Professor Jeff Wooldridge or anybody else who knows the answer, or has encountered the topic before and can provide a reference:
I have a Generated Regressor X, and a set of standard variables W (which might be one, or might be more, let's say for the sake of concreteness that they are two W1 and W2). The generated regressor X is orthogonal to the set of standard regressors W. I am fitting the regression
Y = a + b*X + c*W1 + g*W2 + e
and I want to test Ho: c=g=0.
Would the test statistics, i.e., the variance matrix of W1 and W2 and the F/Chi statistic for joint significance, from the regression output be correct for this Ho?
In other words, is
Code:
reg Y X W1 W2 test W1 W2
I think the answer is Yes, because there is such a result in the measurement error literature. If X is subject to measurement error, and W1 and W2 are both orthogonal to X, then if I am interested in testing hypothesis only on W1 and W2 I can proceed disregarding the fact that X is subject to measurement error.
But I was not able to find any result like this for generated regressors after a day of googling.
0 Response to If a Generated Regressor X is orthogonal to a non-generated regressor W, would the variance and tests on W be correct?
Post a Comment