topic Re: Simulating piecewise loglogistic data in SAS in Statistical Procedures

Simulating piecewise loglogistic data in SAS

EricL — Thu, 03 Dec 2020 18:52:48 GMT

hi-

I am simulating log-logistic data based on the below method but does anyone have SAS code for simulating a piecewise loglogistic model with 2 parts?

thanks

*Using F = 1 / ((1 + x/alpha)**beta), CDF for LLOGISTIC (LL)*;
*Simulate a uniform (0,1) random variable, so then the LL would correspond to*;
*x=alpha*((u/(1-u))**(1/beta))*;

Re: Simulating piecewise loglogistic data in SAS

Ksharp — Fri, 04 Dec 2020 13:15:39 GMT

calling @Rick_SAS

Re: Simulating piecewise loglogistic data in SAS

Rick_SAS — Tue, 08 Dec 2020 14:30:15 GMT

Can you provide more details, please? Provide a reference or more details about what you mean by a "2-part piecewise log-logistic model." How many explanatory variables? How many are continuous?

One interpretation is that you want to use a segmented model where the model has one form for x < x0 and another form for x > x0. You can use PROC NLIN to fit models like this.

For simulation, you can generate a random value for x. If x < x0 then the mean value (eta) is given by the first model. Otherwise, the mean is given by the second model. The apply logistic simulation as usual: p=logistic(eta); Y = rand("Bernoulli", p); For more details, see "How to simulate data from a generalized linear model."

Re: Simulating piecewise loglogistic data in SAS

EricL — Wed, 09 Dec 2020 17:17:32 GMT

thanks for response. Figured out what I needed to do to simulate data I needed. I new the mixture probability so I just did

if u<=p then x=scale1*((u/(1-u))**(1/shape1)), else x=scale2*((u/(1-u))**(1/shape2)) where U is uniform(0,1) like you said.