Hi,
Can someone share some perspective on the Compare Slopes option available in the Mixed Model personality of Fit Model? It seems the support it limited to an ANOM-like comparison. Are there ways to get information out of Fit Mixed if I was interested in pairwise comparisons of slopes (Tukey or Student's) or in a comparison of most of the slopes vs. one of the slopes that is acting as a control (Dunnett's)?
Thanks,
Michael
Hi Laura,
This information is useful. I was still on JMP14 EA7 when I wrote my original post.
In my situation, I really do need the mixed model as I do have a random effect that needs to be included. So, as I am reading it, I don't think Julian's scripts apply, although I think those scripts are close to what I would like to do.
Thanks again for adding something for this case in the first place. It fills a gap in my toolbox.
Michael
Hi Michael,
Thank you for trying our JMP 14 Early Adapter version. I am not sure which EA version you are using, but our latest EA (version 8) also added this option to Fit Least Squares and REML in case you are interested. We did only add the ANOM comparison test here for now. If you are interested, Julian also created a JMP add-in for comparing slopes in Fit Least Squares that gives you some other comparison options.
https://community.jmp.com/t5/JMP-Add-Ins/Compare-Simple-Slopes-in-One-Way-ANCOVA-w-Interaction-add-in/ta-p/23903
Also see the following previous discussion for even more information:
https://community.jmp.com/t5/Discussions/comparison-of-slopes/m-p/18674#M17024
I hope this is helpful.
Laura
I'll take the first stab at an answer. I'm using a table from the sample data.
Open("$SAMPLE_DATA/Growth Measurements.jmp")
And I'm going to run this model:
Fit Model(
Y( :Growth ),Effects( :Gender ),Personality( "Mixed Model" ),
Subject( :Subject ),Repeated Effects( :Age ),
Repeated Structure( "Toeplitz Unequal Variances" ),
Run( Random Effects Covariance Parameter Estimates( 0 ) )
)
From the report's red triangle menu I clicked on Multiple Comparisons. 
This selection adds this to the bottom of the report:
From the least squares Estimates:
Compare the ranges of the confidence intervals. Note that the upper 95% CI of Gender=F is
less than the lower 95%CI of Gender=M. Good chance there is a significant difference.
From the All Pairwise Differences report:
note that the difference between Gender=F and Gender=M is =3.07...
The same difference between the estimate of F and M in the report above
The null hypothesis, that F and M aren't different is rejected. (H; F-M=0) Zero is not contained
within the bounds of the upper and lower 95%CI so we reject the null. (difference is significant)
I love the Pairwise Comparison Scatterplot. Look at it this way, if you plot a line with a slope of 1 and an intercept of 0, and then plot F vs. M on the graph with the confidence interval perpendicular to the diagonal line, its easy to see that the difference between the two levels is different.
In this last figure, I'm comparing fill volumes across three lines. Its easy to see where my problem is. My cursor is on the red dot in the graph, so the label pops up. The result of the effects of the multiple comparisions are pretty easy to see that B and C are likely different. In the stats table, the interval for that comparison does not contain zero, and the p-value is hilighted in red.
so, I'm not sure that I really answered your question, but this is how I use it.
Cheers,
-B
