Wednesday, March 09, 2005

Push polling

I did an opinion poll this morning.

I'm registered with YouGov a UK opinion poll service and get occassional polls sent to me - they have either a prize draw or a 50p (NSD1.45) bonus attached, as well as sometimes being quite interesting.

This one was in the latter category:

They started out by asking various questions on identity theft: had I been a victim of, were the banks/government/telcos doing enough to prevent it, did I feel confident with internet shopping, etc.

After about 5 questions of this nature, they had one similar to this:
Select one or more of the following:
- I think ID cards will help reduce identity theft
- I think ID cards will make banking more secure
- I think ID cards will help combat terrorism
- None of the above

Now, for those who don't know, one of the many illiberal measures being brought in by the UK government is a mandatory ID card scheme. I strongly suspect that the above poll was commissioned by the very same UK government. A poll like this is called a "push poll" - the idea is that the preceding questions put an initially undecided subject in a frame of mind to give the "right" answer - which in this case can be headlined as "XX% support ID cards - poll" where XX% is all those who didn't answer "None of the above".

I wonder how many of the "impartial surveys", both in the UK and NZ, are in fact organised along these lines. (YouGov is regarded as a reputable source - although I and others have doubts about their methodology of taking a self-selecting minority of internet users and then adjusting the figures to correct for sample bias).

2 comments:

Joe Hendren said...

Well spotted.

It concerns me that the media encourage such polls as larger percentages make more dramatic headlines. Polls sell news, critique of dodgy polls sells less news.

I assume that as you have to register for the YouGov polls, they are in effect self selecting.

Rich said...

YouGov is self-selecting, but the fact that they pay for opinions does slightly increase the variablity of the sample from just the opinionated. (the computer also selects you for a poll - you don't get to see one you have a view on and vote).

For their usual political and commercial research, they claim to have a statistical model that corrects for sample bias. I'm a bit sceptical - I can't see how you deal with the fact that they probably get zero respondents from, for instance, C2DEs aged over 60.