Monday, 28 April 2008

The battle of the polls

In the London mayoral election internet poll YouGov is forecasting a big victory for Bozza while telephone based Mori puts Ken ahead. This may suggest that in fact Boris has a small lead, but one of the polling methods and companies is going to end up with egg on its face.

The YouGov panel does tend to attract people who have a higher than average interest in politics. Some of the differences between the two polls are being discussed on the political betting site - I cant give the reference with this Quebecker computer, but we have referred to it before.

7 comments:

carman said...

there is a good discussion of the differences in polling firm weighting procedures (and subsequent implications for poll results) on the UK Polling Report:
http://ukpollingreport.co.uk/blog/index.php

Justin Greaves said...

I am probably totally wrong but I predict that Boris will win by 4% after 2nd preferences are taken into account. Not too long to find out...

Wyn Grant said...

I think that Boris is going to win - then the fun will start.

Justin Greaves said...

On my calculations Boris has won by 6%, exactly in line with YouGov's last poll. Therefore, they called this election far better than the other polling organisations. I think this will help put to rest any doubts about YouGov's polling methods.

Wyn Grant said...

I remain unconvinced. Why were they showing such a big lead for Boris earlier on

Justin Greaves said...

I assume because of a late swing back to Ken in the last few days?

I think we also have to see this in the context of YouGov being only 1% out in the 2004 mayoral election and their good track record (more generally) in the UK. Of course, pollsters can get lucky but not I feel to this extent (I recall you saying a YouGov spin off hasn't so successful in the US, so I am talking just of polls in the UK here).

Wyn Grant said...

Statistically one would need a large N of comparisons between different polling methods to come to a valid conclusion. I don't have that data, but I would be worried about committing an individualistic fallacy. It is a concerning implication for the way in which polls have been done - but I doubt whether probability theory is wrong.