1. Home
  2. All Subsets for REML Fixed Terms

All Subsets for REML Fixed Terms

Use this to search through all sub-models for the Fixed model of the Linear Mixed Model (REML) menus. This uses the VALLSUBSETS procedure to fit all subsets of the fixed model and then 1 or 2 selection criteria to select the best model. This is accessed from the Linear Mixed Model Further Output dialog.

Forced terms

It is sometimes desirable to include specific terms in every model. Such terms may be specified by in this box, which may be set using a model formula or using a list of terms separated by commas.

Factorial limit on forced terms

You can control the factorial limit on model terms to be generated when you use model-formula operators like ‘*’.

Number of models

This specifies the number of models to print, sorted on the selection criteria. For example, if this was 5, only the 5 top models would be printed. If this is set to missing, *, then all models are printed.

Marginal terms

How to treat terms that are marginal to other terms.

Forced Terms that are marginal to another fixed term are treated as forced
Free Terms that are marginal to another fixed term are retained in the “free” terms that are used to form the subsets

Selection criteria

Specifies the model selection criteria used to pick the best models. You may select 1 or 2 or these criteria.

R-squared % sum of squares accounted for (taking the total sum of squares as the residual from the forced model) = 100 × [1 – Dev / Dev0]
Adjusted R-squared % variance accounted for (compared to the residual mean square from the forced model) = 100 × [1 – (Dev / (n-p)) / (Dev0 / (n-p0))]
Mallows Cp Mallows Cp = Dev / f + 2 × p – n
Mean squared prediction error Mean squared prediction error = Dev × (n+1) × (n-2) / [n × (n-p) × (n-p-1)]
Akaike information criterion Akaike information criterion = Dev / f + 2 × p
Schwarz/Bayesian information criterion Schwarz or Bayesian information criterion = Dev / f + Ln(n) × p
Residual mean square/deviance Residual mean square or deviance = Dev / (n-p)
Residual sum of squares/deviance Residual sum of squares or deviance = Dev

where Dev is the deviance or residual sums of squares of the current model, Dev0 is the deviance or residual sums of squares of the null model, p is the number of fitted parameters of the current model, p0 is the number of fitted parameters of the null model, n is the number of units and f is the dispersion parameter.


Use these fields to save results from the all subsets analysis in Genstat data structures. After selecting the appropriate boxes, type the identifiers of the data structures into the corresponding In: fields.

Results Pointer Pointer to save variates and scalars containing the criteria for the sets, and F and Wald statistics for the terms that they contain
Best model Pointer Pointer to formula of the best models according to the selected criteria

Display in spreadsheet

Select this to display the results in a new spreadsheet window. Note as the best model contains formulae, it is not displayed in a spreadsheet. The results are displayed on two pages as some are vectors and some are scalars.

See also

Updated on February 24, 2021

Was this article helpful?