Should `var(B::LogBinner)` give naive (uncorrelated) variance?
carstenbauer opened this issue · comments
Maybe it's a bit strange that var(B) != var(data)
.
Isn't that what var(B, 1)
gives?
Yes it is. The question is whether var(B) = var(B, 1)
would be a better default than var(B) = var(B, _reliable_lvl(B))
.
I think it should be consistent with varN
, std_error
and tau
. As we discussed in #4
Yeah. Let's leave it the way it is.