« Poll of Polls | Main | Ohio »

October 19, 2004

NYT: Why Can't Pollsters Agree?

As a companion piece to their poll story, the New York Times' Jim Rutenberg weighs in this morning on why the polls disagree. It touches on much of what we discuss here. Two money quotes:

While the headlines they produce may diverge, the actual findings of these polls may not be so different. The differing conclusions reflect how different pollsters use complex formulas to interpret very similar findings among self-described registered voters and try to come up with a result they think best accounts for who will actually show up at the polls...
"Science is put in place and then the pollster has to exercise judgment about how to define likely voters," said Nancy Belden, president of the American Association for Public Opinion Research. "And every polling organization may define a likely voter slightly differently, or in some cases, more than slightly differently than the next polling organizations." [link added]

I'll say. Rutenberg's piece is a good primer for the likely voter issues I'll be covering over the next week. As they say, read it all (and soon while it's still free!)

Related Entries - Divergent Polls, Likely Voters, Sampling Error

Posted by Mark Blumenthal on October 19, 2004 at 07:16 AM in Divergent Polls, Likely Voters, Sampling Error | Permalink


Why don't pollsters cite poll numbers from previous elections more often as a means for providing some context and basis of comparison? I'd like to revisit the polls for Gore-Bush (2000) this far out from election day.

Posted by: Clark | Oct 19, 2004 9:15:51 AM

Over at Donkey Rising, one poster showed Gallup's numbers, as well as a list of other polls, from 2000. If the numbers are correct, it is striking how far off the mark they were. They had Bush in many cases up by a significant lead. These numbers held true throughout. There is an argument that the DUI story hurt him in the last weekend, but I believe the polls don't bare that out- especially given the numeral shift. For example, I believe in one case, again Gallup, Bush was up 13 pts.

Posted by: bruhrabbit | Oct 19, 2004 3:04:54 PM

Most pollsters' RV and LV findings are generally similar, differing only by a point or two. Gallup is among the few -- if not the only -- that consistently seems to show marked differences between its RV and LV numbers.

As conveyed in a Sept. 20 Wall Street Journal article by John Harwood, Gallup's Frank Newport has defiantly denied any problem with his firm's likely voter model:

" 'We're open to any scientific evidence that would point to our modifying our likely-voter model,' responds Frank Newport, editor-in-chief of the Gallup Poll. Mr. Newport says so far he hasn't seen any."

How about this? Over at the Emerging Democratic Majority blog (link below), the following figures were revealed, regarding the performance of the final Gallup RV and LV numbers in calling the last four presidential elections. In short, in THREE OF THE LAST FOUR presidential elections, Gallup's REGISTERED VOTER numbers more accurately forecast the election than did Newport's vaunted likely voter numbers.

2000: RVs, +1D; LVs, +2R; actual result, +.5D

1996: RVs, +16D; LVs, +11D; actual result, +8.3D

1992: RVs, +8D; LVs, +12D; actual result, +5.5D

1988: RVs, +8R; LVs, +12R; actual result, +7.7R


Posted by: Alan R. | Oct 19, 2004 3:15:10 PM

The Gallup poll has now swung from even after the democratic convention to Bush +10 after the republican convention to even after first two debates and now to Bush +8 after lesbian-gate.

This seems suggest that there is a segment of the population who are not "undecided" but rather "fickle". How else to explain these 10 points swings while the official undecided remains very low.

This must mean that these leads are phony because the difference is made up entirely by a set of people who change their minds with wind and yet, when asked, affirm that they have made a "decision".

So there are in my mind three types of voters:

1) partisans who have made of their mind, firmly.

2) the true undecided.

3) the "fickle". The folks who change their mind on the basis of the last conversation they had but nevertheless do not accurately report this fact to the pollsters.

Posted by: Brad | Oct 19, 2004 8:08:16 PM

I think the real reason for Gallup's swings is that it doesn't weight its samples to ensure a consistent distribution from poll to poll of Republicans, Democrats, and Independents. Gallup just takes the sample it gets at face value. Slight shifts in party ID may occur from time to time, but not as radically as Gallup sometimes shows. This issue has been discussed extensively in the blogosphere. See my sample weighting website and the links in the Resources section below my introductory essay:


Posted by: Alan R. | Oct 20, 2004 11:00:10 AM

The comments to this entry are closed.