a joint effort of malpractice by grantcart and phrigndumass
Pollster Dilemma: Weighting Demographics in a Change Election - Texas as a Case Study
Summary
The changing identification of the American population with the two parties is giving pollsters a real problem in weighting responses. These problems are exacerbated in national polls. State polls show Obama holding a much wider and more solid lead than the national polls.
The pollsters have even more difficulty in trying to determine what the voting turnout in minority groups is likely to be and this further under reports Obama's strength. In looking closely at Texas, for example, it shows that McCain's biggest electoral vote state would become much more competitive in the polls if the weighting of AA and Hispanics was adjusted and simply bringing Texas into swing state status would be significant.
Pollsters face increasingly difficult choices in giving weight to different demographic groups in a change election.
It has become an increasingly frustrating experience to try an understand contradictory polling results that tend to show a run away electoral college result with national polls that are tied.
This can be seen from the two examples of poll of polls that are using similar data bases but vary only with the addition of 3rd party candidates:
More over if you take time to extrapolate back from the state polls from the same pollsters, it shows that there is much more consistency than a 'global' national poll as illustrated by phrigndumass's projected poll that comes from his daily thread, "The Daily Widget". (If you have gotten this far in the thread then you will certainly prefer his daily unbiased professional snapshot of the latest numbers).
The problem of weighting demographics.
Pollsters do not simply ask the first 1500 people they contact. They are divided into groups and a representation from that group is included. In order to get a feel for how complicated that process is take a look at an academics summary of weighting from the last election.
http://www.depts.ttu.edu/hs/hdfs3390/weighting.htm If you scroll down on the page you will find Dr. Reifman's article "Weighting Pre-Election Polls for Party Composition:
Should Pollsters Do It or Not?" based on the 2004 elections.
What makes this an even more difficult question this year is that it is clear that the party's are realigning. So what percentage of Republicans and Democrats should be included?
Well Rasmussen has re-ignited the discussion and pollster.com has been discussing it
here are some of their articles:
http://www.mysterypollster.com/main/2004/09/weighting_b... http://www.mysterypollster.com/main/2004/09/my_post_on_... http://www.mysterypollster.com/main/2004/09/why_how_pol... http://www.mysterypollster.com/main/2004/10/why_how_pol... http://www.mysterypollster.com/main/2004/10/why_how_pol... http://www.mysterypollster.com/main/2004/10/why_how_pol...
Here is pollster.com's general conclusion on party identification in weighting
The issue of weighting by party is a prime example of the tension between science and art in political polling. When it comes to art, judgments are always subjective. Keeping that in mind, here is my take: I am most comfortable with a combination of the approaches of John Gorman and Peter Hart (of the Fox News and NBC/Wall Street Journal surveys respectively) described in the last post. Gorman's approach of stratifying his samples by actual turnout statistics, despite the risk that past turnout is not a perfect guide to the future, forces a defensible regional consistency across surveys that makes weighting by party less necessary. If weighting is ever necessary, I prefer Peter Hart's cautious, ad-hoc "dynamic weighting in reserve" approach.
One thing to keep in perspective: The debate over party identification is important, but those who weight by party have no magic "fix" to the sometimes random variation in surveys and those who avoid weighting are not overlooking some obvious methodological flaw.
With no 'magical fix' the pollsters are admitting that in an election where so many of the basic assumptions are in flux it is very difficult to certify any scientific basis for their results. More to the point, state polls where those assumptions can be more closely analyzed are going to be a lot more accurate than the national polls that simply mash their assumptions together.
What the state polls are showing?
Again using the same polling company's and the same assumptions the state by state background gives us a much clearer picture, again from phrigndumass' award winning daily thread is this helpful chart:
It shows that Obama not only has a significant lead but that it is solid.
Obama's safe states are a lot safer than McCains. Obama has 100 EVs in New York, California and Illinois alone, which have been conceded by McCain outright. Taking out Iowa Obama has a rock solid 230+ EVs. McCain has 110+.
McCain has to take all of the swing states and steal a couple thought likely for Obama. This chart really shows how critical Florida and Virginia is to McCain. He loses either of those and there is no real way that he can get to 270. Senator Obama has been making his largest media buys in Florida and it has gotten steadily closer. Virginia gets constant attention and will (grantcart wild speculation) benefit from Obama's VP selection of Kaine.
