decisionSupport

Decision-makers often wish to have a quantitative basis for their decisions, but many variables they have to consider cannot be precisely quantified, at least not without unreasonable effort. The major objective of (prescriptive) decision analysis is to support decision-making processes faced with this problem (Luedeling and Shepherd, 2016).

Decision analysis can make forecasts of decision outcomes without precise numbers, as long as probability distributions describing the possible values for all variables can be estimated. The decisionSupport package (Luedeling and Göhring, 2017) implements this as a Monte Carlo simulation, which generates a large number of plausible system outcomes, based on random numbers for each input variable that are drawn from user-specified probability distributions.

decisionSupport also conducts a sensitivity analysis (based on Projection-to-Latent-Structures regression) to highlight uncertain variables that have large impacts on model outputs (Luedeling and Gassner, 2012; Wold et al., 2001). If the distribution of predicted system outcomes doesn’t imply a clearly preferable decision option, these variables can then be targeted by decision-supporting research (Luedeling et al., 2015).

References :

Luedeling, E. and Gassner, A., 2012. Partial Least Squares regression for analyzing walnut phenology in California. Agricultural and Forest Meteorology, 158: 43-52. 

Luedeling, E. and Göhring, L., 2017. decisionSupport: Quantitative Support of Decision Making under Uncertainty, CRAN archive; https://cran.r-project.org/web/packages/decisionSupport/.

Luedeling, E. et al., 2015. Fresh groundwater for Wajir – ex-ante assessment of uncertain benefits for multiple stakeholders in a water supply project in Northern Kenya. Frontiers in Environmental Science, 3: Article 16, 1-18.

Luedeling, E. and Shepherd, K., 2016. Decision-focused agricultural research. Solutions, 7(5): 46-54.

Wold, S., Sjostrom, M. and Eriksson, L., 2001. PLS-regression: a basic tool of chemometrics. Chemom. Intell. Lab. Syst., 58(2): 109-130.

The ecology of the conifer forests of the American west is largely shaped by wildfires. Many ecologists recommend regular and low-intensity burns to manage the build-up of combustible understory materials (shrubs, fallen branches). Not all municipalities or regions implement these practices. Failure to follow recommended controlled burning may lead to the build-up of fire stock and increase the frequency and severity of large wildfires. Such fires have destroyed many western ecosystems in the recent past.

The code provided simulates the decision of forest managers to use controlled fires in conifer forests vs. running the risk of severe fire. This is done through the use of a Monte Carlo analysis with 10,000 model runs based on probability distributions for all uncertain variables. These distributions aim to represent the full range of possible values for each component of the model. For example, the cost of controlled burning may vary greatly with remoteness and topography of forests. Also the frequency at which these controlled fires are necessary may vary with weather events such as storms and droughts. There is often no ‘hard data’ for many such variables, which can paralyze the decision-making processes or lead decision-makers to conclude that large research efforts are needed, before a decision can be made. The decisionSupport package allows simulation of the costs and benefits of complex decisions based on the full range of values deemed plausible for all variables. This is useful for determining whether a clearly preferable course of action can already be delineated based on the present state of knowledge. It also allows identifying key uncertainties that should be reduced by decision-supporting research.