Wedgewood Polls said the vast majority of its election releases lacked genuine survey data, prompting polling services to remove its entries and drawing attention to how published figures are checked.
Wedgewood described the operation as an experiment to show weaknesses in oversight of political polling. Its results had appeared in polling databases and news coverage before services removed them. The Guardian reported that Newsweek, The Hill, Breitbart, The Daily Beast and The Washington Examiner had also published Wedgewood polling.
Different tracker counts refer to separate services. The New York Times said Wedgewood surveys were no longer in its tracking project. Axios counted 15 Wedgewood entries on the Times' tracker, spanning releases from June through October. Decision Desk HQ said it removed 17 Wedgewood polls from its race-level lists and averages.
FiftyPlusOne, a polling aggregator, said its review of raw data for Wedgewood's Mississippi Senate poll found that all 511 respondent IDs in the file had nine digits. Wedgewood said it fielded the poll Oct. 12-14 using Pollfish. FiftyPlusOne estimated the chance of that pattern occurring randomly at less than 10^-170 under its stated assumptions.
How the surveys entered the record and what their removal did to forecasting are distinct questions. The Times said Wedgewood had answered its questions about pollster identities, sampling, how respondents were contacted, weighting and funding before the surveys were included. Nate Silver said his FLIPR model generally used midterm polls from the Times feed and did not usually apply a second screening step after inclusion. Silver said removing Wedgewood's data left the model's averages and forecasts basically unchanged.
Michigan state Sen. Mallory McMorrow said Wedgewood's figures harmed her campaign, with contributions falling and reporters repeatedly asking whether she would end her run. McMorrow's earlier account of the campaign impact described her response to the poll's effects.
The Associated Press reported in August that a separate operation, Median Strategies, acknowledged fabricating at least three polls across three states. The Federal Election Commission lists Nov. 3 as the next regularly scheduled federal general election.
