Hey gang. I've been looking at election statistics for a long time, and do you know what I've learned? By and large, candidates who are leading in the polls actually end up winning.*
Those dots are each a 2018 election result (Sen, Gov, House), compared with late, high-quality (B or better) polling. I average them, weight them by date (> 1 month are basically ignored) and plot them against the results. It's a straight line, with no bias and a systematic noise of $\sigma=3\%$.
While I'm plotting 2018, the same has been true for elections in 2017 (NJ, VA), 2019 (KY, LA), and special elections (AL). A wide variety of states, candidates, etc. Polls, by and large, are pretty good at predicting the future.
But wait!" you say, the polls totally screwed up 2016! Well, yes and no. Nationals were off by 2, state polls, in high-consequence states, were off by 3, which is actually pretty typical. Putting in ~3% systematic noise, here are the odds you might have given Hillary:
There's a lot going on here, but looking at just the quality polls, Hillary was at about 84% and falling on E Day. 538 gave her 70%. Betting markets gave her 80. More likely than not, but simply saying "polls were wrong" is like saying a .160 hitter NEVER gets a hit.
I like to think of the state of polls as "$\sigma$s" above/below a coin flip. Hillary was at +1 on E Day. Here are the last 6 months of $\sigma$s for the last 4 elections (along with results).
Again, a lot going on, but the upshot is that races typically bounce around within a 1-sigma range, and that the outcomes tend to be close to the final predictions. 2008 and 2004 were dead on. Ds overperformed by about a point (1-sigma) in 2012. Rs by a point in 2016.
So where are we now?
Remember, the model is simple:
Simple, but very powerful. While the "unlikely" happen occasionally, they are rare. For instance, here's another view of those Senate, House, Gov races.
A candidate who has a 10% chance (from polling) will win ~10% of the time.
Here is the last 4 months of data for Biden, Sanders, Warren. All 3 could be expected to beat Trump (and the 1st two pretty handily) if the election were held tomorrow.
But what's more, we've seen that the actual amount of fluctuation in a race is quite small.
I should note that impeachment hasn't really moved the dial one way or another. There's a lot that's baked in. It's been like that in the past. The only events that moved polling more than 0.5$\sigma$ were when at least 1 party settled on a candidate.
The Republican party has already settled (in every way imaginable). Based on history, Dems picking theirs will, if anything _improve_ their relative standing.
Don't get complacent, but don't assume we're going to lose. The narrative that Dems are in trouble or that this is Trump's race to lose is frankly insane. Based on history, on actual data, the Dems are the overwhelming favorites to win next year. Please act like it.
---------
*This post originally appeared as a tweet storm.
While I'm plotting 2018, the same has been true for elections in 2017 (NJ, VA), 2019 (KY, LA), and special elections (AL). A wide variety of states, candidates, etc. Polls, by and large, are pretty good at predicting the future.
But wait!" you say, the polls totally screwed up 2016! Well, yes and no. Nationals were off by 2, state polls, in high-consequence states, were off by 3, which is actually pretty typical. Putting in ~3% systematic noise, here are the odds you might have given Hillary:
There's a lot going on here, but looking at just the quality polls, Hillary was at about 84% and falling on E Day. 538 gave her 70%. Betting markets gave her 80. More likely than not, but simply saying "polls were wrong" is like saying a .160 hitter NEVER gets a hit.
I like to think of the state of polls as "$\sigma$s" above/below a coin flip. Hillary was at +1 on E Day. Here are the last 6 months of $\sigma$s for the last 4 elections (along with results).
Again, a lot going on, but the upshot is that races typically bounce around within a 1-sigma range, and that the outcomes tend to be close to the final predictions. 2008 and 2004 were dead on. Ds overperformed by about a point (1-sigma) in 2012. Rs by a point in 2016.
So where are we now?
Remember, the model is simple:
- Average recent, high-quality polls by state and nationally.
- If there are no polls in a state, assume national sets the trend.
- Assume there is some unknown systematic error of ~3%.
Simple, but very powerful. While the "unlikely" happen occasionally, they are rare. For instance, here's another view of those Senate, House, Gov races.
A candidate who has a 10% chance (from polling) will win ~10% of the time.
Here is the last 4 months of data for Biden, Sanders, Warren. All 3 could be expected to beat Trump (and the 1st two pretty handily) if the election were held tomorrow.
I should note that impeachment hasn't really moved the dial one way or another. There's a lot that's baked in. It's been like that in the past. The only events that moved polling more than 0.5$\sigma$ were when at least 1 party settled on a candidate.
The Republican party has already settled (in every way imaginable). Based on history, Dems picking theirs will, if anything _improve_ their relative standing.
Don't get complacent, but don't assume we're going to lose. The narrative that Dems are in trouble or that this is Trump's race to lose is frankly insane. Based on history, on actual data, the Dems are the overwhelming favorites to win next year. Please act like it.
---------
*This post originally appeared as a tweet storm.
I plan on expanding out some of these analyses because they are super-interesting, but figured I'd lay out the basics for now.A thread about misplaced cynicism:— Dave Goldberg (@askaphysicist) December 13, 2019
1. Hey gang. I've been looking at election statistics for a long time, and do you know what I've learned? By and large, candidates who are leading in the polls actually end up winning. pic.twitter.com/pLpR0EyQke
Comments
Post a Comment