Garbage Can University

Studying complex systems via computational modelling can lead you far afield to interesting ideas. This week I read several papers on the Garbage Can Model and how it might describe the decision-making processes (or lack thereof) in universities. I first read the original 1972 paper describing the model. (Abstract and citation in figure below.)

The system being modeled is “organized anarchy”; the authors claim that universities are a prototype of organized anarchy. Such systems have three general properties. First, problematic preferences means that the organization “operates on the basis of a variety of inconsistent and ill-defined preferences [and] can be described better as a loose collection of ideas than as a coherent structure…” Second, unclear technology refers to a lack of understanding among members of the organization. The system therefore “operates on the basis of simple trial-and-error procedures, the residue of learning from accidents of past experience, and pragmatic inventions of necessity.” Third, fluid participation means that time, effort and involvement may have wide variation, and therefore “the boundaries of the organization are uncertain and changing; the audiences and decision makers for any particular kind of choice changes capriciously.” Sound familiar?

In particular, the authors are interested in studying (1) how such organizations make choices in the absence of consensus, (2) how participants become “actively” involved when “not everyone is attending to everything all of the time”. At its base, the model assumes four independent streams that vary temporally: Problems, Solutions, Participants, and Choice Opportunities. There are varying flow rates of different streams, and a nebulous “energy” term is needed to solve (or in most cases) resolve a problem. Sounds like a multidimensional kinetics problem of a complex system, at least to me, the physical chemist. The 1972 paper details the model and includes a full Fortran program for users to tinker with the model and its parameters.

How can decisions be made in the model? The one we all assume: By resolving a problem that shows up, after working on it for some time. But there are two others. By oversight – there’s a norm, and it is simply applied without much time and energy. Or by flight, i.e., the problem is simply punted and not resolved, as new problems come in. The participants have simply re-attached themselves to new and different incoming problems. To see the model and simulation statistics, I recommend reading the paper in full. For this blog post, I wanted to highlight the implications of the results as described by the authors. A note of caution: the model is very simplistic, so one should be wary about the strength of the conclusions. That being said, as an academic I find the qualitative descriptions familiar-sounding.

“University decision making frequently does not resolve problems… Decisions whose interpretations continuously change during the process of resolution… Problems, choices, and decision makers arrange and rearrange themselves. In the course of these [re]arrangements the meaning of a choice can change several times… Problems are often solved, but rarely by the choice to which they are first attached. A choice that might, under some circumstances, be made with little effort becomes an arena for many problems… The matching of problems, choices, and decision makers is partly controlled by attributes of content, relevance and competence; but it is also quite sensitive to attributes of timing, the particular combinations of current garbage cans, and the overall load on the system.”

The 1972 paper has some interesting graphs. Above is Figure 5 from the paper looking at how different hypothetically sized schools with different baseline resources might change the way decisions are made depending on whether times are plentiful or lean. The authors make some predictions based on their model. “As adversity continues… all schools, and particularly rich schools, will experience improvement [i.e. resort to ‘higher efficiency’] in their position… Presidents of such organizations might feel a sense of success in their efforts to tighten up the organization in response to resource contraction.” Interestingly, small selective liberal arts colleges with large endowments are in the ‘unsegmented’ decision structure, i.e., they remain less hierarchical than their counterparts although as the famine approaches, they may cross the border into the ‘hierarchical’ decision structure.

The authors conclude: “It is clear that the garbage can process does not resolve problems well. But it does enable choices to be made and problems resolved, even when the organization is plagued with goal ambiguity and conflict, with poorly understood problems that wander in and out of the system, with a variable environment, and with decision makers who may have other things on their minds.” Sound familiar?

