Quantcast

joint effect of two endogenous variables

classic Classic list List threaded Threaded
6 messages Options
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

joint effect of two endogenous variables

xueliansharon
Dear all,

I want to estimate the following model:

ivreg2 y z1 z2 (y1 y2= z3 z4 z5), cluster(mm)

However, my instruments z3, z4 and z5 don't have enough independent variations (i.e.we don't have instruments that are correlated to y1 but not correlated to y2, or just correlated to y2 but not correlated to y1, z3, z4 and z5 all affect both y1 and y2) and thus the effects of y1 and y2 on the dependent variable y can't be isolated, so I want to compute and test the significance of the joint effect of y1 and y2, i.e. the effect on dependent variable when both y1 and y2 increase by 1 unit. Does anybody know how to realize this idea?


Many thanks,
Sharon
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: joint effect of two endogenous variables

Maarten buis
--- On Tue, 24/8/10, xueliansharon wrote:

> I want to estimate the following model:
>
> ivreg2 y z1 z2 (y1 y2= z3 z4 z5), cluster(mm)
>
> However, my instruments z3, z4 and z5 don't have enough
> independent variations (i.e.we don't have instruments that are
> correlated to y1 but not correlated to y2, or just correlated
> to y2 but not correlated to y1, z3, z4 and z5 all affect both
> y1 and y2) and thus the effects of y1 and y2 on the dependent
> variable y can't be isolated, so I want to compute and test the
> significance of the joint effect of y1 and y2, i.e. the
> effect on dependent variable when both y1 and y2 increase by 1
> unit. Does anybody know how to realize this idea?

The first thing that comes to mind is that you will need to make
sure that the unit of y1 and y2 are equal, if y1 is in seconds and
y2 is in liters, than what does a unit change mean? A common
approach is to standardize variables, i.e. subtract the mean and
divide by the standard deviation.

What you could do is constrain the effects to be equal. A quick
scan of -help ivreg2- gave me the impression that it doesn't
allow for the -constraint()- option. You might be able to use
an old trick: you can constrain the effects of two variables to
be equal by adding the sum of these two variables to your model.
You can see that as follows. Start with a regular regression:

y = b0 + b1 x1 + b2 x2 + e

we want to constrain b1 and b2 to be equal, so we can write:

y = b0 + b1 x1 + b1 x2 + e
  = b0 + b1 (x1 + x2) + e

So application of this trick to your model would been that
you generate a new variable y_comb = y1 + y2, and use that
variable instead of y1 and y2.

However, I don't know much about -ivreg2-, so other people
who know more about it, will need to confirm that this trick
will have the desired properties before I would recommend
it.

Hope this helps,
Maarten

--------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany

http://www.maartenbuis.nl
--------------------------




     

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: stratification and model checking after imputation

Kimberly Forde
Dear Statalisters,

I am conducting an analysis in a large database and not surprisingly there are a number of missing variables.  Body mass index is important to my study and unfortuantely is missing in about 20% of records.  I have chosen to impute this data using the mi commands (STATA 11). After imputing this data, I have encountered 2 distinct problems:

1. I am having difficulty running stratified analyses because "by" statements cannot be used with the mi command.  I was thinking that perhaps I could use an interaction term and then use a lincom command to obtain the actual estimates thereafter.  Are there other ways to circumvent the limitations of the mi commands with stratified analyses?

2. Is there any way to do model checking for survival analyses after the data has been imputed using "mi"?  In the complete case analyses, I was about to compare AICs and also use visual inspection of log-log plots to determine if proportional hazards assumptions were being met.  I have not been able to do so after imputation.  Does anyone know of a way to get around this issue? Would you suggest using the best model obtained from the complete case analysis and then using the same covariates to construct a final model after imputation?

Any assistance you may have to offer would be greatly appreciated!!!!!

Thanks,
Kim  


     

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: joint effect of two endogenous variables

xueliansharon
In reply to this post by Maarten buis
>The first thing that comes to mind is that you will need to make
>sure that the unit of y1 and y2 are equal, if y1 is in seconds and
>y2 is in liters, than what does a unit change mean? A common
>approach is to standardize variables, i.e. subtract the mean and
>divide by the standard deviation.

Maarten, my y1 is people's age started to work, y2 is years of schooling, what I want to get is the joint effect when people started to enter the labor market one year later and receive one more year of schooling.

But I didn't understand why "constraining the effects to be equal" could compute the joint effect, could you explain it more explicitly?

Thank you in advance.
Sharon
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: joint effect of two endogenous variables

Maarten buis
--- On Tue, 24/8/10, xueliansharon wrote:
> my y1 is people's age started to work, y2 is years of schooling,
> what I want to get is the joint effect when people started to
> enter the labor market one year later and receive one more year of
> schooling.

It sounds like the unit of both variables is years, but they are
obviously still different, a year of education is not the same
thing as getting a year older, so you'll still need to standardize.
 
> But I didn't understand why "constraining the effects to be
> equal" could compute the joint effect, could you explain it more
> explicitly?

You'll need to conceptually define what a "joint effect" is, and
you could see constraining the effects to be equal as one such
definition.

By proposing one such definition, I side steped the most important
part: thinking about what it is that you want to measure. So
instead of focussing on technique, I would first try to find a
reason why one might want to know the effect of the combination of
these two variables. If you can come up with such a story, then
you can often derive the type of constraint that you need to apply.
If you cannot come up with such a story, then estimating a "joint
effect" just doesn't make sense, and you'll need to find another
solution.

Hope this helps,
Maarten

--------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany

http://www.maartenbuis.nl
--------------------------


     

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: joint effect of two endogenous variables

xueliansharon
OK, thanks, Maarten, I will try your method.


Best,
Sharon

Loading...