Posted by Peter Ceresole on October 21, 2003 at 23:04:39 from 80.177.22.49 user PeterC.
In Reply to: Re: Social Obstacles, RP & good old Cook posted by Prue Eckett on October 21, 2003 at 20:37:05:
I beg to differ Peter. As a former B.Ed student we had to complete surveys using both contemporary statistical methods and random sampling. Quite different results eventuated, and our professor commented anyway that no survey is ever completely accurate given the human conditions of alternative agendas, lying and simply answering anything to get rid of the surveyer.
As usual, it depends on the skill of the surveyor and the accuracy you're after. It is perfectly possible to frame questions that will detect and correct for those activities and the larger the survey, the lower the uncertainties. It just takes more skill and costs more money. And if the run of results over many years by different organisations tend to reinforce each other, that's a good indication that they are usually pretty robust. Certainly a great deal better than wild guesses and opinions.
If the survey is intended to discover information rather than to make a point, it is a very good tool. For instance, checkable surveys include public opinion surveys at election time. And despite occasional errors in the overall result (mostly they're amazingly accurate), due to the British electoral system having more to do with getting a clear majority than with delivering the choice of the electorate, the predictions of the voting numbers have always been within a couple of percentage points of the outcome. That's pretty good.
This is usually ignored by commentators who know nothing about it, but it's the only thing that an honest pollster will claim to be able to achieve.