Tag Archive | "sampling variation"

Pollwatch: How the general election scores could change by 6 May (6 April 2010)

Tags: , , , , ,

Pollwatch: How the general election scores could change by 6 May (6 April 2010)

Posted on 06 April 2010 by admin

The polls published yesterday may have appeared to be all over the place – but were more consistent than they looked

The polls published just before the election was called may have appeared to be all over the place, but were more consistent than they looked. Conservative support stands at around 38-39%, and Labour somewhere around 31%. The Lib Dems stand firmly on 20% of the vote.

The Saturday and Tuesday ICM polls for the Guardian appeared rather different, as Labour’s rating was a below-average 29 points in the first poll and 33 points in the second. If this is a trend, Labour has reason for confidence, but it may just be sampling variation around a constant figure of 31.

However, polls measure how things stand now. Will the campaign change this, or will the next month of hectic campaigning, saturation media coverage and frenetic expenditure just confirm opinions held already?

Fortunately for my trade, campaigns make a difference. Of the last 10 elections, the only one in which there was no change between the polls at the start and the end of the campaign was October 1974. Every other election has seen significant movement in the polls during the campaign (although in 1970 it was self-cancelling and brought the final polls back to where they started).

Some elections, particularly 1992 and 1970, have seen significant differences between the final polls and the outcome, either because of late swing or methodological faults in the polls. However, the final polls were nearly spot on in 2005.

Comparing poll averages at the start and finish of the campaign in elections gives us some clues as to how things may change between now and 6 May.

If there is a general rule, it is that the Liberal Democrats (and their predecessors) gain support during the weeks just before polling day. In some elections this has been dramatic (eight-point gains in February 1974 and 1983) but a three-four point gain, as in the last four elections, is the least that can be expected. The only major exception was the disastrous Alliance effort in 1987, when Labour’s strong campaign and the increasingly obvious differences between the “Two Davids” (Steel and Owen) led to a five-point fall. In October 1974 an over-hyped Liberal campaign, featuring a campaign hovercraft, ran aground and left the Liberals a little behind where they started.

The main explanation for Lib Dem growth during campaigns is in the broadcasters’ adoption of stricter balance rules, so that Lib Dems are included in panel discussions. This campaign will have more potential to boost the Lib Dems than most, provided Nick Clegg makes a decent showing in the leaders’ debates.

The other reason is that voters become more aware of the tactical situation in their constituencies as the campaign progresses, and in constituencies where the Lib Dems are a presence this tends to build their support. For this reason, one should discount any constituency polls early in the campaign showing the Lib Dems way down, and their third place rival looking as if they will overtake them from where they were in 2005.

The other fairly consistent trend is that the party that starts the campaign season ahead will tend to drop in the polls by election day. If the party’s recorded support is particularly high, it will drop by more than average (as for Labour in 1997 and 2001, and the Conservatives in 1979, all of whom dropped five points). However, Labour managed to defy this pattern in 2005, with a small but crucial increase and a larger Conservative decline during the campaign. There is no consistent pattern of incumbent governments doing better or worse in the final polls than the first polls of the campaign.

How does all this apply to 2010? With fewer people, year by year, expressing strong attachment to one or other of the parties, more choice of parties, and lower turnout, there is certainly the potential for a campaign to shake things up, particularly given the unknown effect of the first leaders’ debates in a British election. There must be the possibility of something dramatic happening.

However, assuming this is an “average” campaign and using my two rough rules for changes during the campaign, the final polls would end up something like Con 37, Lab 31, Lib Dem 24. Allowing for a larger pro-Tory swing in the marginals, this is probably enough for a Conservative minority government in a hung parliament. As several sessions at last week’s Political Studies Association conference in Edinburgh demonstrated, this is where most academic experts’ rules of thumb and statistical models end up pointing; but we are all aware that these rules can suddenly stop working. Even so, prepare for a bumpy ride.


Comments Off on Pollwatch: How the general election scores could change by 6 May (6 April 2010)