Thursday, May 30, 2013

Polls of Journalists

There's a story about an Australian poll of journalists that deserves a brief mention.  A lot of it is caught up in that country's politics, which won't mean a hell of a lot to U.S. readers, but there is one small methodological point I want to explore. See below:
On the topic of voting intentions, only 61 per cent of the 605 respondents were willing to reveal their voting intention, but of those who did, 43 per cent said they would vote Labor, 30 per cent Coalition and 19 per cent Green.
For a lot of us those party affiliations mean little.  The real point here is only 61 percent gave their voting intentions and from that small subsample, we have a list of partisan preferences.  So basically we're operating with a non-random sample of 369 journalists, because there's no reason to suspect people randomly choose to give their political preference.  A number that small, the margin of error balloons.  Plus it's become so non-random as to be questionable whether margin of error (about 5 percent here) even matters.

In other words, beware.

Now you could do a quick-and-dirty analysis to see if those who agreed to reveal their party preferences differed from those who did not on such factors as age, education, etc.  If you did not see significant differences there, you have more confidence in the meaningfulness of your results.  The story here doesn't go that deep, so it's hard to say.  The real point is, beware subsamples when people assign themselves through a willingness to reveal something about themselves.  My hunch is the journalists who didn't reveal their partisan leanings differ wildly from those who did -- and that throws the whole analysis out the window.

No comments: