How to fit a polynomial degree 2 models with by-groups then capture the parameter estimates

[#1]

I would like to create polynomial degree 2 by each (SERIAL, P1, P2, P3) and get the coefficient and constant then put in each column by using JSL.

The example data was attached in this post, but actually I will do with the data about 1 million rows.

[#2]

It would be possible to develop the script below to get what you need:

NamesDefaultToHere(1);
ClearLog();
dt = DataTable("Example data.jmp");
col = Column(dt, "Predicted y By SERIAL By P1 By P2 By P3");
// Get the formula generated by the platform
f = col << getFormula;
// Start to parse this formula expression
for(g = 1, g <= Narg(f) - 1, g = g + 2,
	f1 = Arg(f, g);
	f2 = Arg(f, g+1);
	Print(f1, f2, "\!n");
);

But I suspect it might be simpler to send the 'MakeCombinedDataTable' message to the appropriate table in the report window if that's possible. Look in 'Help > Scripting Guide'.

[#3]

Also, if you are using the Fit Y by X platform / Bivariate platform to generate the polynomial fits, make sure to do a "Fit Special" and use uncentered polynomials.  Otherwise the fitted quadratic will be of the form

f(x) = a + b*x + c*(x-xbar)

where xbar is the mean of the x values used in the regression.

Doing the fitting with a By variable and then right clicking the parameter estimates table to make a combined data table is a very efficient way to do this.

[#4]

Thanks MathStatChem, let me try for your suggestion.

[#5]

Thanks ian_jmp, let me try.

[#6 · Accepted solution]

Try this script:

_biv=Bivariate(
	Y( :y ),
	X( :x ),
	Fit Special(
		Degree( 2 ),
		Centered Polynomial( 0 ),
		{Line Color( {213, 72, 87} )}
	),
	By( :SERIAL, :P1, :P2, :P3 )
);


_rbiv=Report(_biv[1]);

_dtParameterEst = _rbiv[Table Box(3)]<< Make Combined Data Table;

_dtParameterEst << Split(
	Split By( :Term ),
	Split( :Estimate ),
	Group( :P3, :P2, :P1, :SERIAL ),
	Remaining Columns( Drop All ),
	Sort by Column Property
)

[#7]

Hi MathStatChem,

Your solution it work.

[#8]

Hi ian_jmp, My objective is get the coefficient and constant of "y = ax^2+bx+c" but in fit polynomial is "y = a(x-d)^2+bx+c". However I got the solution from user# MathStatChem by fit special.