Need Complex Ordinal Analysis

#1
My question outlines the need for a non-parametric analysis of ratio data.

I have a data set composed of apnea counts for individual mice during different breathing conditions and a drug. This boils down to a three factor ANOVA style data set

Factor 1 -> Mouse Strain
Factor 2 -> Gas Challenge
Factor 3 -> Drug

The data are no where near parametric. Is there an non-parametric/ordinal test that could be used in this case?

Note: We need to test factor interactions.
 

CowboyBear

Super Moderator
#2
I have a data set composed of apnea counts....

The data are no where near parametric.
It can be more helpful to think about the specific characteristics of a variable, rather than just trying to broadly classify it as parametric or non-parametric. In your case your response variable is a count variable, so rather than hunting for a non-parametric test, it might be more helpful to find a model specifically suited for count data.

(I'm assuming that your apnea counts are counts of the number of episodes of apnea for each mouse?)

Two count models that might be useful are the Poisson and the negative binomial. You could still do factorial type analysis, but where an ANOVA would assume a normal distribution of the response variable within each condition, you would instead assume a Poisson or negative binomial distribution. Note that these are parametric models, but the parametric assumptions being made are ones that are plausible for count data.
 
#3
Thank you!

The apnea counts, are the number of apneas for each mouse over a series of gas challenges.

I'm not immediately familiar with poisson and negative binomial, but I'm a quick study. Can you suggest a platform for poison or fractal analysis?
 
Last edited:

CowboyBear

Super Moderator
#4
Can you suggest a platform for poison or fractal analysis?
I don't know anything about fractal analysis, sorry. You can fit a generalized linear model with a Poisson distribution ("Poisson regression") in pretty much any mainstream stats package - e.g., SPSS, R, SAS, etc.
 

CowboyBear

Super Moderator
#5
Can you suggest a platform for poison or fractal analysis?
I don't know anything about fractal analysis, sorry. You can fit a generalized linear model with a Poisson distribution ("Poisson regression") in pretty much any mainstream stats package - e.g., SPSS, R, SAS, etc.