Beliefs, Policy Preferences, and Political Behavior During the COVID-19 Pandemic: Evidence from a Survey Experiment

The first two treatments are intended to generate exogenous variation in respondents' level of distrust and the structure of paranoia. In turn, this will enable us to estimate the impact of distrust and paranoia on our final outcomes. The last two treatments are intended to generate exogenous variation in respondents' economic beliefs (both with respect to their own expected income and the economic outlook in the country as a whole). In turn, this will enable us to estimate the impact of these beliefs on our final outcomes.

2020-04-16

2020-04-26

We will measure the following primary outcomes:
* respondents' support for social distancing (in how many months it should stop), and requiring the Chinese government to pay for the pandemic
* respondents' expected support for the Democratic vs. the Republican candidate in the upcoming gubernatorial election; their expected support for Trump vs. Biden in the upcoming presidential election; and their likelihood to vote in the upcoming presidential election

We will measure the following secondary and intermediate outcomes:
* respondents' beliefs on the change between their 2019 and 2020 household pre-tax income (and on their own likelihood to be unemployed by June 30 and by December 31 and to receive unemployment insurance in that case) and on the economic outlook in the country as a whole
* respondents' beliefs on the responsibility of the Chinese government in the spread of the crisis; on whether the U.S. federal and state governments were influenced by pressure groups or following their own agenda in their response to the crisis; and on whether the media exaggerated the crisis to hurt President Trump

In the context of the coronavirus epidemic, we measure people's expectations on their future income and the economic outlook in the country as a whole; their beliefs on the responsibility of the Chinese and U.S. federal and state governments for the crisis; their support for certain policies (social distancing, and requiring reparations from China); and their expected vote choice and participation in the 2020 U.S. elections. We identify the sociodemographic and attitudinal factors that best predict people's attitudes about the crisis and their expected voting behavior. Using a survey experiment, we provide causal evidence on the impact of four treatments on these outcomes. The first two treatments are intended to generate exogenous variation in respondents' level of distrust and paranoia (towards both domestic and foreign actors, and concerning both the origin and handling of the crisis). In turn, this will enable us to estimate the impact of distrust and paranoia on our final outcomes. The last two treatments are intended to generate exogenous variation in respondents' economic beliefs (both with respect to their own expected income and the economic outlook in the country as a whole). In turn, this will enable us to estimate the impact of these beliefs on our final outcomes. We will compare the magnitude of the impact of distrust and paranoia to the magnitude of the impact of economic beliefs. We will measure treatment impact heterogeneity along economic situation, assessed and actual seriousness of the crisis, ideology, preexisting distrust and paranoia, and predicted populist vote.

The survey experiment is conducted with two distinct samples: a sample of about 10,000 M-Turk respondents; and (in a different format, only including the two first treatments and without stratification), a nationally representative online sample of about 2,000 respondents, tracked over time (these respondents were surveyed a first time about three weeks before the experiment and they will be surveyed again twice after the experiment).

Respondents are randomized into four groups, after stratifying by gender, age, the number of children living in the household, and people's views on poverty. Each respondent will be assigned with equal (25%) probability to each of the four groups.

For the part of the experiment on M-Turk, the randomization is done by the randomizer of the survey software (Qualtrics).

For the part of the experiment in a nationally representative online sample, the randomization is done by the randomizer of the survey company (CSA).

The randomization is done at the individual level.

No

About 10,000 individuals (M-Turk) and 2,000 individuals (nationally representative sample): 12,000 in total

About 10,000 individuals (M-Turk) and 2,000 individuals (nationally representative sample): 12,000 in total

Treatment 1a: about 2,500 + 500 = 3,000 in total
Treatment 2a: about 2,500 + 500 = 3,000 in total
Treatment 1b: about 2,500 + 500 = 3,000 in total
Treatment 1b: about 2,500 + 500 = 3,000 in total