What is variance-based sensitivity analysis?

Imagine you have a model with uncertain parameters and intend to perform a sensitivity analysis. In particular in combination with non-linear models, a local derivative-based sensitivity analysis does very likely not fit the purpose of the intended sensitivity analysis. Instead, for the sensitivity analysis of probabilistic models, a global sensitivity analysis is often preferred. A common choice is to conduct a variance-based sensitivity analysis, which is based on a decomposition of the variance of the model output. 

Variance-based sensitivity analysis does not characterize a single sensitivity measure but constitutes a collection of sensitivity measures that assess how model parameters (or groups of model parameters) contribute to the variance of the model output. This includes:

• the first-order sensitivity indices for factor prioritization
• the total-effect sensitivity indices for factor fixing