the reason for the large coffienct

[Paris Rira]

Hi Profs and colleagues,

* Example generated by -dataex-. For more info, type help dataex
clear
input float mig_jump double(immi_sh firm_age) float(foreign_aff sector)

0 .0009228768042223834  38 1  7
0 .0009228768042223834  15 1 12
0 .0009228768042223834  11 1 12
0 .0065667845980023600  37 1  7
0 .0065667845980023600  44 1 13
0 .0065667845980023600  26 1  7
end

------------------ copy up to and including the previous line ------------------

Listed ... out of 107659 observations

reg  mig_jump immi_sh firm_age foreign_aff i.sector
. reg  mig_jump immi_sh firm_age foreign_aff i.sector

      Source |       SS           df       MS      Number of obs   =   107,659
-------------+----------------------------------   F(10, 107648)   =     30.20
       Model |  6.99315118        10  .699315118   Prob > F        =    0.0000
    Residual |   2493.0856   107,648  .023159609   R-squared       =    0.0028
-------------+----------------------------------   Adj R-squared   =    0.0027
       Total |  2500.07875   107,658  .023222415   Root MSE        =    .15218

------------------------------------------------------------------------------
    mig_jump | Coefficient  Std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
     immi_sh |   2.549468    .383791     6.64   0.000     1.797243    3.301693
    firm_age |   .0001465   .0000291     5.03   0.000     .0000894    .0002035
 foreign_aff |  -.0068377   .0021686    -3.15   0.002    -.0110882   -.0025872
             |
      sector |
          6  |  -.0068547   .0020272    -3.38   0.001    -.0108279   -.0028815
          7  |  -.0137221   .0012101   -11.34   0.000    -.0160939   -.0113504
          9  |   .0017617   .0016331     1.08   0.281    -.0014392    .0049626
         10  |   .0010783   .0035832     0.30   0.763    -.0059447    .0081014
         11  |  -.0072893   .0032724    -2.23   0.026    -.0137032   -.0008755
         12  |  -.0112979   .0017646    -6.40   0.000    -.0147564   -.0078393
         13  |  -.0103782   .0026364    -3.94   0.000    -.0155455    -.005211
             |
       _cons |   .0272185   .0026262    10.36   0.000     .0220711    .0323659
------------------------------------------------------------------------------

The coefficient of interest is 2.5 which is a pretty large number (much above the literature). I tried several ways to figure out the reason. i.e check for outliers, collinearity, and scales of variables. Nothing is wrong. Although there is endogeneity, after running 2sls with IV even coefficient becomes larger (11). (IV variable has been checked. its reliable). So please share your ideas what can be the reason of the large coefficient and what are the possible solution.
Reaaly appreciated.

Cheers,
Paris

[Rich Goldstein]

there is no variation within your -dataex- example on the outcome variable so that is not helpful

however, note that the values of immi_sh are tiny and that the coefficient is telling you about differences in the outcome for a one unit difference in immi_sh; however, at least as compared with your data example, a one unit change in immi_sh (say from 0 to 1) is enormous - try re-scaling the immi_sh variable so that a one unit difference is substantively meaningful

[Paris Rira]

Thank you so much Rich. I did imm_sh*100. The coefficient became 0.025 (makes sense).

reg  mig_jump immi_sh_new firm_age foreign_aff i.sector

      Source |       SS           df       MS      Number of obs   =   107,659
-------------+----------------------------------   F(10, 107648)   =     30.20
       Model |  6.99315153        10  .699315153   Prob > F        =    0.0000
    Residual |   2493.0856   107,648  .023159609   R-squared       =    0.0028
-------------+----------------------------------   Adj R-squared   =    0.0027
       Total |  2500.07875   107,658  .023222415   Root MSE        =    .15218

------------------------------------------------------------------------------
    mig_jump | Coefficient  Std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
 immi_sh_new |   .0254947   .0038379     6.64   0.000     .0179724    .0330169
    firm_age |   .0001465   .0000291     5.03   0.000     .0000894    .0002035
 foreign_aff |  -.0068377   .0021686    -3.15   0.002    -.0110882   -.0025872
             |
      sector |
          6  |  -.0068547   .0020272    -3.38   0.001    -.0108279   -.0028815
          7  |  -.0137221   .0012101   -11.34   0.000    -.0160939   -.0113504
          9  |   .0017617   .0016331     1.08   0.281    -.0014392    .0049626
         10  |   .0010783   .0035832     0.30   0.763    -.0059447    .0081014
         11  |  -.0072893   .0032724    -2.23   0.026    -.0137032   -.0008755
         12  |  -.0112979   .0017646    -6.40   0.000    -.0147564   -.0078393
         13  |  -.0103782   .0026364    -3.94   0.000    -.0155455    -.005211
             |
       _cons |   .0272185   .0026262    10.36   0.000     .0220711    .0323659
------------------------------------------------------------------------------

Now, I can interpret the results by -Percentage change- am I right?

[Rich Goldstein]

unless there is something you are not telling us (e.g., a variable is log-transformed), I don't see how this can be interpreted as percentage change - but I am not expert in the substance here so maybe something is implied by your language that I am not interpreting correctly

[Paris Rira]

In this estimation, the dependent variable is a dummy variable, the explanatory available is the share of immigrants, which now I multiplied by 100, and the control variable firm_age is the age of firms (inter numbers), forein_aff is a dummy variable and sectors are integers from 1-7.

But there is another estimation (I did not bring it here) in which the dependent variable is the log of the number of immigrants and the explanatory variable and control variables remain the same.

So I kindly ask you about your interpretation. Because I am not export as well.