topic Re: Statistical Procedure for longitudinal population in Statistical Procedures

Statistical Procedure for longitudinal population

desireatem — Mon, 19 Jul 2021 15:11:45 GMT

Hello -,

I have this longitudinal data with population as outcome, I wish to ask the best statistical procedure. Treatment, sex, Pre/post, time are covariate and population is the outcome.

Treatment	Sex	Pre/Post	time	Population	id
Yes	Female	Pre	1	196	1
Yes	Female	Pre	3	132	1
Yes	Female	Pre	5	216	1
Yes	Female	Pre	10	520	1
Yes	Female	Post	1	2021	2
Yes	Female	Post	3	2551	2
Yes	Female	Post	5	3050	2
Yes	Female	Post	10	4811	2
Yes	Male	Post	1	6777	3
Yes	Male	Post	3	1165	3
Yes	Male	Post	5	2142	3
Yes	Male	Post	10	3135	3
No	Female	Pre	1	5376	4
No	Female	Pre	3	12736	4
No	Female	Pre	5	15278	4
No	Female	Pre	10	16896	4
No	Female	Post	1	68548	5
No	Female	Post	3	74211	5
No	Female	Post	5	101223	5
No	Female	Post	10	30569	5
No	Male	Pre	1	12141	6
No	Male	Pre	3	13280	6
No	Male	Pre	5	13869	6
No	Male	Pre	10	41440	6

Re: Statistical Procedure for longitudinal population

Ksharp — Mon, 19 Jul 2021 12:13:27 GMT

PROC GLIMMIX

Re: Statistical Procedure for longitudinal population

Rick_SAS — Mon, 19 Jul 2021 14:43:29 GMT

The ID for the last observation looks wrong. I think you intended ID=6.

Re: Statistical Procedure for longitudinal population

desireatem — Mon, 19 Jul 2021 15:12:55 GMT

Thank you! Will the GLIMMIX with identity link works?????

Re: Statistical Procedure for longitudinal population

Ksharp — Tue, 20 Jul 2021 12:12:13 GMT

You could check option " DIST=possion | gamma" .

Or calling experts @StatDave @lvm @SteveDenham

Re: Statistical Procedure for longitudinal population

SteveDenham — Tue, 20 Jul 2021 12:22:54 GMT

I think a gamma distribution with a log link or a lognormal distribution with an identity link are probably best here, as the values seem well bounded away from zero.

SteveDenham