search

fivethirtyeight / data

Data and code behind the articles and graphics at FiveThirtyEight
https://data.fivethirtyeight.com/
Creative Commons Attribution 4.0 International
16.7k stars 10.96k forks source link

blending predictit market data #269

Closed blazespinnaker closed 1 year ago

blazespinnaker commented 3 years ago

I've had a chance to look over the Nov 2nd prediction market data.

It called ME-2, Florida, NC, Georgia for R, versus the D calls by 538 (https://projects.fivethirtyeight.com/2020-election-forecast/)

AZ, NE-2, PA, etc were called same by PI/538. Same for Ohio, Iowa, Texas, etc.

Georgia is probably going to be D, but still I think it's pretty clear PI out-performed this time in the prez market.

PI had a slight R house effect. Similar in 2018

Are you blending PI data? latest is easily available here -https://www.predictit.org/api/marketdata/all/ (pipes out json if you request from python)

blazespinnaker commented 3 years ago

Senate predictions look the same, but PI caught some house markets that 538 did not - NM and Oklahoma didn't flip to D. The rest of the house seem to be the same calls.

https://fivethirtyeight.com/features/final-2020-house-forecast/ https://fivethirtyeight.com/features/final-2020-senate-forecast/

blazespinnaker commented 3 years ago

Here's the data from Nov2. I captured it the same time 538 froze their predictions.

https://github.com/blazespinnaker/predict/blob/main/Nov2PI

If you can upload the poll data from 2020, that'd be helpful.