jhc (03/22/83)
I have heard (anecdotes again) that there are fewer than 2000 Nielson homes (I heard 1500, but this could have been under-rated). Doesn't sound like a good sample size to me, but then I don't get billions of dollars every year in advertising revenue for pumping trash onto the airwaves 24 hours a day (I exaggerate, very slightly). JOnathan CLark ABI Holmdel [houx*|lime]!hou5a!jhc ~v <blast...>
smb (03/23/83)
1500 is around the standard sample size for most national opinion polls; I suspect that they know what they're doing. The question, though, is whether they have adequate samples for smaller subgroups. --Steve
tihor (03/25/83)
#R:hou5a:-31600:cmcl2:11000001:000:615 cmcl2!tihor Mar 24 21:03:00 1983 Unfortunately I don't have my statistics books at the office but as I recall frpm my Politics and Polling class at Princeton the number of people needed for a good nationwide poll (and indeed for alomst any subgroup of a million or more) was about 3000 at the 5% confidence level. (There are several fairly elegant theorems which show that the number of members need from an IID population need goes asymptotic for a fix confidence level as the size of the underlying population goes to infinity.) The NY Times Polls which my prof was advising in the late 70s went for about 7.5% at (as I recall) ~2200 samples.
wapd (03/26/83)
Requiring a sample size of 3000 people for a 5% confidence level on a national survey doesn't agree with what I have seen in Time magazine. Their surveys are always somewhere around 1000 people, and they usually claim a 2-3% error margin. Maybe they claim to be sampling registered voters only ? Is a "2-3% error margin" the same as a "5% confidence level" ? I only took one statistics course and I hated it. Bill Dietrich houxj!wapd
mcewan (04/01/83)
#R:hou5a:-31600:uiucdcs:19200008:000:89 uiucdcs!mcewan Mar 31 14:12:00 1983 Nielson works at a *33%* confidence level. Doesn't inspire too much confidence, does it?