Quote:
Originally Posted by ernie
It isn't just the presidential election though that he called. In 2012 he called many many more down ballot races as well. They talked about it on the Daily Show at the time.
Also to call 9 swing states correctly, something that no pollster nor any other aggregate statistician did, is not something that can just be written off.
In 2008 he called 49 of 50 states (50 of 51 if you include DC) for the presidential race. He called every single Senate race correctly.
In 2010 he called 36 of 37 gubernatorial races. 34 of 37 senate races
IN 2012 he called all 51 of 51 for the presidential election. No pollster or other aggregate models predicted every single swing state but 538/Nate did.
That is very unlikely to be random chance. He has proven to have a better model than anyone else at this point. Doesn't mean he is always correct or that his current model tweakings will be the best now but his track record does indeed speak for itself.
|
I know that and in general I agree with you,
And in general I agree with you that his model is reasonably accurate and more importantly the underdogs win at about he expected rates. However going 34/37 on senate races is likely a coin flip as there are only 5-6 competitive races each term. So if you consider 6 meaningful predictions each time and run the odds. He has gone 25/30 which if you give the random selection a 60% chance of being right has about a 1/500 chance.
The other part is that his model could be just as accurate as it is right now and he could only have predicted 20/30 close races correctly. His predictions would be just as accurate but he wouldn't be as famous. An example of this is his Obama / Florida race which his model showed a .1% edge for Obama. This is a meaningless difference between a .1% edge for Romney yet allows him to go perfect.
I think in general the last Month he has been trying to get this point across that when Trump had a 15% chance and if he were to win that that would be well within his model prediction.
He gets too much credit when he is right and not enough credit if he is wrong even though his model predicts he should be wrong regularly.