-regress- allows both robust/cluster and mse1 as options, it should not, this is a trap

[Joro Kolev]

Good afternoon,

-regress- has the option mse1 which makes it easy to get the inv(X’X) matrix, which you might want to use in further calculations.

To my dismay today I discovered that

1. Stata does not throw an error, when you specify both cluster/robust and mse1. E.g. -reg y x, robust mse1- does not generate an error or a warning.
2. Not only this, but Stata also reports MSE = 1 in the regression header, when you specify both, e.g., -reg y x, robust mse1-.

This behaviour can only lead to problems and errors, because when you specify both mse1 and robust/cluster, Stata does not calculate inv(X’X) in e(V), but rather the standard robust/cluster robust variance is placed in e(V). Not only this, but you would not notice anything irregular from the regression table because the header of the regression table claims that MSE =1, whatever that might mean when you have in e(V) the robust variance placed.
The only way to detect that you are not getting what you asked for is to issue the calls sequentially and compare the results, e.g.
reg y x, mse1
matlist e(V)
reg y x, robust
matlist e(V)
reg y x, mse1 robust
matlist e(V)

See the code below:

. reg price mpg headroom if !missing(rep), mse1
 
      Source |       SS           df       MS      Number of obs   =        69
-------------+----------------------------------   F(2, 69)        >  99999.00
       Model |   123364948         2  61682473.8   Prob > F        =    0.0000
    Residual |   453432011        69  6571478.42   R-squared       =    0.2139
-------------+----------------------------------   Adj R-squared   =    0.2253
       Total |   576796959        68  8482308.22   Root MSE        =         1
 
------------------------------------------------------------------------------
       price | Coefficient  Std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
         mpg |  -243.1092   .0225501 -1.1e+04   0.000    -243.1542   -243.0643
    headroom |  -288.1991   .1550501 -1858.75   0.000    -288.5084   -287.8898
       _cons |    12186.4   .7998794  1.5e+04   0.000     12184.81       12188
------------------------------------------------------------------------------
 
. matlist e(V)
 
             |       mpg   headroom      _cons
-------------+---------------------------------
         mpg |  .0005085                      
    headroom |  .0013971   .0240405          
       _cons | -.0150173  -.1018658    .639807
 
. reg price mpg headroom if !missing(rep), mse1 robust
 
Linear regression                               Number of obs     =         69
                                                F(2, 66)          =       7.49
                                                Prob > F          =     0.0012
                                                R-squared         =     0.2139
                                                Root MSE          =          1
 
------------------------------------------------------------------------------
             |               Robust
       price | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
         mpg |  -243.1092   66.25903    -3.67   0.000    -375.3997   -110.8188
    headroom |  -288.1991    326.297    -0.88   0.380    -939.6722    363.2739
       _cons |    12186.4   2197.378     5.55   0.000     7799.194    16573.61
------------------------------------------------------------------------------
 
. matlist e(V)
 
             |       mpg   headroom      _cons
-------------+---------------------------------
         mpg |  4390.259                      
    headroom |  11389.16   106469.8          
       _cons | -134681.3  -566200.5    4828471
 
. reg price mpg headroom if !missing(rep), robust
 
Linear regression                               Number of obs     =         69
                                                F(2, 66)          =       7.49
                                                Prob > F          =     0.0012
                                                R-squared         =     0.2139
                                                Root MSE          =     2621.1
 
------------------------------------------------------------------------------
             |               Robust
       price | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
         mpg |  -243.1092   66.25903    -3.67   0.000    -375.3997   -110.8188
    headroom |  -288.1991    326.297    -0.88   0.380    -939.6722    363.2739
       _cons |    12186.4   2197.378     5.55   0.000     7799.194    16573.61
------------------------------------------------------------------------------
 
. matlist e(V)
 
             |       mpg   headroom      _cons
-------------+---------------------------------
         mpg |  4390.259                      
    headroom |  11389.16   106469.8          
       _cons | -134681.3  -566200.5    4828471

[ericmelse]

Dear Joro,

I suppose that this should be reported to 'tech support' to have a close look at.

Best,
Eric