topic Re: Calculate Moving R2 in Discussions

Calculate Moving R2

shasheminassab — Mon, 07 Oct 2019 16:32:18 GMT

How can I calculate R2 between two variables (i.e. columns) in a moving window (e.g., 10 pairs of data at a time)?

Sina

Re: Calculate Moving R2

Mark_Bailey — Mon, 07 Oct 2019 18:06:00 GMT

See this discussion about performing linear regression in a column formula. This approach could be extended to obtain R square or other results.

Re: Calculate Moving R2

shasheminassab — Mon, 07 Oct 2019 18:46:00 GMT

Thanks @Mark_Bailey. What is the function to calculate R2 in JSL? (sorry I am new to JSL scripting and couldn't find it online).

Re: Calculate Moving R2

ron_horne — Mon, 07 Oct 2019 19:43:28 GMT

Hi @shasheminassab

take a look at this:
http://www.pega-analytics.co.uk/blog/linear-regression-matrix-form/

Re: Calculate Moving R2

Mark_Bailey — Mon, 07 Oct 2019 20:41:47 GMT

I understand.

Here is an example based on the Big Class data table in the Sample Data folder. It fits a moving linear regression with a span of 5. It returns the R square value in each row. You can run this script to see how it works.

Names Default To Here( 1 );

dt = Open( "$SAMPLE_DATA/Big Class.jmp" );

// example creates a new column with a moving R square with a span of 5
// for regression of :weight versus :height
dt << New Column( "Moving R Square",
	Numeric,
	Continuous,
	Formula(
		If( Row() > 2 & Row() < (N Row()-1),
			x = J( 5, 1, 1 ) || :height[Index( Row() - 2, Row() + 2 )]`;
			y = :weight[Index( Row() - 2, Row() + 2 )]`;
			sst = Sum( (y - Mean( y )) ^ 2 );
			b = (Inv( x` * x ) * x` * y);
			yhat = x * b;
			sse = Sum( (y - yhat) ^ 2 );
			(sst-sse) / sst;
		,
			Empty()
		)
	)
);

You can select the part of this script that is within the parentheses after the Formula argument and past that expression into a new column formula in your data table, then change the column names to match your X and Y data columns.

Re: Calculate Moving R2

shasheminassab — Sun, 13 Oct 2019 21:27:12 GMT

Thank you @Mark_Bailey and @ron_horne . This was very helpful!!