Demographic Profiles and Health Care

"Ashley Woolmore"
Junior Member
Posts: 14
Joined: Fri Mar 29, 2002 3:39 am

Demographic Profiles and Health Care

Post by "Ashley Woolmore" »

I have just joined the mailing list and thought I would take the opportunity
of introduing myself.

I currently work for the National Health Service in the UK, and have a keen
interest in Systems Dynamics
and Systems Thinking. One of the things that I am doing at the moment is
working with small team of managers, both in the health service and in
social services to develop insight into the ways that these organisations
currently function, and the potential benefits and costs of closer

I have been experimenting with a range of techniques - I have found the
ithink software a flexible and powerful tool for this sort of work. The
sorts of models that I have been building with it range from the very simple
coin sorter type - as a tool for learning about basic systems dynamics and
their applications in organisations - to the more complex dynamic models. I
have also been interested in the use of these tools for public health
decision making and health care strategy.

One question that I would like to ask relates to a simple method of
constructing ageing and death in a population. One thing that I tried to
model early on was the gradual ageing and dying of a population. However,
if I use a simple drain from a stock I get a typical exponential curve.
This does not match my mental model of this situation. Instead of getting a
curve that looks like the first half of a U I want to get a curve that
looks like the latter half of a n - that is people gradually age, then
death rate for the population increases with time. You may realise already
that I am self-taught, i.e. I hope this is not a totally naive question! I
have modelled the sort of behaviour that I want - but only with a quite
complicated structure - surely there is a very simple technique?

I hope soon to take part in some of the on-going discussions - I am very
interested in what people have recently contribute to the ideas on diffusion
of innovation. One of the ideas that we are playing with at present is the
use of complexity science as a tool or framework for understanding/promoting
diffsion of innovation in health care.


Ashley Woolmore
From: "Ashley Woolmore" <>

"Jim Hines"
Junior Member
Posts: 14
Joined: Fri Mar 29, 2002 3:39 am

Demographic Profiles and Health Care

Post by "Jim Hines" »

You want to use an "aging chain". An aging chain is simply a several
material delays hooked together. (A material delay is a decay with an
input - probably what youre using right now). So the input to the first
stock is whatever your current input is (births?), the output of the first
stock is the stock divided by the average time in that stock. The input to
the second stock is the output from the first stock, etc. The more stocks
you put together, the more the output will look like the second part of an
"n" (Assuming that all stocks except the first start at zero). The sum of
the delays will equal the average residence time in the chain; the delay for
any one stock will equal the average residence time in that stock. So you
could have three stocks: Young people, Middle aged people, and old people.
If you are young until thirty-five, the first time constant would be 35
years. If you are middle aged until you hit sixty, the second stock would
have a time constant of 25 years. If the average life expectancy of a 60
year old is to live until 90, the third time constant would be thirty years.

Jay Forresters rule of thumb is that a third order delay is usually
sufficient for the delay to communicate the proper dynamics to the rest of
the model. Jim Lyneis tells me that to get an "echo" (e.g. an upsurge in
births when the baby-boomers have babies) you need at least a sixth-order

Jim Hines
MIT and LeapTec

Bill Braun
Junior Member
Posts: 14
Joined: Fri Mar 29, 2002 3:39 am

Demographic Profiles and Health Care

Post by Bill Braun »

>normally death rate would be defined as the number of deaths/population. in
>itself this is not time dependant, but a constant representing the nature of
>the environment being examined (ie how good the health care is, the
>environmental hazards etc).

Ive struggled with this, as much a function of my mental model as one of
technical accuracy. Several issues, if discussed, would help with my

One, if we assume that life expectancy is constant and the distribution of
chronic and acute illness across age/gender stocks is constant, then using
death rates as a fraction of the population seems reasonable. What happens
if any or all of those change? Are the potential implications (moderated by
the dynamics the population model is driving) material to the insights the
overall model reveals?

Two, (correspondingly) social norms reflective of shifts in culture mix
could materially affect birthrates. For example, european,
african-american, hispanic, pacific rim, balkan, (name your culture) etc.
peoples have different attitudes toward pregnancy (family size, conditions
under which it takes place, etc). Assuming these attitudes remain constant
but the culture mix changes, are the implications likewise material? If the
attitudes themselves change (one or more cultures raise or lower their
reproduction attitudes), does that further complicate the mix?

Third, what if a sudden epidemic occurs that disproportionately affects the
population based on age and/or gender and/or culture mix, etc., etc.

