Two issues to round up. How did the Monster’s Poll do, and how did all the other pollsters do?
For the answer to the second question, I defer to the human calculator that is Nate Silver of FiveThirtyEight.com.
First thing to say is I’m delighted that MORI did well, as they’ve been something of a whipping boy over the past few years. ICM too retains their crown as being seen as the “gold standard”. And credit to the newbie RNB Research, an Indian based polling firm. It seems they all deserve high praise for getting the Tories and Labour spot on.
The Lib Dem figure, however, is cause for embarrassment for some firms. How did they get them so wildly wrong, appears to be the universally asked question.
But let’s not be too harsh. Polls are +/- 3%. Those who had them on 26% could say they were within the margin of error. Even 27% is not hugely wrong. But 29%? 32?!
What seems to have happened is certainly that the LD waverers did not go to the Conservatives. The table above suggests that, in fact, around 1.5% of the total vote moved from the LD to Labour, and the rest either didn’t vote at all or went to another party (as “others” seems to have been underestimated). Perhaps that is a reflection of the fact that most Lib Dem support seemed to be coming from sections of the electorate (i.e. the youth) that are totally unreliable.
But the chunk that moved to Labour… my gut feeling is that it was LD sympathisers that nevertheless chose to rally back behind Labour in order to stop a Cameron advance. Sometimes this occurred regardless of the actual situation on the ground. That seems to account for the LDs bizarre inability to take the low-hanging fruits off Labour in Edinburgh South, Islington South, Watford, Aberdeen South, Hampstead & Kilburn and Edinburgh North.
Then there was the movement to the others. If we use RNB as our new gold standard, there was a 0.8% increase in Others. Now it seems that barely any of that went to the SNP or Plaid. Perhaps the more likely explanation is that from the general rise of turnout some of it went to the other others… UKIP and the BNP both had decent elections when examined on a GB basis. It would seem odd that LD support would go in this direction though.
Either way, the pollsters all have methodological revisions to do. The era of Labour overstatement appears to be over – and I now feel vindicated in dropping the Mike Smithson Golden Rule from my own polling prediction. Angus Reid on 24% (ignore the nonsense OnePoll) was clearly very wrong. I suspect they have a lot to learn about how to poll British elections.
Likewise, the era of Lib Dem understatement is also in jeopardy.
As for the final Monster’s Poll prediction, it doesn’t seem too bad. You know what they say, garbage in, garbage out.
The Tories over-performed their median figure by nearly 2%. Labour also over-performed by the same amount. The LDs underperformed their median by just over 4%. If I’d stuck to the Labour minimum and LD maximum, however, I would have been way, way out.
With that in mind, what might the prediction have been if I’d used Con MAX, Lab MAX and LD max over the final 10 days of polling…
LIB DEMS: 24%
The answer is that it would have given us Labour and LD almost spot on, and the Tories slightly over-stated. The total of them is 92%, which is too high. Approximately 2% needs to be removed from them and given to others. How about deducting that straight from the Tories and giving it to Others? Doing that would give an estimate that is pretty much exactly right.
Food for thought for future predictions. But would it still apply in future elections? Surely the conditions for this one were unique?
Probably. But hey, it was fun analysing it!