OR/MS Today - August 2008



Organizational Anarchy


INNOVATIVE EDUCATION

Understanding Organizational Anarchy

Inside the "garbage can": agent-based models help explain well-known dysfunctions in managers' decision-making processes.

By Douglas A. Samuelson


Both real-life managers and teachers of management stand to benefit from recent research in organizational decision-making. What began as a whimsical, almost tongue-in-cheek characterization of how organizations decide — or, more often, fail to decide — now offers a promising basis for both new theory and improved practice.

Every manager has experienced the problem: deadlock, confusion and bickering, with important items seeming to take forever to get considered, let alone resolved Sometimes when a decision does get made, it seems not to be the one preferred by most of the people who supposedly collaborated in making it.

In turn, experienced managers who have tried to teach management have encountered the difficulty of conveying to students just how much real management typically differs from theory. Managers usually don't optimize, they muddle through, simply trying to stay one jump ahead of the issues that threaten immediate disaster. As C. Northcote Parkinson, renowned both as a historian and as a satirist, explained, "Many of the decisions in real management are like the one a person faces when, while crossing the street, he becomes aware of a runaway truck bearing down on him. It doesn't matter much which way you jump; the important thing is to choose one and do it quickly." Most management textbooks neglect to convey the proportion of the time many organizations spend deciding how to decide, sometimes accomplishing little else.

In a classic 1972 article, Cohen, March and Olsen argued that many organizations' decision process resembles throwing all the problems and all the solutions into a garbage can, where only chance meetings of problems and solutions produce anything of value. While this is a rather pessimistic view of performance and of the potential contribution of management, the predicted level of chaos and confusion appeals to many knowledgeable people as far more accurate than the characterizations most formal theories yield.

In all too many organizations, what we see is:

  • They don't make good use of the information they have.

  • Management seems unaware of what information subordinates have.

  • They alternate between indecision and impulse.

  • Internal divisions and factional conflict cause oscillation.

  • Organization-wide commitment to decisions is hard to achieve and maintain.

  • The focus often is not on the most important matters.

  • Resources are mis-allocated relative to likely benefits.

  • Some participants in the process seem to be favored over others although the favorites' recommendations work out relatively badly.

Management theories taught in business schools generally attribute such effects as these to failures of leadership, but rarely focus on the decision-making structure as the culprit. Thus, the would-be manager often emerges knowing how to recognize some signs of poor decision-making, but without much insight about how to improve it. To students, the ideas embodied in the garbage can model are abstractions that do not connect with their experience. Now, with the model available in a form easy to learn and modify, some teachers may find it easier to engage students, most likely through term projects, in learning how decision-making processes work and how they can be improved.

An Economic Theory


To understand how the decision-making process really works, imagine that each person present in a decision meeting must pay to participate actively, and the group then selects among the proposals entered into the discussion. The "token" required as payment to speak can take many forms, such as political standing or the perception that one is taking up too much time, but the effect is the same as if it were monetary. If one or more proposals to address the issue at hand have already been entered into consideration, each person compares the value, from his point of view, of the new proposal he or she most prefers to that of the most acceptable proposal already in consideration. If the difference is less than the value of the "token," he or she will remain silent on this question.

This simple assumption suffices to account for several common phenomena. A meeting can adopt proposals that are, in fact, not preferred by the majority of those present. Those who get to speak first, or are otherwise favored by a lower cost to enter the discussion, have a disproportionately high rate of success in getting their proposals adopted. The chairperson may deliberately manipulate the cost of entry to move the group toward proposals he or she favors, without being as obvious as direct advocacy would be. By setting a high cost of entry, on the other hand, the chair or another influential person at the meeting can prevent any decision from being made, or ensure that the decision will be simply reached but then widely resisted when the organization attempts to implement it.

Now suppose, in addition, that the chair of the meeting derives some benefit, presumably more or less (but probably not entirely) aligned with the interest of the organization, from each decision made. Each proposal has a cost to consider it — again, this may be largely non-monetary, requiring time and political capital, for example. The senior decision-maker seeks, therefore, to maximize the expected total benefit from the decisions made, subject to a constraint on the time, money and other resources consumed by considering the proposals. This is consistent with a well-known observation by cognitive psychologist and Nobel laureate (economics, 1978) Herbert Simon: "The time and attention of senior decision-makers is the scarcest and most valuable resource in most organizations."

