Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: IVQR with two endogenous regressors using -ivqreg-

From   Nick Cox <>
To   "" <>
Subject   Re: st: IVQR with two endogenous regressors using -ivqreg-
Date   Fri, 31 May 2013 17:57:54 +0100

Thanks for the details.  I note there is no help file.

On your question (2)

qui regress `endo' `allinstr' if `touse'
predict dhat, xb

-dhat- is a predicted response after the previous regression.

qui qreg `lhs' `dhat' `inexog' if `touse', q(`quantil')
predict iner if `touse', resid

-iner- ditto. They should be dropped later in the program, so should
be temporary variables.

Presumably the program crashed for you before that.


On 31 May 2013 17:42, Katie Farrin <> wrote:
> Thank you, Nick,
> I posted earlier regarding this program and have contacted the author,
> but have not heard back.  The program is accessible not through Stata
> but through an online search (I am not sure why this is the case,
> either).
> The (unpublished) Stata Journal article written by Do Won Kwak is
> where the author describes the program; I downloaded the program from
> this professor's page (requires some scrolling but it's through a link
> in the fourth option for "IVQR"):
> Reference: Kwak, D. (2010). Implementation of instrumental variable
> quantile regression (IVQR) methods.  Accessible through Kwak's job
> market page
> I would prefer to use an IVQR method in my work as I have evidence of
> heterogeneity of effects for my regressors of interest; however, I
> also have an endogeneity problem.  This user-written code is the only
> thing I've found so far that I can work with to address such a model.
> On another note, I guess I was also hoping that the conformability
> error might be common to instrumental variables methods.  I have read
> elsewhere (Statalist archives) that it may be due to a variance
> covariance matrix operation problem; I have income variables that I
> have scaled down (from dollars to thousands of dollars) and several
> indicator variables, so it may be an issue of different scale.
> However, I'm confused as to why the scale wouldn't be a problem in
> previous regressions (using -regress- and -bsqreg-) that I have
> performed using the same set of regressors.
> Any help is much appreciated.
> Best,
> Katie
> On Fri, May 31, 2013 at 12:24 PM, Nick Cox <> wrote:
>> You are in strange territory here.
>> First, it is hard for anyone else even to look at this code as
>> -findit- can't find it and you don't give a location. Longstanding
>> Statalist practice is to explain where user-written software can be
>> found, as the FAQ does spell out.  You do clearly state that you only
>> hope for attention from people familiar with the program, but that's
>> our convention. (Sometimes people not familiar with a program can look
>> inside quickly and say something helpful.)
>> Second, it sounds as if -ivqreg- creates permanent variables without
>> reference to what you asked. Is that documented in the help? It's not
>> going to help you at all, but I'll opine that's poor programming
>> practice if undocumented, and a sign that you have (what shall we say)
>> a program that is work in progress.
>> Third, if this program is documented then there should be an author
>> email for support. I have not been following closely but there have
>> been various questions about this program and I've not noticed an
>> author replying. If that's so, the clear implication is that the
>> author is not on Statalist and it is a good idea to contact him, her
>> or them directly.
>> Nick
>> On 31 May 2013 17:08, Katie Farrin <> wrote:
>>> Hi, Statalisters,
>>> I am using the -ivqreg- user-written command to evaluate the effects
>>> of source-specific income increases on nutritional outcomes.  However,
>>> I have two endogenous regressors, a dummy that indicates that a
>>> household has chosen to farm tobacco and a (continuous) variable for
>>> tobacco profits; these endogenous regressors are clearly related but
>>> not the same.
>>> I would like to instrument the endogenous dummy variable with GIS data
>>> on the distance to the nearest tobacco auction floor and the tobacco
>>> profits with the deviation from ten-year average rainfall.   I use the
>>> following code in Stata:
>>> *instumental variables quantile regression
>>> ivqreg ln_dietarydiversity maize_thousands emp_inc_thousands
>>> bus_profits_thousands livestock_thousands remittances_thousands /*
>>> */ ln_price_tomato age_hh_head male married durables_thousands
>>> years_education hh_size /*
>>> */ chitipa karonga nkhatabay rumphi mzimba kasungu nkhotakota ntchisi
>>> dowa salima mchinji dedza ntcheu /*
>>> */ mangochi machinga zomba chiradzulu blanytyre mwanza thyolo mulanje
>>> phalombe chikwawa nsanje balaka /*
>>> */ March April May June August September October November
>>> (tobacco_thousands tobacco=deviation_tot_rainfall_wet_2010
>>> dist_auction), robust first
>>> I have two questions:
>>> (1) when running this code I get a conformability error (r(503)), and
>>> I do not know why this is occurring (this is very similar to the code
>>> I run for regular QR using -bsqreg- without any problems) - any advice
>>> would be greatly appreciated on how to run this code without
>>> generating such an error;
>>> (2) when I run the code, before the error message, Stata saves two new
>>> variables to my working data set, "dhat" and "iner."  I have looked at
>>> the user-written -ivqreg- but cannot figure out what these are (I am
>>> assuming they are predicted values, but of what I do not know because
>>> they are not similar in range to my endogenous variable values) - if
>>> anyone is familiar with -ivqreg- and could give me some insight as to
>>> what these variables are I'd appreciate it.
>> *
>> *   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   |   Site index