Interpret results

Larissa MM

Join Date: Dec 2018
Posts: 14

27 Apr 2019, 11:06

Hi, I am quite new in regression, still trying to learn. Initially, i use five independent variables, and after showing to my supervisor, he asked me to add more variables but as sensitivity check so I don't need to redo the whole thing.

I am wondering if it makes sense to add my additional three variables as sensitivity check, as I normally read papers that only change one variables or so when doing sensitivity check.

If yes, then how do I interpret the outcome of the new addition? Does that mean my model is robust? Thank you

Code:

 . reg mtd prof size tang growth liq dc1 dc2 dc3 dc4 i.industry i.year, vce(robust)

Linear regression                               Number of obs     =      4,820
                                                F(27, 4792)       =      80.96
                                                Prob > F          =     0.0000
                                                R-squared         =     0.3569
                                                Root MSE          =     .18635

------------------------------------------------------------------------------
             |               Robust
         mtd |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
        prof |  -.3832766   .0686802    -5.58   0.000    -.5179214   -.2486318
        size |   .0361722    .002791    12.96   0.000     .0307006    .0416438
        tang |   .1248675   .0183476     6.81   0.000     .0888977    .1608372
      growth |  -.0083867    .002782    -3.01   0.003    -.0138407   -.0029326
         liq |  -.0102962   .0036118    -2.85   0.004     -.017377   -.0032155
         dc1 |   .0615475   .0120496     5.11   0.000     .0379248    .0851702
         dc2 |   .0420181   .0074354     5.65   0.000     .0274413    .0565949
         dc3 |   .0213742    .008977     2.38   0.017     .0037751    .0389732
         dc4 |   .0781161   .0124258     6.29   0.000     .0537559    .1024763
             |
    industry |
       9991  |  -.0240652   .0212285    -1.13   0.257    -.0656827    .0175523
       9992  |    .044358   .0242483     1.83   0.067    -.0031799    .0918958
       9993  |   -.189325   .0237744    -7.96   0.000    -.2359337   -.1427163
       9994  |   .0256848   .0208052     1.23   0.217     -.015103    .0664726
       9995  |  -.1267229   .0254384    -4.98   0.000    -.1765938    -.076852
       9996  |    .069327   .0208692     3.32   0.001     .0284139    .1102402
       9997  |  -.0659017   .0222793    -2.96   0.003    -.1095794    -.022224
       9998  |  -.0668792   .0243165    -2.75   0.006    -.1145507   -.0192076
       9999  |  -.0184298   .0227901    -0.81   0.419    -.0631089    .0262493
             |
        year |
       2009  |  -.0585273   .0127497    -4.59   0.000    -.0835225   -.0335321
       2010  |  -.0834637   .0124864    -6.68   0.000    -.1079427   -.0589847
       2011  |  -.0741264   .0125177    -5.92   0.000    -.0986667    -.049586
       2012  |  -.0744085    .012312    -6.04   0.000    -.0985456   -.0502714
       2013  |  -.1149802   .0122686    -9.37   0.000    -.1390323   -.0909281
       2014  |  -.1233646   .0124957    -9.87   0.000    -.1478619   -.0988673
       2015  |  -.1275209   .0127533   -10.00   0.000    -.1525233   -.1025186
       2016  |  -.1244362   .0129507    -9.61   0.000    -.1498255   -.0990469
       2017  |   -.132976    .012769   -10.41   0.000    -.1580091   -.1079429
             |
       _cons |   .0279881   .0510031     0.55   0.583    -.0720014    .1279777
------------------------------------------------------------------------------

Code:

 . reg mtd prof size tang growth liq dividendpayout ntds intcoverage dc1 dc2 dc3 dc4 i.industry i.year, vce(robust)

Linear regression                               Number of obs     =      4,820
                                                F(30, 4789)       =      75.58
                                                Prob > F          =     0.0000
                                                R-squared         =     0.3601
                                                Root MSE          =     .18593

--------------------------------------------------------------------------------
               |               Robust
           mtd |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
---------------+----------------------------------------------------------------
          prof |  -.3903113   .0725805    -5.38   0.000    -.5326024   -.2480202
          size |   .0354715   .0028602    12.40   0.000     .0298643    .0410787
          tang |   .1397064   .0171498     8.15   0.000     .1060849     .173328
        growth |  -.0078057   .0028427    -2.75   0.006    -.0133787   -.0022327
           liq |  -.0104448   .0036926    -2.83   0.005     -.017684   -.0032056
dividendpayout |  -.0018741   .0010861    -1.73   0.085    -.0040034    .0002552
          ntds |  -.3816906   .1609167    -2.37   0.018    -.6971612     -.06622
   intcoverage |  -1.42e-06   1.22e-06    -1.16   0.246    -3.81e-06    9.78e-07
           dc1 |   .0564218   .0129878     4.34   0.000     .0309596    .0818839
           dc2 |   .0397018   .0076943     5.16   0.000     .0246175    .0547861
           dc3 |    .019007   .0089529     2.12   0.034     .0014553    .0365588
           dc4 |    .074056    .012708     5.83   0.000     .0491424    .0989696
               |
      industry |
         9991  |  -.0253018   .0210576    -1.20   0.230    -.0665843    .0159807
         9992  |   .0452068   .0240588     1.88   0.060    -.0019594     .092373
         9993  |  -.1899307   .0236659    -8.03   0.000    -.2363267   -.1435348
         9994  |   .0250902   .0206563     1.21   0.225    -.0154057    .0655861
         9995  |  -.1287157   .0254462    -5.06   0.000    -.1786019   -.0788295
         9996  |   .0628662    .020813     3.02   0.003     .0220633    .1036692
         9997  |  -.0576092    .022123    -2.60   0.009    -.1009805   -.0142379
         9998  |  -.0559507    .024707    -2.26   0.024    -.1043879   -.0075136
         9999  |  -.0166227   .0226275    -0.73   0.463    -.0609829    .0277375
               |
          year |
         2009  |  -.0591305   .0127057    -4.65   0.000    -.0840396   -.0342214
         2010  |  -.0841003   .0124556    -6.75   0.000     -.108519   -.0596816
         2011  |  -.0743277    .012466    -5.96   0.000    -.0987668   -.0498887
         2012  |  -.0741434    .012242    -6.06   0.000    -.0981434   -.0501435
         2013  |  -.1163413   .0121812    -9.55   0.000    -.1402221   -.0924605
         2014  |  -.1243697   .0123403   -10.08   0.000    -.1485625    -.100177
         2015  |  -.1280807   .0126523   -10.12   0.000     -.152885   -.1032764
         2016  |  -.1253472   .0128311    -9.77   0.000    -.1505021   -.1001923
         2017  |  -.1342228   .0126918   -10.58   0.000    -.1591047    -.109341
               |
         _cons |   .0459086   .0528146     0.87   0.385    -.0576323    .1494496

