Read on Twitter
Brief thread which diagnoses betting market differences with 538 as resulting from reliance on RCP (possibly bogus) polls.
First the "puzzling" fact—538 is generally more optimistic about individual state odds than betting markets are:
First the "puzzling" fact—538 is generally more optimistic about individual state odds than betting markets are:
The odds here come from a mix of the state PredictIt contracts/Betfair, and so are less sensitive to concerns about maxing out the number of entrants (not true in the PredictIt states) or liquidity/transaction costs (not really a problem in Betfair).
538's vote share estimate also differs from the betting market forecast. In particular you can see this key quadrant, on bottom right — GA, FL, NC, AZ — any of which would guarantee Biden win; which 538 thinks Biden will get and betting markets not.
To some extent, this is betting markets just being uniformly more optimistic about Trump across states.
But if we look at the size of the 538-betting market bias state by state, there is an interesting structure:
But if we look at the size of the 538-betting market bias state by state, there is an interesting structure:
All of these high-disagreement states — MI, WI, PA, FL, MN — were pivotal last time. Trump won all of them except MN, where he did unusually well.
There is a mild correlation between the size of polling errors in 2016 with the discrepancy between betting markets and 538 today. The geographic structure of past polling errors doesn't account for much of the discrepancy.
Now, another theory is that betting markets are just following RCP, which is different this cycle as they are weighting more iffy pollsters like Trafalgar relative to 538. https://twitter.com/NateSilver538/status/1323645591885414415
I think this is basically what's going on. The size of the RCP polling average - 538 predicts reasonably well the betting market - 538 difference.
We'll see what happens tonight, but overall this doesn't look great for betting markets. They just happen to be weighting pollsters (like Trafalgar) which have a strong Trump leaning; relative to the 538 average.
If Trafalgar is closer, that will validate the betting markets.
If Trafalgar is closer, that will validate the betting markets.
I'd like to thank nonymous referees ( @azsavov) who greatly improved the quality of these tweets.
