The majority of our decision-making is informal, barely structured with little explicit deliberation and made in ways of which we may be barely conscious; of course, the majority of our decision-making concerns things such as when to eat lunch or the route to take across a station concourse. Consideration of the consequences of different outcomes may be barely noticeable. The degree to which risks may be considered depends on how “lucky one feels”, which is an important motivator as to whether systematic approaches are used (risk aversion) or not (risk tolerant). In this case, risk aversion is less about fear and avoidance but more about determining that the consequence of a risk being realized cannot be tolerated. More systematic approaches may be something like needing to get to a meeting on time and considering the timeliness of different options for routes across the city. For significant decisions and in a professional context, we usually seek to make our decisions more formally, more “rationally” (“blind luck” is not an option), and in an auditable way, perhaps supported by some form of decision analysis and evidence. In groups we deliberate and seek to resolve differences of views. In organizations and governments there are formal governance rules and constitutions determining the decision processes, authorities and accountabilities; but alongside these, informal discussions inevitably take place, influencing the outcome.

That any form of decision analysis necessarily imposes some form of consistency and rationality upon the explicit modeling of the objectives and uncertainties is often not appreciated. Moreover, the consistency and rationality assumed by some approaches may contradict those assumed by others, making the use of some tools incompatible with the use of others. Thus users need to check that they not only understand but agree with the assumptions underpinning those methods that they adopt, or they may be misled by or misuse the results. A very important aspect that distinguishes methodologies of decision analysis is whether they seek to be objective or instead render subjective judgements explicit, taking into consideration diverse values and uncertainties. That said, the application of any method, whether its assumptions fit with the users' perspectives or not, stimulates discussion and focuses attention on understanding the issues, and that alone can be enormously beneficial.

Informal decision-making may not naturally satisfy many of the assumptions made by a decision analytic method. The simplistic response is that informal decision-making is about securing an outcome that may not be easily justified, and that formal decision-making embodies principles to ensure that important decisions are made soundly and rationally. However, in practice, informality and formality run side by side and can be more harmonious, making it not quite so easy to make such a ready distinction (Hodgkinson and Starbuck, 2008; Gregory et al., 2012; French and Argyris, 2018). At the individual level, Kahneman, Tversky and many others have investigated the differences between informal and formal decision making (Kahneman and Tversky, 1974; Morton and Fasolo, 2009; Kahneman, 2011; Montibeller and Winterfeldt, 2015). For many years this work was discussed under the general heading of heuristics and biases, recognizing that informal decision-making uses “quick and dirty” heuristics to make choices, but at the risk of biasing choices on average away from what various principles of rationality would suggest. More recently, the terminology has changed to talking about:

Whether there is a true dichotomy here is moot and there are many other subtleties discussed in the literature (Shleifer, 2012; Evans and Stanovich, 2013). However, this simple distinction is sufficient for our purposes. Decision analysis seeks to encourage System 2 Thinking, helping decision-makers, their advisors and stakeholders each individually think through and reflect on the issues. However, it is easy in discussions and specifically in articulating probability and value judgements, to slip into System 1 Thinking. Better methodologies and tools have elicitation processes for nudging and challenging participants to think carefully and explicitly when giving judgements, but weaker ones simply take the responses and build them into the analysis.

Most decisions are made by groups and there are informal and formal aspects to their interactions and deliberation. Sometimes a decision is reached by simple discussion and consensus, or maybe an informal vote. Other times, “horse-trading” and other agreements can connect decisions (“… and you vote with me next time”). Less democratically, there may be a dominant leader who influences agreement with their views. Business, organizational and political/government decision making are bound by more formal governance structures and constitutions, which prescribe who can take part, what interactions are allowed, how stakeholders may have their voice heard, voting systems, etc. In many societies, decision-making has become more inclusive with stakeholders and the public being consulted formally (Bayley, 2008; Renn, 2008; Rios Insua and French, 2010). This is particularly true in the environmental domain in which many modern techniques of stakeholder engagement, public participation and deliberative democracy have been developed (Beierle and Cayford, 2002; Gregory et al., 2012). Alongside such inclusive deliberations, inevitably informal discussions are also influential. In businesses and organizations, these may be no more than “water-cooler” conversations; but the advent of social media has allowed much wider, often very influential discussion to take place for all types and scales of decision-making. To parallel the distinction between System 1 and 2 Thinking, French and Argyris (2018) have introduced the terminology:

