Should the pollsters be fired?
Should the pollsters be fired?
The polls promised us a landslide victory. Instead we were left with days on the edge of our seats. After a disaster in 2016, is it time to fire the pollsters?
Only if you consider a 90% strike rate bad. It’s better than I have done in many exams...
Popular vote versus the electoral college
Coming into election day, national polls gave Biden a decent 7% lead. Investors were wrong to extrapolate a “Blue Wave” – where the Democrats take the presidency and Congress - from a large lead in the national polls. The national poll itself is almost impossible to map on the electoral college, which is crucial, as we were reminded of in 2016.
Four years ago Donald Trump lost the popular vote but won the election. In other words, polls appear to better reflect the popular vote rather than the nuanced way the electoral college works.
Polls correct in most swing states
Polls in the swing states – which, stayed true to their “battleground” name, did pretty well. Out of the 12 swing states we identified, 11 out of 12 had the correct prediction. Only Florida was wrong, pointing to a Biden lead. Even then, the predicted margin from pollsters of 1.4% was well within the 3% margin of error, so arguably should be excused.
Pollsters got the outcome of the swing states correct, but the vote margin was where they tripped up. For example, polls suggested Biden would win Florida by 1.4%, but instead he lost by 3.3%. So the margin of error for Florida – the most important swing state - was wrong by 4.7%.
Various states also had large errors (see chart below), more than creeping outside the supposed 3% margin of error. This error was not symmetric either; across the swing states, polls, on average polled Biden too favourably, by 3.4%.
If we had the benefit of hindsight and were able to eliminate the bias towards by adding 3.4% to Trump in each state, would polls have done better? Polls would have correctly predicted Florida but, in an effort to do so, would have incorrectly predicted Arizona. But with Florida worth more electoral college votes, this change would have better captured 3% of the electoral college vote.
Michigan and Wisconsin – worth another 5% of the vote - would have then also flagged up as narrow races. But most importantly, the polls would still have pointed to a convincing Biden victory.
Stepping back, it seems that investors were caught off guard by the large margin of error in Florida, which was widely seen as Biden’s ticket to a quick victory on the day of the election. Instead it was the combination of the sheer number of mail votes and the order by which states and mail votes were counted, that created the real uncertainty.
Pollsters should not be fired yet
With the correct outcome for the President and a 90% hit rate in the swing states it would be unfair dismissal to fire the pollsters. But they do need to do some more work on eliminating bias, which is beginning to look more systemic than at first glance. We will put them on performance review and check-in in another four years.
This article is issued by Cazenove Capital which is part of the Schroder Group and a trading name of Schroder & Co. Limited, 1 London Wall Place, London EC2Y 5AU. Authorised by the Prudential Regulation Authority and regulated by the Financial Conduct Authority and the Prudential Regulation Authority. Nothing in this document should be deemed to constitute the provision of financial, investment or other professional advice in any way. Past performance is not a guide to future performance. The value of an investment and the income from it may go down as well as up and investors may not get back the amount originally invested. This document may include forward-looking statements that are based upon our current opinions, expectations and projections. We undertake no obligation to update or revise any forward-looking statements. Actual results could differ materially from those anticipated in the forward-looking statements. All data contained within this document is sourced from Cazenove Capital unless otherwise stated.