Quick guide to why polling goes wrong – part 1

Posted this on twitter yesterday and it had cause to be mentioned by me on a blog-site as a comment (currently under moderation). So here it is in a nutshell.

• Labour supporters Mr Smith turns out 9/10 of the time; Mrs Jones 6/10 due to differing certainty.
• This ‘inconsistency’ (heteroscedasticity) gives BIASED results (Yatchew & Griliches 1985) if you aggregate ‘discrete’ (vote/not vote or Labour/Conservative) answers – not just problems in ‘margin of error’
• You need a choice model : understand inconsistency within a person’s head, adjust for it, then aggregate
• This (partly) explains why polls these days are so variable and get the answer wrong so often.