Decision analysis is aimed primarily at supporting System 2 Societal Deliberation helping those charged with the responsibility of taking the decision to do so in an informed, auditable and explicit way. It should, however, recognize the information sources provided by System 1 Societal deliberation such as social media, from which the decision-makers can learn about stakeholder values and other perspectives on the issues, thus ensuring that they are aware of breadth and depth of the issues that they face. Indeed, decision analysis can help in using its tools to communicate the decision-makers' reasoning to their stakeholders, particularly in the formulation and implementation stages of decision-making (French et al., 2005; Morton et al., 2009). In more inclusive decision-making, decision analysis can articulate discussions between decision-makers, stakeholders and experts, drawing the System 1 Societal Deliberations into the formal System 2 process (Mustajoki et al., 2004; Gregory et al., 2012; French and Argyris, 2018). It is important to recognize that broad processes which support this transition to System 2 Thinking and Social Deliberation are context dependent and depend on the skills of the analysis teams rather than something that can be achieved in an almost mechanical way alongside the modeling, computations and analysis. Effective decision analysis requires many more diverse skills than some mathematical and algorithmic introductions to decision analysis suggest.

Decision analysis requires two forms of modeling. First, there is a need to model the external context and the physical issues being addressed; in our case, some climate change impacts. Such modeling is descriptive of the context and can be validated empirically if there are data available, though in many examples of risk management and mitigation, preventive actions are needed before full data may become available. Such models are constructs of two or more entities and the relationships and influences between them. In managing risks, the model may simply be: “if we choose to undertake this action to ameliorate the unacceptable risk, then these consequences will arise because of these reasons.” This model may be founded on the professional judgement of decision makers and their experts or more empirically-based. Options for actions may be further elaborated by alternative beliefs or observations relating actions, the system being managed, and the consequences. The model can be made more robust by assembling knowledge relating to:

These questions may be explored with heuristic models, network (pathway) models, statistical models, dynamic mathematical models or a mixture of types, depending on the available knowledge and data. Some approaches to decision analysis are limited to specific types of models, while others are more flexible. The robustness of a model for decision making is determined by the degree to which an action will be systematically chosen for the task and correctly ameliorate risks as expected. Box 2 illustrates the development of a model highlighting some connections of different parts of the management and Earth Systems in managing the risks of damage from flooding.

The second form of modeling related to the decision-makers' and stakeholders' beliefs, values and objectives. These are more subjective and do not allow empirical validation. Moreover, the modeling is not descriptive in the sense that beliefs, values and objectives exist fully and explicitly before the modeling process begins. Rather the process of elicitation helps the participants reflect on what they are truly seeking to achieve and constructs the detailed objectives for the analysis (Keeney, 1992; Lichtenstein and Slovic, 2006; French, 2021). This form of modeling is particularly focused on helping the participants move from System 1 Thinking toward System 2 Thinking as it generally introduces rationality conditions that help their values become more consistent. Such modeling is known as prescriptive rather than descriptive modeling. Comparing and deliberating on different prescriptive models may also be important in Social Deliberation if some stakeholders hold to dogma that conflicts with current established approaches. While good decision-making depends on sound empirical description of the context, the process of reaching a decision and consensual acceptance of the selected course of action are not necessarily helped by effectively informing some of the stakeholders that they are “simply wrong” (French and Argyris, 2018), so the deliberations around prescriptive models need to be carefully and sensitively facilitated.

