Proof?
by matt at 6:49 am on October 25th, 2004 in PoliticsIt’s no secret that 1115.org has a love/hate relationship with polls. While they are the closest thing the political process has to a scoreboard, they are often unreliable. We have written about how pollsters aren’t reaching people who only have a cell-phone, poorly timed polls, and polls that over-represent members of one party or another. By way of a prediction, I’ll go on the record with my belief that the polls aren’t even in the same ballpark as the actual reality of voter preference. There will be several polling companies forced to hang their heads in shame and re-evaluate their methods and models come November 3rd.
While doing some research this past weekend, I found a press release from the Consumer Electronics Association detailing findings in a recent poll they commissioned.
The survey found while likely voters, both Republican and Democrat, reported receiving an equal number of polling calls, Republicans are 25 percent more likely than Democrats to have responded to at least one poll. The difference was just as pronounced for polls related to the national election where Republicans were more than 23 percent more likely to have responded to a poll call.
Some of the difference may be attributed to the survey findings showing that Democrats are much more likely than Republicans to indicate they screen their calls (44 percent and 27 percent respectively).
This data point is one that became very clear to us during the first night of our phone bank. Our list was almost entirely made up of registered Democrats, and despite the fact that we were calling after normal working hours, we spoke to as many answering machines (or more) as we did people.
If I made my living as a pollster, this information would haunt me.
jamie beth wrote:
i’m considering asking everyone i know to vote BEFORE work so that “exit poll” numbers tip early and so that no one, god forbid, figures “well bush is up, i shouldn’t bother to go.” what do you think about this strategy?
Posted 25 Oct 2004 at 11:40 am ¶
matt wrote:
As with the 2002 Congressional election, there will be no published exit poll numbers. Voting early is always a good strategy however.
Posted 25 Oct 2004 at 12:22 pm ¶
sarabeth guthberg wrote:
Please forgive the very naive question: don’t the polls adjust their numbers before they are released to make the proportion of Democrats in their sample conform to whatever the actual proportion in the relevant population is supposed to be?
Assuming for the sake of argument that the population is 50% republican and 50% Democrat, what I mean is that if Republicans are 23% more likely to respond then in the poll you just count each responding Democrat 1.23 times to balance that.
Sarabeth
Posted 26 Oct 2004 at 2:37 pm ¶
matt wrote:
Latest Florida Gallup Poll:
Likely Voter Sample
TOTAL: 768
Rep: 341 (44%)
Dem: 273 (36%)
Ind: 146 (19%)
Registered Voter Sample
TOTAL: 909
Rep: 387 (43%)
Dem: 321 (35%)
Ind: 187 (21%)
There have been more actual Dem voters in just about every election. Gallup is predicting the most massive realignment in party ID voting in over 100 years. They’ve offered no explanation or proof of this.
Posted 26 Oct 2004 at 2:44 pm ¶
sarabeth guthberg wrote:
Would you happen to know if the likely voter sample is a subset of the registered voter sample?
If so, how representative can the samples be since 768 out of 909 is roughly 84%?
Sarabeth
Posted 26 Oct 2004 at 2:51 pm ¶
matt wrote:
The above numbers are backwards, but they are from this site. Steve Soto of the Left Coaster has been digging into the sample formulation over the last few weeks.
Gallup / CNN / USAT poll here.
Posted 26 Oct 2004 at 2:58 pm ¶