Closed danielwilhelm closed 11 months ago
So I guess, that all we have to do is negate the regressor to be ranked, i.e.
lmranks(r(y, increasing=FALSE)~...)
is the same as lmranks(r(-y, increasing=TRUE)~...)
.
I will prepare an end2end test with indicators reversed, implement increasing=FALSE by negation and check whether results are the same.
This issue has been addressed on the vignette branch, correct? If so, we can close this one ...
Exactly.
Implement lmranks with increasing=FALSE:
the formulas for the asymptotic variance stay the same as for increasing=TRUE, except that in the indicators I(...), inequalities are reversed, so <= becomes >= and < becomes > everywhere.