Friday, April 23, 2010

So where is that swing? Looking at the 1997 UK election and what it might tell us for 2010

I am very fascinated by the UK election and been looking more closely into the past results and the polls and I have come up with some interesting factoids.  I am quickly looking at the 1997 election because it is one that defied the normal assumptions in the conventional thinking of how UK elections work.

They speak of swing from Conservatives to Labour or vice versa in their elections.   As a model it worked reasonably well as long as two major parties were dominant in the results and the voter turnout stayed consistent.     This started to change in the 1997 UK election.   Here is how the total votes changed for each of the major parties from 1992 to 1997:

  • Labour + 2,000,000
  • Conservatives - 4,500,000
  • Liberal Democrats -750,000
  • Other Parties +1,250,000
  • Total vote -2,250,00
A swing normally means a voter goes from party A to party B.   This not what happened in the 1997 election.   More people chose not to vote than changed their vote to Labour in that election.   For every vote Labour did gain, almost 2/3s of a vote went to smaller party.

While it is true that Labour's share of the vote rose by 25%, in terms of rise in actual votes it was only 17%.  As it was, the 25% increase in their vote share when used as a swing did not come up with the actual result on the night of the election.

What is the much more staggering result in the '97 election is that the Conservatives lost four and half million votes.  One in three people that voted Conservative in '92 did not vote for them in '97.   Less than half of these people could have voted for Labour as the Labour vote only increased by two million votes.  What traditional UK models miss out is the impact of voters choosing to stay home.

To illustrate this I would like to look at Enfield Southgate in the 1997 election and compare the swing model to a voter migration model that works with people choosing not to vote.  This is the seat that surprised everyone when Michael Portillo was defeated on election night in 1997.

Swing model gives us:
  • Labour 34%
  • Conservatives 51%
If instead we view it through 1/3 of 1992 Conservative voters not voting Conservative, and 1/8 of 1992 Liberal Democrats not voting Liberal Democrat, the whole result begins to look closer to what happened.
  • Conservatives go from 28,422 to 19,042 in this model - the real vote was 19,137
  • Liberal Democrats go from 7080 to 6160 - the real vote was 4966
Net vote change from 1992 to 1997
  • Conservatives lost 9285
  • Liberal Democrats lost 2114
  • Combined loss in vote 11399
  • Labour only gained a total of  7711
  • Other candidates gained 1164
  • Combined gain in vote 8875
If we assume that a large amount of the LibDem vote went to Labour, say half of their lost vote, and another 1/4 went to other parties, and one quarter stayed home this means only 6670 votes could have come from the Conservatives to Labour.   This means close to one third of the the 1992 Conservative voters did not vote in 1997.  

It is these Conservatives supporters that chose not to vote in 1997 that made it possible for Michael Portillo to be defeated.   The assumptions of the model hold true for most of the seats in the 1997 election.

Background on this Model
I am working from the research of Werner Antweiler at UBC on the impact of non-voters on the results of elections.  This modelling as an addition to the sort of crude voter transition matrix that Electoral Calculus provides will help to detail potential scenarios for an election outcome in the UK.   

The UBC Election Stock Market has an example of a very good voter transition matrix that incorporates non-voters.  If you want to see the dramatic impact of including non-voters or not, try some random forecasts for the Canadian Federal Election with and without the impact of non voters.   I am sure only completely geeky political junkies will try out.

What does this say to us for the 2010 election?
First off, the polling in the UK for the election does not give enough detail for the regions to be certain of what is going on in Scotland versus Cornwall.  It is a limitation in any assumptions of what might happen in an election.  Second, polling in all countries that I am aware does not give any sort of accurate sense of what a person will do on voting day with respect to voting or not.

Most importantly 1997 tells us that we need to consider the actions of the Labour supporters that can not vote for Labour but will not vote for anyone else.   How many of them will not vote this time?  Based on the last generation of elections in the UK, I would estimate the core Labour vote to be about 11.5 million people. 2005 saw Labour go two million votes below that already.

The UK saw dramatic drop in voters between 1997 and 2001, five million people stayed home in '01 that voted in '97, close to a sixth of the electorate.   There was only a marginal gain in 2005 in total vote.   The UK has some six million people that voted regularly in the past that did not vote in the last two elections.   What will these people do?

What will bring non voters back?  A message of change, of something different, of hope of some kind.   This is what Nick Clegg seems to be to a lot of people in the UK.  

My take on the polls in the UK is that the LibDem rise is in part Labour supporters changing to them, but at least as important is people that did not vote in the last several elections.   A high voter turnout on election day will most likely indicate a strong vote for the Liberal Democrats.

With these limitations, I am going to try and lay out some scenarios of possible election outcomes on May 6th.  The first one I will be working with is one in which voter turn out increases by three million votes and splits as follows:

  • Labour  7,500,000
  • Conservatives 10,200,000
  • Liberal Democrats 9,900,000
Off the back of an envelope, the numbers are not good at all for Labour in this scenario, really not very good at all.

No comments: