'UPSETS' tonight? Are many midterm polls biased 4-6 points AGAINST Ds because land-line phones

[ProgressiveEconomist]

land-line phones are less prevalent among Democratic-leaning younger people than among independents or among Tea Partiers and other Republicans?



Remember the "Dewey Defeats Truman" Chicago Tribune headline Harry Truman holds up in his famous Presidential victory photo? IMO, if Democrats pull off some "upsets" tomorrow, something will have occurred eerily similar to what happened in 1948, when those who had (land-line) telephones tended to be wealthier and more Republican than those who turned out to vote.

Pew Research suggests the magnitude of landline bias may be FOUR to SIX percentage points against Democratic candidates for midterm election polles this year.

If this research is accurate, IMO the media tomorrow may be reporting "unexpected surges toward Democrats of eight to twelve percentage points" in many states.

WHAT'S YOUR OPINION?

--------------------------------------------------------------------------------------------------
From http://www.physorg.com/news/2010-10-political-polls-cell-proliferate-lines.html

"October 20, 2010

Political polling in the U.S. is undergoing significant changes because of the growing popularity of cell phones and the diminishing number of Americans with traditional land lines, says Brian F. Schaffner, a political scientist at the University of Massachusetts Amherst. He says the shift to cell phones means that traditional public opinion polls based on calling large random samples of the population that have been familiar since the 1950s really don�t exist anymore. 'Polling is getting more difficult," Schaffner says. "No matter how you do it, you can�t get a truly random sample.'

Adjusting polling methods to reflect these new demographic realities requires understanding what those changes are and which groups of people are affected, he says. For example, Schaffner estimates that between 35 and 40 percent of Americans are very difficult or impossible to reach on those traditional land line telephones. He says, 'People who can be reached by land lines tend to be older, have families, and are more connected to their communities.'

The other key element is the growing number of people, including many young people, who use only cell phones. In a recent study Shaffner conducted with a colleague ... they found, "One in five households relies exclusively on cell phones for telecommunications. That fact has created coverage problems for phone surveys, and the demographics of this population--younger, mobile, less socially connected--may create biases in political surveys limited to random dialing samples." Schaffner says many polling organizations now adjust for these changes by including cell phones users, but not all. In fact some well-known pollsters, such as Rasmussen and Survey USA, don�t include cell phone users and these companies produce many of the statewide polls that political onlookers are watching during this election season. According to Schaffner, "These surveys run the risk of being biased in favor of Republicans because they are more likely to exclude groups, like younger adults, who vote more Democratic." Schaffner notes that while polls attempt to use statistical techniques to adjust for the fact that they are missing those with only cell phones, these adjustments are become less effective with time, forcing pollsters to rethink their approach to polling. ...

Provided by University of Massachusetts Amherst (news : web)"

--------------------------------------------------------------------------------------------------
From http://pewresearch.org/pubs/1761/cell-phones-and-election-polls-2010-midterm-elections

"Cell Phones and Election Polls: An Update

October 13, 2010 People & the Press Pew Research Center for the People & the Press

Cell-only adults are demographically and politically different from those who live in landline households; as a result, election polls that rely only on landline samples may be biased. Although some survey organizations now include cell phones in their samples, many -- including virtually all of the automated polls -- do not include interviews with people on their cell phones. ...

It is possible to estimate the size of this potential bias. The Pew Research Center for the People & the Press conducts surveys with samples of landline and cell phones, which allow for comparisons of findings from combined landline and cell interviews with those only from landline interviews. Data from Pew Research Center polling this year suggest that the bias is as large, and potentially even larger, than it was in 2008. ...

In three of four election polls conducted since the spring of this year, estimates from the landline samples alone produced slightly more support for Republican candidates and less support for Democratic candidates, resulting in differences of four to six points in the margin. ...

[struggle4progress]

1. gotv

[BzaDem]

2. Gallup uses live interviewers and polls cell phones. They show a greater gain for Republicans than

any election since 1894. If their forecast is correct, high double digit losses in the House is likely (and close to triple digits is not out of the question).

And they have not been wrong by more than 3 points in any direction for any midterm ever. Since they started midterm polling in 1950. That's right -- 1950.

Now personally, I believe they are wrong this time around. But it has nothing do to with robopolls and cell phones, since they are not a robopoll and they poll cell phones. Rather, I believe it has to do with their likely voter model, which (while extremely accurate in all past midterms) might be missing some unique factors in this election.

But that means that we are likely facing a 50 or 60 seat loss, rather than a 90 seat loss. Not that we are going to keep the House.

