If I estimate the following regression on a large data set: $y = b_0x_0 + b_1x_1 + b_2x_2 + b_3x_3 + b_4x_4$ where $x_0, x_1,...x_4$ are all dummy variables indicating group membership, and I create two new variables, $x_5$ and $x_6$, that indicate membership such that $x_5$ indicates $x_0, x_1$, or $x_2$ being switched on, and $x_6$ indicates $x_3$ or $x_4$ being switched on, and then I estimate $y = b_5x_5 + b_6x_6$.

What would be the intuitive difference between testing the significance of $b_5$ and $b_6$ using t-stats vs. doing two separate F-tests, one on $b_0, b_1$, and $b_2$, and the other on $b_3$ and $b_4$? Is there any difference? Would the F-tests just be the t-stats squared in this case? Any help is appreciated.


1 Answer 1


$x_5 = 1-(1-x_0) (1-x_1) (1-x_2)$; $x_6= 1- (1-x_3)(1-x_4)$

As such,

$$b_5 x_5 = b_5 [1-(1-x_0) (1-x_1) (1-x_2)] \\ = b_5 x0 +b_5 x_1 + b5 x_2-b_5x_0x_1 -b_5x_0x_2 -b_5x_1x_2 +b_5x_0x_1x_2$$

That is not the same as $b_0x_0+b_1x_1+b_2x_2$.

Similarly with $x_6$ in terms of $x_3$ and $x_4$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.