I realize that this should be probably be an easy question, but I'm
drawing a mental block. Any suggestions on the following problem
would be appreciated.

Problem: I have a large simulation program that, given a set of fixed
input conditions, provides a random response D.

The general problem is structured: P(D|a,g,h) P(g) P(h) P(a | b) P(b|
c) P( c) . G, H, C are random variables with random statistical
charateristics, e.g. mean and variance are random variables. Once C is
sampled, I know P(b|c) and P(a|b) [fixed probabilities]. One goal of
the analysis is to characterize the CDF of P(D| c) .

It seems to me that this should be relatively easy to setup in
Winbugs, but I seem to be making it more complicated than it should
be. Can I treat the parameters P(a | b) and P(b|c) as just weights?
Any suggestions on a WinBugs formulation would be very welcome.

No, they're stochastic nodes: draw the DAG and it should become clear.

Jeanwrote:
I realize that this should be probably be an easy question, but I'm
drawing a mental block. Any suggestions on the following problem
would be appreciated.

Problem: I have a large simulation program that, given a set of fixed
input conditions, provides a random response D.

The general problem is structured: P(D|a,g,h) P(g) P(h) P(a | b) P(b|
c) P( c) . G, H, C are random variables with random statistical
charateristics, e.g. mean and variance are random variables. Once C is
sampled, I know P(b|c) and P(a|b) [fixed probabilities]. One goal of
the analysis is to characterize the CDF of P(D| c) .

It seems to me that this should be relatively easy to setup in
Winbugs, but I seem to be making it more complicated than it should
be. Can I treat the parameters P(a | b) and P(b|c) as just weights?
Any suggestions on a WinBugs formulation would be very welcome.

No, they're stochastic nodes: draw the DAG and it should become clear.
Of course, your proability densities have to be ones that BUGS supports
(unless you want to start playing with the ones trick).

It's difficult to give more precise advice, without knowing the details
of the model.

Bob

--
Bob O'Hara
Department of Mathematics and Statistics
P.O. Box 68 (Gustaf Hällströmin katu 2b)
FIN-00014 University of Helsinki
Finland

Telephone: +358-9-191 51479
Mobile: +358 50 599 0540
Fax: +358-9-191 51400
WWW: http://www.RNI.Helsinki.FI/~boh/
Journal of Negative Results - EEB:www.jnr-eeb.org
Thanks Prof O'Hara. After thinking about this a bit more, I think what

On Mar 16, 1:51 pm, "Anon." <bob.oh...@helsinki.fi> wrote:
Jeanwrote:
I realize that this should be probably be an easy question, but I'm
drawing a mental block. Any suggestions on the following problem
would be appreciated.
Problem: I have a large simulation program that, given a set of fixed
input conditions, provides a random response D.
The general problem is structured: P(D|a,g,h) P(g) P(h) P(a | b) P(b|
c) P( c) . G, H, C are random variables with random statistical
charateristics, e.g. mean and variance are random variables. Once C is
sampled, I know P(b|c) and P(a|b) [fixed probabilities]. One goal of
the analysis is to characterize the CDF of P(D| c) .
It seems to me that this should be relatively easy to setup in
Winbugs, but I seem to be making it more complicated than it should
be. Can I treat the parameters P(a | b) and P(b|c) as just weights?
Any suggestions on a WinBugs formulation would be very welcome.
No, they're stochastic nodes: draw the DAG and it should become clear.
Of course, your proability densities have to be ones that BUGS supports
(unless you want to start playing with the ones trick).

It's difficult to give more precise advice, without knowing the details
of the model.

Bob

--
Bob O'Hara
Department of Mathematics and Statistics
P.O. Box 68 (Gustaf Hällströmin katu 2b)
FIN-00014 University of Helsinki
Finland

Telephone: +358-9-191 51479
Mobile: +358 50 599 0540
Fax: +358-9-191 51400
WWW: http://www.RNI.Helsinki.FI/~boh/
Journal of Negative Results - EEB:www.jnr-eeb.org
Thanks Prof O'Hara. After thinking about this a bit more, I think what
I have is a situation involving stratified sampling, probably with
clusters of two different variables. The probabilities are actually
the probability of a sample being from one of the clusters for
variable 1 and again for variable 2. The 5 clusters for variable 1 are
broken down into 10 clusters for variable 2 and a random sample is
taken from each of the 50 resulting cells. Are you aware of any simple
examples of stratified sampling with MCMC? I've gone through the
examples that come with Winbugs.

