somewhat compromised?
|
| Author | Content |
|---|---|
| dinotrac Sep 02, 2006 10:34 PM EDT |
Sigh. Self selected polls are not somewhat compromised. Aside from being fun for the participants, they are worthless. I take that back -- presuming there is some mechanism to prevent double-voting that actually works (there isn't) -- they will tell you about the people who took part in the poll. The reason is simple: random selection is the key to extrapolating any poll out to a larger population. It is what allows you to calculate all those nifty plus or minus three percentage point (or whatever) margins of error, etc. Knowing that the distribution of a characteristic in the population is random and knowing that you have randomly selected your sample allows you to calculate the probability that your sample is representative of the population at large. |
| jimf Sep 02, 2006 10:50 PM EDT |
Interpretation of polls by various 'experts' makes it even worse. Take a questionable poll, spin it around a bit, and, you can make it say whatever you please. As an example Mr Smith says, 'When OpenSuse and Novell's Suse Linux Enterprise Desktop are taken together Suse Linux with 13 percent of the votes cast gets just ahead of Debian.' One could also say if you add all the Debian derived distros together, you get well over 40%. Gee, that's cool... So, does either interpretation really tell you anything. I don't think so. |
| jdixon Sep 03, 2006 8:23 AM EDT |
> Knowing that the distribution of a characteristic in the population is random and knowing that you have randomly selected your sample allows you to calculate the probability that your sample is representative of the population at large. Which is, of course, a major problem with most polls. Neither Linux usage nor more general political leanings are randomly distributed. Political leanings in particular are highly geographicly dependent. Which probably explains why the major polls have underpredicted the Republican presidental vote by about 5% or so during my lifespan. |
| dinotrac Sep 03, 2006 9:39 AM EDT |
jdixon - Yes and no. Yes, they are randomly distributed, but, yes again, you have to pick your population correctly. The most striking example of that, I think, was the exit poll fiasco in the last election. In the morning, Kerry looked like a convincing winner. In the morning, however, east coast voters were prevalent -- as were students and other people who didn't have to rush off for the office. And, of course, exit polls have a self-selection bias, also: of those who vote, they can only talk to the people who will take the time to talk to them. Not much opportunity to correct for refusal, either. |
| devnet Sep 03, 2006 9:42 AM EDT |
So, Lxer should have a poll to end all polls requiring registration, one email/IP per registrant on what distro reigns supreme among its readers. That would be fun :D |
| dinotrac Sep 03, 2006 10:14 AM EDT |
devnet - Nah. We should do it Chicago style: stuff the ballot box for whomever provides the right price. |
| lordshipmayhem Sep 03, 2006 10:38 AM EDT |
"We should do it Chicago style: stuff the ballot box for whomever provides the right price." I thought that was the "Get the Facts" TCO Whitepaper style? |
| dinotrac Sep 03, 2006 11:09 AM EDT |
>I thought that was the "Get the Facts" TCO Whitepaper style? Amateurs. |
| Scott_Ruecker Sep 03, 2006 4:27 PM EDT |
It would be nice to have a poll that only allowed for one vote for each visitor. It would be interesting to poll the users of and visitors to LXer, to see what there viewpoints are on certain issues, what OS they use, plan to use, and all that jazz. Even then, the margin for error would be relatively high, someone can create multiple user-names and I am sure there are other ways around any limitations that could be enforced. Still, if it could be done, would you participate in it? I am not trying to call anyone out, just seeing if it might be something worth doing here. |
| devnet Sep 03, 2006 5:46 PM EDT |
I would participate. I'd like to see the results too. :D |
| jdixon Sep 03, 2006 8:28 PM EDT |
> Yes, they are randomly distributed, but, yes again, you have to pick your population correctly. Agreed. The distribution of the parties in a given region will be random, but unless you take into account the percentage variantions in a geographical area, you won't get a valid result. From the historical results, I'd say the national polls are doing a very poor job of accounting for those types of variations. As I noted above, they have consistently underestimated the Republican vote, usually by about 5%. I can't say if the same is true for more regional polls, but I suspect not. I suspect it's a simple city/rural divide, and that the national pollsters simply aren't properly allowing for the greater Republican trend of rural areas. |
Posting in this forum is limited to members of the group: [ForumMods, SITEADMINS, MEMBERS.]
Becoming a member of LXer is easy and free. Join Us!