Please panel composed of 28 countries and 13 years but when i introduce difference GMM it become 11 period included
1- i found Sargan p-value for over-identifying restrictions p-value =0.52 and 0.66 for my 2 models so how interpret those p-value? are they good?
2- my data are integrated after applying first difference since my model include fiirst difference, but i dont no if i say my variables are integrated of ordre I(0) or should i say first difference I(1)?
Thank you
Sargan test of overidentifying restrictions
H0: overidentifying restrictions are valid
The null hypothesis of the test implies all instruments are valid
p value greater than 5% (0.05) implies, we accept the Ho, that is all instruments are valid.
@ Augustine Ujunwa, please a reference to your answer will be appreciated.
Goodman, D. (2009). How to do xtabond2: An Introduction to Difference and System GMM in Stata. The Stata Journal 9(1): 86 -136
There are three conditions in applying Sargan's test. First, the p-value must be greater that 5%. Second, the p-value must not be less than 0.1. Third, the p-value must be greater than 0.25. In the extant case, p-values of 0.52 and 0.66 is a serious call for concern.
… Read more
Sargan test has a null hypothesis (Ho): The Instruments as a group are exogenous. Sargan p-value must not be less < 5% and > 10%. The higher the p-value of the sargan statistic the better. However according to Roodman (2006) , it is recommended that sargan p-value should be greater than 0.25.
This does not invalidate other results that rejects the null hypothesis.
So, for me the above results with p-values of 0.52 and 0.66 indicate the group of instruments used in the analysis are validly exogenous.
Dear Osarumwense Osabuohien-Irabor . Thank you for your answer. Roodman has several publications in 2006. Would you please specify the publication and page number of the article? Thank you in advance.
Dear Abida,
According to Roodman, D. (2009). A note on the theme of too many instruments. Oxford Bulletin of Economics and Statistics, 71(1), 135-158,
"A p-value as high as, say, 0.25 should be viewed with concern" (Page 142) and using too many instruments can lead to several statistical problems, especially an overfitting bias of the coefficients of the endogenous variables and a weak Hensen test of instrument validity. As a solution Roodman (2009, p. 156) suggests the following “Results should be aggressively tested for sensitivity to reductions in the number of instruments.”
I recommend this paper to process with reducing the number of instruments.
Best,
Faten Ben Bouheni
There are some misconceptions about the interpretation of the Sargan test, even here in this post. Piper 2014 (https://mpra.ub.uni-muenchen.de/59556/) discusses it in footnote 11. Roodman has several publications attending this matter (2006, 2007, 2009) and is almost always referenced when the test is used in empirical papers, but sometimes Roodman's notes are understood incorrectly. When Roodman says that p-values (for the sargan test) over 0.25 should be "viewed with concern", like in Faten Ben Bouheni answer, he means that even at 25% error you should not be satisfied with the certainty that you reject the null hypothesis, as it represents a probability of 1 in 4 to obtain a J-score that large. Remembering that Ho is: Instruments as a group are exogenous (as Osarumwense Osabuohien-Irabor correctly points out) and that this is what you want (you don't want to reject that the instruments are valid!), 25% chance it is still too low!
In summary, look for p-values 0.25 and above.
Alternatively, use robust standard errors and your sargan worries are over.
Hi Oluwaseun Oyebamiji, if we use robust standard errors in system GMM than the STATA does not provide the Sargan test results. Can please guide me in this context.
Hello Augustine Ujunwa . I found p-value = 0.255514 Can we say that this is within acceptable limits?
- Badges
- Science topic
- Similar topics
- Mathematics