Andrew Preston wrote: >On Tue, 18 Mar 2003 10:55:10 -0500, "John Hayes" <[EMAIL PROTECTED]> >said: > > >>> Such a poll would be useless given selection bias issues inherent to >>>nonrandom sampling.... >>> >>> >Would you expand on this a little? > In statistics, people frequently focus solely on the number of observations in the sample (eg how many people did you poll.) However, to have external validity, you need both a large enough sample *and* an unbiased sample. To protect against a biased sample, we go to great lengths to make sure we draw a random sample. A small unbiased sample is better than a large biased sample. Maybe an example would help.
If I want to know how American's feel about gun control, I can't ask everyone because that would cost too much and take to much time. So instead, I decide to take a much smaller sample and extrapolate to the general population. But to have external validity, that is, to have confidence in generalizing from my sample to the population, I need an unbiased sample. If I call 2000 NRA members and ask them how they feel about gun-control I'll get a very different answer than if I use a stratified random sample and call just 200 adults from every state and socio-economic bracket. It should be obvious that most ethical pollsters wouldn't intentionally select a biased sample. However, sometimes the factors which induce such biases are hidden, hence the great lengths we go to ensure a truly random sample. Which brings me to the problem with web polls. By their very nature, they suffer from 2 separate flaws. First, because participation requires some effort on the part of the participant, it tends to select for the extreme views on either end of the spectrum and underestimates the middle because only individuals with strong opinions on either side can be bothered to take part. Second, a poll on the journeytoforever website is a non-random sample due to the nature of individuals that choose to visit that type of site in the first place. Thus, it has what we would call "limited external validity". That is, is would do a great job of telling you what the opinions of the people that responded but you can't generalize from those results to a population. Without the ability to generalize, it has little utility and, as such, is a waste of time. Given that most reasonable readers of the list would probably be willing to stipulate that more than 50% and less than 80 or 90% of the list readership is opposed to this war, it makes little sense to expend the effort to take a poll. Put another way, even if we did a complete census, that is, asked for and got a response for every single list member, what would we gain? Knowing whether 25, 15, 5 or even 0.1% of the readership supported the war wouldn't change the way we interact on the list or change our opinions of biofuels, so why bother. John ------------------------ Yahoo! Groups Sponsor ---------------------~--> Your own Online Store Selling our Overstock. http://us.click.yahoo.com/rZll0B/4ftFAA/46VHAA/FGYolB/TM ---------------------------------------------------------------------~-> Biofuel at Journey to Forever: http://journeytoforever.org/biofuel.html Biofuels list archives: http://archive.nnytech.net/ Please do NOT send Unsubscribe messages to the list address. To unsubscribe, send an email to: [EMAIL PROTECTED] Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/
