REPLY Meaning of Stock/Level (SD6964)

[SDMAIL Ralph Levine]

Posted by  "Ralph Levine" <leviner at msu.edu>

As the discussion about representing soft variables as stocks continues, 
a number of new points about the usefulness of including soft variables 
in models have emerged.  There are many points we agree with.  First, we 
agree that the modeler must formulate and understand the specific goals 
of the model, and, at least, at first, not push the model outside of its 
usefulness, as Jean-Jacques Lauble suggests.  This rule of thumb applies 
to the inclusion of any variable, intangible or not.  We suggest 
starting from formulating the dynamic hypothesis representing the 
processes modeled in the loop structures.  Again, Jean-Jacques Lauble’s 
example of getting a satisfactory result with a small model that was 
composed of only 8 loops, without the inclusion of soft variables and 
additional processes (13 loops), makes a lot of sense to us. From a 
slightly different example, going back to the history of system 
dynamics, the urban dynamics model was criticized by some reviewers as 
being deficient because the model did not include a suburban component.  
One could argue that the structure formulated by Forrester to account 
for urban growth and decay was sufficient to evaluate suggested policy 
interventions without  adding the suburbs. This was essentially a 
boundary issue.  As in Jean-Jacques Lauble’s case, adding  loop 
structure to capture the dynamics of the suburbs and the interaction of 
the suburbs with the city  was not necessary to achieve the original set 
of goals. Walter Schroeder expanded the boundaries of the original urban 
model by including a suburban component.  The new model was 
significantly more complex. However, the augmented model essentially 
"reaffirms the policy recommendations presented Urban Dynamics." 
 (Schrioeder, 1975).  Forrester’s original dynamic hypothesis was 
sufficient to account for the problem.

The other side of the coin: When there needs to be more structure. 
Getting back to the topic of including soft variables in one’s models, 
perhaps the best reason for doing that is that the inclusion of soft 
variables may lead to better understanding and insight into what caused 
the problem and perhaps what changes in policies should be initiated to 
make things better in the long run.  Here is an example of such a case 
where including soft variables paid off. The work of John Heinbokel and 
Jeff Potash, who developed a SD model of the pneumonic plague comes to 
mind.  Epidemiological compartment (S-E-I-R) models are well known to 
system dynamicists (e.g., see John Sterman’s text, Chapter 9). John and 
Jeff’s model was formulated with the best science available.  The first 
iteration contained no subjective variables.   After developing the 
model, they looked for data to see if the model fit a real instance of 
the plague.  They found time series data from an outbreak of the plague 
in Surat, India in 1994. Their first model fit those data miserably.  
The model predicted many more new cases and a much longer course of the 
disease than was found in the time series data. Why did the epidemic die 
out so fast?  John and Jeff began to get the details of what happened by 
rereading a very detailed account of the incident written by Ghanshyam 
Shah   Shah’s book contained both quantitative data and as well as a 
vast amount of qualitative data dealing with differences in coping 
responses of the residents and the decision processes used by officials 
and health care providers.  In addition to the eventual distribution of 
prophylactic antibiotics (and the indication that those antibiotics were 
not typically used correctly), they discovered that a large proportion 
of the population fled the city, and those who remained self-isolated 
themselves.  Epidemiological models, in general, do not include these 
behavioral mechanisms to cope within the framework of the model.  John 
and Jeff then (1) estimated antibiotic usage, (2) included a drainage 
(fleeing) process out of the stocks of (a) people at risk and (b) people 
incubating the disease, and (3) incorporated a change in the mixing 
coefficient to capture the effect of self-isolation. A combination of 
both processes accounted for an excellent  fit to the quantitative data.  

