How to draw random multibinomial samples with specified correlation?

Question

Suppose I have three variables with a correlation matrix like this:

    y1  y2  y3
y1 1.0 0.2 0.3
y2 0.2 1.0 0.4
y3 0.3 0.4 1.0

How could I draw a set of n random bernoulli samples of y1, y2, y3 under this specified correlation parameters (maybe also with specified probability for each of them)? such like:

No, they are different. This one is with specified correlation and that one is just for not iid. — David Z, Jul 10 '14 at 20:20
I've created an answer using code directly from the linked question that solves your issue. — josliber, Jul 10 '14 at 20:31

score 1 · Accepted Answer · edited May 23 '17 at 11:52

1

As in this post, you could use the rmvbin function:

# Correlation and marginal probabilities
cor.mat <- matrix(c(1, .2, .3, .2, 1, .4, .3, .4, 1), nrow=3)
marg.prob <- margprob <- c(.1, .2, .3)

# Generate random data
library(bindata)
set.seed(144)
res <- rmvbin(10000, margprob, bincorr=cor.mat)

# Check things are working properly
cor(res)
#           [,1]      [,2]      [,3]
# [1,] 1.0000000 0.1891508 0.2965542
# [2,] 0.1891508 1.0000000 0.3927966
# [3,] 0.2965542 0.3927966 1.0000000
colMeans(res)
# [1] 0.1013 0.1980 0.3029

edited May 23 '17 at 11:52

Community

1
1

answered Jul 10 '14 at 20:31

josliber

43,891
12
98
133

What is the exact equivalent of this in python ? Anyone ... ? – user1301295 Jun 16 '16 at 18:58
@user1301295 If you have a new question you should use the "Ask Question" button instead of asking in the comments. – josliber Jun 16 '16 at 20:04

How to draw random multibinomial samples with specified correlation?

1 Answers1