🚀 go-pugleaf

A colleague needs to adjust a bunch of t-tests in a hurry -
the journal wants the MS back yesterday!
He has measured what percentage of department budgets go
on each of five areas, once this year and once last year.
He wants to look at change in percentage allocation of
funds. Now, doing 5 t-tests, with a Bonferroni correction,
still leaves two problems:
.1. A percentage is the ratio between two variables that
may not be independent. I vaguely remember a tweak you can
do based on the covariance - is this necessary here? Does anyone
have the formula?
.2. If you add the five percentages, they must come to 100, so
in fact only four of them can vary before the last one is fixed.
Is it sufficient to subtract N/5 from the degrees of freedom to
take into account the lower-than-normal number of parameters
needed for a model that completely specifies the data?

There's bound to be some bells-and-whistles method that cracks
this whole problem, but this unfortunate just needs to patch a
hole in a small and unimportant area of a larger paper and
consequently just wants to do the minimum. With the tweaks
above, is this an acceptable minimum?

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
+ "Apres l'heure c'est l'apres-heure, mais apres Loire,       +
+  c'est l'Eure et Loire"    J P Sartre                       +
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Ronan Conroy
Lecturer in computing and         Royal College of Surgeons in Ireland
biostatistics                     123, St Stephen's Green, Dublin 2.
                                  (353)-(1)-784422 [fax:780934]