On Private polling and incumbency effects

Earlier this week some eyebrows were raised over an article by Dan Hodges which quotes a "Tory analyst" saying he re-ran Lord Ashcroft's poll of ultra-marginal Conservative seats and claiming "We reran it in the seats we hold [...] but included the name of the sitting MP. We were ahead by 2 per cent" as evidence of the Conservative party's belief in the power of incumbency.

Now, I've written about political parties' private polls and the strange mystique they seem to hold over the commentariat here before. Essentially, if you see media claims that party's private polling shows some wonderful picture that contrasts with the public polling it's best ignored unless they actually cough up some numbers and tables so you can see what they are up to. Your response should always be "give us the tables or they don't exist". Part of the British Polling Council's disclosure rules were intended to stamp out this sort of thing: if a poll is published, the company that did it must release the figures. It does, however, only apply to companies that are members of the British Polling Council - if parties use companies like CrosbyTextor or GreenbergQuinlanRosner then even if their numbers are leaked, the companies wouldn't be obliged to release numbers. Lord Ashcroft here writes about it being a bad sign if the Tories are resorting to comfort polling, but I think that's rather off the mark. Back in the day the then Tory Chairman Lord Saatchi himself used to come out with dubious morale boosting figures, this is a reference from an unnamed "Tory analyst" - I suspect, like Andrew Cooper suggests here that Dan's source was exaggerating or doesn't know what they are talking about.

But putting the mystery "private polling" aside, what about the substantive issue, what about the incumbency effect? This is supposedly the bonus that sitting MPs get - the benefit they'll have from people who aren't voting for the party they represent, but because of name recognition, or because people think the MP is hard working or decent, or helped the voter with a problem they contacted them with, what is sometimes called the "personal vote".

Let's start with Ashcroft's August poll. In contrast to the implications of what Dan's source said, Ashcroft's poll already included an attempt to get at the incumbency factor. Rather than just asking straight voting intention, it double-asked the question, first using the normal poll wording, and then asking people to think specifically about their own constituency and the candidates standing there. The intention of this - something I first came up with for a PoliticsHome poll of marginal seats back in 2008 - is to try and get at tactical considerations and personal votes. The idea is that some people may not give their *actual* voting intention in the main question, they'll give the party they support nationally, even if at a local level they might vote tactically or might vote for the local MP they like.

This approach appears to work - or at least, some people do give different answers, especially in seats where the Liberal Democrats are players. In Con-LD seats Ashcroft found standard voting intentions of CON 33%, LAB 24%, LDEM 18%, UKIP 14%, but when he prompted to get people to think about their own constituency it became CON 32%, LAB 18%, LDEM 29%, UKIP 12%. That the

difference is concentrated in Lib Dem seats does imply that it is picking up people's tactical voting considerations, and perhaps personal vote considerations (Ashcroft wasn't polling Lib Dem held seats, so he obviously wasn't, but I've done the same in Lib Dem held seats in the past and the two-question approach makes a huge difference). In Ashcroft's August poll in Con-Lab marginals though the two stage question did not produce any drop in Labour's lead in the second, more "locally prompted" question. No sign of an incumbency vote there.

By definition, any incumbency bonus should be the most obvious in seats with new incumbents. In seats where the MP was already there at the previous election any personal vote should already be taken into account, in a seat with a new incumbent any personal vote should be above and beyond what the party got last time, so if there was an incumbency effect we would expect to see a party doing better than average in seats with new incumbents like the ones Ashcroft polled. Again, this was not the case - in Ashcroft's August poll the swing to Labour was about 2 points higher in the marginals than the GB polls at the time were showing.

That bit of evidence seems clear then. The problem is, the incumbency effect is not a vague theory, the evidence for it happening at past elections is pretty damn solid. In 2010 Labour lost an additional 2.2 percentage points in seats where an incumbent MP stood down compared to seats where the incumbent contested the seat. The Conservatives vote rose by 2.9 percentage points in seats where an incumbent stood down and they lost the incumbency effect. In seats with an existing incumbent it rose by 3.8%, in seats with a new incumbent first elected in 2005 the Conservatives increased their vote by 5.6%. The same happened in 2005, the Conservatives did 1% worse where an MP retired, 2.5% better where there was a new incumbent.

Going further back the record is more patchy (while the Conservatives did worse where MPs stood down in 1997 and 2001, they didn't do better in seats with new incumbents), but there is a strong and consistent effect in elections following on from an election where a party gained many seats. It is even clearer in the seats that a party gained at the previous election (the so-called double incumbency bonus), so in 2010 Conservative MPs who gained seats from opposing parties in 2005 did 1.9% better than average, in 2001 Labour MPs who gained their seats in the 1997 landslide did 2% better, in 1997 Labour MPs who gained a seat in 1993 got a bonus of 4.3%, in 1992 Labour candidates who gained seats in 1987 got a bonus of 2.7%, in 1987 Conservatives who gained seats in 1983 enjoyed an extra boost of 3.9%. For Liberal Democrats, incidentally, the figures are even bigger. (*)

There is no reason to think the same will not happen next time. New incumbent MPs will do better than average, seats where an MP retires will see the party of the retiring MP do worse. It will affect all parties, but because most of the battleground seats between Labour and the Conservatives are ones with new incumbent Conservative MPs who gained the seat in 2010, on this occassion it will favour the Conservatives.

If that is the case though, why didn't the Ashcroft polling pick it up?

Well, there are a couple of potential reasons. One is obviously that the incumbent effect has vanished for some reason. Personally I think this unlikely - the general trend has been for MPs to devote ever more of their time to constituency matters and their "social worker" type role - but it is certainly possible. Alternatively it could be that it's there, but that the polls don't pick it up. Perhaps such things don't really emerge until the election campaign itself, or perhaps the big differences the "two-stage question" approach produces in Lib Dem seats are just tactical considerations and there isn't really a good way we can get at incumbency effects. Alternatively, you need to remember that the August Ashcroft poll is just one poll - the previous Ashcroft poll of marginal seats in January 2013 (the bigger one that did lots of different groups of marginal seats) did actually show a very small pro-Conservative incumbent effect from the two-stage question in Con-Lab seats, and did show the Conservatives doing slightly better in the marginals they were defending than in the country as a whole, the patten we would expect to find given the past history of incumbency effects.

And that brings us back to the original question and the orginal polling that whoever was talking to Dan thought they had seen. If there is an incumbency effect, then on past experience it will be a couple of percentage points. If asking a question a different way changes a 14 point deficit into a 2 point lead, then that's not an incumbency effect, that's a shonky question.(**)

I expect there to be an incumbency effect, but it's going to make a couple of points of difference, not 16 points. It's one of the reasons I think the Conservatives could probably get a majority with a 7 or 8 point lead rather than 11 point lead they need on paper, why the Labour party need a slightly bigger lead than what a uniform swing suggests. It'll be a small effect at the margins though, not something that transforms the election. It makes the election a bit harder for Labour to win than uniform swing suggests, but not much. It makes seats a little easier for the Conservatives to defend, but they still have a mountain to climb.

(*) The 2010 figures are from Curtice, Ford and Fisher in Cowley & Kavanagh's British General Election 2010, the figures for 1983 to 2005 are from a research note by Tim Hallam Smith here, alas behind an expensive academic paywall.

(**) At least, it is with a Tory 14 point deficit. As we've seen from the two-stage questions above, in Lib Dem seats it's a bit of a different story, but that will be tactical considerations as well as incumbency