No I'm not, and it's not too clear to me what the full model is (I don't

Jean wrote:
I realize that this should be probably be an easy question, but I'm
drawing a mental block. Any suggestions on the following problem
would be appreciated.

Problem: I have a large simulation program that, given a set of fixed
input conditions, provides a random response D.

The general problem is structured: P(D|a,g,h) P(g) P(h) P(a | b) P(b|
c) P( c) . G, H, C are random variables with random statistical
charateristics, e.g. mean and variance are random variables. Once C is
sampled, I know P(b|c) and P(a|b) [fixed probabilities]. One goal of
the analysis is to characterize the CDF of P(D| c) .

It seems to me that this should be relatively easy to setup in
Winbugs, but I seem to be making it more complicated than it should
be. Can I treat the parameters P(a | b) and P(b|c) as just weights?
Any suggestions on aWinBugsformulation would be very welcome.

No, they're stochastic nodes: draw the DAG and it should become clear.
Of course, your proability densities have to be ones that BUGS supports
(unless you want to start playing with the ones trick).

It's difficult to give more precise advice, without knowing the details
of the model.

Bob

--
Bob O'Hara
Department of Mathematics and Statistics
P.O. Box 68 (Gustaf Hällströmin katu 2b)
FIN-00014 University of Helsinki
Finland

Telephone: +358-9-191 51479
Mobile: +358 50 599 0540
Fax: +358-9-191 51400
WWW: http://www.RNI.Helsinki.FI/~boh/
Journal of Negative Results - EEB:www.jnr-eeb.org

On Mar 16, 1:51 pm, "Anon." <bob.oh...@helsinki.fi> wrote:
Jean wrote:
I realize that this should be probably be an easy question, but I'm
drawing a mental block. Any suggestions on the following problem
would be appreciated.
Problem: I have a large simulation program that, given a set of fixed
input conditions, provides a random response D.
The general problem is structured: P(D|a,g,h) P(g) P(h) P(a | b) P(b|
c) P( c) . G, H, C are random variables with random statistical
charateristics, e.g. mean and variance are random variables. Once C is
sampled, I know P(b|c) and P(a|b) [fixed probabilities]. One goal of
the analysis is to characterize the CDF of P(D| c) .
It seems to me that this should be relatively easy to setup in
Winbugs, but I seem to be making it more complicated than it should
be. Can I treat the parameters P(a | b) and P(b|c) as just weights?
Any suggestions on aWinBugsformulation would be very welcome.
No, they're stochastic nodes: draw the DAG and it should become clear.
Of course, your proability densities have to be ones that BUGS supports
(unless you want to start playing with the ones trick).

It's difficult to give more precise advice, without knowing the details
of the model.

Bob

--
Bob O'Hara
Department of Mathematics and Statistics
P.O. Box 68 (Gustaf Hällströmin katu 2b)
FIN-00014 University of Helsinki
Finland

Telephone: +358-9-191 51479
Mobile: +358 50 599 0540
Fax: +358-9-191 51400
WWW: http://www.RNI.Helsinki.FI/~boh/
Journal of Negative Results - EEB:www.jnr-eeb.org

Your suggestion that the nodes were stochastic was interesting and,
regardless of my change in formulation of the original problem, I am
curious about your comment.

Simplifying a bit (and hopefully explaining a bit better), define the
problem as:
P(response|action A occurs)P(action A occurs|action B occurs) P(action
B occurs)

The response variable is continuous, and actions A, B either happen or
don't happen. The probabilities of happening are known, e.g. 0.8 and
0.9 . (I'm working on characterizing the P(response|action A occurs)
but that's another part of the problem.)

I'm new at this Winbugs stuff and I guess I don't see how this would
be modeled using WinBugs. I would really appreciate some insight.

I don't get the model: what if actions A or B don't occur?

Jean wrote:
On Mar 16, 1:51 pm, "Anon." <bob.oh...@helsinki.fi> wrote:
Jean wrote:
I realize that this should be probably be an easy question, but I'm
drawing a mental block. Any suggestions on the following problem
would be appreciated.
Problem: I have a large simulation program that, given a set of fixed
input conditions, provides a random response D.
The general problem is structured: P(D|a,g,h) P(g) P(h) P(a | b) P(b|
c) P( c) . G, H, C are random variables with random statistical
charateristics, e.g. mean and variance are random variables. Once C is
sampled, I know P(b|c) and P(a|b) [fixed probabilities]. One goal of
the analysis is to characterize the CDF of P(D| c) .
It seems to me that this should be relatively easy to setup in
Winbugs, but I seem to be making it more complicated than it should
be. Can I treat the parameters P(a | b) and P(b|c) as just weights?
Any suggestions on aWinBugsformulation would be very welcome.
No, they're stochastic nodes: draw the DAG and it should become clear.
Of course, your proability densities have to be ones that BUGS supports
(unless you want to start playing with the ones trick).

