Chapter Contents |
Previous |
Next |

Details of the OPTEX Procedure |

**MODEL***effects < / options >***;**-
You use the MODEL statement to specify the independent effects used to
model data that are to be collected with the design that is being
constructed. The
*effects*can be- simple continuous regressor effects
- polynomial continuous effects
- main effects of classification variables
- interactions of classification variables
- continuous-by-class effects

The variables used to form*effects*in the MODEL statement must be present in all input data sets. For details on input data sets, see "Input Data Sets" . For details on the specification of different types of effects and on how the design matrix is defined with respect to the effects, see "Specifying Effects in MODEL Statements" .

If you specify a data set containing fixed covariate effects with a DESIGN= data set in the BLOCKS statement, then a CLASS or MODEL statement that*follows*the BLOCKS statement refers to the model for the fixed covariates. A CLASS or MODEL statement that defines the model for the candidate points (treatment model) should occur*before*the BLOCKS statement.

The following options can be used in the MODEL statement: **NOINT**-
excludes the intercept parameter from the model. By default,
the OPTEX procedure includes the intercept parameter in the model.
**PRIOR=***num-list*-
specifies prior precision values corresponding to groups of effects
in the model. Groups of effects in the MODEL statement with the same
prior precision must be separated by commas. Then use the PRIOR= option,
listing as many prior precision values as there are groups of effects.
See Example 24.6 for an example.

When you specify prior precision values, the information matrix for estimating the linear parameters is*X*'*X*+*P*, where*X*is the design matrix and*P*is a diagonal matrix with the prior precision values that you specify on the diagonal. Thus, in terms of a prior distribution, the inverses of the prior precision values can be interpreted as prior variances for the linear parameters corresponding to each effect. As an alternative interpretation, note that with orthogonal coding the value of the prior for an effect says roughly how many prior "observations' worth" of information you have for that effect. See "Design Coding" for details on orthogonal coding.

Chapter Contents |
Previous |
Next |
Top |

Copyright © 1999 by SAS Institute Inc., Cary, NC, USA. All rights reserved.