Discussion about this post

User's avatar
Michael Kelly's avatar

Or perhaps Republicans have found their own way to 'stick it to the man.'

Expand full comment
mdhh's avatar

Thanks for the post.

"It’s fairly easy to argue that adding the previous polling bias is the mean of what we can expect when we’re still too far in advance of the next election of what it would happen and therefore minimizes least squares across the different ways Alaska and America could diverge in the future."

Could you be a bit more precise about this? I don't really understand what you mean.

Based on the results in the table, previous bias doesn't seem to predict the magnitude or direction of future bias well at all (in fact, simply adding as you suggest would lead to worse predictions more often than not).

I also think I disagree with your overall criticism of pollsters pretending to be scientific. More than most areas, polling seems like a field where the proof is in the pudding, and nobody denies the subjectivity of modelling decisions. The sheer fact there are a variety of different polling companies who claim to be more accurate than each other suggests that this is well understood.

I also think it's important to take into account the fact that the business model of polling companies relies on them being able to sample public opinion on a range of different things, not just predict elections. Elections are where their methods are publicly tested, but if they're systematically missing some of the population, then fudging the numbers as you suggest might give a good prediction, but it won't be much good when Kellogg's need to understand the impact of their new marketing campaign for Cornflakes.

Expand full comment
4 more comments...

No posts