On this page, we offer additional information supplementing the analyses shared in our Monkey Cage post about Facebook and COVID-19 vaccination rates. For more details, read the COVID States Project Report #57:
The figures below present results from logistic regressions with identical independent variables, estimating the likelihoods of saying one will not get vaccinated or has already received at least one vaccine dose, respectively. The vertical reference lines at zero indicate no effect; coefficients whose confidence intervals touch that line are not statistically distinguishable from zero. Coefficients that are to the right of the line are associated with a higher marginal probability of observing the outcome. For instance, as an otherwise-identical respondent moves from the youngest age group (18-24 years old) to the oldest age group (65 and over), the odds of being vaccinated increase and the odds of saying one does not intend to get vaccinated decrease.
The key independent variables of interest in these models are COVID news sources, based on an item that asked the respondent to indicate which of six sources, if any, they had used for COVID news or information in the previous 24 hours. These sources are CNN, Fox News, MSNBC, Facebook, Newsmax, and the Biden Administration. For modeling purposes, we construct a categorical variable with separate categories for selecting single sources; a category for selecting multiple sources where at least one was Facebook, Fox, or Newsmax; a category for selecting multiple sources, none of which were Facebook, Fox, or Newsmax, and a final category for those who did not select any of these sources. For the purposes of modeling, we use the category representing multiple sources, none of Facebook, Fox, or Newsmax as the reference category.
We include a variety of additional independent variables which are associated with vaccine attitudes and uptake. These include geographic region and context, age group, race, gender, education, household income, parental status, and partisan identification. Continuous variables (partisan identification and income) are recoded to represent the percentage of the maximum value, putting them on 0-1 scales such that coefficients are associated with the marginal change in log-odds of observing the outcome when moving from the minimum to maximum value.
These results are based on data collected between Jine 9, 2021 and July 7, 2021. We surveyed 20,669 individuals across all 50 states plus the District of Columbia. The survey was conducted by PureSpectrum via an online, non-probability sample, with state-level representative quotas for race/ethnicity, age, and gender. In addition to balancing on these dimensions, we reweighted our data using demographic characteristics to match the U.S. population with respect to race/ethnicity, age, gender, education, and living in urban, suburban, or rural areas. This was the latest in a series of surveys we have been conducting since April 2020, examining attitudes and behaviors regarding COVID-19 in the United States.