[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: Endogenous count variable in a binary model

From   "Leda Inga" <>
Subject   Re: st: Endogenous count variable in a binary model
Date   Tue, 27 May 2008 18:46:20 -0500

Thanks Viktor. I've read qvf help and understand that it allows for
different family distribution and that has support for instrumental
variables. Nevertheless, I still have some doubts:

1. In previous emails
(, it has
been said that qvf cannot be used for endogeneity if running a
probit/logit models (which is my case). Is this fact also true for
endogeneity caused by omitted variables instead of simultaneity??

2. I've read qvf help and the paper wrote about it
( and
havenīt found what estimation method is used for the
instrumentalization. I think that it is OLS since the endogenous
variables of the examples provided in both mentioned sources are
continuous. Is this is the case, I wouldn't be using an appropiate
model for my endogenous variable, since it is count data. Is there a
way of managing with an endogenous (by omitted variables) count data?

3. What would be the procedure if the endogenous count variable also
enters the model with a square term?



2008/5/27 Viktor Slavtchev <>:
> perhaps you can use -qvf-
> search qvf
> findit qvf
> hth
> viktor
> Leda Inga wrote:
>> Hi,
>> I'm runing a binary regression with survey data and I'm interested in
>> the effect of a count explanatory variable (called CP) which might
>> have a diminishing impact. Theoretically both variables, the dependent
>> and the explanatory variable of my interest, could be determined by
>> some others factors which can't be controlled and so go in the error
>> terms.  My objectives are: 1) Test if CP has significant effect on the
>> probability of ocurrence of the event I'm studying, 2) If so, test if
>> the effect is diminishing, based on the significance of a cuadractic
>> term (CP^2), and finally 3) Know at which point the effect of CP
>> reaches it peak.
>> Since ivprobit doesn't allow for a cuadractic term and is for
>> continuos data, I didn`t use it. Instead I ran a count data model,
>> saved the predicted values and generated a new variable equal to the
>> square of these. Then I ran a binary regression including both, the
>> predicted values of CP and the square of them. But I'm not sure if
>> this estimation procedure is correct and if I'm really getting
>> consistent betas and standard errors.
>> Just to give some details, CP takes values from 0 to 20, has a mean
>> 7.48 and a variance of 12.8.
>> I would really appreciate any help,
>> Leda
>> *
>> *   For searches and help try:
>> *
>> *
>> *
> *
> *   For searches and help try:
> *
> *
> *

*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index