Prediction performance

According to SurveyUSA's pollster scorecards, which track 8 different measures of scoring polling results:

North Carolina: PPP was the most accurate by five measures, Zogby by three.

Indiana: PPP was the most accurate by five measures, InsiderAdvantage by three.

Overall, Research 2000, who does polling for Daily Kos, was second best in North Carolina (out of 13) and fourth best in Indiana (out of 10).

Meanwhile, here are the results of my last three predictions:

    My prediction      Actual      Margin of error
PA  C:54.0 O:46.0   C:54.6 O:45.6   1.0 point
NC  C:43.9 O:56.1   C:41.5 O:56.2   2.5 points
IN  C:51.1 O:48.9   C:50.7 O:49.3   0.8 points

Damn NC burns me up, since I was one tenth of a percent off on Obama, but forgot to account for "no preference" and Mike Gravel on the ballot. Combined, they got 2.4 percent. I would've guessed 1 percent, so that would've brought my margin of error to under 2 points. But that's just me being greedy.

Given that I pull these out of you-know-where, it seems my ass is more accurate than the pollsters! I jest, I jest. The polling composites were a big part of my prediction system. But if you want to know why I abandoned SUSA for the NC predictions, something I'm loathe to do, check this out:

             SUSA       Actual
Charlotte  C:52 O:42   C:29 O:70
Raleigh    C:42 O:54   C:34 O:65

Of course, it's hard to say what SUSA meant by "Charlotte" and "Raleigh". I assumed metro area, while those actual results are from the counties those cities reside in. But if we take "Raleigh" to mean the research triangle, that was always a region that Obama was going to win by about 30 points. And there was no way that Clinton was going to win the Charlotte region.

In any case, don't take much stock in my predictions. I've blown way too many to pretend that I'm much good. But it does feel good to hit a couple once in a while. But there's one more theory of mine that got additional support: that the only endorsements that matter are those from politicians that have machines.

Obama won Connecticut on the strength of the New Haven machine. Clinton won California on the strength of the Los Angeles Villaraigosa machine. So despite the hooplah, I immediately discounted the Sen. Bob Casey endorsement in Pennsylvania for Obama, and the Sen. Evan Bayh and Gov. Mike Easley endorsements in Indiana and North Carolina for Clinton. Then I gave Obama a point in both states for the quality of his mayoral endorsements in Indiana and North Carolina. Looking at Obama's performance in northeastern Indiana, I think the mayors may have been worth two points to his margin. Indeed, SUSA predicted a reasonable African American turnout of 10 percent. Exit polls pegged it at 18 percent, nearly double their state's percentage (about 9.5 percent).

I think it's pretty clear that the only endorsements that truly matter are those that bring patronage machines into the fray.

Update: Here are SUSA's regional definitions for NC, and the rest of the country can be found here.