Error on running ppmlhdfe command

[Janine Boshoff]

Hi Statalists!

I am trying to run a Poisson pseudo-likelihood regression to understand the impact of an intervention on the occurrence of juvenile crime. My initial investigation used the reghdfe command as follows:

reghdfe youthcrime intervention, absorb(date) cluster(region)

This produced output without any issue, but when I substitute the reghdfe command for the ppmlhdfe command, I get the following error message:
always_run_lsmr_preconditioner not found in class FixedEffects
(278 lines skipped)
(error occurred while loading ppmlhdfe.ado)
r(3000);

I've done a bit of searching on Google and found that someone had older package dependencies that may have caused an issue. I have the following versions loaded:

• ftools: version 2.48.0
• reghdfe: version 6.12.1
• ppmlhdfe: version 2.2.0

Has anyone else had a similar issue that can advise on how they solved it?

Thank you!

[Andrew Musau]

ppmlhdfe is from SSC (FAQ Advice #12). First test whether the command works on a publicly accessible dataset.

use "http://fmwww.bc.edu/RePEc/bocode/e/EXAMPLE_TRADE_FTA_DATA" if category=="TOTAL", clear
egen imp = group(isoimp)
egen exp = group(isoexp)
ppmlhdfe trade fta, a(imp#year exp#year imp#exp) cluster(imp#exp)

Res.:

. ppmlhdfe trade fta, a(imp#year exp#year imp#exp) cluster(imp#exp)
Iteration 1:   deviance = 3.3342e+09  eps = .         iters = 5    tol = 1.0e-04  min(eta) =  -3.58  P   
Iteration 2:   deviance = 8.4785e+08  eps = 2.93e+00  iters = 5    tol = 1.0e-04  min(eta) =  -4.93      
Iteration 3:   deviance = 4.4486e+08  eps = 9.06e-01  iters = 5    tol = 1.0e-04  min(eta) =  -6.27      
Iteration 4:   deviance = 3.8460e+08  eps = 1.57e-01  iters = 4    tol = 1.0e-04  min(eta) =  -7.32      
Iteration 5:   deviance = 3.7800e+08  eps = 1.74e-02  iters = 4    tol = 1.0e-04  min(eta) =  -8.38      
Iteration 6:   deviance = 3.7738e+08  eps = 1.65e-03  iters = 3    tol = 1.0e-04  min(eta) =  -9.01      
Iteration 7:   deviance = 3.7733e+08  eps = 1.29e-04  iters = 2    tol = 1.0e-04  min(eta) =  -9.30      
Iteration 8:   deviance = 3.7733e+08  eps = 4.91e-06  iters = 2    tol = 1.0e-04  min(eta) =  -9.37      
Iteration 9:   deviance = 3.7733e+08  eps = 3.03e-08  iters = 2    tol = 1.0e-05  min(eta) =  -9.39   S  
Iteration 10:  deviance = 3.7733e+08  eps = 3.10e-12  iters = 2    tol = 1.0e-07  min(eta) =  -9.39   S  
Iteration 11:  deviance = 3.7733e+08  eps = 0.00e+00  iters = 1    tol = 1.0e-09  min(eta) =  -9.39   S O
------------------------------------------------------------------------------------------------------------
(legend: p: exact partial-out   s: exact solver   h: step-halving   o: epsilon below tolerance)
Converged in 11 iterations and 35 HDFE sub-iterations (tol = 1.0e-08)

HDFE PPML regression                              No. of obs      =      5,950
Absorbing 3 HDFE groups                           Residual df     =      1,189
Statistics robust to heteroskedasticity           Wald chi2(1)    =      21.04
Deviance             =  377332502.3               Prob > chi2     =     0.0000
Log pseudolikelihood = -188710931.7               Pseudo R2       =     0.9938

Number of clusters (imp#exp)=      1,190
                            (Std. Err. adjusted for 1,190 clusters in imp#exp)
------------------------------------------------------------------------------
             |               Robust
       trade |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         fta |   .1924455   .0419527     4.59   0.000     .1102197    .2746713
       _cons |   16.45706   .0217308   757.32   0.000     16.41447    16.49965
------------------------------------------------------------------------------

Absorbed degrees of freedom:
-----------------------------------------------------+
 Absorbed FE | Categories  - Redundant  = Num. Coefs |
-------------+---------------------------------------|
    imp#year |       175           0         175     |
    exp#year |       175           5         170     |
     imp#exp |      1190        1190           0    *|
-----------------------------------------------------+
* = FE nested within cluster; treated as redundant for DoF computation

.

If so, then check what is peculiar with your dataset. If not, uninstall the command and all of its dependencies and then reinstall.

[Janine Boshoff]

Hi Andrew,

Thank you for your message. I followed your advice to uninstall and reinstall the package and its dependencies, but this appears to be an environment-specific issue. I used the auto.dta dataset preloaded into Stata on both my personal computer and my office-based computer. I tested the following command in both environments:

ppmlhdfe price mpg, a(rep78) cluster(weight)

When running the ppmlhdfe command on my personal computer, it produces output as it should. When I replicate the command in Stata on my office computer, I get the error message mentioned above
always_run_lsmr_preconditioner not found in class FixedEffects
(278 lines skipped)
(error occurred while loading ppmlhdfe.ado)
r(3000);

The versions of the package and its dependencies are as follows:

1. On my personal computer:
   • ftools: version 2.37.0 16aug2019
   • reghdfe: version 5.7.3 13nov2019
   • ppmlhdfe: version 2.2.0 02aug2019
2. On my office computer:
   • ftools: version 2.49.0 06may2022
   • reghdfe: version 6.12.2 02nov2021
   • ppmlhdfe: version 2.2.0 02aug2019

If you have any further advice, I would be very grateful.

[Joao Santos Silva]

Dear Janine Boshoff,

Following up on Andrew's helpful advice, I suggest that you make sure you have an up-to-date version of Stata.

Best wishes,

Joao