🚀 go-pugleaf

In article <1992Apr15.233135.2693@newshost.anu.edu.au> Andrew.Robinson@anu.edu.au writes:
>In article <OLIVER.92Apr15054328@ballard1.mit.edu> oliver@athena.mit.edu (James
>D. Oliver III) writes:
>>I was using a packaged statistical program to do some simple means testing,
>>and notice that in several cases the program didn't do a straightforward
>>t-test, but instead "computed the standard error of the difference of means
>>assuming IQR(pop.1)/IQR(pop.2)= X".
>>
>>I found out what IQR is (interquartine(sp?) range) and have a general
>                          ^^^^^^^^^^^^^  [interquartile]?
>>understanding of what it is, but can find no references on the specifics of
>>this particular adjustment and/or the justification for using it.
>>
>Sounds to me like the package is trying to determine the relative spreads of
>the populations because the t-test fails when testing on populations with
>differing variances.  The ANOVA does as well, but I don't know of a comparable
>procedure.  In fact, that's the first time I've heard of a package which tries
>to do this.  Nice.
>
>But I have _no_ idea where to go from here.
>
If indeed the variances are so different the package sees the t-test as
inappropriate, you might consider looking at the distributions
themselves, then looking toward some applicable non-parametric test.