Bayesian modeling and model–based decision analysis 193
much simpler, is equivalent to the original BB, in view of the construction of
the Dirichlet distribution from independent exponential variables. The BB
metho d is easy to implement as long as fitting the model with weighted data
is technically easy. This includes GLM models, line ar mixed models, GLMM
fitted with GEE approaches, and nonlinear models with nonsparse data.
There is an obvious link between BB and the standard bootstrap (SB),
which is equivalent to using integer (including 0) weights to fit the model in
the same way as BB. Therefore, BB can also be considered as a smoothed
version of SB. One technical advantage of BB is that it is also more stable
than SB, a s it ensures each pair having ...