topic Simulate blocks and treatments in SAS Programming

Simulate blocks and treatments

vdfdd — Mon, 21 Jan 2019 04:29:00 GMT

Please, I want to simulate a dataset with 4 blocks and 4 treatments and counts following a Poisson and negative binomial distribution. Is it possible is SAS?

Re: Simulate blocks and treatments

PGStats — Mon, 21 Jan 2019 05:25:49 GMT

Sure,

function RAND generates pseudo random numbers from many distributions

Pois = rand("Poisson", m); /* Generates a Poisson variate */

NegBin = rand("NegBinomial", p, k); /* Genarates a negative binomial variate */

Check the documentation for the precise parameterisation.

Re: Simulate blocks and treatments

Rick_SAS — Mon, 21 Jan 2019 14:04:31 GMT

Here is an example. Your model will determine the definitions of mu, r, and p. I just invented a few values to demonstrate the basic idea.

data SimBlock;
call streaminit(123);
do Block = 1 to 4;
   do Treatment = 1 to 4;
      /* invent a formula for the expected count */
      mu = ( Block + (Treatment-2.5)**2 );
      PoisCount = rand("Poisson", mu);
      /* invent a probability and number of trials */
      r = mu / 8; /* prob of success */
      k = 1;        /* number of successes */
      NBCount = rand("NegBin", r, k); /* number of failures before k successes */
      output;
   end;
end;
drop mu k r;
run;

proc means data=SimBlock;
run;

Re: Simulate blocks and treatments

vdfdd — Wed, 06 Feb 2019 05:33:38 GMT

Thank you for your response.

Re: Simulate blocks and treatments

vdfdd — Wed, 06 Feb 2019 05:38:26 GMT

Thank you for your response. Please I have 2 questions:
1. In defining the mu (mu = ( Block + (Treatment-2.5)**2 )), you added 2.5. Why?
2. You defined r = mu / 8; Why the value 8?

Re: Simulate blocks and treatments

Rick_SAS — Wed, 06 Feb 2019 11:35:33 GMT

No reason. I just invented some numbers because you did not provide any information about the design of the simulation. The number 2.5 represents the average treatment effect. I used 8 to make sure r is a probability in [0, 1].

Re: Simulate blocks and treatments

vdfdd — Wed, 06 Feb 2019 15:09:57 GMT

Thank you for your prompt response. Here is the description of the experiment :
i. block - 4
ii. treatments = 4
iii. repetition = 8
iv. counts = containing some zeros

Question: Simulate the count as coming from
i. a NegBin distribution
ii. a Poisson distribution

Thank you, again, in anticipation.