There is much emphasis currently on evidence-based decision-making and we would certainly echo this, but with a careful interpretation. Evidence and the knowledge that it supports is often encoded in descriptive models. We only have direct observations about the past. Decision-making is about planning for the future and so to use observations, we must make judgements about its relevance to the future: do we believe that things will continue in this way or that? Moreover, in many cases relating to risk management, what evidence we have is partial, if indeed we have sufficient data to claim any validated evidence at all. This inevitably means that there is a tension between scientific advisors, who want more time to accumulate and validate evidence, and the decision-makers, who need to make urgent decisions to mitigate risks. Decision analysts need to appreciate this tension in managing the deliberations between decision makers, their scientific experts and stakeholders.

A tension exists between observations, evidence, models about the future and effective risk management related to climate change; what constitutes evidence in climate-related risk management? In the last three decades, the Drivers-Pressures-State change-Impact-Response (DPSIR) framework has grown to underpin evidence-based management (Patrício et al., 2016). At its heart, is the need to attribute change to a driving cause; attribution of climate change to greenhouse gas emmissions from human activity has been a central theme of the Intergovernmental Panel on Climate Change (Bindoff et al., 2013). The application of DPSIR is usually at small spatial scales (10,000 km² at most) (Patrício et al., 2016) with a view that impacts can be detected and, once detected, restoration would be possible within a similar time frame. Such an approach implies that failure to not have a significant impact can be easily detected and rectified. In climate change, this is equivalent to accepting that an overshoot of a target global mean temperature would not be a disaster and that the climate can be restored before disastrous effects would arise. Yet, we know that the effects of greenhouse gases emitted now will take many decades before their effects will be diminished. Managing this risk requires decisions on actions well in advance of observations demonstrating impacts. Timeliness for action in risk management is an important factor not usually associated with DPSIR analyses. Uncertainty in both descriptive and prescriptive modeling increases the risks of failure. Evidence, therefore, needs to comprise not only observations of the state of the system but consider their power for detecting change and attributing it to the causes, as well as the degree to which models can help manage future risks.

Lastly, we are aware that the prescriptive modeling and analysis used to support decision-making can be interpreted naively as algorithmic computations on simple, often linear models without concern for the wider processes that these calculations support. To be frank, one of us has seen many naïve analyses performed by quantitatively adept scientists which have not truly supported the socio-political processes that surround any complex decision and which consequently have not really informed the decision-makers and stakeholders. Effective decision analysis uses the prescriptive modeling to articulate the deliberations between the participants building a shared understanding of the issues and each other's perceptions and values. It is through that shared understanding that a decision emerges, not simply from the maximization of some objective function. So in the next sections we emphasize many such “softer” socio-political issues that need to be reflected upon and brought into the deliberations in developing appropriate decision analyses for a set of issues.

Decisions are affected by many contextual factors beyond the formality and constraints of the processes used (see French and Geldermann, 2005; Hodgkinson and Starbuck, 2008; French et al., 2009 for a wider discussion of contextual issues) and can be explicitly incorporated in the analyses leading to and supporting decisions. Some important issues to consider are:

Box 2 develops these concepts as part of a control system, using a simplified example of managing the risks associated with flooding based on a model of how the management and Earth system might work. The contextual issues that arise in mitigating and adapting to climate change are important in determining suitable decision processes and tools. Firstly, the threatened impacts are global, but with many regional and continental differences in the scale and type of impact. Different areas of societal and business activity will be affected differently. A wide range of stakeholder interests will need to be considered in almost all cases, including those of humanity itself. There are ethical and moral issues to consider alongside more prosaic objectives. The breadth of uncertainties, many of which are deep and difficult to assess, is huge. While less so than in previous decades, climate science is controversial, and consensus will be hard to achieve across all stakeholders.

Decision making is driven by values, by what the decision-makers want to achieve. Values are necessarily subjective, but in societies that seek to avoid explicit subjectivity in their decision-making, economics and financial theory provide ways of costing many climate change impacts in a broadly objective manner; but there are some “intangible” impacts such as the loss of a historic site or natural estate, or the cultural impact of moving communities that are difficult to cost. The ability to assess intangible impacts, albeit subjectively, is one of the characteristics that distinguish different schools of decision analysis.

