I noticed a difference in BIC between SEM.jl and lavaan. I traced this back to a data set, in which there were some rows that had NA on all observed variables. It seems that lavaan ignores those and thus has a different count for nsamples(sem_fit) when computing BIC.
It could be a good idea if SEMObservedMissing filtered out all cases that are all NA because those do not contribute to any (mis)fit but distort fit indices.
I noticed a difference in BIC between SEM.jl and lavaan. I traced this back to a data set, in which there were some rows that had NA on all observed variables. It seems that lavaan ignores those and thus has a different count for
nsamples(sem_fit)when computing BIC.It could be a good idea if SEMObservedMissing filtered out all cases that are all NA because those do not contribute to any (mis)fit but distort fit indices.