0

I have 6 input variables, each of which is normally distributed. Can I use Sobol variance-based sensitivity analysis? I have read some articles where they said that input variables must have uniform distributions. Is that correct?

gung - Reinstate Monica
  • 132,789
  • 81
  • 357
  • 650
cemik88
  • 1
  • 2

1 Answers1

1

That is not correct.

Variance-based sensitivity indices can work for data distributed according to any probability distribution as long as you know what they are. And, of course, normally distributed ones.

You can find more about this, for example, in this article.

You want to use R for computation, I recommend you the sensitivity package.

PS: I am very interested to know in which articles you have seen that input variables must have uniform distributions

Pop
  • 1,446
  • 9
  • 21
  • As you seem very interesting in investigating sobol indices, I would like to ask you a little question : Can sobol indices be applied to discrete random varibales ? Thank you in advance . – Nizar Jan 19 '16 at 13:06
  • 1
    It can be applied to discrete **input** random variables without any change. – Pop Jan 19 '16 at 16:03
  • Thank you very much for your reply ! I would like to know if you have any references for this idea ( examples where sobol indices are applied to discrete input variables) ? Thank you again :) – Nizar Jan 19 '16 at 16:06
  • 1
    here a paper where it is stated that we can use Sobol with discrete inputs (table 1): http://onlinelibrary.wiley.com/doi/10.1002/psp4.6/pdf And here a paper in which it is actually done on a small example (section 5): http://www.informs-sim.org/wsc97papers/0261.PDF – Pop Jan 19 '16 at 16:10
  • 1
    OH HH Thank you again, why not adding your answer on [my question](http://stats.stackexchange.com/questions/190843/sobol-indices-for-discrete-variables) so I can marked it as an answered question. – Nizar Jan 19 '16 at 16:12