Missing data methods in longitudinal studies - Milk Production - Dairy Farms

[Rodolfo_Mingoti]

Hello, I would like to help for analyzing of database in dairy farms (Longitudinal analysis).

In my database, I have daily milk production for 115 days. However, some cows stop producing milk at 20d, 35d, 80 days (random days). Therefore, as time go on, naturally the cows production less milk until to stop completely the milk prodution.

Ex: On the first day I have 1000 cows producing milk. At the end of the 115 days, only 150 cows remain producing milk.

My greatest doubt occurs when an animal stops producing milk. As the cows are removed of the database (cows stop producing milk) the average from this moment changes (increases) causing to curve error (because the cows less productive stop the production).

How can I solve it?

PROC GLIMMIX DATA=Dairy.Cows;
	CLASS ID Group Time;
		MODEL MilkProduction = Group Time Group*Time / DDFM=KENWARDROGER DIST=GAUSSIAN ;
			RANDOM Time / RESIDUAL TYPE=UN SUBJECT=ID(Group);
RUN;

[Norman21]

It looks like you might need a "time to failure" analysis, described in some detail here:

https://stats.idre.ucla.edu/sas/seminars/sas-survival/

[Rodolfo_Mingoti]

Hello Norman21, firstly thanks for your attention. I saw your recommendation about survival analysis using Proc Lifetest but unfortunately, that procedure is not that I need.

The topic listed is very close to what I need, but do not have the solution yet: https://communities.sas.com/t5/Statistical-Procedures/Using-PROC-NLIN-to-fit-curves-for-dairy-cattle...

In my database, after the middle of lactation the numbers of cows decrease. The begin of lactation (day 1) I have 1000 observations (cows) and the end of lactation (day 115) I have only 150 observations. The response variable is milk_production (continuous variable).

Some people said that I can not compare the groups, because I have not completed basedata. The solution for that is filter the dataset just for animal that have complete lactation. Using this methodology, I will lose information about 750 cows.

My doubt is: Are there another way to solve it?