Theory Prediction Projection

From SystemsWiki

Jump to: navigation, search

Posted by Monte Kietpawpan Mar 12.2006

1.I am confused between the two concepts: prediction and projection.

D. H. Meadows once said the World II model is not intended to predict, but to PROJECT the behavior of the system, while R. N. Giere regards that as prediction. I don't know what are the differences between projection and prediction in the SD tradition.

2.It seems to me that SD is a theoretical tradition;

An SD model is considered a theory. I am not certain which part of the SD model is a theory. If an SD Model consists of a dynamic hypothesis, a flow diagram, model equations, and model behaviors, which of them is a theory?

3.Is an SD model a theory of structure or of behavior or of both?

I must understand this so that I can name a theory properly. Is there any convention for naming a theory constructed under the SD tradition? Let there be a theory of A. What is A? Is A the name of a problematic level, a principal loop, a system, or a problematic behavior being modeled?

Posted by David Gillespie Mar 13.2006

My understanding is that prediction states a condition of reality at some future point in time while projection states a possible condition of reality at some future point in time based on a set of assumptions. If the assumptions are correct, then the projection will be accurate, otherwise not.

I believe that an SD model is a theory in the sense that it specifies the variables and relationships necessary and sufficient to account for the behavior it represents. The theory is the complete model including all equations and the assumptions underlying those equations.

It is my understanding that an SD model is a theory of both structure and behavior because it is the structure that causes the behavior. So the structure and the behavior it produces go together. I do not know of a convention for naming an SD theory, but you might get some help by looking into the literature on archetypes.

Posted by Erik Pruyt Mar 13.2006

These are important questions.

You ask whether SD models are theories. SD models could be seen as micro-content theories, but not as general theories (see the many papers of David Lane on this topic). The combination of the structural model and the resulting behaviour is then a micro-theory explaining the particular dynamics of the modelled system.

(Related to this, one could also ask whether SD itself is a theory or not. Now, the answer to this question is rather difficult and there is confusion as to what SD really is: a (modelling) paradigm, a philosophy, a theory of structure, an epistemological theory, a methodology, a method, etc. Personally, I see SD as a very specific paradigm because of its very specific paradigmatic basic assumptions (ontological, epistemological, etc). Digging into these basic assumptions would also clarify the answer to your first question and would clarify the difference between SD and many ((post)positivist) domains of science which are characterised by very different basic assumptions. If you are interested in this question, then please let me know, because I have looked into this very question some time ago and might give you some help.)

Posted by Erling Moxnes Mar 13.2006

Good questions from Monte Kietpawpan about something that causes conceptual problems to most system dynamicists. Hopefully the following can be helpful in itself, or provoke better answers from others.

1. There is an important difference between projection and prediction, although the word projection may not be an obvious candidate to highlight this difference. When making a prediction, the purpose is to predict the future as accurately as possible. This task requires accurate, large and complex models, with no definite end to the amount of work that should go into the task.

The alternative to accurate prediction (projection) is simply to demonstrate that a given structure gives rise to a certain behaviour. (The simulation proves a theorem about the resulting behaviour). This can be a quite simple task where very precise models are not needed. It is also a very valuable task whenever decision makers misperceive the system and draw wrong conclusions about behaviour and therefore about choice of policies.

2. A theory usually denotes an established thesis or mental view, and is typically of a somewhat general nature. I recall J.W.Forrester being sceptical of general theories in social systems. Hypothesis seems a better word to use about a SD-model because a hypothesis typically denotes a case (often derived from some general theory), and because the hypothesis is subject to testing (against data and to see if it is useful for policy design). In SD literature the hypothesis is often referred to as a "dynamic hypothesis". To me this sounds as if the hypothesis is subject to change (the hypothesis is dynamic, which it typically is), however it seems more appropriate to use the term "hypothesis of the system dynamics".

Avoiding grand theories about social systems in general, the basic principles of system dynamics could be seen as theories from which hypotheses can be inspired but not deducted. E.g. search for important stock and flow relationships, feedback loops and nonlinearities.

