18.52 0.47% College eductation 78.90 82.14 0.22 81.98 79.08 0.27% Currently working 58.46 51.63 0.04 55.38 54.68 0.83% Democrat 38.90 41.18 0.48 37.58 42.48 0.13% Republican 33.41 32.90 0.87 34.95 31.37 0.25% High trust in science 1.98 0.22 0.01 0.88 1.31 0.53p-value of joint significance 0.00 0.74
Notes:
Online Appendix Table A.13 displays balance tests for the experimental sample. The data were collected on March 5. Columns (1)to (3) show means for both experimental groups in the mortality information experiment and the p-value for a test of equality of meansacross samples. Columns (4) to (6) show means for both experimental groups in the contagiousness experiment and the p-value for a testof equality of means across samples. p-values are obtained using heteroskedasticity robust standard errors. p-values for the test of jointsignificance are based on the F-statistic obtained by regressing all observables on the treatment indicators.
Panel A: No control variables
High relative mortality 0.2833 ∗∗∗ ∗∗∗ (0.0655) (0.0658)R Panel B: Including control variables
High relative mortality 0.2705 ∗∗∗ ∗∗∗ (0.0659) (0.0651)R Notes:
Online Appendix Table A.14 displays the impact of information about the coronavirus on the perceived severity of the impacts ofthe coronavirus. The data were collected on March 5. Panel A shows the results without control variables. Panel B shows the results con-trolling for gender, age bin dummies, log income, log income squared, dummies for having a high school degree and having some collegeeducation, dummies for being unemployed, currently working, a student and dummies for self-identifying as Democrat or Republican.The table shows coefficients estimated using linear regressions that compare respondents who were truthfully informed that the death rateof the coronavirus is either “20 times higher than for the flu” (high mortality treatment), or “5 times lower than for SARS” (low mortalitytreatment). The dependent variables in columns (1) to (2) are agreement on a five-point Likert-scale (from “strongly disagree ” to “stronglyagree”) with the statements “The world will be severely affected by the coronavirus.” (column (1)), and “The US will be severely affectedby the coronavirus.” (column (2)). All outcomes are standardized to have mean 0 and standard deviation 1. Heteroskedasticity robuststandard errors are presented in parentheses. ∗ p < ∗∗ p < ∗∗∗ p < Worry about (standardized)(1) (2)US Economy Pers. Economic Sit.
Panel A: No control variables
High relative mortality 0.1557 ∗∗ ∗∗ (0.0660) (0.0660)R ∗∗ (0.0434) (0.0415)R Panel B: Including control variables
High relative mortality 0.1556 ∗∗ ∗∗∗ (0.0660) (0.0645)R ∗∗ (0.0435) (0.0416)R Notes:
Online Appendix Table A.15 displays the impact of information about the coronavirus on economic anxiety. The data were collectedon March 5. The results show that information about coronavirus causally affects economic anxiety. Panel A shows the results withoutcontrol variables. Panel B shows the results controlling for gender, age bin dummies, log income, log income squared, dummies for havinga high school degree and having some college education, dummies for being unemployed, currently working, a student and dummiesfor self-identifying as Democrat or Republican.
High relative mortality shows coefficients estimated using linear regressions that comparerespondents who were truthfully informed that the death rate of the coronavirus is either “20 times higher than for the flu” (high mortalitytreatment), or “5 times lower than for SARS” (low mortality treatment).
