Can non-probability sampling compete with RDD and other probability polls? In the 2020 election, yes -- in fact, they did better!
New research with @jake_rothschild@3Streamsblog: medium.com/3streams/revis… (a short thread)
We looked at every national poll in the @FiveThirtyEight 2020 forecast in the last 2 months of the election. After coding each sampling method, we found that non-probability polls had less overall error than RDD/other probability methods.
Aug 29, 2020 • 4 tweets • 3 min read
2020 Election Forecast with @JuliusLagodny@cornellgov forthcoming in @ps_polisci: about 6-in-10 chance #BidenHarris win the Electoral College, about 1/5 chance #TrumpPence get exactly 270 EC votes. (1/4)
Our state-by-state forecast, Dark blue=expected Dem win, Light red=expected Rep win; distributions centered around 50% could go either way; distribution height =Electoral College importance. (2/4)