In swift succession, YouGov has been accused of being anti-Labour, anti-Conservative, and (yesterday), anti-LibDem. Apparently, at various stages, we have had Labour / Conservatives / LibDems too low, and this is because we’re unrepresentative / skewing the weights / push-polling. I think we’ve dealt with the first two, but now we have to deal with the third – which is easy to do, since it just requires a simple explanation of the difference between push polling and message testing.
Most professionally-run campaigns will at some time want to test public reaction to a variety of messages, to see if its communication strategy is likely to work. This applies to companies and brands, and of course political parties. Typically, they will want to test both positive and negative messages, either to promote those messages or find out which they most need to defend themselves against. Market researchers will run those tests to specifically or randomly selected respondents. Typically some respondents will see one set of questions, others will see another set. When you see these questions, you can’t be sure who the client is, or what the reason for the test is.
Voting intention questions
Voting intention questions intended for publication must always come at the start of a poll, so that respondents cannot be affected by anything else that might be in the poll. In fact all questions asked in research that is for publication is carefully ordered so as to minimise the effect of any question on the questions that follow.
Push-polling is a very different beast. Push-polling is unethical. Push-polling is conducting a poll to influence a respondent for some particular purpose. Mainly it is when people are pretending to conduct a poll, but actually they are contacting hundreds of thousands of people to repeat attack lines – it’s campaigning masquerading as polling, and in New Hampshire it’s even illegal (and quite right too).
Another variant of push-polling (at least that’s how the phrase is often used) is when you ask ‘questions’ designed to influence the outcome of a poll. For example, if I ask you to choose which you like best from a list of positive attributes about a candidate and then ask you who you would want to vote for.
Message testing is an extremely valuable and reasonable form of research. Push-polling of any kind is plain wrong. YouGov, like all members of the British Polling Council, does lots of message-testing, and zero push-polling.
I hope that’s clear.
I obviously can’t tell you about who our clients might be or what they were interested in, but I can categorically state that all our published voting intention questions come at the start of our polls, as is standard practice. I also remind you that you can never tell as a respondent to a survey whether the questions you are seeing are the same as the questions a different respondent is seeing, nor whether the purpose is attack or defence or something quite different. YouGov conducts polls for at least three different political parties in this election, as well as for third parties that are interested in these things, such as academics. I hope that clears things up.