What does the US election tell us about the future of polling?

With Joe Biden now the confirmed winner of the US presidential election, and a recount in Democrat-leaning Georgia underway, focus has returned to the polling industry.

Prior to election night, the overriding expectation was for a ‘blue wave’ that would see Biden sweep past Donald Trump with a comprehensive victory. However, the results were much closer, with Trump’s support overperforming expectations. An anticipated Democratic clean sweep of the main legislative and executive branches of government did not materialise.

The results follow several high-profile polling misses in several countries. Trump’s 2016 victory, the UK elections of 2015 and 2017, the EU referendum of 2016 and last year’s Australian elections all saw the polling industry criticised due to unexpected results. How should the polling industry react to the latest US presidential results?

Deborah Mattinson, founding partner, BritainThinks
Elections in the US are big business – meaning that there are many agencies that only do political work, unlike in the UK where political insight can be a useful ‘shop window’ and an opportunity to showcase the work that we do, but is unlikely to sustain an agency of any size. Given this, it is interesting that the US polling industry did not manage to call this election, or the one before, more accurately – I am sure that there will be many post-mortems and something of a reset.

I think some of the answer will lie in the disconnect between Capitol Hill and the voters in the so-called ‘rust belt’ states, just as the fall of the ‘red wall’ here defied much conventional wisdom. I think insight companies can tackle this by a) recognising the problem b) reviewing their techniques, gaining deeper insight through more qualitative and ethnographic techniques, and c) recruiting a more diverse workforce that reflects the whole country better.

Jean-Marc Leger, founding president, Leger
Pollsters were pretty good in evaluating Biden’s support but underestimated Trump’s voters by 3%. We also observed the same differences in the 15 key states, where pollsters precisely measured Biden’s votes but underestimated Trump’s support by 2.5%. Overall, pollsters, as an average, were in the margin of error in nine states out of the 15. They were outside the margin of error in Texas ( 4%), Virginia ( 4%), Pennsylvania ( 4%), Wisconsin ( 5%), Iowa ( 5%) and Ohio ( 6%).

We must take responsibility for some failure. We are very good at measuring yesterday’s behaviour, not so bad in measuring today’s attitudes, but not the best at predicting the future. This is the area where pollsters should work. We still have issues that we must address: we are publishing straight numbers without the margin of error; we have difficulties in identifying who are the likely voters; we have issues predicting the voting turnout; and we have a hard time reaching less well-educated people, as well as those who don't speak English, such as Latino and Asian voters.

The industry should better predict the outcome of the election by integrating other sources of data including social media, past behaviours, election issues, voting turnout, state results, money raised, advertising budget, media trends and economic indicators.

Chris Curtis, senior research manager, Opinium
The first thing to bear in mind is that the research shows polling is not getting any more or less accurate over time. There are some examples, and the US election is a good example, of where the polling is a little less accurate than it should be. 

Polling is an imperfect science. Despite our best efforts there will always be a margin of error and this should always be considered when people read about polling results. Secondly, polling is only ever a snapshot of where public opinion currently stands and this might change in the future.

It is important for the industry to do what we can to clearly communicate the results from our research. This means ensuring we correctly present all the correct caveats surrounding the margin of error and potential for the polling to move. It is also important that we continue to engage and work with journalists to ensure that polling is correctly reported.

While polling will never be perfect, it is only by constantly questioning and updating our research practices that we can ensure our research is as robust and representative as possible.

Ben Page, chief executive officer, Ipsos Mori
Professor Will Jennings’ analysis shows that polls globally have tended to average about two percentage point error for each party every election since 1942. This is despite the world and technology becoming far more complex over the last 70 years.

The public think that polling is mostly about politics and the ‘horse race’. We should do more to explain that the vast majority of polling is not about that, and is useful even if we aren’t always aiming for pinpoint accuracy. We need to remind people that sampling tolerances exist even for perfect polls, so we are less surprised when the margin of error turns out to be larger than we expected. We ourselves should often say more loudly “too close to call”.

So far there is no magic bullet for being right in election polling – what works in some countries doesn’t work in others. We will continue to develop tools like passive measurement, social media analysis and more – my Chinese colleagues said they could predict it would be a tight race in the US because sales of Trump merchandise were holding up. Underpinning it all needs to be getting the basics right: making sure you have representative samples of voters; and experimenting with different methods of accounting for differential turnout.