Onepoll survey in the People
Tomorrow there is apparently a OnePoll survey in the People. Until some tables or full methodological details appear, I'm wary. In the past I've been rather rude about the sort of PR-puff polls that get churned out just to fill column inches, which celebrity would you most like to be stuck on a desert island with, which celeb has the nicest bottom, that sort of rubbish. If you are a company who would like to raise public awareness of their brand, an easy way to get column inches is to commission a quirky or unusual poll question, write a fun couple of paragraphs about it, and send it off to the tabloids who need to fill column inches. The company raises their profile, the paper gets a nice little story (and the polling company gets the money for doing it, let's not be shy about it - PR companies doing this sort of thing provide a steady income stream for polling companies).
Ben Goldacre calls this sort of stuff "PR reviewed data", and has been particularly disparaging about OnePoll, who specialise in producing this sort of polling for PR purposes (they will write up copy in the style of the newspapers targetted for example, and website says they offer advice on writing question to get "newsworthy" results, which isn't necessarily the same as if you prioritising accuracy). Anyway, when Ben Goldacre wrote the article I linked to Harriet Crosse of OnePoll wrote this article in response, which essentially said that OnePoll did produce cheap and cheerful surveys to drive PR campaigns and get fun articles in the paper and so what? If people don't like that sort of survey, don't read the article. That's a fair argument.
However, if you have a panel of 50,000 people for doing PR puff surveys on, there isn't necessarily any reason why you can't do proper surveys on them too. It takes different skills of course - a nicest bottom survey doesn't need political weighting for example - but these are not insurmountable. The biggest barrier of entry to doing online polling is getting a panel.
The reported OnePoll figures in the People seem sensible enough - CON 37%, LAB 31%, LDEM 21% and presumably others on 12%. Anyone can do post-hoc political weighting of a poll to get results that look sensible though - to judge it properly, we need to know more about how it was done, exactly what weighting variables were used, what sort of political weighting was used, how it is sampled and so on. Without anything to judge it by, we should be wary.
