Sensitivity analysis?

[David Dodemaide]

Hello,

I'm not sure if this is still active but I may as well give it a go!

I am modelling some wetland invertebrates, using various environmental variables as predictors. I have reached the point of model evaluation/sensitivity analysis but am running into a little hurdle. For a few of the predictors in two of my models I am getting sensitivity's greater than 1. From what I can understand of the analysis, the sensitivities should only range from 0 - 1. I am using the Sensitivity formula 1.

This may or may not be relevant but I ran a principal components analysis on the environmental variables first and am using the principal components as predictors.

Kind Regards,
David

[Bruce McCune]

David, Sure, sensitivities greater than 1 are possible. For example, if you had a sensitivity of 2, it would mean that nudging a predictor by a small amount would result in a response with twice the magnitude of the predictor, when both are scaled by their ranges. It is true, however, seem that sensitivities less than one are most common, simply because there is usually a fair amount of noise in the response. For a perfect straight-line relationship the sensitivity would be 1, but if the relationship is nonlinear, the proportionate change in the response can be greater than the proportionate change in the predictor.

Bruce McCune

--
You received this message because you are subscribed to the Google Groups "HyperNiche and NPMR" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hyperniche+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Daniel Palacios]

Hi everyone,

I'm glad this question was asked before. I had been struggling to understand how sensitivity can be greater than 1, and as I was about to post the question to the group I found Bruce's helpful answer at the very top!

Best,

Daniel

To unsubscribe from this group and stop receiving emails from it, send an email to hyperniche+...@googlegroups.com.