nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

Test for homo/heteroscedacity in JMP

kaylesawyer

Hi, I'm trying to check for homo- or heteroscedasticity in a least squares regression model. I can subjectively examine the residuals, but I'd prefer to additionally report a p-value, such as one generated by the White test or the Breusch–Pagan test. I cannot find documentation of how to do this in the JMP help or documentation, or on the web. Should I save the residuals, square them, and then run a new regression with the same predictors as the original, but with the outcome as the squared residuals? Would the Breusch–Pagan p-value be the significance of the omnibus F in that model? This is the closest I've found: http://www.augustana.edu/users/ecmarme/using%20JMP%20to%20check%20model%20for%20heteroskedasticity.ppt

Accepted answers

ron_horne

Hi Kayle,

I think JMP doesn't have it built in but you can do it "manually" by using the R2 from the auxiliary regression (multiplied by the number of parameters) and comparing it with the appropriate chi2 value.

if you need to do it frequently just script your own process.

All comments

ron_horne

hi Kayle,

take a look at this link:

Regression Diagnostics using JMP - Equality of Error Variance - YouTube

perhaps it can give you a starting point.

ron

kaylesawyer

Hi Ron,

Thanks for the link. The video is helpful. Julian Parris suggests using a median split for predicted, and examining whether the residuals in the upper half have different variability than those in the lower half (using Levene, Brown-Forsythe, Bartlett, or another categorical equality of variability test). However, collapsing continuous data into categories is not ideal. Is it possible to calculate the White test or the Breusch-Pagan test in JMP? Would squaring the residuals and correlating them against predicteds yield p-values for one of these tests?

ron_horne

Hi Kayle,

if you need to do it frequently just script your own process.

kaylesawyer

Hi Ron, I'm having trouble specifying the auxiliary regression and obtaining the chi-square p-value. These are the steps I'm doing:

1. Fit Model of interest.

2. Save residuals.

3. Square residuals.

4. Square original predictors.

5. Fit Model, put squared residuals in Y, and add these Model Effects:

A. The original predictors

B. The predictors squared (using the red triangle next to transform)

C. All possible pairwise crosses of the predictors (using the Macro "factorial to degree", with 2 selected below)

6. Use the r-squared to determine the p-value under a chi-square distribution.

Here is an example with JSL:

dt = Open("$Sample_Data/Fitness.jmp");
fm = Fit Model(
  Y( :Runtime ),
  Effects( :Age, :Weight ),
  Personality( Standard Least Squares ),
  Emphasis( Effect Leverage ),
  Run(
  :Runtime << {Lack of Fit( 0 ), Plot Actual by Predicted( 1 ),
  Plot Regression( 0 ), Plot Residual by Predicted( 1 ),
  Plot Effect Leverage( 1 ), Show VIF( 1 )}
  ),
);
fm << residuals;
dt << New Column ("res2",formula(:Residual Runtime^2));
dt << New Column ("age2",formula(:Age^2));
dt << New Column ("weight2",formula(:Weight^2));

However, at this point I'm stuck. After a lot of reading online and textbooks, I think I'll forgo a hypothesis test and just stick with subjective visual inspection to confirm homoscedasticity for all my analyses. Thanks for your help.