Thread by @MarcoGBiagi

With a lot of coverage of the latest dramatic Scottish poll results, it's time to remind everyone that predicting results in individual seats or seat totals (and overall majorities or lack thereof) is MUCH harder than predicting national shares of the vote.

This is REALLY IMPORTANT when people will be holding forth on the supposed certainty or otherwise of a single-party majority, or of a single-party whitewash in a particular region, and perhaps encouraging people to change their votes in light of this supposed certainty.

Let's assume you could go to the start of 2016 with the vote shares from the election, Biff Tannen from Back to the Future style. You are far ahead of any pollster - you know the *exact* national vote totals. What would they have told you would be the result in terms of MSPs?

Well, they'd have said it was going to be an SNP majority. Plug it in and you would have seen SNP 66, Con 30, Lab 23, Grn 5, LD 4. Close, but off in that one key detail. And remember, this is assuming you actually had the correct national numbers rather than just poll estimates.

Why? Basically, all the seat predictors out there for Holyrood have a big simplifcation. They assume all parties' shares of votes will change by equally in all constituencies. If the SNP are up 1% nationally they increase that vote by 1% everywhere and work out the new 'winner'.

As a rule of thumb it works reasonably well. But you might be looking at one of those all-yellow maps and thinking the SNP is set for a sweep. Perhaps, but here are a few 2016 reminders of the locality of some contests to shake overly confident/complacent predictions.

Nationally, the Lib Dem constituency vote share was down 0.1%. In North East Fife it was up by 15%, giving them a gain. Tories nationally were up 8% but won Aberdeenshire West up 17%. Labour was down 9% nationally but managed to be up 8% in Edinburgh Southern, taking the seat.

The method in 2016 would have got the winners of 63/73 constituencies right. Pretty good - like I said, it works reasonably well - but it was the results in the other 10 and the decline in the SNP list vote that meant the SNP majority predicted by polls didn't materialise.

Basically, if you want to look at these polls and say SNP are set for a majority, beware of three things: the inaccuracy of any poll; possibility of change before election day and, if any of this has sunk in, the difficulty of turning polling numbers into *exact* seat results.

Addendum: SNP 67 including Presiding Officer.