I'm trying to solve a delivery problem which involves transportation of goods from a set of sources to a set of destinations within a time window. Moreover, I'm trying to check which transportation model is more efficient: 1) small number of large vehicles or 2) large number of small vehicles.

To check which way of delivering is most efficient I've created a mathematical formalization (MILP) where I'm trying to maximize the delivery efficiency (measured between 0% and 100%).

Finally, I've generated 30 datasets and solved (with CPLEX) each one 2 times: one with large vehicles and another with small vehicles. In the end, I've obtained 2 sets of solutions for the same delivery problem.

Now the questions:

  • Due to LP relaxation and other global search approaches (like branch and cut) is it valid to do statistical analysis over the results? For example: can I do a t-test to check if the mean efficiency of two approaches is significantly different from each other?

  • Can I consider CPLEX solver as a blackbox and just do hypothesis testing over the results?

  • Is there any article that applies some statistical analysis over the results obtained from a MILP?


You can and should do statistical analysis on this sort of experiment. (But unfortunately, many operations researchers [including me] fail to do so regularly.)

Anyway, this paper might be of interest: https://pubsonline.informs.org/doi/abs/10.1287/ijoc.

  • $\begingroup$ Thank you very for your response and the link that you've provided. Is there any specific reason for not doing statistical analysis? $\endgroup$ – CrzCki May 20 at 17:01
  • $\begingroup$ No good reason, IMO. This is something that our field should get better about. $\endgroup$ – LarrySnyder610 May 20 at 17:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.