Random effects model with dummy variable as control variable

[Jihad El Yaagoubi]

Hello, everyone,
I run a study on the impact of CSR (Corporate Social Responsibility) on FP(Financial Performance|), N=28, T=5 years. The independent variable is a dummy variable which has the value 1 if the company is CSR labelled, and 0 if not. the dependent variable I have ROA. I have three control variables (size, risk and industry).
When I run a random effects model, Stata drop out an industry (for example I have IND1(chemicals), IND 2(technological) and IND 3 (food industry)), Stata gives the p-value for IND1and IND2 but IND3 is omitted. So how can I evaluate the significance of IND3?

[Carlo Lazzaro]

Jihad:
welcome to this forum.
Stata is as always right: it omits one of the level of the categorical variable to protect you against the dummy trap problem (https://en.wikipedia.org/wiki/Dummy_...le_(statistics)).
You can change your reference category via -fvvarlist- notation, but onel level will be always omitted.
For the future, as reminded by the FAQ, please post waht you typed and what Stata gave you back. Thanks.

[Jihad El Yaagoubi]

Thank you, Mr.Carlo
I typed in command: xtreg ROA CSR Risk Size IND1 IND2 IND3 IND4,re
And Stata 14 omits IND4. so in my interpretation of results, I should report the significance of variables. when I run another analysis by changing the reference category, I have p-value changed for the others. So what's the significance of IND4 in my model?

I have another problem, rho is 0. But R-sq overall = 0.2257. (I read that I should report the value of rho in a random effects model). That means my results aren't valid?

Thank you
Jihad.

[Carlo Lazzaro]

Jihad:
on of the first thing that I would do is reorganizing your IND1-2-3-4 variables in an unique categorical variable (see -help fvvarlist).
If your data are not in the -long- format yest, -reshape- them accordingly, as panel data analysis do prefer the -long- format.
Last but not least, your chances of getting (more) helpful replies are conditional on posting what you typed and what Stata gave you back via CODE delimiters (see the FAQ on this and other posting-related topics).
As a sidelight, please call me Carlo, as all on (and many more off) the list do. Thanks.

[Jihad El Yaagoubi]

Thank you Carlo. Im working on this. Next time I will share my results, Im very interesting in your opinion.
jihad

[Jihad El Yaagoubi]

Hello Carlo,
Here are my results and what I typed. I need to know if my results are valid and if I typed correctly the dummy variable.

 xtset Companyscode Year, yearly
       panel variable:  Companyscode (strongly balanced)
        time variable:  Year, 2012 to 2016
                delta:  1 year

. 
. tabulate Industry,generate(Ind)

   Industry |      Freq.     Percent        Cum.
------------+-----------------------------------
          1 |         35       25.00       25.00
          2 |         25       17.86       42.86
          3 |         50       35.71       78.57
          4 |         30       21.43      100.00
------------+-----------------------------------
      Total |        140      100.00

. list Industry i.Industry in 1/4

     +---------------------+
     |                   3.|
     | Industry   Industry |
     |---------------------|
  1. |        3          1 |
  2. |        3          1 |
  3. |        3          1 |
  4. |        3          1 |
     +---------------------+

 xtreg ROA i.CSR Risk Size ib2.Industry,re

Random-effects GLS regression                   Number of obs     =        139
Group variable: Companyscode                    Number of groups  =         28

R-sq:                                           Obs per group:
     within  = 0.6424                                         min =          4
     between = 0.7522                                         avg =        5.0
     overall = 0.6649                                         max =          5

                                                Wald chi2(6)      =     261.88
corr(u_i, X)   = 0 (assumed)                    Prob > chi2       =     0.0000

