Skip to content

Quasi-Binomial simulation

2 messages · saber fallahpour, Ben Bolker

#
saber fallahpour <s_fallahpour <at> yahoo.com> writes:
There is no such thing as a quasi-binomial distribution, but if you
parameterized the beta-binomial distribution appropriately I think it
would be straightforward to generate discrete data with a specified
maximum value, a mean that was specified by an inverse-link function
and a design matrix applied to the covariates, and had a variance proportional
(but not equal) to n*p*(1-p).  For comparison, you might want to look
up the "negative binomial type I" as defined by Hardin and Hilde, which
is "quasi-Poisson" in the same sense.

  See

  ?model.matrix
  ?plogis
  ?dbetabinom in the emdbook package (and probably elsewhere: install
the sos package and try findFn("beta-binomial")