Expert Insights on The Future of U.S. Polling and Forecasting

One month after this year’s Presidential election, the ARF returned to the subject explored in 2017 after the 2016 Election which considered the predictive accuracy of different methods of political and business forecasting.  This year’s fascinating event featured some of the best in the business who addressed the advances in the field and lessons from the vastly different 2020 election and a world swamped in a pandemic.

Kathy Frankovic, Consultant, YouGov America, Inc. used her deep experience at CBS News to answer, “What Did Pollsters Learn From the 2020 Election Polls?”  Clearly, this election was like no other:  A pandemic; new ways of voting, notably by mail at truly significant levels in many States that experienced heavy mail voting for the first time; massive early voting; extensive fake news about polls allowed on social sites; underestimating the Trump brand effects; different and evolving swing States; etc. 

“FiveThirtyEight,” the polls averaging site that accounts for differential polling quality, had Biden at 51.8% versus 43.4% for Trump just before the election.  The totals finished up with Biden at 51.3% versus 46.9%.  Similar to 2016, the majority of Biden’s overall total edge came from California and New York.  His wins in the ultimate swing States were narrow and he lost Florida but still earned 308 Electoral College Votes.  Nearly 160 million voted and the turnout was the highest since 1908 at approximately 65%. 



Why were the polls off?  A late swing to the President?  Kathy suggested, “likely not, except for maybe Iowa, Maine and Wisconsin.”  She noted the Trump win in Iowa was forecast by J. Ann Selzer, President of the Des Moines, Iowa-based polling firm Selzer & Company against all the general predictions!  According to some polls, mail voters favored Biden by close to 70% and in hindsight should have been discounted.  How mail voters can be more accurately surveyed with all mail’s peculiarities by state (some is never counted) needs to be addressed.  Trump voters were missed and likely shy or heavily non-respondents due to the anti-polling messages on social messages.  Despite extensive polling by state, many of the state polls were significantly off especially in states comfortably carried by the President. 

“There Were Always Mixed Signals” was the theme for Cliff Young, President, Public Affairs, Ipsos which is wrestling with the “optimal multi-input approach to forecast and understand uncertainty and how to best present the results of any forecast.”  In recognizing that no one research approach/input “can be the be all and end all,” Ipsos is continuing its post-mortem recognizing that it needs to, “Beware of horse races!”  One input beyond a poll that has demonstrated merit is the approval rating on key issues for the incumbent relative to the challenger, e.g., Covid #1 issue - Biden, the economy #2 issue - Trump, etc.  This was found to be predictive for 2020 in the key swing States of Arizona, Wisconsin, Michigan, Pennsylvania, North Carolina, and Florida.  Consequently, it also showed that Trump was stronger in those states than the polls were showing.  Cliff underlined that Ipsos did get the overall results right although they predicted that the Republicans would retain the Senate.  TBD!  He later underlined the need to capture the energy, intensity, and enthusiasm of two highly polarized political groups that demonstrated mutual contempt and their effects on turnout state by state.   

David Dutwin, senior vice president, strategic initiatives, National Opinion Research Center (NORC) at the University of Chicago defended polling via some heavy questioning from Scott McDonald, President & CEO, ARF.  “We cannot function as a democracy without an independent source of opinion.  How can we fix polling?” McDonld asked.

David reflected that polling is in a continual state of enhancements, adjustments, and improvements in techniques as errors based on sampling, respondent veracity, response rates and ultimate voter turnout are inevitable.  Every election brings new and different issues.  He noted that there has been a significant increase in the sheer number of polls especially at the state level which are relatively new.  These are critical based on the US Electoral College structure.  “The challenges are always greatest when it’s close and errors identified in previous polls always offer learning opportunities,” he said.  David, noted the consequences of the dramatic reduction in survey response rates (for all surveys) and probably the number one problem, disproportionate non-response error.  Both aspects can drive significant result bias that needs to be corrected by adjusting those responding to be reflective of the population being measured.  Tricky stuff as these issues are becoming worse.  Is polling even worth it?”  A resounding, Yes!

Robin Garfield, executive vice president, research & scheduling, CNN exquisitely summarized an extraordinary election year and the variety of events, activities, data points and signals that provided clues to where it was all going.  During, “Insights into the Path to the 2020 Presidential Election” it was easy to understand their impact post November 3 and see how they help tell the final story.  As she hinted, “It’s not just about polls!”  She highlighted CNN’s coverage of: Impeachment; Key indicators favoring the President; George Floyd; Black Lives Matter – the largest social movement in US history; Covid explosion; unusually high interest in the election with donations reaching all-time highs.  And then, the October surprise arrived.  A Super Spreader event like no other – at the White House!  The celebration of Amy Coney Barrett’s confirmation by the Senate to the Supreme Court which, in and of itself, had a parallel chilling effect on American democracy. 

By then millions had already voted versus the thousands of early voters in 2016 which was to continue apace.  TV audience levels gave Biden the edge in the competing Town Halls and CNN viewing analysis indicated which way the main swing States would go.  Echoing Cliff Young of Ipsos, Robin suggested there is too much pressure on polling and reminded that businesses run on multiple sources and inputs. 

In contrast, the more general issue of how to manage stochastic, or unpredictable, shocks to forecasting models for brand marketing were addressed respectively by Rex Briggs, founder & executive chairman, Marketing Evolution, Inc., Diana Saafi, data science lead, Discovery and Mark Wilson, associate vice president, Analytic Partners. 

Rex used Marketing Evolution’s tracking and predictions on COVID-19 to demonstrate the value to business planning and the need for executional flexibility and adjustment over time as forecasts are produced and presented.  His model’s forecast suggest that COVID “recovery” could be around 40% by mid-summer which potentially would signal marketing investment boost for Q3 and Q4 2021.  In understanding his business opportunity, he quoted Bill Gibson, “The future is here it is just unevenly distributed.” 

Discovery Channel is, “Improving Viewership Projections”the bane of TV buyers, and sellers, as brand TV/video campaigns run, and their delivery pacing must be maintained.  According to Saafi, Discovery implemented an ensemble of traditional time series forecasting and machine learning models that together perform significantly better than their traditional models and benchmarks. 

Wilson suggested that global uncertainty represents a key moment in brand business strategy.  As any disruption continues to accelerate, scenario planning is becoming even more of a tactical and strategic necessity.  His presentation, “Forecasting Outcomes for Optimized Impact In A Time Of Uncertainty” stressed the importance of multiple longitudinal datasets, best practices and innovative expertise from across a wide array of segments that are needed to translate the resulting wisdom for business executives. 

Clearly polling and forecasting is here to stay and will continually be refined and improved.  Any doubts?  Perhaps ask J. Ann Selzer.


Next story loading loading..