Dear all,
I think I solved my own previous posting.
If the problem of using 1-binormal(a,b;rho) as an approximation for
1-Pr(x>a,y>b;rho) is created by cancellation error (when
binormal(a,b;rho) is close to 1), then the former quantity can be
approximated by
binormal(-a,b,-rho) + binormal(a,-b,-rho) + binormal(-a,-b,rho)
Since this only involves additions of small terms it shouldn't face
the same accuracy problems.
robert
On 2/5/07, Robert Duval <[email protected]> wrote:
Sorry there was a typo in my previous posting... here is the correct one
On 2/5/07, Robert Duval <[email protected]> wrote:
> Dear friends
>
> In the ml book (Gould, Pitblado and Sribney) it is mentioned that
>
> 1-normal(a) is a poor numerical approximation for Pr(x>a) and instead
> they suggest using
>
> normal(-a).
>
> Does anyone knows of a similar expression for the bivariate normal
> case giving 1-Pr(x>a,y>b;rho) without using
>
> 1-binormal(a,b;rho)
>
> i.e. without using substractions (or additions) that make the
> approximation numerically poor.?
>
> Any help would be greatly appreciated.
>
> Robert
> *
>
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