[ProgressiveEconomist]

4. Have you seen the issue of cell-phone-only bias covered in any media during tthe past

six months? It's surely more relevant to predicting outcomes tomorrow than many other "issues" the media have focused on.

I was surprised to see the Prew Research report linked in the OP. Pew evidently have changed their tune on this during the past two years--Factcheck (at http://www.factcheck.org/askfactcheck/are_polls_skewed_because_many_people_only.html cite a January 2008 Pew report as evidence the bias may be nothing to worry about.

[BzaDem]

5. My point is that there are polls with cell phones that show EVEN WORSE results for us than

other polls without cellphones.

Pew's results only showed that there was a cell phone bias with their weighting model. It is entirely possible that such bias is removed (or even overcompensated for) depending on your weighting model. PPP for example does not poll cell phones, but they got Scott Brown's margin exactly right, got within a few points of both Christie and McDonnell in 2009, and got within a few points (or nailed exactly) almost every state in 2008.

So what Pew says might apply to other pollsters. Or it might not. Or the reverse might apply. It completely depends on how they weight.

But regardless, there are reputable polls with cellphones that show apocalyptic results, and polls without cell phones that show just plain bad (as opposed to apocalyptic) results. This is not a question with a clear cut answer.

[ProgressiveEconomist]

6. "This is not a question with a clear cut answer". Not tonight, anyway.

[ProgressiveEconomist]

8. "It completely depends on how they weight". IMO perfect weights could come from

a lsrge dataset such as the latest National Center for Health Statistics survey, which has the demographics of the quarter of the population that can't be reached on landlines, according to the Pew research link in the OP.

Political polls that don't have any cellphone data could use those landline coverage surveys to adjust their poll results.

If demographics alone predict landline coverage probabilities well, then perfect weights for demographic cells in political polls would be the inverses of probabilities people in those cells can be reached on landlines.

I wonder which political polls use coverage data from other surveys in this way. Do you know?

[Kalyke]

10. No they don't.

The polls that include cell phones show both Democratic WINS and/or smaller Republican gains. Only one - Gallup - who is weighting, not necessarily contacting, likely voters - show any sort of Republican tsunami.

(See below)

[ProgressiveEconomist]

7. "Rather, I believe it has to do with their likely voter model". AIUI, only Pew has compared

polls where every other detail of methodology was the same except that one poll used cell-phone data along with land-line data and the other used only land-line data, which misses at least a fifth of the population. Pew found no difference in early 2008, but now says midterm poll results with no cell-phone data may be biased against Democrats by four to six percentage points.

As you point out, even polls with no cell-phone data may apply undisclosed wights to try to counteract this bias. It's a shame pollsters don't make their weighting schemes public so that this issue could be analyzed using all potentially-available data.

[Lexwasp]

9. The question is...

"Where will the number a seats lost be in the 45-85 seat range?"

Not "Will the Dems lose the house?".

[krabigirl]

3. I think so. Even the ones who use cell phones don't sample too many of them.

Except that SurveyUSA one.

[Statistical]

11. Unlikely.

1) Some polling firms DO poll cellphones
2) Some polling firms instead weight their results. The raw data is weighted to represent all subpopulations. Thus if Democrats make up 47% of electorate but the raw data in the poll only shows 41% then it is "boosted" in the final result. Same thing applies to other demographics (age, race, gender, etc). So if 18-25 year olds are projected to make up 18% of the population but only make up 12% of the raw data each 18-25 "vote" counts as 1.5 votes.

The only way you have a bias is if for example 18-25 male Democrats living in urban area with a landline are MORE/LESS likely to vote for one candidate than 18-25 male Democrats living in urban area without a landline.

Wishful thinking nothing more.

Democrats will lose the house. Democrats will most likely barely keep Senate.

I call:
51 seats in Senate (+/- 2 seats).
200 seats in House (+/- 10 seats).

[ProgressiveEconomist]

12. In practice, is lack of cellphone coverage lumped in with nonresponse corrections

in political polls, as you are suggesting? Is being out of the frame because you can't be reached on a landline the same as nonreponse for people in the landline sampling frame?

The procedure you describe sounds like a correction for random nonresponse, relabeled as a correction for random landline noncoverage.

[ProgressiveEconomist]

13. I can't imagine Pew Research did not use the usual nonresponse corrections in their comparative

polls with and without cell-phone-only respndents. Thus their finding of 4-6 point biases against Democrats already would have included the wiehghtings you recommend. Thus cell-phone-only appears not to be random but rather correlated positively with voting for Democrats.