It’s Labor Day weekend and the midterm elections are roughly 10 weeks away. That means political polling season is upon us. Are you interested in the polls, but not quite sure if you should trust them? Please read on.
After the 2016 presidential election, pollsters were criticized for failing to detect higher support for Donald Trump where it mattered, in the battleground states. Consequently, a number of assessments were conducted to explore what happened.
A blue-ribbon panel of experts convened by the American Assn. for Public Opinion Research concluded, “National polls were generally correct and accurate by historical standards,” which perhaps is contrary to the widespread perception. They also found that state polls “clearly underestimated Trump’s support in the Upper Midwest.” A few reasons were proffered for that shortcoming: voters changed their minds at the last minute, polls overrepresented more educated voters without adjusting for it, and some polling respondents simply did not tell the truth (e.g., “shy Trump supporters”).
Outright hacking of voting machines could lead to a clear difference between a poll that accurately reflects voter preferences and the official result.
Historically, in case you’re wondering, there’s been no consistent partisan slant in favor of Democrats. The 2016 national and state-level polls underestimated support for the Republican nominee, while in 2000 and 2012, they underestimated support for the Democrat.
Other research corroborated a general conclusion that, for the most part, the polls were fine, and that there is no crisis in polling accuracy. I guess we can say the polls were accurate except when they weren’t, which is always true.
That’s the thing to keep in mind: Polling results are always subject to imperfections and uncertainty, and any professional pollster understands this.
Imperfections are baked-in limitations that may include an over-reliance on unrepresentative samples without adequate correction, or a failure to accurately predict likely voters. Another ever-present problem is that respondents aren’t necessarily willing to tell the truth.
Polls use a margin of error (e.g., “plus or minus five points”) to quantify uncertainty, which can be controlled by using a large enough sample. But margins of error cannot fully account for the aforementioned imperfections inherent in all polls.
As we head into the 2018 midterm elections, there are two somewhat novel factors that could influence polling performance.
First, external meddling. Provocative late-in-the-game social media postings, for instance, could influence some reluctant voters to show up at the polls and sway others to give up on voting altogether, thus disrupting likely voter models. More extreme, outright hacking of voting machines could lead to a clear difference between a poll that accurately reflects voter preferences and the official result.
Second, there is the possibility of voter suppression. Recently, there was an attempt in Randolph County, Ga., to eliminate local voting sites. Earlier this year, more than 100,000 voters were inexplicably purged in Los Angeles County. And voter ID laws in a number of states will surely discourage some registered voters from casting a ballot. Again, voter suppression could render good, methodologically sound polls inaccurate if people who intend to vote cannot, for whatever reason, succeed in doing so.
Now, some wise words for the midterms:
—Remember that no poll is right all the time. Statistical uncertainty is inherent in all polls and it is often greater than what is typically posted.
—Keep in mind that polls capture voter sentiment at a point in time, and opinions and intentions can and do change from one day to the next. Polls are not designed to forecast the future.
—Participants’ answers reflect and are influenced by many external factors, including the news of the day (October surprises), views of friends, their willingness to tell the truth when asked a direct question, even their understanding of the question being asked. The best polls are designed to mitigate these factors, but they can’t account for day-of problems, including: voting machine breakdowns (accidental or from hacking), voter suppression at the polling location (e.g., voter roll purges), or even the weather (e.g., thunderstorms, cold fronts).
So, should you trust polls? Yes — provided that you understand they come in many quality flavors and that even the good ones will be off the mark at some point. Never, ever use polling results to decide whether or not to vote. Use them to understand the kaleidoscope of our electorate’s sentiment on various issues. And remember, while you can’t control what others do, you can control your own uncertainty by committing to vote no matter what. You owe it to yourself and you owe it to your country.
Robert Santos is vice president and chief methodologist of the Urban Institute in Washington, D.C.