This is the season of political polling. Between now and Election Day, voters will be told almost daily that someone is surging, collapsing, closing the gap, pulling away, or suddenly vulnerable.
Read more No matter how hard Americans work, the government spends more
Polls appear scientific. They come with precise percentages, sample sizes, field dates, and margins of error. “Our candidate is gaining” sounds like campaign advertising. A poll showing the candidate gaining three points looks like objective evidence.
Polling was developed to measure political reality. Increasingly, it is also being used to manufacture political reality.
A favorable poll can attract donors, produce endorsements, generate national attention, discourage opponents, and persuade political organizations to invest in a race. Once repeated by campaigns, news organizations, aggregators, and prediction markets, one poll can create the appearance of an established trend.
In the wrong hands, polling is becoming a new form of deepfake. A fabricated video places recognizable people inside an invented event. A misleading poll places plausible percentages and scientific language inside an invented or distorted political reality. Both gain credibility because they look like familiar forms of evidence.
As is often said, every good lie contains a kernel of truth. The most effective misleading poll may contain real respondents, calculations, and percentages. What makes it misleading is how those elements were selected, adjusted, presented, and used.
Pollsters have many technical ways to influence results: wording or ordering questions to encourage certain answers, deciding whom to contact and classify as a likely voter, oversampling particular groups, weighting demographic or partisan categories, reallocating undecided respondents, and selectively releasing favorable findings.
Some are legitimate methodological judgments made in good faith. Every serious poll requires decisions about sampling, weighting, and turnout. But the same tools necessary to conduct a legitimate poll can push the results toward a preferred conclusion. Published percentages rarely tell readers which is which.
Recent elections have provided unusually clear examples.
Median Strategies released purported polls showing Los Angeles mayor Karen Bass leading challenger Nithya Raman by nearly 12 points and Wisconsin gubernatorial candidate Francesca Hong ahead by a commanding margin. Bass promoted her favorable result. Hong appeared so inevitable that another prominent candidate, Mandela Barnes, withdrew and identified her as the likely nominee.
Hong lost the primary to David Crowley.
Median then admitted that the polls were fabricated, withdrew them, and shut down its website. It called the operation a “short-term social experiment” designed to see how easily false polling could enter the political information system.
Whatever its actual motive, the experiment answered the question. Invented polls from an unknown organization were circulated by campaigns, reported by news outlets, and absorbed into the political narrative before anyone established that the surveys were real. The Bass poll even produced activity on political prediction markets.
Florida provided a different form of manipulation. During the Republican gubernatorial primary, a poll associated with James Fishback showed him running a strong second to eventual nominee Byron Donalds. The pollster later acknowledged that its respondents had placed Lieutenant Governor Jay Collins in the twenties. Because that did not agree with the firm’s “read of the race,” it moved some Collins supporters toward Fishback.
In plain English, respondents who selected Collins were reported as supporting someone else.
Read more Catch-22: Third-worldism (or is it Islam?) is a ‘mental illness’
The election supplied a useful reality check. Donalds won with approximately 48 percent. Collins finished second with about 25 percent. Fishback received approximately 10.5 percent. The original responses were much closer to what voters actually did than the pollster’s preferred interpretation.
Complete fabrication and reassignment of answers are extreme. More commonly, a poll can create a false impression without falsifying the interviews.
Consider Pennsylvania’s First Congressional District. Two internal polls by Global Strategy Group showed Republican Congressman Brian Fitzpatrick leading Democrat Bob Harvie. The first reported 48-42; the second reported 47-43. Fitzpatrick’s measured lead changed from six points to four.
That small difference was within normal sampling variation. It did not establish that Harvie had gained or Fitzpatrick had declined. Both polls showed Fitzpatrick leading.
Nevertheless, the second produced claims that Harvie was “closing the gap,” was “within the margin of error,” and had made the district increasingly flippable. Those claims generated coverage and could influence prediction markets and race ratings, which may then be cited as additional evidence that the race is changing.
There is another danger between the poll and the public: the person interpreting it. Candidates, journalists, and commentators routinely pontificate about candidates being “tied,” “surging,” or “within the margin of error” without understanding what those terms mean.
A small change between polls may be statistical variation rather than voter movement. “Within the margin of error” does not mean the race is tied. A poll showing one candidate ahead still shows that candidate ahead, although it may not measure the lead precisely. A confident voice is no substitute for competent analysis.
None of this means all polling is fraudulent or useless. Properly designed polls remain valuable. Nor does using a partisan pollster automatically make a survey false. Republican and Democratic campaigns both commission legitimate research.
Voters do not need to become statisticians, but they should ask:
Who paid for the poll and who conducted it, and are they related?
Were the questions relevant, and did the sample represent those likely to vote?
Do the numbers support the headline, campaign claim, or commentator’s interpretation?
Are several reports independent, or are they repeating one original poll?
A legitimate poll is a useful but imperfect snapshot. A campaign poll may be both research and advertising. A manipulated poll is political messaging disguised as mathematics. A fabricated poll is a deepfake made of numbers.
During this polling season, voters should examine not only what the numbers appear to say, but whether those numbers — and the claims made about them — are reasonable and true, just as they should with every other piece of campaign information.
Read more Is Ossoff the ‘macho’ the Dems have been looking for?
Image via PickPik.