RE: ID cards - What wasn't mentioned about the results of the consultation exercise

["Owen Lewis" <[EMAIL PROTECTED]>]

> -----Original Message-----
> From: [EMAIL PROTECTED]
> [mailto:[EMAIL PROTECTED] Behalf Of Andrew McLean
> Sent: 01 November 2004 23:21
> To: [EMAIL PROTECTED]
> Subject: Re: ID cards - What wasn't mentioned about the results of the
> consultation exercise
>
>
> In message <[EMAIL PROTECTED]>, Roland
> Perry <[EMAIL PROTECTED]> writes
> >The American pollsters on TV this morning are saying that a *carefully
> >selected* representative sample of 1000 voters [asked one question]
> >will result in a +/- 3% accuracy. As one of then said "if the result
> >appears to say Bush will win by 2%, then problems with the sample could
> >mean that actually Kerry might win by 1%".
> >
> >They then said that to get 1% accuracy you'd need a sample of 3,000,
> >which is suspiciously linear, and therefore I doubt it.
>
> You are right to doubt it. Assuming the sample is a small fraction of
> the total population, then the error due to sampling effects is, to a
> good approximation, inversely proportional to the *square root* of the
> sample size.

Could you give an example of that? E.g. Population is 75,000. Sample size is
2,500. Sqrt of sample is 50. So forecast accuracy is? And when sample is the
same but the population is 750,000,000?

> Of course, this error estimate is an underestimate, as it
> completely ignores the systematic effects due to not having a
> representative sample.

This is the devil, isn't it? What *is* a representative sample? Is a sample
representative in different places and at the same time? Or in the same
place at different times?

Owen