1. What shapes support for travel bans?

International border closure to China was among the first visible policy steps taken by the Trump Administration to mitigate the COVID-19 pandemic, following by a rollout of travel bans in response to emerging global peaks in the pandemic. By mid-March 2020, the entry ban was extended to include 38 countries. Globally, 127 countries had similar no-entry policies in place (Dzankic et al., Reference Piccoli, Dzankic, Perret, Ruedin and Jacob-Owens2022).

The China travel ban was popular among Americans. Even as early as 2–4 February 2020, which precedes the (known) outbreak of COVID in the United States by a month, 51 percent of Americans supported denying entry to foreigners who have coronavirus (17 percent reporting “not sure”) and 41 percent of Americans supported blocking admittance to individuals traveling from China (32 percent “not sure”).Footnote ¹ In emergencies, the public gives presidents more leeway to make policy, especially in foreign affairs, and often presidential approval increases (Mueller, Reference Mueller1973). But political actors remain sensitive to domestic public opinion. Even then-candidate Joe Biden backed the travel ban, with his deputy campaign manager carefully noting “Science supported this ban, therefore he did too.”Footnote ²

We consider two broad explanations to account for public support for travel bans. First, in the US, immigration attitudes are specifically shaped by group attitudes (Nelson and Kinder, Reference Nelson and Kinder1996). For instance, negative views of Latino/a/x people increase support for restrictive immigration policy generally (Valentino et al., Reference Valentino, Brader and Jardina2013; Pérez, Reference Pérez2016). We draw on a rich literature showing how racial attitudes shape public opinion, including vote choice (Tesler, Reference Tesler2016) but also attitudes beyond race-based policies, exhibiting what Tesler (Reference Tesler2012b) calls a “spillover of racialization”. Work in this vein specifically illustrates how racial resentment, measured as attitudes about Black Americans, shapes opposition to immigration (Kinder and Sanders, Reference Kinder and Sanders1996; Kinder and Kam, Reference Kinder and Kam2010).Footnote ³

We therefore expect those with high racial resentment to exhibit higher support for COVID travel bans. As these attitudes vary by the national group in question, so too will support for travel bans vary by target country in the context of a global pandemic. Specifically, we expect those higher in racial resentment to support bans against China and Brazil more strongly than they do predominately white, high-income European countries (Great Britain, Italy). In this approach, China and Brazil represent racial outsider groups for white Americans, compared to European countries.

Our second theoretical expectation is that individuals support travel bans in response to their fears about the pandemic. Behavioral immune system theory predicts that individuals seek to mitigate risk (real or perceived) and participate in high pathogen-avoidance behavior, including taking on conservative attitudes (Rodriguez et al., Reference Rodriguez, Gadarian, Goodman and Pepinsky2022) and opposing immigration (Aarøe et al., Reference Aarøe, Petersen and Arceneaux2017). Large, disruptive events—including terrorism (Huddy et al., Reference Huddy, Feldman, Taber and Lahav2005), economic crises (Creighton et al., Reference Creighton, Jamal and Malancu2015), and previous pandemics like Ebola (Adida et al., Reference Adida, Dionne and Platas2018)—create uncertainty and anxiety, which can lead to patriotic rallying but also negative attitudes toward outsiders like immigrants (Hellwig and Sinno, Reference Hellwig and Sinno2017), as well as higher support for restrictive policies (Albertson and Gadarian, Reference Albertson and Gadarian2014). And where work from Pew Research Center showed that as COVID-19 infection rates and death tallies climbed, individuals reported greater health concerns (2020) and an aggregate increase in worries (2021), we might expect protectionist attitudes. Therefore, we predict Americans will support bans during a pandemic against travelers from countries experiencing high caseloads.

In testing the effects of racial attitudes and pandemic threat on support for travel bans, we include partisanship as a moderator. Partisanship is a social identity comprised of an increasingly aligned set of attitudes and positions (Mason, Reference Mason2018) and is a significant explanation for differences in public attitudes toward immigration policy (Levy and Wright, Reference Levy and Wright2020), including during health scares (Adida et al., Reference Adida, Dionne and Platas2018). Evidence from the pandemic already shows individuals follow ideological positions in evaluating border policy (Williams et al., Reference Williams, Gravelle and Klar2022). We therefore expect that the effects of racial attitudes and pandemic threat will vary by party affiliation: the relationship between racial resentment and support for travel bans will be higher among Republicans than Democrats, whereas the relationship between pandemic threat and support for travel bans may be stronger among Democrats than Republicans.

