Dear Statalist users,
I am currently working on a study on how to increase complementor participation in platform ecosystems. Complementors is the count of third-party producers of complementary products over time for various platforms (panel), and my dependent variable. As the variance (roughly 275) is bigger than the mean (21.5), and based on a histogram of complementors, I concluded that a negative-binomial distribution would be best-suited:

I would like to analyse a three-way interaction between independent variables, two of which are ratios (Lbreadthcontent, Lpropexcl), one of which is a count (Lgameengines). I was hoping to follow a procedure similar to this one (computing margins at one standard deviations below and above means), but after reading Leitgöb’s 2014 presentation, I understood that it wouldn’t be as simple. Could anyone advise on how to correctly analyse threeway interactions with a count as dependent variable?

This is an example of the data I am using:
Code:
* Example generated by -dataex-. To install: ssc install dataex
clear
input int complementors float(Lbreadthcontent Lgameengines Lpropexcl instbase_ln) byte(genlaunch season) float(platformd Quarterly)
16         .  .         .         0 0 1 1 187
11  .6363636 13 .22857143 14.040285 0 0 1 188
 7 .54545456 13 .26923078 14.865415 0 0 1 189
13  .4545455 14 .04761905  15.15477 0 0 1 190
30      .625 17  .2413793  15.42962 0 1 1 191
22  .8461539 17 .07936508 16.031548 0 0 1 192
18        .5 18 .07692308 16.273754 0 0 1 193
21  .6363636 20  .1724138 16.442444 0 0 1 194
44  .7692308 21 .12962963 16.558687 0 1 1 195
23  .7692308 22  .1025641 16.792501 0 0 1 196
end
format %tq Quarterly
And the code (I use Stata 15 64 bit):
Code:
xtset platformd Quarterly
xtnbreg complementors c.Lbreadthcontent##c.Lgameengines##c.Lpropexcl instbase_ln genlaunch season ,fe
Many thanks in advance,
Best regards,
Vladimir Sobota