iebaltab: F-test for joint significance of balance variables

[luizaandrade]

John has pointed out that we are using an unconventional method for the F-test of joint significance of balances variables in a group. Under the conventional way, it would still be possible to include observations that have missings in one of the balance variables to the test.

The way to implement the conventional way in Stata would be something like this:

gen random_group =  uniform()
gen group               = (random_group < .5)

gen random_miss = uniform()
replace         mpg = . if random_miss < 0.05 

foreach var in mpg rep78 headroom {
               reg `var' i.group
   est sto reg_`var'
}

suest reg_mpg reg_rep78 reg_headroom
test [reg_mpg_mean]group   = [reg_rep78_mean]group
test [reg_rep78_mean]group = [reg_headroom]group, accum

[kbjarkefur]

I trust John more than I trust myself when it comes to this, but I have also heard different economists claiming that contradicting practices are the convention for different tests. So while I agree that this should be included, I wonder what should be the default.

Maybe we should test if there are missing values, and if there are any missing values in any variable an error will be thrown forcing the user to make an active choice between this approach and the approach currently implemented in iebaltab. What do you think?

[luizaandrade]

That seems reasonable. I guess we should also test that they give the same result when there are no missings