情趣用品,情趣用品,情趣用品,情趣用品,情趣用品,情趣用品,情趣用品,情趣用品,情趣用品,情趣用品,...

Hmm....

I built a simple Bayesian model, in which the "Macro" variables (incumbency, economy, etc.) predict the winner of the election as a prior Gaussian distribution with mean M and variance vM.

I then consider a poll taken t days before the election, with result P and variance vP + t*vT.

I get the following expected election result:

Most likely election result =

M * (vP + t*vT) / (vM + vP + t*vT)
+
P * vM / (vM + vP + t*vT)

Looking at the data from their paper, we can estimate that

vP / vM = 1/4
vT / vM = 1/112
vT / vP = 1/28

This corresponds very well to the time-evolution of their regression coefficient, and gives us an idea of the rate at which polling information decays. The amazing thing is that it suggests that the weight of the poll should be cut in 1/2 after four weeks, which is almost exactly what Nate uses for this site. However, it suggests that after 8 weeks, the weight of the poll should be 1/3, not 1/4.

Another interesting result is that the total variance of the poll equals the total variance of the Macro model about 12 weeks before the election. Before this, the model is more accurate, and afterwards the polls are.

Determining the intercept is a little bit trickier, but I think the data actually suggest that M is best interpreted as the expected two-way vote share of the incumbent, minus 49. If this is the case, then M should be roughly 3, meaning that the prior expected vote share for the incumbent party is about 52%. This is a very rough estimate, as the intercept data in their paper is a little weird. I suspect that this variable is more susceptible to short-term news events, such as conventions and debates.

I think the real lesson is that although this "Macro" models may be useful initially, there is still a lot of variance in these models and polls quickly become more important. Remember that in 2000, the models predicted that Al Gore would win in a landslide.

Forgot to add that the 1984 (Reagan incumbent) ele...

2008-07-09T17:32:26.192-04:00

Forgot to add that the 1984 (Reagan incumbent) election was also of the "sustained and/or ever-widening lead" type. That makes it seven of that type, out of the last fourteen

@ counsellorben: What a great paper you have linke...

2008-07-09T17:29:05.821-04:00

@ counsellorben: What a great paper you have linked to! Great information content, great analysis (in particular for somebody who has not followed previous elections in detail). Thanks a lot.

Nate: When looking at the paper's diagrams on poll movements for previous presidential elections, it appears that the majority of past elections have indeed displayed the 'regression-to-mean' pattern that you are applying to your projection. However, at least five of the last fourteen elections show a different pattern, which may be either described as "ever-widening lead', or as 'permanent advantage', or somewhere in-between those two. These were the elections in 1956 (Truman incumbent), 1964 (LBJ, post JFK), 1972 (Nixon incumbent), 1984 (Reagan incumbent) and 1988 (W. Bush, quasi-incumbent). The respective Wikipedia article suggests that the 1996 election (B. Clinton incumbent) has as well been of the 'permanent advantage' type: "Throughout the run-up to the general election, Clinton maintained comfortable leads in the polls".

In other words: You either have the (quasi-)incumbent advantage, in which he keeps a sustained and/or expanding lead, or a 'change' election that displays a 'regression-to-means' pattern and becomes a nail-biter in the end. While I was a bit sceptical about your 'regression-to-means' initially, after having looked at the data I now believe it is adequate and justified for 2008. However, there is no justification at all to apply an additional 'incumbency' correction to the data, in particular since McCain does not display any of the characteristics of an incumbent (presidential experience/ track record, power to implement pre-election etc.).

[This also shows how extra-ordinary the 2004 election has been. It has been the only election in the USA since pre-election polling exists, which the incumbent only won narrowly. All other incumbents were either defeated or won convincingly.]

PPP is out with MO poll, showing McCain up!!!!!!!!...

2008-07-09T16:56:23.727-04:00

PPP is out with MO poll, showing McCain up!!!!!!!!!!!!

When the economy's crappy, the incumbent party los...

2008-07-09T16:25:41.693-04:00

When the economy's crappy, the incumbent party loses. It's pretty much that simple.

thanks anon at 2:30, I guess I just was not seeing...

2008-07-09T16:24:22.116-04:00

thanks anon at 2:30, I guess I just was not seeing the current trend as being either pro Obama or McCain, kind of stagnant lately, but that may just be a lack of polling. Can there be a trend absent polling?

If you're going to try to incorporate all sorts of...

2008-07-09T15:55:53.887-04:00

If you're going to try to incorporate all sorts of candidate-specific variables like party ID, incumbency, etc. and base your coefficients on historical elections data, you're likely to end up with something a lot like Ray Fair's presidential election predictor (at www.fairmodel.com). He's a Yale Economics prof who uses some pretty complex econometric modeling to isolate a set of variables -- some political, but mostly economic -- that he uses to predict the 2-party vote share for each candidate.

In my opinion, it's better to leave that kind of stuff to him and to devote this site instead toward detailed analysis of polls and the kind of predictive value they may have. Of course, there's some overlap if you're trying to determine whether some variables affect the predictive power of polling data, but as you said, the sample size problem makes it really difficult to say anything definitively about their impacts.

