# Calculating Correlations and Sensitivity Coefficients

1. May 31, 2014

### SFA10

I have a numerical simulation where I'm randomly varying an input parameter.

This results in variations in three output parameters: A, B and C. The output parameters can be assumed to have normal distributions, but are correlated to each other.

If I calculate the mean and standard deviation on the range of values of A, B and C is there any way I could estimate the change in the mean of B and C due to a small change in the mean of A, making some account of the correlations? I was thinking of an approach based on sensitivity coefficients, but I'm not sure that's appropriate.

Any help/advice would be very welcome! Thanks!

2. May 31, 2014

### harmonic_lens

It sounds to me like the only way you could make a valuable correlation would be if you ran your experiment again with slightly tweaked input parameter, giving you new outputs A', B', C'. Then I'd calculate mean and SD and whatever else you think is good for the new data, and then see the changes between correlated outputs (i.e. A vs A').

I have no idea about the order of magnitude for the relationship of the correlation, or if one could exist, but as an engineer who has often been presented with data that is "supposed to be useful", this is how I would define useful, meaning that the correlation between data is clear and not based on nested calculations if possible.