2. Research design

To examine American support for COVID-19 travel bans, we first fielded a large survey of American adults (N = 3000) between March 20 and 23 March 2020, collected by YouGov using its standard methodology (see the Appendix for sample details and recruitment procedures). We fielded a revised version of the survey experiment re-interviewing the same respondents four times: 6–25 June 2020 (Wave 2), 4–14 August 2020 (Wave 3), 15–21 October 2020 (Wave 4), and 24 March–5 April 2021 (Wave 5). The five-panel wave structure allows us to follow support for entry bans by country of origin over time.

In Wave 1, we embedded an experiment to investigate whether support for travel bans varied by the target country. Initially focusing on China, Italy, and Great Britain enables us to distinguish between racial attitudes and pandemic threat, as both Italian and Chinese outbreaks peaked in caseloads during our first survey wave, whereas Great Britain's first peak occurred over a month later (See Figure S1, though early peak sizes pale in comparison to later waves). Figure S1 also shows that global hotspots changed as the pandemic wore on. As rates in Italy, China, and Great Britain began to decline, cases in Brazil started to surge, leading it to be added to the U.S. entry ban list on 24 May. In Waves 2–5, respondents were randomly assigned to consider a travel ban from China, Great Britain, and Brazil (with Italy omitted).

Our dependent variable measures respondents' support for policies that restrict entry of citizens to manage the spread of COVID-19 in the United States. In Wave 1, the question reads: “The United States must continue to ban the entry of citizens of [China/Great Britain/ Italy] into the United States.” In Waves 2–5, the question replaces Brazil for Italy. Response options ranged from 1 to 5 on a Likert scale, where 1 indicates “strongly disagree” and 5 indicates “strongly agree.” Our treatment variable is the group to which this policy would be applied, represented by the terms in italics above. These were randomized with equal probability across all respondents. We report balance tests among treatment groups across all five survey waves in the Appendix (see Table S1).

We estimate the effects of varying targets countries on support for entry bans using ordinary least squares regression with robust standard errors. Let State ∈ {China, Italy, Great Britain} denote the randomly assigned state to which the entry ban is applied. Our baseline empirical model for Wave 1 is

(1)$$Y = \alpha + \beta _1{\rm State} + \gamma X + \varepsilon $$

Our baseline results adjust for no control variables. In further analyses, the elements of X adjust for pretreatment covariates: respondent age, gender, race, marital status, education, income, urban-rural status, and state of residence.

To explore heterogeneity in treatment effects by racial resentment and partisanship, we interact the treatment variable with measures of racial resentment and partisanship. We measure partisanship as a categorical variable of Democrats (with leaners), Republicans (with leaners), and Others (i.e., Independents and those without a partisan identification). We measure racial resentment using a survey item asking respondents the extent to which they agree or disagree with the view that, like other minorities, Black Americans should work their way up without special favors (for the complete question wording, see Appendix Table S2). This item is highly correlated with the other questions in the racial resentment scale (r = 0.75 in the 2020 American National Election Studies) and has been used as a proxy for the full scale in prior research (Tesler, Reference Tesler2012a). In Appendix Table S3, we validate our racial resentment measure as a predictor for general outgroup dispositions like anti-immigrant attitudes, where racial attitudes are tightly correlated to xenophobia and other ethnocentric positions (Kinder and Kam, Reference Kinder and Kam2010).

