No, the Liberal Democrats are NOT on 8%

Many years ago Teletext (those of you over 25 will remember it) used to have phone in polls on issues of the day. On occassion they would ask voting intention, and it would invariably show the Conservatives on about 80% of the vote even in the midst of Blair's greatest popularity - presumably because only elderly Tory voters bothered to ring into Teletext polls.

I was rather reminded of it by this from Sky News. Conducted on their own panel it has voting intentions of Conservative 43%, Labour 24%, Liberal Democrats 8% - repercentaged to exclude don't knows and wouldn't votes, it works out at CON 50%, LAB 28%, LDEM 10%, Others 13% - so while reputable pollsters are showing a Conservative lead of between 2 and 6 points, Sky's panel are showing a lead of 22 points. That rings alarm bells to say the least.

This isn't actually a voodoo poll in the purest sense, it was conducted using a panel, rather than an open access "red button" poll (although there is no indication of whether there was an attempt to draw a representative sample from within the wider panel) - but the sample looks very ropey and there is no apparent attempt at proper political weighting. There are sparse demographic details in the results, but the 2010 recalled vote break shows 44% of the sample voted Tory, compared to 17% Labour and 18% Lib Dem. For context, established polling companies like ICM weight their polls so that 25% of the sample is people who voted Tory in 2010, 21% Labour and 16% Lib Dem.

You sometimes get fun little red button polls on media websites, but they normally come with disclaimers that they are not properly represenative polls. In contrast, Sky have it as the headline on their website, liberally sprinkled with quotes from their Chief Political Correspondent Jon Craig about what it would mean if repeated at a general election. Sigh.

Ignore (and for journalists out there, this summary by Peter Kellner from the BPC website about when to pay attention to a poll is always worth revisiting).

UPDATE: Jon Craig's blog here at least starts by acknowledging "Now I know the sniffy ones among you - yes, you know who you are - will say it's not a wholly scientific, weighted opinion poll and all that." On one hand, I'm pleased he's added the caveat. On the other hand, one is rather tempted to reply that you shouldn't bloody publish it then. Wanting polls to be scientifically weighted is not some odd personal fetish or the pedantry of pollsters and statisticians in ivory towers, it's that all that makes a poll meaningful is that it is representative of the wider population, through proper weighting and/or sampling. A poll that doesn't do that is just the views of an arbitary 1500 people, who do not necessary represent anyone but themselves.