"anja.ende@googlemail.com" <anja.ende@googlemail.com> wrote in message <63344cb9ead94235bb13de80516e0c66@q11g2000vbu.googlegroups.com>...
> Hello all,
>
> This is not a matlab question per se, but I was hoping someone here
> might have a good idea.
>
> I have a joint entropy like expression as follows:
>
> F(BA) = integral[inf +inf] P(B>bA) log P(B>bA)
>
> To implement this, I have a 2D array where the elements represent this
> joint PDF (I compute the conditional survival function and its log
> basically).
>
> Now, what would be the expected value of this expression i.e. E(F(B
> A)). Would it be simply scaling over the sum of each of the rows and
> columns? I am somehow getting very confused with this.
>
> Thanks for any help you can give me.
>
> Anja
         
I don't think the notion of expected value applies to your F(BA) quantity. B and A are stochastic but probabilities involving them are not. They are definite numerical quantities. If I flip a penny it is legitimate to ask for the expected number of heads, namely 1/2, but it is not legitimate to ask for the expected probability of throwing a head. That is not a quantity subject to statistical variation.
Roger Stafford
