Oversampling is a myth

[Onlooker]

After reading this rather hard to follow Gallup article -- http://pollingmatters.gallup.com/ -- it is clear that pollsters do not poll by party, but instead poll randomly using other factors.

With Obama consistently leading in most polls, Republicans are desperate for good news. So, they are trying to create a myth of oversampling of Democrats. This is the absurd notion that pollsters intentionally call a lot more Democrats than Republicans, and low and behold, Obama is ahead!

The reality is that polling is random and not based on party affiliation, but on education level, gender, race, and other fairly fixed and knowable factors. The reality is that this random phone calling is turning up consistently more Obama than Rmoney supporters. Of course, if you ask someone who says they support Obama which Party they lean toward, more often than not they will say "Democratic" or at least "Independent," not "Republican." This results in the appearance of oversampling of Democrats.

So, if there is oversampling that favors the Democrats, it would not be that pollsters are calling up more Democrats, but that they are calling up a disproportionate number of African Americans, women, gays, union workers, etc. So far, there does not seem to be evidence of that, but perhaps it's simpler for the simple-minded to simply believe that pollsters are intentionally calling up more Democrats than Republicans, rather than engaged in some form of random phone calling.

The oversampling claim is nonsense that the MSM is doing a very poor job dispelling.

[Loudly]

1. Another possible explanation is that there are simply more Democrats.

Do the voter rolls of states which register with expressly declared party affiliation bear this out?

[jody]

2. Credible sampling methods must be random but judgment when trying to stratify the sample for

multiple groups is even more critical.

Oversampling can be a problem if sample results are not properly applied to each sub population within the target population.

[Xipe Totec]

3. There is one population that's being undersampled: cell phone only hipsters nt

[jody]

4. What pollster knows the % of the population is "cell phone only hipsters" and its complement?

Absent that minimal knowledge at the sub-population level, then sample data can not be collected for the population.

[Xipe Totec]

5. They're not sampled at all because you can't poll cell phones w/o permision

So people such as myself who do not have a land line and use cell phones exclusively, are not being sampled at all

People who embrace modern technology tend to be more progressive by definition.


8% of the population uses cell phones exclusively

http://electronics.howstuffworks.com/how-to-tech/how-to-go-completely-mobile.htm

[jody]

6. Thanks, I knew that and my point is still valid. nt

[Xipe Totec]

7. I just updated my post. The answer is 8%

of the population (and increasing) uses cell phones exclusively.

[jody]

8. Understand but you used "hipster" which is a subset of "uses cell phones exclusively". Not trying to

derail your thought but points like that do matter when designing a sampling plan and writing conclusions.

[Xipe Totec]

9. I understand the confusion

I meant "uses cell phones exclusively" = hipster.
So the sub-sample is 100%

[jody]

10. Understand. Thanks for the exchange because it's easy for readers to misunderstand. When it comes to

journalists they often add an innocuous adjective that in no way relates to the stats being discussed.

[Xipe Totec]

11. When it comes to precision, English is a blunt instrument

But this is an interesting demographic that is opaque to polling because of the stringent rules regarding soliciting and cell phones, and the political makeup of this population is very likely different from the population of people with land lines.

It will make things more interesting as time goes by.

[HopeHoops]

12. Well, if the sample group is actually random, the larger n is the lower the MOE. HOWEVER...

I recently had a GOP pollster come around. She had a printed list (obviously voter role) and only my wife and eldest daughter were on it. I wasn't. I'm a registered Republican (mostly so I could write in Green Party candidate Jill Stein in the GOP primary - okay, only because of that). I wasn't listed. I pointed out to her that the poll results would, by the very nature of the list, be way off base. I know my statistics. That's a skewed poll by its very nature. "Oversampling" is a good thing if the base is even, but it only makes results less accurate if the base is selective.