UndefVarError:r(999)-running reghdfejl in stata

[semiloreadegbonmire]

Hi David,

I am running a replication package for one of the papers we recently submitted. I have a master.do file that has a master cleaning and analysis file. When I run the whole master.do file, I get the UndefVarError:r(999) when the master analysis file is about to start running as it uses reghdfejl. I'm a bit confused on where this error came from. The weird part is if I run the master do file in a new stata session commenting out the master cleaning file, the master analysis and reghdfejl works fine. I recently started having this issue and I'm looking for how to fix it as ideally I would want the master.do file to be able to run everything. Here's an example:

. ****************************************************************
. *** Main table (IV/RF/OLS)
. ****************************************************************
. if rf_iv_ols'==1 | run_all'==1 {
.
. label var any_inst "Post expansion"
. label var cahoots_any "CRT"
. local ncol = 7
. local tabname "$TABLES/main/table_3.tex"
.
. MultiPartTabStart, ///

ncol(`ncol') tabname(`tabname') ///
caption("Effects of CRT response on call outcomes using expansions as instruments") ///
label("tab:dind_main") scalebox(.8) ///
colnames("CRT" "Police" "Any response" "Arrest" "Invol. hold" "Detox" "Warrant" ) ///
mgroups(`""Call responses" "Call outcomes""') pattern(1 0 0 1 0 0 0) tabular

1 4 8
1,4
4,8

.
.
. // panel A
. MultiPartTabPanelStart, ///

ncol(`ncol') tabname(`tabname') ///
panelstring("Panel A: Reduced form effect of expansion")        

. eststo clear
. foreach y in cahoots_any police_any outcome_any_response outcome_arrest arrest_rept_nch arrest_rept_detox arrest_rept_warrant {
2. eststo: reghdfejl y' any_inst [aw=rewt], absorb(${dind_base_fes}) ${se_dind} 3. qui su y' [aw=rewt] if treat==1 & any_inst==0
4. estadd scalar depmean = r(mean)
5. }
UndefVarError:r(999);
. MultiPartTabPanelEnd, ///

ncol(`ncol') tabname(`tabname') keep(any_inst) ///
stats(depmean, fmt(%9.3f) ///
labels("Untreated mean")) b(%9.3f) substitute(" ." "0." "-." "-0.")

.
. // panel B
. MultiPartTabPanelStart, ///

ncol(`ncol') tabname(`tabname') ///
panelstring("Panel B: IV estimates of effect of CRT response")          

. eststo clear
. eststo: reg police_any
. foreach y in police_any outcome_any_response outcome_arrest arrest_rept_nch arrest_rept_detox arrest_rept_warrant {
2. // hausman test
. ivreghdfe y' (cahoots_any = inst*), absorb(${dind_base_fes}) ${se_dind_iv} endog(cahoots_any) 3. local htest = e(estat) 4. local hpval = e(estatp) 5. . calc_ccm y', d(cahoots_any) z(inst*) absorb(${dind_base_fes}) ${se_dind}
6. local ccm = r(ccm)
7. eststo: ivreghdfe y' (cahoots_any = inst*), absorb(${dind_base_fes}) ${se_dind} 8. estadd scalar ccm = ccm'
9. estadd scalar fstat = e(widstat)
10. estadd local newline_lab ""
11. estadd scalar jtest = e(j)
12. estadd scalar jpval = e(jp)
13. estadd scalar htest = htest' 14. estadd scalar hpval = hpval'
15. }
. MultiPartTabPanelEnd, ///

ncol(`ncol') tabname(`tabname') keep(cahoots_any) ///
stats(ccm fstat newline_lab jtest jpval newline_lab htest hpval, ///
      fmt(%9.3f %9.2f %9.2f %9.3f %9.3f %9.3f %9.3f %9.3f) ///
      labels("Control complier mean" "First-stage F-stat" ///
      "\(J\) test of overidentification" "\quad \(\chi^2\) statistic" "\quad \(p\)-value" ///
      "Hausman test" "\quad \(\chi^2\) statistic" "\quad \(p\)-value" )) ///
b(%9.3f) substitute(" ." "0." "-." "-0.")

