Skip to content

How to test a difference in ratios of count data in R

1 message · Greg Snow

#
It is usually best to keep these discussions on the list.  Someone
else may have a better answer than mine, or be able to respond
quicker, and if I answer on R-help then it is community
service/involvement.  If I respond directly then it is consulting and
then we need a contract and I have to charge you (not that I would see
the money myself, but it would help my budget be a little less red).

For your question, your fit4 and my fit2 are just 2 different ways of
fitting the exact same model to the exact same data, so there is no
surprise that the results match.  Which one to use is personal
preference.

The line that starts with "treatmentB" is the coefficient
(log-odds-ratio) for B compared to A, so that is the main line to look
at for interpretation.

The correlation of the fixed effects is mainly there for diagnostics,
if it is too close to -1 or 1 then that indicates that assumptions may
not hold, or computations may be in doubt.  Your value is not of
concern.
On Wed, Sep 28, 2016 at 2:14 PM, Shuhua Zhan <szhan at uoguelph.ca> wrote: