turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

- Home
- /
- SAS Programming
- /
- SAS Procedures
- /
- Proc MCMC Error: doesn't produce a positive log li...

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

06-30-2016 09:37 AM

Hi,

I am applying MCMC in SAS University edition, I got the error that the log likelihood is not positive.

This is a part of the code.

```
p= A**(x+B)**C + D*exp(-1*E*(log(x)-log(F))**2) + G*(H**x)/(1+G*(H**x));
model y ~ binomial(n,p);
```

A to H are parameters which have prior distribution of uniform.

Thanks.

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

Posted in reply to Marwah

06-30-2016 11:27 AM

What is the range and typical values of x? Obviously x must be greater than 0, but also x > B so that the expression (x-B)**C is defined. Is B always bounded above by some B0 and x > B0? Otherwise you're going to have problems.

It also looks like it is possible to choose parameters such that p is greater than 1, which will cause problems in the binomial(n,p) computation.

What are you trying to accomplish?

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

Posted in reply to Rick_SAS

07-05-2016 08:40 AM

x is age grouped in five years, it ranges from 0 to 110 (eg. 0,5,10,...110). B ranges from 0 to 1. The exponent is (x+B) not (x-B) so it is always defined.

It rarely happens in the models that p is greater than 1 but anyway I put a condition that if it is greater than 1 then substitute with 0.999.

I am trying to approximate a posterior distribution of the parameters (A to H) through slice sampling or Metropolis Hastings method. but the problem here is that the deaths is binomial and the prior distributions are not for the probability of success of the binomial, they are for the parameters that are in Heligman Pollard model (the p equation).

Sorry for replying late and thank you for your time.