Models of Decisionmaking: Their Uses and Limitations

This paper critically surveys explicit models of actual decisionmaking. How are the models derived? How are the models used and evaluated? What conceivable value can the models have? Are there better practices for creating and using models of decisionmaking? Contact: Patrick D. Larkey Heinz School of Public Policy and Management Carnegie Mellon University Email: [email protected] Models of Decisionmaking: Their Uses and Limitations Patrick D. Larkey In his autobiography, Models of My Life (1991, p. 370), Simon describes how he acquired a research problem in 1935 at the age of 19 that sustained him intellectually for life. The empirical question posed to him was: How are funds divided between playground maintenance -planting trees, cutting grass, etc. -and playground activity leadership -planning and running programs -in Milwaukee? Pondering this somewhat mundane question about allocation behavior, Simon concluded that the ready hypothesis from his then-limited study of economics, “Divide the funds so that the next dollar spent for maintenance will produce the same return as the next dollar spent for leaders’ salaries”, was obviously deficient in two respects. First, no one seemed to be thinking about the decision in this way; the hypothesis failed descriptively. Second, he could not see how to weigh the value of one against the value of the other; the hypothesis failed prescriptively. For Simon, the specific problem about behavior on allocations to playgrounds in Milwaukee became a much more general problem: “How do human beings reason when the conditions for rationality postulated by neoclassical economics are not met?” In much of his subsequent research, Simon labeled the behavior of interest as decisionmaking, choice, or problem-solving interchangeably as synonyms for “reasoning.” From the discovery of this problem in 1935 until his death in 2001, most of what Simon published can be understood as contributing more or less directly to answering this question. “Decision” is a useful, albeit labile, frame for thinking about many instances of human behavior. It is used to organize our thinking about volitional circumstances from simple gambles to the use of nuclear weapons. In the modal application of the decision frame, an individual or group has knowledge and control of the available courses of action, estimates of the consequences of choosing any particular action, ordered preferences (values on estimated future consequences), and a necessity to act or not. Even where these concepts are not very descriptive of what precedes an action, they may be descriptive of the post hoc explanations for actions – rationalizations of actions. Jose Ortega Y Gasset once observed that “Living is a constant process of deciding what we are going to do.” Behavior that can potentially be framed as a “decision” is everywhere and always. You, for example, can be understood to have chosen to read the previous sentence and this one. In doing so, you allocated some time to this reading behavior and not to an infinite set of possible alternatives, some of which may have produced greater value than the reading. You can also be said to have chosen a time and place for reading this document. You can also be said to have made a series of larger decisions about the course of your life that make any reading about any aspect of decisionmaking an option at this moment. Now, about the next thirty minutes of your life and the need to be rational......... What we choose to hack out of the complex thicket of behavior in real time and call a “decision” is somewhat arbitrary. The decision frame is most persuasive and probably most useful for consequential, discrete choices that entail prior, conscious deliberations about the alternative futures as a consequence of particular actions. The decision frame is less persuasive and probably less useful for thinking about behaviors that are inconsequential, habitual, highly emotional, or socially deviant. Decisionmaking is an important function for leaders in public affairs, business and the professions. Leaders in all types of organizations have central roles in deciding what the “business” is (strategic decisions), how the business is conducted (operating and resource allocation decisions), who participates in conducting the business (personnel decisions), what those participating do in the conduct of the business (organization and tasking decisions), and how the value produced by the business is distributed to groups and individuals (distributive decisions). Leaders as human beings also have the usual array of personal decisions from the mundane, e.g., what to have for lunch today, to the significant, e.g., college or spousal choice, with everything in between. The “decision sciences” are the accumulated and accumulating knowledge across several fields with respect to three main questions: 1. How do human beings, individually and collectively, make decisions? 2. How should human beings, individually and collectively, make decisions? 3. How can human beings, individually and collectively, make better decisions? The literature associated with each of these questions is very large and diverse. The most relevant materials for the first question are found in psychology, organization theory, political science, economics, history, and ethics. The most relevant materials for the second question are the products of economists, statisticians, mathematicians, and philosophers. The most relevant materials for the third question are found in operations research, statistics, psychology, and organization theory. The extent to which much of the knowledge on decisionmaking is “scientific” is debatable. The epistemological evolution and issues are complicated. Much of the empirical work on the first question, how decisions are made, is not very rigorous. The bulk of the work, at least by volume, consists of interpretations of various historical “decisions” based on limited observation, self reports, and fragmentary records of nonexperimental behaviors. The behaviors examined range from the monumental to the mundane. While the details vary widely, our reasoning about human reasoning has a certain constancy across many diverse fields. Often there is an individual identified with more control than others. There is a tendency to treat observed effects as intended and behavior as goal-seeking. There is a fascination with obvious mistakes, Napoleon at Waterloo; Lee at Gettysburg; MacArthur at the Yalu River; Kennedy and the Bay of Pigs; Nixon and Watergate; Reagan and IranContra, balancing the budget and regulating the savings & loan industry; Clinton and Monica-gate; Ford and the Edsel; Coca Cola and the New Coca Cola, because the effects are not easily understood as intended and because these