Marginal Effects from Interaction Terms in OLS

[Kristian Pal]

Hi,

This was previously posted on cross validated.

I have the following model, as you can see I have included an interaction between i.own for 10 dummies and innoscore which is grand mean centred as a way of making interpretation easier.

 
 regress logrdintassets ib(freq).own##c.innoscorecentred lowtech midlowtech midhightech hightech lemployees salestoassets firmage size HHI5 d_marketbased i.year, vce(cluster id)

I end up with 2 significant interaction terms and I run margins command after the regression and get the following estimates.
But I am unsure how to interpret this avarage marginal effect of the interaction between an indicator for a specific owner type and a point increase in from the mean of the centred variable? My dependent variable is log transformed as well.

. margins own, dydx(innoscorecentred) post

Average marginal effects                          Number of obs   =        821
Model VCE    : Robust

Expression   : Linear prediction, predict()
dy/dx w.r.t. : innoscorecentred

----------------------------------------------------------------------------------
                 |            Delta-method
                 |      dy/dx   Std. Err.      t    P>|t|     [95% Conf. Interval]
-----------------+----------------------------------------------------------------
innoscorecentred |
             own |
           Bank  |   .0738848   .0250683     2.95   0.004      .024463    .1233067
      Corporate  |   .0171173   .0259077     0.66   0.510    -.0339595    .0681942
         Family  |   .0175771   .0254447     0.69   0.490     -.032587    .0677411
      Financial  |   -.000999   .0323958    -0.03   0.975     -.064867     .062869
     Foundation  |    .116375   .0956555     1.22   0.225    -.0722089    .3049588
     Government  |   .0370958   .0477014     0.78   0.438     -.056947    .1311386
         Mutual  |  -.0688859   .0426249    -1.62   0.108    -.1529204    .0151486
 Private equity  |  -.0005615    .029716    -0.02   0.985    -.0591464    .0580234
         Public  |   .0116551   .0695254     0.17   0.867    -.1254135    .1487237
Venture capital  |   .2090188   .0353349     5.92   0.000     .1393564    .2786812
----------------------------------------------------------------------------------

Thanks for considering the question.

[Marcos Almeida]

In general, - marginsplot - will provide a friendlier perspective.

That said, I wonder whether by using "year" you wish to express information concerning a longitudinal study with repeated measures for some variables. If so, you may consider using a panel data analysis instead.

[Kristian Pal]

In general, - marginsplot - will provide a friendlier perspective.

That said, I wonder whether by using "year" you wish to express information concerning a longitudinal study with repeated measures for some variables. If so, you may consider using a panel data analysis instead.

Yes, I also used marginsplot, but didnt know how to include it here. Actually, I am using panel data. So, i take it I made a mistake to use i.year in this case. This is the marginsplot: