INSTITUTIONAL DECISION-MAKING HEURISTICS

The city of London has a problem. The River Thames flows through central London and, like many rivers, the Thames is prone to flooding. Currently, the only thing preventing London from going the way of Atlantis is an estuary wall along the Thames. With climate change leading to higher and more unpredictable water levels, the city of London needs to build a taller wall in order to avoid catastrophic flooding.

The Thames Estuary 2100 Project set out to answer the following question: how high should the estuary wall be built in order for it to last until the year 2100? Building the wall too low could lead to catastrophic flooding, but building it too high would unduly burden taxpayers. The Project’s task was made more difficult by the fact that our best scientific models of climate change do not deliver specific and reliable guidance to the city. Leading general circulation models struggle to make predictions over such a long time scale and are not reliable guides to local climate (Frigg et al. [2015]; Thompson et al. [2016]).

The problem facing the Thames Estuary 2100 Project is an example of decision-making under deep uncertainty. Under conditions of deep uncertainty, standard econometric and decision-theoretic models of rational choice begin to give advice that is unreliable or unhelpful, so decision makers must devise new methods for making decisions that will perform satisfactorily over a wide range of plausible future scenarios. A suite of novel methods for decision-making under deep uncertainty have been developed by risk analysts, decision analysts, and government planners, among others, to tackle institutional challenges under deep uncertainty.

For example, we might guide policy choice using the method of robust decision-making (Lempert et al. [2003]). We could build a system model of the target system and a set of policy alternatives to choose between. Perhaps we could model system features such as water levels and wall heights, letting policy alternatives be candidate wall heights. We could use exploratory computational models to construct a landscape of plausible futures across which policy alternatives can be evaluated. For example, we might evaluate wall heights across a range of changes in water levels and patterns of urban development.

We could then assess the performance of policy alternatives across the landscape of plausible futures using one or more decision-theoretic criteria, such as regret-avoidance or minimizing the probability of flooding. We could use this preliminary analysis to identify new policy alternatives that had not been considered, such as novel wall configurations. We could also use the preliminary analysis to revise the landscape of plausible futures in order to highlight decision-relevant differences between future scenarios. Analysis could repeat as before, until we are satisfied that we have a good understanding of relevant policy alternatives and the trade-offs between them.

How should institutional decision-making methods such as robust decision-making be understood? One option would be to interpret them as novel criteria of rightness for decision-making, which tell us what it means for institutions to act correctly under conditions of deep uncertainty. So understood, novel approaches to decision-making under deep uncertainty would be competitors to traditional theories of rightness, such as decision-theoretic specifications of consequentialism or deontological approaches.

Alternatively, we might interpret methods of decision-making under deep uncertainty as decision procedures, mental processes that agents use to select actions. Well-known decision procedures include heuristics such as satisficing or default choice, as well as non-heuristic processes such as calculating expected utilities. While, in principle, it is possible to defend any given decision procedure as a criterion of rightness, this is not always attractive—for example, based on the fact that agents perform fairly well in a given context by choosing default options, we would not want to defend defaultness as a novel criterion of rightness against traditional theories such as consequentialism.

In my article, I defend a heuristic interpretation on which methods such as robust decision-making are not merely decision procedures, but rather decision procedures of a familiar kind: institutional decision-making heuristics. Activities such as identifying policy alternatives and constructing and analysing landscapes of plausible futures are components of heuristic processes that institutional actors can use to make effective decisions under conditions of deep uncertainty.

In support of the heuristic interpretation, I argue that methods such as robust decision-making bear two characteristic marks of heuristic decision procedures. First, like paradigmatic heuristics, recently proposed processes of decision-making under deep uncertainty are frugal. They neglect relevant information and they fail to draw relevant inferences in order to highlight the most important information and inferences. Second, processes of decision-making under deep uncertainty are thickly procedural, showing detailed concern for elements of the decision-making process such as search, model construction, and model revision that are traditionally neglected by non-heuristic processes.

It has long been recognized that institutions, like individuals, often do and should use heuristics to make decisions. Sometimes institutions use the same heuristics that individuals use. However, this is not always a good idea. Institutions differ from individuals in many ways: they have different cognitive abilities, use different decision-making structures, and face different problems. For these reasons, most work on institutional decision-making heuristics has focused on identifying new heuristics that are well suited to the challenges facing institutional decision makers.

The dominant strand of research on institutional heuristics is the simple rules approach of Kathleen Eisenhardt and colleagues (Eisenhardt and Sull [2001]). The simple rules approach stresses the need for organizations to balance flexibility and efficiency in decision-making by developing firm-specific portfolios of simple decision rules. For example, Cisco applied the ‘75% rule’ of acquiring only companies with 75 or fewer employees, at least 75% of whom are engineers.

It should not be denied that simple rules are sometimes valuable. However, simple rules heuristics are less general than traditional heuristics in at least two ways. First, simple rules are limited in their domain of applicability, the problems to which they can be applied at all. Cisco’s 75% rule governs only the decision of whether to acquire a company. Second, simple rules are limited in their domain of rationality, the subset of their domain of applicability where it would be rational to use them. Companies in other industries might balk at Cisco’s 75% rule because they are less interested in hiring engineers.

That is not to say that companies do not or should not sometimes apply simple rules to make decisions. Simple rules have a valuable place in decision making. However, the analogy between simple rules and rules of thumb suggests that we should expect to find institutional heuristics that are more general than simple rules, being both applicable to, and rational in, a wide range of circumstances. 

I suggested that novel processes for decision-making under deep uncertainty, such as robust decision-making, should be viewed as heuristics for institutional decision-making. An important feature of these processes is that they are quite general. Robust decision-making has a broad domain of applicability: it can be applied to any problem that is sufficiently well understood to construct and computationally analyse the performance of alternatives over landscapes of possible futures. Robust decision-making may also have a broad domain of rationality: robust decision-making has been applied to problems as diverse as reinforcing the Thames estuary wall (Ranger et al. [2013]), planning Israel’s long-term energy future (Popper et al. [2009]), and managing municipal water infrastructure (Groves et al. [2015]).

In particular, I argue, processes such as robust decision-making are well adapted to many institutional decision-making situations because they address the specific needs of institutional actors that many traditional heuristics neglect. Institutions, as compared to individuals, have high cognitive abilities and an increased ability to bear deliberative costs; deliberate collectively rather than individually; learn rather than evolve heuristic strategies; face higher-stakes situations; and have a heightened need to explain their decisions.

Traditional heuristics for individual decision-making require limited cognitive abilities and impose low costs; are suitable for individual rather than collective deliberation; are often evolved rather than learned; are less-suitable for high-stakes situations; and do not lead always to explainable decisions. By contrast, I argue that processes such as robust decision-making make good use of institutions’ heightened deliberative abilities; adapt well to collective deliberation; are learned rather than evolved; are suitable for high-stakes situations; and lead to highly explainable decisions.

This is not to say that any one heuristic is suitable for all processes of institutional decision-making under deep uncertainty. A great number of complementary processes have recently been developed for institutional decision-making under deep uncertainty, and there are doubtless more to be found. However, insofar as processes such as robust decision-making are often applicable to, and the rational choice for, the problems faced by deeply uncertain institutional decision-makers, they provide a necessary supplement to less-general institutional heuristics such as simple rules.