Examples (MULTIPLE IMPUTATION command)

Automatic Imputation 

DATASET DECLARE imputedData.
MULTIPLE IMPUTATION x y z
  /OUTFILE IMPUTATIONS = imputedData.
  • The command requests multiple imputation of scale variables x, y, and z.
  • By default, the procedure chooses an imputation method based on a scan of the data. Five imputations are produced by default.
  • As all variables are scale, a linear regression model is used for each. When a particular variable is imputed, all other variables are treated as covariates in the imputation model if FCS is automatically chosen as the imputation method. If the monotone method is chosen, only preceding variables are used as predictors.
  • The imputed data are written to a new dataset imputedData.
  • By default, an overall summary of missingness is displayed, as well as information about how values were imputed.

Analysis of Missingness 

MULTIPLE IMPUTATION x y z
  /IMPUTE METHOD=NONE
  /MISSINGSUMMARIES VARIABLES.
  • The command requests analysis of missing values for variables x, y, and z.
  • No imputation is performed.

Custom Imputation Model 

DATASET DECLARE imputedData.
MULTIPLE IMPUTATION x y z
  /IMPUTE METHOD=FCS
  /CONSTRAINTS x y (RND=1 MIN=1)
  /CONSTRAINTS x   (MAX=5)
  /CONSTRAINTS y   (MAX=10)
  /OUTFILE IMPUTATIONS = imputedData.
  • Scale variables x, y, and z are defined as analysis variables.
  • Missing values are imputed using the FCS method. Variables are imputed in order of their appearance in the variable list. Variable x is imputed first, using x and z as predictors. Then y is imputed, treating x and z as predictors. Finally, z is imputed using x and y as predictors.
  • CONSTRAINTS subcommands specify constraints on imputed values. Imputed values of x and y are rounded to the nearest integer and are constrained to be no less than 1. Maximum values of 5 and 10 are specified for x and y, respectively. No constraints are defined for variable z.
  • Imputed data are written to a new dataset imputedData.
  • By default, an overall summary of missingness is displayed, as well as information about how values were imputed.