Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | Vidhya Soundararajan <vidhyahere@gmail.com> |
To | statalist@hsphsun2.harvard.edu |
Subject | st: AP F-statistic is 0 with multiple endogenous regressors that are correlated |
Date | Thu, 4 Jul 2013 15:18:28 -0400 |
Dear statalist users, I have a model that has three endogenous regressors. Three because I have one endogenous variable (E), another that is an interaction of E with an included instrument (MW), which makes it endogenous (E * MW), and third one which is another interaction term (E*MW*MW). My model is: L = β0 + β1 * MW + β2 * MW*MW + β3 * E + β4 * E*MW + β5 * E*MW*MW + β6 * X + ε I have three instruments for these say F, G, and H. Please note that these instruments are "not" F, F*MW, and F*MW*MW. Rather, they are three separate instruments F,G, and H. There are three first stage regressions and so the test for weak instruments is not a simple test based on the F-statistic of the first stage regression but the one based on partialling out other endogenous regressors from one, as suggested in Angrist and Pischke (2009) in page 217-218. I am using the ivreg2 command in stata for obtaining these AP F-statistic and a bunch of other test statistics. The problem I encounter is that the F-statistic from individual first stage regressions for each of the three endogenous regressors are high, which is good! But the AP F-statistic are very small - literally 0. So I guess I should conclude now that my system is weakly identified (or even under-identified because the AP F-stat is 0). But I am thinking may be that AP F-statistic is low because my three endogenous regressors are probably correlated (that two endogenous regressors are just interaction terms of E)? Does this interaction play a role in why AP F-stat is 0? Best, Vidhya * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/faqs/resources/statalist-faq/ * http://www.ats.ucla.edu/stat/stata/