Redc Research

Polling Techniques

Whichever method RED C utilise to conduct any poll, there are a number of techniques we use in an effort to achieve the most accurate result.

These include:-

Careful Question Wording
Representative by Past Voting
Analysing by those Likely to Vote
Adjusting for Refusers or Don't Know's

Question Wording
The introduction to a poll and the question wording has been shown to be vital in its impact on voter's response. Immediately announcing you are doing a political poll when you knock at the door or ring someone up, can lead to a high refusal rate and a biased sample, whereas more careful introduction and emphasis on the value of each individual's opinions ensures a more accurate sample.

Weighting by Past Voting
When we undertake any poll we set quotas for different demographic groups and the sample is then weighted by demographics (age, sex, tenure etc) to ensure it is representative of the total electorate within the prescribed area.

We then also ask respondents whether they voted in the last general election and if they did which party they voted for.

Even after weighting by demographics, the declared past votes may not match, exactly, the results of the last election. Partly this is because demographics (by which RED C control the sample) are relatively poorly correlated with vote behaviour. Nevertheless past vote weighting has to be used with caution as some people genuinely forget how they voted. We cannot simply weight the data to the actual results last time.

Instead RED C compares the declared past votes to the actual result of the last general election and weights to the mid point of the two, thereby assuming that half of the difference between declared past votes can be attributed to faulty recall and half to remaining political imbalances in the sample.

Likelihood to Vote
It is also important that we only give credence to the voting intentions of those who will actually vote in the next election. As you know, in any one election turnout might be at only around 60% of all eligible to vote. It is important therefore that we control our voting intention questions by how likely people are to vote.

RED C ask respondents to say how likely it is that they will go and vote in a new general election using a ten point scale where 10 means they would be absolutely certain to vote and 1 means they would be certain not to vote.

We then exclude anyone who gives a score of 4 or below from our voting intention calculations, as these people will not vote in an election anyway. This normally brings our national turnout to about 80%. This may be higher than the actual turnout. But 100% turnout could not be achieved because of faults in the electoral register. Evidence also suggests that polls tend to interview slightly more voters than exist in the whole population.

Adjusting for Refusers/Don't Knows
Some people tell RED C that they intend to vote, and have a good past voting record, but say they don't know who they would vote for or refuse to answer the question.

Whenever such people say which party they voted for in the last election, RED C ascribe 50% to the party they voted for previously. We have seen from analysis of past elections that this is the most likely outcome and overcomes the effect of the "spiral of silence" where voters are do not want to admit who they will vote for.