ARCHIVES Problems with causal-loop ... - systems …

1 ARCHIVES Problems with causal-loop diagrams (D-3312) George P. RichardsonFounded in the mid-1950s, the field ofsystem dynamicshas intellectualroots reaching muchfurther into the Archivessection of theReview seeks topublish materialfrom that past whichcan contribute tocurrent theory andpractice. The sectionwelcomespreviouslyunpublished but de-serving system dy-namics work, clas-sics from pastsystem dynamics lit-erature that shouldreceive renewed at-tention, and previ-ously published arti-cles from otherdisciplines of partic-ular significance tocurrent systemdynamicists. Contri-butionsemphasizing thephilosophy andtheory of modelbuilding, validation,implementation, ed-ucation, and genericstructures are partic-ularly mayrange in length fromnotes to main arti-cles but may be ed-ited.

2 Send suggestedmaterial, togetherwith a brief intro-duction placing it inhistorical and cur-rent contexts, toJohn D. Sterman, system DynamicsGroup, MIT E40-294, Cambridge,MA 02139, USAThe first system dynamics work did not include the use of causal-loopdiagrams. Feedback structure was portrayed by equations orstock-and-flow diagrams . Such representations were natural forengineers. In an attempt to make system dynamics accessible to a widerrange of people, causal-loop diagrams have become increasinglypopular. In many texts and courses they are the first tool , recently several analysts have proposed that system dynamicsstudies can be carried out without the development of formal models atall (Morecroft 1985; Wolstenholme and Coyle 1983; Wolstenholme1985).

3 causal-loop diagrams often figure prominently in such even those who advocate the use of qualitative system dynamics arecareful to point out that in all the successful applications of suchqualitative methods the analysts have had extensive experience withformal model building. Nevertheless, it seems inevitable that people at allexperience levels will continue to rely on causal-loop diagrams . In the following paper dating from 1976, George Richardson describes avariety of Problems which often arise in causal-loop diagramming, both inthe development of the diagrams and the explication of behavior fromthem.

4 The main difficulties arise because causal-loop diagrams obscure thestock and flow structure of systems . We sometimes emphasize so heavily therole of feedback structure in generating behavior that the crucial role ofaccumulation processes is lost. Even experienced modelers are easilymisled by causal-loop diagrams . I suggest the following experiment: takethe causal-loop diagram for the family feud described in Richardson'spaper and ask a random sample of system dynamics modelers or studentshow it will behave. In my experience, one will not only receive a widerange of answers but most of these will be incorrect.

5 Then repeat theexperiment using the stock-and-flow diagram ( with a different group ofpeople, obviously). While answers will still vary, the number of correctresponses should rise. In recognition of these difficulties, there has been arevival of stock-and-flow diagrams as a means of communicating structure(Morecroft 1982). Richardson's paper should not be taken as an argumentto abandon causal-loop diagrams or qualitative system dynamics,however. But it serves as a caution to the facile use of an easily abusedtechnique. Despite their Problems , causal-loop diagrams are likely toremain important tools for the communication of feedback D.

6 Sterman, EditorIntroductionPositive and negative feedback loops are the building blocks of systemdynamics. While a complete specification of the feedback structure of asystem requires specifying levels (states) and rates, the essentialcomponents and interactions in a system can be communicated quicklyand concisely in a causal-loop diagram. The simplicity of causal-loopdiagrams has led to their use in the early stages of modelconceptualization, in introductory curriculum material in system dynamicsfrom the fifth grade to graduate school, and in presentations of systemdynamics studies in both technical and popular Dynamics Review 2 (no.)

7 2, Summer 1986):158-170 ISSN 0883-7066 1986 by the158 system Dynamics : Problems with causal-loop diagrams 159 George P. Richardsonis associate professorof mathematics andcomputer science atWheaton College. Heholds an inmathematics from theUniversity of Chicagoand a in systemdynamics from theSloan School ofManagement at is co-author ofIntroduction toSystem DynamicsModeling withDYNAMO. Currently,he serves the SystemDynamics Society asvice president forpublications and asresearch problemseditor for the SystemDynamics :Science Center 105B,Wheaton College,Norton, MA 02766,USAFig. 1. A typicalcausal-loop(influence) diagramused to define positiveand negative causallinks (influences)The simplicity of causal-loop diagrams hides a subtlety, however, whichposes Problems which have not been adequately acknowledged.

8 The crux of the problem with causal-loop diagrams , which this paperexplores in some detail, is that they make no distinction betweeninformation links and rate-to-level links (sometimes called "conservedflows"). That simplification is usually thought to be one of the advantagesof causal-loop diagrams , but it has a rather dramatic disadvantage: in casesinvolving rate-to-level links the standard characterizations of positive andnegative polarities in causal-loop diagrams are false. This paper first exposes the difficulties in the traditional definitions ofpositive and negative links in causal-loop diagrams .

9 Several possibleimprovements are suggested. loops are then considered, and flaws in theirtraditional definitions and characterizations are uncovered, leading to theconclusion that definitions and characterizations in terms of dynamicbehavior are not possible. While the observations in this paper may have some significance forpracticing system dynamicists, particularly in their writings for generalaudiences, the paper's major purpose is to clarify ideas important in theteaching of system dynamics. I should say at the outset that the basic ideaspresented here are not new, most having appeared in some formpreviously.

10 However, the seductive simplicity of causal-loop diagrams hasled to general practice which is sometimes too casual and which may leadto of positive and negative linksTraditional definitionsThe following definitions of positive and negative influences incausal-loop diagrams are representative of the general following figure (Figure 1) shows a possible set of causalrelationships. The arrows indicate the causal direction of signs adjacent to the arrows indicate the polarity. A plus (+) signimplies that a change in the variable at the end of the arrow will cause achange in the variable at the top of the arrow in the same direction.