14.03.2014 Views

Modeling and Multivariate Methods - SAS

Modeling and Multivariate Methods - SAS

Modeling and Multivariate Methods - SAS

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

680 Statistical Details Appendix A<br />

<strong>Multivariate</strong> Details<br />

Figure A.10 Cases of Significant, Borderline, <strong>and</strong> Nonsignificant Confidence Curves.<br />

Significant Borderline Not Significant<br />

confidence curve<br />

crosses horizontal<br />

line<br />

confidence curve<br />

asymptotic to<br />

horizontal line<br />

confidence curve<br />

does not cross<br />

horizontal line<br />

Leverage plots make use of the same device by calculating a confidence function with respect to one variable<br />

while holding all other variables constant at their sample mean value.<br />

For general hypothesis tests, JMP can display a curve with the same properties: the confidence function<br />

shows at the mean value in the middle, with adjusted curvature so that it crosses the horizon if the F-test is<br />

significant at some α-level like 0.05.<br />

Consider the functions<br />

Upper(z) = z s 2 2 F α<br />

+ t α ⁄ 2<br />

h + ------ 2 z<br />

F<br />

<strong>and</strong><br />

Lower (z) = z s 2 2 F α<br />

– t α ⁄ 2<br />

h + ------ z 2<br />

F<br />

where F is the F-statistic for the hypothesis, F α is the reference value for significance α, <strong>and</strong><br />

h = xX'X ( ) – 1 x' , where x is the regressors at a suitable middle value, such as the mean.<br />

These functions serve the same properties listed above. If the F-statistic is greater than the reference value,<br />

the confidence functions cross the x-axis. If the F-statistic is less than the reference value, the confidence<br />

functions do not cross. If the F-statistic is equal to the reference value, the confidence functions have the<br />

x-axis as an asymptote. And the range between the confidence functions at the middle value is a valid<br />

confidence region of the predicted value at the point.<br />

<strong>Multivariate</strong> Details<br />

The following sections show computations used for multivariate tests <strong>and</strong> related, exact <strong>and</strong> approximate<br />

F-statistics, canonical details, <strong>and</strong> discriminant functions. In the following sections, E is the residual cross<br />

E<br />

product matrix <strong>and</strong> ----------- estimates the residual covariance matrix. Diagonal elements of E are the sum of<br />

n – 1<br />

squares for each variable. In discriminant analysis literature, this is often called W, where W st<strong>and</strong>s for<br />

within.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!