Controls the operation of the REML algorithm.
Options
CONVERGENCE = string token |
Type of criterion for assessing convergence (deviance, parameter); default * uses the deviance with the average-information algorithm, and the variance parameter values for the Fisher scoring algorithm |
|---|---|
CRITERIONVALUE = scalar |
Sets the convergence criterion value; default * i.e. determined automatically |
STEPLENGTH = scalar |
Sets the default relative step size for the average-information algorithm; default * i.e. determined automatically |
NDENSE = scalar |
Number of equations to use as dense in the average-information algorithm; default * uses all fixed model terms as dense |
EQORDER = string token |
Method to use to reorder the mixed model equations for fitting (none, a, b); default b |
No parameters
Description
VCYCLE allows you to modify various aspects of the REML algorithm. The CONVERGENCE option specifies the type of criterion to use to assess convergence. There are two possibilities, each of which is used as the default for one of the fitting algorithms. For the average-information algorithm the default is to check for convergence in deviance, whereas the Fisher method checks the variance parameter values. The criterion value can be specified by the CRITERIONVALUE option. The defaults differ according to the type of criterion. For assessing changes in variance parameter values a multiplier of 0.005 is used. So, for convergence, the change in every variance parameter s must be less than 0.005 × s. When assessing change in deviance, convergence occurs when the absolute change in the deviance is less than 0.0001.
The STEPLENGTH option allows you to change the default step size for the average-information algorithm. Valid values are between zero and one, and the value is the proportion of the average-information step taken. The default is to start with small steps and work up to full steps.
The NDENSE option allows you to manipulate the number of equations used as dense in the average-information algorithm. The default includes all the fixed model terms. This option is likely to be used only by advanced users. If NDENSE is set, the value may be modified by the algorithm so that model terms are not split between the dense and sparse sections. Note that Wald tests (dropping terms) are not available for terms in the sparse section.
The EQORDER option controls the order in which the mixed model equations are solved, with settings:
none |
processes the equations in the order in which they are specified in the model; |
|---|---|
a |
method A; and |
b |
method B (default). |
This option needs to be set only rarely as method B, which corresponds to the ASReml option setting !EQORDER 3 (introduced to become the default in ASReml Release 2), is generally the best. Method A corresponds to the ASReml option setting !EQORDER 1 (which was the default in ASReml Release 1). For further details, see ASReml User Guide Release 2.
Options: CONVERGENCE, CRITERIONVALUE, STEPLENGTH, NDENSE, EQORDER.
Parameters: none.
See also
Directive: REML.
Commands for: REML analysis of linear mixed models.