Skip to content
Oct 6 / Barry Clemson

Specifying a System: How science really works

Barry Clemson. Volume 1. Issue 3. October 4, 2012

Abstract: How do we specify a system for study? Science proceeds largely by inventing an increasingly rigorous answer to what is meant by “the system”. In general, we begin by pointing at or naming some natural phenomena, then we define a boundary, and finally we list the variables and relationships that constitute the conceptual system. The result is called a model or a theory.

An Overview

This section provides an example to get us started. The example proceeds in nice neat stages. Please keep in mind that science rarely proceeds so neatly, but this pretense simplifies the discussion.

A doctor (a medical researcher) might say:

“The system I study is the human body and its state of health”

We understand, at least in general terms, what the doctor means. But a personal trainer might ask “Is the exercise the body gets also part of the system?

The doctor might be a little more precise by saying:

“The system I study is the human body, defined by it’s skin”

Now the system is defined by its boundary. It seems that the answer to the trainer’s question is that the gym is outside the system but the strength and stamina of muscles is part of the system.

A biologist might ask if bacteria and viruses are part of the system. What about the microbes that live on the skin? What about the multitudes of microbes that live in symbiosis with the human body and without which we quickly die?

A nutritionist might want to know if the food eaten by the human is part of the system? And what about the digestive tract … strictly speaking everything in the stomach and the intestines is OUTSIDE of the skin (topologically, the human body is a donut and the hole in the donut is the digestive tract)?

It seems that specifying a boundary still leaves quite a bit of ambiguity about what is and is not part of the system under study.

Our doctor might try again. She might say:

“The system I study consists of the following variables:
·      White cell count in the blood
·      Red cell count in the blood
·      T-Cell count in the blood
·      HIV cell count in the blood
·      and so on.

This is a partial list of the variables that a doctor dealing with HIV positive patients might study. Such a list, when completed, provides a rather rigorous and unambiguous definition for a system.

Let’s summarize. Our medical researcher provided three ways of specifying a system for study. In increasing degree of rigor, they were:

1.   Name a system
2.   Specify a boundary for a system
3.   Specify a set of variables.

Specifying an exhaustive set of variables works very well so long as we are dealing with systems that are not overwhelmingly large and complex. It becomes quite unwieldy (probably impossible) if we are studying General Motors, a national political system, or a neighborhood.

Ross Ashby (1956, 1960) suggested that for every system that exhibits homeostasis (i.e., a dynamic stability maintained by a complex web of inter-connected feedback loops) there will be a set of essential variables that must be kept within physiological limits. For these systems, listing the essential variables is a good way of defining the system.

Thus, for complex systems, instead of listing all of the variables, we list only the essential variables which, at this point are our best estimate of those variables regulated by the system’s own internal homeostasis and that becomes our system definition.

Depending upon our practical purposes, a wide variety of other “relevant variables” may also be important in addition to the essential variables. For a trivial example, consider a clock: our purpose for the clock might be either as a decoration in the home or for timing training runs on a racing yacht. These two purposes imply quite different “relevant variables” in terms of, for example, appearance or corrosion resistance.

Thus our next stage in defining a system is:

3a. List the essential variables maintained by homeostasis and those other “relevant variables” required for our specific practical purposes.

We also need to know the relationships among the variables: what effects what else?  Given only three variables and the assumption that the relationships are asymmetric, there are 21 possible different configurations of relationships. Therefore, figuring out which relationships are important is generally a non-trivial exercise.

4. List the key relationships among the (essential and relevant) variables.

This gives us four steps or stages in defining a system. These steps correspond to gaining a better and better understanding of that system. When we succeed in doing all this, we usually have a pretty good model or theory for that natural system.

For complex dynamic systems, when we know the essential variables maintained by homeostasis we are well on the way to being able to say what constitutes “health” or “sickness” for that natural system. Further, understanding homeostasis for a given natural system means we are getting close to being able to identify basins of stability for that natural system.

When we can identify basins of stability for a natural system, we can then identify the stressors which push the system from one basin to another. Doctors can do this for some of the basins that are important for the human body, e.g. “normal”, hypothermia, heat stroke, diabetic coma, etc.

If we could do this for Gaia (Lovelock, 2000), we would know whether or not humankind is threatening balance of the planetary ecosystem. Unfortunately, our understanding of Gaia is still pretty rudimentary so all we can say is that humankind is stressing Gaia … but we do not know if we are close to critical thresholds.

“System” used to mean different things

The word “system” is used in several quite different ways.

