Logit F-Test for joint significance of subset of coefficients

[Henry Strawforrd]

I am running a Logit with many variables. I want to compute the F-test for joint significance for a subset of these variables.

Questions:
1. Is that what I would get using lincom (see below)?
2. How can I do it by hand?

Example:

sysuse auto.dta, replace
logit foreign price mpg headroom trunk weight length turn
margins, dydx(*) atmeans
lincom price + mpg

[Joseph Coveney]

I want to compute the F-test for joint significance for a subset of these variables.

Questions:
1. Is that what I would get using lincom (see below)?

No. You would get test of the sum of the coefficients.

2. How can I do it by hand?

With official Stata, you can get a test for joint significance in either of two ways, a Wald test or a likelihood-ratio test. See below.

quietly sysuse auto, clear
logit foreign c.(price mpg headroom trunk weight length turn), nolog

// Wald
test price mpg

// LR
estimates store Full
quietly logit foreign c.( /* price mpg */ headroom trunk weight length turn), nolog
lrtest Full

There is also a user-written command lmtest, which is available on SSC, to perform a Lagrange multiplier (score) test.

constraint define 1 _b[price] = 0
constraint define 2 _b[mpg] = 0
logit foreign c.(price mpg headroom trunk weight length turn), constraints(1/2) nolog
lmtest