Would you please deny ignorance?
I find myself giving this speech way too often on ATS, and more often than not let it go, but here's another freebie.
While it's true that anyone who understands a little
about polling knows that you can make a poll say anything you want depending on who you
ask and how you phrase the question. It's also true, that anyone that knows alot
about the polling industry knows that bias and purposeful
misrepresentation is not going to be the case with a reputable independent
MRA/CASRO firm like Gallup.
While the client, which in this case is CNN/USA Today may choose to editorialize on the results by making the headline "Majority Call War a Mistake"
rather than "Confidence in Economy Improving" (also reported in the same Gallup Poll), that does not impune the integrity of the independent polling
organization or it's strict adherance to statistical and ethical guidelines strongly enforced in our industry.
Speaking of which, I'd stake my life on Gallup's sampling method here. I don't even have to look it up (which you are always welcome to do when you
see a poll) to know it's standard RDD (random digit dialing) among prefixes pulled to statisitically represent the population of the US. Not to
weight toward Democrats, not puppy owners, not oil barons... a time tested statistical representation of the population of the US.
I mean they aren't making this up as they go along. If multi-million dollar rollouts of new products are based on public opinion sciences, I think
they can handle a damn political poll. They aren't that big a deal.
As to sample size. Anyone that thinks you need to interview 51% or more of a population to get any meaningful, reliable, statistically significant and
time tested results needs a math class.
And here it is...
In this case, I'd call the 1005 sample overkill. But that's what media clients like to report in findings since the public can't imagine how
homogeneous and predictable they really are.
I'd wager this poll had the same percentage results after the first 400 completes. I KNOW after the first 600 out of a population of 280,000,000 they
had a highly reliable representation of the American public at a 95% confidence level and +/- 4% interval. By taking it up to 1005 they got the
interval down around +/- 3.1
Completely redundant, since I guarantee nothing changed between 400 to 600 to 1000 completes, but that's how statistics work. And news media like the
lowest MOE number they can report.
To put it ALL in perspective, had Gallup interviewed 9,999 people (10 times the amount they did) the reported interval would have only gone down to
3.0. THAT'S IT! From 3.1 to 3.0
And it would take millions of interviews to get it much lower than that. You approach the point of diminished returns very quickly in sampling.
It's redundant. It's predictable. It's math.