Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | "Wooldridge, Jeffrey" <wooldri1@msu.edu> |
To | <statalist@hsphsun2.harvard.edu> |
Subject | RE: Interpretation RESET-Test: Problems with different test options |
Date | Fri, 4 Mar 2011 11:45:21 -0500 |
Properly using RESET can be tricky. I have argued in a paper, "Score Diagnostics for Linear Models Estimated by Two Stage Least Squares," in Advances in Econometrics and Quantitative Economics. G.S. Maddala, P.C.B. Phillips, and T.N. Srinivasan (eds.), 66-87. Oxford: Blackwell, 1995, that RESET is purely a functional form test. It should not be expected to have systematic power for omitted variables or for detecting heteroskedasticity. (A test for heteroskedasticity only makes sense using squared residuals, not the residuals themselves, which is what RESET uses.) The argument is simple. Suppose E(y|x,z) = b0 + b1x + b2z and also E(z|x) is linear in x. Then E(y|x) is linear in x -- and, of course, the estimator is generally inconsistent for b1. That E(y|x) is linear in x means no other functions of x will matter. So RESET will fail to reject (more precisely, have power equal to size, asymptotically). That using polynomials in fitted values leads to a rejection whereas polynomials in the xj themselves is not surprising. They are different functions of the explanatory variables, and often the first choice conserves on degrees of freedom. Note that the squared fitted value, (b0 + xi*b1)^2 includes interactions as well as squares of all variables. The fact that adding another variable causes RESET to reject is also easily explained. For example, suppose E(y|x,z) = b0 + b1*x + b2*z + b3*z^2 If b3 != 0 and you regress y on x, z then RESET should reject because the linear functional form is incorrect. But suppose z and x are independent. Then E(y|x) is linear and RESET will not reject. I hope this helps. JMW Jeffrey M. Wooldridge University Distinguished Professor Department of Economics Michigan State University 110 Marshall-Adams Hall East Lansing, MI 48824-1038 Phone: 517-353-5972 Fax: 517-432-1068 http://www.msu.edu/~ec/faculty/wooldridge/wooldridge.html -----Original Message----- From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of matthias.stoetzer@bw.fh-jena.de Sent: Friday, March 04, 2011 8:18 AM To: statalist@hsphsun2.harvard.edu Subject: Interpretation RESET-Test: Problems with different test options Dear STATA community, in order to test for specification errors I use the RESET-test. I understand that RESET is not a test for omitted variables but more a test looking for non-linearities. I have two problems as to the interpretation of my results: 1. There is a striking difference of the RESET test using powers of the fitted values of the dependent variable (default-option) and using powers of the independent variables (rhs-option). In the first case with my data set the Ho has to be rejected (5%-level). In the second case the Ho can not be rejected even at the 40%-level. For sure this has to do with a difference of the dependent and independet variables but is there any precise explanation as to this difference? 2. I run a multiple regression. The RESET-test (default-option) does not reject the Ho of correct specification (10% level). I run a regression with the same dependent variable and the same independent variables but in addition I include one other independent variable: Now the RESET-test (default-option) rejects the Ho of correct specification. To my opinion including additional variables should reduce the probability of omitting variables and neglecting non-linearities. In advance many thanks for any suggestions! Matthias S. * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/ * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/