Problem Using reghdfe with only an Intercept and Double Clustering

[David Kust]

Hello Statalist members,

for my research I want to test if the mean of my dependent variable is different from zero, while additionally clustering the standard errors. I want to use reghdfe and double cluster my standard errors. Originally, I use panel data, but I have created an example that leads to the same error message as with my own data set, so that you can better understand my issue. Here’s my approach:

sysuse auto, clear
gen wc = wordcount(make)
su wc, meanonly
local max = r(max)
gen wanted = substr(make, 1, 1)
 
reghdfe price, vce(cluster foreign wanted)

However, there seems to be a problem with reghdfe when only using the intercept while clustering on multiple dimensions. I get the following output:

. reghdfe price, vce(cluster foreign wanted)
(MWFE estimator converged in 1 iterations)
reghdfe_fix_psd():  3301  subscript invalid
reghdfe_vce_cluster():     -  function returned error
reghdfe_solve_ols():     -  function returned error
<istmt>:     -  function returned error
r(3301);

I have already tried to generate a constant and add it to the regression, but this didn’t resolve my issue. Any help would be highly appreciated! Thank you!

[Andrew Musau]

reghdfe is from https://github.com/sergiocorreia/reghdfe (FAQ Advice #12). Try an older version of the command or regress (Stata 18+).

webuse grunfeld, clear
reghdfe invest, noabsorb vce(cluster company year) version(5)
regress invest, vce(cluster company year)

Res.:

. reghdfe invest, noabsorb vce(cluster company year) version(5)
(running historical version of reghdfe: 5)
(MWFE estimator converged in 1 iterations)

HDFE Linear regression                            Number of obs   =        200
Absorbing 1 HDFE group                            F(   0,      9) =          .
Statistics robust to heteroskedasticity           Prob > F        =          .
                                                  R-squared       =     0.0000
                                                  Adj R-squared   =     0.0000
Number of clusters (company) =         10         Within R-sq.    =     0.0000
Number of clusters (year)    =         20         Root MSE        =   216.8753

                          (Std. err. adjusted for 10 clusters in company year)
------------------------------------------------------------------------------
             |               Robust
      invest | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
       _cons |   145.9583   62.19122     2.35   0.044     5.271936    286.6446
------------------------------------------------------------------------------

.
. regress invest, vce(cluster company year)

Linear regression                                     Number of obs =      200
Clusters per comb.:                                   Cluster comb. =        3
  min =  10                                           F(0, 9)       =        .
  avg =  77                                           Prob > F      =        .
  max = 200                                           R-squared     =   0.0000
                                                      Adj R-squared =   0.0000
                                                      Root MSE      = 216.8753

                                  (Std. err. adjusted for multiway clustering)
------------------------------------------------------------------------------
             |               Robust
      invest | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
       _cons |   145.9583   62.31677     2.34   0.044     4.987914    286.9286
------------------------------------------------------------------------------
Cluster combinations formed by company and year.

[David Kust]

Thank you very much for your fast reply! This helps.