Although we have discussed contextual issues and uncertainties first, good decision making in practice follows value-focused thinking. Keeney (1992) describes this as “first deciding what you want then figuring out how to get it.” This runs counter to the more usual alternative-focused approach: namely first identifying some alternatives, then deciding between them. However, value-focused decision making is more creative, not being confined to a set of pre-defined alternatives. Moreover, being aware of the objectives of a decision analysis at the outset means that analytic effort can be focused on what matters, avoiding irrelevancies and providing the means by which any options can be evaluated for their contribution to a solution. In particular, since climate change, environmental and economic models can be very computationally expensive to analyse, value-focused thinking can direct effort to calculating what matters; the scale of the effort required to address the problem can be more easily identified.

Economic and financial methods provide one way of exploring values and objectives: e.g. cost-benefit methods seek for each alternative to evaluate the total cost of implementation and consider it relative to the cost of potential impacts that the alternative ameliorates (Boardman et al., 2017; OECD, 2018). More generally, value and utility methods offer ways of assessing both tangible and intangible costs and benefits, albeit relying more of subjective or, at least, less objective inputs (Keeney, 1992; Keeney and Raiffa, 1993; Bedford et al., 2005).

In passing, we note that many perspectives on rational decision-making separate the Science from the Values that need to be balanced in making a choice. By “Science” we mean the knowledge and investigations that can be brought to bear on resolving the issue and addressing the uncertainties. By “Values”, we mean the decision-makers' objectives that the ultimate choice seeks to meet. Of course, in any democratic society addressing climate change issues, the decision-makers should draw stakeholders' objectives into the ones that they use in the analysis (Renn, 2008; Rios Insua and French, 2010).

Almost every writer on decision analysis has summarized the decision-making process as a cyclic iteration of several phases (see the many citations to decision analysis in this paper). Here we use three broad phases, which are adequate for our purposes:

Figure 5 relates the three phases to the use of data from the real world, and choosing from available options to meet the policy objectives. The left-hand side of the graphic corresponds to the discussions, deliberations, analyses and studies that support the decision making. The right-hand side relates to the real world, which is always too complex to be perfectly modeled or analyzed. We emphasize that the real world includes not just relevant changes in climate, but human society, the environment, business, industry and agriculture and all the systems that need be considered in developing policies in mitigation and adaption.

Generally, the three phases of decision-making proceed from the top to the bottom of the graphic and are indicated by the bulleted lists, but we recognize that analyses, discussions and deliberation will iterate backwords and forwards as understanding of the issue grows. The “decision-maker” at the bottom of the graphic is to be understood as the person or, more likely, group, who are responsible and accountable for the decision under the appropriate governance structure. We emphasize that the interactions of this decision-maker with the real world include appropriate consultations and engagement with stakeholders.

This apparently linear approach from problem to decision implies risk management is organized, that all risks are identified and the processes set in train are carried out with some order, including the monitoring of success. Yet this is obviously rarely the case. Invariably, the process iterates within and between phases as thinking about one issue catalyses further thoughts about other issues or reflections during one phase indicate that other issues should have been considered in an earlier one. Further, problems may be latent, arising at seemingly random times, decisions postponed, and attention of scarce resources diverted to other purposes some way during the process. Moreover, many risks will be interrelated, and will be dealt with on differing timelines and urgencies. Being mindful of these relationships between risks and between their management process can help reduce tensions between them, take advantage of synergies in activities and processes, and avoid inadvertent negative consequences between risks.

Simplistically, decision analytic studies tend to be conducted in one of two modes (Franco and Montibeller, 2010).

This rough dichotomy is an oversimplification; many studies involve elements of both. Large projects dealing with complex issues, and integrating across related risks, may begin with several facilitated workshops to explore and identify issues, creating a series of questions. These questions are then explored through sophisticated modeling studies carried out in the expert mode. Later, there may be a return to facilitated modeling to share what has been learnt and evaluate possible strategies, providing guidance to the decision-makers. Some or all of the workshops might be conducted as face-to-face events or remotely (Coakes et al., 2002; French et al., 2009; Nunamaker et al., 2014; Pyrko et al., 2019).

