How many to select?

[bartholomew]

The sample size has to be 100 patients (N=100).
The population is composed of 80% of White people and 20% of non-White.
The aim is to compare the two groups' reaction to a new drug.
How many people would you advise I should select from the first
group and the second one?

[fo3nix]

if you choose to test 80 non-whites, and 20 whites, it'll compensate for the difference:

20% * 80 = 16
80% * 20 = 16

[WaldoMcFroog]

What is the population you're measuring? Unless you know that, how can you know whether your sample size is suitable? I assume you're going for a 95% confidence level (This means you're 95% confident that your responses are an accurate reflection of the population as a whole).

There's a sample size generator here:

http://www.cawalker.com/tools_calc_samplesize.html

If you go want a 95% confidence level with +/- 1% error (this is the default for most surveys) , and a population of 1000, you need to interview 906 people, of which 725 would need to be white, 181 non-white.

[WaldoMcFroog]

Sorry - an addition and a correction:

If you allow for a 5% margin of error (and to correct my previous comment, 95% with +/- 5% is typical) you would have to interview 278; 222 white, 56 non-white. If you increase the margin of error beyond that, you're going to be using statistics that are so unreliable you might as well not bother.

[Peter Pedant]

Hi bart - you're not having very good luck with stats are you ?

Presumably you are random sampling in a 80.20 population and you want your sample to match 80.20 and not be something else.

Surprsingly this is a quite a complex question. MORI and GALLUP were the first to tackle this question and whether it was acceptable to go out and get a few Irish, if your quota was a lacking the Irish.....in order to balance up numbers. I thought it was pretty obvious that selection in this way scre*ed up randomisation .

I think you need proper advice.....