Hello guys: thanks for the great responses. I am collecting them and sending them back to the List for the benefit of all (I'm including the sender's name in case anyone wants to follow up on some issue directly with them). Discussions such as these, are great for students. If they can raise such an interest amongst ourselves, who see the subject every day, imagine what they can do for them! Cheers and thanks again/jorge
Sent: Sun 4/18/2004 1:40 PM
Subject: surveys and rooselelt's 1936 election
Hi, guys: need some help with this annecdote that, perhaps, someone there remembers.
Per my recolection, Mr. Roosevelt was the underdog candidate in the polls for the 1936 election, which, needless to say, he carried hands down. The survey, as I recall, was done by phone and, of course, at the time this artifact was a middle/upper class device. Therefore, the poll accuracy reported on the expectation of this social class -not of the entire nation. I have been told this is erroneous; that the poll was done by a magazine and that GAllup predicted this result via their phone poll. Please advise.
Check out Freedman Pisani Purves. Statistics. At least the earlier
editions dealt with this.
I believe the Truman Dewey election was predicted erroneously for Dewey on the basis of phone surveys. Newspapers came out with "Dewey Wins" headlines. There's a famous photo of Harry Truman holding one of these newspapers up and laughing.
Here's some info, you can probably find a more fuller description on the
>Per my recolection, Mr. Roosevelt was the underdog candidate in the polls
>for the 1936 election, which, needless to say, he carried hands down. The
>survey, as I recall, was done by phone and, of course, at the time this
>artifact was a middle/upper class
there was a mail-in component and the response rate was very low. names
and addresses were taken from vehicle registration lists and country club
memberships is where you really got the class divide
roosevelt was also the current president so the voluntary response was much
more in favor of the new opponent
>device. Therefore, the poll accuracy reported on the expectation of this
>social class -not of the entire nation. I have been told this is
>erroneous; that the poll was done by a magazine
>and that GAllup predicted this result via their phone poll. Please advise.
yup, gallup was able to predict the outcome of the Lit Digest poll (who had
been pretty successful up until this election), as well as of the actual
election, with much smaller (but more random) sample sizes
Hope this helps,
Beth L. Chance
Department of Mathematics
St. Olaf College
Northfield, MN 55057
I just came home from the office, so cannot get access to my file on this. As I recall, the magazine Literary Digest made the wrong call, and that preceded the magazine's demise, and is said to have hastened or caused it. There is a conventional wisdom on the matter, but a decade or so ago a corrective article on the matter appeared, and I believe that's what at my office. Sorry. If nothing turns up on ISOstat by tomorrow afternoon, I'll try to find it and let you know.
Mathematics and Computer Science
Amherst CollegeHuff talks about it in "How to Lie with Statistics", but he (like you said)
mistakenly reports the problem to be about the upper class.
In fact, the real problem is the response rate. The poll was done via mail,
but sent to lists of people from the telephone books (which is why it gets
mistaken as a telephone poll.) Anyway, they get a pretty small response
rate, and the people responding are all the people who want FDR out of there
(all the people who want change.) Those who were happy with the status quo
didn't bother to respond to the questionnaire.
The American Statistician has a good article on it: November 1976, vol 30,
no 4, pgs 184-185.
Department of Mathematics
Check out http://www.ciadvertising.org/student_account/fall_00/adv382j/jeffrey01/gallup/presi.html
the phone list gaff was in 1936: http://www.abilene2000.com/elec/net0520.html and
it was Literary Digest's error ... http://www.studyworksonline.com/cda/content/worksheet/0%2C1600%2CEXP545_NAV2-76_SWK543%2C00.shtml
Finally, go to Yahoo.com and search the topic literary digest presidential poll, to find specific details at a number of websites. The following is one of several returns: http://muweb.millersv.edu/~politics/June182002.htm
Dept of Math Sci
Saginaw Valley State Univ
University Center MI 48710
I looked into this a few years ago. The survey was run by a magazine
called the Literary Digest. As I recall, they had multiple sampling
frames, including automobile registration lists and telephone
listings. They had correctly predicted the winner of all the
elections from around 1912 to 1928. Gallup fairly accurately
predicted both their prediction and the outcome. The other propable
source of systematic error was non-response bias (republicans in the
sample responded at a higher rate than democrats in the sample; only
22% of those surveyed responded). The Literary Digest went bankrupt
within a short time :-)
I don't think Gallup used a phone poll, but rather a kind of
stratified sampling designed to get the right proportion of
respondents in various categories. Gallup was not got embarrased in
turn in the `48 election (Truman vs Dewey), most commonly attributed
to timing (a last minute surge of undecided voters).
Albyn Jones email@example.com
Reed College, Portland OR 97202 (503)-771-1112 x7418
Hi, guys: thanks for the great response and the many excellent pieces of info regarding the Roosevelt poll; I am going to collect them and send them back to the group on wednesday.
Another, related issue: there is some concern among pollsters regarding the (potentially negative) effect that gadgets such as cell phone and their corresponding numbers, due to the way that disposable phones and changing providers' sometimes effects on rapid turnaround and short life of such numbers. Also answering machines and caller IDs (that allow phone users to screen out unwanted calls). can effect the survey and or produce a significant drop in the survey response especially if such drop is correlated with some users' type of responses (e.g. say those favoring negative response to survey question tend to not answer phone or those perhaps younger population favoring a survey question tend to use and often change mobile phones and may therefore be less prone to be selected in the sample).
Any references or tips about these issues? By the way, especially for general stats and "literacy" stats courses posing such questions to students may raise their motivation and interest in the subject/
The New York Times published an interesting article on the effects of
cellphones and caller ID on the phone polling process. I often have my
students read and discuss the article. Here is the reference:
Nagourney, Adam (November 5, 2002), "Cellphones and Caller ID Are Making
Pollsters' Jobs Harder," New York Times
(Unfortunately, you can't access the article through the NYT website
archives without paying for it, but perhaps a librarian can help you
track it down.)
