Details


Missing Values

If one of the scoring variables in the DATA= data set has a missing value for an observation, all the scores have missing values for that observation. The exception to this criterion is if the PREDICT option is specified, the variable with a coefficient of ˆ’ 1 can tolerate a missing value and still produce a prediction score. Also, a variable with a coefficient of 0 can tolerate a missing value.

If a scoring coefficient in the SCORE= data set has a missing value for an observation, the coefficient is not used in creating the new score variable for the observation. In other words, missing values of scoring coefficients are treated as zeros. This treatment affects only the observation in which the missing value occurs.

Regression Parameter Estimates from PROC REG

If the SCORE= data set is an OUTEST= data set produced by PROC REG and if you specify TYPE=PARMS, the interpretation of the new score variables depends on the PROC SCORE options chosen and the variables listed in the VAR statement. If the VAR statement contains only the independent variables used in a model in PROC REG, the new score variables give the predicted values. If the VAR statement contains the dependent variables and the independent variables used in a model in PROC REG, the interpretation of the new score variables depends on the PROC SCORE options chosen . If you omit both the PREDICT and the RESIDUAL options, the new score variables give negative residuals (PREDICT-ACTUAL). If you specify the RESIDUAL option, the new score variables give positive residuals (ACTUAL-PREDICT). If you specify the PREDICT option, the new score variables give predicted values.

Unless you specify the NOINT option for PROC REG, the OUTEST= data set contains the variable Intercept . The SCORE procedure uses the intercept value in computing the scores.

Output Data Set

PROC SCORE produces an output data set but displays no output. The output OUT= data set contains the following:

  • the ID variables, if any

  • all variables from the DATA= data set, if no ID variables are specified

  • the BY variables, if any

  • the new score variables, named from the _NAME_ or _MODEL_ values in the SCORE= data set

Computational Resources

Let

v = number of variables used in computing scores

s = number of new score variables

b = maximum number of new score variables in a BY group

n = number of observations

Memory

The array storage required is approximately 8(4 v + (3 + v ) b + s ) bytes. When you do not use BY processing, the array storage required is approximately 8(4 v + (4 + v ) s ) bytes.

Time

The time required to construct the scoring matrix is roughly proportional to vs and the time needed to compute the scores is roughly proportional to nvs .




SAS.STAT 9.1 Users Guide (Vol. 6)
SAS.STAT 9.1 Users Guide (Vol. 6)
ISBN: N/A
EAN: N/A
Year: 2004
Pages: 127

flylib.com © 2008-2017.
If you may any questions please contact us: flylib@qtcs.net