LISTSERV at the University of Georgia
Menubar Imagemap
Home Browse Manage Request Manuals Register
Previous messageNext messagePrevious in topicNext in topicPrevious by same authorNext by same authorPrevious page (April 2007, week 2)Back to main SAS-L pageJoin or leave SAS-L (or change settings)ReplyPost a new messageSearchProportional fontNon-proportional font
Date:         Sun, 8 Apr 2007 21:37:23 -0700
Reply-To:     David L Cassell <davidlcassell@MSN.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         David L Cassell <davidlcassell@MSN.COM>
Subject:      Re: Help with NLMIXED or GLIMMIX for multilevel GLMM
In-Reply-To:  <>
Content-Type: text/plain; format=flowed

qxu@UCDAVIS.EDU wrote: > >Dear all, > >I have a time allocation multinomial response data set in Anthropology. It >was a two year study on twenty families: from A to T; each family had >unequal number of individuals, for example, family A had individual A-1, A- >2, A-3, and family B had B-1, B-2. Each individual was observed by >anthropologists every six days at a random time during that day. Their >activities were recorded. For example, there are three activity codes: A- >doing agriculture work, L-Leisure, M-Marketing (there were more categories >in the original data, up to 26). The response will look like: >A-1: A A A M L ... (over two years) >A-2: A M M L M ... >. >. >. > >Now I want to fit a generalized linear mixed model for this response >against fixed effects, such as gender, age, with random effect for family >and person. (since each person has repeated observations over 2 years and >individuals were clustered in families.) >I attampted to fit the model by GLIMMIX using multinomial distribution and >glogit link, but convergence could not be reached. I could not figure >outhow to fit more than one level random effects in NLMIXED either. >Any suggestion is highly appreciated.

[1] How were the families selected? Isn't this really a survey sample? So isn't this something other than a multilevel GLMM? It looks to me (from way over here) like a simple cluster sample with lots of data over time. So it may be that you should be using PROC SURVEYLOGISTIC instead.

[2] If I'm right, then the random effects you perceive are simply survey sampling structure, and you have no random effects to handle in your model.

[3] Before I can get much further, I really need to know what your hypotheses are, and how you are testing them. It may be a royal pain to analyze as a survey sample model, or it may be a breeze. It depends on your hypothese and your data usage.

HTH, David -- David L. Cassell mathematical statistician Design Pathways 3115 NW Norwood Pl. Corvallis OR 97330

_________________________________________________________________ Get a FREE Web site, company branded e-mail and more from Microsoft Office Live!

Back to: Top of message | Previous page | Main SAS-L page