I understand the basics to Latin hypercube sampling, such as implemented by the algorithm LHSA mentioned in the book Design and Modeling for Computer Experiments.

But I'd like to make sure:

1, n samples for s independent variables from LHS are not necessarily uncorrelated with each other. 2, these s independent variables can have interactions in a variance based sensitivity analysis.

  • $\begingroup$ What exactly do you mean by "are samples from LHS necessarily independent of each other"? Generally, the point of design of experiments is to ensure the variables / conditions are independent of each other, & the independence of the data is an assumption. $\endgroup$ – gung - Reinstate Monica Apr 22 '15 at 17:47
  1. The s random variables are independent in most LHS implementations. The variables may have a non-zero sample correlation, but they are independent in the statistical sense.
  2. The model or computer experiment's dependent variable may include interactions between the LHS variables. You may also induce correlation or interaction by how you transform or draw each LHS variable (e.g. draw a sample for parameter 2 conditional on parameter 1).

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.