A couple of days ago I was lucky enough (genuinely) to attend (sneak into the back of) a polling seminar hosted at the LSE featuring professional pollsters, academics, and a few assorted others like Mike (also complimentary wine). To be specific it was hosted somewhere in the depths of an especially labyrinthine and oddly signposted LSE building that left attendees wandering the corridors in search of rescue. LSE is one of the universities that is attempting to engage with the non-academic community, with a pretty extensive set of blogs, events like this, and the promise of more online resources to come.
In the interests of brevity (and getting two posts for the price of one) I’ve split this write-up between the presentations of the pollsters and academics (brilliantly mirroring the format of the seminar itself).
Mike has already covered Nick Moon’s low opinion of Populus’ party id methodology, I happened to be sitting next to the Populus representative, (@richardbridger, I thought his efforts earned a plug) who defended his company’s honour, using largely the same arguments as Rick Nye essentially that they stood by making a judgement call that worked when backfitted into previous election results.
There was a general consensus that the challenges of polling (and related seat predictions, which pollsters were pleased to note weren’t really their problem) was becoming harder as the two party voting dominance at elections had declined. Polling benefits from being tested against results, and as the situation changes then previous assumptions become less applicable and leads to greater uncertainty, UKIP being noted as likely to offer the biggest difficulty in accurate polling.
(Due to space constrictions and sitting far away from powerpoint slides written in small type I’m heavily simplifying all over the place).
There were also some concerns raised with how the media covered polling, with emphasis being placed on outlying results, dramatic headlines based on stretched interpretations and the dabbling in the political arena of PR-offshoot companies using methodologies that are often less than transparent and rigorous.
The tone from all the pollsters was that of caution, that it’s likely to get harder to be accurate but with election polling being essentially a marketing exercise there isn’t the funding available for most of the techniques to improve accuracy (increasing sample size etc). All this alongside an acknowledgement that polls is measuring a hypothetical, they ask about voting intention if there were a General Election tomorrow in full knowledge that if there were one the campaign would mean different results. Peace-time polls are not only not a prediction but at best a snapshot of an near impossible event (i.e. an election without a campaign).
Last of all (perhaps inevitably) came questions about predictions, which after all talk of caution saw some pretty interesting and predictions. Joe Twyman (Head of Political and Social Research at YouGov) predicted (and had Nick Moon agreeing with him) Labour winning most seats in a hung parliament with the Conservatives winning most votes?
Any takers? I haven’t checked but I suspect there’d be some nice odds available.