Decision analyses comprise many families of techniques, some with sufficient philosophical and methodological underpinnings to be called a “school”; while others are more collections of techniques with enough common qualities to be grouped together. We have categorized seven broad classes of techniques that support decision making and give details of each approach in Table 3, identifying how they relate to the general considerations in our earlier discussion.

No “one-size-fits-all” tool is available for managing every climate risk or, indeed, managing the same risk but in different contexts, urgencies or availabilities of resources. This section aims to provide a means by which a climate risk manager may appraise the value of different analytic techniques for their situation. We encourage a prospective user of these techniques to consider the nature of the control system they are dealing with, such as described in the box, the Cynefin context in which they find themselves, and the types of uncertainties most conspicuous in their case. Table 3 can then be used to assess the appropriateness, or not, of different groups of techniques described above. The Table lists the various forms of decision analysis, indicating how they manage uncertainties, how they may be used in the different Cynefin contexts, how they fit into the different phases of decision-making and the resources needed in each use. In order to dig deeper into whether an approach may be suitable, citations are given to relevant literature to support our comments. In addition, we cite some relevant case studies in the application of the tools to climate-related risk management. We make no claims of exhaustiveness, and recognize that in identifying these characteristics we are making many subjective choices, but we hope that they offer a constructive guide into the literature that may help problem-owners and analysts find tools potentially valuable in their context.

While once-intractable, Bayesian Methods have made huge strides becoming computationally tractable and transparent to non-specialist users since the last century (Edwards et al., 2007). Moreover, developments in elicitation can be used to address behavioral and cognitive issues that can bias judgemental inputs (Dias et al., 2018; Turkman et al., 2019; Hanea et al., 2020). Many of the other methods evolved before these advances. Thus, Bayesian ideas should not be dismissed on those grounds; the main issue in using them is that they are explicitly subjective, emphasizing transparency, consensus, impartiality, and correspondence to observable reality instead of objectivity (Gelman and Hennig, 2017). Different cultures recognize and value subjectivity and objectivity differently. Some demand that subjective judgements are recognized explicitly, while others only acknowledge objective issues explicitly.

Decisions are based on analyses of the knowledge and information at hand to the decision-maker. The context of the decision process described above influences what can be done in each phase of decision making. Figure 6 illustrates how knowledge and uncertainty of the different subsystems in the example control system for managing the risk of consequences of flooding (Figure 3) influence the Cynefin space that the management problem may fall within, as well as indicating the analytic techniques that may be available.

Decision analyses used to support decision-making on climate-related risks shown as case studies in Table 3 were assessed for the circumstances in which they were used. The first dimension of the assessment was the geo-political scale to which the decision was intended to apply—household (or individual), community (village or neighborhood), city (including the greater city jurisdiction), sub-national region (a state, province), nation, trans-national regions (within a continent), international (through global agreements, organizations and the like). This scale differs from the type of body making the decisions, which is reflective of whether the outcome is intended as top-down, autonomous, or bottom-up. Here, a top-down decision is one that applies from a body that is autonomous at higher geopolitical scale to lower scales, whereas a bottom up decision is one made by a body autonomous at lower scales intended to influencing high scale outcomes.

The second dimension relates to the contribution of the technique to decision outcomes. These contributions relate to phases in the decision process but, as described previously, may not be implemented in a set sequence. The types of contributions include:

The results of this assessment are shown in Figure 7. Evidence of the basis of actual decisions and whether decision-analytic techniques were used to support the making of those decisions is difficult to find in the peer-reviewed literature. Most of the case studies were related to theoretical studies with realistic data, reflecting that most literature on climate change is about scenarios and the consequences of those scenarios. Many fewer studies address the actual decision processes of managing climate-related risks. Moreover, the spectrum of different types of contribution to the decision process seem more focussed at subnational/national levels.