Dear Statalisters,

I have data of 1323 firms, from 2014-2018 and within 30 industries, my data is unbalanced in which some firms don’t have data for some years. I got a total of 5,108 observations. I have run the below command to set panel data.

Code:
. xtset firmcode year
       panel variable:  firmcode (unbalanced)
        time variable:  year, 2014 to 2018, but with gaps
                delta:  1 unit
Previous research on similar topic used Random Effect to run regression. However, followed the suggestion of Hausman test, the Prob>chi2=0.0000 indicated that Fixed effect model is better for my model.

Code:
. xtreg roa_w fnsub indb bsize oeb c.fnsub##c.indb c.fnsub##c.bsize c.fnsub##c.oeb rdratio fage fsize lev ib1.y
> eardummy ib1.inddummy,fe

. estimates store fixed

. xtreg roa_w fnsub indb bsize oeb c.fnsub##c.indb c.fnsub##c.bsize c.fnsub##c.oeb rdratio fage fsize lev ib1.y
> eardummy ib1.inddummy,re

. estimates store random

. hausman fixed random
    b = consistent under Ho and Ha; obtained from xtreg
            B = inconsistent under Ha, efficient under Ho; obtained from xtreg

    Test:  Ho:  difference in coefficients not systematic

                 chi2(12) = (b-B)'[(V_b-V_B)^(-1)](b-B)
                          =      131.79
                Prob>chi2 =      0.0000
                (V_b-V_B is not positive definite)
Before running Fixed effect model, I have run the following tests:
- VIF: to check the multicollinearity test, the lowest VIF = 1.04, the highest VIF= 1.82, both are less than 10, so my variables don’t have multicollinearity problem.
- Ramsey RESET test: for omitted variable, result Prob>F = 0.0000
- White’s test for Heteroskedasicity: Prob>chi2 = 0.0000
- Breusch-Godfrey LM test for autocorrelation: Prob>chi2 = 0.0000

There are 3 models for my study:
M1:
Code:
 xtreg roa_w fnsub rdratio fage fsize lev ib1.yeardummy ib1.inddummy,fe
M2:
Code:
 xtreg roa_w fnsub indb bsize oeb rdratio fage fsize lev ib1.yeardummy ib1.inddummy,fe
M3:
Code:
 xtreg roa_w fnsub indb bsize oeb c.fnsub##c.indb c.fnsub##c.bsize c.fnsub##c.oeb rdratio fage fsize lev ib1.yeardummy ib1.inddummy,fe
Y= roa (Return on assets)
X = fnsub (Number of foreign subsidiaries)
Moderating variables = indb (Independent board), bsize (Board size), oeb (Oversea experience board)
Control variables = rdratio (R&D ratio), fsize (Firm size), fage (Firm age), lev (Leverage ratio), inddummy (Industry), yeardummy (Year),

The output result of Model 3 (with interaction term), it doesn’t show the coefficient of roa_w and fnsub, and this coefficient is the main point of this study. Then I switch to run the Random Effect model, still I don’t get the coefficient of roa_w and fnsub.

Fnsub (Number of foreign subsidiaries) is omitted because it doesn’t change across time, but sometimes it changes for some years when the firm increases its subsidiary's number. Previous research had balanced penal data, and use a random effect model.

Please help to suggest what Model I should use. I am not sure whether the code that I enter to run the regression for data is correct or not. I am so confused with the dummy variables (year and industry). I am new to Stata! Thank you very much. I really need your help!

Below is the result of fnsub omitted!

Code:
. xtreg roa_w fnsub indb bsize oeb c.fnsub##c.indb c.fnsub##c.bsize c.fnsub##c.oeb rdratio fage fsize lev ib1.year
> dummy ib1.inddummy,fe
note: fnsub omitted because of collinearity
note: indb omitted because of collinearity
note: fnsub omitted because of collinearity
note: bsize omitted because of collinearity
note: fnsub omitted because of collinearity
note: oeb omitted because of collinearity
note: 0.inddummy omitted because of collinearity

Fixed-effects (within) regression               Number of obs     =      5,108
Group variable: firmcode                        Number of groups  =      1,323

R-sq:                                           Obs per group:
     within  = 0.1867                                         min =          1
     between = 0.0730                                         avg =        3.9
     overall = 0.0940                                         max =          5