natural experiments gone awry suggest that we might learn something to improve future behaviors. The most scientific of the evidence on human decisionmaking, i.e., evidence with the highest internal validity, is from experimental and process tracing studies by psychologists of individual behavior in laboratory settings. This work is centered on behavior in structured decision tasks; it, has been extended to simple groups especially with regards to competition and cooperation. A very substantial strand of the experimental work on decisionmaking has tested rational models of behavior and posited “heuristics and biases” that explain subject departures from the predictions of the rational models. Much of the descriptive work on decisionmaking behavior that attempts to model explicitly decision agent behavior is due to Simon and his colleagues and students. Group and organizational decisionmaking poses significant methodological problems; most of the ostensibly scientific evidence is from surveys or observational studies with an intervention here and there. Most of the statistical and econometric work on data from natural experiments – individuals and organizations choosing -is confirmatory; it does not seriously test any theory of or rival hypotheses about human behavior; it does not predict choice or decision behaviors successfully. Regardless of the technical sophistication of the analyses and the persuasiveness of the supporting arguments, explanations of why consumers bought what they did or voters voted as they did are of the same intellectual genre as the reasons that the “talking heads” on the financial networks give for why the Dow Jones Average went up or down yesterday. Such commentaries may be entertaining and profitable for the providers, but we don’t learn much in the end of value to future decisions. The voluminous literatures on the second question, how should decisions be made, are primarily the product of mathematicians, statisticians, philosophers, and economists, although some ethicists, historians and novelists have chipped in. Most of the mathematical work on decisionmaking is normatively appealing, empirically false as a positive theory of human behavior where testable, and not pragmatically useful for actual decisions by real decisionmakers. The rational model in many guises is the most frequently utilized theory in the social sciences, especially in economics and the public choice brand of political science for how decisions are made. Individuals are said to be maximizing expected utility, whatever that is. Firms are maximizing profit, whatever that is. Politicians, government officials and voters are optimizing this or that. While some variant of rationality is a convenient postulate for aggregate analyses, rationality as a positive, predictive theory of behavior does not stand up well upon closer inspection. There is a large body of empirical work, the best of it by experimental psychologists, showing departures from rational behavior on a variety of decision tasks. Maximizing Subjective Expected Utility (SEU) in some form is probably the correct way, normatively speaking, to make a decision, any decision. If you know all of the actions available, can predict, perhaps with some error, the consequences of taking any combination of these actions, can predict, perhaps with some error, how you will value all possible consequences, SEU is logically an attractive way to proceed; alternative processes are hard to defend a priori or post hoc. The transaction costs associated with executing the SEU approach to decisionmaking – the costs of all that knowing, predicting and valuing – may, however, exceed the expected value from using SEU rather than some simpler heuristic approach. Every champion of SEU should at least once seriously and selfconsciously attempt to follow their own decisionmaking advice on a decision important to them personally – perhaps choice of spouse as a contribution to both population control and natural selection – to avoid the age-old trap of anything being possible for someone who does not actually have to do it. It is a truly sobering experience to try to follow the rational process to the letter on even seemingly simple choices. The third question, how can better decisions be made, is related to the second question in that it is normative in thrust but with an important difference. The second question is in principle normative while the third question is in practice normative. Work on the second question might advise, “maximize utility subject to constraints.” Work on the third question might couple this advice with feasible techniques for estimating the necessary probabilities, eliciting the necessary preferences, specifying the necessary constraints, and doing the necessary predictions and computations. In principle is one thing. In practice is quite another thing. Many of the tools for decisionmaking are found in Operations Research (Management Science/Industrial Engineering/Applied Statistics). The tools are quite powerful for making decisions where the criterion is clear, the context is stable, and measures exist. Such decisions are, of course, a small proportion of real decisions. Even where the tools are fully appropriate for decisions and competently applied, the effects of resulting information on the eventual decision in supra-individual settings may be slight; the leaders responsible for the decisions are usually not the analysts and frequently have a hard time understanding analytic arguments. Three inescapable conclusions from a survey of what we know about decisionmaking behavior are: (1) our understanding of how humans individually and collectively make decisions is very incomplete, if not wrong; (2) the technologies for decisionmaking developed thus far are either very general and difficult to apply or very specific and inapplicable to most real decisions; and (3) there are precious few explicit models of decisionmaking that attempt to simulate actual behavior. Most of the decisionmaking technologies that have been developed apply to structured choices with specified mutually exclusive and exhaustive alternative sets and estimable consequences and values. Most real decision problems have interdependent and incomplete alternative sets and unimagined consequences and values. There are no well developed technologies – techniques that could be taught as recipes with good effect -for finding and structuring decision problems. Asking the right questions is usually much more difficult than answering the right questions once posed. This paper critically surveys explicit models of actual decisionmaking. How are the models derived? How are the models used and evaluated? What conceivable value can the models have? Are there better practices for creating and using models of decisionmaking?