Interaction terms in regression

[Tony Lee]

Hello everyone,

I am writing my BA-Thesis in Economics and hope somebody can help me out.

First let me explain the variables:

• tprofits = profit of participants
• class = 1 if participant was in the class treatment (not of interest here, control variable)
• prior_exp = 1 if participant had prior business experience
• mentorL_ba = 1 if mentor is in the lowest 25% of mentor experience
• mentorM_ba = 1 if mentor is in the 25% - 75% of mentor experience
• mentorH_ba = 1 if mentor is in the 75% - 100% of mentor experience
• i.wave, tprofits_b, $controls = control variabels

Here is my regression:

reg tprofits class i.prior_exp i.mentorL_ba#i.prior_exp i.mentorM_ba#i.prior_exp i.mentorH_ba#i.prior_exp i.wave tprofits_b $controls2, cluster(id)
Linear regression                               Number of obs     =      2,276
                                                F(21, 367)        =          .
                                                Prob > F          =          .
                                                R-squared         =     0.1283
                                                Root MSE          =     1706.2

                                           (Std. Err. adjusted for 368 clusters in id)
--------------------------------------------------------------------------------------
                     |               Robust
            tprofits |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
---------------------+----------------------------------------------------------------
               class |    93.1618   141.8026     0.66   0.512    -185.6859    372.0094
         1.prior_exp |   410.4867   124.8495     3.29   0.001     164.9765    655.9969
                     |
mentorL_ba#prior_exp |
                1 0  |   389.6264   263.7742     1.48   0.141    -129.0721    908.3249
                1 1  |   594.4702   224.8253     2.64   0.009     152.3627    1036.578
                     |
mentorM_ba#prior_exp |
                1 0  |   468.3452   257.7104     1.82   0.070    -38.42922    975.1196
                1 1  |   438.0457   202.1707     2.17   0.031     40.48736     835.604
                     |
mentorH_ba#prior_exp |
                1 0  |   1151.504    516.897     2.23   0.027     135.0521    2167.955
                1 1  |   206.6049   254.6565     0.81   0.418    -294.1641    707.3739

I want to know which mentor-mentee combination has the largest impact on the profit. I’m not sure what regression to use here as I'm still very new to STATA.
Alternatively I have created a single variable for mentor experience (mentor_exp=1 if low experience, mentor_exp=2 if medium experience, mentor_exp=3 if high experience) and run another regression.

reg tprofits class i.mentor_exp i.mentor_exp#i.prior_exp i.wave tprofits_b $controls2, cluster(id)
Linear regression                               Number of obs     =        618
                                                F(16, 123)        =          .
                                                Prob > F          =          .
                                                R-squared         =     0.1205
                                                Root MSE          =     2036.6

                                           (Std. Err. adjusted for 124 clusters in id)
--------------------------------------------------------------------------------------
                     |               Robust
            tprofits |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
---------------------+----------------------------------------------------------------
               class |          0  (omitted)
                     |
          mentor_exp |
                  2  |   16.18308   347.0116     0.05   0.963    -670.7051    703.0713
                  3  |   358.0511   484.6994     0.74   0.461    -601.3816    1317.484
                     |
mentor_exp#prior_exp |
                1 1  |    572.767   332.1417     1.72   0.087    -84.68714    1230.221
                2 1  |   392.4994   294.4023     1.33   0.185    -190.2519    975.2508
                3 1  |  -173.9383    454.761    -0.38   0.703     -1074.11    726.2332

And another were I changed the base category to see the effect of no prior experience.

reg tprofits class i.mentor_exp i.mentor_exp#b1.prior_exp i.wave tprofits_b $controls2, cluster(id)
Linear regression                               Number of obs     =        618
                                                F(16, 123)        =          .
                                                Prob > F          =          .
                                                R-squared         =     0.1205
                                                Root MSE          =     2036.6

                                           (Std. Err. adjusted for 124 clusters in id)
--------------------------------------------------------------------------------------
                     |               Robust
            tprofits |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
---------------------+----------------------------------------------------------------
               class |          0  (omitted)
                     |
          mentor_exp |
                  2  |  -164.0845   258.1515    -0.64   0.526    -675.0797    346.9106
                  3  |  -388.6542   285.2731    -1.36   0.176    -953.3348    176.0263
                     |
mentor_exp#prior_exp |
                1 0  |   -572.767   332.1417    -1.72   0.087    -1230.221    84.68714
                2 0  |  -392.4994   294.4023    -1.33   0.185    -975.2508    190.2519
                3 0  |   173.9383    454.761     0.38   0.703    -726.2332     1074.11

I hope somebody can help me find the regression of interest here.

Thank you in advance for your help.