Chapter Contents
Chapter Contents
Previous
Previous
Next
Next
The MIXED Procedure

Changes in Output

The SAS System now features a new Output Delivery System (ODS), replacing the old one used by PROC MIXED in Version 6. The primary changes involve the replacing of the MAKE statement and the _print_ and _disk_ global variables with the new ODS statement. The following table lists some typical conversions you need to make.

Table 41.6: ODS Conversions for PROC MIXED
Version 6 Syntax Versions 7 and 8 Syntax
make 'covparms' out=cp;ods output covparms=cp;
make 'covparms' out=cp noprint;ods listing exclude covparms; ods output covparms=cp;
%global _print_; %let _print_=off;ods listing close;
%global _print_; %let _print_=on;ods listing;

Each table created by PROC MIXED has a name associated with it, and you must use this name to reference the table when using ODS statements. These names are listed in Table 41.7.

Table 41.7: ODS Tables Produced in PROC MIXED
Table Name Description Required Statement / Option
AccRatesacceptance rates for posterior samplingPRIOR
AsyCorrasymptotic correlation matrix of covariance parametersPROC MIXED ASYCORR
AsyCovasymptotic covariance matrix of covariance parametersPROC MIXED ASYCOV
Basebase densities used for posterior samplingPRIOR
Boundcomputed bound for posterior rejection samplingPRIOR
CholGCholesky root of the estimated G matrixRANDOM / GC
CholRCholesky root of blocks of the estimated R matrixREPEATED / RC
CholVCholesky root of blocks of the estimated V matrixRANDOM / VC
ClassLevelslevel information from the CLASS statementdefault output
CoefL matrix coefficientsE option on MODEL, CONTRAST, ESTIMATE, or LSMEANS
Contrastsresults from the CONTRAST statementsCONTRAST
ConvergenceStatusconvergence statusdefault
CorrBapproximate correlation matrix of fixed-effects parameter estimatesMODEL / CORRB
CovBapproximate covariance matrix of fixed-effects parameter estimatesMODEL / COVB
CovParmsestimated covariance parametersdefault output
Diffsdifferences of LS-meansLSMEANS / DIFF (or PDIFF)
Dimensionsdimensions of the modeldefault output
Estimatesresults from ESTIMATE statementsESTIMATE
FitStatisticsfit statisticsdefault
Gestimated G matrixRANDOM / G
GCorrcorrelation matrix from the estimated G matrixRANDOM / GCORR
HLM1Type 1 Hotelling-Lawley-McKeon tests of fixed effectsMODEL / HTYPE=1 and REPEATED / HLM TYPE=UN
HLM2Type 2 Hotelling-Lawley-McKeon tests of fixed effectsMODEL / HTYPE=2 and REPEATED / HLM TYPE=UN
HLM3Type 3 Hotelling-Lawley-McKeon tests of fixed effectsREPEATED / HLM TYPE=UN
HLPS1Type 1 Hotelling-Lawley-Pillai- Samson tests of fixed effectsMODEL / HTYPE=1 and REPEATED / HLPS TYPE=UN
HLPS2Type 2 Hotelling-Lawley-Pillai- Samson tests of fixed effectsMODEL / HTYPE=1 and REPEATED / HLPS TYPE=UN
HLPS3Type 3 Hotelling-Lawley-Pillai- Samson tests of fixed effectsREPEATED / HLPS TYPE=UN
InfoCritinformation criteriaPROC MIXED IC
InvCholGinverse Cholesky root of the estimated G matrixRANDOM / GCI
InvCholRinverse Cholesky root of blocks of the estimated R matrixREPEATED / RCI
InvCholVinverse Cholesky root of blocks of the estimated V matrixRANDOM / VCI
InvCovBinverse of approximate covariance matrix of fixed-effects parameter estimatesMODEL / COVBI
InvGinverse of the estimated G matrixRANDOM / GI
InvRinverse of blocks of the estimated R matrixREPEATED / RI
InvVinverse of blocks of the estimated V matrixRANDOM / VI
IterHistoryiteration historydefault output
LRTlikelihood ratio testdefault output
LSMeansLS-meansLSMEANS
MMEqmixed model equationsPROC MIXED MMEQ
MMEqSolmixed model equations solutionPROC MIXED MMEQSOL
ModelInfomodel informationdefault output
ParmSearchparameter search valuesPARMS
Posteriorposterior sampling informationPRIOR
Rblocks of the estimated R matrixREPEATED / R
RCorrcorrelation matrix from a blocks of the estimated R matrixREPEATED / RCORR
Searchposterior density search tablePRIOR / PSEARCH
Slicestests of LS-means slicesLSMEANS / SLICE=
SolutionFfixed effects solution vectorMODEL / S
SolutionRrandom effects solution vectorRANDOM / S
Tests1Type 1 tests of fixed effectsMODEL / HTYPE=1
Tests2Type 2 tests of fixed effectsMODEL / HTYPE=2
Tests3Type 3 tests of fixed effectsdefault output
Type1Type 1 analysis of variancePROC MIXED METHOD=TYPE1
Type2Type 2 analysis of variancePROC MIXED METHOD=TYPE2
Type3Type 3 analysis of variancePROC MIXED METHOD=TYPE3
Transtransformation of covariance parametersPRIOR / PTRANS
Vblocks of the estimated V matrixRANDOM / V
VCorrcorrelation matrix from blocks of the estimated V matrixRANDOM / VCORR

