HPSEVERITY Procedure

specifies the name of the output data set. If you do not specify this option, then PROC HPSEVERITY names the output data set by using the DATAn convention.

specifies the names of the variables that you want to copy from the input DATA= data set to the OUT= data set. If you want to specify more than one name, then separate them by spaces.

If you specify the BY statement, then the BY variables are not automatically copied to the OUT= data set, so you must specify the BY variables in the COPYVARS= option.

specifies the scoring functions that you want to estimate. For each scoring function that you want to estimate, specify the suffix of the scoring function as the function. For each function that you specify in this option and for each distribution D that you specify in the DIST statement, the FCMP function D_function must be defined in the search path that you specify by using the CMPLIB= system option.

If you want to evaluate the scoring function at a specific value of the response variable, then specify a number arg, which is enclosed in parentheses immediately after the function. If you do not specify arg or if you specify a missing value as arg, then for each observation in the DATA= data set, PROC HPSEVERITY computes the value v by using the following table and evaluates the scoring function at v, where y, r, and l denote the values of the response variable, right-censoring limit, and left-censoring limit, respectively:

Right-Censored	Left-Censored	v
No	No	y
No	Yes	l
Yes	No	r
Yes	Yes

You can specify the suffix of the variable that contains the estimate of the scoring function by specifying a valid SAS name as a variable. If you do not specify a variable, then PROC HPSEVERITY uses function as the suffix of the variable name.

To illustrate the FUNCTIONS= option with an example, assume that you specify the following DIST and OUTPUT statements:

dist exp logn;
output out=score functions=(cdf pdf(1000)=f1000 mean);

Let both exponential (EXP) and lognormal (LOGN) distributions converge. If is the final estimate of the scale parameter of the exponential distribution, then PROC HPSEVERITY creates the following three scoring function variables for the exponential (EXP) distribution in the Work.Score data set:

Similarly, if and are the final estimates of the log-scale and shape parameters of the lognormal distribution, respectively, then PROC HPSEVERITY creates the following three scoring function variables for the lognormal (LOGN) distribution in the Work.Score data set:

If you specify the SCALEMODEL statement, then the value of the scale parameter of a distribution depends on the values of the regression parameters. So it might be different for different observations. In this example, the values of and might vary by observation, which might cause the values of the EXP_F1000, EXP_MEAN, LOGN_F1000, and LOGN_MEAN variables to vary by observation. The values of the EXP_CDF and LOGN_CDF variables might vary not only because of the varying values of v but also because of the varying values of and .

If you do not specify the SCALEMODEL statement, then the values of scoring functions for which you specify a nonmissing argument arg and scoring functions that do not depend on the response variable value do not vary by observation. In this example, the values of the EXP_F1000, EXP_MEAN, LOGN_F1000, and LOGN_MEAN variables do not vary by observation.

If a distribution does not converge, then the scoring function variables for that distribution contain missing values in all observations.

For more information about scoring functions, see the section Scoring Functions.

specifies the quantiles that you want to estimate. To use this option, for each distribution that you specify in the DIST statement, the FCMP function D_QUANTILE must be defined in the search path that you specify by using the CMPLIB= system option.

You can specify the following quantile-options:

For example, assume that you specify the following DIST and OUTPUT statements:

dist burr;
output out=score quantiles=(cdf=0.9 0.975 0.995 names=ninety var);

PROC HPSEVERITY creates three quantile estimate variables, BURR_NINETY, BURR_VAR, and BURR_P99_5, in the Work.Score data set for the Burr distribution. These variables contain the estimates of , for p = 0.9, 0.975, and 0.995, respectively, where denotes the quantile function and , , and denote the parameter estimates of the Burr distribution.

If you specify the SCALEMODEL statement, then the quantile estimate might vary by observation, because the scale parameter of a distribution depends on the values of the regression parameters.

If you do not specify the SCALEMODEL statement, then the quantile estimates do not vary by observation, and if you do not specify any scoring functions in the FUNCTIONS= option whose estimates vary by observation, then the OUT= data set contains only one observation per BY group.

If a distribution does not converge, then the quantile estimate variables for that distribution contain missing values for all observations.