Repeated Measures -- Can Covariates be used?I have a client who has a data set dealing with horse nutrition. The idea is to give the animals a high-glucose feeding and monitor insulin levels over several hours. They then compare 2 groups of animals: one group is the control, and the other has received a special diet over the past month. So they are really looking at the effect of the diet on horse insulin. They took a baseline on all animals at the start of the study. All had been on the same diet up to the start of the study, and they gave them the high glucose feeding and monitored insulin. Of course each horse had a different baseline. Then the put half of the animals on the new diet for a month, and repeated the test. They can see a small difference doing a standard Repeated Measures on the end-of-month data, but they would like to account for individual animal baselines at the start of the month, hoping that this would clean up the signal. Any thoughts on how to do that? Thanks!Re: Repeated Measures -- Can Covariates be used?Jerry,
It's pretty straightforward, just put the covariate in as an effect. I gave a WebEx/Discovery talk on using JMP Pro for repeated measures and the main example did just this. Here's the link to the WebEx. The Discovery talk had the same content. Under simple covariance assumptions (uncorrelated time periods or the same correlation for all time periods, e.g., compound symmetry), this can be done in regular JMP, too.
Don
Re: Repeated Measures -- Can Covariates be used?Thanks Don. Unfortunately my client has standard JMP, not Pro. You mentioned that this could be done in standard JMP... can you point me to a resource describing how to do that?
Thanks,
Jerry
Re: Repeated Measures -- Can Covariates be used?Jerry,
If they set the subject variable (the entity on which the repeated measurements are take) to random, they will get a repeated measures model assuming compound symmetry (CS) as the covariance matrix. This indicates that the time dependent correlation is the same, regardless of the difference in the time periods (e.g., the correlation between the first and second time periods is the same as the first and last time periods). The assumption of CS (or no correlation) is very limiting. Fit Mixed Models in JMP Pro gives them about 10 additional covariance structures to work with (SAS has about 30 or 40 more in PROC MIXED). An alternative is to use the Manova personality to fit a repeated measures models. I typically avoid doing this because I find it difficult to interpret and explain (and it can have considerably lower power than a mixed models approach).
Don
