Rarely used specialist qualifiers

Introduction

Qualifiers are described in 4 sections based on frequency of use.

The major qualifiers are !CONTINUE, !CONTRAST, !FCON, !MAXIT, !SUBSET, !SUM and XYplots ( !X !Y !G !JOIN )

The qualifiers described here are rarely required. They include
    Controlling AI updates: !AISINGULARITIES, !AILOADINGS, !EMFLAG, !HOLD, !STEP, !SLOW, !VCC,
    Exporting the data: !PRINT, !RESIDUALS, !SAVE
    Best Linear Unbiased Prediction: !BLUP
    Modifying model defaults: !DENSE, !EXTRA
    Adjusting residual variance/degrees of freedom: !ADJUST, !DF, !YSS
    Modifying output format: !BRIEF, !PVSFORM, !TABFORM, !TXTFORM, !SLNFORM, !YHTFORM
    Equation ordering: !EQORDER, !LAST,
    Identify MYOWNGDG program: !OWN,
    Variogram settings: !SPATIAL, !TWOWAY, !VGSECTORS,
    Refining GLMM model fitting: !GLMM,
    Regression Screen: !SCREEN

Very rarely used qualifiers

Controlling AI updates

ASReml stops if the Average Information Matrix is singular unless !AISINGULARITIES is specified

Occasionally there are problems with the AI updates for factor analysis loadings which may be modified using !AILOADINGS.

When trying to estimate an unstructured US variance matrix, you may use !EMFLAG to request EM updates if the AI updates would make the matrix non positive definite. Consider using a different parameterization such as XFA.

As an ad hoc procedure, ASReml does not usually take the full AI update for the first few iterations. This is controlled by the !STEPSIZE and !SLOW qualifiers.

!HOLD list allows the user to temporarily fix the parameters listed. Parameter numbers have been added to the reporting of input values to facilitate use of this and other parameter number dependent qualifiers. The list should be in increasing order using colon to indicate a sequence. For example

 !HOLD 1:20 30:40.

The !VCC qualifier indicates how many Variance Component Constraint lines are supplied.

Own F tests

!FOWN terms to test ; background terms allows the user to explicity specify the particular F tests required.

Exporting the data:

!PRINT, controls exporting the (transformed) data to an ASCII file,
!RESIDUALS requests the data and residuals be written to a binary output file,
!SAVE requests the data be written to a binary output file.

!BLUP:

When the user only wants to perform a BLUP analysis and is not interested in estimating variance parameters (assuminng that they are known), the !BLUP qualifier controls how much of an iteration is performed before the BLUP effects are written to file.

Modifying model defaults:

!DENSE, modifiers how many (fixed) effects (and consequently model terms) can be included in the DENSE portion of the mixed model equations; the portion used to generate tests of fixed effects.
!EXTRA requests ASReml do more iterations after the usual convergence criterion have been satisfied.

Adjusting sum of squares and/or residual degrees of freedom:

Sometimes, a weighted analysis is performed where the weights are actually just providing a concise method of data entry. Then the mixed model equations actually represent more records than were actually supplied in the data file. Consequently the residual degrees of freedom will be wrong unless adjusted using either !ADJUST or !DF qualifiers.

When anaylsing summarised data, you may have an independent estimate of the variance. To include that variation in the analysis, use !YSS 100 !DF 20 say, to add the equivalent of 20 observations with a variance of 5.

Modifying output format:

The amount of detail written to the .pvs file can be modified using the !BRIEF qualifier.

The format of the .pvs, .sln, .tab and .yht files can be changed to TAB, COMMA or Ampersand separated ( .tex form) using the !PVSFORM, !TABFORM, !TXTFORM, !SLNFORM, !YHTFORM qualifiers. The file extension is also changed. For example .sln becomes _sln.txt if !SLNFORM is specified.

Equation ordering:

The processing efficiency of ASReml is very dependent on the order the equations are solved. Some user control is provided by the !EQORDER and !LAST qualifiers.

Defining your own variance structure.

If you want to fit a variance structure not already included in ASReml, write a separate program to create it and its derivatives. The program name is specified with !OWN program.

Variogram settings:

The variogram is a popular diagnostic plot for examining spatial dependence. By default ASReml will produce a variogram display for spatially dependent residuals or BLUPs. Some user control is possible with the !SPATIAL, !TWOWAY and !VGSECTORS qualifiers.

Refining GLMM model fitting

!GLMM n sets the number of inner iterations performed when a iteratively weighted least squares analysis is performed. Inner iterations are iterations to estimate the effects in the linear model for the current set of variance parameters. Outer iterations are the AI updates to the variance parameters. The default is to perform 4 inner iterations in the first round and 2 in subsequent rounds of the outer iteration. Set n to 2 or more to increase the number of inner iterations.

Alternative Outlier Model

!OUTLIER invokes a more formal calculation of the relative size of random effects and the residual, and reports values exceeding 3 standard deviations to the .res file. More?

Regression Screen:

A regression screen involves comparing the fit of (a subset of) all possible models obtained by selecting from a set of model terms. This is requested in ASReml with the !SCREEN [ !SMX ] qualifiers.

Return to index