Dear Clarify users,
I have a question on a matter I have already touched upon in a previous message (October
2011). I didn't receive any answer so I am asking again trying to be more clear.
My point is about the calculation of expected values from a linear regression of a logged
variable. -Simqi- with the option -tfunc(exp)- should yeald results transformed back to
the original (unlogged) scale. Stata's official command -adjust- with the option
-exp- should do the same. However the two do not coincide. What Stata appears to do is the
simple exponentiation of the expected value: exp[E(lny)]. This yealds the expected
geometric mean of the unlogged variable (geometric mean is always less than arithmetic
mean and more similar to the median for skewed variables). -Simqi, tfunc(exp)- yealds
results flagged E[exp(lny)]. Actually the latter always display higher values than the
former, as if they were more similar to expected arithmetic mean. So my question is: how
does -simqi- exactly compute expected values when I ask for the -tfunc(exp)- option?
Many thanks
Renzo Carriero
Show replies by date