All standard errors missing

[Neg Kha]

Hi everyone,

I am using the following specification:

 reghdfe     y_hat female_cw i.age  if random==0, absorb(kontor_handel#i.year) vce(cluster sign)

but when Stata shows me the table, there are no standard errors and hence no p-values.

Could someone please explain to me why is that?
kontor_handel#i.year is office by year fixed effects.

[Nick Cox]

I doubt that anyone can explain this except in broad terms. You don't explain the structure of your data, or show an example, or show any output from the command. But odds are that you fitted a model that has no degrees of freedom left to allow such diagnostics to be calculated.

The name y_hat could well mean that you are supplying the model prediction from some previous model as the outcome for this one, so that the variation of the data around that is not visible for this command to assess.

[Hemanshu Kumar]

Perhaps similar to what #2 is suggesting, my first instinct is that your dependent variable y_hat is probably an exact linear combination of at least some of the variables in your current model (perhaps the prediction from a model where the regressors were female_cw and/or i.age), so that there are no non-zero residuals in your current regression.

Consider, for instance:

clear
set obs 1000

set seed 12345

gen x1 = runiform()
gen x2 = runiform()
gen x3 = runiform()

gen y = 5*x1 - 2*x2

regress y x1 x2 x3

which gives:

      Source |       SS           df       MS      Number of obs   =     1,000
-------------+----------------------------------   F(3, 996)       =         .
       Model |  2353.70294         3  784.567646   Prob > F        =         .
    Residual |           0       996           0   R-squared       =    1.0000
-------------+----------------------------------   Adj R-squared   =    1.0000
       Total |  2353.70294       999    2.356059   Root MSE        =         0

------------------------------------------------------------------------------
           y | Coefficient  Std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
          x1 |          5          .        .       .            .           .
          x2 |         -2          .        .       .            .           .
          x3 |   1.37e-09          .        .       .            .           .
       _cons |  -1.63e-09          .        .       .            .           .
------------------------------------------------------------------------------

or take this example:

clear
set obs 1000

set seed 12345

gen x1 = runiform()
gen x2 = runiform()
gen x3 = runiform()
gen u = runiform()
gen y = 5*x1 - 2*x2 + u

qui regress y x1 x2
predict y_hat

regress y_hat x1 x2 x3

which produces:

. regress y_hat x1 x2 x3

      Source |       SS           df       MS      Number of obs   =     1,000
-------------+----------------------------------   F(3, 996)       =         .
       Model |   2389.7036         3  796.567867   Prob > F        =         .
    Residual |           0       996           0   R-squared       =    1.0000
-------------+----------------------------------   Adj R-squared   =    1.0000
       Total |   2389.7036       999   2.3920957   Root MSE        =         0

------------------------------------------------------------------------------
       y_hat | Coefficient  Std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
          x1 |    5.04563          .        .       .            .           .
          x2 |   -1.99396          .        .       .            .           .
          x3 |   7.30e-09          .        .       .            .           .
       _cons |   .4729168          .        .       .            .           .
------------------------------------------------------------------------------