topic Re: Multiple repeated measures glimmix in Statistical Procedures

Multiple repeated measures glimmix

tlyons — Sat, 24 Dec 2016 15:25:07 GMT

I'm trying to run an analysis in glimmix of some ecological data thats throwing me for a loop, but I'm probabaly overthinking it. The data were collected over 4 years, at 4 blocks where each block contained 1 of 4 different vegetative types. Because the same plots were visited each year, I know I can account for potential covaraince at this level using a random effect of study plot or as a repeated measure using one of the following

random int/ subject=plot;

random year/ subject=plot;

However, these plots were also visited twice within a given year. How do I account for this second level of repeated-ness? Is it just as simple as :

random visit/ subject=plot*yr?

and can this kind of statement be combined with one of the above?

Re: Multiple repeated measures glimmix

PGStats — Sat, 24 Dec 2016 23:00:55 GMT

Shouldn't visits be called season instead and considered as a fixed effect?

Re: Multiple repeated measures glimmix

tlyons — Sat, 24 Dec 2016 23:07:06 GMT

I don't think so, they're repeated measurements of the same sample unit. But I'm open to it if you can explain why.

Re: Multiple repeated measures glimmix

PGStats — Sun, 25 Dec 2016 21:13:16 GMT

Well, I am missing some context here.

But ecological systems typically got through a cycle during the year. Observations from the Spring may be systematically different from observations from the Fall. Random parameter distributions available in GLIMMIX are unimodal, they won't fit too well. If your visits occurred at different moments of a systematic cycle, then it would be preferable to model that cycle instead of considering them as independent random realizations of a single distribution.

Re: Multiple repeated measures glimmix

tlyons — Mon, 26 Dec 2016 17:44:26 GMT

I think in this case I can assume that they're not systematically different. The data are sort of a maximum number of unique small mammals trapped during each survey. In this case, we know individuals trapped early in the year show up in surveys later in the year, and that fall abundances ought to be larger than spring abundances because there is the potential for multiple litters. For this reason, I know there is covariance among the two samples within a year and and I am just searching for a way to model that out, in addition to the covariance associated with returning to the exact same plots year after year.

Re: Multiple repeated measures glimmix

SteveDenham — Tue, 27 Dec 2016 19:53:33 GMT

Here's an idea. Throw out year. If you have two visits to a plot per year, then fit visit 1 through 8 as a single repeated measure. You can then look at 'year' effects by combining adjacent visits in an LSMESTIMATE statement.

And if the visits are not equally spaced in time, consider fitting a spline to both the fixed effect and as a random effect (see example 45.15 Comparing Multiple B Splines and Example 45.6 Radial Smoothing of Repeated Measures Data.

Steve Denham