“System” is used in ordinary conversation to point to something like the “education system” or the “criminal justice system”. The speaker and the listener both understand that this use of “system” means a complex set of institutions and processes that is not very well defined.

Scientists use theories or conceptual models to understand natural systems. For example, Newton’s laws of motion (a conceptual model) are used to understand the forces acting on a race car as it accelerates (a natural system). I will use “system” to mean a conceptual system, i.e. I use system interchangeably with model or theory.

“System” is often used rather loosely by scientists when they fail to distinguish between some actual phenomena and the conceptual system which represents our model of that natural system. I use “natural system” to refer to the actual phenomena.

 Examples of How Science Really Works

A Systems Dynamics Example. Bellinger’s “Unleashing Understanding” (2012) provides a nice example to illustrate my main points.

The paper started with a basic question, specifically:
Given a relationship between the pesticide applied and the amount of crop damage, what else is going on?

By repeatedly asking this simple question “and, what else is going on?” a model of the natural system was developed.

Bellinger’s example did not explicitly name the system nor did he begin by explicitly defining a boundary. His focus was on a relationship between two variables.

If we had asked Bellinger to name the system and what the boundary was, he probably would have pointed to the field where the crop was growing or perhaps said something like “those turnips over there”. This is a typically messy starting place for science.

Bellinger’s “Unleashing Understanding” nicely illustrates the four stages in specifying a system listed above. However, the paper also makes clear that the process is not one of moving sequentially through the stages. Rather, it is a rather messy process in which an insight in one stage illuminates another stage.

In Bellinger’s example, boundary definition, variable definitions, and relationship definitions were all mixed up and inter-twined. Further, this inter-twined messiness is the usual situation.

A Physical Science Example. I argue that the process I describe is actually how science works, although the textbook description of science is quite different. Consider a simple pendulum.

1.   I already did step one: I named it.

But now imagine that we know nothing about pendulums and have just invented one by hanging a rock on a string attached to a rafter.

I notice that the rock, when displaced and released, follows some interesting cycles. Assume I record, every second, the velocity and position of the rock. This gives me a long table of data. I can easily verify that my pendulum is regular and deterministic in its behavior, i.e. a given position and velocity always goes to the same next state and there is no variation in this behavior. Unfortunately, there is no way to summarize my table of data, no mathematical formula that describes the system as defined so far.

Suppose Frank, one of my friends, wanders in and asks me about the boundary of this “pendulum” thing. My first thought is that the boundary is what we can see: a piece of string, a rafter where the string is fastened, and a rock.

Frank whips out his stop-watch and iPad and a minute later says “you’re looking at the wrong system.  The system you should be looking at consists of these three variables: length of the string, gravitational field, and period of the pendulum”

I respond, “Why would I do that?”

Frank says “Look at this elegant equation. It describes the pendulum perfectly”.

Ashby (1956) pointed out that science might well be thought of as a search for systems that can be simply described (preferably with an equation or two). It is considered a great triumph when science succeeds in defining a system that can be described simply.

To summarize the pendulum example, the science proceeded as follows:
1.   Naming the thing
2.   Defining the boundaries (incorrectly) and re-defining the boundaries to include gravity
3.   Listing the variables that enabled a simple mathematical description of the system., i.e., length of string, gravitational field, period of the pendulum, and
4.  Defining the relationships among the variables (i.e., with the equation).

For this simple system, there is no need to go any further: the list of variables is the same as the set of essential variables.

Please notice that there are innumerable ways to draw a boundary around the thing we name a “pendulum”. Further, even with an appropriate boundary, there are a number of variables that could have been chosen (e.g. mass of the rock) that would not have resulted in a simple, easy to describe system.

In scientific / engineering circles a “pendulum” is defined to be a system consisting of three variables (gravity, length, and period) and we normally think of this as an objective definition.

I assert that the three variables used to define a “pendulum” were subjectively selected. Imagine that the rewards within science were for descriptions of systems that required tabular form. In this case a “pendulum” defined as position and velocity of the rock would be valued and the three variable system we normally teach in textbooks would not be valued. The selection of pendulum = f(gravity, length, and period) in preference to pendulum = f(position, velocity) is based entirely on the reward system within science because the two are equally true and complete.

To assert that the system definition is subjective is not to say it is arbitrary. The system definition is always selected by some observer(s) with some purpose(s) in mind. Further, whatever “reality” consists of, it clearly includes some roadblocks that reveal at least some of our errors in thinking.

The real importance of experiments is not that they prove hypotheses (experiments cannot prove anything) but that they demonstrate that some hypotheses are wrong. Thus experimental science should not be thought of as a process of proving things, but rather as a process of moving away from errors.

