The package currently supports “case weights” - weights on individual observations. I wonder if there is a way to tweak the code so that I can put weights on clusters?
Specifically, the package currently can take a vector of observation weights
w , and calculate the total log-likelihood as
\sum log(w_ij * P(y_ij)) , where
y_ij are the weight and response for observation
j from cluster
However, instead of weighting the probabilities on the observation level, I want to weight the log-likelihood on the cluster level, i.e., I want to calculate
\sum w_i l_i ? Here
l_i are the weight and the log-likelihood for cluster
Can anyone give a pointer on where I should get started? Thank you.