4816d039@news.povray.org...
> That's not very absurd at all, unless the elections were really close, 
> asking 12 totally random people would probably give the same result as 
> asking the whole population.  Raise that number to just 30 or 50 people 
> and you'd almost certainly get the same result every time.  If you're good 
> at stats you can work out the figures.

Actually no. A sample of 12 people is completely useless for surveying a 
large population (>1000). Typically, for a national population, a minimum 
size is 1000 (confidence interval of 3 and confidence level of 95%), so that 
you're 95% sure that the actual result lies between x-3 and x+3 where x is 
your survey result.  That's already a large interval (47-53 for a close 
election) and you'll need more for a smaller confidence interval (2500 for a 
CI of 2). And that's assuming that your sample is truly random and not 
biaised in some way, which is the big problem here since people aren't lab 
rats.
A sample of 12 would give an confidence interval of 30, i.e. the true answer 
will be between x-30 and x+30. Not very useful... Of course this never 
prevented people from deriving large trends from what their 12-people-large 
circle of acquaintances, including themselves + mom and dad, think or do ;)
Not that rounding up 1000 people for a single jury duty would be much 
practical, and the 12-people jury thing isn't meant to be a survey anyway.

G.

Post a reply to this message