The real issue that the pollsters are missing is weighting of minority participation in the GE.
The pollsters are basically assuming that the participation in the GE will be similar to past elections with a slight increase in minority votes. If we take a closer look however it shows that pollsters may be making a major error not by weighting party participation incorrectly but by not giving enough weight to increased minority voter turnout.
A Closer Look at Texas.
Looking again at the breakdown by state it is clear that if McCain were to lose Texas it would be a landslide for Obama.
Phrigndumass, being as kind as he is wise, is able to give us additional background on the numbers in Texas, where Obama is consistently trailing McCain by 9 points.
http://www.pollster.com/polls/tx/08-tx-pres-ge-mvo.php Figuring in a 10% increase in African American voters, and a 15% increase in Latino voters, and a small 2% increase in white, non-Hispanic voters, we get the following turnout in Texas:
Projected turnout 2008 in Texas:
Black - 914,461
Latino - 1,440,557
White - 5,602,023
Other - 50,501
Total Votes 2008 - 8,007,542
Assume the following votes for Obama in Texas:
Black - 95%
Latino - 70%
White - 37.35%
Other - 70%
Obama wins Texas with 50.01% of the vote. If the white percentage is lower than 37.35% for Obama, Obama loses Texas.
Further, if the Latino vote is only 65% for Obama, then Obama must win at least 38.62% of the white vote to win Texas.
Interestingly BRC, a highly respected pollster in AZ, is showing that McCain's numbers among Hispanics are plummeting in AZ where you would expect them to be strong.
Growing evidence of stronger movement of Hispanics to Obama
McCain only polls 9% of the Hispanics in his home county
http://www.brcpolls.com/08/RMP%202008-III-01.pdf Finally, even though Latino voters have in the past displayed a tendency to favor Democratic
candidates, John McCain has long been able to attract from a quarter to a third of their vote. This appears not
to be the case today. Seventy- nine percent of Latinos in this survey say they will vote for Barack Obama while
McCain’s share registers at only nine percent.
If we plug in more realistic minority participation in Texas then Obama can win with white support as low as 38% and certainly makes it close with 35%. Winning in Texas would, obviously, be huge but simply putting Texas into play would be very significant. If Obama's VP selection and convention bounce were to give him 4-5% it would be substantial.
Here is how phrigndumass illustrates and explains it:
This graph shows what the makeup of the voting populace in Texas will be in November if we go by state trends and interest by ethnic group:
This second graph shows the breakdown of the Texas population by ethnicity, and the trended percentage of those ethnic groups who will actually turn out to vote in a presidential election. Taken into account are a 10% increase in African American voters, a 15% increase in Latino voters, and a 2% increase in white voters over 2004. The graph shows room to grow by ethnic group, so a higher turnout percentage by minority group than shown below will require a smaller percentage of the white vote for Obama to win Texas.
If Obama wins 40% of the white vote in Texas at this point, the Texas race will be a toss-up. The average polling nationwide for white voters is currently 36.6% Obama, 44.6% McCain, and 18.7% Undecided. Since Texas is polling 5 points right of center, we can assume average polling of 31.6% Obama, 49.6% McCain, and 18.7% Undecided for white voters in Texas. Obama and McCain would need to split the Undecided voters evenly for Texas to be a toss-up.
A more realistic appraisal of minority turnout and a convention bounce and Texas comes into play and McCain has to start worrying about keeping Texas - the foundation for any Republican GE campaign.
It is interesting to note that many red states that have no minority population, like Montana, show Obama outperforming past campaigns and it has to raise questions about how much weight pollsters are giving to minority turnout in Georgia, North Carolina and Florida. With Florida now polling McCain +4% and Obama reporting more than 100,000 new voters registered you have to wonder if the pollsters can keep up with the radically changing dynamics of weighting for demographics that this General Election campaign brings. As they admit they have no 'magic fix'.
Printer-friendly format
Email this thread to a friend
Bookmark this thread
(1000+ posts)
Send PM |
Profile |
Ignore
