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

[Roger Hird <[EMAIL PROTECTED]>]

In article <[EMAIL PROTECTED]>,
   Owen Lewis <[EMAIL PROTECTED]> wrote:
> >
> > It doesn't matter what the population size is.

> That's where I suspect that reality and theory may part company. If I
> try to balance a sample of 2500 from a population of 75,000 (a large
> town), there's a fair chance that I can do that. Balance the same sized
> sample for a population of 750,000? ..... Glad I don't have to try.

That's the skill polling companies have developed.  They have ways of
constructing samples by crude bandings (eg by sex, age, class) to ensure
approximation to a true random sample.  But the sqrt of the sample size is
only the MINIMUM error (crude terminology, I know: its the standard
deviation, if I remember) from a perfect sample if everything else is
right.  Getting the sample wrong makes the error bigger. In political
polls, people lying does too, as does not knowing what "don't knows" will
do, not knowing whether or not people will actually vote, etc etc.  

It does work very well with measurements of radioactive decay!

-- 
Roger Hird
[EMAIL PROTECTED]

Running Voyager 2.07 and RISCOS 4.02 on an Acorn StrongARM RiscPC