I am accomplishing some hypotheses tests related to a growth model (y =
a*exp(bt)) to calculate adjusted growth rates. We can interpret the
significant values of the growth rates easily, considering the p-values
as being the probabilities of type I errors.
But when the values are not significant, how I should treat them and to
interpret them statistically? I cannot simply despise them because they
are the best estimates, in spite of they be not significant values.
Evidently they cannot be considered as being null values of growth rates
because the type II errors exists and we have to analyze the power of
the tests. My problem is as proceeding to an analysis of power of these
tests, considering that the data come from a sample in three stages ("
cluster sampling "). How should I treat the " effect size " in this case
to analyze the power of a certain test? Is it possible to have, for
example, an answer for each non significant rate of the type " the rate
is not significant but does it exist a probability of beta percent that
it has been generated by a value of the population the same to the
value of Ho plus the " effect size "? Which would the best treatment be
for this problem and for this test? There is some program that works
with this test type executing the power analysis?
Respectfully
Henrique Dantas Neder
Uberlāndia Federal University - Economics Institute - Brazil
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|