.jpg%20Polls&w=1920&q=75)
It must be getting close to Election Day because the polling world is getting messy again. One pollster this week admitted to switching some participants’ vote choices to increase support for a different candidate, while another firm published a statement on social media saying all its surveys were complete fabrications designed as a “short term social experiment.” How many more are out there spreading junk numbers and distorting public perception? If history is any indicator, probably several. Maybe that’s a good thing.
Trolling With Polls
First, let’s get a quick rundown of what happened this week. On August 9, Median Strategies published a post on social media saying, “Polling has a credibility problem. Ours is the fix.” Soon after, Median released a poll on the Wisconsin gubernatorial race showing Democratic socialist Francesca Hong ahead by more than 20 points; she narrowly lost to David Crowley.
“We got Wisconsin wrong- and by quite a margin too,” the firm posted on August 12. The following day, Median released a poll for the LA mayoral race, showing Los Angeles Mayor Karen Bass leading Nithya Raman by nearly 12 points. The results in both were completely fabricated.
On Monday, Median Strategies told The Los Angeles Times that the firm was “created as a short term social experiment to examine how purported polling information could enter and spread through the political information ecosystem without independent verification.”
Was it really a “social experiment”? If it was all done to prove a point, what did it achieve? It’s no secret how easy it is to whip up a polling firm and toss some numbers onto social media to manipulate opinions. That’s why some aggregators do a little research before publishing a pollster’s results on their page. But strong vetting isn’t always enough.
Bait and Switch
Just a couple of days ago, FiftyPlusOne published a piece about a tip it received last week, claiming that The Public Sentiment Institute had released “campaign internals for the James Fishback campaign” but didn’t disclose that “those polls were funded by a candidate.” On Tuesday, TPSI published a final pre-election survey of its Florida polls and confessed that it significantly overstated support for gubernatorial candidate James Fishback in the Republican primary.
“TPSI admits that the firm changed the stated vote choice of certain respondents who favored a different candidate, current Lieutenant Governor Jay Collins, turning these respondents into supporters of James Fishback based on their belief that Fishback should have been polling higher,” explained FiftyPlusOne. So TPSI was paid by Fishback, didn’t disclose it, and manipulated the results to make it seem as if Fishback had more support than he did.
The reason this is all so concerning is that polls reflect public opinion, which can also alter how people perceive candidates and races. It’s the bandwagon effect: When people see a candidate leading in a poll, it alters their perception of the political hopeful, and sometimes influences voters to choose the candidate just because they appear popular. Fabricated results can also help shape media narratives and make candidates appear stronger than they are, which can give a campaign momentum and drive fundraising.
But a poll doesn’t have to be fake to be misleading.
Decoder Rings Not Included
Many factors must be taken into consideration when conducting and interpreting polls. Primary polls, for instance, are typically off by an average of 13 points, according to a recent analysis by pollster G. Elliott Morris over at Strength In Numbers. This rule usually holds, but Wisconsin’s gubernatorial primary was off by more than that, leaving many to wonder if polls are useless. Several post-mortems suggest multiple factors were responsible for the polling failures there.
One reason some polls missed badly in Wisconsin might’ve been because they unintentionally included too many young people, inflating Hong’s lead. Late movement toward Crowley – after pollsters released their final polls – probably pushed the race a few points, too. Plus, as Morris pointed out, most voters for Francesca Hong were likely eager to respond to polls, while Crowley supporters were less likely to respond.
Pollsters often mathematically adjust survey responses so that the final sample matches the actual demographics of the real-world population. But they can only do so much, especially if the data they’re working with is faulty, and “the Wisconsin voter file is notorious for having a lot of missing and incorrect data,” explained Morris. “One of the variables that is particularly errant is the age variable.”
Many voters on Wisconsin’s voter file “have a birth year of 1800,” explained the polling firm State Navigate, “and we don’t believe in vampires.” Apparently, “these generally are newer voters who skew very young; about half of voters who we polled with a birth year of 1800 say they are in the 18-29 age bracket. Thus, if you were to simply take the voter file’s birth year data, you’ll get a Democratic primary electorate that is majority 65+ and has less than 2% of voters under the age of 30.”
However, just as important as knowing the demographics of samples is knowing what language is used in the surveys. Not all questions are equal.
Words Matter
The standard language for almost all survey questions for presidential approval is “Do you approve or disapprove of the way Donald Trump is handling his job as President?” The slightest change in wording can give drastically different responses. For instance, McLaughlin & Associates released a poll in July showing Trump’s approval at 49%. But the way it words the question and how it interprets the data is misleading:
“Which of the following statements regarding President Trump and his policies comes closest to your opinion?
- I approve of President Trump and his policies.
- I disapprove of President Trump personally, but I like a lot of his policies.
- I disapprove of President Trump and his policies.”
The results for questions 1 (36%) and 2 (13%) were then combined, giving Trump a 49% job approval rating, far higher than any other poll. Nowhere in these questions does it ask whether participants approve of the president’s job performance. “I disapprove of Trump personally, but I like a lot of his policies” has little to do with how Trump is handling the executive position.
Another factor to consider, especially with congressional ballot polls, is the sample size and whether the respondents were registered or likely voters. Obtaining a sample of 1,000 adults probably won’t be as accurate as one from 1,000 likely voters. After all, how much can somebody’s opinion be worth in a poll about voting if they’re not even a registered voter or don’t intend to vote? This is why Liberty Nation News’ polling aggregator only includes polls with samples of 1,000 or more registered or likely voters.
Of course, some polls deceive because someone made them up; others mislead because no one bothered to ask the right question. So many factors come into play, but a poll doesn't need to be fake to shift an election – people only need to believe the results. But if more people are aware that fake polls are relatively normal in the industry, perhaps they will approach real ones with a skeptical eye and act accordingly.
