At Red State, Blue State it’s about politics, here at Statistical Modeling it’s about survey sampling. Was it all based on a sample of size 6?

Posted by Andrew on 2 November 2008, 9:32 am

At Red State, Blue State it’s about politics, here at Statistical Modeling it’s about survey sampling. Was it all based on a sample of size 6?

## Recent Comments

- Sameera Daniels on The statistical significance filter leads to overoptimistic expectations of replicability
- Carlos Ungil on The statistical significance filter leads to overoptimistic expectations of replicability
- Daniel Lakeland on The statistical significance filter leads to overoptimistic expectations of replicability
- Andrew on Click here to find out how these 2 top researchers hyped their work in a NYT op-ed!
- Emmanuel Charpentier on X spotted in L’Aimant, par Lucas Harari
- Carlos Ungil on The statistical significance filter leads to overoptimistic expectations of replicability
- Carlos Ungil on The statistical significance filter leads to overoptimistic expectations of replicability
- Anonymous on Click here to find out how these 2 top researchers hyped their work in a NYT op-ed!
- Andrew on Click here to find out how these 2 top researchers hyped their work in a NYT op-ed!
- Sameera Daniels on The statistical significance filter leads to overoptimistic expectations of replicability
- AB on Click here to find out how these 2 top researchers hyped their work in a NYT op-ed!
- Anonymous on Click here to find out how these 2 top researchers hyped their work in a NYT op-ed!
- Anonymous on Click here to find out how these 2 top researchers hyped their work in a NYT op-ed!
- Sander Greenland on The statistical significance filter leads to overoptimistic expectations of replicability
- Xi'an on X spotted in L’Aimant, par Lucas Harari
- Jeff on Comment of the year
- Andrew on X spotted in L’Aimant, par Lucas Harari
- Anonymous on X spotted in L’Aimant, par Lucas Harari
- Shecky R on Comment of the year

## Categories

Huh?

Nothing wrong with McCain:

http://fafblog.blogspot.com/2008/11/fafblog-inter…

Let's look at the superrich percentages cited:

"tax policies ranked last, with only 16% citing them as important. “Social issues” ranked first, with “policies dealing with wars” ranking second (67%)"

That's consistent with a sample of 6 (although 1/6 should really be 17%)

Someone identifying himself as Frank comments on WSJ: "To CP and anonymous. You’re absolutely right about the small sample size but such is the nature of all wealth surveys, given the demographic. As I’ve written before, these surveys are useful for directional trends (if findings are repeated), not as absolutes.

Comment by Robert Frank – October 13, 2008 at 6:17 pm

See: http://blogs.wsj.com/wealth/2008/10/13/the-rich-s…

Unlike this blog page [which contains a little dingbat next to Andrew's name if he replies], you can't be sure this is really posted by Frank.

Let's, for a moment, give Prince and Frank the benefit of the doubt. Let's say Price has modeled the %McCain versus net worth level and it shows a trend strong enough to make the point.

There's then the issue as to how to present this to an intelligent audience (WSJ readers) who nevertheless aren't stat experts.

What would be recommended here? Should be create deciles of the 493 people (so 49 in a group) and then run a smoothed line through these ten points? That's a little more complicated, but might work. But that means the top group isn't "super-rich".

ZBicyclist: It's hard for me to believe they'd have the chutzpah to report findings based on an n=6. I'm guessing that the 16% and 67% are just coincidences and it's really n=30 or something like that. In any case, though, the real question is what is the "purposive" sampling method they used. Especially if the person who did the survey knows what he's looking for, I'd think it would be easy to end up with an extremely biased sample.

I agree with you about the hazy definition of "purposive" — Prince is probably right that you can't really do a "random" sample of the super-rich, but "purposive" covers a multitude of sins.

Even in the more mundane world, it's one thing to pick a sample well. It's another thing to be able to execute it. For example this quote:

"survey cooperation rates from CMOR, the Council for Market and Opinion Research, averaged only 14.7 percent." (Jim Spaeth, Advertising Research Foundation, 2002)

The possible biases from a cooperation rate this low are huge.

I'm not trying to say that because there are problems with completion rates we should just forget about sample design — more the opposite, that we need to be sure we try to estimate as much bias as possible, and that sometimes the best you can do for a sample design is something that's not very good, and then footnote it. If you are going to study the super-rich, you are going to have to be creative.

It's possible Prince isn't being evasive with not revealing his method, and only treating it as a trade secret / proprietary method.