Shock poll predicts Tory losses

[AtW]

Hung Parliament, that's something I'd vote for!!!

[eek]

Wait to you see the methodology they are using

[chopper]

I assume this is just to make sure Tory voters go out and vote?

[northernladyuk]

[SueEllen]

http://youtube.com/watch?v=ROKXlvYMKQc

The British electorate doesn't like bald men.

[jamesbrown]

I'm quite surprised that the Times went with this. Or, rather, I would be if it weren't a Murdoch rag It can't be described as a poll. It's a constituency model whose statistical parameters are estimated with polling data. These type of models have a very poor track record.

[eek]

I'm quite surprised that the Times went with this. Or, rather, I would be if it weren't a Murdoch rag It can't be described as a poll. It's a constituency model whose statistical parameters are estimated with polling data. These type of models have a very poor track record.

The argument for publishing is that it was accurate in the referendum...

That only leaves likelihood to vote and numerous other flows in the model. Its like using the huffingtonpost poll of polls that lists Tories 33%, Labour 31%, undecided 21%, rest 15%... Unless you know what that 21% is going to do and trust that everyone stating their choice is going to vote its pointless...

[jamesbrown]

The argument for publishing is that it was accurate in the referendum...

That only leaves likelihood to vote and numerous other flows in the model. Its like using the huffingtonpost poll of polls that lists Tories 33%, Labour 31%, undecided 21%, rest 15%... Unless you know what that 21% is going to do and trust that everyone stating their choice is going to vote its pointless...

The inherent problem here is that these polling events cannot be treated as realisations of some stationary random process. Every poll has unique context, particularly at a constituency level, and that context is essentially invisible to the pollsters. They're constantly tweaking the structure of their models (e.g. turnout filters) to better map the most recent polling bias, but that is difficult to justify for the above reason. In this context, both their expectations and the estimates of uncertainty surrounding those expectations are highly questionable. Polling is a pseudo-science at best (i.e. a business ), and constituency models take this to a whole new level. It's the illusion of precision.

[Fronttoback]

The inherent problem here is that these polling events cannot be treated as realisations of some stationary random process. Every poll has unique context, particularly at a constituency level, and that context is essentially invisible to the pollsters. They're constantly tweaking the structure of their models (e.g. turnout filters) to better map the most recent polling bias, but that is difficult to justify for the above reason. In this context, both their expectations and the estimates of uncertainty surrounding those expectations are highly questionable. Polling is a pseudo-science at best (i.e. a business ), and constituency models take this to a whole new level. It's the illusion of precision.

Each poll has a unique context that makes the result inaccurate.

But if this is true of all polls, wouldn't they all be equally wrong by the same degree, on average?

So, one could draw a useful trend but not an accurate snapshot consensus.

And that, my friend, is why I have a GCSE in Maffs.

[northernladyuk]

Each poll has a unique context that makes the result inaccurate.

But if this is true of all polls, wouldn't they all be equally wrong by the same degree, on average?

So, one could draw a useful trend but not an accurate snapshot consensus.

And that, my friend, is why I have a GCSE in Maffs.

But what does it all mean for Kebuber?