Misinformation During a Pandemic

Leonardo Bursztyn, Aakaash Rao, Christopher Roth, David Yanagizawa-Drott
2020 Social Science Research Network  
Media outlets often present diverging, even conflicting, perspectives on reality -not only informing, but potentially misinforming audiences. We study the extent to which misinformation broadcast on mass media at the early stages of the coronavirus pandemic influenced health outcomes. We first document large differences in content between the two most popular cable news shows in the US, both on the same network, and in the adoption of preventative behaviors among viewers of these shows. Through
more » ... both a selection-on-observables strategy and an instrumental variable approach, we find that areas with greater exposure to the show downplaying the threat of COVID-19 experienced a greater number of cases and deaths. We assess magnitudes through an epidemiological model highlighting the role of externalities and provide evidence that contemporaneous information exposure is a key underlying mechanism. Abstract Media outlets often present diverging, even conflicting, perspectives on reality -not only informing, but potentially misinforming audiences. We study the extent to which misinformation broadcast on mass media at the early stages of the coronavirus pandemic influenced health outcomes. We first document large differences in content between the two most popular cable news shows in the US, both on the same network, and in the adoption of preventative behaviors among viewers of these shows. Through both a selection-on-observables strategy and an instrumental variable approach, we find that areas with greater exposure to the show downplaying the threat of COVID-19 experienced a greater number of cases and deaths. We assess magnitudes through an epidemiological model highlighting the role of externalities and provide evidence that contemporaneous information exposure is a key underlying mechanism. JEL Codes: D1, I31, Z13. viewers; while viewership of Tucker Carlson Tonight is associated with changing behavior three days earlier (controlling for demographics and viewership of other shows and networks). Given the critical importance of early preventive measures (Bootsma and Ferguson, 2007; Markel et al., 2007) , these differences in the timing of adoption of cautious behavior may have significant consequences for health outcomes. For example, Pei et al. (2020) estimate that approximately half of all COVID-19 deaths in the United States at the early stages of the pandemic could have been prevented had non-pharmaceutical interventions (NPIs) such as mandated social distancing and stay-at-home orders been implemented one week earlier. While the behavioral changes our survey respondents report are likely not as extreme, and our survey is representative only of Republicans over the age of 55, this evidence nonetheless suggests that these differences in timing may have directly affected the spread of the pandemic. Motivated by our survey evidence of persuasive content, we examine disease trajectories in the broader population using county-level data on COVID-19 cases and deaths. In our primary analysis, we focus on health outcomes during the early stages of the pandemic where we would expect first-order effects of treatment -late February to mid-April -though in additional analyses we report our main outcomes until the time of writing. 2 We first show that, controlling for a rich set of county-level demographics (including the local market share of Fox News), greater local viewership of Hannity relative to Tucker Carlson Tonight is associated with a greater number of COVID-19 cases starting in early March and a greater number of deaths resulting from COVID-19 starting in mid-March. In a set of permutation tests across socio-economic, demographic, political, and health-related covariates, as well as across geographical fixed effects accounting for unobservable factors, we show that the established relationship is highly robust. 3 Even so, it is likely that areas where people prefer Hannity over Carlson might differ on a number of unobservable dimensions that could independently affect the spread of the virus. Thus, to identify our effect of interest, we employ an instrumental variable approach that shifts relative viewership of the two shows, yet is plausibly orthogonal to local preferences for the two shows and to any other county-level characteristics that might affect the virus' spread. In particular, we predict this difference in viewership using the product of (i) the fraction of TVs on during the start time of Hannity (leaving out TVs watching Hannity) and (ii) the local market share of Fox News (leaving out Hannity and Tucker Carlson Tonight). The idea of our instrument is simple: if people like to turn on their TVs to watch something when Hannity happens to be on instead of Tucker Carlson Tonight, the likelihood that viewers are shifted to watch Hannity is disproportionately large in areas where Fox News is popular in general. 4 We show that, conditional on a minimal set of controls and the main effects, the interaction term is uncorrelated with any among a larger number of variables that might independently affect the local spread of the coronavirus. We then show it strongly predicts viewership in the hypothesized direction. Using this instrument, we confirm the OLS findings that greater exposure to Hannity relative to Tucker Carlson Tonight is associated with a greater number of COVID-19 cases and deaths. Our results 2 In principle, there could be second-order effects due to behavioral adjustments and policy responses when local infections and deaths rise sharply due to the treatment. To estimate these endogenous dynamic effects is beyond the scope of the paper, which is why we focus on the early time period. 3 Indeed, an exercise following Oster (2019) to estimate the bias generated by omitted variables suggests that our estimated coefficients are negatively biased. 4 Leaving out Fox News from the first term and Hannity and Tucker Carlson Tonight from the second allows us to ensure that the variation we exploit is driven by general preferences for when to watch TV and general preferences for watching Fox News, rather than specific, potentially endogenous, preferences for the two shows.
doi:10.2139/ssrn.3580487 fatcat:cs5ywgkyzfdkpn56e4u5e5lary