Journal on Policy and Complex Systems
Since the R of XLRM already refers to models and relationships , why do we need the meta models ? The reason is practical : once we allow for fundamentally different models of the problem , it affects “ everything ”: the uncertainty factors , the policy levers , the computational model , and the measures of effectiveness . Thus , in practice , it is useful to show these separately for each meta model . Fortunately , addressing just two or three substantially different views of the problem often suffices .
4.2 . Methods for Dealing with Model Uncertainty
Let us now ask about specific mechanisms for considering model uncertainty . We suggest the methods in Table 1 , all of which we have used . Some have led to different meta models , some to alternative formulations within a single meta model .
Table 1 . Methods for Addressing Model Uncertainty Method
Description
Scenarios Use scenarios to group alternative assumption sets . Example : strategic planning that describes the world as potentially unfolding by a no-surprises scenario or a scenario with technological breakthroughs and economic growth
Competitive Models
Bounding Models
Use competing models . Often , the most important disagreements or other uncertainties can be grouped into two or three clumps . This sharpens issues . Example : in contemplating coercive strategy , use alternative cognitive models of how the adversary reasons . Is the adversary fearfully attempting to deter or is the adversary contemptuous of others and planning aggression ? 6
Identify models that bound the range of not-implausible model-uncertainty consequences ( easiest if the issue has only one primary dimension ). 7 Example : in estimating outcome of a group ’ s internal debates , one model might assume an outcome compromising across factions ; another might assume that the outcome is the strident outcome of the winning faction .
Elicit diverse alternative models and consider all of them . Example : in the study of climate-change , consider results of models from different universities , government laboratories , and private institutions . 8
6 An early example used competing “ red agents ” in a simulation to study Cold War deterrence ( Davis , 1989 ). Later , competing models of Saddam Hussein helped anticipate Saddam ’ s 1990 aggression against Kuwait ( Davis & Arquilla , 1991 ; National Research Council , 2014 ). These efforts reflected cognitive biases and misperceptions . Later , competing models of Kim Jong Il were used to study possible negotiation options , although the study concluded that Kim was very unlikely ( across plausible models ) to truly give up his nuclear program ( Arquilla & Davis , 1994 ). Recent studies compare alternative social-science theories in agent-based modeling of social systems ( Gunaratne & Garibay , 2018 ; Hadzikadic & Whitmeyer , 2019 ).
7 The method was used for a model of public support for terrorism ( Davis & O ' Mahony , 2013 ) and in a study of heterogeneous fusion of information , bearing on whether an individual should be judged as a potential terrorist threat given inconsistent information ( some of it wrong or deceptive ) ( Davis , Perry , Hollywood , & Manheim , 2016 ).
8 See the Intergovernmental Panel on Climate Change ( IPCC ; 2010 , 2014 ).
188