« Twitter roundup | Main | Twitter roundup »

November 10, 2011


One quick correction, I use the preference poll and not the presidential approval rating at Labor Day.

My major problem with Silver's model (other than not seeing the data and the estimation) is that it assumes a Downsian centrist strategy is optimal. I think we know now that base voters cannot be taken for granted. There is a price to be paid for being too centrist. As irrational as it may seem, extremists in both parties will stay home if they think their party's candidate does not represent them well enough. In effect, their seemingly irrational behavior gives them more leverage in the process. Just ask President McCain. All of this should not come as a surprise since turnout itself only makes sense if it is seen as an expressive rather than instrumental act.

All the models are correlated with each other, since they are fitting to basically the same set of underlying data, and of course the Ensemble is correlated with all of them. When you calculated the error bars for the Ensemble did you take correlations into account?

Apologies to Jim Campbell - the error is corrected above.

This is the coolest thing I've read all week! I wish there were more posts!

Thank you for the corrective.

I'd still like to see someone note that RDI is actually a good measure of the underlying concept we're interested in: is the average person likely to see themselves as better off or not? As economic inequality has grown so dramatically, GDP is increasingly a bad measure of how average people are doing.

Economic conditions and poll results are objective, whereas the public perception of a candidate's extremism is subjective. So inclusion of Extremism Index leads to considerations of Media Bias.

Many conservatives think Obama's previous record before running for President showed him as extreme. They point to:

-- his voting record, though thin, ranked him as the most liberal, according to one measure.
-- his 20-year membership in an extreme church
-- His association with (literal) bomb-thrower Bill Ayers.

However, candidate Obama portrayed himself as moderate and the mainstream media pretty much went along with pooh-poohing these three facts.

OTOH Sarah Palin had a fairly moderate record, again quite thin. Silver doesn't show her extremism index (since she was only running for VP), but I would guess it was higher than Obama's because she got such bad press.

Al Gore was another candidate who whose unkind media treatment was likely the key to his loss.

My conclusion is that replacing Extremism Index with a Media Bias Index might make the models more accurate.

(However, liberal academics don't like to admit a pro-liberal media bias. They don't see a liberal POV as biased; they see it as correct. It probably wouldn't help Brendan's career to emphasize liberal bias.)

Excellent article pointing out the limitations in election prediction models. The aptly named author, Sean Trende, lucidly explains ten assumptions implicit in all these models. He considers all of these assumptions questionable.

Trende argues, as Nate Silver recently argued, that one should not be too certain about the validity of any of these models.

@David in Cal, true that 'liberal academics don't like to admit to a pro-liberal media bias'. On the other hand conservatives do not like to admit a pro-conservative media bias in News Corp holdings either. People of all political beliefs want to read/watch news that reinforces their already held political ideaology and MSNBC/Fox News are there to fill that demand.

For fun check out this old Herman Cain clip singing Imagine There's no Pizza. Any analysis that does not include these singing abilities is missing the big picture.


What a great clip, JP! Thanks for the link.

You can tell that Cain has had some voice training or coaching. On words ending with an "er" sound, he properly sings "uh", so as not to get stuck in the "rrr". And, with dipthongs he holds the first vowel sound as long as possible. E.g., with a long i "ah-ee", he holds the "ah" through most of the note and only goes to "ee" at the very end.

I can't help but notice that in your chart comparing the 7 prediction models over 9 election cycles (for 63 total predictions), the result fell outside of the 90% confidence interval 11 times, or 11/63 = 17%... so Silver's fundamental critique that the modellers underestimate their error bars seems fair.

Although Silver's model falls outside of the 90% confidence interval twice (2/9 = 22%), even with his more generous error bars. :)

The comments to this entry are closed.