Variable Lists (MLP command)

The command line variable lists specify the dependent variables, any categorical predictors (also known as factors), and any scale predictors (also known as covariates).

Dependent Variables

  • A list of one or more dependent variables must be the first specification on the MLP command.
  • Each dependent variable may be followed by the measurement level specification, which contains, in parentheses, the MLEVEL keyword followed by an equals sign and then S for scale, O for ordinal, or N for nominal. MLP treats ordinal and nominal dependent variables equivalently as categorical.
  • If a measurement level is specified, then it temporarily overrides a dependent variable’s setting in the data dictionary.
  • If no measurement level is specified, then MLP defaults to the dictionary setting.
  • If a measurement level is not specified and no setting is recorded in the data dictionary, then a numeric variable is treated as scale and a string variable is treated as categorical.
  • Dependent variables can be numeric or string.
  • A string variable may be defined as ordinal or nominal only.

Predictor Variables

  • The names of the factors, if any, must be preceded by the keyword BY.
  • If keyword BY is specified with no factors, then a warning is issued and BY is ignored.
  • The names of the covariates, if any, must be preceded by the keyword WITH.
  • If keyword WITH is specified with no covariates, then a warning is issued and WITH is ignored.
  • A dependent variable may not be specified within a factor or covariate list. If a dependent variable is specified within one of these lists, then an error is issued.
  • All variables specified within a factor or covariate list must be unique. If duplicate variables are specified within a list, then the duplicates are ignored.
  • If duplicate variables are specified across the factor and covariate lists, then an error is issued.
  • The universal keywords TO and ALL may be specified in the factor and covariate lists.
  • Factor variables can be numeric or string.
  • Covariates must be numeric.
  • If no predictors at all are specified, then the procedure fits an input layer containing only the bias unit—that is, the constant-only input layer.
  • At least one predictor must be specified.