There is no new state polling today, but I have rolled forward the numbers based on the continual good performance for Barack Obama in the national trackers and the latest several editions of the Economist/YouGov poll, which is conducted on a weekly basis but updated on the Economist's website only occasionally. The downtrend in Obama's numbers since the July 4 holiday has now completely flattened out, though we will need a lot more information to determine whether it has reversed itself. I also fixed a small bug in the model that was causing the state-by-state implementation of the trendline adjustment to behave oddly.
A few miscellaneous items:
- Perhaps the most comprehensive study I have seen on the Bradley/Wilder effect so far in this cycle was conducted by Harvard political scientist Dan Hopkins and is available for your perusal here (PDF). Hopkins' conclusion: the Bradley Effect may have been real in the late 1980s and early 1990s, but does not not appear to exist any more, and in fact there is some (weak) evidence of a reverse Bradley effect. This finding is broadly consistent with the (far less rigorous) studies I have done on the issue.
- I will be on the Brian Lehrer Show on WNYC Radio tomorrow morning at approximately 10 AM Eastern.
- Karl Rove is an occasional reader.
- Speaking of conservatives, I saw Reihan Salam on David Gregory's show the other day. He is perhaps the right's best answer to Rachel Maddow -- someone who has a strong point of view without being predictable about it -- and is deserving of more airtime.
Sunday, July 27, 2008
Today's Polls and Miscellaneous Thoughts, 7/27
-- Nate Silver at 4:31 PM
Labels: bradley effect, today's polls
128 comments
I think the Karl Rove link needs fixing.
(And of course he's reading.)
I mentioned this in another thread, but I'm gonna drop it here as well.
Nate: When I see your page, it screams "Link to me on Facebook." Any plans for an app? I'd love to have a updating graphic for my page.
It would be good if there could always be a list of which polls are due to come out in the next 24 or 48 hours, so we know what's coming up. At the moment I keep logging on to fivethirtyeight.com for new polls and then somedays there aren't any. I'd like to know in advance, if that's possible.
Donate to John McCain!
Your Country needs you!
Donate online now!
Help your country!
--- CLICK HERE ---
I don't see how anyone can support Obama's income redistribution plans and raising taxes as well.
A vote for Obama is a vote for a further destroyed economy.
Maybe that someone doesn´t make over 250k and actually gets bigger tax cuts than in McCain´s plan.
I don't see how anyone can support the Bush/McCain economic policies.
We have already seen how well neocon economic policies work. They are disastrous.
McCain supporters support dishonor and yet more tax breaks for the rich. Follow John McCain's plan for financial success: Marry a rich beer heiress!
A vote for McCain is a vote for Bush, Rove and Phil Graham.
Actually some some study showed that US economy was on average much better under Democrat rule. IIRC since WW2 the only republican president ahead of any democratic president in that scale was Nixon ahead of Carter.
1) KARL ROVE WAS WRONG, BECAUSE THAT DOES NOT REPRESENT CURRENT POLLING, BUT NOVEMBER PROJECTION
2) YOU NEED TO FIX THE LINE THAT SAYS MCCAIN LOSES OH/MI, WINS ELECTION.....IT IS LINKING TO MCCAIN LOSES OH/PA AT LEAST FOR THE DENOMINATOR, PERHAPS NOT THE NUMERATOR............106 ITERATIONS OUT OF 10000 MCCAIN WINS PA WHEN LOSING OHIO....WE KNOW THIS BECAUSE OF WHAT IS STATED BELOW: OBAMA WINS OH WHEN LOSING PA--106....5801 (mccain loses oh)-5695(mccain loses oh and pa)= 106
What the LOESS curve does is take discontinuous change and turn it into continuous change. I believe that if you looked, you could find a tool that smooths out the noise in a different way. Your model should look for instantaneous adjustments - mini-landslides in either direction, whose effects are permanent - and assume a 1/f ("fractal") frequency/size distribution.
