Improving Dunnett's Comparison in Fit Y by X

[jszarka]

I was running a training course on one-way ANOVA and was surprised by how underwhelming the Dunnett's output is in Fit Y by X. Below I am using the Typing Data data set with Regal as the control group. When turning all of the options on, there isn't a single interpretable piece of output shown below:

The Fit Model platform can be a bit too much for a basic training, but the output for Dunnett's there is much better.

Using the Multiple Comparisons tab, the output can look like the plot below. For Fit Y by X, the 'Comparisons with Control' output that includes the 'Differences from Control' should be the default output. It gives the 2 relevant comparisons with an easy interpretation of their estimated difference, 95% confidence interval, and adjusted p-value to evaluate their significance.

The Effect Details show similar Dunnett comparisons, but this suffers from errors in the StdErrDif and Cls equaling zero in some areas. These issues were documented 2 years ago in JMP 18 for TS-00178884, but have not been resolved.

My proposal is to include this as the standard Dunnett's output in future versions. While this could be populated as a Wish List item, my concern would be that this isn't a modern or cool fancy thing to put up, but rather a simpler scenario that can just be fixed. It would also mirror the Tukey comparison (which is very interpretable - especially with the red/blue significant/insignificant charts).

The title under the red triangle could be "Dunnett's Ordered Differences Report".