Is sensitivity germane to this discussion (admittedly more or less relative
to the dynamics the population model is affecting and being affected by)
and is selecting the number of stocks a critical decision in designing the

Bill Braun
From: Bill Braun <>

"Jaideep Mukherjee"
Junior Member
Posts: 14
Joined: Fri Mar 29, 2002 3:39 am

Demographic Profiles and Health Care

Post by "Jaideep Mukherjee" »

After reading the many replies, I guess we come down to the usual sensible
agreement. We agree about the appropriate use of data, but that definition
of "appropriate"ness is a very subjective matter (one cohort, 4 or 100???).
I have read somewhere that not even using four cohorts in the World3 model
would not have changed the conclusions - an exogenous forcing function would
have done the trick too. (I used such simplification in my work to simplify
my life, realizing that one-cohort model was appopriate for a particular
purpose - dynamic game optimization analyses). But many times we reach such
conclusions only after doing the modeling, and not before.

To digress a bit, one of my biggest peeves with some styles of modeling is
when people "start" with unshakeable insights and causal-loop diagrams and
dont view the modeling process as an "iterative, incremental" development
which may or may not lead to any earth-shattering insights. (A) Not being
focussed on good data and modeling practice can lead to a tendency to
reinforce your own mental models and insights you started with, but (B)
being focused on data, reality checks, can actually lead to newer insights,
and after/during-the-fact creation of causal-loop-diagrams to communicate
high-level-knowledge. I believe the latter (B) to be the scientific process
anyway. Sometimes I feel the focus shifts too much to causal-loop
diagramming (BEFORE the modeling) and systems dynamics modelers become
trapped with their own mental models - client is wrong, we dont need more
data, they just dont get it, etc. etc. I believe, among other things, it
was "quantitative" modeling of fuzzy data that made system dynamics a very
powerful technique, not causal-loop diagrams.

I also believe that linear programming/modeling may sometimes give even
deeper insights than SD modeling - the insights not come just because we are
using SD, but they come whenever our mental models are challenged and that
can happen whenever we have more than 7+/-2 variables to deal with (refer to
George Millers article on how we cant mentally calculate anything beyond
7+/-2 variables in our head), hence the raison detre for modeling, linear
or nonlinear, in my view.

In case somebody thought so, I dont imply in this email nor did I imply in
my previous post that Prof Coyle, or any specific person, was plucking
numbers out of thin air, or indulges in the kind of practices I am
describing here (some people do and they are welcome to put forth their
rationale & arguments here). But I was asking a more serious question of
alternatives to census data. My take is this - if you dont have any info,
try to gather as much info by talking to people or use what is there (e.g.,
census data), if you have time, do surveys and get better info, if you just
dont have time, pluck numbers out of thin air and say so, and realize the
limitations of such a model. But if good data is availabe, use it whenever
you can. If technology/time allows it, use even fine-grained numerical data
(Vensim I believe has many techniques that allow you to use data in such a
way). We all agree that too much data-massaging can be counterproductive,
but I think as big a danger, much more so, lies in not using data when it is
available. I liked the approach taken in World3 models, as the population
models were built the way they were, after a very serious research and
thinking into different alternatives.

If we focused only on higher level insights, we would have been satisfied by
the minor variations in the speed of light (Michelson-Morley experiments),
we would have been happy with the slight variations in Mercurys orbit
(departing from Newtons predictions), and we would never have discovered
chaos theory (after all, it is only very slight perturbations in initial
conditions, isnt it!!). By intelligent use of data, and not denying the
above as random variations, we now have Einsteins theory of relativity, and
the whole edifice of chaos theory. Similar developments can happen in
applications of SD, especially in studies of structured environments
(current business environments in the developed world, as contrasted with
unstructured ones, such as crime environments in Russia right now).

To further illustrate my concerns:

My peppermint teabox here has a quotation: "We do not weave the web of life,
We are merely a strand in it, Whatever we do to the web, We do to
ourselves." - Chief Seattle, 1854

Very nice, and I admire the vision and how it is put. Many systems
thinkers/modelers would tend to agree with the above. It is a deep insight
which, when followed, can lead to a very different way of life than when it
is not followed. I agree to all of this. This is the essence of systems
thinking (well, sort of).

The problem begins when the above "web of life" degenerates to a spaghetti
of causal-loop diagrams and it gives a warm fuzzy feeling, without helping
us in any way in our decisions to make changes in business or public policy.
For true decision support, this "web of life" must be populated with real,
GOOD data (this is where I think system dynamics modeling comes in, after
systems thinking). By not denying nonlinearities, delays, feedbacks, fuzzy
knowledge, but by quantifiying them and making them precise, we (actually
Forrester) transformed systems thinking art into a system dynamics science.
The bad news is when SD degenerates into half-baked ST and masquerades as SD
science. It happens too often, in my limited eperience, and can give the
science of SD a bad name.

I bet I have opened the can of worms once again - and I must look for a
place to hide now:-)) Why do I invite trouble? Be kind, folks.