Contagion information shows regression coefficients that comparerespondents who were truthfully informed about the estimated contagiousness of coronavirus (R0 ≈
2) to respondents who were givenno information. Estimates for
Contagion information are obtained with an ANCOVA specification using baseline outcomes obtained inthe same survey prior to the information treatment. The dependent variables in columns (1) and (2) are answers on a four-point Likert-scale (from “not at all worried” to “very worried”) to the questions “Are you worried about the effects of the coronavirus on the USeconomy?” (column (1)) and “Are you worried about the effects of the coronavirus on your household’s economic situation?” (column(2)). All outcomes are standardized to have mean 0 and standard deviation 1. Heteroskedasticity robust standard errors are presented inparentheses. ∗ p < ∗∗ p < ∗∗∗ p < Panel A: No control variables
Contagion information -0.0107 -0.0876 ∗∗ (0.0435) (0.0414)R Panel B: Including control variables
Contagion information -0.0052 -0.0836 ∗∗ (0.0436) (0.0415)R Notes:
Online Appendix Table A.16 displays the impact of information on the contagiousness of coronavirus on economic anxiety. Thedata were collected on March 5. All specifications include a dummy for whether the respondent saw the high relative mortality treatmentduring the first experimental variation. Panel A shows the results without control variables. Panel B shows the results controlling forgender, age bin dummies, log income, log income squared, dummies for having a high school degree and having some college education,dummies for being unemployed, currently working, a student and dummies for self-identifying as Democrat or Republican.
Contagioninformation shows regression coefficients that compare respondents who were truthfully informed about the estimated contagiousness ofcoronavirus (R0 ≈
2) to respondents who were given no information. Estimates are obtained with an ANCOVA specification using baselineoutcomes obtained in the same survey prior to the information treatment. The dependent variables in columns (1) and (2) are answers on afour-point Likert-scale (from “not at all worried” to “very worried”) to the questions “Are you worried about the effects of the coronaviruson the US economy?” (column (1)) and “Are you worried about the effects of the coronavirus on your household’s economic situation?”(column (2)). All outcomes are standardized to have mean 0 and standard deviation 1. Heteroskedasticity robust standard errors arepresented in parentheses. ∗ p < ∗∗ p < ∗∗∗ p < Worry about (standardized)(1) (2)US Economy Pers. Economic Sit.
Panel A: No control variables
Contagion information 0.0270 -0.1243 ∗∗ (0.0576) (0.0493)Contagion information × Low relative mortality -0.0759 0.0743(0.0675) (0.0571)Baseline value 0.7534 ∗∗∗ ∗∗∗ (0.0248) (0.0228)R Panel B: Including control variables
Contagion information 0.0427 -0.1188 ∗∗ (0.0576) (0.0494)Contagion information × Low relative mortality -0.0965 0.0709(0.0667) (0.0567)Baseline value 0.7467 ∗∗∗ ∗∗∗ (0.0259) (0.0250)R Notes:
Online Appendix Table A.17 shows the interaction effect of our treatments on economic anxiety. The data were collected on March5. Panel A shows the results without control variables. Panel B shows the results controlling for gender, age bin dummies, log income, logincome squared, dummies for having a high school degree and having some college education, dummies for being unemployed, currentlyworking, a student and dummies for self-identifying as Democrat or Republican. The dependent variables in columns (1) and (2) areanswers on a four-point Likert-scale (from “not at all worried” to “very worried”) to the questions “Are you worried about the effects of thecoronavirus on the US economy?” (column (1)) and “Are you worried about the effects of the coronavirus on your household’s economicsituation?” (column (2)). The outcome variables are elicited after both treatments. Estimates are obtained with an ANCOVA specificationusing baseline outcomes obtained in the same survey after the mortality treatment and prior to the contagiousness information treatment.We do not include a baseline indicator for the mortality treatment as it would be collinear with the baseline measures of economic worrieswhich are themselves sufficient statistics for the assignment probabilities regarding the mortality treatments. For interpretation, we use theinteraction between the low