While almost all public polling predicted a landslide victory for President-elect Joe Biden, RIWI’s technology showed for weeks in advance that the race would be much tighter, in line with the actual results.
Traditional polls underestimated support for Trump…again
In 2016, mainstream public polls systematically underestimated support for President Trump, and as a result, failed to predict the outcome. In 2020, almost all public polls predicted a landslide victory for Mr. Biden. But once the votes were counted, “Americans had not delivered a blunt repudiation of Trump’s values, but had shown themselves to be intractably divided” (T he New Yorker , November 16, 2020 issue). “[We] over-estimated support for Joe Biden”, The Economist wrote. Instead of a landslide, Mr. Biden beat Mr. Trump by less than two percentage points in the states that decided the election. Public polling aggregator FiveThirtyEight and The Economist’ s U.S. elections forecasting project under-estimated support for Trump in every battleground state, and by at least five percent in Florida, Ohio, Iowa, Michigan, and Wisconsin. In Wisconsin, polls said Mr. Biden was ahead by ten percentage points, and he won by less than one, a huge polling miss.
According to The New York Times, “pollsters spent much of the last four years trying to fix the central problem of 2016 — the underestimation of the Republican vote in multiple states — and they failed.”
Clients asked RIWI to provide an alternative lens for 2020
In the months leading up to the election, several global finance firms hired RIWI to provide an alternative, evidence-based lens for their election-related investment decisions. They were skeptical about the reliability of public polls after 2016 and wanted to confirm or challenge their investment theses.
RIWI collected data across the entire country, but the main challenge and area of client interest was to identify whether there was a risk that the polls were missing something in the contested races. As the polls began to show comfortable leads for Mr. Biden in these states, RIWI was asked to look for evidence to confirm, nuance, or reject what the polls were seeing.
RIWI engaged those who don’t typically answer polls
In total, RIWI randomly engaged 100,584 Americans, half of those in the final week before the election.
To increase the likelihood of truthful responses, RIWI asked respondents to forecast the outcome in their state, in addition to their preferred candidate and voting likelihood. To further increase the chance of truthfulness and reduce the chance of any “shy” Trump or “shy” Biden voter effects, RIWI did not collect any personally identifiable information from respondents, unlike mainstream polls.
RIWI forecasters anticipated a much tighter race for weeks in advance of the election
While polls showed a consistently strong Biden lead, data from RIWI forecasters showed a tighter race than the conventional poll-of-polls data throughout the pre-election period, both overall and in many of the contested states. As RIWI wrote in its September 25th election report, “there is a broad-based perception among knowledgeable [RIWI] forecasters of an ‘undetected’ GOP vote.”
RIWI’s data identified and showed consistently tight races in Michigan, Wisconsin, Nevada, and Pennsylvania, all of which Mr. Biden won by less than two percentage points (polls had expected a much more comfortable margin). In Florida and North Carolina, RIWI forecasters found enough undetected support for Mr. Trump to correctly anticipate him winning those states (polls had expected Mr. Biden would win these states).2 Each day, RIWI checked these findings by surveying a new randomly engaged group of forecasters in those states, and the results held firm. RIWI forecasters overestimated support for Mr. Trump in Arizona and Georgia which ultimately went to Mr. Biden in exceptionally tight races (the final margin of victory in both states was 0.3 percent).
RIWI data consistently cast doubt on the conventional polling wisdom
The consensus polling wisdom was that Biden would win comfortably in the contested states critical to the election. RIWI relied on a truly random sample of Americans — including the perspectives of those who do not respond to traditional polling methods — rather than trying to sample, or over-sample, various demographic groups based on past voting patterns. Each day in the seven weeks before the election, RIWI tested the prevailing wisdom by canvassing the views of a unique, randomly engaged cohort, and each day these random cohorts cast doubt on the consensus. This approach provided a check on public polling results, and showed clients that a “Blue Wave” was not a forgone conclusion. Clients who knew this in advance were able to leverage this knowledge for increased confidence in their investment decisions.
1 Conventional polls draw on a pre-identified sample or voter database, which does not represent a truly random sample of the population. As a result, these approaches risk failing to identify new or changing coalitions of support (a key factor in the 2016 polling miss). Pollsters tried to correct for the 2016 error by overweighting non-College educated white males, but it appears that Mr. Trump may have expanded his voter turnout in new demographic groups, and the polls missed this.
2 The same was true for the North Carolina Senate race: while no public polls correctly called the North Carolina Senate race Republican, RIWI forecasters continued to point to enough undetected support for Republican Senator Thom Tillis to win. Both RIWI and the poll aggregators anticipated the results of the other Senate races, except for the race in Maine.