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
- /
- Glimmix: Using covtest to compare the matrices for...

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

04-10-2015 10:20 AM

When using Glimmix, it is possible to test if the matrices for the R side random effects for different treatments are equal. For example, if you measure three variables per individual (id), with the individual receiving one of two treatments (for example high or low food) this would be done by including:

random _residual_ / type=un group=treatment subject=id;

Covtest homogeneity;

Now say that individuals belong to different families (g side random effect):

random variable / type=un group=treatment subject=family;

random _residual_ / type=un group=treatment subject=family(id);

Covtest homogeneity;

Question 1:

In the second code, what is the covtest statement actually requesting? Will this just show if the matrices for the r side random effects are different for each treatment?

Question 2:

I would like to know specifically if the matrices for the G side random effects (family) are different for each treatment, is it possible to specify this? Is the only way to do this by using ‘covtest general’ and then specify elements from the ‘Covariance Parameter Estimates’ table you want to compare?

Accepted Solutions

Solution

04-10-2015
01:39 PM

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

04-10-2015 01:39 PM

I think the covtest general approach is the only one you can use to get at your question. The other options available for G side hypotheses don't necessarily fit.

One thing you could do is look at the information criteria obtained when fitting models with and without the group=treatment option. This would give an idea of how much information is lost by assuming that all groups have equal variances, as opposed to all groups having different variances. Note that this is not a statistical test that will return a probability value.

Steve Denham

All Replies

Solution

04-10-2015
01:39 PM

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

04-10-2015 01:39 PM

I think the covtest general approach is the only one you can use to get at your question. The other options available for G side hypotheses don't necessarily fit.

One thing you could do is look at the information criteria obtained when fitting models with and without the group=treatment option. This would give an idea of how much information is lost by assuming that all groups have equal variances, as opposed to all groups having different variances. Note that this is not a statistical test that will return a probability value.

Steve Denham