R help - Jun 2010 - Fw: simulating binary random numbers using 'mvtnorm'

Hi R-users,

I just started learning R. I have a project on lot quality assurance sampling
(LQAS). In this project I have to develop LQAS plans to make decision on
stopping / continuing a programme. The LQAS plans is based on cluster
sampling: selection of k clusters (villages) of m children each for a total
sample of n. In this connection I would like to estimate the classification
errors ( alpha and beta) that would be expected as a consequence of sampling
clusters (villages / households) instead of usual SRS under traditional LQAS. In
order to estimate the classification errors I would like to simulate correlated
binary outcomes for a given level of prevalence and intracluster correlation
(rho).

Specifically I would like simulate an LQAS plan (33 clusters of 6 children
each, n =198 children) to generate clusters with specific intercluster (0.05)
and intracluster correlation (0.1) subject to the constraint that
the intercluster correlation is less than or equal to intracluster correlation
for a given level of prevalence (upper threshold =1% and lower threshold=0.1%).
For this purpose I have used the following command (which is under
mvtBinaryEP in mvtnorm library in R):

ep0 = ep(mu=0.1, rho=0.05, n=6, nRep=33, seed=NULL)
apply(ep0$y, c(1,2), mean); cor(ep0$y)

The above commands produce 198 binary outcomes (33 clusters with cluster size of
6 each) and a matrix (6x6) of bivariate correlations. Since the input among
others requires values either for 'rho' along with 'mu' or an
'R' matrix and a column vector of 'mu', I wonder whether the
binary outcomes meets my specification about inter-cluster correlation. I
would be very much grateful to R-community, if someone helps me out.

thanking you in advance and I am looking forward to R-community

subramanian
=========Dr S.Subramanian
Assistant Director
Vector Control Research Centre
Indira Nagar
Pondicherry - 605 006
INDIA 


      
	[[alternative HTML version deleted]]