turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

- Home
- /
- Analytics
- /
- Stat Procs
- /
- Mixed model analysis of experiment in GENMOD

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

07-16-2010 03:26 PM

My lab is conducting an experiment with a nested, repeated measures design, and would like feedback on the proper design for the analysis.

Each plot receives 3 treatments (food provisioning from a central feeder, order randomized among plots) over 3 consecutive sessions. Within each plot, a binary response (animal detected or not) is measured repeatedly (several checks) at points arranged at several distances from the feeder. There are multiple replicate points per distance, but the number of replicate points differs among distances. Points are numbered 1,2,3,... in each distance and plot. The experiment was run concurrently for all plots and, for all plots, the response variable (detection) is generally greater in the second and third sessions than the first. We're trying this as a GEE analysis in GENMOD, treating plot and point as random subjects, and treating treatments, distances, and session as fixed effects. The scientific hypothesis being tested indicates that the relationship between detection rate and distance from the feeder should differ among treatments.

Proposed analysis code:

*Proc Genmod;*

Class distance treatment session plot;

Model detections / n_checks = distance treatment distance*treatment session /link=logit dist=binomial;

Repeated subject = point(plot*distance) / type=exch;

run;

Questions:

1) Would "LOGOR=" be more appropriate than "type=" in this case?

2) Should we account for serial autocorrelation (type=AR1)?

3) Would we be better off using GLIMMIX? Pros / cons?

4) Are there other important questions that I'm not asking?

Thanks!

Each plot receives 3 treatments (food provisioning from a central feeder, order randomized among plots) over 3 consecutive sessions. Within each plot, a binary response (animal detected or not) is measured repeatedly (several checks) at points arranged at several distances from the feeder. There are multiple replicate points per distance, but the number of replicate points differs among distances. Points are numbered 1,2,3,... in each distance and plot. The experiment was run concurrently for all plots and, for all plots, the response variable (detection) is generally greater in the second and third sessions than the first. We're trying this as a GEE analysis in GENMOD, treating plot and point as random subjects, and treating treatments, distances, and session as fixed effects. The scientific hypothesis being tested indicates that the relationship between detection rate and distance from the feeder should differ among treatments.

Proposed analysis code:

Class distance treatment session plot;

Model detections / n_checks = distance treatment distance*treatment session /link=logit dist=binomial;

Repeated subject = point(plot*distance) / type=exch;

run;

Questions:

1) Would "LOGOR=" be more appropriate than "type=" in this case?

2) Should we account for serial autocorrelation (type=AR1)?

3) Would we be better off using GLIMMIX? Pros / cons?

4) Are there other important questions that I'm not asking?

Thanks!