The subjective nature of our definition of a system for a “pendulum” is also true for any system we may define. Remember, we are using “system” to refer to the mental construct that the scientist / observer abstracts from the real world phenomena, i.e, from the natural system. Further, please notice that it is impossible to completely describe any real world object … all of our descriptions of reality are partial and biased in unknowable ways.

Deciding how to start

Imagine you are confronted with a natural system that is truly novel, i.e., you truly know nothing about it. How do you begin to study this?

First, name the thing. You have to be able to think about it, you need to tell your colleagues and your husband about it, and your field notes need a way to refer to it. So, give it a name, for example, “blorp”.

Second, you need a tentative scope of study, so define a tentative boundary for it. If it is an alien being, it’s skin is an attractive boundary. If it is an ecosystem, some sort of geographic limits are an appropriate starting point, e.g. a watershed or continental boundaries. If you really know nothing about this “blorp”, then you can not strategize about how best to study it. All you can do is observe it or randomly prod it in some way. Eventually, you begin to list what seem to be important variables for this “blorp”.

Next, list the variables that define this “blorp” as a system.

If this “blorp” turns out to be truly different from any phenomena you are familiar with, this will be a long and difficult business. All your assumptions and expectations will mislead you. The situation is quite different and easier if the new natural system is similar to one you are already familiar with ( e.g. an expert on earth eco-systems is studying a Martian eco-system). In this case, the process is likely to go more quickly and smoothly.

There is, of course, a danger here: we might mistakenly think the new natural system is similar to one we already know about when it is in fact quite different. In this case, our previous knowledge will lead us astray and will almost certainly blind us to important aspects of the new natural system.

The seductiveness of boundaries

Our language and our training encourages us to think in terms of “things”. Even the textbook descriptions of science (hypothesis, experiment, etc.) is couched in terms of things. While real science rarely if ever actually works anything like the textbook description, that description is still buried deeply in most of us who consider ourselves scientists.

So, we are accustomed to think in terms of “things” and these “things” generally have nice hard boundaries, e.g.:
·      Cells are bounded by membranes
·      Animals are bounded by skin
·      Groups have boundaries (members vs non-members)
·      Neighborhoods, cities, states, nations all have well-defined geographic boundaries
·      Organizations are nicely bounded by employees, volunteers, products, ownership of physical and intellectual property.

Our usual habits of thought leads us to think that these “things” in the natural world are well-defined by their boundaries.

For the scientist, defining a system by specifying the boundary is quite ambiguous and unsatisfactory. Specifying a boundary may be the best that can be done at a given point in time, but the systems scientist should be very clear that this is merely an interim step in coming to an adequate system definition.

In general, by the time an observer is ready to suggest a boundary for a system, he or she probably already has in mind a set of variables that seem important. Therefore, in addition to specifying a boundary, the observer should also list the variables and the important relationships among them that seem important.

Of course, initially you are almost certain to be wrong in both your boundary definition and your list of variables … but being explicit about them will make it easier to spot your mistakes. This entire process of defining a system, finding errors in our thinking, re-doing the system definition, etc. is a recursive, iterative process of unfolding a mystery via correcting errors in our thinking.

A somewhat later question is “how do we know when to stop?” Systems thinking tells us everything is connected to everything else. A slightly demented System Dynamics modeler could begin with Bellinger’s original two variable relationship, discussed above, and end with a model of the global economy or a model of the ecosystem of the Mississippi river basin (given sufficient funding!).

Thus an important question is the boundary we put around our inquiry. The simple answer to “where do we stop?” is that our practical purpose tells us what is irrelevant and our model tells us what can be safely ignored (i.e., our model tells us what relationships are important for our practical purposes).

 Policy Analysis and Design Practices

Policy analysis is an important special case of system definition. Design (e.g., architecture or engineering design) is, from this perspective, a similar process. Both policy analysis and design begin with a consideration of system effects (R. Wright, personal communication).

Policy analysis typically begins with one of these questions:
•   How can we mitigate THIS undesirable outcome? Or,
•   How can we get more of THIS desirable outcome?

Some concrete examples include:
•   How can we reduce illegal drug use?
•   How can we improve the math ability of graduating high school seniors?
•   How can we reduce the infant mortality rate?
•   How can we improve the health outcomes for US citizens?
There are of course a great many such questions that a society considers from time to time.

Defining a system for policy analysis purposes generally starts with consideration of the effects produced by that system. If our concern is that high school graduates are mathematically illiterate, then we need to understand the system that produces this outcome. We will also want to invent and then hopefully implement a system to produce better outcomes.

