12.12.2012 Views

SAS/STAT 922 User's Guide: The MIXED Procedure (Book Excerpt)

SAS/STAT 922 User's Guide: The MIXED Procedure (Book Excerpt)

SAS/STAT 922 User's Guide: The MIXED Procedure (Book Excerpt)

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.

E<br />

E1<br />

E2<br />

E3<br />

MODEL Statement ✦ 4557<br />

c matrices that are the size of the mixed model equations, where c is the number of covariance<br />

parameters. In the notation of Table 56.25, this is approximately 8q.p C g/.p C g/=2 bytes.<br />

Extra computing time is also required to process these matrices. <strong>The</strong> Satterthwaite method<br />

implemented here is intended to produce an accurate F approximation; however, the results<br />

can differ from those produced by PROC GLM. Also, the small sample properties of this<br />

approximation have not been extensively investigated for the various models available with<br />

PROC <strong>MIXED</strong>.<br />

<strong>The</strong> DDFM=KENWARDROGER option performs the degrees of freedom calculations detailed<br />

by Kenward and Roger (1997). This approximation involves inflating the estimated variancecovariance<br />

matrix of the fixed and random effects by the method proposed by Prasad and Rao<br />

(1990) and Harville and Jeske (1992); see also Kackar and Harville (1984). Satterthwaite-type<br />

degrees of freedom are then computed based on this adjustment. By default, the observed<br />

information matrix of the covariance parameter estimates is used in the calculations. For<br />

covariance structures that have nonzero second derivatives with respect to the covariance<br />

parameters, the Kenward-Roger covariance matrix adjustment includes a second-order term.<br />

This term can result in standard error shrinkage. Also, the resulting adjusted covariance matrix<br />

can then be indefinite and is not invariant under reparameterization. <strong>The</strong> FIRSTORDER<br />

suboption of the DDFM=KENWARDROGER option eliminates the second derivatives from<br />

the calculation of the covariance matrix adjustment. For the case of scalar estimable functions,<br />

the resulting estimator is referred to as the Prasad-Rao estimator em @ in Harville and Jeske<br />

(1992). <strong>The</strong> following are examples of covariance structures that generally lead to nonzero<br />

second derivatives: TYPE=ANTE(1), TYPE=AR(1), TYPE=ARH(1), TYPE=ARMA(1,1),<br />

TYPE=CSH, TYPE=FA, TYPE=FA0(q), TYPE=TOEPH, TYPE=UNR, and all TYPE=SP()<br />

structures.<br />

When the asymptotic variance matrix of the covariance parameters is found to be singular,<br />

a generalized inverse is used. Covariance parameters with zero variance then do<br />

not contribute to the degrees-of-freedom adjustment for DDFM=SATTERTHWAITE and<br />

DDFM=KENWARDROGER, and a message is written to the log.<br />

This method changes output in the following tables (listed in Table 56.22): Contrast, CorrB,<br />

CovB, Diffs, Estimates, InvCovB, LSMeans, Slices, SolutionF, SolutionR, Tests1–Tests3. <strong>The</strong><br />

OUTP= and OUTPM= data sets are also affected.<br />

requests that Type 1, Type 2, and Type 3 L matrix coefficients be displayed for all specified<br />

effects. For ODS purposes, the name of the table is “Coef.”<br />

requests that Type 1 L matrix coefficients be displayed for all specified effects. For ODS<br />

purposes, the name of the table is “Coef.”<br />

requests that Type 2 L matrix coefficients be displayed for all specified effects. For ODS<br />

purposes, the name of the table is “Coef.”<br />

requests that Type 3 L matrix coefficients be displayed for all specified effects. For ODS<br />

purposes, the name of the table is “Coef.”

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

Saved successfully!

Ooh no, something went wrong!