------------------------------------------------------------------------------
         ROA |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       1.CSR |   .1158301   .0779946     1.49   0.138    -.0370365    .2686966
        Risk |   .6491344   .0419502    15.47   0.000     .5669136    .7313552
        Size |  -.1009088   .0568764    -1.77   0.076    -.2123845     .010567
             |
    Industry |
          1  |   -.192214   .1219311    -1.58   0.115    -.4311945    .0467665
          3  |  -.1182151   .1472157    -0.80   0.422    -.4067525    .1703223
          4  |  -.1268181   .1254604    -1.01   0.312    -.3727161    .1190798
             |
       _cons |   1.028692   .4946718     2.08   0.038     .0591531    1.998231
-------------+----------------------------------------------------------------
     sigma_u |          0
     sigma_e |  .42755899
         rho |          0   (fraction of variance due to u_i)
------------------------------------------------------------------------------

.
As you see rho=0

Thank you very much

[Carlo Lazzaro]

Jihad:
what does -xttest0- after -xtreg,re- give you back?

[Jihad El Yaagoubi]

It gives me this

xttest0

Breusch and Pagan Lagrangian multiplier test for random effects

        ROA[Companyscode,t] = Xb + u[Companyscode] + e[Companyscode,t]

        Estimated results:
                         |       Var     sd = sqrt(Var)
                ---------+-----------------------------
                     ROA |   .5318008       .7292468
                       e |   .1828067        .427559
                       u |          0              0

        Test:   Var(u) = 0
                             chibar2(01) =     0.00
                          Prob > chibar2 =   1.0000

[Carlo Lazzaro]

Jihad:
no evidence of panel-wise effect.
Go pooled OLS with standard errors clustered on -panelid-.

[Jihad El Yaagoubi]

Ah ok I see. Is it because p= 1 so >0.05?
which command i should use : regress depvar indepvar ?
It would be a pooled OLS model through panel data ?

Thank you very much Carlo for your help

[Carlo Lazzaro]

Jihad:
in your case -xtest0- does not reject the null that Var(u)=0; that result implies that pooled OLS is the way to go to analyse your data.
However, you should acknowledge the panel structure of your data by clustering standard errors on -Companyscode- (ie, your -panelid-).
Hence, your code should be:

reg ROA i.CSR Risk Size ib2.Industry,vce(cluster Companyscode)

[Jihad El Yaagoubi]

Thank you very much Carlo. I will work on this and share it .. Many thanks

[Jihad El Yaagoubi]

Hello, again Carlo,
This is the results I obtained from the pooled OLS estimation, Are they valid? can I have reliable conclusions from it?

 xtset Companyscode Year, yearly
       panel variable:  Companyscode (strongly balanced)
        time variable:  Year, 2012 to 2016
                delta:  1 year

. 
. reg ROA i.CSR Risk Size ib2.Industry,vce(cluster Companyscode)

Linear regression                               Number of obs     =        139
                                                F(6, 27)          =       2.15
                                                Prob > F          =     0.0803
                                                R-squared         =     0.6649
                                                Root MSE          =     .43165

                          (Std. Err. adjusted for 28 clusters in Companyscode)
------------------------------------------------------------------------------
             |               Robust
         ROA |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       1.CSR |   .1158301   .0642641     1.80   0.083    -.0160289    .2476891
        Risk |   .6491344   .3023649     2.15   0.041     .0287329    1.269536
        Size |  -.1009088   .0477627    -2.11   0.044    -.1989097   -.0029079
             |
    Industry |
          1  |   -.192214   .0868874    -2.21   0.036    -.3704923   -.0139358
          3  |  -.1182151   .1022348    -1.16   0.258    -.3279836    .0915534
          4  |  -.1268181    .083513    -1.52   0.141    -.2981726    .0445364
             |
       _cons |   1.028692    .442875     2.32   0.028     .1199876    1.937396
------------------------------------------------------------------------------

.Thanks
Jihad

[Carlo Lazzaro]

Jihad:
at their face-value, your results make sense.
As I do not know your research field (corporate finance?) I cannot help you out on any substantive interpretation.

[Jihad El Yaagoubi]

So statistically they are valid? but I should relate it to the theorical background of corporate social responsibility ?
thank you very much Carlo. Your help was considerable