As system dynamicists, they realized that the changes they had made to 
the model were exogenous in nature. The parameters were in fact dynamic 
and most likely part of the internal loop structure.   Thus, they sought 
to endogenize those behavioral processes by developing an additional set 
of loops that would account for the behavioral responses of fleeing and 
self-isolating.  As part of expanding the boundary of the model, they 
consulted with social scientists (RL was among them) as domain experts 
to help them incorporate such processes as (1) risk perceptions as one 
of the drivers for fleeing and accepting medicine and (2) trust in the 
authorities in communication those risks.  We should emphasize that a 
good fit of the augmented model does not mean that other very different 
mechanisms might also generate an equally good fit. However, the 
behavioral processes used in the model reflected the best scientific 
information currently known and accepted.  The output of the model also 
was consistent with a recent study by the WHO on those behavioral 
processes included in the model, any one of which, if mishandled, could 
result in undermining effective communication and response strategies. 
This provides at least an initial basis for validating the model from an 
independent source.

Briefly, we want to comment about the issue of when to use a soft 
variable as a stock and when to treat it as a weight or index that is a 
function of conserved stocks.  We differ somewhat from John Barton’s 
handling of the intangible variable, "reputation." He points out that 
reputation is a function of "easily recognized measurables."  
Presumably, if those input variables change, the organization’s 
reputation will change.  However, in many cases, the input variable can 
change up and down, but it take times to generate a good reputation and 
respond to variability in the input variables that would make up the 
index.  In this particular case, we suggest thinking about making 
reputation an information delay of some order. This is what we described 
in our last post. It was also suggested by Jean-Jacques Lauble in 
handling the willingness of the client to fulfill needs his or her needs 
with the considered project. 

The use of the smooth in this case may work and be simple enough to 
capture the delay in the cognitive system and be consistent with 
original goals of the model.  Let us point out, however, that for 
variables, such as reputation, the dynamics may be a bit more 
complicated.  For example, reputations have to be maintained, which may 
make including a drainage process desirable. The drainage process would 
represent those factors that are unaccountable by the structural loops 
associated with the inputs, such as perhaps competition from other 
firms, if that process is not in the model.  In our estimation, the 
adding a drainage process to a smooth is certainly possible ( which 
would generate a steady state error )is problematical, but that is 
material for another post.

 

Forrester, J. (1969). Urban Dynamics. Cambridge, MA: Wright Allen Press

Heinbokel,  J, & Potash, J. (2003). Modeling behavior ;as a factor in 
the Dynamics of an outbreak of pneumonic plague. Proceedings of the 
international Conference of the System Dynamics Society, New York, July 
20-24.

Heinbokel, J., & Potash, J. Endogenous ;human behaviors in pneumonic 
plague simulation: Psychological & behavioral theories as small "generic 
models.  Proceedings of the international Conference of the System 
Dynamics Society, Boston, July 17-21.

Shah, G. (1997). Public health and urban development: The plague in 
Surat. New Delhi: Sage Publication

Schroeder,W,. W. (1975). Urban dynamics and the suburbs.  In W. W. 
Schroeder , R. E. Sweeney, L.  E.  

  Alfeld (Eds.) Readings in Urban Dynamics: Vol. 2, Cambridge, MA: 
Wright-Allen Press

Sterman, J.D. (2000) Business dynamics: Systems thinking and modeling 
for a complex world.  Boston: McGraw Hill.

World Health Organization. (2005) WHO outbreak communication guidelines.

Respectfully,

Ralph Levine, Ph.D.
Professor Emeritus (On Call)
Departments of Community, Agriculture, Recreation, and Resource Studies And
Department of Psychology
Michigan State University
East Lansing, MI 48823

AND

David W. Lounsbury, Ph.D.
Asst. Attending Psychologist, Beh. Sci. Service
Community Outreach and Health Disparities
Dept. of Psychiatry & Behavioral Sciences
Memorial Sloan-Kettering Cancer Center
641 Lexington Ave, 7th Floor
New York, NY 10022

Posted by  "Ralph Levine" <leviner at msu.edu>
posting date  Wed, 23 Apr 2008 18:07:15 -0400