In article <o7IiCDAWaQ$yMwen@bmco.demon.co.uk>, Heidi Tang <HAT@bmco.demon.co.uk> writes: > In article <1997Feb8.140954@mint.hnrc.tufts.edu>, Jerry Dallal > <jerry@mint.hnrc.tufts.edu> writes > >>It means that the assumptions upon which the method for calculating the >>CI is based are not satisfied by the data. Typically, the sample size >>is too small and/or the value of th eparameter is too extreme. In many >>cases there are other methods that avoid these problems. > > Many thanks for the reply. > > I wonder whether I dare ask one more question?

Certainly . . . or even 2!

> If a negative lower CI limit indicates that the sub-group size is too > small to analyse, what other method might be applicable (in this > instance, sample size was only 3 and standard deviation large). How > large should sub-group size be before 95% CI can be used - otherwise, > what tests might I try? <snip giving details of problem>

Perhaps someone else will tackle this. It's not an answer I can type in a few lines. A small sample analysis will have to formally incorporate the limits on the data. One rule-of-thumb (other than n=30!) is that the sample size is large enough if the CI lies within the permissible range of the data! However, its application falls under the banner, "Don't try this at home!" Or, in a more serious vein, you've got a problem that is not straightforward. It requires the hands-on assistance of someone with experience in these types of situations. Small sample solutions will always depend on what you can say about the underlying distribution of the individual data.