Higgly:trend adjustments work for individual polls...

2008-07-09T15:30:33.009-04:00

Higgly:

trend adjustments work for individual polls.

A poll was taken in February. The trend then favored Obama. The trend adjustment gets the Obama numbers down.

A poll was taken in April. The trend then favored McCain. The trend adjustment gets the McCain number down.

It works out that we are still in an environment where trend adjustment, on the whole, tends to help Obama. But this may well change in a few weeks.

I note you are still making trend adjustment numbe...

2008-07-09T15:27:19.713-04:00

I note you are still making trend adjustment numbers in all states. How does this work. It would seem the new polls in single states would be self adjusting to trends within the state and not blindly following the bump Obama got post primary. Or is the Trend adjusment something I just do not understand.

I guess the question really is why do trend adjustments still take place after new polls take place after the adjustment in trend was made? Or is this trend adjustment going to be a feature based only on new trends in National Polling? In other words now that Obama is down 1 to 2% this week in the trends will there be a negative trend adjustment to new polls to account for this?

the other really interesting thing in this MO poll...

2008-07-09T15:19:23.342-04:00

the other really interesting thing in this MO poll. 67% support drilling for oil, but only 54% believe this could improve the cost of gas. Interesting that 13-14% feel that drilling for oil is appropriate just to reduce dependancy by some margin, regardless of the price of gas.

I prefer Obama to Hillary but she was stronger ele...

2008-07-09T15:15:22.465-04:00

I prefer Obama to Hillary but she was stronger electorally. McCain was dead in Pennsylvania and Ohio, Hillary was winning Florida, she'd bring Arkansas and was very competitive in West Virginia. I know Barack is stronger in Colorado and Virginia but Hillary would have gotten a bump from winning the nomination too. She'd be up by 8-9 points right now. about 4 points better than Obama. Obama is a much riskier candidate. He can win, Hillary would win easily.

While everyone appears to know the Rasmussen MO po...

2008-07-09T15:13:51.716-04:00

While everyone appears to know the Rasmussen MO poll exists with McCain plus five, some of the answers in it are suprising as well. For one Bush got 48% fair or better approval raiting in the state. With 35% good or better. Still 52% poor, but drastically different than the national average, and interesting giving this Quibble string.

Second 50% stated Obama is too inexperienced to be president. While only 28% felt McCain was too old.

You should all take a look over there. Its only 500 voters so probably not great weight, but interesting none the less.

correction to my @:59should be "maintaining the cu...

2008-07-09T15:12:49.661-04:00

correction to my @:59

should be "maintaining the current polling or *not much worse* by Election Day"

(point being that the polling may go down by 4-5 points and Obama is still likely to win, which is why Nate's model believes that one can say with 75% or so that the current polling will "hold up" to the extent of an Obama win).

Nate Silver: If you want to look at something like...

2008-07-09T15:11:26.208-04:00

Nate Silver: If you want to look at something like "Democrat challenging Republican quasi-incumbent in wartime with crappy economy", your dataset gets down to zero pretty quickly.

True.

But if you want to look at something like "Does the challenging or incumbent party do better in times of a crashing economy, quickly rising food & fuel costs, and multiple crises in the Middle East?", then the 1976 and 1980 elections might be a good place to begin.

.

Thank you, Cugel. I was beginning to wonder if I w...

2008-07-09T15:07:56.298-04:00

Thank you, Cugel. I was beginning to wonder if I was the only person who remembered that McCain isn't the incumbent. Nate says it in his post.

(Although, I guess the Obama campaign must be getting somewhere if the brainy commenters of 538 are writing as though a McCain election would in fact be Bush's third term...)

(on a completely unrelated note, are any of you coming here to Austin next week? Would love to meet some of my new pals--plus, this is my hometown. While we have no baseball, I'm sure we can get together and obsess about the weight that just being a Yankee should be given in All-Star voting.)

Couldn't this just as easily be attributed this to...

2008-07-09T15:06:02.482-04:00

Couldn't this just as easily be attributed this to name recognition or "the-devil-you-know" phenomena?

Voters make last second decisions at the polls simply to fill holes in their ballots. I'd like to see the data that suggests that this isn't what is really happening here.

McCain's numbers may improve by November but I dou...

2008-07-09T15:05:54.980-04:00

McCain's numbers may improve by November but I doubt you could attribute this to an incumbency effect, just because he's in the incumbent's party. Given the all-time low approval rating of the incumbent President, any ties to the incumbent are probably more harmful than helpful. I'm not sure we have really good historical examples of this, an incredibly unpopular term-limited President with two non-incumbents. Does the non-incumbent in the sitting President's party really get a boost in November?

Comments on FiveThirtyEight: Politics Done Right: Quibble

酒店經紀人, 菲梵酒店經紀, 酒店經紀, 禮服酒店上班, 酒店小姐兼職, 便服酒店經紀, ...