Possible bug in for jackknife clustered errors with a string clustering variable

[Matt Webb]

Hello,

I think I have found a bug with Stata's built in cluster jackknife. This bug seems to occur when the clustering variable is a string. Curiously, the string variable does not cause an issue for jackknife: reg y x, cluster, but fails for reg y x, cluster vce(jackknife).

I get the same error in both Stata 16 and Stata 17. Specifically, I get the error "no observations". The code below gives the same standard errors for the first three regressions, but fails on the fourth.


clear

set obs 1000

gen clusnum = ceil(_n/100)

gen clusname="."

replace clusname = "A" if clusnum==1
replace clusname = "B" if clusnum==2
replace clusname = "C" if clusnum==3
replace clusname = "D" if clusnum==4
replace clusname = "E" if clusnum==5
replace clusname = "F" if clusnum==6
replace clusname = "G" if clusnum==7
replace clusname = "H" if clusnum==8
replace clusname = "I" if clusnum==9
replace clusname = "J" if clusnum==10

gen y = rnormal()
gen x = rnormal()
gen w = rnormal()

/*jackknife with numeric clustering variable*/
reg y x w, cluster(clusnum) vce(jackknife)

jackknife: reg y x w, cluster(clusnum)

/*jackknife with string clustering variable*/

jackknife: reg y x w, cluster(clusname)

/*error - no observations*/
reg y x w, cluster(clusname) vce(jackknife)

Is this actually a bug, or am I missing something?

Cheers,

Matt

[FernandoRios]

I think the problem is more of syntax.
When using "vce" the syntax is a bit different, the correct command should be

reg y x w, vce(jackknife , cluster (clusname) )

I think the new syntax (VCE) does not "play well" with the older syntax (option 4).

[Matt Webb]

Thank you Fernando.