Let Party ∈ {Republican, Democrat, Other} denote the respondent's party and RR represent our measure of racial attitudes. Our model is

(2)$$\eqalign{Y = & \,\alpha + \beta _1{\rm State} + \beta _2{\rm Party} + \beta _3{\rm State} \times {\rm Party} + \beta _4{\rm RR} \cr & \quad + \beta _5{\rm RR} \times {\rm Party} + \beta _6{\rm RR} \times {\rm State} + \beta _7{\rm RR} \times {\rm State} \times {\rm Party} + \gamma X + \varepsilon } $$

which allows us to estimate a separate effect of State for members of each party and across the levels of RR. We include the triple interaction because we expect the relationship between racial resentment and travel bans to vary by party and target country. Although random assignment allows us to estimate the effect of State without controls, including X helps to ensure that the Party- and RR-specific effects that we estimate do not reflect respondents' other demographic characteristics. The survey items asking about racial resentment appeared prior to the experiment to ensure that they capture respondents' pre-treatment attitudes.Footnote ⁴

Last, for our four-wave (Waves 2–5), over-time analysis, we estimate least squares regressions that interact treatment status, partisanship, and RR with survey wave. With i denoting individuals and t waves, we estimate

(3)$$\eqalign{Y_{{\rm it}} = & \,\alpha + \beta _1{\rm Stat}{\rm e}_{{\rm it}} + \beta _2{\rm Part}{\rm y}_{\rm i} + \beta _3{\rm R}{\rm R}_{\rm i} + \beta _4{\rm Wav}{\rm e}_{\rm t} + \beta _5{\rm Stat}{\rm e}_{{\rm it}} \times {\rm Part}{\rm y}_{\rm i} \cr & \quad + \beta _6{\rm Stat}{\rm e}_{{\rm it}} \times {\rm R}{\rm R}_{\rm i} + \beta _7{\rm Stat}{\rm e}_{{\rm it}} \times {\rm Wav}{\rm e}_{\rm t} + \beta _8{\rm Part}{\rm y}_{\rm i} \times {\rm R}{\rm R}_{\rm i} + \beta _9{\rm Part}{\rm y}_{\rm i} \cr & \quad \times {\rm Wav}{\rm e}_{\rm t} + \beta _{10}{\rm R}{\rm R}_{\rm i} \times {\rm Wav}{\rm e}_{\rm t} + \beta _{11}{\rm Stat}{\rm e}_{{\rm it}} \times {\rm Part}{\rm y}_{\rm i} \times RR_{\rm i} + \beta _{12}{\rm Stat}{\rm e}_{{\rm it}} \cr & \quad \times {\rm Part}{\rm y}_{\rm i} \times {\rm Wav}{\rm e}_{\rm t} + \beta _{13}{\rm Part}{\rm y}_{\rm i} \times {\rm R}{\rm R}_{\rm i} \times {\rm Wav}{\rm e}_t + \beta _{14}{\rm Stat}{\rm e}_{{\rm it}} + {\rm Part}{\rm y}_{\rm i} \times {\rm R}{\rm R}_{\rm i} \cr & \quad \times {\rm Wav}{\rm e}_{\rm t} + \gamma X_{\rm i} + \varphi _{\rm i} + \varepsilon _{{\rm it}}} $$

Individual random effects φ_i capture individual-level heterogeneity.Footnote ⁵

To test our threat hypothesis, we consider the threat from caseloads in the countries mentioned in the experiment. In Waves 2–5 we measure objective threat through COVID-19 caseloads, using data on the smoothed daily number of new cases, collected by ourworldindata.org and represented in logarithmic terms.Footnote ⁶ To model how new cases condition our results, we replace Wave_t in equation (3) with Cases_State,t, which records the number of cases in the randomly assigned state at the beginning of that survey wave. We analyze these data using random effects regressions that interact target country (the treatment status) with that country's caseload and racial resentment. We measure objective caseloads as the daily number of new cases smoothed over a seven-day window.Footnote ⁷ In the analyses below we include all respondents in our surveys, but to ensure that our findings are not driven by the racial composition of our sample, we repeat all analyses with nonwhite respondents excluded where racial resentment has been found to be most predictive for white respondents (Kam and Burge, Reference Kam and Burge2018). These results can be found in the Appendix (Tables S10-S15).

We use a graphical approach to interpret the results of our analysis, estimating the predicted mean and standard error of each dependent variable for combinations defined by State, Party, Wave (where applicable), and RR dichotomized around its sample median.