This package provides various relative importance measures for regression explanatory variables and shows how regression coefficients vary as the model size changes.
- IBM SPSS Statistics 18 or later and the corresponding IBM SPSS Statistics-Integration Plug-in for R.
Note: For users with IBM SPSS Statistics version 21 or higher, the STATS RELIMP extension is installed as part of IBM SPSS Statistics-Essentials for R.
- Open IBM SPSS Statistics
- Navigate to Utilities -> Extension Bundles -> Download and Install Extension Bundles
- Search for the name of the extension and click Ok. Your extension will be available.
STATS RELIMP DEPENDENT=variable^* FORCED=variables ENTER=variables^* MEASURE=LMG^** FIRST LAST BETASQ PRATT
/OPTIONS SCALE=NO^** or YES RANKS=NO^** or YES MISSING=LISTWISE^** or STOP
[/HELP]
STATS RELIMP /HELP prints this information and does nothing else.
^* Required
^** Default
Example:
STATS DEPENDENT=y ENTER=x1 x2 x3 MEASURE=LMG FIRST LAST
DEPENDENT and ENTER specify the dependent variable and the candiate predictors.
FORCED variables are always included in the equation. The importance measures and related statistics are calculated for the ENTER variables. Categorical variables are converted to factors.
Occasionally the computations fail with a singularity message. If this occurs, it may help to rescale variables with very large values.
MEASURE specifies one or more importance measure calculated for each ENTER variable. The choices are
- LMG: also know as the Shapley value - the incremental R2 for the variable averaged over all models
- FIRST: the R2 when only that variable is entered
- LAST: the incremental R2 when the variable is entered last
- BETASQ: the square of the standardized coefficient
- PRATT: the standardized coefficient times the correlation
- PMVD: the proportional marginal variance decomposition as proposed by Feldman. It can be interpreted as a weighted average over orderings among regressors, with data-dependent weights. PMVD is only available in the non-US version of the relaimpo package which must be obtained from http://prof.beuth-hochschule.de/groemping/relaimpo and is licensed for use only outside the United States. For this reason, this extension command has not been tested with this option.
All the measure except for FIRST and LAST sum to the overall R2.
SCALE=YES scales the importance measure to sum to 100%. This is not meaningful for FIRST and LAST.
If RANKS=YES, a table is produced showing the rank of each variable for each importance measure specified.
By default, the calculations are carried out only on complete cases. MISSING=STOP stops the calculation if any missing values are encountered.
- Apache 2.0
- JKP, IBM SPSS