"We have already seen how well neocon economic policies work. They are disastrous."
The neo-cons have enough to answer for without blaming them for Bush's economic policies. The neo-con position centers around foreign policy and domestic security, and is relative neutral on economics. While some hae indeed gone whole hog to the right (like the switch of some Trotskyites from the 30's to rabid right wingers in the 50's) others don't really care so much. People like Doug Feith have always been hard shell Republicans; so was his father. The fact that their foreign policy is the same as the neo-cpns does not make all neo-cons hard shell anti=-welfare sate types.
Nate, have you looked at the internals of the YouGov polls at all? I just took a look at the most recent one, and the sample seems crazy. There were 921 individuals in the sample and, 768 of them where White for 83% of the population.
And going back, it's pretty similar, they all have about 83% Whites in their sample. That is way too high and makes me think it's completely useless as a poll.
Pollingreport.com has a new result (7/27) from Research2000: 1100 LV, Obama 51%, McCain 39%. This parallels Obama's upswing on Gallup and Rasmussen.
In latest YouGov poll online, out of 921 respondents the N's are 768 whites, 93 blacks, 60 Hispanic.
If these numbers were the unweighted numbers, well that happens. But in a weighted sample there should be roughly 11% blacks and 12% Hispanics, plus perhaps 3% Asian Americans and others. (Weighting could vary depending on whether there is further selection for likely voters.)
So the question is: Are the data weighted in the analysis but not in the crosstabs/toplines that are shown? I suspect so. Why?
The authors of the survey are certainly very reputable researchers: "Survey was designed by Samuel Popkins of the University of California, San Diego, and Douglas Rivers
of YouGovPolimetrix and Stanford University."
But the use of weights to correct for the undersampling of African Americans and Latinos really ought to be confirmed by others.
It's time to give up, conservatives.
No one likes John McCain and you won't be able to instill enough fear and hatred to defeat Obama.
Change is coming.
Nate,
I am glad you have incorporated the Economist/YouGov national polling.
I analyzed some of it from linking through the Pollster site earlier this weekend.
Quibbles about some data & maybe a small size [1000], but that is more than offset by the frequency [weekly] and consistency, and some methodology choices they have made.
I prefer the high rate of UNDECIDED that they allow to flow through. At this stage, pushing someone to 'lean' creates noise & unreliability.
The trends on Economist/YouGov make since for the lasy 6 weeks. Both candidates slightly up, but they draw their increases from the 'NOT VOTING' category.
Keep following them for the trends. There is also loads of non-push opinions & favorables available in their data to mine on a weekly basis.
THAT info is invaluable to anyone trying to analyze or project for this election even if not available at the micro-state level.
Thanks, Nate !
Anon@5:40,
You must be speaking to stubborn Republicans who've sabotaged conservative ideals....
One of these days, you'll no longer be able to use the "Blame Bush" and "McSame" mantra as they'll be long gone...so you'll have to come up with some constructive or original argument, which for liberal idiots is impossible. Then, YOU won't be able to instill enough fear and hatred to whomever it is restores true conservative ideals to this great nation.
Don't worry...change REALLY IS coming...it likely won't happen until at least 2012 or 2016 after the Obama presidency has further neutered this country of any social or economic relevance.
I have a bit more to add to the Loess discussion
Judas_priest I think said that the only measure on the correctness of smoothing techniques we adopt is how well they fit the data; Of course this is right in a sense but the qualification is important.
To fit the data perfectly, we could after all have had the set of straight lines joining all points - the unsmoothed spiky graph. Why do we avoid this? Not because it's unwieldy but because we *independently believe it to be false* - we assume that the underlying dynamics rule out such a spiky sequence which therefore MUST be noise to be corrected.
Any reduction of observations to an underlying curve depends on assumptions concerning underlying dynamics. And statistical tools incorporate, implicitly, such underlying assumptions.
Let me note that this entire field of statistics emerges from the analysis of astronomical observations, where the goal w