BookmarkSubscribeSubscribe to RSS Feed
Highlighted
ghartel

Community Trekker

Joined:

Mar 5, 2017

Generalized regression parameterization of nested effects

I'm using nested effects in the Generalized Regression and find that the parametrization seems strange, maybe wrong?  In the parameter estimates I would expect one row for each level of the nesting factor, with just the name of that level in []'s.  Instead it uses the difference between that level and the reference level.  For the last rows then you get an entries with [reference level - reference level].  That doesn't make sense to me.  To illustrate run the attached code and you'll see rows in the parameter estimates table including these:

 

Term

Estimate

Std Error

Wald ChiSquare

Prob > ChiSquare

Lower 95%

Upper 95%

...       

      

Region[South-Southwest]:State[South Carolina-Wyoming]

 -3.902821

115.44404

0.0011429

0.9730

 -230.169

222.36334

 ... 

Region[Southwest-Southwest]:State[New Mexico-Wyoming]

1547.9147

121.83319

161.42203

<.0001*

1309.1261

1786.7034

Region[Southwest-Southwest]:State[Texas-Wyoming]

1854.3527

115.99596

255.56355

<.0001*

1627.0048

2081.7006

 

the subtraction of the reference category for State is fine by me, that's like how SAS parametrizes (I know that's different than other JMP platforms), but for the Region, it doesn't make sense? How do I interpret that?

thanks

Gunter

dt = Open( "$SAMPLE_DATA/Seasonal Flu.jmp" );
Fit Model(
	Y( :Flu Cases ),
	Effects( :Region, :State[:Region] ),
	Personality( "Generalized Regression" ),
	Generalized Distribution( "Normal" ),
	Run(
		Fit(
			Estimation Method( Standard Least Squares ),
			Validation Method( None ),
			Normal Quantile Plot( 1 ),
			Active Parameter Estimates( 1 )
		)
	),
	SendToReport( Dispatch( {}, "Model Launch", OutlineBox, {Close( 0 )} ) )
);