A dispute over one Florida governor’s race poll has become a larger test of how political surveys are funded, adjusted and presented to the public. The pollster accepts a disclosure failure and methodological error but rejects the claim that it deliberately fabricated responses.
A second polling firm admitted to fabricating data, according to the allegation that emerged after the Florida primary, but the firm at the center of the dispute, The Public Sentiment Institute, disputes that description. The firm acknowledged a methodological error and did not disclose that James Fishback’s campaign had funded its survey in Florida’s Republican gubernatorial primary.
Polling tracker FiftyPlusOne has permanently barred TPSI data from its database, saying the adjustment appeared to improve Fishback’s standing. TPSI says no respondent was removed, no response was altered and no data was fabricated — a distinction that now sits at the center of the controversy.
How the Florida poll changed
TPSI’s survey placed Rep. Byron Donalds first in Florida’s GOP contest for governor. Its published result put Fishback in second place and Florida Lt. Gov. Jay Collins in third.
After the election, TPSI said it reprocessed the raw data as part of an accuracy review. The firm said the unadjusted responses produced measurements in line with the certified results, within the applicable margin of error.
The problem, by TPSI’s own account, came from what it called an “experimental layer.” The firm said it had applied prior survey results after the poll’s initial findings did not match its view of the race.
That process identified respondents who selected Collins but whose broader profiles appeared to resemble a different political coalition. TPSI said it then leaned those respondents toward Fishback. The adjustment changed the reported order of the candidates, moving Fishback ahead of Collins.
Why FiftyPlusOne called it fabrication
FiftyPlusOne reached a far more severe conclusion. The polling tracker said it found evidence that TPSI fabricated vote-choice data for selected respondents and said the action appeared to boost the candidate who paid the firm for polling services.
On that basis, FiftyPlusOne said it would permanently stop using TPSI’s polling. For a tracker that aggregates surveys into broader race estimates, removing a pollster’s work is a consequential judgment: it signals that the organization no longer considers the firm’s published numbers reliable enough to include.
The tracker also emphasized the undisclosed financial relationship. TPSI’s Florida survey was funded by Fishback’s campaign, according to FiftyPlusOne, yet the poll was not publicly labeled as campaign-sponsored when released.
That disclosure issue matters independently of the data dispute. A poll commissioned by a candidate can still provide useful information, but readers and analysts need to know who paid for it before deciding how much weight to give it.
TPSI accepts errors, rejects allegation
TPSI has accepted responsibility for failing to disclose the funding source and for the experimental adjustment. It said that layer has been discontinued and will not be used in future releases.
But the firm rejects the claim that it deliberately falsified a survey. In its response, TPSI said no respondent had been removed, no individual response had been altered and no data had been fabricated.
Its defense rests on a difference between the raw survey record and a later model-based modification. TPSI maintains that the raw data, when rerun after the election, aligned with the eventual certified outcome within the survey’s margin of error.
Critics see the issue differently: once a pollster shifts reported vote choice because respondents do not fit its expectations of the race, the published poll no longer cleanly reflects the answers collected. That is why the phrase “experimental layer” has drawn such scrutiny.
A second controversy in days
The Florida dispute followed a separate episode involving Median Strategies, which said it had spread false polling on the Los Angeles mayoral race as a short-term social experiment about the circulation of unverified information.
That earlier admission is why reports have described TPSI as a second polling firm caught up in a fabricated-data controversy. Yet the cases are not identical. Median Strategies said it had disseminated false polling; TPSI says its underlying raw responses were real and argues that its failure was an improper adjustment and inadequate disclosure.
The distinction may sound technical, but it is material. Deliberately inventing survey results, changing the interpretation of real responses after collection, and failing to disclose a client are different breaches — even if each can undermine confidence in the final number.
For voters, campaigns and news organizations, the common lesson is simpler: a poll should explain who funded it, how respondents were selected, what questions were asked and how any weighting or adjustment was performed.
Other polling concerns remain
FiftyPlusOne also said it had temporarily suspended use of polling from Patriot Polling over concerns that it, too, did not disclose Fishback’s campaign as a client in the Florida race. The tracker said surveys from both firms already in its database would be marked partisan because it could not be confident they were conducted on a nonpartisan basis as claimed.
Patriot Polling had not publicly responded in the material reviewed by The Hill. That makes its situation unresolved, unlike TPSI’s acknowledgment of an undisclosed sponsor and an erroneous experimental process.
VoteHub co-founder and head of research Austin Park said his organization had already excluded TPSI from its polling averages for months. His reaction reflects the practical consequence for pollsters: once a firm’s methods are challenged, even accurate-looking future results may receive less attention or be excluded entirely.
The larger credibility test
Donalds won Florida’s Republican gubernatorial primary, defeating both Fishback and Collins. The certified result limits the immediate electoral impact of one flawed poll, but it does not erase the broader concern.
Polls can shape campaign coverage, donor perceptions and expectations before ballots are cast. A survey that moves a candidate into second place can affect the story of a race even when it does not change the winner.
The unresolved question is whether TPSI’s explanation satisfies analysts who view the adjustment as a serious but correctable methodological failure, or whether its process crossed the line into data falsification. FiftyPlusOne has clearly chosen the latter interpretation; TPSI has clearly rejected it.
What is settled is narrower but still significant: the firm did not disclose its campaign client, used an experimental process that altered the reported contest, and has said it will not repeat that process. In polling, trust depends not only on getting close to the final result, but on showing the public exactly how a number was produced.
Leave a Reply