Blog

Why Survation does not adjust our constituency telephone polls based on how people voted in 2010.

Survation does not adjust our constituency telephone polls based on how people voted in 2010. Methodological decisions such as this can divide pollsters, so here our Director of Research, Patrick Briône, explains Survation’s approach.
 

  • The demographic makeup of the average constituency has changed significantly in the last 5 years, with an average of 20% moving away and a further 6% passing away per constituency.

 

  • In what’s known as ‘false recall’, people often don’t remember or don’t want to tell polling companies which party they voted for at the last election.

 

Weighting (1)

 

Why Do Polling Companies Use Past Vote Weighting?

Polling companies try to ensure their samples are politically representative. Adjusting samples to how the public in a seat voted in 2010 seeks not to over-represent certain political views – a balanced sample. Having an unrepresentative sample could potentially skew results, decreasing the accuracy of the poll’s findings.

However, Survation believe there are a number of very good reasons, often stark at the constituency level, why weighting the sample’s by the voter’s recall of their 2010 vote to the actual results of the 2010 general election can make a seat poll less accurate rather than more accurate.

 

Internal Migration

The Office for National Statistics suggests that each year approximately 4% of the resident population of any given constituency move at least 10 km from their former residence. This means that over the five years since the last election, approximately 20% of the residents in a constituency have moved elsewhere.

We have no reliable way of knowing how new residents compare to former residents in terms of their voting behaviour in 2010.

 

Deceased Voters

Since May 2010, an average of 4,000 people have passed away per UK constituency. As an example, in South Thanet, a constituency with a lower-than-national average life expectancy and a more aged population, it is possible that more than 5,000 of 2010’s voters are deceased.

ONS statistics suggest at least 6% of voters in a constituency such as South Thanet have passed away since 2010 – about half of whom voted Conservative at the last election. Around 3 percentage points of the Conservative’s 2010 vote are now likely deceased.

Although UKIP voters are also disproportionately older, there were far fewer of them in 2010 as the UKIP surge is only a recent phenomenon. If around 6 percent of both Conservative and UKIP 2010 voters in South Thanet passed away, that is 3 percentage points off the Conservative vote share, but only around 0.3 percentage points off the UKIP vote share.

 

How does this affect 2010 vote weighting?

A quarter of the 2010 population therefore might no longer live in the same constituency and have been replaced by a different 25% consisting of newly eligible young adults and those moving into the area.

Weighting most new voters in a constituency to how previous residents voted five years ago does not therefore seem sensible to us.

 

“False Recall”

People often don’t remember or don’t tell polling companies accurately which party they voted for at the last general election five years ago – what’s known as “false recall”.

There is evidence from the independence referendum in Scotland and polling across Great Britain to suggest that recall of 2010 general election vote is not wholly reliable.

In the run-up to the Scottish Independence Referendum, 2010 vote in Scotland was found to be so poorly recalled that it was abandoned as a weighting target by most British Polling Council (BPC) pollsters.

Survation and Ipsos Mori had the most accurate final polls ahead of the referendum in Scotland, despite not using 2010 vote weighting in Scotland. We used 2011 vote weighting instead, which had very little impact on the results  suggesting that recall was relatively accurate in this case.

One reason 2010 recall in Scotland is poor 5 years on is that people there remember their vote in the 2011 Scottish Parliament election as their “vote at the last election” rather than their 2010 general election vote.

Across Great Britain, there have been at least two elections in each constituency since 2010. Constituents of South Thanet have had the 2013 County Council elections and the 2014 European Parliament elections. Some of those who voted UKIP these more recent elections (both instances where the party performed particularly well) may have forgotten that they voted for a different party in 2010.

Furthermore, in our South Thanet poll (February 2015), 8% of respondents told us they could not remember who they voted for in 2010, whilst 7% refused to say. There is no reliable way for us to know how proportionate these voters are to voters as a whole.

False recall of 2010 vote is particularly strong for parties which have gained or lost considerable support in the intervening years. This means that recall of voting UKIP and SNP in 2010 is consistently higher than the actual outcome. Conversely, Liberal Democrat recall is consistently lower than the actual outcome –  a significant proportion of those who voted Liberal Democrat in 2010 no longer remember – or want to tell polling companies – that they did.

 

In Clacton by-election polling by both Survation and Lord Ashcroft, 8% of respondents said they had voted UKIP in 2010. However, UKIP did not stand in Clacton in 2010.

Applying a blunt 2010 recall weighting to these polls would therefore have weighed these voters to zero, despite the fact that they do exist and lived in the constituency when polled. These voters may have voted UKIP in a more recent local election, or moved to the constituencyafter the election in 2010. The fact that the Clacton polls quite accurately reflected the result of the by-election supports the inclusion of these respondents.

 

Conclusion

Differences between recall of 2010 vote in a given poll in 2015 and the actual result of the last general election in that particular constituency are not necessarily evidence that the sample is unrepresentative.

Furthermore, adjusting such a poll may have the adverse effect of making the reported results less representative.

 

————————————————

 Addendum

How Survation strive to ensure  the sample of people in our telephone polling is representative.

  • We begin with a demographically-balanced sample of the public to minimise the effects of weighting after the data has been collected.
  • We use a combination of landline and mobile telephone numbers in order to represent harder-to-reach demographics. 
  • Results are then weighted by gender, age and ward group to ensure a demographically and geographically representative sample based on 2013 ONS data.

 

 


< Back