strangegrey wrote:RossValoryRocks wrote:But Rasmussen is far from a Democratic operative, while CBS is...so how do you explain the relative sameness (within margin of error) of all these polls?
Im not saying they're all doing it. Perhaps Rasmussen and CBS are *not* doing anything wrong with their polling and they're both asking the right questions.
The point is that some of the polling data is so far outside of what one would consider the margin of error for this sorta thing, that someone's gotta be taking poor samples or asking skewed questions....
For the most part, I think pollsters with household names (and/or the ones affiliated with the major media) try to do the best job they can. Some have a better track record of predicting electoral outcomes (Gallup), some don't (Zogby). The differences in outcomes can often be easily explained by assumptions they use.
For example, does the pollster use "registered voters" as the population sampled, or "likely voters"? Using the latter makes more sense, because a lot of registered voters don't vote. But how will you determine who is a "likely voter"? This is where sampling error is injected into the process. Often a lot of assumptions have to be made here about the
predicted electoral turnout on election day.
Here's another (big) problem. In the past, pollsters used to call people at home, using easily available telephone directories. Stratified sampling and cross-tabulations (i.e., Democrats vs. Republicans, men vs. women, whites vs. non-whites) could be easily planned in advance, using these directories. But now, people often don't have a home phone -- they only have a cell phone, and those directories are far more difficult to compile or get a hold of. Also, with cheap voicemail for home phones and cell phones, it's easier for people to screen their calls. Are you to assume, as a pollster making 100 calls, that the 49 people you reached by phone are statistically similar to the 51 people who screened your call or let it roll into voicemail? There could be a lot riding on your answer to this question ......
Another point. You win the presidency by winning electoral votes, which means winning individual states. A close poll (say, 48% to 46%) may not really predict that the candidate with the higher vote totals wins. Let's say Obama is ahead 48% to McCain with 46%. But let's also say that in 3 large states (California, New York and Illinois), Obama is
far ahead in the state polls (say, 70% to 24%). Let's also assume that McCain has leads in states, but no leads this large (or perhaps, no leads among states with that many electoral votes in total). Because Obama has built (in this hypothetical example) large leads in states that he was going to win anyway, his national poll results are distorted and might not reflect an actual electoral victory. Obama in this example could lose a close electoral vote, even if the national poll accurately reflects public opinion on the eve of election day. The point is: look at how the state polls too. That's where the ground action is ......
In my opinion, based on the last few elections, the two organizations that seemed to have good predictive ability in their national polls are (1) Gallup and (2) Rasmussen. I add them together and divide by two.
Posted: Tue Sep 09, 2008 12:19 am
