23391_en-US

liliconcom

JMP 14 Academic Feature Requests

Requests:

• SEM
  • Market research, behavioral sciences
  • Gaining ground (SPSS!)
• Interactive model builder, that builds different types of models, including SEM, PLS,..., other fitting options
  • Along the lines of the Partial Correlation Diagram Add-in (Partial Correlation Diagram)
• The new Fit Curve - option to fit all and find the best?
• Simple way of specifying repeated measures (seeFull Factorial Repeated Measures ANOVA Add-In)
• Hierarchical linear mixed models (HLMM) - we do HLM, and we do Mixed Models, but not HLMM
• Time Series
  • Validation
  • Winters Method multiplicative
  • Simple Exponential Smoothing - save out the formula in the expected form (EWMA with the smoothing parameter)
• Fit Model
  • Show effect size (eta^squared, Cohen's d) in the effect test output (maybe a right-click > Columns option to display)
• Export to LaTeX file
  • Several tools exist for R
• Existing requests, keyword TEACHINGBA2015 (copy and past this link in your browser to view - http://darts.na.sas.com/darts/dashboard.php?mode=view&vREPORT_ID=20837)

Inconsistencies/improvements:

• RASE and RMSE
  • Having two terms is confusing.  RASE is only reported in Least Squares (for validation) and Model Comparison, otherwise its reported as RMSE.  RASE is just root SSE/n - the biased estimator of RMSE - in most data mining literature it is simply referred to as RMSE.  Whatever we do, we need to be consistent.
• Fit and Summary statistics for the different predictive modeling platforms are different:
  • For example, for continuous responses (when there is validation):
    • Fit model - RSquare and RASE
    • Gen Reg - Generalized RSquare, -LogLikelihood, BIC, AIC (and RMSE is reported as Scale in the parameter estimates table)
    • Partition - RSquare, RMSE, AICc
    • Neural - RSquare, RMSE, MAD, -LogLikelihood, SSE
    • Model Comparison - RASE, RSquare, AAE.
  • Add MAPE in Model Comparison for continuous response (mean absolute percent error) - this is available in Time Series, but that's the only platform.
• Residual plots in Fit Y by X and Fit Model
  • People like the variety of the residual plots in Fit Y by X, but it would be nice if they were prettier (like in the new Mixed platform)
• Fit Model, with SLR, it would be nice if the regression plot has more functionality - confidence bands fit and prediction with the ability to shade
• The option to save prediction or probability formulas - what the option is called, and where its found, is based on the platform.  It would be nice if we had more consistency here
• Naming of saved prediction and probability formulas
  • Grouping by saved model, with the ability to name the group
  • Take advantage of virtual join for scoring with new data?  (Via Model Depot)

Re: JMP 14 Academic Feature Requests<meta http-equiv="Content-Type" content="text/html; charset=UTF-8" />

Next steps:

1. Mia, Volker, Julian and Curt go through these items to list names of professors/schools for different requests
2. Develop mock up/prototype of interactive modeler - Julian lead

Re: JMP 14 Academic Feature Requests

1. SEM/Model Builder

2. Repeated measures, + effect sizes

3.  Time Series

Re: JMP 14 Academic Feature Requests

SEM / Interactive Model Building

Effect Size Indices

Easier Repeated Measures

     (Example: SPSS interactive)

    

Re: JMP 14 Academic Feature Requests

My top 3:

SEM

Interactive model builder

More consistent Fit and Summary statistics for the different predictive modeling platforms

Re: JMP 14 Academic Feature Requests

My top 3 are:

SEM

Making the summary statistics for predictive models more consistent

Adding functionality to Time Series