the whole deal with that CNN/Gallup poll

[ButterflyBlood]

with all the screaming here about how either Kerry is doomed or its a media conspiracy against him, here's a summary of what I heard non-partisan polling expert say on it.

Gallup has a good methodology. However, this methodology is designed for predicting results right before the election. Determining a "likely voter" 3 months before the election is not an easy task. That's why in 2000 Gallup got the results before the election pretty good, but had Dumbass leading by 13 points at one point.

Another point is that Gallup uses one of the "purest" samples possible. They basically take a totally random sample. Rasmussen deliberately attempts to get a sample equivalent to party registration levels, and Zogby tallies the demographics polled, and adjusts the results a little "weighting" certain demographics depending on how well they were represented in the sample to their actual numbers. Gallup basically takes what they get. One advantage this has over the other methods is they can't predict turnout well, there's no way of knowing whether the exact percentage of parties or demographics in the poll will be the actual turnout. However, this means that they often get a sample which is just plain bad. For example, if you flip a coin 10 times in a row, and keep doing this, the average number of heads in your flip sequence will almost certainly be around 5 if you do it a lot. However you probably will almost certainly also get sequences with 7-8 or 2-3 heads. Those sequences are not representative of what normally is gotten.

so let's settle down, and wait for what the other polls say. But odds are this isn't an omen of Kerry's doom, or a plot against him.

[GainesT1958]

1. Gallup was a "far outlier" in North Carolina earlier in July...

Because of the methodology your expert describes. That methodology is probably why this poll will prove to be a far outlier, too.

B-)

[Gloria]

2. Gallup was sold years ago, so it is not the same outfit I worked for

back around 1980, when old man Gallup still walked the halls of Bank St. in Princeton. Back then, it still had a largely academic bent and was small and had a sense of pride.
When I was there, the Presidency was alternated between Leonard Wood and Andy Kohut, the guy you will see on TV all the time. (Gallup, Sr. was still around, but the two guys did the real operations.) Andy went on to found his own outfit, and then went on to Pew.

Just because I worked for Andy, I would tend to trust him. One never knows though, over the years, things change. But, having come from the original Gallup roots, I would hope that he's still doing good work...

Samples of "likely voters" are more accurate in the fall. Gallup continues to use them because they started out with this method and keep using it for continuity's sake.

Back when I was there, they were just starting phone interviewing and the Presidential polling was done by in-person interviewing as part of an ongoing, weekly "omnibus" survey, with all sort of client questions lumped into one survey.

As far as this CNN/Gallup poll: it probably isn't a plot. However, when CNN makes this a "news item" and then making their own poll a continuing subject of biased commentary, etc.-- it can and is being used in ways that go far beyond a simple poll to report on.

[oscar111]

3. Gloria, super site you have

what a lot of work you do! Keep it up!
Nice addition would be some note as to what online radio is receivable by webtv. I only find one, the AZ or NM gal, name oddly spelled Mary, forget just how spelled. "alive and kicking" is her motto.

Others , including white rose archives , all unusable to webtv.

[Gloria]

8. Thanks for visiting it...in all honesty, except for the WMW I've

been very remiss in updating the radio-internet section. I started a revamp but got distracted. SIGH......

My mother has WebTV. It can do the BBC, I know that. But not much else. Even if you get something to stream, WebTV cuts it off in 10 minutes, unless you click to keep it going.......But not much else.
Sometimes I wonder if liberal radio streams are being deliberately blocked.

BTW, I think you're thinking of Meria Heller out of AZ...........

[Edgewater_Joe]

4. We Already KNOW It's An Aberrant Poll

Both Zogby and Rasmussen show Kerry up around 8 points -- and an MSNBC poll that was broadcast on Saturday said the lead was 13 points.

What I found odd was their admission that they based the story of "no bounce" on 'likely' voters -- a smaller sample by something like 150-200 respondents from "registered" voters, which showed a lead for Kerry.

I worked for a few years at the U. of Chicago's National Opinion Research Center, and one lesson I got is that the smaller the number of responses, the less accurate the result is. Based on that, I am deeply suspicious of this poll since they admit they are using a smaller sample that gives them a completely different result than the larger sample -- and aren't working too hard to clarify the polling distinctions, to put it mildly.

It does seem like the storywriters, at least, are bending over backwards to make this look more pro-Bush. Now, whether it's because AOL/Time-Warner wants Bush in there or that ratings will be higher in the fall if the perception is that the race is thisclose (and I suspect more of the latter than the former) I can only guess.

But as Hank Hill might put it, somethin' ain't right with this ...

[depakid]

5. If Gallop/CNN simply takes what they get

then the question becomes where do they get it? That's where multistage clustering comes into play. you can and will get wildly differing results, depending on where the individual "clusters" are located.

Also, with true randomization, the variations among individual subjects cancel each other out- provided that the sample size is large enough (if you flip a coin enough times, the results will very closely approach 50%). Statistical methods take into account chance variations- and that's what the margin of error does.

Now, I'm not certain how close to the election the 13% Bush lead was, but assuming that the margin of error was + or - 4 points, a swing of 13 points is pretty extreme- especially for a purely random sample. Something like that a week or two before the election would suggest that more the swing had to do with factors other than turnout. A month or two out and the matter becomes much more clouded.

Obviously, this particular poll means nothing in isolation and the people crowing over it on TV are just doing what they always do- accentuating any negative news that they can find to further their own agenda.

[grasswire]

6. even old George Gallup..

...was a conservative, though. When I worked for a very conservative non-profit organization in D.C., Gallup was a friend of the management. I recall him and his daughter coming to meetings.

[Gloria]

9. Yes, and his son, George, Jr., is still running his little shop (although

I think it's not really independent anymore) out in Hopewell, NJ doing religious polling (at least he was still running it a few years ago.) I always thought George, Jr. was a bit weird, even back when I knew him.

[tritsofme]

7. This is what they recently said about turnout

All signs point to a higher election turnout this fall. To whose advantage would that be?

It's not absolutely certain that turnout this year will be higher than usual, but it's certainly a possibility given the high levels of interest we have seen so far. Generally speaking, our working assumption is that high turnout helps the Democratic candidate, given that the GOP base is likely to turn out regardless. But at this point, Gallup's modeling shows little difference between turnout assumptions of 50% and 55%, and only a minor difference if turnout gets as high as 60%
<snip>

This poll seems like an outlier to me, and I really don't understand why it was so urgent to release it today instead of polling another day and waiting until tomorrow to release it like they usually do.