Tuesday 29 April 2008
Below is an extract from a presentation outlining our methodology for political polls, and busting some common myths about political polling.
Political Polling Overview
Introduction
Ipsos MORI has been conducting political polling since 1979.
We run a monthly Political Monitor, which includes questions on:
- Voting intention
- Likelihood to vote
- Satisfaction with Government and party leaders
- Most important issues facing Britain
- Economic optimism
Ipsos MORI has the most long-term trend data and comprehensive set of political polling data of all polling agencies in the UK
Election Polling
Most political polls ("peacetime" polls, which do not occur during election campaigns) are a snapshot indicator of how the public (or a defined section of the public) thinks they would vote at a given moment.
- Not "what would happen" — this is too hypothetical to be meaningful
- If different companies design their polls in different ways they are measuring different things
- Ipsos MORI measures two separate things:
- How the whole public thinks they would vote; and
- How those who say they are certain to vote think they will vote.
It is useful to think of polls as a barometer.
- Barometers do not predict the weather, they measure something that is useful to know in predicting the weather
What do voting intention polls measure?
Election campaign polls measure what the public think they are going to do — which may or may not be accurate.
Only the final election poll, usually conducted the day before the election itself, is a "prediction".
- The final prediction may include adjustments not applied to other polls — because the final prediction is not "pure" polling
- Polls do not predict (though pollsters do, from the data they hold, the models they develop about elections, and from their own expertise)
This is in contrast to "peacetime polls", in which we are asking about a hypothetical election, and so the answers are therefore hypothetical too 10% of voters say they make up their mind on whether and/or how they'll vote in the final 24 hours of the election, and another 10% during the final week — this makes it even more difficult for pollsters to assess opinions accurately
Polls of polls
A poll of polls is useful if it allows separate polls to be combined for bigger sample size. This requires that:
- The polls are measuring the same thing;
- The sampling methodologies are broadly comparable; and
- Either fieldwork dates are similar, or nothing has happened to change ublic opinion between the two (unlikely given the immediacy of the media today).
A "poll of polls" combining incompatible surveys may be useful as a diagnostic tool (e.g. how well did the polls collectively perform, which company is most out of line with others), but not as a means of increasing accuracy of measurement.