,
3 tweets,
1 min read
Read on Twitter
So, we published an interactive today that basically looks at every probabilistic forecast that FiveThirtyEight has ever made (dating back to 2008).
Here's what we found: When we say 70 percent, it *really does* mean 70 percent. 53eig.ht/2OWhD33
Here's what we found: When we say 70 percent, it *really does* mean 70 percent. 53eig.ht/2OWhD33
Technically speaking, our elections forecasts have been a bit *under-confident*, meaning favorites have won a bit MORE often than they're supposed to, but the differences from perfect calibration aren't statistically significant. Overall, the forecasts are really well-calibrated. 
Here's the interactive itself. Obviously, there are lots of ways to criticize 538. But some of the more philosophical criticisms about probabilistic forecasting just don't hold up very well if you look at the actual track record. 53eig.ht/2uNtE1c
