CONSTRAINTS Subcommand (MULTIPLE IMPUTATION command)

The CONSTRAINTS subcommand specifies constraints on imputed values. CONSTRAINTS specifications are ignored when imputation is turned off.

To define constraints, specify one or more analysis variables and one or more of following in parentheses:

MIN = NONE | num. Minimum allowable imputed value for scale variables. Specify a number. If an imputed value is less than the minimum, the procedure draws another value until it finds one that is greater than or equal to MIN or the MAXCASEDRAWS or MAXPARAMDRAWS threshold is reached (See METHOD subcommand). There is no default minimum. MIN is ignored when predictive mean matching is used or when applied to a categorical variable. An error occurs if MIN is greater than or equal to MAX. For date format variables, values must be enclosed in single or double quotes and expressed in the same date format as the defined date format for the variable.

MAX = NONE | num. Maximum allowable imputed value for scale variables. Specify a number. If an imputed value is greater than the maximum, the procedure draws another value until it finds one that is less than or equal to MAX or the MAXCASEDRAWS or MAXPARAMDRAWS threshold is reached (See METHOD subcommand). There is no default maximum. MAX is ignored when predictive mean matching is used or when applied to a categorical variable. For date format variables, values must be enclosed in single or double quotes and expressed in the same date format as the defined date format for the variable.

RND = NO | num. Controls rounding of imputed values. By default imputed values are not rounded. To round imputed values, specify a numeric multiplier value. Imputed values are rounded to the nearest integer multiple of the specified multiplier. For example, if you specify RND=1, imputed values are rounded to the nearest integer; if you specify RND=0.1, imputed values are rounded to the nearest tenth. Note that no fuzz factor is added before rounding (that is, rounding is 'fuzzless'). RND is ignored when applied to a categorical or date format variable.

ROLE = BOTH | DEP | IND. Note: Missing values of variables that are defined as predictor-only (ROLE=IND) variables are imputed, but only for internal use (otherwise those values of the variable could not be used to impute other variables). Imputed values of such variables are not saved to the output dataset. When imputing predictor-only variables, a main-effects only model is always used; the INTERACTIONS keyword is ignored.

The variable list honors the TO and ALL keywords. Constraints are ignored when applied to a variable that is not an analysis variable.

The CONSTRAINTS subcommand can be specified more than once, and its settings are cumulative. That is, if more than one set of specifications is provided for a variable across CONSTRAINTS subcommands, the specifications are merged. When merging, if a particular keyword has different values for the same variable, the value specified in the last CONSTRAINTS subcommand for that variable is honored.

Cumulative Effects of Multiple Constraints

MULTIPLE IMPUTATION x y z
  /CONSTRAINTS x y z (RND=1 MIN=1)
  /CONSTRAINTS x     (MAX=5)
  /CONSTRAINTS y     (MIN=2 MAX=10)
  /CONSTRAINTS z     (ROLE=DEP)
  /OUTFILE IMPUTATIONS = imputedData.

• The first CONSTRAINTS subcommand specifies that imputed values of x, y, and z are rounded to the nearest integer and are constrained to be no less than 1.
• The second CONSTRAINTS subcommand specifies a maximum value of 5 for x.
• The third CONSTRAINTS subcommand specifies a maximum value of 10 for y, and overrules the previous minimum of 1 with a minimum of 2 for y.
• The fourth CONSTRAINTS subcommand specifies that z should be used as a dependent variable only in the imputation models.

The constraints can be summarized as follows: