The front page of The Times last Thursday (1 June) led with ‘Pollsters predict shock Tory crash’, sending shockwaves through Westminster and the financial markets.
The forecast in question is described by YouGov as a ‘new model’ that guesses what a large poll would show based on ’Mulitilevel Regression and Post-stratification’ analysis which is jargon for profiling – taking what one group of potential voters say and applying it to others.
So in short it is educated guess work, but still guess work.
Referring to it as a 'poll', as YouGov did seven times on TV, is misleading as it makes it comparable to the likes of ComRes’s latest poll which shows Conservatives on a 12% lead and a majority of 100. Kantar yesterday announced a poll with an extended lead of 10% and ICM today announced a 12% lead. None of these got much media coverage as they did not have the shock value of a ‘hung parliament’.
YouGov’s new technique projects the outcome in individual seats based on a sample size of just c.75 per seat which is not statistically robust. YouGov claim they applied it in the EU Referendum and US election but in the former they publically predicted on the day of the vote, Remain would win by 4%. In their final US polls that Clinton would win by 4% and “Trump would come up short in key battleground states”.
As a result of yesterday’s ‘rogue’ forecast the pound dropped sharply. We have seen this before when YouGov forecast that ‘Remain’ would comfortably win on the day of the Referendum which wrong-footed the media and financial markets. YouGov’s last poll for the 2015 general election was out by a massive 6%. These were not ‘snapshots’ of changing opinion during the campaign but at the very end of them.
YouGov have form in running outlier rogue polls which have had significant consequences. Their forecast that a Yes vote was 2% ahead in the Scottish referendum, resulted in our politicians making expensive commitments that were needless as the Scots were never going to leave. My own prediction was that “the Yes Vote would not get more than 45%”. The result was 44.7%.
I am by no means a sage with all the answers but an individual without “hundreds and thousands of pounds” to spend on polling. I do not always get it right, and like many I got Trump wrong, but of the seven elections I have predicted I was right in six with a high level of accuracy (e.g. Vote Leave would win by 52%” - the result was 51.9%).
All I do is apply consumer behavioural insight to publically available polling data. The difference is that YouGov charge millions of pounds for their forecasts and I provide mine for free.
You would have more success tossing a coin to guess the results of the above elections than relying on YouGov’s predictions. Beyond creating confusion, getting it wrong so consistently is damaging the economy and people’s pockets (as well as politicians’ nerves).
My own prediction is a Conservative majority of 103-108 seats but whether I am wrong or right is in the hands of the electorate, which is where it should be.
Michael Moszynski is CEO and founder of London Advertising.