How has the Garbage Can model fared over time? Not very well because there are fundamental flaws as described in a 2001 paper by Bendor, Moe and Shotts. (See abstract and citation in figure above.) The authors think this is unfortunate because the “[original] paper is brimming with provocative insights that offer a promising basis for theory, but thus far their potential has largely gone untapped.” Their review and critique is comprehensive; I recommend reading the paper in full. I will highlight two significant issues. First, there is a disconnect between the verbal form of the theory and its heavily constrained incarnation within the model. Everyone should remember the dictum: All models are wrong, but some are useful. But in this case, the conclusions drawn may have well exceeded what the model is able to do given its limitations and strictures. The verbal articulations do not match up well with the mathematical constraints and rules. Second, the model neglects the role of individuals and how they actually behave when problems arrive. Depending on their role in the organization and their relative expertise, there could be a number of specific feedback loops crucial to the model. The independent stream assumption coupled with a linear temporal flow is unrealistic, possibly to the breaking point. However, the authors hope that the model will be revamped significantly so that it might actually prove insightful.

One example that tries to improve on the model is recasting it as agent-based. The authors, Fioretti and Lomi, also add several features to the model. (Abstract and citations shown below.) Flight (i.e., not resolving the problem) can now be postponement (per the original) but also passing-the-buck. I can personally attest that both happen very often at multiple administrative levels in the university. At first glance, this might seem to be a bad thing – but the simulation shows some interesting results.

The authors also compare and contrast organized anarchy with two different hierarchical setups. (In a hierarchy, “participants are only allowed to make decisions on choice opportunities that are equally or less important than their own hierarchical level.”) In the competent hierarchy, participants higher up in the hierarchy have the greatest ability to ‘solve problems’. In the incompetent hierarchy, the opposite is true. The higher-ups have lower ability to ‘solve problems’. Problems are narrowly defined to be technical, and solutions to problems also have narrow characteristics –read the paper in full for details. Again, there are some surprises. The incompetent hierarchy may actually outperform the other systems under some desired metrics/outcomes.

Fioretti and Lomi have worked on confirming the decision-making characteristics outlined by the original 1972 model. Three are highlighted by the authors. (I have modified their quotes slightly for clarity.)

·      Decisions by oversight are much more common than decisions made by resolution suggesting that the rational mode of decision-making is rare (< 20%). Most decisions are socially induced acts, made with the purpose of obtaining legitimacy by conforming to required rituals.

·      In a hierarchy, top executives are busy gaining legitimacy for their organization by means of decisions by oversight, whereas the [lower-level] bottom line cares about solving [technical] problems.

·      Organizations make themselves busy with a few problems that present themselves again and again. So participants have the impression of facing the same problems repeatedly.

Sound familiar?

There are several interesting conclusions (again, subject to the limitations of the extended model). Postponement and passing-the-buck can be “beneficial to an organization, since they avoid members wasting time on problems they cannot solve… they channel the most difficult problems to the best problem-solvers, creating opportunities for them to display their abilities.” Another surprising observation is that contrary to common wisdom, it is NOT necessary that the most capable problem solvers sit at the top of the hierarchy. Instead, good “socializers” who can “obtain legitimacy” for the organization should be the high-level “managers”. That already happens in many cases, but that means that the different roles in a hierarchical structure simply emphasize different strengths. The disparity in compensation seems particularly outlandish in this light – unless one values the social aspect much more highly over the technical aspects. The authors carefully state that “incompetence at problem solving should not be confused with [overall] incompetence… [and] top decision-makers should be good at gaining legitimacy for their organization, which is possibly the kind of ability they should be selected for.”

While I think the garbage can model is still too crude to establish particular practices for organized anarchy and hierarchical systems, it did remind me that I should be a little more understanding when the system doesn’t “work” as efficiently – i.e., the person I think should solve a problem postpones or passes-the-buck. Not because I think that’s the right response in the particular cases I’m thinking about, but that being in the midst of an organized anarchy leads to certain behavioral ‘ruts’ – explicable to some extent by systemic issues. The papers also made me think about how encountering different modeling approaches to complex systems in areas vastly different from my own, has expanded my horizons. Organized anarchy seems oxymoronic, but it’s also a good description for thermodynamics. Are managers simply needed to control entropy?