Uri:

374 LV for a NATIONAL dem poll? WTF?

All the others use at least 1,000 for these and that's with tracking.

Posted on May 9, 2008 4:11 PM

 RS:

@Uri:
Yes, that's low. But the recent CBS/NYT poll had just 283!!
The only poll that uses over 1000 voters is Gallup. Rasmussen has 900, AP/Ipsos and USAT/Gallup had about 500 "adults."
400-500 seems to be quite common. Look up the Pollster.com page for US Dems.

Personally, yes, I prefer bigger samples, of course.

Posted on May 9, 2008 4:50 PM

 Uri:

@RS: So none of these polls make any sense.

In a country with 50 distinct state + 3 more that vote in primaries (Guam, PR, DC) you pick up less than 350 vote? That's not even enough to cover one person out of every city of 50k+.

There is no way to get close to grabbing the correct population on such tiny sample sizes over such a heterogenous population.

Posted on May 9, 2008 5:22 PM

 HESSIAN:

Uri,
Your post was a bit vague on this point, but just to clarify, DC has 3 electoral votes in the general election whereas Guam and PR have zero.

Posted on May 9, 2008 5:50 PM

 Uri:

@HESSIAN:
Thanks for the clarification. I still can't figure out the system. So DC has EVs but no senate seats and no congressional seats?

Either way, my point about a limited sample are still the same for 50.

Posted on May 9, 2008 7:32 PM

 Shadar:

1. This poll is old, in the field April 30 - May 3rd.

2. It is indeed a small sample, that said, it doesn't make it horribly unreliable. It just gives it a higher margin of error. They cite a margin of error of 5.1% for the dem only numbers and 3.5 for the numbers against mccain.

3. You don't need to poll a person from every state necessarily to get a good national sample. You just need to poll similar voting demographics to match the national profile. The more states you poll from the more accurate you will be but some people here seem to not understand the process well. If, for example, you polled 1 person from every city over 50k in size you'd have horribly inaccurate results. You need to poll from all over. But if you poll a black urban voter from NY you will get similar results as a black urban voter from Atlanta. But if you poll a rural white voter from Iowa you may get a very different answer than a white rural voter from WV for example and different still from a rural voter from California or Wisconsin.

You need to know your demographics so you make sure to poll a representative section of the population as a whole. If you could do this perfectly you could probably poll as few as 100 people and get a very accurate result. The problem is no one can do it perfectly so you add more and more people to gradually reduce the margin of error in your polling due to demographic errors.

Posted on May 11, 2008 2:15 PM