For a frequency distribution to be accepted through the COUNTSTORE= option of PROC HPCDM, you need to estimate it with PROC COUNTREG. Unfortunately, PROC COUNTREG does not support binomial distribution. However, it does support the negative binomial distribution, which is more commonly used to model count data than the binomial distribution. If you cannot use the negative binomial distribution, then you will need to use the external counts feature of PROC HPCDM. Steps to take: Estimate the binomial distribution by using something other than PROC COUNTREG. Simulate a sample data set of counts by using your estimated binomial distribution. You can use DATA step to do this. Say this data set is called WORK.FOO and stores the generated count in a variable called COUNT. Invoke PROC HPCDM with DATA=WORK.FOO along with other options as needed and specify "EXTERNALCOUNTS COUNT=COUNT;" statement to tell PROC HPCDM which variable contains the simulated counts.   I do not understand your question about "use a constant. Can you please elaborate? Do you want to simulate with a constant value of loss count? In that case, you can create a data set that contains one row and one column with column value being the constant value of the count you want to use, and then use the steps that are listed above to simulate with that constant count. If you want the sample size to be a multiple of the constant count, specifying the multiplier as the NREPLICATES= option.   Hope this answers your questions. ... View more

There are many ways to fit distributional parameters to data: method of moments, optimization of the likelihhod equation, optimization of an approximate likelihood, matching percentiles, and so forth.  The estimates can also depend on the initial guess and the numerical optimization technique.   The documentation for the procedures descrbe how the parameters are estimated. The doc tor PROC SEVERITY shows how parameters are initialized and what estimation techniques are used. For example, for a gamma distribution, an APPROXIMATE MLE is formed by approximating the digamma function.  This results in a fast estimation. In contrast, the UNIVARIATE procedure solves the true MLE by using the full digamma fnuction.  It is slower, but more accurate.   In short, you are not going to be able to get exactly the same answers for the distributions for which SEVERITY and UNIVARIATE use different estimation methods. For distributions like the gamma, the documentation indicates that the estimates might be off by a few percentage points.        ... View more

There is not procedure that fits multiple discrete families and chooses the best. However, you can use PROC GENMOD to fit many discrete distributions to data. An example is given in the article "Fitting a Poisson distribution to data in SAS."    /* number of emails received in each half-hour period 8:00am - 5:30pm on a weekday. */ data MyData; input N @@; /* 8a 9a 10a 11a 12p 1p 2p 3p 4p 5p */ datalines; 7 7 13 9 8 8 9 9 5 6 6 9 5 10 4 5 3 8 4 ; /* 1. Estimate the rate parameter with PROC GENMOD: http://support.sas.com/kb/24/166.html */ proc genmod data=MyData; model N = / dist=poisson; output out=PoissonFit p=lambda; run; The results of PROC GENMOD include various goodness-of-fit (GOF) statistics succh as AIC, AICC, and BIC.  If you rerun the analysis for another distribution such as negative binomial (DIST=NEGBIN) or geometric (DIST=NEGBIN SCALE=1 NOSCALE) you can compare the various GOF statistics and choose the distribution that gives the smallest value.     Regarding QQ or PP plots, discrete families do not have them (see the previous article). Some researchers have attempted to provide other graphical techniques that represent the goodness of fit, but these attempts have caught on nor gained general acceptance in the statistical community. ... View more

I would like the hcpdm for severity only, i would a frequency equal to a constant 1.   ... View more