This website uses Cookies. Click Accept to agree to our website's cookie use as described in our Privacy Policy. Click Preferences to customize your cookie settings.
JMP is taking Discovery online, April 16 and 18. Register today and join us for interactive sessions featuring popular presentation topics, networking, and discussions with the experts.
Created:
Aug 17, 2014 11:40 AM
| Last Modified: Nov 11, 2017 04:39 PM
model-diagnostics.jmpaddin
November 2017: note from the author:
This add-in doesn't do a good job of handling Fit Group windows. This type of output used to only appear when using stepwise regression but now JMP uses them for standard least squares when multiple responses are present. In this instance the add-in will probably only detect the first response. For those of you who have the intellectual capacity to model multiple responses simultaneously I will provide an update to the add-in - but will probably wait and combine it with support for version 14 of JMP.
Latest Update (May 2017)
In the original release of this add-in, models were discovered based on open windows. Now,. if there are no model windows open, the add-in will scan scripts attached to the current data table. Any scripts that have been saved from the Fit Model platform will be used to generate summary statistics and diagnostic information:
In addition the following enhancements have been implemented :-
Residual plots now have a smoother curve to help identify curvature
It is now possible to plot residuals versus variables not included in the model
The partition platform is used to model residual variation to help indentify latent variables
The add-in can now perform a check to verify that you have the latest version
This add-in aggregates various model diagnostic information into a single tabbed window. The add-in also explores various transformation options that can be applied to the response(s).
The add-in automatically discovers model windows (least square models generated using Fit Model or via Stepwise), and generates the following information:
Models
List of models grouped by response, with summary fit statistics for each model
Terms
List of terms for the selected model
Effects
Half normal probability plot of effects. To decrease clutter, abbreviations are used on the plot with a legend showing the associated factor names
Residuals
Normal probability plot of residuals
Residuals versus response
Residuals by run order
Cook's Distance
Residuals versus each predictor
Residual versus non-model variables
Transformations
Box-Cox transformation test with user-friendly labelling for the lambda parameter
Model comparison statistics - various transformations are performed and comparison statistics generated, included a measure of complexity
Links from the model comparison tabulation allow these new models to be quickly launched and reviewed