Yes, definitely for tolerance intervals, you need a deeper
understanding of the distribution. Nonparametric tolerance invervals
require much larger samples to produce tight and stable results than do
parametric tolerance interval procedures -- and the validity will be
only as good as the parametric assumptions.
Thanks for that counterexample.
Best Wishes,
Spencer Graves
Brian S Cade wrote:
Perhaps the important thing to keep in mind here is that our data can often
deviate alot from normality yet inferences about means, changes in means,
etc. will not be too far astray. But start asking questions that require
estimating other statistics like percentiles (which are needed to estimate
prediction intervals for individual units or to estimate tolerance
intervals for a proportion of the population) then deviations from
normality can make a big difference. My experience is that many
biological/ecological questions would be more realistically solved by
tolerance interval estimates - how does some large (e.g, 90%) proportion of
the population respond.
Brian S. Cade
U. S. Geological Survey
Fort Collins Science Center
2150 Centre Ave., Bldg. C
Fort Collins, CO 80526-8818
email: brian_cade@usgs.gov
tel: 970 226-9326
|