Tags: None

Carlo Lazzaro

Join Date: Apr 2014

Posts: 17671
#2

27 Apr 2019, 11:25

Larissa:
results are basically similar.
The main issue is that you've seemingly performed a fixed effect panel regression via -regress- that rarely outperforms -xtreg,fe- when it come to panel data regression.

Kind regards,
Carlo
(StataNow 18.5)
Comment
Larissa MM

Join Date: Dec 2018

Posts: 14
#3

27 Apr 2019, 11:36

Originally posted by Carlo Lazzaro View Post

Larissa:
results are basically similar.
The main issue is that you've seemingly performed a fixed effect panel regression via -regress- that rarely outperforms -xtreg,fe- when it come to panel data regression.

Hi Carlo, noted on the xtreg fe. I will rerun again. Meanwhile, if the results are similar, what can I say about my model? Can I say that the model is robust?
Comment

Carlo Lazzaro

Join Date: Apr 2014
Posts: 17671

27 Apr 2019, 11:48

Larissa:
if you mean that results basically does not change when you add other predictors, your original model seems robust.
One of the usual way to compare two (or more) OLS models is to take a look at their adjusted_R_sq (that gives an idea of the efficiency of the model).
Adjusted_R_sq is not reported by default if you run an OLS with non-default standard errors, but can be retrieved via -ereturn list- after -regress-, as you can see from the following toy-example:

Code:

use "http://www.stata-press.com/data/r15/union.dta"
. reg age black, cluster( idcode)

Linear regression                               Number of obs     =     26,200
                                                F(1, 4433)        =       3.36
                                                Prob > F          =     0.0670
                                                R-squared         =     0.0002
                                                Root MSE          =     6.4884

                             (Std. Err. adjusted for 4,434 clusters in idcode)
------------------------------------------------------------------------------
             |               Robust
         age |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       black |   -.227264   .1240422    -1.83   0.067    -.4704486    .0159206
       _cons |   30.49461   .0686782   444.02   0.000     30.35996    30.62925
------------------------------------------------------------------------------

. ereturn list

scalars:
            e(N_clust) =  4434
                  e(N) =  26200
               e(df_m) =  1
               e(df_r) =  4433
                  e(F) =  3.356776197760231
                 e(r2) =  .000244306956287
               e(rmse) =  6.488387438062084
                e(mss) =  269.5154303763993
                e(rss) =  1102914.096172636
               e(r2_a) =  .000206145428955
                 e(ll) =  -86169.55714742451
               e(ll_0) =  -86172.75795955812
               e(rank) =  2

      *<snip>*

That said, as robustness can have different meanings, I think that in your research report you should be explicit about that.
As an aside, I would take the -xtreg,fe- issue up with your supervisor.

Kind regards,
Carlo
(StataNow 18.5)

Comment

Larissa MM

Join Date: Dec 2018
Posts: 14

27 Apr 2019, 11:59

Originally posted by Carlo Lazzaro View Post

Code:

use "http://www.stata-press.com/data/r15/union.dta"
. reg age black, cluster( idcode)

Linear regression Number of obs = 26,200
F(1, 4433) = 3.36
Prob > F = 0.0670
R-squared = 0.0002
Root MSE = 6.4884

(Std. Err. adjusted for 4,434 clusters in idcode)
------------------------------------------------------------------------------
| Robust
age | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
black | -.227264 .1240422 -1.83 0.067 -.4704486 .0159206
_cons | 30.49461 .0686782 444.02 0.000 30.35996 30.62925
------------------------------------------------------------------------------

. ereturn list

scalars:
e(N_clust) = 4434
e(N) = 26200
e(df_m) = 1
e(df_r) = 4433
e(F) = 3.356776197760231
e(r2) = .000244306956287
e(rmse) = 6.488387438062084
e(mss) = 269.5154303763993
e(rss) = 1102914.096172636
e(r2_a) = .000206145428955
e(ll) = -86169.55714742451
e(ll_0) = -86172.75795955812
e(rank) = 2

*<snip>*

That said, as robustness can have different meanings, I think that in your research report you should be explicit about that.
As an aside, I would take the -xtreg,fe- issue up with your supervisor.

You explained clearly on the interpretation. I will ask him. Thank you so much Mr. Carlo.

Comment

Carlo Lazzaro

Join Date: Apr 2014

Posts: 17671
#6

28 Apr 2019, 03:34

Larissa:
Carlo is far enough!

Kind regards,
Carlo
(StataNow 18.5)
Comment

Announcement