"Peter Heffron"
Junior Member
Posts: 14
Joined: Fri Mar 29, 2002 3:39 am

Demographic Profiles and Health Care

Post by "Peter Heffron" »

Im responding primarily to Jaideep Mukherjees interesting observations of
10 February 2000 regarding demographic modeling issues.

It seems to me that regardless of whether one starts with causal diagramming
or modeling, ones subjectivity, including her experience and world view,
will naturally influence the diagrams or the models shape. If one is
somewhat humble, reflective, and curious, then the causal diagramming and/or
the modeling process will spark questions and insights that will in turn
lead to changes in the diagram and/or model. Of course if one is not
sufficiently humble, reflective, and curious then, yes, the causal diagram
and/or model will merely reinforce what the person "knew all along."

In either case (humble/not humble), without the participation of others in
the modeling process, ones solo model should be more suspect than a truly
participatory model would be. Group models should of course be suspect (and
rigorously critiqued) too, but by selecting a modeling team that embodies a
variety of disciplines and perspectives, the odds are that there will be
fewer errors and greater robustness in models generated that way. But more
on participation later.

Although for the sake of variety it may be useful to try different modeling
approaches and sequences when one feels the need (i.e., model first, then
diagram, as Jaideep suggests), the "Modeling Process Guidelines" section in
the 1997 edition of "An Introduction to Systems Thinking," by High
Performance Systems, suggest this order:

"1. Define the Issue/Problem (Explicitly State the Purpose, Develop a
Reference Behavior Pattern, Develop a System Diagram)
"2. Develop and Represent Hypotheses (Seek a Dynamic Organizing Principle,
Map the Hypotheses, Make the Map Simulatable)
"3. Test Hypotheses (Mechanical Mistake Tests, Robustness Tests, Reference
Behavior Tests)
"4. Design and Test Policies (Policy Tests, Sensitivity Tests, Scenario
"5. Challenge the Boundaries (Extensive Boundary, Intensive Boundary...and
from the 1994 edition: Challenge Whats Been Left Out and Challenge Whats
Been Put In)
"6. Make Learning Available (Develop a Drama, Design a Learning Progression,
Implement the Progression, Create In-character Feedback and Coaching

The above modeling sequence is based on a great deal of collective lessons
learned/best practice over the years and has worked well for a great many

Although it is tempting to "just do it," if we skip the first step, above
("Define the Issue/Problem"), we may end up solving the wrong problem. If
the problem is that we dont know what the problem is, then it makes sense
to start in brainstorm mode as a precursor to step one.

To end, I feel any modeling effort by an individual, no matter how
intelligent she is and how expert in systems thinking and modeling she is,
the model will lack the balance, richness, and greater validity that comes
about through the participation of people of different socio-economic
backgrounds, different disciplines, different genders, and who, in some
cases, at least initially, disagree entirely with the premises held by the
coordinating modeler.

Thus, no matter how "good" the data is, the lone modeler behind her computer
printing out stocks and flows, versus the modeler facilitating a "messy"
(but following the six steps above) modeling workshop with 30 people of
various colors and persuasions, and where there is all kinds of action,
questions, debate-much of which is reflected in the model... Which approach
is better for gaining a better understanding of a system and for solving
real-world problems?

Thanks, Jaideep.


Peter Heffron
95 West Naauao Street
Hilo, Hawaii 96720 USA
Telephone: 808-934-0527

L J Wilkinson
Junior Member
Posts: 14
Joined: Fri Mar 29, 2002 3:39 am

Demographic Profiles and Health Care

Post by L J Wilkinson »

hi ashley

without seeing your actual model i can not really prove specific help but
there are a number of things to bear in mind

sd is very much about understanding and mental models. i wonder if your
mental model of the situation is reflected in your results?

normally death rate would be defined as the number of deaths/population. in
itself this is not time dependant, but a constant representing the nature of
the environment being examined (ie how good the health care is, the
environmental hazards etc).

however, if you are refering to the total number of deaths from the
population increasing with the number in the population then this is a
different situation. normally the rate of dying (ie the number of deaths
from the population in a given time period) is represented by the death rate
* population (ie a feed forward of information )

population can usually be thought of as a simple resevoir where people have
a residence time = life span . thus it would be modelled as having an
inflow (number of births per time period (influenced by the number in the
population and the birth rate), a residence time in the source and an
outflow of deaths per time period (influence by the death rate and the
number in the population) . thus the system has both feedforward and feed

please feel free to get back to me as i am interested in sd and the nhs.
incidentally i use stella 2


Larry J Wilkinson Room 321
Newcastle Business School Northumberland Building
University of Northumbria
Newcastle upon Tyne 0191 2274374

"geoff coyle"
Junior Member
Posts: 14
Joined: Fri Mar 29, 2002 3:39 am

Demographic Profiles and Health Care

Post by "geoff coyle" »

My good friend Keith Linard makes some interesting points about using arrays
to model census-based population data but, equally, his reply raises some
questions in my mind.

