Testing for a Valid Instrumental Variable

[Hellie Flint]

Hi Statalisters,

I am in the process of looking for a suitable instrument for gender inequality, as I believe it is endogenous when looking at its effect on economic growth. I have been using % of women in parliament as an instrument but what assumptions must it satisfy for me to know it's an appropriate instrument? I ran the following but I'm not sure how to interpret it:

reg lagGII lagWIP, robust

Linear regression                               Number of obs     =      1,974
                                                F(1, 1972)        =     556.04
                                                Prob > F          =     0.0000
                                                R-squared         =     0.2684
                                                Root MSE          =     .17316

------------------------------------------------------------------------------
             |               Robust
      lagGII |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      lagWIP |  -.0095053   .0004031   -23.58   0.000    -.0102959   -.0087148
       _cons |   .5679747   .0070715    80.32   0.000     .5541063    .5818431
------------------------------------------------------------------------------

. test lagWIP

 ( 1)  lagWIP = 0

       F(  1,  1972) =  556.04
            Prob > F =    0.0000

as the lagWIP is significant in the GII regression and the consequent F-Test value is greater than 10, does this indicate the instrument is appropriate? Then how do I also test that women in parliament is not a determinant of my dependent variable, economic growth?

Many thanks

[Phil Bromiley]

You should look at ivreg2 which has many tools for such problems. Whether this is a high enough r-square to avoid the weak instruments problem is not clear. By the way, you can just use L.WIP instead of lagging things previously.

[Hellie Flint]

Thanks Phil!