j***m
发帖数: 74 | 1 if I cannot find an instrument variable, is there anyway to test endogeneity
? |
s*****w
发帖数: 2065 | 2 how about thinking of the intuition
do you mean testing or solving?
endogeneity
【在 j***m 的大作中提到】
: if I cannot find an instrument variable, is there anyway to test endogeneity
: ?
|
j***m
发帖数: 74 | 3 I think test first and then solve.
any advice? thanks |
i*******e
发帖数: 349 | 4 If I remember correctly, you need more instruments than possible endogenous
variables to test for endogeneity. |
a**n
发帖数: 3801 | 5 why don't you just plot x vs. residuals?
endogeneity
【在 j***m 的大作中提到】
: if I cannot find an instrument variable, is there anyway to test endogeneity
: ?
|
j***m
发帖数: 74 | 6 sounds good idea. is it common and convincible?
【在 a**n 的大作中提到】
: why don't you just plot x vs. residuals?
:
: endogeneity
|
a**n
发帖数: 3801 | 7 not so formal..
if you can observe obvious trend, it's quite convincing that
endogeity exists..
however it's not so safe to claim no endogeneity if no trend
is detected by your eyes.. my 2 cents..
this paper maybe relevant
http://ideas.repec.org/p/hhs/umnees/0553.html
【在 j***m 的大作中提到】
: sounds good idea. is it common and convincible?
|
w****y
发帖数: 53 | 8 You could only have a sense of the serverity of the endogeneity problem.
Try to include additional variables in your model, and see how the results
will change. Another way is to assume some form of endogeneity, and then to
assess the robustness of your results when you vary the endogeneity level.
Per's paper is another useful exercise. |
f*******r
发帖数: 257 | 9 That's correct. There is no way to test endogeneity without a valid
instrument, unfortunately. Checking endogeneity based on graphs is at least
not main-stream.
endogenous
【在 i*******e 的大作中提到】
: If I remember correctly, you need more instruments than possible endogenous
: variables to test for endogeneity.
|
j***m
发帖数: 74 | 10 Thanks for all the inputs. really appreciate them |
h****i
发帖数: 1674 | 11 Is there any method to identify the endogeneity without extra instruments?
If we can easily find the instruments, we partly solve the problem already.
Still confusing |
f***5
发帖数: 1569 | 12 who is per? where I can find it
【在 w****y 的大作中提到】
: You could only have a sense of the serverity of the endogeneity problem.
: Try to include additional variables in your model, and see how the results
: will change. Another way is to assume some form of endogeneity, and then to
: assess the robustness of your results when you vary the endogeneity level.
: Per's paper is another useful exercise.
|
f*******r
发帖数: 257 | 13 What I should have said is that you need instruments as least as many as
endogenous variables to do IV regression. With valid instruments, you can
do Hausman-Wu test to check whether IV regression is indeed needed. With
extra instruments, you can do over-identification test to check each
instrument's exogeneity. These in my opinion, it's not really a test of
endogeneity. But we don't have anything better... |
