Logo

FiveThirtyEight's Pollster Ratings

FiveThirtyEight's Pollster Ratings

This directory contains the data behind FiveThirtyEight's pollster ratings.

See also:

Past data can be found in the directories in this repository. The codebook for data calculated between 2014 and 2023 can be found here.


raw-polls.csv contains all of the polls analyzed to give each pollster a rating. Descriptions for each column are in the table below.

HeaderDefinition
poll_idFiveThirtyEight poll ID number.
question_idFiveThirtyEight question ID number. Note that there may be more than one question from the same poll included in the data.
race_idFiveThirtyEight race ID number.
cycleElection cycle.
locationGeography of the question. This may be a state, House district, or the US.
type_simpleCategory of the race. This value is structured with the type of election (Sen, Gov, Pres, or House) followed by a dash, and then either P or G to indicate if the survey is of a primary or general election, respectively. House generic ballot polls are noted as House-G-US.
raceRace for this question. This is structured as the year, followed by an underscore and the simple type, followed by an underscore and the location. For presidential primaries, the party is indicated with a -R or -D after Pres.
pollsterPollster name.
pollster_rating_idFiveThirtyEight pollster rating ID number.
aapor_roperBoolean flag for whether the pollster is a member of the AAPOR Transparency Initiative or contributes polls to the Roper Center.
inactiveBoolean flag for whether the pollster appears to still be conducting publicly available opinion polls.
methodologyMode used to conduct this poll. This should be a combination of one or more of the following values, separated by a forward slash:
  • Live Phone — Live telephone interviews, may or may not include calls to cell phones
  • IVR — Interactive voice response, otherwise known as automated polls or "robopolls"
  • Online Panel — Opt-in online panel, either proprietary or contracted to a panel provider
  • Text-to-Web — Recruitment via text messages that direct participants to a website to complete the survey
  • Probability PanelProbability based online panel
  • Email — Recruitment via emails that direct participants to a website to complete the survey
  • Mail-to-Web — Recruitment via snail mail that directs participants to a website to complete the survey
  • Mail-to-Phone — Recruitment via snail mail that directs participants to call a phone number to complete the survey
  • Text — Recruitment via text message in which questions are asked directly over SMS
  • Mail— By U.S. mail or other “snail mail” service
  • Online Ad — Recruitment via advertisement on websites, mobile games, or other online medium
  • App Panel — Opt-in panel using a proprietary app
  • Face-to-face — Poll conducted in person
partisanFlag for internal/partisan poll. "D" indicates Democratic poll, "R" indicates Republican poll, "I" indicates poll put out by independent candidate's campaign. Note that different sources define these categories differently and our categorization will often reflect the original source's definition. In other words, these definitions may be inconsistent and should be used carefully.
polldateMedian field date of the poll.
electiondateDate of the election.
time_to_electionNumber of days between the median field date of the poll and the election date.
samplesizeSample size of the question. For questions with no sample size available, this is approximated as the median sample size for races of the same type.
cand1_nameName of Candidate #1. Candidates #1 and #2 are defined as the top two finishers in the election (regardless of whether or not they were the top two candidates in the poll). In races where a Democrat and a Republican were the top two finishers, Candidate #1 is the Democrat.
cand1_idFiveThirtyEight ID number for Candidate #1.
cand1_partyParty code for Candidate #1.
cand1_pctCandidate #1's share of the vote in the poll.
cand1_actualActual share of vote for Candidate #1
cand2_nameName of Candidate #2.
cand2_idFiveThirtyEight ID number for Candidate #2.
cand2_partyParty of Candidate #2.
cand2_pctCandidate #2's share of the vote in the poll.
cand2_actualActual share of vote for Candidate #2
margin_pollProjected margin of victory (defeat) for Candidate #1. This is calculated as cand1_pct - cand2_pct. In races between a Democrat and a Republican, positive values indicate a Democratic lead; negative values a Repubican lead.
margin_actualActual margin in the election. This is calculated as cand1_actual - cand2_actual. In races between a Democrat and a Republican, positive values indicate a Democratic win; negative values a Republican win.

pollster-ratings-combined.csv contains the output of the pollster ratings algorithm for each pollster. A description of the columns is in the table below.

HeaderDefinition
pollsterPollster name.
pollster_rating_idFiveThirtyEight pollster rating ID number.
aapor_roperBoolean flag for whether the pollster is a member of the AAPOR Transparency Initiative or contributes polls to the Roper Center.
inactiveBoolean flag for whether the pollster appears to still be conducting publicly available opinion polls.
numeric_gradeFiveThirtyEight numeric overall rating for the pollster.
rankFiveThirtyEight pollster rank based on numeric grade. Note that pollsters that are flagged as inactive do not have a numeric grade and appear at the end of the rankings; this is not intended to indicate anything about their skill level.
POLLSCOREScore including both error and bias (but not transparency). Lower numbers are better. Stands for Predictive Optimization of Latent skill Level in Surveys, Considering Overall Record, Empirically.
wtd_avg_transparencyAverage transparency score for polls conducted by this pollster, weighted by the age of the survey.
number_polls_pollster_totalNumber of questions used to evaluate the pollster.
percent_partisan_workPercent of the pollster's polls that are considered partisan by FiveThirtyEight's standards.
error_ppmPredictive Plus-Minus for pollster's absolute error, as described in the pollster ratings methodology.
bias_ppmPredictive Plus-Minus for pollster bias, as described in the pollster ratings methodology.
number_polls_pollster_time_weightedNumber of questions used to evaluate the pollster, weighted by the age of the survey.
ranking_valueValue used to determine pollster rank, as described in the pollster ratings methodology.

© 2025 All rights reservedBuilt with DataHub Cloud

Built with LogoDataHub Cloud