10-05-2015 04:41 PM
Hi, I'm trying to assess the changes of my categorical data (MOV; with 5 ordinal scores) over time (day; 5 days) in 3 differents groups of animals (group). So, each animal whithin of each group was recorded by MOV score over 5 repeated days. I would like to know if the animals in each group has changed from score 1 to 5, for example over the days.
I know that is posible to to this using PROC GLIMMIX with day as random effect, but I don't know how. Does anybody know how to do this?
10-06-2015 08:39 AM - edited 10-06-2015 08:40 AM
I would like to know if the animals in each group has changed from score 1 to 5, for example over the days.
Is this literally what you mean, did the score change from 1 to 5? Or do you mean did the mean score in each group change? Or do you mean something else?
10-06-2015 01:45 PM
Hi, the variable MOV consisted of 5 scores (1: animal stood still for the entire assessment period; 2: animal stood still for most of the assessment period; 3: animal reacted calmly at least half of the assessment period; 4: animal reacted intensely more than half of the assessment period; 5: the animal jumped, raised its limbs off the ground). So, I would like to see if the animals changed their scores over time (day 1, day 2, day3, day 4, and day5) by each breed (breed 1, 2 and 3).
My appologizes for my poor English.
10-06-2015 03:49 PM - edited 10-06-2015 03:50 PM
My answer is: I don't know what the best analysis is.
PROC GLIMMIX does not have a REPEATED statement, so you could include the days into the model as an ordinary (not repeated measure) factor.
PROC GLM has a REPEATED statement, and so you could consider the response to be continuous and use PROC GLM.
PROC CATMOD has a REPEATED statement, and allows you to handle your response as ordinal rather than continuous; this seems to come the closest to giving you the proper analysis, but I guess I'd need to think about using this some more.
10-06-2015 04:56 PM
Thank you for your answer, but I think that is not apropriate consider MOV score as continuous data. The distribution is not even close of normal distribution.I saw on this book (Generalized Linear Mixed Models, Walter W. Stroup, 2012) that I can use proc glimmix with multinomial distribution, but I don't know yet how to do the model. I saw that could be something like this:
proc glimmix data=mydata;
class animal breed day;
model mov (order=data) = breed day breed*day / dist=multinomial solution oddsratio;
random intercept / subject = animal;
But I am not so sure if is right and I know that I missed something in this model.