Jialu Zhang <[EMAIL PROTECTED]> wrote in message news:<[EMAIL PROTECTED]>...
> Hello,
> 
> I have a question on how to generate random binomial samples with known
> variance-covariance structure. For example, y1, y2, ..., y10 all follow
> binomial distribution Bin(n,p). n and p are given. Also some observations
> are correlated. corr(yi,yi+1)=0.1, i=1,2,...9.
> 
> I wonder if someone knows how to generate these random numbers.
> 
> Many thanks in advance.
> 
> Jialu

Let X1,X2,X3,X4,Y be independent and identical random variables from
any distribution which has a mean m and a variance v. Now let Z1 be Y
+ X2 and Z1 be Y + X4 with probability p, else let Z1 be X1 + x2 and
Z2 be X3 + X4.

Then E(Z1.Z1)-E(Z1).E(Z2)
= p.E(Y^2+Y(x2+x4)+X2.X4) + (1-p).E((X1+X2).(X3+X4)) - 4.m^2
= p.[v+m^2+2m^2+m^2]+(1-p).4.m^2-4.m^2
= p.v

and the correlation is p/2.

So 1 possibility for n even is as follows.
Generate a uniform rv U. If U < 0.2 then let
Z1 = X1+X2, Z2 = X2+X3, ...Z9 = X9+X10, Z10 = X10+X11 where X1..X11
are generated as B(n/2,p) rvs else generate Z1..Z10 as B(n,p) rvs.

Ian Smith
.
.
=================================================================
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