Reghdfe vce(robust) or vce(cluster var)

Jan Geurst

Join Date: May 2018

Posts: 18
#1

Reghdfe vce(robust) or vce(cluster var)

14 Jun 2018, 12:27

Hello,

Could somebody explain to me the difference between the two commands:

reghdfe y var1 var2, absorb(var3, var4) vce(robust)

reghdfe y var1 var2, absorb(var3, var4) vce(cluster var4)

Thank you!
Tags: None

Jan Geurst

Join Date: May 2018
Posts: 18

14 Jun 2018, 12:29

Within my analysis with the vce(robust) my variable is signifiact and with the vce(cluster var) command the variable becomes insignificant. Can someone explain to me the difference?

Code:

 reghdfe bda announcement_eligible eligible announcement l_lnassets l_roa, absorb(c incorp indus t) vce(robust)
(converged in 3 iterations)
note: eligible omitted because of collinearity
note: announcement omitted because of collinearity

HDFE Linear regression                            Number of obs   =      3,944
Absorbing 4 HDFE groups                           F(   3,   3441) =       3.18
Statistics robust to heteroskedasticity           Prob > F        =     0.0230
                                                  R-squared       =     0.8970
                                                  Adj R-squared   =     0.8820
                                                  Within R-sq.    =     0.0393
                                                  Root MSE        =     0.1505

---------------------------------------------------------------------------------------
                      |               Robust
                  bda |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
----------------------+----------------------------------------------------------------
announcement_eligible |   .0123368   .0070483     1.75   0.080    -.0014824    .0261561
             eligible |          0  (omitted)
         announcement |          0  (omitted)
           l_lnassets |  -.0151639    .021781    -0.70   0.486     -.057869    .0275411
                l_roa |  -.8755667   .3424687    -2.56   0.011    -1.547029   -.2041042
---------------------------------------------------------------------------------------

Absorbed degrees of freedom:
------------------------------------------------------------------------+
          Absorbed FE |  Num. Coefs.  =   Categories  -   Redundant     |
----------------------+-------------------------------------------------|
                    c |          493             493              0     |
               incorp |            0              16             16     |
                indus |            0              56             56 ?   |
                    t |            7               8              1 ?   |
------------------------------------------------------------------------+
? = number of redundant parameters may be higher

Code:

 reghdfe bda announcement_eligible eligible announcement l_lnassets l_roa, absorb(c incorp indus t) vce(cluster c)
(converged in 3 iterations)
note: eligible omitted because of collinearity
note: announcement omitted because of collinearity

HDFE Linear regression                            Number of obs   =      3,944
Absorbing 4 HDFE groups                           F(   3,    492) =       1.63
Statistics robust to heteroskedasticity           Prob > F        =     0.1826
                                                  R-squared       =     0.8970
                                                  Adj R-squared   =     0.8796
                                                  Within R-sq.    =     0.0393
Number of clusters (c)       =        493         Root MSE        =     0.1521

                                             (Std. Err. adjusted for 493 clusters in c)
---------------------------------------------------------------------------------------
                      |               Robust
                  bda |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
----------------------+----------------------------------------------------------------
announcement_eligible |   .0123368   .0149394     0.83   0.409     -.017016    .0416897
             eligible |          0  (omitted)
         announcement |          0  (omitted)
           l_lnassets |  -.0151639   .0309505    -0.49   0.624    -.0759755    .0456476
                l_roa |  -.8755667   .5327249    -1.64   0.101    -1.922263    .1711297
---------------------------------------------------------------------------------------

Absorbed degrees of freedom:
------------------------------------------------------------------------+
          Absorbed FE |  Num. Coefs.  =   Categories  -   Redundant     |
----------------------+-------------------------------------------------|
                    c |            0             493            493 *   |
               incorp |           15              16              1     |
                indus |           55              56              1     |
                    t |            7               8              1 ?   |
------------------------------------------------------------------------+
? = number of redundant parameters may be higher
* = fixed effect nested within cluster; treated as redundant for DoF computation

.

Comment

Andrew Musau

Join Date: Oct 2014

Posts: 9134
#3

14 Jun 2018, 13:27

Assuming that var4 is your panel variable, the default behavior for panel data commands in Stata is to use

Code:

cluster(var4)

You can find the literature relating to (the justification of) this in the manual of xtreg.

-vce(robust)- in reghdfe will cluster by observation (i.e., you assume that each observation is independent) which (arguably) is not correct for panel data. There is some dependence between observations of an entity across time. So these two are equivalent

Code:

gen n= _n reghdfe...., vce(robust) reghdfe...., cluster(n)
Comment

Jan Geurst

Join Date: May 2018
Posts: 18

15 Jun 2018, 02:14

But how come both the commands give different outputs?
reghdfe....vce(robust)
reghdfe...vce(cluster c)

Code:

. reghdfe bda announcement_eligible announcement eligible l_lnassets l_roa, absorb(c t indus incorp) vce(robust)
(converged in 3 iterations)
note: announcement omitted because of collinearity
note: eligible omitted because of collinearity

