BoxCoxSSE - SSE and confidence interval for Box-Cox transformations

Syntax

Description

This is a method of mbcmodel.linearmodel.

[sse, ci, lambda] = BoxCoxSSE(Model, lambda) computes the sum of squares error (sse) and confidence interval (ci) for values of the model under different Box-Cox transforms (as given by the parameter lambda). The data used is that which was used to fit the model. sse is a vector the same size as lambda and ci is a scalar. There is no statistical difference between the Box-Cox transforms where sse less than ci.

[sse, ci, lambda] = BoxCoxSSE(Model) If lambda is not specified, then default values for are used and these are returned in third output argument.

BoxCoxSSE(Model, ...) If no output arguments are requested then a plot of SSE versus lambda is displayed. The confidence intervals are also displayed on this plot.

Examples

To try several different values, of the Box-Cox parameter and plot the results:

lambda = -3:0.5:3;
[sse, ci] = BoxCoxSSE( M, lambda);
semilogy( lambda, sse, 'bo-', lambda([1,end]), [ci, ci], 'r--' );
xlabel( 'Box-Cox parameter, \lambda' );
ylabel( 'SSE' );

Note that BoxCoxSSE does not set a Box-Cox transform in the model. To do this use:

M.Properties.BoxCox = 0; 
[S,M] = M.Fit;

See Also

ParameterStatistics

© 1984-2012- The MathWorks, Inc.    -   Site Help   -   Patents   -   Trademarks   -   Privacy Policy   -   Preventing Piracy   -   RSS