Hypothesis testing

[Mariela Dal Borgo]

Dear all,

I am estimating the same regression model for two subgroups, where "Positive" and "Negative" are the positive and negative components of the same shock (each takes value zero when the other is positive or negative, as indicated):


Y = a1+ a2* Positive + a3* Negative + a4*X + u (if group = 1)

Y = b1 + b2* Positive + b3* Negative + b4*X + v (if group = 0)


And I want to test the following hypotheses:

H0: a2 = a3 (symmetric effects in first group)

H0: b2 = b3 (symmetric effects in second group)

H0: a2 = b2 (same positive shock across the two groups)

H0: a3 = b3 (same negative shock across the two groups)


I was using the simple "test" command, but some of the results are inconsistent (e.g., I do not reject three of the nulls, which implies that I shouldn't reject the four either, but the test leads to rejection).(*)

So, I was wondering how to handle this. Do I need to conduct multiple hypothesis testing? In that case, what would be a good way to do it for this purpose?

Thanks a lot for any guidance on this.

(*) I get the same results by using xlincome as follows (with fixed effects and clustered s.e. ommitted for simplicity and the user-written command "reghdfe"):

reg Y i.group##c.(Positive Negative X)
xlincome (c.Positive + 1.group#c.Positive - c.Negative - 1.group#c.Negative) (c.Positive - c.Negative) (1.group#c.Positive) (1.group#c.Negative)

[Carlo Lazzaro]

Mariela:
I'm not clear whether you perforemed two different regressions or not.
If the former were the case, you may want to take a look at: https://stats.oarc.ucla.edu/stata/co...ps-using-suest

[Mariela Dal Borgo]

Dear Carlo,

That was very helpful, thank you very much! I had to do some adjustments because I am using pweights and I am clustering the s.e., but it worked well I think. Below is the simplified code.

svyset clustervar [pw=ipw]
svy:reg Y positive negative X i.timeFE i.stateFE if group == 1
eststo group1

svy:reg Y positive negative X i.timeFE i.stateFE if group == 0
eststo group0

suest group1 group0, svy
test [group1]positive=[group1]negative
test [group0]positive=[group0]negative, accum
test [group1]positive=[group0]positive, accum
test [group1]negative=[group0]negative, accum