How might we improve decision making in the face of
uncertainty? I’ve thought about this a great deal throughout my career since
uncertainty exists, whether acknowledged or not, in all decisions concerning
proposed actions to protect water quality.
In past assessments (http://kreckhow.blogspot.com/2014/05/assessment-of-value-of-new-information.html ) of this issue, I have used decision
analysis as my prescriptive model for how to consider uncertainty in model
forecasts. That led me to focus on the value of new information that might
reduce uncertainty.
However, there is another way to think about uncertainty in
decision making, particularly when several management options are being
considered. To see this, consider the figure below.
The probability distributions in the figure represent the predicted outcome for two management options affecting the concentration of a key response variable for which lower concentration is better. The peaks of the distributions represent the most likely outcome; this is the only information that would be generated by a deterministic model. Thus, on the basis of a deterministic model prediction alone, the likely decision would be to select management option B, if the management objective is to reduce concentration.
However, note that the prediction for option A has far less
uncertainty than that for option B, even though option A is predicted to have a
most likely concentration (the peak of the probability distribution) that
exceeds the most likely prediction for option B. The uncertainty analysis
provides additional information crucial to the decision. To be specific, do we
want to select an option that has substantial nonzero probability (represented
by the portion of distribution B that exceeds the concentration covered by
distribution A) of exceeding the outcome predicted for option A, or do we want
to select an option (A) that has a likely predicted higher concentration than option
B, but has a lower probability of higher concentrations?
How might uncertainty differences arise between two
management options? If these options represent nutrient levels in a lake, for
example, option B may represent uncertain nonpoint source controls, while option
A may represent more certain point source controls.
Obviously, cost of pollutant control is an essential
component of decision making. So, cost of pollutant control may favor either option
A or option B, but that does not take away from the fact that uncertainty in
the concentration resulting from pollutant control also is useful information. Beyond
that, the distribution of costs and benefits to the constituent groups and
jurisdictions affected by the decision may be important.
One “take home” message from this hypothetical example is
that public sector decision making is complex. To add to this complexity, I am
suggesting that the uncertainty in the response to management actions is yet
another attribute that should be considered. So, should we ignore prediction
uncertainty because the issues are just too complex? Of course not! Public
sector decision makers can always choose how to weight the information
presented by their support staff. Indeed, they can choose to down weight
information on scientific prediction uncertainty. Yet that does not mean that
uncertainty no longer exists. As the public, and as decision makers, we lose if
decision-relevant information is not available for consideration. Uncertainty in
the impact of management decisions should be part of that decision-relevant
information.