                                                F(12,3773)        =      72.16
corr(u_i, Xb)  = -0.5890                        Prob > F          =     0.0000

---------------------------------------------------------------------------------
          roa_w |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
----------------+----------------------------------------------------------------
          fnsub |  -.0032305    .001998    -1.62   0.106    -.0071478    .0006867
           indb |  -.0188497   .0254971    -0.74   0.460    -.0688392    .0311398
          bsize |   .0001165   .0011576     0.10   0.920    -.0021531    .0023861
            oeb |  -.0245969   .0064587    -3.81   0.000    -.0372597    -.011934
          fnsub |          0  (omitted)
           indb |          0  (omitted)
                |
 c.fnsub#c.indb |   .0041808   .0031404     1.33   0.183    -.0019762    .0103379
                |
          fnsub |          0  (omitted)
          bsize |          0  (omitted)
                |
c.fnsub#c.bsize |   .0000774    .000132     0.59   0.557    -.0001813    .0003362
                |
          fnsub |          0  (omitted)
            oeb |          0  (omitted)
                |
  c.fnsub#c.oeb |   .0015363   .0004951     3.10   0.002     .0005656     .002507
                |
        rdratio |  -.2935473   .0370724    -7.92   0.000    -.3662313   -.2208634
           fage |  -.0924686   .0092464   -10.00   0.000     -.110597   -.0743402
          fsize |   .0365267   .0021533    16.96   0.000     .0323051    .0407484
            lev |   -.176563   .0079622   -22.18   0.000    -.1921737   -.1609523
    0.yeardummy |   .0020265   .0019373     1.05   0.296    -.0017717    .0058247
     0.inddummy |          0  (omitted)
          _cons |   -.384708   .0416697    -9.23   0.000    -.4664053   -.3030107
----------------+----------------------------------------------------------------
        sigma_u |  .05817084
        sigma_e |  .03557531
            rho |  .72779517   (fraction of variance due to u_i)
---------------------------------------------------------------------------------
F test that all u_i=0: F(1322, 3773) = 4.73                  Prob > F = 0.0000
Code:
(1) (2) (3) (4) (5) (6)
roa_w roa_w roa_w tobinq_w tobinq_w tobinq_w
fnsub -0.000458* -0.000451* 0 0.00612 0.00611 0
(0.020) (0.022) (.) (0.280) (0.281) (.)
rdratio -0.290*** -0.290*** -0.294*** -6.662*** -6.658*** -6.682***
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
fage -0.0936*** -0.0941*** -0.0925*** -9.418*** -9.428*** -9.429***
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
fsize 0.0362*** 0.0364*** 0.0365*** -0.516*** -0.512*** -0.509***
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
lev -0.178*** -0.177*** -0.177*** -3.051*** -3.045*** -3.046***
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
0.yeardummy 0.00187 0.00206 0.00203 1.836*** 1.840*** 1.840***
(0.335) (0.288) (0.296) (0.000) (0.000) (0.000)
1.yeardummy 0 0 0 0 0 0
(.) (.) (.) (.) (.) (.)
0.inddummy 0 0 0 0 0 0
(.) (.) (.) (.) (.) (.)
1.inddummy 0 0 0 0 0 0
(.) (.) (.) (.) (.) (.)
indb -0.00460 0 -0.366 0
(0.837) (.) (0.570) (.)
bsize 0.000314 0 -0.00679 0
(0.754) (.) (0.815) (.)
oeb -0.0168** 0 -0.292 0
(0.005) (.) (0.089) (.)
c.fnsub#c.indb 0.00418 -0.00483
(0.183) (0.957)
c.fnsub#c.bsize 0.0000774 0.00333
(0.557) (0.382)
c.fnsub#c.oeb 0.00154** 0.00274
(0.002) (0.848)
_cons -0.386*** -0.386*** -0.385*** 39.83*** 40.02*** 40.09***
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
N 5108 5108 5108 5108 5108 5108
R2 0.183 0.184 0.187 0.446 0.447 0.447
adj. R2 -0.105 -0.103 -0.101 0.252 0.252 0.251
F 140.6 94.85 72.16 507.5 338.8 254.0
p-values in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001