.
. // panel C
. MultiPartTabPanelStart, ///

ncol(`ncol') tabname(`tabname') ///
panelstring("Panel C: OLS estimates of effect of CRT response")        

. eststo clear
. eststo: reg police_any
. foreach y in police_any outcome_any_response outcome_arrest arrest_rept_nch arrest_rept_detox arrest_rept_warrant {
2. eststo: reghdfejl y' cahoots_any, absorb(${dind_base_fes}) ${se_dind} 3. su y' if cahoots_any==0
4. estadd scalar depmean = r(mean)
5. estadd scalar obs_full = e(N_full)
6.
. }
. MultiPartTabPanelEnd, ///

ncol(`ncol') tabname(`tabname') keep(cahoots_any) ///
stats(depmean obs_full, fmt(%9.3f %9.0fc) ///
labels("Untreated mean" "Observations")) ///
b(%9.3f) substitute(" ." "0." "-." "-0.")

. MultiPartTabEnd, ///

ncol(`ncol') tabname(`tabname') scalebox width(1) notes(" ") tabular

. }
r(999);
r(999);
r(999);

end of do-file

r(999);

[droodman]

Are you able to send me exact data and code to reproduce this?
If not, can you start a log, do set trace on and send me the log for a case when this error happens?
What OS are you using, and which version of Stata? What do you get from which reghdfejl and which jl?

[semiloreadegbonmire]

Hey David, I am not able to send either but I will start a log with set trace on and send over the log when it happens again. I am using Stata SE 19.5. I'm running this on a windows. when i do which reghdfejl and which jl , I get the stuff below:

. which jl
C:\Users\John Adegbonmire\ado\plus\j\jl.ado
*! jl 2.0.0 18 January 2026
*! Copyright (C) 2023-26 David Roodman

. which reghdfejl
C:\Users\John Adegbonmire\ado\plus\r\reghdfejl.ado
*! reghdfejl 1.1.8 12 January 2026

[droodman]

Does the data cleaning script include any calls to reghdfejl or jl? Or does this problem never happen if you run the cleaning and the estimation in a new Stata session?

I'm worried that the problem here is something I can't do much about. I learned a year ago that Stata will sometimes unload plugins as it manages memory. For the jl plugin, this is a big problem. Even if the plugin can be reloaded, it's like restarting a Julia session. All data and loaded packages--all context--is lost. A Stata employee told me they might create an option to prevent this, but not until Stata 20.

If this is the problem, then doing something in Julia, then doing a lot of stuff not using Julia, then coming back to Julia could fail.

I think the crash might be in the first jl command in reghdfejl.ado, around line 449, jl PutVarsToDF....

That said, I might be able to build a work-around if this proves to be the issue.

[semiloreadegbonmire]

Hi David, the data cleaning files only use reghdfe not jl or reghdfejl. I still encounter the same issue if i run the cleaning and estimation in a new stata session. I am running the code right now with set trace on so i should be able to get something more informative back to you soon. I also figured the issue you mentioned might be whats happening as cleaning probably takes up a lot of memory as it involves lots of tempfiles. I do find it weird that cleaning doesnt use julia in any form tho but reghdfejl still dies.

[semiloreadegbonmire]

Hi David , I have the master log file attached below as well as an image of my stata interface.

master.log

[droodman]

John, are you willing and able to iterate with me on this? It looks like my previous theory is wrong, which is a relief. However, I am still mystified. I think the real crash is coming in these commands, which, deep down, are triggered by your first call to reghdfejl:

plugin call _julia, start "C:\Users\John Adegbonmire\.julia\juliaup\julia-1.12.5+0.x64.w64.mingw32\bin/libjulia.dll" "C:\Users\John Adegbonmire\.julia\juliaup\julia-1.12.5+0.x64.w64.mingw32\bin"
plugin call _julia, evalqui "using Pkg"

The first line should load and launch Julia. The second sends a first, simple command to Julia; I think it is crashing. This suggests that the first line is failing to load libjulia.dll, which is the core of Julia.

Would you be able to insert these lines right before the first call to MultiPartTabPanelStart?

program _julia, plugin using(jl.plugin)
plugin call _julia, start "C:\Users\John Adegbonmire\.julia\juliaup\julia-1.12.5+0.x64.w64.mingw32\bin/libjulia.dll" "C:\Users\John Adegbonmire\.julia\juliaup\julia-1.12.5+0.x64.w64.mingw32\bin"
plugin call _julia, eval "using Pkg"

And actually maybe force the code to stop there by commenting out or deleting the rest of the script or inserting a bad command like "xxx".

I predict that the last line will produce an error message. If true, then...I'll still have a challenge.

Can you confirm that "C:\Users\John Adegbonmire.julia\juliaup\julia-1.12.5+0.x64.w64.mingw32\bin\libjulia.dll" exists and that you have read access to it, e.g., by making a copy of it? Is this an organization-owned computer with a lot of security controls on it?
How much RAM do you have?