Lets look, first, at the hidden assumptions in such an apparently precise
model with 100 age categories and 2 sexes. One is that census data is never
accurate (about 10% too low, I seem to recall) and, in any case, it could be
anything up to 10 years out of date. Another is that not everyone in, say,
the 40 year-old category (Keith is much younger than that, of course) is the
same age. 1/365 of them will be 40 years and 1 day, 40 years and 2 days and
so on (Oh, dear, there are leap years as well) so they dont all move to
being 41 at the same time. Three, what kind of delays are involved?
Pipelines of 1 year or delay1 with 1 years delay? What is DT? Is this a
stiff system? How long is LENGTH?

I have a horrible feeling that such a model will be illusory in its image of

In any case, surely a model is a simplification of reality designed to
answer some well-defined questions, as Richardson and Pugh have so well
expressed it. What government policies might a population model need to
address? Here are a few which come to mind:

Should we restrict/promote immigration and if, so, in which categories of
people? The US might want to do the first and Australia the second but that
might need two very different models.

Should we promote health care for the old or should we restrict it so that
we save on the pension budget?

How can we finance pensions with an ageing population?

What provision might be made for higher education, should it be free, who is
going to pay for it?

All these might require different models and I doubt if they could be well
addressed by a model which starts off with 200 levels. One recalls the
beautiful elegance of the World Model - 7 levels and written (in pencil)
during an air journey. That model was designed to answer questions and we
lose sight of that approach at our peril. The software we now have is
immensely powerful but, like all string medicine, it needs to be used with


Professor Geoff Coyle
Consultant in System Dynamics and Strategic Analysis
Tel: (44) 01793 782817 Fax: 01793 783188

Keith Linard
Junior Member
Posts: 14
Joined: Fri Mar 29, 2002 3:39 am

Demographic Profiles and Health Care

Post by Keith Linard »

I agree with Geoff Coyles point (which cannot be emphasised too often)
that the question being addressed (or at least the underlying issues ...
which may indeed be quite different) should drive the modelling approach.

My suggestions re using a full age spectrum census model rather than the
"young-middleaged-elderly" concept espoused in many examples and texts
derives from 12 years preparing briefings for Government Ministers and for
the Cabinet Expenditure Review Committee in policy areas including
immigration, education, pensions, employment, health care etc. Age
specific targeting impacts and the likely age specific feedback impacts on
non-targeted (or excluded) groups, in both short term and longer term, are
of invariably of concern to Government, the parliamentary opposition and
lobby groups. My point was that there are advantages in using a common
population molecule among the building blocks of the policy model. That
way one does not have to reestablish credibility of that essential element
in each policy context. Obviously the rest of the model and the way it
impacts on the population molecule will vary according to the context and

Regarding model accuracy, were I ever to achieve +/- 10% I would be
ecstatic. My ex-post analyses of government benefit-cost studies show that
+/- 100% is par for the course for that respected genre of modelling. In
my experience the accuracy of the population census module of policy
models is generally of much lesser concern than the validity / accuracy
of the policy feedback modules and their impacts on population parameters.
That said, I remain to be convinced that simplistic
"young-middleaged-elderly" models give superior accuracy to an annual
census type model.

Keith Linard
Senior Lecturer
University of New South Wales
From: Keith Linard <>

"geoff coyle"
Junior Member
Posts: 14
Joined: Fri Mar 29, 2002 3:39 am

Demographic Profiles and Health Care

Post by "geoff coyle" »

This correspondence gets more interesting by the day. I bow to Keiths
practical experience in government work though it has not been the same as
mine. Usually, the people I work with are not too interested in the 49 year
olds as opposed to the 50s, 51s and so on. They want to talk about the
middle-aged, elderly, pensioners and so forth - they work in categories, not
annual cohorts - and the difference between UK and Australian work is

Keith says that +/- 10% accuracy would be great though I dont really
understand how that ties in to the apparent +/- 1% accuracy of a centennial
age model.

I am very impressed if the Australians get within +/- 100% in cost-benefit
work. The last major CB study in the UK was for another London airport and
Saxon churches, 1000 and more years old, which would have had to have been
destroyed to
build it were included at their fire insurance value! I share Keiths
cynicism at that genre of analysis.

Keith mentions simplistic "young-middle-aged-elderly" models but I dont
think one can say a priori that a given type of model is simplistic. The
question is whats appropriate to the issues. An age-group model is just
different from an annual census type model; I dont believe that either can
be said to be more accurate than the other without defining a context.

Keith, thanks for the debate. I get more and more convinced that data in
models would be a smashing topic for a top-flight PhD student.


Professor Geoff Coyle
Consultant in System Dynamics and Strategic Analysis
Tel: (44) 01793 782817 Fax: 01793 783188