It's difficult to give more precise advice, without knowing the details
of the model.

Bob

--
Bob O'Hara
Department of Mathematics and Statistics
P.O. Box 68 (Gustaf Hällströmin katu 2b)
FIN-00014 University of Helsinki
Finland

Telephone: +358-9-191 51479
Mobile: +358 50 599 0540
Fax: +358-9-191 51400
WWW: http://www.RNI.Helsinki.FI/~boh/
Journal of Negative Results - EEB:www.jnr-eeb.org

Your suggestion that the nodes were stochastic was interesting and,
regardless of my change in formulation of the original problem, I am
curious about your comment.

Simplifying a bit (and hopefully explaining a bit better), define the
problem as:
P(response|action A occurs)P(action A occurs|action B occurs) P(action
B occurs)

The response variable is continuous, and actions A, B either happen or
don't happen. The probabilities of happening are known, e.g. 0.8 and
0.9 . (I'm working on characterizing the P(response|action A occurs)
but that's another part of the problem.)

I'm new at thisWinbugsstuff and I guess I don't see how this would
be modeled usingWinBugs. I would really appreciate some insight.

I don't get the model: what if actions A or B don't occur?

Bob

--
Bob O'Hara
Department of Mathematics and Statistics
P.O. Box 68 (Gustaf Hällströmin katu 2b)
FIN-00014 University of Helsinki
Finland

Telephone: +358-9-191 51479
Mobile: +358 50 599 0540
Fax: +358-9-191 51400
WWW: http://www.RNI.Helsinki.FI/~boh/
Journal of Negative Results - EEB:www.jnr-eeb.org

On Mar 23, 2:34 pm, "Anon." <bob.oh...@helsinki.fi> wrote:
Jean wrote:
On Mar 16, 1:51 pm, "Anon." <bob.oh...@helsinki.fi> wrote:
Jean wrote:
I realize that this should be probably be an easy question, but I'm
drawing a mental block. Any suggestions on the following problem
would be appreciated.
Problem: I have a large simulation program that, given a set of fixed
input conditions, provides a random response D.
The general problem is structured: P(D|a,g,h) P(g) P(h) P(a | b) P(b|
c) P( c) . G, H, C are random variables with random statistical
charateristics, e.g. mean and variance are random variables. Once C is
sampled, I know P(b|c) and P(a|b) [fixed probabilities]. One goal of
the analysis is to characterize the CDF of P(D| c) .
It seems to me that this should be relatively easy to setup in
Winbugs, but I seem to be making it more complicated than it should
be. Can I treat the parameters P(a | b) and P(b|c) as just weights?
Any suggestions on aWinBugsformulation would be very welcome.
No, they're stochastic nodes: draw the DAG and it should become clear.
Of course, your proability densities have to be ones that BUGS supports
(unless you want to start playing with the ones trick).
It's difficult to give more precise advice, without knowing the details
of the model.
Bob
--
Bob O'Hara
Department of Mathematics and Statistics
P.O. Box 68 (Gustaf Hällströmin katu 2b)
FIN-00014 University of Helsinki
Finland
Telephone: +358-9-191 51479
Mobile: +358 50 599 0540
Fax: +358-9-191 51400
WWW: http://www.RNI.Helsinki.FI/~boh/
Journal of Negative Results - EEB:www.jnr-eeb.org
Your suggestion that the nodes were stochastic was interesting and,
regardless of my change in formulation of the original problem, I am
curious about your comment.
Simplifying a bit (and hopefully explaining a bit better), define the
problem as:
P(response|action A occurs)P(action A occurs|action B occurs) P(action
B occurs)
The response variable is continuous, and actions A, B either happen or
don't happen. The probabilities of happening are known, e.g. 0.8 and
0.9 . (I'm working on characterizing the P(response|action A occurs)
but that's another part of the problem.)
I'm new at thisWinbugsstuff and I guess I don't see how this would
be modeled usingWinBugs. I would really appreciate some insight.
I don't get the model: what if actions A or B don't occur?

snip

Then there is no response.

In which case, there is no information from that part of the data, so