In Table 41.7, "Coefficients" refers to multiple tables produced by the E, E1, E2, or E3 options in the MODEL statement and the E option in the CONTRAST, ESTIMATE, and LSMEANS statements. You can create one large data set of these tables with a statement similar to

   ods output Coefficients=c;

Be aware that the number of variables in this data set is determined by the first table created. To create separate data sets, use

   ods output Coefficients(match_all)=c;

Here the resulting data sets are named C1, C2, C3, etc. The same principles apply to data sets created from the "R," "CholR," "InvCholR," "RCorr," "InvR," "V," "CholV," "InvCholV," "VCorr," and "InvV" tables.

In Table 41.7, the following changes have occured from Version 6. The "Predicted," "PredMeans," and "Sample" tables from Version 6 no longer exist and have been replaced by output data sets; see descriptions of the MODEL statement options OUTPRED= and OUTPREDM= and the PRIOR statement option OUT= for more details. The "ML" and "REML" tables from Version 6 have been replaced by the "IterHistory" table. The "Tests," "HLM," and "HLPS" tables from Version 6 have been renamed "Tests3," "HLM3," and "HLPS3."

Table 41.8 lists the variable names associated with the data sets created when you use the ODS OUTPUT option in conjunction with the preceding tables. In Table 41.8, n is used to denote a generic number that is dependent upon the particular data set and model you select, and it can assume a different value each time it is used (even within the same table). The phrase model specific appears in rows of the affected tables to indicate that columns in these tables depend upon the variables you specify in the model.

Caution: There exists a danger of name collisions with the variables in the model specific tables in Table 41.8 and variables in your input data set. You should avoid using input variables with the same names as the variables in these tables.

