Dynamic Decision Making

Dynamic decision problems are those in which the decision maker may act repeatedly on an environment that responds to his or her actions and also changes independently over time, both endogenously and exogenously (Edwards, 1962), An example might be a senior manager's efforts to improve low morale in an organization. She may, over a period of months,

try a number of different interventions, scaling up successes and abandoning failures. Over the same period various fac­tors internal and external to the organization may also affect morale. Clearly, such problems set decision makers extra­ordinary challenges.

They have also proved difficult for researchers, partly be­cause of their inherent complexity and partly because of the experimenter's partial lack of control. Complexity implies difficulty in deriving optimal strategies. Lack of control arises from the fact that the problem facing the decision maker at time / is partially the consequence of his or her earlier deci­sions, as well as of the experimental conditions imposed. On the positive side, the growing availability of computers has helped both in the creation of realistically complex experi­mental environments and in the analysis of strategic alterna­tives. Some examples of the sorts of studies this allows include the following:

1. Simulated medical diagnosis. Kleinmuntz and Kleinmuntz (1981) created a diagnostic task in which simulated doc­tors attempted to treat simulated patients on the basis of their initial symptoms and of the results of any tests the doctor chose to order. They could also act at any point to administer "treatments" that might or might not improve the patient's health. Health fluctuated, over the 60 time periods of each trial, both in response to the doctor's interventions and to the preset (downward) course of the disease. The simulated strategies explored included Bayesian revision, a heuristic hypothesis-testing strategy, and a simple trial and error approach. The computationally intensive Bayesian strategy yielded only modest improve­ments over the heuristic strategy in this environment, and even the simplistic trial and error approach did well on some cases. Further simulation results are reported in Kleinmuntz (1985), and experimental results with real subjects are in Kleinmuntz and Thomas (1987).

2. Artificial worlds. A number of European researchers (see Mahon, 2000, for a review) have explored dynamic deci­sion problems with the aid of simulated worlds: fire fighting in simulated forests (B. Brehmer, 1990), economic devel­opment in a simulated third-world country (Reither, 1981), control of a simulated smallpox epidemic (Hesse, 1982), and so on. Funke (1995) provided an extensive review, with studies classified as to the person, task, and systems factors each examined. Typical findings are those of B. Brehmer (1990) from his simulated fire-fighting task. Subjects ini­tially perform quite poorly but can learn this complex task with repeated play. Feedback delays impede learning sub­stantially. Opportunities to offset feedback delay by decen­tralizing decision making were mainly ignored.

3. Systems dynamics. A group strongly associated with MIT (Diehl & Sterman, 1993;Paich&Sterman, 1993;Sterman. 1987, 1989) base their dynamic decision-making tasks on feedback dynamics models in which coupled feedback processes make response over time extremely nonintuitive to most subjects. For example, in Sterman (1987) subjects faced a capital budgeting task in which there was signifi­cant lag between ordering new equipment and having it available to meet increased demand. Most subjects in this task generated very large and costly oscillations, despite instruction in system linkages

As this sampling suggests, empirical studies of dynamic decision tasks are difficult. The tasks themselves are quite complex, even if they are greatly oversimplified versions of real-world analogs. Amateur subjects are thus easily over-_k whelmed, whereas expert subjects object to the unreality of f the tasks. Findings thus tend to be task specific and difficult to aggregate over different studies. Progress, clearly, is being made, but there are important challenges in this area.

Date: 2016-03-03; view: 785