I am dealing with a bivariate sample (~1000 elements, variables $A,B$) whose marginal distributions are known.

Each marginal distribution is assumed to be drawn from a semicontinuous distribution with positive mass at 0 and a two-parameter gamma for strictly positive values.

What is the simplest (in terms of no further assumptions) way to produce samples of $A,B$ which follow the explained distributions in a way they reproduce the observed spearman correlation?

Ideally, I would like to draw samples from A and then obtain samples of $B|A$ so that the correlation is kept.

  • 1
    $\begingroup$ The Spearman correlation doesn't care what the marginals are. A simpler formulation of this problem, then, supposes the marginals are Uniform$(0,1)$ variables. Now see stats.stackexchange.com/questions/168566. $\endgroup$
    – whuber
    Commented Jun 28, 2021 at 17:13


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.