topic How to perform Ordered logit model? in Statistical Procedures

How to perform Ordered logit model?

Education — Wed, 07 Jun 2017 19:22:58 GMT

I conducted an experiment measuring the libido of my animals as a score from 0 to 3 in different season. i collected **bleep** for 11 days consecutively during each season now i want to see if season has an effect on libido. This is what i did:

Proc glimmix data= semendata;
class maleid age season;
model libido= season age season*age/ dist=multinomial link=cumlogit;
random maleid;
run; quit;

I found that there is a significant differences between seasons but when i want to perform the mean separation SAS gives me this error:

ERROR: Least-squares means are not available for the multinomial distribution.

Now my question is, is this the right way to run the odered logit model for such kind of data and secondly if it is, then how do i continue to compare the differences between the seasons since Type 3 revealed a significanct difference?.

Any help will be appreciated.

Thank you

Re: How to perform Ordered logit model?

Rick_SAS — Wed, 07 Jun 2017 19:34:02 GMT

A multinomial model just means that the response is one of k categories. The response could take on the values "red", "green", and blue, which is why least squares means do not always make sense.

An ordinal response is a special kind of multinomial model. To learn more about the various ordinal models and link functions that you can use in SAS, I recommend High (2013) "Models for Ordinal Response Data." The author shows several ESTIMATE statements for each kind of model.

Re: How to perform Ordered logit model?

Education — Wed, 07 Jun 2017 21:19:52 GMT

Hi Rick

Thank you so much for the paper you recommended. From reading that paper i added the ESTIMATES as follows:

Proc glimmix data= semendata;
class maleid age season;
model libido= season age season*age/ dist=multinomial link=cumlogit;
random maleid;
Estimate "summer vs winter " season 1 1 -1-1;
Estimate "spring vs winter" season 1 1 -1-1;
Estimate "autumn vs winter" season 1 1 -1-1;
Estimate "summer vs spring" season 1 1 -1-1;
Estimate "summer vs autumn" season 1 1 -1-1;
Estimate "spring vs autumn" season 1 1 -1-1;
run; quit;

And the results of the estimates all looks identical like this:

SAS Output

Estimates Label: Estimate summer vs winter: 0.6118	Standard Error 0.7608	DF 422	t Value 0.80	Pr > \|t\| 0.4218
spring vs winter: 0.6118	0.7608	422	0.80	0.4218
autumn vs winter: 0.6118	0.7608	422	0.80	0.4218
summer vs spring: 0.6118	0.7608	422	0.80	0.4218
summer vs autumn: 0.6118	0.7608	422	0.80	0.4218
spring vs autumn: 0.6118	0.7608	422	0.80	0.4218

Hence, below is my Type 3 results:

SAS Output

Type III Tests of Fixed Effects

Effect: Num DF

Season: 3

Den DF

422

F Value

26.48

Pr > F

<.0001

age: 1

422

0.11

0.7428

age*Season: 3

422

14.24

<.0001

I am abit worried about the identical estimates between seasons, did i do it the right way?

Thank you!