# Ecolego Wiki

### Site Tools

sensitivity_analysis

# Sensitivity analysis

Sensitivity analysis can be performed on probalistic simulation outputs (see probabilistic simulation). After a probabilistic Monte Carlo simulation, sensitivity indices can be calculated between simulation inputs (parameters) and simulation outputs (blocks).

## Performing a sensitivity analysis

 Note The sensitivity analysis toolbox offers many sensitivity analysis methods and several additional charts.
1. Run a probabilistic simulation as explained in probabilistic simulation.
2. Create tornado charts or correlation tables to graphically display sensitivity indices. Seven different types of sensitivity indices are available in Ecolego.

## Sensitivity indices

Method When to Use
Pearson product-moment correlation coefficient (Pearson)Computes the correlation coefficient. Interesting when model is linearly depending on parameters.
Spearman's rank correlation coefficient (Spearman) Computes the ranked correlation coefficient. Interesting when model is linearly or monotonically depending on parameters.
Standardized Regression Coefficient (SRC) Sets up a regression model from the raw data. Interesting when model is linearly depending on parameters.
Standardized Rank Regression Coefficient (SRRC) Sets up a regression model from the ranked data. Interesting when model is linearly or monotonically depending on parameters.
Partial Correlation Coefficient (PCC) Computes the correlation coefficient taking into account the rest of the varying parameters. Interesting when model is linearly depending on parameters.
Partial Rank Correlation Coefficient (PRCC) Computes the ranked correlation coefficient taking into account the rest of the varying parameters. Interesting when model is linearly or monotonically depending on parameters.
EASI first order correlation coefficient (EASI) Quantifies the first order effects, if model is additive this will take into account the full variance.
XEASI Higher Order Sensitivity Indices(XEASI) XEASI indices can assist in model calibration and validation processes by revealing which parameters have the most substantial impact on model predictions.