Fits terms from a REML fixed model in a Genstat regression (R.W. Payne).
Options
PRINT = string tokens |
Controls printed output (model, deviance, summary, estimates, correlations, fittedvalues, accumulated); default mode, summ, esti, accu |
FACTORIAL = scalar |
Limit for expansion of terms; default 3 |
DENOMINATOR = string token |
Whether to base ratios in accumulated summary on rms from model with smallest residual ss or smallest residual ms (ss, ms); default ss |
SELECTION = string tokens |
One or two criteria to be printed with the models (%variance, %ss, adjustedr2, r2, dispersion, aic, sic, bic); default %var, aic, sic |
Parameter
TERMS = formula |
Fixed terms to be dropped |
Description
VRFIT is one of several procedures designed to improve the process of determining the appropriate fixed terms to include in a REML analysis. (The others are VRADD, VRDROP, VRDISPLAY, VRKEEP, VRSETUP, VRSWITCH and VRTRY.) They do this by a generalized regression analysis, with a weight matrix based on variances estimated from the original REML analysis (with the full fixed model). You can use the mean square of the current model to assess each change as in ordinary regression. However, as this is a weighted regression, the mean square from the full fixed model is one. A convenient alternative might therefore be to use this mean square (of one), to assess the terms with the same measure of random variation as in analysis of variance. (Conversely, if you were to assess the fixed model by changing the fixed model in a sequence of VCOMPONENTS commands, the fixed terms that are not fitted will be included in the random variation. This will then vary from fit to fit, making it difficult to reach a clear and consistent conclusion.) Having used these procedures to decide on the important fixed terms, you can use VPREDICT to form predicted means.
Before fitting the terms, the VRSETUP procedure must be called to make some checks, and initialize the regression by specifying a MODEL command with the necessary weight matrix and a TERMS command with the full fixed model. It also uses the WORKSPACE directive to set up a Genstat workspace structure to store control information and results. VRFIT will call VRSETUP for you, if you have not done so already. The analysis will then be based on the most recent REML analysis. To use an earlier analysis, you should call VRSETUP yourself, setting its SAVE option set to the save structure of the required REML analysis.
The terms to be fitted are specified by the TERMS parameter, in a similar way to the FIT directive. The FACTORIAL option sets a limit (by default 3) on the number of factors and variates in each term. Terms containing more than that number are omitted.
The PRINT option controls printed output as in the regression directives, except that some irrelevant settings are omitted. (For example, grid is relevant only to the fitting of nonlinear models.) See VRDISPLAY for more details.
The DENOMINATOR option specifies how the residual is selected for the accumulated analysis of variance By default it is taken from the model with the smallest number of residual degrees of freedom. However, you can set DENOMINATOR=ms to take it from the model with the smallest residual mean square.
The SELECTION option specifies the statistics to be displayed in the summary of analysis as in the regression directives, except that again some irrelevant settings are omitted. See VRDISPLAY for more details.
Options: PRINT, FACTORIAL, DENOMINATOR, SELECTION.
Parameter: TERMS.
Method
VRFIT calls the VRSETUP procedure to initialize the regression, if this has not been done already. It then uses the directives FIT and ADD to fit the terms one at a time, storing the results in the workspace together with the denominator degrees of freedom for each term if available from the original REML analysis. (It is the need to use the REML denominator degrees of freedom regression for the terms, instead of the residual degrees of freedom from the regression, that prevents you from using the regression commands directly.) Finally, VRFIT calls VRDISPLAY to print the results.
Action with RESTRICT
Any restriction applied to vectors used in the REML analysis will apply also to the results from VRFIT.
See also
Directives: FIT, REML.
Procedures: VRADD, VRDROP, VRDISPLAY, VRKEEP, VRSETUP, VRSWITCH, VRTRY, VALLSUBSETS, VSCREEN.
Commands for: REML analysis of linear mixed models.
Example
CAPTION 'VRFIT example',\
'Example 5.3.6 from The Guide to Genstat, Part 2 Statistics';\
STYLE=meta,plain
FACTOR [NVALUES=322; LEVELS=27] Dam
& [NVALUES=322; LEVELS=18] Pup
FACTOR [NVALUES=322; LEVELS=2; LABELS=!T('M','F')] Sex
FACTOR [NVALUES=322; LEVELS=3; LABELS=!T('C','Low','High')] Dose
VARIATE [NVALUES=322] Littersize,Weight
OPEN '%GENDIR%/Examples/GuidePart2/Rats.dat'; CHANNEL=chan
READ [CHANNEL=chan] Dose,Sex,Littersize,Dam,Pup,Weight; \
FREPRESENTATION=2(labels),4(levels)
CLOSE chan
VCOMPONENTS [FIXED=Littersize+Dose*Sex] RANDOM=Dam/Pup
REML [PRINT=model,components,wald] Weight
VRFIT [PRINT=estimates,accumulated] Littersize+Sex*Dose