Setting predict points


!PVAL covariate values
is a mechanism for specifying the particular points to be predicted for covariates modelled using fac(.), leg(.), spl(.) or pol(.)   model terms. The actualpoints predicted is controlled by the PREDICT directive. The points are specified here so that they can be included in the appropriate design matrices.
    covariate is the name of a data field.
    values is the list of values at which prediction is required.

!PVAL filename varlist
is used to read predictpints for several variables from a file filename. varlist is the names of the variables having values defined. If the file contains unwanted fields, put the pseudo variate label skip in the appropriate position in varlist to ignore them. The file should only have numeric values. predictpints cannot be specified for design factors.


!GKRIGE p controls the expansion of !PVAL lists for fac(X,Y) model terms.
    For kriging prediction in 2 dimensions (X,Y), the user will typically want to predict at a grid of values, not necessarily just at data combinations. The values at which the prediction is required can be specified separately for X and Y using two !PVAL statements. Normally, predict points will be defined for all combinations of X and Y values. This qualifier is required (with optional argument 1 ) to specify the lists are to be taken in parallel. The lists must be the same length if to be taken in parallel.


!PPOINTS n influences the number of points used when predicting splines and polynomials. The design matrix generated by the leg() , pol() and spl() functions are modified to include extra rows that are accessed by the PREDICT directive. The default value of n is 21 if there is no !PPOINTS qualifier.

The range of the data is divided by n-1 to give a step size i . For each point p in the list of unique covariate values, a predict point is inserted at p + i if there is no data value in the interval [ p , p +1.1 i ].

!PPOINTS is ignored if !PVAL is specified for the variable. This process also effects the number of levels identified by the fac() model term.

Return to start