SAMPLING Subcommand (BOOTSTRAP command)

The SAMPLING subcommand is used to specify the sampling method and any associated variables.

  • If SAMPLING is not specified, the procedure performs simple bootstrap resampling..

SIMPLE. Simple resampling. This performs case resampling with replacement from the original dataset. This is the default.

STRATIFIED (STRATA = varlist). Stratified resampling. Specify one or more variables that define strata within the dataset. This performs case resampling with replacement from the original dataset, within the strata defined by the cross-classification of strata variables, preserving the size of each stratum. Stratified bootstrap sampling can be useful when units within strata are relatively homogeneous while units across strata are very different.

RESIDUAL (RESIDUALS=varlist | PREDICTED=varlist). Residual resampling. Specify one or more variables containing residuals from fitting a model to the data. The model that produced the residuals should ideally be the same model that follows BOOTSTRAP. A residual sample is drawn by replacing each target variable value with that case's predicted value plus a residual sampled from the entire original set of residuals.

Specify PREDICTED as an alternative to RESIDUALS when the model residuals are not immediately available but the predicted values are. Specify one or more variables containing predicted values from fitting a model to the data.

If RESIDUAL is specified, the TARGET keyword is required and the variables specified on RESIDUAL should be the residuals (or predicted values) for, and match the order of, the variables specified on TARGET.

WILD (RESIDUALS=varlist | PREDICTED=varlist). Wild bootstrap resampling. Specify one or more variables containing residuals from fitting a model to the data. The model that produced the residuals should ideally be the same model that follows BOOTSTRAP. A wild sample is drawn by replacing each target variable value with that case's predicted value plus either the case's residual or the negative of the case's residual.

Specify PREDICTED as an alternative to RESIDUALS when the model residuals are not immediately available but the predicted values are. Specify one or more variables containing predicted values from fitting a model to the data.

If WILD is specified, the TARGET keyword is required and the variables specified on WILD should be the residuals (or predicted values) for, and match the order of, the variables specified on TARGET.