Xtabond2

[Mareike]

Hello everybody.

I am new with Stata and I need help with the xtabond2 command.

Here are my questions:

1. Year dummies, do I have to include the year dummies before the comma or just in the iv( ) command?
2. How do I interpret a result that looks like this:

------------------------------------------------------------------------------
Arellano-Bond test for AR(1) in first differences: z = -0.96 Pr > z = 0.336
Arellano-Bond test for AR(2) in first differences: z = -0.32 Pr > z = 0.753
------------------------------------------------------------------------------
Sargan test of overid. restrictions: chi2(219) =2606.37 Prob > chi2 = 0.000
(Not robust, but not weakened by many instruments.)

Difference-in-Sargan tests of exogeneity of instrument subsets:
GMM instruments for levels
Sargan test excluding group: chi2(200) =2364.99 Prob > chi2 = 0.000
Difference (null H = exogenous): chi2(19) = 241.38 Prob > chi2 = 0.000
gmm()
Sargan test excluding group: chi2(10) = 37.70 Prob > chi2 = 0.000
Difference (null H = exogenous): chi2(209) =2568.68 Prob > chi2 = 0.000
iv()
Sargan test excluding group: chi2(203) =2365.18 Prob > chi2 = 0.000
Difference (null H = exogenous): chi2(16) = 241.19 Prob > chi2 = 0.000

3. My p values are very close to 1, so highly signifcant. Can you recommend a paper where I can find the leves for significance level?

Thanks a lot!

Mareike

[Nick Cox]

Some pointers from the FAQ Advice:

1. You are asked to explain where user-written programs you refer to come from. Thus xtabond2 is from SJ, updated on SSC.

2. We ask posters to use full real names, e.g. given name and family name.

Your last question (your #3) seems confused:

My p values are very close to 1, so highly signifcant. Can you recommend a paper where I can find the leves for significance level?

None of your P-values is close to 1, nor would such values be reported as "highly significant". Perhaps 1 here is a typo for 0.

More crucially, the P-values are cited as calculated P-values, so published tables of critical values at conventional levels are not needed. If you feel an urge to reject at a certain critical level, e.g. 5% or 1%, you have all the information you need.

(I don't use this stuff, so have no comment on the other questions.)

[Mareike]

Thanks for the fast reply.
I tried to add my last name but I cant happen to find where to change it!

For #3. Sorry, I wasnt referring to the results from the tests listed above but to the following:

Coef.	Std. Err	t	P>|t]	[95% Conf. Interval]
4.074176	197.5678	0.02	0.984	-383.4825 391.6309
229.66247	224.2042	0.13	0.895	-410.1453 469.4702
91.80167	1136.746	0.08	0.936	-2138.084 2321.687

[Nick Cox]

Do please read the FAQ Advice as requested of posters. Section 6 explains how to change your name. http://www.statalist.org/forums/help

You are now citing P-values you didn't show us, and although they are close to 1, the same remark applies: You hardly need tables of critical values, and in any case Stata's own statistical functions can be used. See the functions for t distributions under help density functions

[Richard Williams]

Unfortunately I can't help you much with the substance of your question. I can say that the output you provide would be much easier to read if you used code tags. When typing a message, click on the underlined A next to the smiley, then click #. Copy and paste your code and output between the two code tags that are generated.

For the p values you show in post #3, the effects are highly insignificant. None of the three effects shown significantly differs from 0 or even comes close to significantly differing from 0.

[David Roodman]

1. You should include year dummies before the comma as well.
2. The "AR()" (really, serial correlation) tests are showing no sign of first-order serial correlation, which is good. The Sargan test results are telling you that the instruments are collectively invalid. But this might change once you add year dummies. They are also telling me that you did not specify "robust" for the standard errors, which is unusual, and un-conservative. If you specify robust, then Hansen statistics will be reported instead.

[Ramzi Trabelsi]

if the hansen test p value is greater than 0,25 , can we accept that instuments are valid?