State polls about average_of_polls HOT 9 OPEN

So I built a module for these state primaries and loaded it locally last night. It'll go live probably by this weekend. Anyway, the ratio of noise to signal in the state primary/caucuses is really high and the smoothing algorithm doesn't really know how to handle it. It's a cool idea but I don't think it'll be workable until right before each state's primary, if ever.

Working on a few other neat ideas though. Stay tuned.

from average_of_polls.

I'll look into it. If I can find a good data source that compiles all polls together by state, I'll make a module for it. Upkeep will be a pain though. I think you're right that it could be essential, especially if we wanted to do a 'polls-only' model of who is going to win the nomination.

from average_of_polls.

Okay, added the first four primaries--doesn't seem like there's a big database for the remaining 53 states/territories. I'll update the site when I get to my desktop tonight.

from average_of_polls.

Yeah the RCP early primary polls are pretty much useless with how sensitive the model is

from average_of_polls.

Nate Silver's model (yet unpublished for this season) is approx 9,000 lines of stata code, so that should give some indication of what exactly we're working with in terms of complexity.

I can't even get the damn thing to control for house effects. Their primary polling model is coming soon and will do this at which point this repo will be a bit redundant/obsolete.

from average_of_polls.

Does RCP use likely voters / all voters / registered voters? I've noticed different numbers than 538 sometimes.

Also what do you think about RCP excluding some pollsters? Ipsos for example

from average_of_polls.

Also I think 538 just updated their pollster ratings, so the house effects probably are different now too.

Yeah, just checked, they are all different.

from average_of_polls.

(1) RCP uses different things because polls report different things... this is a source of measurement error that is systematic and one I would like to eliminate with mixed effects if / when I have time to sit down and figure it out.
(2) in general, RCP is reporting democratic primary voters.

from average_of_polls.

(3) will get into house effects the next time I sit down for sustained work on this project. Thanks!

from average_of_polls.