Modifies a model-definition structure for a REML analysis to estimate BLUPs instead of BLUEs (R.W. Payne).
Options
PRINT = string tokens |
Controls printed output (model, components, effects, means, stratumvariances, monitoring, vcovariance, deviance, Waldtests, missingvalues, covariancemodels); default mode, comp, effe |
RANDOM = formula |
Additional random model terms |
PTERMS = formula |
Terms (fixed or random) for which effects or means are to be printed; default prints them for the terms specified by RANDOM |
PSE = string token |
Standard errors to be printed with tables of effects and means (differences, estimates, alldifferences, allestimates, none); default diff |
VCONSTRAINTS = string token |
Modifies the constraints on the variance components (none, positive); default * keeps the existing constraints |
Parameters
Y = variates |
Response variates |
MODELSTRUCTURE = pointer |
Specifies the model-definition structure of the original analysis; no default (must be specified) |
NEWMODELSTRUCTURE = pointer |
Saves the modified model-definition structure, with terms now fitted as random |
SAVE = REML save structures |
Save structure from the modified analysis |
Description
REML model-definition structures are designed to simplify the assessment of alternative models for a REML analysis. They can be defined by the VFMODEL and VFSTRUCTURE procedures, and used as input to procedures like VARANDOM to run the various analyses and select the one with the most effective random model. This process is used, for example, within the VABLOCKDESIGN and VAROWCOLUMNDESIGN procedures to analyse data from block and row-and-column designs.
When assessing the random model it may be best to concentrate only on the “design” aspects, and fit terms like genotype or variety as fixed. VA2BLUPS can then be used to redo the analysis, but with terms like genotype or variety now fitted as random terms to obtain best linear unbiased predictors (BLUPs) instead of best linear unbiased estimates (BLUEs).
The model-definition structure for the original analysis must be specified by the MODELSTRUCTURE parameter. The terms that should now be fitted as random instead of fixed are specified by the RANDOM option. The VCONSTRAINTS option allows you to modify the constraints on the variance components; the default is to keep the existing constraints. The modified model-definition structure, with these terms now fitted as random, can be saved by the NEWMODELSTRUCTURE parameter.
The Y parameter specifies the response variate, and you can save the REML save structure from the analysis with the SAVE parameter. The PRINT option controls the printed output from the analysis, as in REML. The PTERMS option specifies the terms whose means and effects are to be printed; the default is to print them for the terms specified by the RANDOM option. The PSE option controls the type of standard error that is displayed with the means and effects; the default is to give a summary of the standard errors of differences.
Options: PRINT, RANDOM, PTERMS, PSE, VCONSTRAINTS.
Parameters: Y, MODELSTRUCTURE, NEWMODELSTRUCTURE, SAVE.
See Also
Directive: REML.
Procedures: VFMODEL, VFSTRUCTURE, VABLOCKDESIGN, VAROWCOLUMNDESIGN, VARANDOM.
Commands for: REML analysis of linear mixed models.
Example
CAPTION 'VA2BLUPS example',\
'Slate Hall Farm data (Guide to REML in Genstat, Section 1.8).';\
STYLE=meta,plain
SPLOAD '%data%/Slatehall.gsh'
" find best row-column model (with variety as a fixed term) "
VAROWCOLUMNDESIGN [PRINT=best,description,deviance,aic,sic,dfrandom;\
PTRY=*; FIXED=variety; REPLICATES=replicates;\
ROWS=fieldrow; COLUMNS=fieldcolumn;\
TRYSPATIAL=always; TRYTRENDS=yes;\
RSTRATEGY=fastoptimal]\
Y=yield; BESTMODEL=bestmodel; SAVE=savebest
VDISPLAY [PRINT=model,components,wald] savebest
" form BLUPS for variety (with same covariance model) "
VA2BLUPS [RANDOM=variety] yield; MODELSTRUCTURE=bestmodel