Subscribe Bookmark RSS Feed

Does JMP Pro FIT MIXED personality allow for 3-level hierarchical models ?

mjyork0

Community Trekker

Joined:

Jan 15, 2016

Does JMP Pro FIT MIXED personality allow for 3-level hierarchical models (e.g., subject nested within unit nested within division)?  I would like to run the following random intercepts model:

PROC MIXED data=survey covtest noclprint method=REML;

class DEPT DIV ;

model SATISF = SUPV COWORKERS INST JOBFIT  ;

random intercept / sub = DIV type = UN ;

random intercept / sub = DEPT(DIV) type = UN;

Would the corresponding JMP JSL script look something like this?

Fit Model(

               Y( :SATISF ),

               Effects( :SUPV, :COWORKERS, :INST, :JOBFIT ),

               Random Effects(

                              Intercept[:Business Unit] & Random Coefficients( 2 ),

                              Intercept[:Business Unit, :DEPT] & Random Coefficients( 2 )

               ),

               Personality( "Mixed Model" ),

               Run( Repeated Effects Covariance Parameter Estimates( 0 ) )

)

1 ACCEPTED SOLUTION

Accepted Solutions
Solution

I am curious about the covariance results you get from running the script as shown above.

Random effects in a three-level model are typically be modeled like this

    Random Effects(

      Intercept[:DIV] & Random Coefficients( 1 ),

      Intercept[:DIV, DEPT] & Random Coefficients( 2 ),

      ),

Since you don't estimate random slopes, an alternative specification would be just simply add DIV, and DEPT[DIV] as two random effects like the following (assuming DEPT nested within DIV)  :

    Random Effects( :DIV, :DEPT[:DIV] ),

7 REPLIES
mjyork0

Community Trekker

Joined:

Jan 15, 2016

Correction to my original post: 

Would the JMP JSL Script look something like this?

Fit Model(

  Y( :SATISF ),

  Effects( :SUPV, :COWORKERS, :INST, :JOBFIT ),

  Random Effects(

  Intercept[:DIV] & Random Coefficients( 1 ),

  Intercept[:DIV, DEPT] & Random Coefficients( 1 ),

  ),

  Personality( "Mixed Model" ),

  Run( Repeated Effects Covariance Parameter Estimates( 0 ) )

);

txnelson

Super User

Joined:

Jun 22, 2012

I would suggest that you need to let JMP create the script for you.  Go into:    Analyze==>Fit Model.  Generate your "Mixed Model". Yes JMP will handle 3 level hierarchical nesting.  Once you have defined the model, then run the analysis.  Under the red triangle at the top of the output, Go to:   Script==>Write Script to Script Window.     It will then generate the script required to create the model you defined.

Jim
mjyork0

Community Trekker

Joined:

Jan 15, 2016

Hi Jim, Thanks for the response.  However, that is what I did to get the script.  I just want to make sure I specified the correct model.

I went to Analyze / Fit Model

- selected Personality = Mixed Model

-- specified the Fixed and Random Effects

-- then saved the Script to the Script Window, copied and pasted

Solution

I am curious about the covariance results you get from running the script as shown above.

Random effects in a three-level model are typically be modeled like this

    Random Effects(

      Intercept[:DIV] & Random Coefficients( 1 ),

      Intercept[:DIV, DEPT] & Random Coefficients( 2 ),

      ),

Since you don't estimate random slopes, an alternative specification would be just simply add DIV, and DEPT[DIV] as two random effects like the following (assuming DEPT nested within DIV)  :

    Random Effects( :DIV, :DEPT[:DIV] ),

mjyork0

Community Trekker

Joined:

Jan 15, 2016

Thank you!  This is very helpful.  What do the numbers after "Random Coefficients" mean (e.g., 1 or 2)?  I can't find any documentation for the JSL script for Mixed Models.

If I wanted to add random slopes at the DEPT level, would the script look like this?

Y( :SATISF ),

Effects( :Factor1, :Factor2, :Factor3, :Factor4 ),

Intercept[:DIV] & Random Coefficients( 1 ),

Intercept[:DIV, :DEPT] & Random Coefficients( 2 ),

:Factor1[:DIV, :DEPT] & Random Coefficients( 3 )

),

Personality( "Mixed Model" ),

Run( Repeated Effects Covariance Parameter Estimates( 0 ) )

);

jiancao

Staff

Joined:

Jul 7, 2014

What do the numbers after "Random Coefficients" mean (e.g., 1 or 2)?

They indicate hierarchical levels, e.g., DIV, DEPT within DIV. They are assigned when using the Nest Random Coefficient tab to add random effects.


If I wanted to add random slopes at the DEPT level, would the script look like this?


Your random effects would look like this


Random Effects(

Intercept[:DIV] & Random Coefficients( 1 ),

Intercept[:DIV, :DEPT] & Random Coefficients( 2 ),

:Factor1[:DIV, :DEPT] & Random Coefficients( 2 )

),

mjyork0

Community Trekker

Joined:

Jan 15, 2016

Thank you!