Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Cragg Donald tests

From   Abekah Nkrumah <>
Subject   Re: st: Cragg Donald tests
Date   Thu, 22 Mar 2012 14:19:15 +0000

Dear Karan,

I have looked at Cameron and Trivedi's book and the understanding I
get is that it depends on the number of endogenous regressors in your

If you have only one regressor then the minimum eigen value derived
will be the same as the  Cragg-Donald F statistic. Thus you compare
the F statistic to the thresholds you have stated in your mail which
is actually the Stock and Yogo's thresholds. The principle is that you
first decided on a tolerable bias level say 10% which in your
calculation is 16.38 of the Stock and Yogo threshold. Thus if your F
statistic is higher than the 16.38 then one can cautiously conclude
that the instruments are not weak

If you have more than one endogenous regressors say 2, then you will
have two first stage and therefore 2 F statistics. You can then
calculate the minimum eigen value of the two statistics which is then
compared to the the Stock and Yogo threshold values. Remember that the
minimum eigen value is calculated by stata. Remember also that
theoretically this is based on the assumption of iid errors.

Thank you very much and hope this helps


On Thu, Mar 22, 2012 at 12:14 AM, Karan Nagpal <> wrote:
> I'm trying to test the strength of my instruments, but not sure how to
> interpret the Cragg-Donald IV values. I implemented ivreg2, and
> received the following output:
> Stock-Yogo weak ID test critical values:
> 10% maximal IV size             16.38
>                                          15% maximal IV size              8.96
>                                          20% maximal IV size              6.66
>                                          25% maximal IV size              5.53
> I'm not sure what I should be comparing these values to and how I can
> use the "main output" to find the Cragg-Donald F statistic, which is
> what stata advised.
> Thanks,
> Karan
> *
> *   For searches and help try:
> *
> *
> *

*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index