Value of Information - zward/Amua GitHub Wiki
Drawing on statistical decision theory, we can define the value of information (VOI) of new research as the expected improvement in our outcome of interest (e.g. social welfare) generated by the information. That is, gathering more information before a decision is made may potentially reduce the probability that a suboptimal decision is made, and therefore improve the expected outcomes.
Expected Value of Perfect Information (EVPI)
The expected value of perfect information (EVPI) provides an upper bound on the value of new research, and estimates the value of (simultaneously) eliminating all uncertainty around the model parameters. EVPI can be estimated from the output of a probabilistic sensitivity analysis (PSA). In Amua the user specifies the number of PSA iterations to run and EVPI will be estimated.
Expected Value of Partial Perfect Information (EVPPI)
The expected value of partial perfect information (EVPPI) estimates the value of removing the uncertainty around individual model parameters. It is calculated for each parameter as the expected difference between the value of the optimal decision based on perfect information about that parameter, compared with the value of the decision based only on prior information.
To calculate EVPPI Amua uses the algorithm described by Strong and Oakley (2013). This approach estimates EVPPI from output obtained from a PSA. While more efficient than a 2-level nested Monte Carlo procedure, the estimates of EVPPI are sensitivity to the number of bins used. Choosing too few bins will bias estimates of EVPPI downwards, while too many bins will bias EVPPI upwards. Amua therefore plots estimates of EVPPI over a range of various bin sizes so the user can assess how sensitive the estimates may be.
