Linda,

I assume you have 1000 triples (Power, Delay,Angle).
What to do may depend on what you feel you can assume on the basis of
knowledge of the subject.

At the non-parametric extreme, for each pair (Delay,Angle) and each p you
have the proportion of triples with Power <=p. That is the empiric cdf of
Power given  (Delay,Angle).

Presumably, you would like something a little more useful in the sense that
it can be used to create predictions. If you have a physical basis to assume
that,
there are functions f, g and h such that f( Power) is a normal variable with
mean a linear combination of g(Delay) and h(Angle) with variance independent
of Delay and Angle, you could do a regression
f(Power) = b0+b1*f(Delay)+b2*h(Angle).

You might want to plot some graphs holding Delay(Angle) constant to see how
Power varies and get ideas for appropriate functions. Also, check if there
are engineering or physics  texts on the subject already that give the
distribution of Power given  Delay and Angle  for data similar to your
sample

Good Luck.
Ellen Hertz

"Linda" <[EMAIL PROTECTED]> wrote in message
news:[EMAIL PROTECTED]...
> Helloo..
>
> I am having some problems in understanding this and I hope someone out
> there are willing to explain it to me.
>
> I have 3 RVs (Power,P Delay,t and Angle,a) each with 1000 samples
> estimated from real time measurements. I would like to find PDF of the
> Power given their corresponding Delay & Angle i.e. f(P|t,a). How do I
> do that practically?
>
> Thanks.
>
> Linda


.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to