Austin, so let me make sure with you with the info from my ivreg2
below that indeed my identification is strong as the CraggDonald F
stat is above the critical values from OLS bias and that I am
rejecting my hypothesis of weak instruments actually at 20%.
Much thanks,
Number of observations N = 728
Number of regressors K = 57
Number of instruments L = 109
Number of excluded instruments L1 = 53
Number of partialled-out regressors/IVs = 1
------------------------------------------------------------------------------
Underidentification test (Anderson canon. corr. LM statistic): 607.412
Chi-sq(53) P-val = 0.0000
------------------------------------------------------------------------------
Weak identification test (Cragg-Donald Wald F statistic): 58.734
Stock-Yogo weak ID test critical values: 5% maximal IV relative bias 21.30
10% maximal IV relative bias 11.10
20% maximal IV relative bias 5.86
30% maximal IV relative bias 4.07
10% maximal IV size 141.08
15% maximal IV size 72.44
20% maximal IV size 49.26
25% maximal IV size 37.65
Source: Stock-Yogo (2005). Reproduced by permission.
------------------------------------------------------------------------------
Sargan statistic (overidentification test of all instruments): 58.317
Chi-sq(52) P-val = 0.2544
------------------------------------------------------------------------------
On 9/25/07, Austin Nichols <[email protected]> wrote:
> Nirina F <[email protected]>:
> The IV estimate is always biased, but is less biased than OLS to the
> extent that identification is strong; in the limit of weak
> instruments, there would be no improvement over OLS in terms of bias
> and the bias would be 100% of OLS, and in the other limit, the bias
> would be zero percent. Clearly, you'd like to know where you are on
> that spectrum, even if only approximately. There is also a problem
> with the size of tests after IV--you think you are rejecting a
> hypothesis using a 5% alpha, but it is really 10% or 20%. Stock and
> Yogo did simulations to provide "rule of thumb" critical values (or
> rule of thumb "critical values" perhaps). Table 1 on page 39 of Stock
> and Yogo (http://ksghome.harvard.edu/~jstock/pdf/rfa_6.pdf) shows the
> value of the SY stat (a measure of the "strength of identification" or
> the predictive power of the excluded instruments) to limit the bias to
> 20% of OLS for two endogenous variables and three excluded instruments
> (n=2, K2=5) is 5.91 (similar values are given to limit the size of
> Wald tests in table 2, and stats for LIML estimates in tables 3 and
> 4).
>
> It's not clear what you don't understand from the Stock and Yogo
> paper, so it's hard to comment directly. If you gave your n and K2
> values and the SY stat from the output of -ivreg2- I suppose one could
> describe directly the situation in which you find yourself.
>
> But the key point is that all IV and IV-type specifications suffer
> from bias and size distortions, not to mention inefficiency and
> sometimes failures of exclusion restrictions. The SY stat gives you
> some measure of how strong your identification is in your sample, but
> no information about the validity of your instruments. Hope you also
> read BSS2007:
> http://econpapers.repec.org/paper/bocbocoec/667.htm
> and I expect -ranktest- will generate a new set of papers eventually
> (-ssc install ranktest- and -help ranktest- for more).
>
> On 9/25/07, Nirina F <[email protected]> wrote:
> > Dear all,
> >
> > Yes, I read
> > Stock, J.H. and Yogo, M. 2005. " Testing for Weak Instruments in
> > Linear IV Regression"
> >
> > but I still don't understand how do I interpret the critical values of
> > the Stock&Yogo- maximal IV size when using ivreg2.
> >
> > I think I compare the critical values from x%maximal IV relative bias
> > with the Cragg-Donald stat to test for weak instruments but the
> > critical values for the size with what and what it means?
> *
> * For searches and help try:
> * http://www.stata.com/support/faqs/res/findit.html
> * http://www.stata.com/support/statalist/faq
> * http://www.ats.ucla.edu/stat/stata/
>
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
