SENSITIVITY ANALYSIS IN THE PRESENCE OF CORRELATED PARAMETER ESTIMATES

Sensitivity analysis is usually conducted independently for each parameter in turn. This can give misleading results when parameter estimates are correlated. Such correlations can be accounted for by adjusting particular parameter values so that the new combination of parameter values lies on a contour of equal likelihood. Alternatively, all combinations of parameter values may be considered, subject to the constraint of equal likelihood, to find the combination which gives rise to maximum change to the model predictions. This combination indicates the direction in parameter space to which model predictions are most sensitive given the data used for parameter estimation, as well as indicating the potential magnitude of prediction error which may arise from the error in estimating this combination.