Will polling inaccuaracy and tactical voting cause an upset?

One of our regular contributors, Andy Cooke, produced the following well-worked out analysis of the coming election and posted it in our comments section. We think it should be given greater prominence.

The volatility of this upcoming election is largely down to two major factors:

Opinion Poll accuracy (or lack) – will they be closer than in 01?

Tactical Voting – will it unwind, and if so, how far?

On the first, the widespread belief is that the opinion polls err in Labour’s favour. The “failure of the polls” in 92 did lead to major investigations by the polling companies and changes of strategy by many. Nevertheless, in 97 and 01, they still overestimated the Labour lead. Have they got it right this time?

In 97, the polling booths produced 103 Tory voters, 94 Labour voters and 112 Lib Dems for every 100 of each that the pollsters found (from the Martin Baxter historical poll of polls at the time)

In 01, there were 108 Tories, 91 Labour and 107 Lib Dems for every 100 of each that the poll of polls predicted. (In 92, the figures were 116 Tories, 90 Labour and only 92 Lib Dems per hundred of each predicted).

On Baxter’s current poll-of-polls, we have Con 32.07, Lab 36.97, LD 21.04. If that is as accurate as in 97, it would reflect a “polling booth” value of Con 33.03, Lab 34.75, LD 23.56. If it is as accurate as the 01 polls, it would be a “true” value of Con 34.64, Lab 33.64, LD 22.51.

The latter figures give a wafer-thin Labour majority of 8 (with no TV unwind): Con 231, Lab 327, LD 57. So we have (without Tactical Voting addressed) a range of Labour majority of 8 to 116 (the “8″ assuming accuracy equivalent to the 2001 polls, “116″ assuming the poll-of-polls to be spot on).

How about Tactical Voting?
The accumulated Tactical Voting squeeze, from the 92, 97 and 01 elections, equates to a total of +3.6 Lab to Lib and +10.2 Lib to Lab (that is, to get the current status quo on the 2001 vote shares from a hypothetical baseline of no tactical voting, you’d need to input +3.6 and +10.2 into the relevant boxes).

This means that if anti-Tory Tactical Voting were to disappear completely (which is unfeasibly unlikely – voter inertia should lead some to continue), the values for these boxes would be -3.6 and -10.2 respectively. Putting these in should give the absolute limits of Tactical Voting unwind.

For the “97-class accuracy” projected vote shares of Con 33.03, Lab 34.75, LD 23.56, it gives Con 285, Lab 277, LD 54 [Highly non-intuitive – the Con seat share is higher than Labour on less of the vote! Then again, with vote shares between the top two parties this low, it all depends on which seats are lost where]

For the “01-class accuracy” projected vote shares of Con 34.64, Lab 33.64, LD 22.51, it gives Con 315, Lab 248, LD 52. Please note that a full tactical Vote unwind in one fell swoop such as this is really unlikely – I’m trying to show the limits of the possible.

So, on the poll-of-polls as of 9th January, we can have Con at 180-315 seats, Lab at 248-381 seats an LDs at 52-58 seats. This is a range from Labour Majority of 116, to Conservatives minority of 9 seats. Now that’s volatile. Also note that the polls could be inaccurate the other way – the pollsters might have overcorrected. In which case, that same poll-of-polls could reflect a Labour majority of 130+

As a PS, I put in “Mean” values – projecting vote share using derived accuracy midway between the 97-class and 01-class accuracy polls, with TV unwind at 50% (-1.8 and -5.1). This gives a hung Parliament: Con 262, Lab 295 and LD 58.

My main observation is that Andy is basing his polling accuracy figures on Martin Baxter’s “poll of polls” from previus elections. Since then we have seen the development of YouGov which, with its large sample, represents about a quarter of Martin’s figures and probably counter-balances some of the pro-Labour bias that Andy cites.

Mike Smithson