OR/MS analysts will readily recognize that the senior decision-maker's choice is a knapsack problem, also known as a bin-packing problem. A reasonable but sub-optimal heuristic is to take the most valuable item first, then the most valuable item that will fit in the remaining space in the knapsack, and so on. Assuming that value is roughly correlated with size, this procedure results in taking a few high-value items, then filling the remaining space with small, low-value items. In the meeting context, this corresponds to solving a few issues of great importance and filling the remaining decision-making time with rather trivial matters. The results could be improved by a formal optimization, which might sacrifice a few large items to deal with a number of slightly less important ones. The predicted behavior seems intuitively accurate, in the view of many experienced managers. Michael Cohen, first author of the famous article mentioned earlier, has stated in a personal communication that this line of reasoning appears to provide a theoretical explanation, previously lacking, for why the garbage can characterization seems to fit real organizations so well.

Support from Biology


Additional insight comes from a recent model of human immune response. The blood contains numerous cells that identify unfamiliar substances and organisms and drag them to lymph nodes, located throughout the body, and then, if needed, to large lymph glands located in the neck, the maxilla (armpits) and groin. Lymph nodes contain a variety of specialized cells that can dispose of certain foreign substances and organisms. When one of these cells meets an intruder the cell "knows" how to deal with, it does so. Clearly, this model of the lymph node closely resembles the garbage can model of organizational problem-solving.

When the body is overwhelmed by some type of intruder, it begins producing all kinds of these specialized cells. This is why your lymph glands swell, to the point of becoming sore to the touch, when you have an infection. When the threat is sufficiently reduced, the body proceeds to dispatch other cells to kill many of the now-surplus specialized cells. This corresponds nicely to many organizations' tendency to hire more people and/or engage consultants, more or less willy-nilly, when in crisis, then to downsize when times get better — a feature not contemplated in the original article.

From this and other evidence, then, the garbage can model has gained considerable credibility as a serious characterization of how organizations decide. The next development, quite recent, is its translation into a form more suitable for experimentation.

Agent-Based Implementation


Klaus Troitzsch, an internationally renowned expert on social simulation, now has done an agent-based implementation of the garbage can model, incorporating some notable improvements and greatly elucidating the theoretical foundations of its assumptions. Following a suggestion he graciously attributes to this reporter, he has recast the model so that what problems require is not just "energy" but some mix of skills, and what floats around in the "can" with the problems is a variety of people with various combinations of skills. He proceeded to test various metrics of how well a problem-solver's skills corresponded with those required by the problem. As in the original model, problems arrive at random, with no guidance matching up problems and solutions. Instead, problem-solver agents have some rule by which they decide whether a problem fits their skills closely enough that they should tackle it. In this version, exactly one problem-solver works on a problem from match until completion.

For some cases, the rate at which problems arrive is low enough, relative to problem-solving capacity, that the organization solves its problems using any selection rule. For other cases, the organization is overwhelmed no matter which selection rule it uses. In some cases, however, the selection rule is sufficient to make the difference between an organization that can eliminate its backlog of unsolved problems and one that gets a backlog growing indefinitely. The rules that yield best performance maximize alignment of skills and minimize wasted talent, rather than minimizing time to complete the task. Put another way, a problem-solver with lower skill levels, but a mix close to the mix required for the task, is the best choice.

The model is written in NetLogo, which is easy to learn and modify. The code is freely available from Professor Troitzsch or this reporter.

Directions for Teaching and Further Research


Many variations are conceivable; more important, many of them are easy to implement. Problems could be modeled as imposing increasing costs over time until they are solved, which most likely would change the relative benefits of selection rules. "Manager" agents could be added, attempting to expedite matches via a variety of selection rules. These manager agents could be given various assumed amounts of knowledge and ability to discern where problem-solving agents are. Problem-solver agents could be grouped into teams; with or without such groupings, multiple problem-solvers could be permitted to work on each problem.

In the somewhat longer term, as researchers work out how to represent interactions among teams, this modeling approach may form the basis for a systematic investigation of the insight that motivates current U.S. combat doctrine: in an information-rich environment, networks out-perform hierarchies. Just how information-rich the environment has to be, and what processes in a hierarchy impede its performance, are among the open questions.

