MyDD :: The polls DID predict a close race! (UPDATED)

The polls DID predict a close race! (UPDATED)

by ghost 2, Fri Jan 11, 2008 at 01:14:52 AM EST

(also posted at Taylor Marsh / Hot Topics Diaries)

Tuesday night was great. What a roller-coaster! Everone expected an Obama win. The question was: by how much?

That didn't happen. By now, all kinds of explanations and conspiracy theories are thrown out there, among them the Bradley effect and the Diebold effect. Various explanations, such as the high number of new voters, the difficulty of predicting who would turn out, and polling in such a narrow window of time and with so many events compressed together, are given by pollsters. But one thing is missed amid all the noise, and it could turn out to be a beauty.

The polls DID predict a tight race MOVING TOWARDS Hillary Clinton. The pollsters missed it not because the polls were wrong, but because their methodology obscured it.

Am I crazy? No.

Yesterday, Zogby mentioned that there had been a late surge towards Hillary Sunday night and Monday morning showing up in their numbers, but because they averaged the data over 2 days, that effect was washed out by the larger numbers for Obama earlier.

Wait, it gets better.

Look at the Rasmussen numbers on RealClearPolitics. Rasmussen interviewed lots of voters, and so had good sample size for each day. Now, look closely. Rasmussen reported the January 4th number by itself. Then, it reported the tracking poll on the next few days. If you do the math (very easy), you get the following daily numbers for Obama's lead:

January 4th (10)
January 5th (14)
January 6th (6)
January 7th (1)
Sample size is about 500-700LV daily.

So here it is. The highest spread according to Rasmussen was on the 5th, and came down very quickly. The number on the last day was close to what Zogby cited (+2). There was a very good movement in numbers favoring Hillary Clinton. If you extrapolate Rasmussen numbers to the 8th, the race is statistically tied, with a slight advantage to Hillary.

So: Yes, the polls moved very very fast from Friday to Tuesday. The Obama bounce was a very short pulse, and came down. Considering the very rapid sequence of events in that span of time, this is not at all surprising. The pollsters' methodology of averaging over 2 or 3 days completely obscured the speed with which the wave dissipated.

How do we prove that this is right? Well, let's ask Rasmussen to give out the numbers for each day and then see. They don't give them here. Anyone has access to the daily numbers in their tracking polls?

Other polls have such small daily sample size that the fluctuations may cover this effect. Yet, the daily numbers may be worth seeing.

UPDATE: I just thought of this, and it's amazing. Imagine that my math is correct, and the numbers above are Rasmussen daily numbers. IF Rasmussen had reported the rolling average for the 6th and 7th, and left out the 5th in the last number, their official (rolling averages, not daily) numbers would have been like this:

January 4th (10)
January 5th (12)
January 6th (10)
January 7th (4) (average of 6 and 1)

Wow!!!!!!!!! Then they could have been the only poll that saw the strong moving of numbers towards Hillary.

Tags: Election 2008, New Hampshire Primary (all tags)

Permalink :: 13 Comments :: Post a Comment

Re: The polls DID predict a close race! (2.00 / 1)

Lori has access to Rasmussen's premium service and probably would be able to access the individual day numbers.

I don't quite understand why Zogby and Rasmussen published the MONDAY findings by themselves that evening (or Tuesday morning)? a 4-day rolling sample is worthless in that type of environment. Should the day-before-primary-day finding not be pinpointed and shown? Hopefully they'll do better next time and change their system. Show the damn ONE-DAY numbers right before caucus/primary day, and let us be the judge whether it is from a big enough sample to make any type of judgement calls about it. Now they are just trying to minimize the damage by claiming that the rolling averaging obscured strong Clinton momentum. Then SHOW US when it happens, so we buy it (and take it into consideration) on election day.

by georgep on Fri Jan 11, 2008 at 01:24:52 AM EST

Re: The polls DID predict a close race! (2.00 / 1)

should have been:

I don't quite understand why Zogby and Rasmussen did not publish the MONDAY findings by themselves that evening (or Tuesday morning.)

by georgep on Fri Jan 11, 2008 at 01:26:04 AM EST
[ Parent ]

Re: The polls DID predict a close race! (none / 0)

George,

Could you ask Lori about this?? Thanks.

I have yet to see what [Obama] has done to take the highest office in the land. He is no Martin Luther King. --Helen Thomas

by ghost 2 on Fri Jan 11, 2008 at 01:53:55 PM EST
[ Parent ]

Re: The polls DID predict a close race! (none / 0)

well, one-day numbers jump around a lot (larger MOE), and they probably don't want untrained people to interpret random fluctuation as real movement.

it's a tradeoff. multiple-day samples have a smaller MOE because you're increasing the sample size, but less precision chronologically. think of it as the Heisenberg uncertainty principle of polling- you can know either the exact percentages or the precise time, but not both at once ;-)

by campskunk on Fri Jan 11, 2008 at 01:33:19 AM EST

Re: The polls DID predict a close race! (2.00 / 1)

Good one.

That's a catch phrase with a lot of potential!

I have yet to see what [Obama] has done to take the highest office in the land. He is no Martin Luther King. --Helen Thomas

by ghost 2 on Fri Jan 11, 2008 at 01:37:22 AM EST
[ Parent ]

Re: The polls DID predict a close race! (2.00 / 1)

So, the way to deal with that is to publish BOTH numbers on the morning of the caucus. The overall 4-day roll, and the final day findings, with the caveat that one-day samples can be more volatile, less precise. At least we would have seen a trend or movement towards Clinton, as that is what actually occured. Hope they learn from this and publish both in the future. What do they have to lose? The one-day sample comes with the caveat to treat it with extra care, but if the overall movement turns out to be correct, it will serve to cover their butts better.

by georgep on Fri Jan 11, 2008 at 01:50:45 AM EST
[ Parent ]

Re: The polls DID predict a close race! (2.00 / 1)

The best way is to simply have a larger sample size and poll each day. Then they can catch the movements in the poll.

