Struggeling with esstab output for regression

[Boris Cornelis]

Hi guys!

I have a question might be extremely easy fix.
I am trying to create a regression table for several models that displays (R-squared and F for my OLS regressions and AIC en loglikelihood voor Logit regressions).

Only issue is this is how it is reporting it atm.

And this is the command I use (Maybe you guys can help me how I can fix it such that it actually shows F statistic with p value, and the log likelihood values

Would appreciate any help

[Andrew Musau]

estout is from SSC (FAQ Advice #12). If the P value is less than 0.001, then you will have a lot of "0.000" values in the P value row. It's just better to star the statistic, same as the coefficients. regress does not store the P value corresponding to the model's F-statistic, so you can use testparm to retrieve it. Here is an example that stars the F-statistic.

sysuse auto, clear
eststo m1: regress mpg weight disp
testparm *

estadd local Fstat = cond(r(p)<0.01,"`:di %5.3f `=e(F)''***", ///
cond(r(p)<0.05,"`:di %5.3f `=e(F)''**", ///
cond(r(p)<0.1,"`:di %5.3f `=e(F)''*",  "`:di %5.3f `=e(F)'''")))

esttab m1, stats(N r2 aic Fstat, labels(Observations R-squared AIC F))

Res.:

. esttab m1, stats(N r2 aic Fstat, labels(Observations R-squared AIC F))

----------------------------
                      (1)  
                      mpg  
----------------------------
weight           -0.00657***
                  (-5.63)  

displacement      0.00528  
                   (0.54)  

_cons               40.08***
                  (19.84)  
----------------------------
Observations           74  
R-squared           0.653  
AIC                 396.5  
F               66.785***  
----------------------------
t statistics in parentheses
* p<0.05, ** p<0.01, *** p<0.001

[Boris Cornelis]

estout is from SSC (FAQ Advice #12). If the P value is less than 0.001, then you will have a lot of "0.000" values in the P value row. It's just better to star the statistic, same as the coefficients. regress does not store the P value corresponding to the model's F-statistic, so you can use testparm to retrieve it. Here is an example that stars the F-statistic.

sysuse auto, clear
eststo m1: regress mpg weight disp
testparm *

estadd local Fstat = cond(r(p)<0.01,"`:di %5.3f `=e(F)''***", ///
cond(r(p)<0.05,"`:di %5.3f `=e(F)''**", ///
cond(r(p)<0.1,"`:di %5.3f `=e(F)''*", "`:di %5.3f `=e(F)'''")))

esttab m1, stats(N r2 aic Fstat, labels(Observations R-squared AIC F))

Res.:

. esttab m1, stats(N r2 aic Fstat, labels(Observations R-squared AIC F))

----------------------------
(1)
mpg
----------------------------
weight -0.00657***
(-5.63)

displacement 0.00528
(0.54)

_cons 40.08***
(19.84)
----------------------------
Observations 74
R-squared 0.653
AIC 396.5
F 66.785***
----------------------------
t statistics in parentheses
* p<0.05, ** p<0.01, *** p<0.001

Awesome! truly appreciate the effort, thanks!
And how could I now add log likelihood to that?

[Andrew Musau]

esttab m1, stats(N r2 aic Fstat ll, labels(Observations R-squared AIC F Log-likelihood))

[Boris Cornelis]

esttab m1, stats(N r2 aic Fstat ll, labels(Observations R-squared AIC F Log-likelihood))

Hi Andrew,
I still need your help: See image (is the code I have at the moment)

I tried to visualise the issue in the image below:

Only R squared is correct
AIC continues (also for the regression models for which it doesnt apply)
F statistic doesnt show for logit (which is good) but also only shows 1x for 1 model, whereas there are a lot more empty fields wr it should show.
Loglikelihood shows everywere (but it should only show for logit)

Could you please help me with this issue, rather urgent

[Andrew Musau]

AIC continues (also for the regression models for which it doesnt apply)
Loglikelihood shows everywere (but it should only show for logit)

Not so. Linear models can also be estimated using maximum likelihood, so you can compute a log-likelihood and AIC. Here is an example:

sysuse auto, clear
regress mpg weight disp i.rep78
glm mpg weight disp i.rep78

Res.:

. regress mpg weight disp i.rep78

      Source |       SS           df       MS      Number of obs   =        69
-------------+----------------------------------   F(6, 62)        =     21.30
       Model |  1575.69146         6  262.615243   Prob > F        =    0.0000
    Residual |   764.51144        62  12.3308297   R-squared       =    0.6733
-------------+----------------------------------   Adj R-squared   =    0.6417
       Total |   2340.2029        68  34.4147485   Root MSE        =    3.5115

------------------------------------------------------------------------------
         mpg |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      weight |  -.0062559   .0014844    -4.21   0.000    -.0092232   -.0032886
displacement |   .0071103   .0128045     0.56   0.581    -.0184855    .0327061
             |
       rep78 |
          2  |  -.6519631   2.797811    -0.23   0.817    -6.244708    4.940782
          3  |  -.5992772   2.577557    -0.23   0.817    -5.751741    4.553187
          4  |  -.6856857   2.625668    -0.26   0.795    -5.934321     4.56295
          5  |   2.069259    2.74007     0.76   0.453    -3.408062    7.546581
             |
       _cons |   39.03528   3.572599    10.93   0.000     31.89375     46.1768
------------------------------------------------------------------------------

. 
. glm mpg weight disp i.rep78

Iteration 0:   log likelihood = -180.88376  

Generalized linear models                         Number of obs   =         69
Optimization     : ML                             Residual df     =         62
                                                  Scale parameter =   12.33083
Deviance         =  764.5114402                   (1/df) Deviance =   12.33083
Pearson          =  764.5114402                   (1/df) Pearson  =   12.33083

Variance function: V(u) = 1                       [Gaussian]
Link function    : g(u) = u                       [Identity]

                                                  AIC             =   5.445906
Log likelihood   = -180.8837605                   BIC             =   501.9968

------------------------------------------------------------------------------
             |                 OIM
         mpg |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      weight |  -.0062559   .0014844    -4.21   0.000    -.0091653   -.0033465
displacement |   .0071103   .0128045     0.56   0.579     -.017986    .0322066
             |
       rep78 |
          2  |  -.6519631   2.797811    -0.23   0.816    -6.135572    4.831646
          3  |  -.5992772   2.577557    -0.23   0.816    -5.651197    4.452643
          4  |  -.6856857   2.625668    -0.26   0.794      -5.8319    4.460529
          5  |   2.069259    2.74007     0.76   0.450    -3.301179    7.439697
             |
       _cons |   39.03528   3.572599    10.93   0.000     32.03311    46.03744
------------------------------------------------------------------------------

.

F statistic doesnt show for logit (which is good) but also only shows 1x for 1 model, whereas there are a lot more empty fields wr it should show.

logit reports a likelihood ratio (LR) chi-square test statistic. So you should run the code that I show in #2 only after estimating a linear model. Why it does not display in your results, I cannot comment without a reproducible example.