weakivtest

[ALKEBSEE RADWAN]

Hello everyone
I know that there exist a lot of posts on the weak instrument test and how to interpret the results.
but actually i am still not able to recognize whether my instrument is strong or weak from my results


here is the results that i got from applying the command weakivtest

Montiel-Pflueger robust weak instrument test
--------------------------------------------
Effective F statistic:    2,515.878
Confidence level alpha:          5%
--------------------------------------------

--------------------------------------------
Critical Values             TSLS      LIML
--------------------------------------------
% of Worst Case Bias
tau=5%                    37.418    37.418
tau=10%                   23.109    23.109
tau=20%                   15.062    15.062
tau=30%                   12.039    12.039
--------------------------------------------

please tell me whether my instrument is strong or not ?
and how can i explain such results

@Joro Kolev

[Joro Kolev]

From the table we see that your instruments are very strong.

How you read the table:

The test tests the null hypothesis that your instruments are weak, therefore you would wish to reject it.

To reject it, the calculated statistic, which in your case is 2,515.878, should be larger than the critical value.

The critical value depends on whether you use TSLS or LIML, and depends on what bias are you willing to tolerate. The bias is given as a percentage of the worst case scenario bias, which for homoskedastic model is the OLS bias. As you can see, the critical values are lower (and hence it is easier to reject weak instrument null) if you are willing to tolerate larger bias.

[ALKEBSEE RADWAN]

Thank you so much For the great clarification.
once again I am grateful to you

[Mauricio Carvalho]

Dear Prof Kolev Joro Kolev

Considering this case of a very high Effective-F such as 2,515.878. It is enough to say our instrument is strong (not weak), regardless of the size of the coefficient from the first-stage?

Thank you !

[Joro Kolev]

Hi Mauricio Carvalho , these are two separate issues: one is whether the test comes out as you want it to come out, and two whether the instrument is doing what you are supposing that it should be doing.
If the instrument has the correct sign in the first stage (that is, the sign that you thought it should have based on prior knowledge), you might want to report that. In effect you would be saying "My instrument passes the test, and has the sign predicted by theory or prior reasoning."

[Mauricio Carvalho]

Thank you very much, Professor!

Just to clarify.

So, in a context of high Effective-F even if the size of the estimated coefficient from the first-stage is too small (assume very close to zero) but significant and with the "right" sign, we are good? I mean: ""My instrument passes the test (e.g, High effective-F), and has the sign predicted by theory, despite a low coefficient from the first-stage" ?

[Joro Kolev]

The size of a regression estimate can be only economically interpreted as small or big in the context of units of measurement of the dependent and independent variable. The estimate tells you by how many units the dependent variable will change, when the independent changes by one unit.

Assuming that you know how to do what I said in the previous paragraph, you could say what you want to say. But my guess is that you might be wrong, and the estimate is not as small as you think. I guess that because statistical significance and economic size tend to go hand in hand.

[Jongeun Park]

Hi,

When I run weakivtest, it keeps running without showing the results. Is there any solution to this?

use http://fmwww.bc.edu/ec-p/data/wooldridge/mroz.dta, clear
ivreg2 lwage exper expersq (educ=fatheduc motheduc)
weakivtest

Interestingly, if I add one more IV age, then it works fine.

ivreg2 lwage exper expersq (educ=fatheduc motheduc age)
weakivtest