Table 41.8: Variable Names for the ODS Tables Produced in PROC MIXED
Table Name Variables
AsyCorrRow, CovParm, CovP1 -CovPn
AsyCovRow, CovParm, CovP1 -CovPn
BaseDenType, Parm1 -Parmn
BoundTechnique, Converge, Iterations, Evaluations, LogBound, CovP1 -CovPn, TCovP1 -TCovPn
CholGmodel specific, Effect, Subject, Sub1 -Subn, Group, Group1 -Groupn, Row, Col1 -Coln
CholRIndex, Row, Col1 -Coln
CholVIndex, Row, Col1 -Coln
ClassLevelsClass, Levels, Values
Coefficientsmodel specific, LMatrix, Effect, Subject, Sub1 -Subn, Group, Group1 -Groupn, Row1 -Rown
ContrastsLabel, NumDF, DenDF, ChiSquare, FValue, ProbChiSq, ProbF
CorrBmodel specific, Effect, Row, Col1 -Coln
CovBmodel specific, Effect, Row, Col1 -Coln
CovParmsCovParm, Subject, Group, Estimate, StandardError, ZValue, ProbZ, Alpha, Lower, Upper
Diffsmodel specific, Effect, Margins, ByLevel, AT variables, Diff, StandardError, DF, tValue, Tails, Probt, Adjustment, Adjp, Alpha, Lower, Upper, AdjLow, AdjUpp
DimensionsDescr, Value
EstimatesLabel, Estimate, StandardError, DF, tValue, Tails, Probt, Alpha, Lower, Upper
FitStatisticsDescr, Value
Gmodel specific, Effect, Subject, Sub1 -Subn, Group, Group1 -Groupn, Row, Col1 -Coln
GCorrmodel specific, Effect, Subject, Sub1 -Subn, Group, Group1 -Groupn, Row, Col1 -Coln
HLM1Effect, NumDF, DenDF, FValue, ProbF
HLM2Effect, NumDF, DenDF, FValue, ProbF
HLM3Effect, NumDF, DenDF, FValue, ProbF
HLPS1Effect, NumDF, DenDF, FValue, ProbF
HLPS2Effect, NumDF, DenDF, FValue, ProbF
HLPS3Effect, NumDF, DenDF, FValue, ProbF
InfoCritBetter, CovParms, MeanParms, Likelihood, AIC, HQIC, BIC, CAIC
InvCholGmodel specific, Effect, Subject, Sub1 -Subn, Group, Group1 -Groupn, Row, Col1 -Coln
InvCholRIndex, Row, Col1 -Coln
InvCholVIndex, Row, Col1 -Coln
InvCovBmodel specific, Effect, Row, Col1 -Coln
InvGmodel specific, Effect, Subject, Sub1 -Subn, Group, Group1 -Groupn, Row, Col1 -Coln
InvRIndex, Row, Col1 -Coln
InvVIndex, Row, Col1 -Coln
IterHistoryCovP1 -CovPn, Iteration, Evaluations, M2ResLogLike, M2LogLike, Criterion
LRTDF, ChiSquare, ProbChiSq
LSMeansmodel specific, Effect, Margins, ByLevel, AT variables, Estimate, StandardError, DF, tValue, Probt, Alpha, Lower, Upper, Cov1 -Covn, Corr1 -Corrn
MMEqmodel specific, Effect, Subject, Sub1 -Subn, Group, Group1 -Groupn, Row, Col1 -Coln
MMEqSolmodel specific, Effect, Subject, Sub1 -Subn, Group, Group1 -Groupn, Row, Col1 -Coln
ModelInfoDescr, Value
ParmSearchCovP1 -CovPn, Var, ResLogLike, M2ResLogLike2, LogLike, M2LogLike, LogDetH
PosteriorDescr, Value
RIndex, Row, Col1 -Coln
RCorrIndex, Row, Col1 -Coln
SearchParm, TCovP1 -TCovPn, Posterior
Slicesmodel specific, Effect, Margins, ByLevel, AT variables, NumDF, DenDF, FValue, ProbF
SolutionFmodel specific, Effect, Estimate, StandardError, DF, tValue, Probt, Alpha, Lower, Upper
SolutionRmodel specific, Effect, Subject, Sub1 -Subn, Group, Group1 -Groupn, Estimate, StdErrPred, DF, tValue, Probt, Alpha, Lower, Upper
Tests1Effect, NumDF, DenDF, ChiSquare, FValue, ProbChiSq, ProbF
Tests2Effect, NumDF, DenDF, ChiSquare, FValue, ProbChiSq, ProbF
Tests3Effect, NumDF, DenDF, ChiSquare, FValue, ProbChiSq, ProbF
Type1Source, DF, SS, MS, EMS, ErrorTerm, ErrorDF, FValue, ProbF
Type2Source, DF, SS, MS, EMS, ErrorTerm, ErrorDF, FValue, ProbF
Type3Source, DF, SS, MS, EMS, ErrorTerm, ErrorDF, FValue, ProbF
TransPrior, TCovP, CovP1 -CovPn
VIndex, Row, Col1 -Coln
VCorrIndex, Row, Col1 -Coln

Some of the variables listed in Table 41.8 are created only when you have specified certain options in the relevant PROC MIXED statements.

The following changes have occurred in these variables from Version 6. Nearly all underscores have been removed from variable names in order to be compatible and consistent with other procedures. Some of the variable names have been changed (for example, T has been changed to tValue and PT to Probt) for the same reason. You may have to modify some of your Version 6 code to accommodate these changes.

Chapter Contents
Chapter Contents
Previous
Previous
Next
Next
Top
Top

Copyright © 1999 by SAS Institute Inc., Cary, NC, USA. All rights reserved.