topic How to randomly generate Y from zero-inflated negative binomial dbn fit with proc genmod? in Statistical Procedures

How to randomly generate Y from zero-inflated negative binomial dbn fit with proc genmod?

RobertWF1 — Mon, 10 Jan 2022 13:25:39 GMT

I've recently fit a zero-inflated negative binomial distribution to hospital inpatient visit data using proc genmod:

ods trace on;
proc genmod data = IP_count_data;
  	model Y = / dist=zinb;
  	zeromodel;
	output out=ZINBFit p=p pzero=pzero;
	ods output ZeroParameterEstimates = ZeroParameterEstimates; 
	ods output ParameterEstimates = ParameterEstimates;
run;

I'd like to randomly generate observations Y from this same zero-inflated negative binomial distribution. This can be done using the following expression, pulled from an earlier question in the forum:

Y_simulated = rand('BERNOULLI', 1-pzero)*rand('NEGBINOMIAL', k, p);

I can use the pzero = zero inflation probability parameter generated from proc genmod.

However, proc genmod does not produce the parameters k (= # of successes) and p (probability of success) which are necessary in the rand('NEGBINOMIAL', k, p) function. Can k and p be manually calculated from the proc genmod output?

Re: How to randomly generate Y from zero-inflated negative binomial dbn fit with proc genmod?

WeiChen — Mon, 10 Jan 2022 14:17:53 GMT

@Rick_SAS might know how to do this.

Re: How to randomly generate Y from zero-inflated negative binomial dbn fit with proc genmod?

RobertWF1 — Mon, 10 Jan 2022 14:33:24 GMT

Maybe I should repost my question in the Statistical Procedures forum?

https://communities.sas.com/t5/Statistical-Procedures/bd-p/statistical_procedures

Re: How to randomly generate Y from zero-inflated negative binomial dbn fit with proc genmod?

StatDave — Mon, 10 Jan 2022 16:22:41 GMT

See this note that shows how to estimate distribution parameters for use in DATA step functions, including the negative binomial.

Re: How to randomly generate Y from zero-inflated negative binomial dbn fit with proc genmod?

RobertWF1 — Mon, 10 Jan 2022 22:38:37 GMT

Fantastic, thank you!