🚀 go-pugleaf

In article <STAT-L%93072610012027@VM1.MCGILL.CA>, Robyn Lee <lee_rb@MRICD1.APGEA.ARMY.MIL> writes:
> To Stat-L members,
>       I have been confronted with yet another interesting challenge.  I have
> an entemologist that is using pesticides on field strain and susceptible
> strain cockroaches to test if there is increased resistance to the pesticides.
> Currently, nine batches of ten roaches each are placed in containers with the
> pesticide.  They are observed every 15 minutes and the numbers of dead roaches
> are counted at each observation time.  When graphing the cumulative number of
> dead roaches by observation time, the graph makes a fairly good sigmoid curve.
> A probit approach is currently being used, but I know that it is not
> appropriate since the counts at the observation times are not independent. So
> how should I proceed from here? My entemologist wants to be able to compare
> the response from the field strain to the susceptible strain as well as
> compare how the field strain may change over time ( more exposure to the
> perticide). In addition, how might I calculate whether the nine groups of ten
> is an adequate sample size?
> Robyn Lee
> lee_rb@mricd1.apgea.army.mil

One approach would be to directly model the survival time
[or tolerance to time!] distribution, on which you have
grouped [and possibly censored] observations. A normal
or other appropriate distribution could be fitted by
likelihood methods. Various different models could be
compared by likelihood ratio tests.

For planning sizes of treatment groups it may be
easiest just to ignore the grouping and use midpoints
for the survival times of each cockroach.
-- 
Murray A. Jorgensen [ maj@waikato.ac.nz ]           University of Waikato
Department of Mathematics and Statistics            Hamilton, New Zealand      
-------------------------------------------------------------------------
    Theory eleison     *    Mystery eleison     *   Theory eleison