1. Version 9 introduces namespaces for variables. This is a great way to control scope of variables, but there is one unfortunate side effect - it is nolonger possible to simply inspect variable values (in v8 you just hover the cursor over the variable name). I'd therefore like to have some sort of diagnostic tool for inspecting variables declared within a namespace (without having to use the show function)
2. Inclusion of additional event callbacks. Version 9 introduces an OnClose callback for a window. Other such events would be useful e.g. if a user resizes a window or resizes a graphbox within a window.
3. Ability to define a menu specific to a window e.g. creating a custom platform and including menu options for that platform available only from the parent window as opposed to generically via the addins options.
1. The FIT MODEL dialog could benefit from some usability enhancements. If you have a model with main effects and you want to add cross terms then the terms (or atleast one of them) have to be selected from the main list of columns. If you have 100s of columns this is extremely cumbersome. Here are 3 potential approaches to simplify use:
a. Allow the 2 variables to be selected from the list of current model terms, then click the cross button
b. Provide a filter for the column list to show only those columns that correspond to current model terms
c. Allow the creation of a 3rd list which would act as candidate columns for the model. That is, I select the columns that I am interested in as potential model terms and then I create the model terms from this candidate list. This is actually quite an attractive option, because then you could extend functionality. For example, you could provide an option for (say) the Partition platform to create a list of candidate columns based on the contents of the decision tree.
2. A dedicated platform for residual analysis - although I am happy to build an add-in for this if it doesn't become part of the product :)
I think your (a) would be a welcome addition. Regarding the other 2, and maybe you already know this, you may be able to get something like you want by excluding columns you don't want in the column panel. As long as they are excluded they will not show up in any dialog for column selection. The initial effort would involve selecting all of the columns (pretty easy) and then deselecting all the columns you want to work with (could be tedious), and finally right clicking and excluding the remaining selected columns. If you don't want to see the excluded columns even in the table, then you can hide them as well.
Regarding residual analysis, are you wanting something that implements some of the methodology in, say, Barnett's book, Outliers in Statistical Data?
I'm not thinking of anything elaborate, and nothing that can't be done currently - just making the diagnostics more accessible and in a single place:
residuals (raw or studentised) : - by row - by predictor variables - by predicted response - histogram and quantile plots - graphical representations of leverage and influence e.g. residuals versus leverage with contours for Cook's distance
It is quite possible that I am missing something, but I wish for more relational capabilities. If JMP could decipher the relationships in a database I would not have to be so explicit about table and column names when I do a big data grab. Particularly if I am setting up a gui for someone else to get to the data of interest.
More accessible (context specific) explanation to help with practical interpretation of statistics and how various parameters are calculated. I know most, if not all, is buried in the help but it often seems not directly at your fingertips. For example, if I make a control chart I want to see clearly how the control limits are calculated without scrollling through a help.
For any statistical test I want to see exactly (case specific) what assumptions are made and the practical significance.
And for Pareto charts I would like to make with absolute % for each detractor (% of total including non-failures, but non failures not plotted). I know this not exactly standard, but it is the way the vast majority of people want to see a Pareto.