In the wake of the 2016 election, some observers, including many in the broader public, were surprised by Donald Trump’s victory. Much of this surprise stemmed from polling data from several swing states that showed Hillary Clinton with a solid lead, which turned out to be inaccurate – or at least misleading – on election day.
In Wisconsin, polling data in the weeks leading up to the election gave Clinton an average of a 6.5% lead, with some individual polls as high as 8%. In Michigan, similarly, Clinton had a 3.6% lead over Trump. According to the final results, Trump won Wisconsin by 0.7% and Michigan by 0.3%.
In Ohio, where Trump was polling ahead of Clinton with a 2.2% lead going into the election, Trump ended up winning by a larger margin than expected. On election day he took Ohio by 8.1%.
If the average polling results for Wisconsin, Michigan and Pennsylvania – three vital swing states – in the week leading up to the election had been correct, Clinton would have won the election. As a result, after the election there was considerable handwringing over the trustworthiness of public polls.
While some have argued that polls are weak sources of information to base election predictions due to problems with their methodology, others say that they still point to important trends and indicators, but admit that the design of the poll is crucial.
What went wrong?
According to Courtney Kennedy, the director of survey research at Pew Research Center, a nonpartisan polling organization, national polls in 2016 were largely correct, many of which showed Clinton with a marginal lead over Trump nationally, but there were issues with state polling.
In a system where state vote tallies can determine the winner of an election, as opposed to the overall popular vote, which is the case in the US presidential election, this matters. As polls predicted, Clinton won the popular vote by a few percentage points. On the state level, however, there were inaccuracies.
“The major problem was with state-level polls, many of which missed a late swing to Trump among undecided voters and did not correct for the fact that their responding samples contained proportionally too many college-educated voters (who were more likely to favor Clinton),” argued Kennedy.
For those hoping for more accurate polls, Kennedy added that “more rigorous survey weighting” and “heightened attention to the possibility of late shifts in voter preferences” could help alleviate these issues in the future.
Emerging 2020 polling data is already being utilized by partisans who claim to have momentum on their side.
According to compiled data from different polling organizations, Biden seems to have a strong lead over Trump in several important swing states.
Recent polls from nine different organizations covering seven different states, including Wisconsin, Michigan, Pennsylvania, North Carolina and Florida, show Biden with a lead in all but one.
The polling organization, Hodas and Associates, gives Trump a 5% lead in Pennsylvania in their poll conducted from May 9-13. Polls from another organization, Redfield and Wilton Strategies, however, show Biden with a 9% lead in Pennsylvania in a poll conducted from May 10-14.
Although the trends seem to be clear overall, individual results like these may help fuel unease about the reliability of polls in some voters’ minds. They also provide fodder for candidates in search of narratives that indicate their popularity.
President Trump, for instance, has used ambivalence over polling data not only to sow doubts about their accuracy, but also to cherry pick the ones that show him in a positive light when deemed convenient.
“I don’t believe the polls,” Trump said in late April after he was asked about Biden’s purported lead, adding in a tweet that the polls weren’t based in reality.
“FAKE POLLING, just like 2016 (but worse)!” he tweeted on April 30.
Despite the confusion in 2016 and the tendency for polling data to be politicized, pollsters seem to agree that if done correctly they can be useful tools that document changes in attitudes and perceptions over time.
As for election polling, despite periodic mistakes they have shown to be quite accurate overall.
In a 2018 study that looked at election polling data from 45 countries between the years of 1942 and 2017 and then compared them to the actual results, polls were found to generally be good indicators of the final outcome.
Despite some forecasting misses in the past few years that stick in the public’s mind, such as Brexit, according to the study the average level of accuracy of polls has not declined over the years.
As for the status of polling in 2020, polling organizations look poised to make sure the problems with state-level data are improved upon. As for the guaranteed accuracy of polls, however, even the most meticulous designs are likely to fall short.
“The things that we can fix, we’ve fixed,” said Patrick Murray, director of the Monmouth University Polling Institute in New Jersey. “But then there are things that are never fixable because each election presents its own sets of challenges,” he added.
Have a tip or story? Get in touch with our reporters at [email protected]
Sign up for daily news briefs from The Millennial Source here!