This the approach I was considering. I was just concerned that I would
end up with a poor fit unless I run a large set of data.

I will give it a try and see what the results are

Thanks for the help


In cbug2@yahoogroups.com, "jamesamurtha" <jmurtha@...> wrote:
>
> Hi, Jimmy,
> Try this: Run your model, including the correlations using 20,000 or
> more iterations. Extract trial values for F, X, Y, Z. Sort the data on
> X. Locate a subset of the data whose X-values are, say, + or – 1%
> from the desired X-value. Sort the reduced data set on Y and locate a
> subset of this data whose Y-values are "close" to the desired value of
> Y. Plot F against Z.
> You can see from this experiment that you will not have any values in
> the database where the X and Y values are exactly what you specify.
> Depending on your distributions and the functional relationship, you
> might get close enough for your purposes. I tried the experiment with
> a couple of functions and got very different results. Had X and Y been
> discrete variables, you might have found some exact matches.
> Regards,
> Jim
>
> --- In cbug2@yahoogroups.com, "asujames" <asujames@> wrote:
> >
> > The distribution are continious gamma functions.
> >
> > I do not get what you mean by
> >
> > "If X and Y are continuous, then the probability of their assuming
> > particular fixed values is of course 0"
> >
> > How does that help me ... I'm sure it does I just do see it
> >
> > Thanks
> >
> >
> >
> >
> > --- In cbug2@yahoogroups.com, "jamesamurtha" <jmurtha@> wrote:
> > >
> > > Hi, Jimmy,
> > > If X and Y are continuous, then the probability of their assuming
> > > particular fixed values is of course 0.
> > > If X and Y are discrete, then you can run the simulation with
lots of
> > > iterations (depending on the number of possible combinations of
values
> > > of X and Y), extract the simulation data, and sort on X then Y.
Simple
> > > counting will allow you to calculate the desired conditional
> > probability.
> > > Regards,
> > > Jim
> > >
> > > --- In cbug2@yahoogroups.com, "asujames" <asujames@> wrote:
> > > >
> > > > Hello all,
> > > >
> > > > I'm trying to model conditional proabailities in CB. Quick
> > > > desscribtion of my problem.
> > > >
> > > > I have three distributions as inputs (X, Y, Z) and I run my MC
> to get
> > > > output F. There is a correlation between X and Z and Y and Z. I
> want
> > > > to be able to specify X and Y and run my simulation, but at the
> same
> > > > time I want the correleation between the distributiuons to
remain.
> > > > Because of this I can't just set X and Y to a value and run
because
> > > > CB defines the correlations to assumptions.
> > > >
> > > > What I have done. I tried to use the 2D simulation tool to
perform
> > > > thsi task, which seemed liek it shodul work because it will set
> X and
> > > > Y and then run the MC for Z and forcast my output F. The
porbelm is
> > > > the corrleatiosn do not remain between X and Z or Y and Z. So
I can
> > > > get a conditional probability for F but it doesn't include a
> > > > crrelation ebwteen my inputs.
> > > >
> > > > The goals is to determin the 99th confidence for F given a X
and Y,
> > > > and running MC all all of my other inputs in thsi case just Z and
> > > > keeping the correlations between my inputs.
> > > >
> > > > Has anyone tried to answer a simular type of problem and can
> help or
> > > > maybe you see a solution that I'm missing
> > > >
> > > > Thanks
> > > >
> > > > jimmy
> > > >
> > >
> >
>