Bookies and in-depth poll suggest that Bush is underestimated... again

Apart from the remark that an opinion poll with a three and a half per cent margin of error is pathetic this poll reveals opinions that seem likely to favour the incumbent at the polling station.

Let's be clear just how poor this polling is: you could predict that Ralph Nader would have got zero votes in 2000, or that Bush would have won five extra states.

I've said on Samizdata that when it comes to forecasting, I trust bookies better than pollsters: we're talking people putting their money where their mouth is, and gamblers are not spin doctors. Wired shows the gap between people voting with money rather than with gratuitous chat.

There have been rumours that some unscrupulous individuals weight opinion poll samples to put more registered voters from a certain political party. As if by magic, this produces the result the media outlet wishes to peddle.

With a U.S. election I'm afraid polling is bound to be tricky. One million extra Democrat voters in California compared with last time don't mean a thing. Six thousand more Republicans in Wisconsin mean a lot more. Now design me an opinion poll which factors these issues in a country with relatively high voter mobility.

My preferred approach is to pick out the swing states and check them out with a mixture of polls, random interviews, even check out the TV advertising schedules.