Request on how to use movestay command without instrument in the selection equation

[Solomon Wassie]

Dears,

I want to estimate the impact of membership in collective action on income using the movestay command of Stata to estimate Endogenous Switching Regression model. But I don't have valid Instrument to be used in the selection model. And I want the model to be identified through non-linearities.

Hence, I would be very happy if you could help me on the command (I have tried to follow the help option of Stata "movestay (y1= x1 x2 x3 x4) (y1= x1 x2 x3 x5), select(regime1)" but I don't exactly understand two points:

1. Is it a must to have two specifications of dependent variables and is it a must to have different explanatory variables for the first two specifications - i.e.

y1= x1 x2 x3 x4) (y1= x1 x2 x3 x5)

?

2. What should I use in the regim1, I was simply using my dummy variable of treatment group (membership status) but it says "too few variables are specified"

Last but not least, I really will appreciate if you could tell me

(or send me documents)

about the difference, if any, between identifying using the instrument and through nonlinearities only (in case of no valid instrument).


Waiting for your help.
Due regards

[FernandoRios]

Hi Solomon,
First of all, i would recommend you to go through the paper that introduces the user written program -movestay-. http://www.stata-journal.com/sjpdf.h...iclenum=st0071
If you follow his example, you will find that similar to the heckman command, you need to specify the model for the first regime selection, even if that contains no instrument.
for example:

movestay (y1= x1 x2 x3 x4) (y1= x1 x2 x3 x5), select(regime1=x1 x2 x3 x4 x5)

Now, move stay is no more than a heckman combine with two outcome regressions, so there is nothing to say (as far as i know) that all variables in the outcome have to be in the decision step of the model, but for most of i have worked, and seen, that is what is done.

HTH
Fernando

[Solomon Wassie]

Dear Fernando,

Thank you very much for your help. I really appreciate.

Sorry but I have one more question. I have one only one continuous outcome variable (y1) so should the first two specifications have different explanatory variables - i.e.

(y1= x1 x2 x3 x4) (y1= x1 x2 x3 x5)

In other words is it possible to estimate for example command,

movestay (y1= x1 x2 x3 x4) (y1= x1 x2 x3 x4), select(regime1=x1 x2 x3 x4)?

Thank you very much in advance.

[FernandoRios]

Hi again,
I dont think there is anything against having the same specification for both models in the regime. I would even say that is my preference when this type of model is being estimated.
Just keep in mind, just as when estimating heckman selection models, you can identify the model based on the non-linearity, but it's hard to justify
Fernando

[Solomon Wassie]

Dear Fernando,

Thank you very much.

Due regards

[Solomon Wassie]

Dears,

Can I use movestay command for a censored continuous outcome? If now what is the alternative command I need to use for running censured endogenous regression model?

Thank you very much

Solomon