Will Donald Trump, now that he is effectively the GOP candidate, win the General Election this November? Well, I’m neither a prophet, nor the son of a prophet. And this blog is not a political blog, far less a prophetic one. But, this blog is a blog about thinking. And especially about critical thinking. And one of the key skills of critical thinking is being able to use statistics to reason well. So with an election in the US in November, and one here in Australia in July, let’s have a chat about opinion polling and what to watch out for.
The closer to the election, the more predictive opinion polling is
One of the first principles of opinion polling is that the closer you are to an election, the more predictive opinion polling becomes. Nate Silver, in his great book The Signal and the Noise has a table showing the probability of an American Senate candidate winning her election given the size of her opinion polling lead at particular points in time. For example:
- A 5 percentage point lead in the polls 1 day before the election gives an 95% chance of winning.
- But, a 5 percentage point lead in the polls 1 week before the election gives an 89% chance of winning.
- And, a 5 percentage point lead in the polls 6 months before the election gives only a 66% chance of winning.
The full table is given in Nate Silver’s book and makes fascinating reading. But, it’s worth noting, this is based on past data: US Senate races in the ten years to 2008. And, more importantly, this is data based on races that are simple seat races, in single states. The US presidential election, of course, is a race across the whole country, and its complexity is compounded by the electoral college system.
The latest polling I’ve seen is an Ipsos Poll showing Clinton almost 10% ahead of Trump. Six months out from November, if I were to apply Nate Silver’s table naively (the emphasis here is naively!) to this result, that would give Clinton a 79% chance of winning.
Of course, you have to wonder how normal this US election will be. PJ O’Rourke has just come out and said he is voting for Clinton. Yes, you read that correctly. The noted conservative is supporting the nemesis of the Right. Why? Because, in his words, “… she’s wrong about everything, but she’s wrong within normal parameters.”
One poll in twenty will be rogue
The other critical point to remember about opinion polling is that 5% of polls will be plain wrong. Not just margin of error wrong. But actually outside of the margin of error.
Every poll has a margin of error. But the thing to remember about this is that it’s a confidence interval – not a hard and fast bound. The best way to explain this is with an example:
- Let’s say an opinion poll gives a candidate a 52% result, with a margin of error of 4%.
- That means the poll estimates the real result for the candidate will be between 48% and 56%, 95% of the time.
- In other words, 5% of the time the actual result will be outside of the 8% range given by the margin of error.
Yes, most margins of error you see in the papers are 95% confidence intervals. So, even taking the margin of error into account will mean that sometimes (1 time in 20) you will still be wrong in your estimate.
How does this affect your thinking in the real world? Well firstly, back in the day when we would get a Newspoll published every fortnight, 26 times a year, that means that at least one Newspoll a year will be entirely wrong. That is, it will give a result that is not even within the margin of error of the true result. And of course for the next two weeks the Australian political debate will be based on completely wrong assumptions.
Of course, in an election campaign I assume we will be getting plenty of polls. In the next eight weeks, you can reliably predict that at least one of the them will be completely wrong.
So what to do? Look at an average of polls. And treat each new poll as just one more piece of information. And don’t think that one new poll, departing from the last one, means anything at all. You will get a lot of volatility within the margin of error that can be mistaken for a trend (remember, we are pattern making creatures), and this is compounded by the 1 poll in 20 being completely wrong principle.
If you want to be updated on Australian and US election polls as they break, follow @GhostWhoVotes on twitter. This person tweets polls as they break. A great public service!
And, if you want to see good predictions on the US election, you can’t go past Nate Silver’s FiveThirtyEight site. Remember, Nate Silver predicted how all 50 states would break in the 2012 Presidential election, even as Fox News was still calling it for Romney.
As always, if you want to sign up for my weekly newsletter on critical and creative thinking, just drop me a line at email@example.com.