mortality treatment (as opposed to the high mortality treatment), as it is complementary to the contagiousnessinformation given that both treatments render the severity of the coronavirus less intense. All outcomes are standardized to have mean 0and standard deviation 1. Heteroskedasticity robust standard errors are presented in parentheses. ∗ p < ∗∗ p < ∗∗∗ p < Log predicted cases after twenty days Exponential mental model(1) (2) (3) (4)OLS Post LASSO OLS Post LASSOMale 0.236 0.213 0.0163(0.287) (0.285) (0.0240)Aged 25 to 34 -0.0312 -0.244 0.000318 -0.0361(0.490) (0.381) (0.0415) (0.0300)Aged 35 to 44 0.153 -0.115 0.0539(0.531) (0.407) (0.0455)Aged 45 to 54 0.296 -0.0193 -0.0603 ∗∗ (0.495) (0.0398) (0.0272)Aged 55 to 64 0.536 0.354 0.0400(0.534) (0.421) (0.0431)Aged 65 above 1.289 ∗∗ ∗∗ ∗∗∗ ∗∗∗ (0.560) (0.450) (0.0472) (0.0353)Log income 2.792 0.370 ∗∗ ∗∗ (3.450) (0.182) (0.271) (0.0139)Log income squared -0.113 -0.0166(0.162) (0.0127)Highschool education 1.265 ∗∗ ∗∗ (0.499) (0.0370)College eductation 2.257 ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ (0.477) (0.314) (0.0354) (0.0255)Currently working -0.687 ∗ -0.667 ∗ -0.0428 -0.0366(0.379) (0.373) (0.0322) (0.0278)Student -0.804 -0.780 -0.0317(0.602) (0.601) (0.0472)Democrat -0.658 ∗∗ -0.642 ∗ -0.0420 -0.0346(0.330) (0.329) (0.0278) (0.0230)Republican -0.359 -0.336 -0.0165(0.376) (0.376) (0.0302)Constant -12.79 1.335 -2.215 -0.250 ∗ (18.34) (1.764) (1.436) (0.140)R Notes:
Online Appendix Table A.18 displays the correlations of covariates with different classifications of mental models. These datawere collected on March 16. Columns (1) and (2) show the correlation with log predicted of cases after 20 days from a fictitious disease.Columns (3) and (4) show the correlation with an indicator variable being classified as having an exponential model by k-means clustering.Columns (1) and (3) show the result for a simple OLS regression. Columns (2) and (4) show results using OLS with only the variablesselected by the LASSO algorithm using α and λ parameters chosen by 10-fold cross-validation. Heteroskedasticity robust standard errorsare presented in parentheses. ∗ p < ∗∗ p < ∗∗∗ p < Predicted impact on (standardized) Worry about (standardized)(1) (2) (3) (4) (5) (6) (7) (8)World World US US US Economy US Economy Pers. Economic Sit. Pers. Economic Sit.
Panel A
Log(estimate day 5)- z-score 0.0971 ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗ ∗∗ Panel B
Log(estimate day 10)- z-score 0.1044 ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗ ∗∗ -0.0229 0.0086(0.0291) (0.0292) (0.0295) (0.0303) (0.0313) (0.0314) (0.0327) (0.0321)R Panel C
Log(estimate day 20)- z-score 0.0828 ∗∗∗ ∗ ∗∗∗ ∗∗∗ Notes:
Online Appendix Table A.19 displays the regression coefficients of perceived severity of the effects of the coronavirus with partic-ipants’ standardized log estimate of the spread of a fictitious disease. The data were collected on March 16. The table shows coefficientsestimated using linear regressions that regress perceived crisis severity and economic anxieties on the z-scored log of estimated infectionsfrom a fictitious disease. The dependent variables in columns (1) to (4) are agreement on a five-point Likert-scale (from “strongly disagree” to “strongly agree”) with the statements “The world will be severely affected by the coronavirus.” (columns (1) and (2)), and “The US willbe severely affected by the coronavirus.” (columns (3) and (4)). The dependent variables in columns (5) to (8) are answers on a four-pointLikert-scale (from “not at all worried” to “very worried”) to the questions “Are you worried about the effects of the coronavirus on theUS economy?” (columns (5) and (6)) and “Are you worried about the effects of the coronavirus on your household’s economic situation?”(columns (7) and (8)). The right-hand-side variables are the standardized log of participants’ estimates for the number of people infectedwith the fictitious disease on day 5, day 10 and day 20, respectively. All outcomes are standardized to have mean 0 and standard deviation1. Even columns control for gender, age bin dummies, log income, log income squared, dummies for having a high school degree and hav-ing some college education, dummies for being unemployed, being currently working, being a student and dummies for self-identifyingas Democrat and Republican. Heteroskedasticity robust standard errors are presented in parentheses. ∗ p < ∗∗ p < ∗∗∗ p <0.01