Global sensitivity analysis with Input and output data only

I have a Matrix of input variables that were drawn from uniform distribution with in a specified range. My matrix is like [10000,30]. And I have several outputs or responses (13) of them. I want to pick one of the output at a time and calculate global sensitivity. All the examples i see refer to a so called model to generate outputs based on inputs but I do not have a simple model and I would like to use the matrix of data instead. Is that possible? For now I am computing corrcoef on matrix and using the R values in lieu of sensitivity coefficients. Not sure how they are related to the so called Global sensitivities. Please help. Thanks in advance.

Answer 1

[Sahil Jain]

Hi. From my understanding, sensitivity analysis determines how changes in input variables affect changes in output variables. To determine this, a relationship between the input and output variables (i.e. a model) is required. I do not know if any techniques for sensitivity analysis exist which do not require a model but as of now, no MATLAB functions exist for them.

[Pappu Murthy]

I understand what you are saying. But can't we get something out of the input/output matrix generated for 1000's of instances of inputs drawn with prescribed pdf. Some kind of variance based sensitivities. Not saying such thing exists or not in Matlab but may be it does.