There was an interesting meeting of psephologists, the folks who study elections, last weekend at the University of Manchester on “Methods and Models for Election Forecasting; the UK General Election of 2010”. It was organised by Rachel Gibson and Ed Fieldhouse of DCERN, the Democracy, Citizens and Elections Research Network, Institute for Social Change and sponsored by the Politics Department at Manchester, EPOP, and Taylor & Francis Publishers.

Some of the usual ‘Prognosticators’ Panel’, ‘Forecasters’ Forum’, and a few of the ‘Sunspots Correlationists’ were present, e.g., Lebo and Norpoth (I usually lay down a bet with Helmut, with whom I’ve had a great relationship since his book on the Falklands 25 years ago), Whiteley  and Clarke of the BES/BPIX group (Harold’s presentations are the highlight of any seminar he attends), all were participative and attentive, most generous in sharing their past failures as well as pointing up their successes (unlike some of the polling fraternity on such occasions).

I was only able to be present for the first panel, as I’d been asked to be the discussant on papers by Michael Lewis-Beck (Iowa) and Mary Stegmaier (Virginia) on ‘Voters as Forecasters: The UK Case’, Clarke & Stewart (Texas), and Sanders & Whitely (Essex) “Valence Politics Scenarios for Electoral Choice in 2010’, and the third paper from Jocelyn Evans (Salford) and Gilles Ivaldi (University of Nice).  I have many of the other papers, as well as two I mentioned to the seminar’s participants and can share them with colleagues, but most are now available at http://www.dcern.org.uk/news/ and more will be posted over the next few days.

To begin with my comments including referencing two other papers in the literature which, being academics, they would likely have overlooked, one in Public Opinion Quarterly (vol 73, No. 5) by David Rothschild (Wharton, UPenn) on “Forecasting Elections comparing predicting markets, polls and their biases”, the other in the International Journal of Market Research (vol 51, issue 6) by Parackal, Harris and Rudd (Otago) on “Election forecasting: Development of the constant sum Scale to be used in telephone surveys”.

Lewis-Beck & Stegmaier described their theme as ‘citizen forecasting’ based on polling data (only looking at academic data which was all they could find) which failed to draw the conclusion that for the most part, as pointed out by the redoubtable David Denver, the public mostly decide who is going to win based on people who either read the opinion polls themselves or listen to the media and other people who do, in addition to those few (for the most part) who say that their party will win, no matter how hopeless the prospect.

In 1983 we found even after the Falklands’ victory that 11% a month before the 1983 election still thought Labour would win, just before the Thatcher landslide against a badly led Labour Party with unpopular policies and a hopeless campaign.  The week before that election it wasn’t a case of Labour winning, it was whether they going to run third, after the Liberal/SDP coalition, and on the day only were two points ahead.  And in 1997, one person in four thought the Tories were going to win against New Labour, or that there would be no overall majority for any party, when Tony Blair had his 179 seat majority. By 2001 the Tory loyalists, more realistically after four years of Labour, were only five percent of the people, one in twenty. (See the Ipsos MORI website for details).

In 1992, which resulted in a surprise win for the Tories to most people, it wasn't a surprise for the 28% of the electorate who expected a Conservative victory.  Our poll in mid March had it as a close race, but with another 18% saying that it would be a hung parliament with no overall majority, but with John Major returned to 10 Downing Street, making 46% who got what they expected. 

In May it was a very close race in seats, as the Tories’ 21 seat majority would have been wiped out if one person in 200, one half of one percent of the voters across the country who voted Tory if they had voted for the second party in their constituency, it would have been a hung Parliament.

In that election, a week before election day, 89% of the public said they had ‘seen or heard’ any national opinion polls in the last week giving the position of the parties, and if yes, 90% remembered Labour in the lead.

Yet one member of the seminar argued that no one except politicians, pundits and psephologists read the opinion polls. He was shouted down (that was when Denver exploded, by the way); the sceptic later recanted.

And it was about that point in the debate I reminded the prognosticators that election after election around one person in ten among those who voted said that they’d finally decided for whom to vote in the last 24 hours of the election. 

To conclude my remarks, I introduced the seminar participants to yet another model which my colleague Dr. Roger Mortimore introduced to the science (?) of forecasting elections:

An Almost Infallible Forecasting Model
(Sweet FA Prediction model)

Roger Mortimore

Professor Robert Mackenzie was once quoted as saying that he enjoyed election nights in the same way other people enjoyed the Cup Final. Perhaps there is more to this remark than meets the eye!

Allow me to present a prediction model for determining the outcomes of British general elections, which over the period since 1950 has a record to match Bob Mackenzie’s swingometer. (See Table.) All you have to do to predict which of the major parties will have an overall majority in the Commons following the election is to note the shirt colours usually worn by the current holders (on election day) of the FA Cup.

Table: The Political Football

If their shirts are predominantly in the Conservative colours of blue or white, a Conservative victory will ensue; on the other hand if the predominant colour is red or yellow, Labour will be successful. (Black stripes are ignored.)

This, which I have christened the Sweet FA Prediction model, has failed only twice over the last sixteen elections; furthermore, the sensitivity of the prediction method is demonstrated by the election of February 1974, which produced the only post-election hung Parliament since the War – that election was fought when the cup holders were Sunderland, whose striped shirts are red and white in equal measure. The obvious improbability of such a pattern arising by chance gives the model a high degree of statistical significance.

There is persuasive evidence that Tony Blair took the Sweet FA model seriously. The 2000 FA Cup winners were Chelsea, who play in blue, and if as was generally expected the 2001 election had been held on 3 May (before the 2001 Cup Final), the model predicts that the Conservatives would have won. Plainly, all the fuss about Foot and Mouth Disease was just spin, giving Mr Blair an excuse to postpone the election until another cup final had been held. Since the 2001 FA Cup Final was won by Liverpool (who are nicknamed the Reds for the obvious reason), Labour duly secured their majority on 7 June.

And what about the 2010 election? If it is held on the date that most pundits expect, 6 May, then the 2010 Cup Final will still be in the future and the cup holders will still be the 2009 winners – Chelsea, once more. So that would indicate a Conservative overall majority. But if Gordon Brown were to ignore the convenience of holding the general and local elections on the same day and hang on until after this year's Cup Final on 15 May, perhaps he still has a chance. True, the odds are still stacked against him, as three of the four semi-finalists play in white or blue; but Aston Villa, whose shirts are predominantly a shade of red (claret, to be precise), still offer him some hope.

Or perhaps not. The point of this jeu d’esprit is to demonstrate that it is possible to find an apparently statistically significant pattern in almost anything, given a sufficiently free hand. (Rather as certain scholars discovered “hidden messages” to prove that Francis Bacon wrote the plays of Shakespeare; Mgr Ronald Knox, when he set his mind to it, was able to use the same methods to “prove” that Queen Victoria wrote Tennyson’s In Memoriam.)

It is always possible to construct a pattern which fits the past. But unless it explains the past, in a way which still applies in the present, it will not help predict the future. The initial test of any model must be its inherent plausibility as a causal explanation, and this is a test that relies on judgment, not mathematics; if this is forgotten, “statistically significant” becomes a meaningless, perhaps dangerously misleading, term. Nor is “track record”, as such, anything more than a perceptional delusion.
 
One thing's for sure, though, all this talk of a hung Parliament is misplaced. The last chance of that disappeared when Stoke City (red and white stripes) were knocked out in the quarter finals!