These guys are in business, they should know this. Events were moving very fast. There was a debate on Saturday night. What was Rasmussen doing mixing Saturday numbers with Monday?

Perhaps there weren't ready for the craziness happening from Friday to Monday. Why not?

As an aside, this is also a testament to the amazing tenacity of Hillary Clinton. Obama had planned it like that. He is a get momentum/ get press candidate. He had tons of money, and the compressed schedule favored him.

Hillary just didn't give up. She is a fighter, always.

I have yet to see what [Obama] has done to take the highest office in the land. He is no Martin Luther King. --Helen Thomas

by ghost 2 on Fri Jan 11, 2008 at 02:02:15 AM EST
[ Parent ]

Re: The polls DID predict a close race! (none / 0)

the problem is that public pollsters are essentially in an adversary relationship with their customers, the public and the campaign partisans who want to focus in on the part that makes their candidate look the best. if the pollsters give people a choice, the partisans will choose the part they like.

it also boils down to money- polls with larger samples cost more to do. you also run into problems in small states like new hampshire in doing a sampling without replacement model (never call someone after you've called them one time, to eliminate test-retest effects) in that you run out of people to call if you do huge samples. they called thousands of people to get those 800 completed calls, i bet. do that every day, and you're scraping for new phone numbers pretty quickly in a state with a small population.

it's not cost-effective double your cost to reduce MOE by a percent or two when there are larger biases you can't control (sample bias due to only sampling land lines, for instance) still affecting the outcomes. by analogy, why by $800 speakers when you're stuck with a $100 radio? it's a waste of money.

by campskunk on Fri Jan 11, 2008 at 11:07:56 AM EST
[ Parent ]

Re: The polls DID predict a close race! (none / 0)

Well, yes, but in this case both Zogby and Rasmussen HAD last-day polling data available that apparently suggested a major Clinton movement (if you look at a graph of those dailies, you'll get about a 2%, 3% Clinton lead for the next day, which is exactly what happened on voting day.) They chose not to publish the findings of that one-day sample, even though it must have been apparent to anyone looking at the data from that one day only that the poll was turning for Clinton in a very dramatic, strong way. That seems like an omission that made little sense, in retrospect. All we got spoonfed publically were double-digit leads based on a 4-day sample, when a closer look at more recent data ALREADY showed a tie.

But, I am not complaining, since this way it made it even more historic, more "take that, frigging media pundocrats"-ish. I was just wondering why Zogby and Rasmussen chose to keep the last-day findings, which so strongly showed a major Clinton surge, hidden, unpublished.

by georgep on Fri Jan 11, 2008 at 12:04:36 PM EST
[ Parent ]

Zogby vs. Rasmussen (none / 0)

1) Rasmussen should have averaged 01/06-01/07, and not included 01/05, for their last report.

2) Zogby came out and said what the internal data looked like on Sunday and Monday. Rasmussen hasn't said peep about their daily data, and their daily sample size is bigger. Why not? Meanwhile, all kinds of conspiracy theories are flying around.

3) my guess for why? in item 2. Rasmussen will have serious egg on their face, and be laughed at by others, if they did.

4) All of this brings attention to the fact that pollsters can have a sample size of 330 (like CBS) for environment like this, and be taken seriously by press (who doesn't know crap about anything, but pretend to) and other pollsters.

I have yet to see what [Obama] has done to take the highest office in the land. He is no Martin Luther King. --Helen Thomas

by ghost 2 on Fri Jan 11, 2008 at 01:52:16 PM EST
[ Parent ]

Re: The polls DID predict a close race! (none / 0)

Camp,

Hope you see this. I don't know how I didn't think of this.

Go back in time and imagined Rasmussen reported 01/06-01/07 for the last report. Then rolling averages would have been:

4th 10
5th 12
6th 10
7th 4 (average of 6 and 1).

Its methodology would have been proper AND it would have validated the results!!!!!!!!!!!! It would have predicted a massive movement towards Hillary!!!!!!!!!

TALK ABOUT EGG ON THE FACE!!!!!!!! Could someone please ask them for the daily numbers?? somebody??

I have yet to see what [Obama] has done to take the highest office in the land. He is no Martin Luther King. --Helen Thomas

by ghost 2 on Fri Jan 11, 2008 at 02:01:38 PM EST
[ Parent ]

Re: The polls DID predict a close race! (2.00 / 1)

Zogby, I understand. Look closely at RealClearPolitics. It's really very interesting. The polls in general had very small sample sizes, which make them almost worthless with fast moving numbers. CBS news had a poll in there with 323 interview over 2 days.

But Rasmussen is interesting. As I said in diary, they have very large daily numbers. I don't understand why their polls have such huge sample sizes. But the trend of their daily numbers are really neat.

I am really keen to know their actual numbers.

I have yet to see what [Obama] has done to take the highest office in the land. He is no Martin Luther King. --Helen Thomas

by ghost 2 on Fri Jan 11, 2008 at 01:36:33 AM EST

Re: The polls DID predict a close race! (2.00 / 2)

Interesting analysis. Thanks for looking behind the numbers a bit. The bottom line--women and Democrats came home for Hillary at the end.

by markjay on Fri Jan 11, 2008 at 07:39:19 AM EST

You are not logged in.

In order to post a comment, you must be logged in. If you have a member account, please log in to comment.

If not, you can make an account right here. It's quick and free.