Skip to content

fivethirtyeight/covid-19-polls

Repository files navigation

covid-19-polls

This repository contains the data behind How Concerned Are Americans About The Coronavirus?

Polls

covid_approval_polls.csv contains polls that ask Americans whether or not they approve of the way the president handling covid-19.

covid_concern_polls.csv contain the polls ask Ameicans how concerned they feel about aspects of the outbreak such as infection and economic impact.

Column Description
subject For approval polls, this column marks whose handling of covid-19 the approval poll is about (e.g. Trump). For concern polls, this column identifies which subject the poll is asking Americans about (for example, concern-infected vs concern-economy)
party Party of respondents
startdate Start date of poll
enddate End date of poll
pollster Organization that conducted the poll
sponsor Organization that sponsored the poll
samplesize Size of polling sample
population A for adults, RV for registered voters, LV for likely voters
tracking TRUE if the poll is a tracking poll, meaning that the pollster is releasing data with overlapping samples
text Text of the question the pollster asked
url Link to the poll

Adjusted Polls

covid_approval_polls_adjusted.csv and covid_concern_polls_adjusted.csv contain the polls after adjustments are applied by our poll-averaging algorithm. Additional columns include:

Column Description
modeldate Date of model run
grade Grade given to the pollster in our Pollster Ratings
weight Weight given to each poll in the model
influence Weight given to each poll, adjusted for recency
multiversions * denotes that multiple versions of a poll in the raw data file were combined (see note below)

Averages

covid_approval_toplines.csv and covid_concern_toplines.csv contain the calculated daily averages for the approval and concern polls respectively.

Our averages are calculated similarly to how we handle presidential approval ratings, which means we account for the quality of the pollster and each pollster’s house effects (whether they seem to yield unusually high or low numbers for each question compared with the polling consensus), in addition to a poll’s recency and sample size. In cases where the pollster did not provide sample sizes by party, they were calculated based on the percentage of total respondents who identified with each party.

If the same poll asked more than one relevant question (using different wording), we included both questions, but the results of those questions were averaged together, then input into the model, so the poll was not double counted.