Chapter 64: The SCORE Procedure

Overview

The SCORE procedure multiplies values from two SAS data sets, one containing coefficients (for example, factor-scoring coefficients or regression coefficients) and the other containing raw data to be scored using the coefficients from the first data set. The result of this multiplication is a SAS data set containing linear combinations of the coefficients and the raw data values.

Many statistical procedures output coefficients that PROC SCORE can apply to raw data to produce scores. The new score variable is formed as a linear combination of raw data and scoring coefficients. For each observation in the raw data set, PROC SCORE multiplies the value of a variable in the raw data set by the matching scoring coefficient from the data set of scoring coefficients. This multiplication process is repeated for each variable in the VAR statement. The resulting products are then summed to produce the value of the new score variable. This entire process is repeated for each observation in the raw data set. In other words, PROC SCORE cross multiplies part of one data set with another.

Raw Data Set

The raw data set can contain the original data used to calculate the scoring coefficients, or it can contain an entirely different data set. The raw data set must contain all the variables needed to produce scores. In addition, the scoring coefficients and the variables in the raw data set that are used in scoring must have the same names . See the section 'Getting Started' beginning on page 4067.

Scoring Coefficients Data Set

The data set containing scoring coefficients must contain two special variables: the _TYPE_ variable and the _NAME_ or _MODEL_ variable.

• The _TYPE_ variable identifies the observations that contain scoring coefficients.

• The _NAME_ or _MODEL_ variable provides a SAS name for the new score variable.

PROC SCORE first looks for a _NAME_ variable in the SCORE= input data set. If there is such a variable, the variable's value is what SCORE uses to name the new score variable. If the SCORE= data set does not have a _NAME_ variable, then PROC SCORE looks for a _MODEL_ variable.

For example, PROC FACTOR produces an output data set that contains factor-scoring coefficients. In this output data set, the scoring coefficients are identified by _TYPE_ ='SCORE'. For _TYPE_ ='SCORE', the _NAME_ variable has values of 'Factor1', 'Factor2', and so forth. PROC SCORE gives the new score variables the names Factor1 , Factor2 , and so forth.

As another example, the REG procedure produces an output data set that contains parameter estimates. In this output data set, the parameter estimates are identified by _TYPE_ ='PARMS'. The _MODEL_ variable contains the label used in the MODEL statement in PROC REG, or it uses MODEL n if no label is specified. This label is the name PROC SCORE gives to the new score variable.

Standardization of Raw Data

PROC SCORE automatically standardizes or centers the DATA= variables for you, based on information from the original variables and analysis from the SCORE= data set.

If the SCORE= scoring coefficients data set contains observations with _TYPE_ ='MEAN' and _TYPE_ ='STD', then PROC SCORE standardizes the raw data before scoring. For example, this type of SCORE= data set can come from PROC PRINCOMP without the COV option.

If the SCORE= scoring coefficients data set contains observations with _TYPE_ ='MEAN' but _TYPE_ ='STD' is absent, then PROC SCORE centers the raw data (the means are substracted) before scoring. For example, this type of SCORE= data set can come from PROC PRINCOMP with the COV option.

If the SCORE= scoring coefficients data set does not contain observations with _TYPE_ ='MEAN' and _TYPE_ ='STD', or if you use the NOSTD option, then PROC SCORE does not center or standardize the raw data.

If the SCORE= scoring coefficients are obtained from observations with _TYPE_ ='USCORE', then PROC SCORE 'standardizes' the raw data using the uncorrected standard deviations identified by _TYPE_ ='USTD', and the means are not subtracted from the raw data. For example, this type of SCORE= data set can come from PROC PRINCOMP with the NOINT option. For more information on _TYPE_ ='USCORE' scoring coefficients in TYPE=UCORR or TYPE=UCOV output data sets, see Appendix A, 'Special SAS Data Sets.'

You can use PROC SCORE to score the data that were also used to generate the scoring coefficents, although more typically, scoring results are directly obtained from the OUT= data set in a procedure that computes scoring coefficients. When scoring new data, it is important to realize that PROC SCORE assumes that the new data have approximately the same scales as the original data. For example, if you specify the COV option with PROC PRINCOMP for the original analysis, the scoring coefficients in the PROC PRINCOMP OUTSTAT= data set are not appropriate for standardized data. With the COV option, PROC PRINCOMP will not output _TYPE_ ='STD' observations to the OUTSTAT= data set, and PROC SCORE will only subtract the means of the original (not new) variables from the new variables before multiplying. Without the COV option in PROC PRINCOMP, both the original variable means and standard deviations will be in the OUTSTAT= data set, and PROC SCORE will subtract the original variable means from the new variables and divide them by the original variable standard deviations before multiplying.

In general, procedures that output scoring coefficients in their OUTSTAT= data sets provide the necessary information for PROC SCORE to determine the appropriate standardization. However, if you use PROC SCORE with a scoring coefficients data set that you constructed without _TYPE_ ='MEAN' and _TYPE_ ='STD' observations, you might have to do the relevant centering or standardization of the new data first. If you do this, you must use the means and standard deviations of the original variables, that is, the variables that were used to generate the coefficients, not the means and standard deviations of the variables to be scored.

See the section 'Getting Started' on page 4067 for further illustration.