When is it possible to get the "Multiple Comparisons" option within "Estimates" for JMP15 Pro?I see the "Multiple Comparisons" option is available mostly when you have categorical variables in your model. It also shows up when you have only continuous variables. It becomes restrictive when you have categorical and continuous variables together in the model because you can only test the overall means of the levels of the categorical variable but not the means at given levels of the continuous variables. Besides, the "Multiple Comparisons" option does not seem to be available if you have mixture variables in the model.
Re: When is it possible to get the "Multiple Comparisons" option within "Estimates" for JMP15 Pro?I believe the issue of the multiple comparisons is an impossibility due to the constraint of components adding to 1. I am not completely certain of this, but believe it to be true. Remember that the parameters of a Scheffe model are really different estimates. The typical statistical theory of linear regression does not really apply. The coefficients are constrained.
As for the profiler, remember that the profiler has a confidence interval on it. Prediction A might be 90 +/- 5. Prediction B might be 87 +/- 5. There is no real difference because 87 is within the range of prediction A. Might not be as rigorous as the multiple comparison, but still very effective.
Re: When is it possible to get the "Multiple Comparisons" option within "Estimates" for JMP15 Pro?Thanks @Dan_Obermiller ,
It is just that the continuous variable won't show up in the multiple comparisons window unless you select "User-Defined Estimates", like in the example you provided.
You mentioned the problem of the mixture model specification. What I wonder is whether there is an impossibility or it is just plain complex but doable. If the latter case, should it be a request for the wish-list?
It seems to me that the fact that e.g. "the main effects include the intercept" does not prevent the definition of a contrast matrix, it just make its interpretation difficult (for the lack of a better word) if you want causal inferences to be drawn. However, from a 100% predictive perspective - there can't be another one with mixture models anyway - the comparison can still be made - e.g. like when you want to optimize a formulation, you play with your model profiler, you have multiple combinations of the components levels for which the predicted response is close to max desirability, and you want to test if the predicted response of one of these combinations is equal to that of some other.
Re: When is it possible to get the "Multiple Comparisons" option within "Estimates" for JMP15 Pro?Fabio,
First, I assume that you are referring to the Multiple Comparisons found within Fit Model > Estimates > Multiple Comparisons.
If you have both categorical and continuous variables in the model, you can still get multiple comparisons involving the continuous factors. You need to choose the "User-Defined Estimates". In my simple example, there is a categorical factor, League with levels American and National. There is a continuous factor, Years in Majors. You must specify a level of the categorical factor over which to test because there is not interpolation allowed for a categorical factor. The testing of the continuous factors only makes sense within a specific level of the categorical factor.
As for mixture variables, the multiple comparisons would be especially problematic because a Scheffe mixture model does not have the typical parameter estimates. For example, main effects include the overall intercept, so multiple comparisons need to remove the overall intercept from some or possibly all of the levels. Bottom line is that the overall testing on mixture components is quite different than in the unconstrained model-building space.
I hope this information helps.
