Yesterday I looked at the presidential predictions. Today I'm looking at the Senate ones, with focus on those of us, individuals or aggregators, who made predictions on the victory margins.
Below are the predictions and average error for me, Nate Silver, Sam Wang, and the Huffington Post, Real Clear Politics, and Talking Points Memo aggregators.
The orange cells are the best calls for each race. The garish red and blues are races were the predictor got the wrong victor. So I erroneously thought Democrats would win Nevada and Republicans would win North Dakota.
Nate, RCP and I each got two of the 16 races wrong. Sam Wang, HuffPo, and TPM didn't get any wrong. Wang didn't make predictions on all of these races, just the ones that looked closest, so his perfect record is worth as much as anyone else's.
I may have had the lowest average error, but missing two races disqualified me from the crown. So of the people who called all the winners correctly, the winner is ... Huffington Post by a sliver.
A few quick notes:
I really don't understand why anyone respects RCP. They cherrypick polls and have a crap methodology. It wasn't so much that they missed two races, since both they missed were hard-to-call ones, but their average error (just like in the presidential results) is far beyond anyone else's.
The error margins are far bigger here than in the presidential race, and the reason is simple: There was far fewer polling for these races. The more polling, the more accurate the aggregation becomes.
Three of us got the North Dakota Senate race wrong. Silver only gave Democrat Heidi Heitkamp an 8 percent chance of winning! Or put another way, Republicans had an 92 percent chance of picking up the seat. So yeah, that was a crazy one. I went back and looked at the polling to see what I missed. Not much—much of the polling was either partisan (Dems had themselves +4, Rasmussen had GOP up +5, and they were both wrong), or unproven. This fresh new Pharos Research outfit had Heitkamp up +2, but they also had the Democrats trailing by just 2.7 in the Nebraska Senate race they lost by 16.2. They weren't looking very credible. But ultimately, even TPM composite had Heitkamp at 46, while HuffPo had her at 47.9. In a Red state, Heitkamp should not have gotten the undecideds, but she's a spectacular politician and somehow pulled it off.
Look how far off everyone was in Missouri. Most polling had the race close, but there were two standouts: SuveyUSA gave McCaskill a 15-point lead, DSCC internals had it at 12, while McCaskill campaign internals had it at 14. They seemed optimistic at the time, but they nailed it. For comic relief, the Republican pollsters at Wenzel Strategies, polling for Citizens United (yeah, that Citizens United), had Akin up four points at the same time as that McCaskill internal. It must've taken some serious unskewing to get those numbers.
I'm still struck by the wide disparity in many of these races among the pure poll aggregators (HuffPo, RCP, TPM). Just because someone says they're averaging out all the polls doesn't mean that they're doing it the same way. Or put another way, RCP really, really sucks.