Assistant Professor of Statistics
Department of Mathematics
Attached is an article from the New York Times about cell phones and
polling. They say the one of the problems with cell phones is that the
owners are selected out, since pollsters don't want to alienate them with
unsolicited calls that use up their minutes.
November 5, 2002
Cellphones and Caller ID Are Making Pollsters' Jobs Harder
By ADAM NAGOURNEY
A rapid rise in the use of cellphones and caller identification technology, along with telemarketing calls that are chasing Americans from their telephones, is making political polling more difficult and increasingly less reliable, pollsters say. A result this Election Day is that it is harder than ever for pollsters to find voters and to get them to say how they intend to vote.
Pollsters say a problem that they first began noting 10 years ago, as Americans realized that answering machines could be used to screen out unwanted solicitations, is today forcing a re-examination of the methods by which they question voters.
In interviews, several pollsters said they now discussed ways to change how they approached a fundamental procedure in politics that has, over 75 years, moved from the mail to door-to-door canvassing to the telephone.
"At some point, there's going to be a crash between what's happening in the country and what's picked up on the phone," Stanley Greenberg, President Bill Clinton's pollster in the White House, said yesterday.
Whit Ayres, a veteran Republican pollster, said: "I can't fathom 20 years from now the telephone remaining the primary means of data collection. This industry is in a transition from telephone data collection to Internet data collection."
"In the meantime," Mr. Ayres said, with a note of frustration in his voice, "we've got to get people to answer the phone."
Pollsters said the increasing difficulty in reaching people was undercutting their efforts to assemble a pool of voters that was scientifically large enough and diverse enough upon which to draw reliable conclusions. While some pollsters said they could compensate for that by staying in the field longer or calling more people, that kind of effort takes time and costs money, two things that are often in short supply at the end of a campaign.
Pollsters said they had tried to respond to the problem with various time-consuming and costly remedies. They have increased the time they spend in the field, employed teams of specialists to methodically call back numbers that are answered by machines, and mathematically adjusted their findings at the end of the survey period to make up for voters they might have missed.
But several described these as stop-gap measures that have been only partly successful. Pollsters are under intense pressure to move quickly and hold down costs. They are uncomfortable with the kind of statistical adjustments used to compensate for missed or refused calls.
In particular, pollsters said they might be undercounting the growing number of younger voters who only have cellphones, as well as elderly voters who, they said, tend to be especially wary of any call that sounds like a solicitation. Several pollsters said the rise in the number of unlisted telephone numbers was more pronounced in minority and low-income neighborhoods.
In a case that drew much notice over the weekend, two polls trying to measure the Senate contest in Minnesota produced opposite results: one had Walter F. Mondale with a six-point lead, while the other had Norm Coleman with a six-point lead.
Howard Wolfson, the executive director of the Democratic Congressional Campaign Committee, said he had seen instances this year where two polls by different pollsters in the same district had produced findings so different that it was as if they had come from different states.
"There is a lot of evidence that all of this is making our life more difficult and hurting our efforts," Mark A. Schulman, the president of the American Association of Public Opinion Research, said yesterday.
But, Mr. Schulman added, "I'm not ready to run up the white flag and concede defeat."
Matthew Dowd, a Republican pollster who advises the White House, said: "Right now, I'd still make the argument that polling is the best way to find things out. You haven't gotten to a point where you can't trust it, but you have gotten to a point you have to weigh it."
One prominent pollster said the number of telephone calls that were not completed — either because no one answers the telephone or because they answer and refuse to participate — had jumped in recent years, to about 30 percent from 10 percent. The number is even higher in New York and in South Florida.
"Response rates are falling," said Tom W. Smith of the National Opinion Research Center at the University of Chicago. "Either you spend a lot more time and money doing your survey or you end up being stuck with a much lower response rate than is traditionally acceptable."
As a rule, political polls are done by candidates, political committees or news media outlets. There was some debate yesterday which of these conductors of polls would prove to be most affected by mounting difficulties.
The technological burdens facing the pollsters are expanding. Telephone answering machines and caller ID make it easy for potential respondents to screen out calls — something that is more likely in urban and suburban areas, where people are more likely to be the target of aggressive commercial telephone solicitations.
Mr. Smith said that to complicate things for people in his profession, new screening machinery had been developed that can, in theory, identify calls that are being made by a mass dialer or can refuse calls from any unknown number.
Cellphones have posed another complication for pollsters. There is no directory of cellphone numbers, and an increasing number of people use cellphones as their home telephones. Mr. Schulman said federal regulations barred pollsters from calling people on their cellphones without permission, because the recipients of the calls are obliged to pay the cost.
Several pollsters said yesterday that they would prefer, if possible, not to conduct interviews with people on cellphones. These interviews typically take 20 minutes or so, and were intended to be done with people sitting at home rather than chatting on a cellphone from a car or restaurant.
"We haven't come to grips with the cellphone issue yet, I'll be honest with you about that," Mr. Schulman said. "Up to this point, the cellphone has generally been the second phone for the hard-wire phone in the household. In the future, we've got to figure out a strategy to deal with this."
I just sent this message to Joy but now I see that you might be interested in it. I do not really understand who gets what when I reply but it looks like it goes only the the sender. Laurie
You can find a discussion of the cell phone problem in the current issue of chance news (www.dartmouth.edu/~chance)
There is a link here to a more recent article in the Arizona Republic on the effect of cell phones that is available.
There is also a discussion of an pbs program that listened to a poll being carried out that I found fascinating. Cheers
Thanks again, all!!!!