Page **1** of **1**

### F-Statistic in 2SLS

Posted: **Wed Dec 07, 2011 4:10 pm**

by **tgmoe**

The 2SLS output provides the F-Statistic, is this the same as the first-stage f-statistic reported in Stata? They are providing different results, should this be the case? If so, can someone explain why? Thanks

### Re: F-Statistic in 2SLS

Posted: **Wed Dec 07, 2011 4:37 pm**

by **startz**

It is not the first-stage F. It's the F-statistic for the hypothesis that all the coefficients except the intercept equal zero.

### Re: F-Statistic in 2SLS

Posted: **Wed Dec 07, 2011 4:39 pm**

by **EViews Glenn**

The EViews 2SLS output produces the F-statistic for the significance of the regressors in the 2SLS specification, not for the first-stage. Bearing in mind that I am not certain what Stata is doing, if they are producing an F-statistic based on the first stage, it would presumably be quite different.

I just noticed that Startz posted a reply to your original question. Coincidentally, I was going to provide you with a reference:

Startz, Richard, "Computation of Linear Hypothesis Tests for Two-Stage Least Squares," Economics Letters 11, 1983, pp. 129-131.

### Re: F-Statistic in 2SLS

Posted: **Wed Dec 07, 2011 5:09 pm**

by **tgmoe**

Alright great, thanks.

### Re: F-Statistic in 2SLS

Posted: **Sun Sep 07, 2014 10:42 pm**

by **kruupy**

Wooldridge(2012) says you shouldnt conduct F-Tests using 2SLS because the R^2 is able to be negative.

Wooldridge suggests that there are other ways to conduct an f-test that avoids this problem and that most statistical software uses this wrap around. Does Eviews correct for this problem?

### Re: F-Statistic in 2SLS

Posted: **Mon Sep 08, 2014 9:42 am**

by **EViews Glenn**

I'm not certain of which Wooldridge reference you are citing, nor his recommeded workarounds. The R2 form of F-statistics is an computational shortcut that one can use in certain, but not all, cases. EViews has never used this shortcut; the reference I provided above describes in great detail what we do.