This model is simple enough to be a basis for teaching, as well, combining ease of learning with greater realism. Students can experiment — under experienced guidance, of course — with different ways of using and transmitting information within a decision structure and gain some understanding of how processes affect outcomes.

Guidance for Managers


The work to date generates a few suggestions for managers:

  • Manage your attention as a critical resource. Good managers are already acutely sensitive to potential wastes of time, but even good managers can do better at allocating more of their time to the problems whose resolution would be most valuable. This usually means devoting more time to setting clear goals, providing appropriate resources and managing relationships ... and less to supervising task activity.

  • Teach and encourage your subordinates to present their recommendation with sufficient supporting information to make your decision quicker and simpler; that is, promote Winston Churchill's concept of "completed staff work." This means rewarding people more for how well they supported their recommendations and less for how closely their recommendations resembled what you would have done anyway.

  • Make more effort to align tasks with the skills and interests of the people responsible for doing them.

  • Reorganize your agendas to allow due consideration of those nagging, not-quite-the-most-important problems. Don't keep letting the urgent crowd out the important.

  • Don't let anyone paralyze decision-making by adding uncertainty to issues her or she has some reason to want to remain unresolved. When you have found indications of some unexplained process that hinders decision-making, consider yourself the most likely culprit.

Conclusion


Agent-based implementations of the garbage can model, via readily available code, offer a useful tool for research, teaching and practice of management. Modifications emphasizing the role of attention and information should prove most fruitful. Results to date indicate that managers can gain much by more specifically and wisely managing their own time and attention. The ease of learning this model and experimenting with it should prove useful in instruction, as well.



Douglas A. Samuelson (samuelsondoug@yahoo.com) is a principal decision scientist at Serco-North America, a general professional services company in Vienna, Va. He is also president of InfoLogix, Inc., a research and consulting company in Annandale, Va., and a frequent contributor to OR/MS Today.

References


  1. Carley, Kathleen, and Michael Prietula, eds., 1994, "Computational Organization Theory," Lawrence Erlbaum.

  2. Michael Cohen, James March, and Johan Olsen, 1972, "A Garbage Can Model of Organizational Choice," Administrative Science Quarterly, Vol. 17.

  3. Guido Fioretti and Allessandro Lomi, 2008, "The Garbage Can Model of Organizational Choice: An Agent-Based Reconstruction," Simulation Modelling Practice and Theory, Vol. 16.

  4. Guido Fioretti and Allessandro Lomi, 2008, "An Agent-Based Representation of the Garbage Can Model of Organizational Choice," Journal of Artificial Societies and Social Simulation, Vol. 11, No. 1, 2008; also http://jasss.soc.surrey.ac.uk/11/1/1.html.

  5. Virginia Folcik Nivar and Charles Orosz, 2006, "An Agent-Based Model Demonstrates that the Immune System Behaves Like a Complex System and a Scale-Free Network," Swarmfest Conference, Notre Dame, Ind., June.

  6. Virginia A. Folcik, Gary C. An and Charles G. Orosz, 2007, "The Basic Immune Simulator: An Agent-Based Model to Study the Interactions Between Innate and Adaptive Immunity," Theoretical Biology and Medical Modelling, Vol. 4, No. 39.

  7. Nigel Gilbert and Klaus Troitzsch, 1999, "Simulation for the Social Scientist," Open University Press; second edition, 2005.

  8. Alessandro Lomi and Stefano Cacciaguerra, 2003, "Organizational Decision Chemistry on a Lattice," Swarmfest Conference, Notre Dame, Ind., April 13-15; also www.nd.edu/~swarm03/Program/Abstracts/LomiSwarm2003.pdf.

  9. Douglas A. Samuelson, 2000, "Designing Organizations," OR/MS Today, December.

  10. Douglas A. Samuelson, 2003, "The NetWar in Iraq," OR/MS Today, June.





  • Table of Contents
  • OR/MS Today Home Page


    OR/MS Today copyright 2008 by the Institute for Operations Research and the Management Sciences. All rights reserved.


    Lionheart Publishing, Inc.
    506 Roswell Rd., Suite 220, Marietta, GA 30060 USA
    Phone: 770-431-0867 | Fax: 770-432-6969
    E-mail: lpi@lionhrtpub.com
    URL: http://www.lionhrtpub.com


    Web Site Copyright 2008 by Lionheart Publishing, Inc. All rights reserved.