3. In my view the hypothesis concerns the model structure (equations or equivalent graphical representations). Behaviour can be derived from structure by simulation (as theorems, or as a series of lemmas (preparatory simulations) to build up the clients understanding of the resulting overall behaviour from the behaviours of carefully chosen submodels).

Posted by Timothy Quinn Mar 13.2006

(1) Prediction versus projection.

Often, SD models are intended to reproduce a behavior mode (e.g., overshoot-and-collapse or damped oscillation with a certain period) rather than to make specific forecasts. For example, I can build a model to demonstrate how different rates of economic growth in India and China will influence the rate at which the world's oil stocks -- both discovered and yet undiscovered -- will ultimately be depleted. Can I determine from my model output a time at which full depletion occurs? Yes. Is this time likely to be correct? No. Is my model "invalid" or "useless" because this full-depletion time exhibits great uncertainty (i.e., wide confidence interval)? Absolutely not. Why? Because I constructed my model with the purpose of determining the influence of economic variables on the rate of oil depletion, with important implications for economic and energy policies. I did not build my model to estimate the full-depletion time.

Dennis Meadows and the rest of the Limits to Growth modeling team faced the problem of having critics who did not understand the purpose of their model. These critics assume that if the WORLD model is not a "miniature" of our real world, then it must be "wrong" or "invalid". Critics want to impose their own purpose on the model -- When will the oil run out? That's what is keeping me awake at night! -- a purpose that the model was NOT built to address. To divert the attention of these critics away from the times of population crash, toward the qualitative shape of the overshoot pattern, they needed to use a word other than "prediction". However, both "prediction" and "projection" suggest making your best judgements (with the help of computer simulation) about some aspect of the future.

(2) What part(s) of an SD model are theory?

First, I recognize that the question "What is a theory?" is not straightforward. But I will assume that our individual conceptions of "theory" are congruent enough to be unproblematic.

Let's say I have the theory "The Earth is flat". It is descriptive of reality "structure", falsifiable, and can serve as the basis for making predictions or counterfactual claims. I see a theory of this type as analogous to SD model structure.

Now let's say I have the theory "A rock dropped from a tower will fall to the ground faster than a feather because it is more massive". This theory comprises a causal story about dynamic physical laws, how things unfold over time, reality's "behavior". Again, it is falsifiable and can serve as the basis for making predictions. I see such a theory analogous to SD model equations or dynamics hypotheses that produce expectations about behavior patterns.

I don't think it is a problem to lump little "structure" and "behavior" theories together and call the whole package a theory as well. Therefore, I think the whole model is the theory. Or, if you like, a collection of little theories. For example, an equation in my model might reduce worker productivity when their level of fatigue increases. This equation represents my theory that "As workers become fatigued, they do not have enough energy to work at the same pace, so they slow down and their output per unit time drops".

(3) Is SD a theory of structure, behavior, or both?

A model is a simplification of reality. How do you know which features of reality to include in your model? How do you know which features "do not matter"? You must have criteria to decide these questions. These criteria are provided by your model purpose.

Let's say that I want to build a model to understand the following: Why does the national unemployment rate exhibit cycles (oscillations)? This purpose guides the formulation of my model. When sufficiently detailed, I can point to my model and say "This model is my theory for why the national unemployment rate exhibits cycles". Note that this statement has the form "<SD model> is a theory for <model purpose>".

The details of the theory for unemployment cycles involve structure and how that structure gives rise to certain behavior. Each piece of your model -- a loop, or an equation, or a stock-flow chain -- may represent a little theory of how you think reality works. You also hypothesize that these little pieces fit together into a large theory that provides an explanation for your purpose.

Posted by Arlen Wolpert Mar 16.2006

"(Einstein) once wrote in his autobiographical notes about the earmarks of a good theory:

There are two. One is that it should not be falsified or disproved by experiments. It doesn't even have to be proved by experiments as long as it doesn't contradict experience. And, secondly, it should have a kind of persuasive rationality, a kind of internal beauty that makes you convinced that you're on the right track - an aesthetic criterion."

(From an interview with Einstein's biographer, Gerald Holton, in the Harvard Gazette of October 2,1992).

Questions & Comments to Geoff McDonnell
Personal tools