Two limit tobit

sahajperon · science forum beginner Joined: 20 May 2006 Posts: 1

Hi, I am trying to derive analitically the marginal effects of a change
in the upperbound limit of a doubly limited censored tobit (data
censored at 0 and b) on the expected dependent variable. This aims at
explorin the sign of variations of founding limits of EU grants on
expected grants)
The problem is partially addressed in Maddala (1983), and so far I get
to the full expression of the expected yi )(under the assumption of
normal error ter1
E(yi|xi)=xiß[F((b- xi ß)/s)-F((-xi ß)/s))]+s[-f((b- xi
ß)/s)+f((-xi ß)/s))]+b(1-F((b- xi ß)/s))
where F is the cumul Norm
f normal density
Now, when I take the derivative with respect to b, I get:
dE(yi|xi)/db=xiß/sf((b- xi ß)/s)+b/sf((b- xi ß)/s)+1-F((b- xi
ß)/s)-b/sf((b- xi ß)/s)=1-F((b- xi ß)/s)+xiß/sf((b- xi ß)/s)
by the way, the most commonly reported effects in the literature report
dE(yi|xi)/db=1-F((b- xi ß)/s)
I don't see why the term
xiß/sf((b- xi ß)/s) has to be null.

Can anyone see or has anyone good suggestions on references/ books or
papers where doubly censored models are analysed?
Thanks in advance
SaraJ

Google