HDFE Linear regression                            Number of obs   =      3,944
Absorbing 4 HDFE groups                           F(   3,   3441) =       3.18
Statistics robust to heteroskedasticity           Prob > F        =     0.0230
                                                  R-squared       =     0.8970
                                                  Adj R-squared   =     0.8820
                                                  Within R-sq.    =     0.0393
                                                  Root MSE        =     0.1505

---------------------------------------------------------------------------------------
                      |               Robust
                  bda |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
----------------------+----------------------------------------------------------------
announcement_eligible |   .0123368   .0070483     1.75   0.080    -.0014824    .0261561
         announcement |          0  (omitted)
             eligible |          0  (omitted)
           l_lnassets |  -.0151639    .021781    -0.70   0.486     -.057869    .0275411
                l_roa |  -.8755667   .3424687    -2.56   0.011    -1.547029   -.2041042
---------------------------------------------------------------------------------------

Absorbed degrees of freedom:
------------------------------------------------------------------------+
          Absorbed FE |  Num. Coefs.  =   Categories  -   Redundant     |
----------------------+-------------------------------------------------|
                    c |          493             493              0     |
                    t |            7               8              1     |
                indus |            0              56             56 ?   |
               incorp |            0              16             16 ?   |
------------------------------------------------------------------------+
? = number of redundant parameters may be higher

Code:

 reghdfe bda announcement_eligible announcement eligible l_lnassets l_roa, absorb(c t indus incorp) vce(cluster c)
(converged in 3 iterations)
note: announcement omitted because of collinearity
note: eligible omitted because of collinearity

HDFE Linear regression                            Number of obs   =      3,944
Absorbing 4 HDFE groups                           F(   3,    492) =       1.63
Statistics robust to heteroskedasticity           Prob > F        =     0.1826
                                                  R-squared       =     0.8970
                                                  Adj R-squared   =     0.8796
                                                  Within R-sq.    =     0.0393
Number of clusters (c)       =        493         Root MSE        =     0.1521

                                             (Std. Err. adjusted for 493 clusters in c)
---------------------------------------------------------------------------------------
                      |               Robust
                  bda |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
----------------------+----------------------------------------------------------------
announcement_eligible |   .0123368   .0149394     0.83   0.409     -.017016    .0416897
         announcement |          0  (omitted)
             eligible |          0  (omitted)
           l_lnassets |  -.0151639   .0309505    -0.49   0.624    -.0759755    .0456476
                l_roa |  -.8755667   .5327249    -1.64   0.101    -1.922263    .1711297
---------------------------------------------------------------------------------------

Absorbed degrees of freedom:
------------------------------------------------------------------------+
          Absorbed FE |  Num. Coefs.  =   Categories  -   Redundant     |
----------------------+-------------------------------------------------|
                    c |            0             493            493 *   |
                    t |            7               8              1     |
                indus |           55              56              1     |
               incorp |           15              16              1 ?   |
------------------------------------------------------------------------+
? = number of redundant parameters may be higher
* = fixed effect nested within cluster; treated as redundant for DoF computation

.

And both

reghdfe...cluster(c) and
reghdfe... vce(cluster c)

Code:

. reghdfe bda announcement_eligible announcement eligible l_lnassets l_roa, absorb(c t indus incorp) vce(cluster c)
(converged in 3 iterations)
note: announcement omitted because of collinearity
note: eligible omitted because of collinearity

HDFE Linear regression                            Number of obs   =      3,944
Absorbing 4 HDFE groups                           F(   3,    492) =       1.63
Statistics robust to heteroskedasticity           Prob > F        =     0.1826
                                                  R-squared       =     0.8970
                                                  Adj R-squared   =     0.8796
                                                  Within R-sq.    =     0.0393
Number of clusters (c)       =        493         Root MSE        =     0.1521

                                             (Std. Err. adjusted for 493 clusters in c)
---------------------------------------------------------------------------------------
                      |               Robust
                  bda |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
----------------------+----------------------------------------------------------------
announcement_eligible |   .0123368   .0149394     0.83   0.409     -.017016    .0416897
         announcement |          0  (omitted)
             eligible |          0  (omitted)
           l_lnassets |  -.0151639   .0309505    -0.49   0.624    -.0759755    .0456476
                l_roa |  -.8755667   .5327249    -1.64   0.101    -1.922263    .1711297
---------------------------------------------------------------------------------------

Absorbed degrees of freedom:
------------------------------------------------------------------------+
          Absorbed FE |  Num. Coefs.  =   Categories  -   Redundant     |
----------------------+-------------------------------------------------|
                    c |            0             493            493 *   |
                    t |            7               8              1     |
                indus |           55              56              1     |
               incorp |           15              16              1 ?   |
------------------------------------------------------------------------+
? = number of redundant parameters may be higher
* = fixed effect nested within cluster; treated as redundant for DoF computation

.

give the same output.

Does it have something to do with the absorbed degrees of freedom. As presented in the two outputs you see that they are different for the two regressions. Can some explain to me what they mean?

Thank you!

Comment

Andrew Musau

Join Date: Oct 2014

Posts: 9134
#5

15 Jun 2018, 13:34

Read #3 again and the suggested xtreg manual reference.
Comment

Announcement