Design, whether of a coke bottle, a race car, or a town, also involves beginning with consideration of system effects. For example the effects desired from a Ferarri are much different than the effects desired from a Ford 150 pickup.

The key point for this paper is simply that beginning with a desired set of effects (whether to avoid them or to achieve them) is a way to focus and thereby substantially reduce the difficulty of specifying the relevant system. For instance, consider a valley that has suffered serious overgrazing and topsoil loss. To improve the valley’s ecosystem, one must find plants and trees that will grow under the current conditions and gradually restore the valley to ecological health. While this is a complex issue, it is probably orders of magnitude simpler than “let’s understand the entire ecology of this valley”.

Similarly, understanding why high school graduates are poor at math and designing interventions to improve math ability is much simpler than trying to understand the sociocultural system of teenagers. Granted that improving math scores requires a certain level of understanding of teen-age culture, but it does not require understanding all of it.

In the abstract, the process for defining a policy analysis system is the same as for any scientific study: system name, system boundary, system variables, and relationships among those variables.

In practice, the difference between system specifications for policy analysis (or design) and for some non-human natural system has to do mainly with the greater complications involved in dealing with human beings, large institutions, and political factions.

Conclusion

It is useful to think about four steps or stages in the rigor and adequacy of defining a system. These steps roughly correspond to our level of understanding of the natural system being studied. The steps are:
1.   Name it
2.   Specify a boundary
3.   List the relevant variables (3a. For very complex systems, list only the essential variables maintained by homeostasis and selected relevant variables needed for your specific purposes).
4. Specify the relationships among the variables.

In practice, scientists probably never go through these stages in order from beginning to end. The stages are inter-related and progress in any one usually illuminates others. The normal path for a study is highly convoluted and iterative through these stages.

The conceptual systems (i.e., the models and theories) that result from these four stages are the products of science. These four stages provide a very general description of what scientists do.

In all of this, it is crucial to remember that our systems are mental constructs and that these mental constructs are less than reality. Our conceptual systems (i.e., theories and models) are subjectively defined descriptions of natural systems that are beyond our ability to fully grasp.

Barry Clemson worked in custom manufacturing, community development, educational evaluation, software development, university teaching (organizations & system thinking), consulting, construction (as a small contractor and carpenter), and as a novelist (in roughly that order). His studies have been as eclectic as his work experience and has ranged broadly over the sciences. He started studying systems / cybernetics in 1967 when he discovered Stafford Beer. Stafford became a friend and his most important intellectual mentor.  “Management cybernetician” indicates Clemson’s worldview / biases. He is committed to helping turn the current global mess into an opportunity for renewal. www.barryclemson.net

Staff helping with this article: Editor: Ivan Taylor.  Reviewers: Gene Bellinger, Anne Maguire, Richard Wright

Acknowledgments

In addition to the Ashby books listed below, this work has been heavily influenced by:
1.  Russell Ackoff, Stafford Beer, and Heinz von Foerster, who provided a general framework for thinking about science (however, which ideas came from where is difficult to sort out).
2. Gene Bellinger’s insistence, that the early drafts were not correct, was critical in achieving whatever clarity this paper has achieved.
3. In addition, several of the STW discussion threads have been challenging in very helpful ways (again, it is no longer possible to sort out which ideas came from where, but Jessie Henshaw made several very helpful points, especially on the need to be clear about the difference between conceptual systems and natural systems).

References

Ashby, Ross. (1956). An Introduction to Cybernetics. London: Chapman & Hall.
Ashby, Ross. (1960) Design for a Brain: the origin of adaptive behavior. London: Chapman & Hall
Bellinger, Gene. (2012). Unleashing Understanding. Systems Thinking World Journal: Reflection in Action. [Online Journal]. Vol. 1 Issue 1. [Referred 2012-04-26]. Available:http://stwj.systemswiki.org . ISSN-L 2242-8577  ISSN 2242-8577
Lovelock, James. (2000). Gaia: A new look at life on earth (Kindle edition). London: Oxford University Press.

Reference details for this article

Clemson, Barry. (2012). Specifying a System: How science really works. Systems Thinking World Journal: Reflection in Action. [Online Journal]. 1(3). [Referred 2012-10-04]. Available:http://stwj.systemswiki.org . ISSN-L 2242-8577  ISSN 2242-8577

Print Friendly

One Comment

leave a comment
  1. Trevor / Jul 24 2014

    Brilliant article Barry, I am much enlightened by reading this concise work. thank you!!

Leave a Comment