jhm:

I thought that I learnt in Statistics class that a major problem with the Truman call was the income disparity between those with telephones (got polled) and those without (left out of survey). Was this not the case, or if it was, do file that under better "likely voter" ID?

Posted on January 10, 2008 7:22 PM

 NB:

Does anyone ever re-poll the people polled before the election, and ask them how they in fact voted? Whether such a post-election poll matches the election results or the pre-election poll, either way would be highly informative. Why don't people do this?

Posted on January 10, 2008 9:47 PM

 Mark Blumenthal:

jhm: I am really not an expert on the debate over the 1948 polls, but you learned right. See this NY Times letter to the editor that summarizes the lesson you remember. This 10-year-old AP story, on the other hand, recounts that the pollsters themselves held to the idea that the big problem was that they stopped interviewing too soon.

NB: Yes, those sorts of polls are done from time to time, and I'd wager that some are probably being done as we speak in New Hampshire.

Posted on January 10, 2008 10:20 PM

 Andric:

I don't understand why pundits and surprisingly, Mark Blumenthal, have to go all the way to 1948 to find flawed polls, choosing to skip 2000 and 2004 as if everything was a-ok with polling accuracy during those primaries. Nobody predicted McCain's 18 pt. victory over Bush Jr. in New Hampshire, as the Daily Howler noted yesterday, Washington Post's David Broder wrote this in 2000:

BRODER (1/31/00):

On a day when the five Republican hopefuls and the two Democrats raced from television studios to town meetings before joining the rest of America in watching the Super Bowl, Texas Gov. George W. Bush and Arizona Sen. John McCain, deadlocked in the latest surveys of the GOP race, took heart from enthusiastic crowds and continued their polite sparring for advantage.

This was written 1 day before the NH GOP primary.
http://dailyhowler.com/dh010908.shtml

Posted on January 10, 2008 10:37 PM

 puzzlegal:

Because the "lesson" he's referring to isn't "polls are sometimes wrong". Read the article. The lesson is "we need to share our methodologies so we can analyze what went wrong". I don't think that's happened recently.

Posted on January 11, 2008 7:06 AM

 Nick Panagakis:

Re: 1948

As you indicated, in addition to methodology issues there was also the issue of timing. I can�t find a citation for this but I believe polling was suspended about one month out from election day. Can you confirm this? This may have fostered a mindset that Dewey would win going into the election, an expectation. Do you know when pollsters stopped polling?

http://www.historybuff.com/archives/view.cgi/11-03-1948-fd-cdt

Re: the Tribune story above.
If you read the story in the right hand column, there were no poll references. So the headline had nothing to do with polls although expectation was probably a factor. What you will see is that an otherwise reputable political reporter was getting much of his election outcome information from Herbert Brownell, Dewey�s campaign manager.

See:
http://www.historybuff.com/library/refdewey.html

There were eleven Tribune editions that night. The first is the one remembered but later editions said otherwise.
"DEWEY DEFEATS TRUMAN" -- (City and Mail editions)
"DEWEY HOLDS NARROW LEAD" (Sports Final -- City edition)
"IT'S TRUMAN BY 150,000" (Flash 4 Star Edition)

Also note: �One of the November 4, 1948 editions of the Chicago Daily Tribune showed a photo with the following caption: "The eleven editions printed by The Tribune yesterday to keep readers abreast of the election results." All eleven editions were shown in a fanned out format.�

There are modern day examples of such headlines, these from 2000:
The St. Louis Post-Dispatch: BUSH WINS A THRILLER;
The Miami Herald: BUSH WINS IT.
The Boston Globe: IT'S BUSH
New York Times fudged a bit: BUSH APPEARS TO DEFEAT GORE.
Source: CJR, Nov/Dec 2001

Nick Panagakis

Posted on January 11, 2008 12:14 PM

 davidgmills:

Always the Poll -- Never the count.

Try this NH coincidence on for size:

Optical Scan Ratio between C & B
Clinton 91,717 52.9507%
Obama 81,495 47.0493%
Total 173,21

Hand Counted Ratio between C & B
Clinton 20,889 47.0494%
Obama 23,509 52.9506%
Total 44,398

Exact inverted ratios to .0001%. Truly amazing!

Still you pollsters want to maintain there is never any computer fraud.

Posted on January 11, 2008 5:31 PM

 davidgmills:

Correction Ratio between C & O, not C & B.

Posted on January 11, 2008 5:33 PM

 davidgmills:

Should have provided the link:

http://www.opednews.com/articles/opedne_bruce_o__080110_obama_clinton_3a_remar.htm

What I would like to know when I see this post about the Truman/Dewey fiasco is whn will the polling community stop beating themselves up? This community is so spineless, it makes the Democrats look like vertebrates.

Pollsters, it is high time to stand up for your profession.

When we get elections where we know the count is reliable and then you get your predictions wrong, well you can crawl on your bellies then. But until such time as we get reliable, auditable counts, get off your slithering bellies and stick up for yourselves.

No more of: "We got it wrong this time because ..."

Posted on January 11, 2008 8:36 PM

 just pixels:

The inverted ratio davidgmills describes is intriguing (if correct). Obviously larger localities use machine counts, while smaller ones hand count so the ratios would be distinctly different for lots of reasons.

Paper versus computer count? In the olde days they didn't stuff the ballot box with memory chips. Beware of too much trust in paper alone -- especially when there's a copier nearby.

Posted on January 12, 2008 2:41 PM

 jawbone:

Seconding commenter's request (Andric | January 10, 2008 10:37 PM) for analysis of the 2000 Republican primary results, which, as Bob Somererby points out, was another primary without incumbents (altho Al Gore was VP on the Dem side).

Polls said very close race--McCain blew Bush our of the water by 19 pts, iirc. Have polling techniques changed radically since then?

How was that polling error, or whatever it was, explained after thorough review of the polling internals?

I listened to WNYC this morning and hoped to hear something about this result--I may have missed it if you brought it up...?

Posted on January 14, 2008 1:32 PM

 jawbone:

Correction: Bob Somerby--at dailyhowler.com

Thanking you in advance, Mr. Blumenthal, for your attention to these requests.

Posted on January 14, 2008 1:35 PM