Ventana software support forum

This is indeed a very important consideration, especially when trying to
promote modeling to colleagues who are already skeptical. WE know that
curves of dynamic behavior can be useful in describing the general BEHAVIOR
of the sytem, sensitivity analysis, influence of feedback, and so on. But
if the Y-axis does not conform with the expected units of the naive
observe, they often interpret the results as academic diddling around.

Accomplishing this can be easier said than done!

Very often we can make reasonable assumptions about the *relative*
interactions between system components, but we have a difficult time
assigning *absolute* values to our parameters. This problem is compounded
by the fact that if those doing the research dont have a knowledge of SD,
then they dont collect the "correct" data. Then when we go to develop a
model, we cant find the "correct" parameters. Catch-22! A good reason
for raising the general appreciation of the modeling process, so that the
experiments and field work are designed correctly in advance.

This is particularly true in the biomedical research arena, where SD
modeling is not (yet) understood or appreciated.

>
>A friend suggested that I acquire the book, Lying with Statistics; I presume he
>meant that it would help me to avoid prevarication.

I believe the correct title is: "How to Lie With Statistics". I cant find
my copy, and dont recall the author. It is a classic, probably about 25
years old, and is excellent.

Despite the provocative title, it is not a subversive book with
instructions on how you can lie with statistics! It does do a good job of
describing how statistics can be so often misunderstood, both numerically
and graphically. An example includes newspaper symbols which describe
yearly housing sales by the size of the houses in the figure. However,
when housing goes up 2-fold, a symbol that is 2X wider and 2X higher,
actually leads to a symbol that is 4X larger (!), appearing to be a much
larger difference to the casual observer. Another example is truncating
the Y-axis. If murders go up from 100 to 105, and the scale shows from
95-110, quick visual inspection will suggest a HUGE increase. In most
cases, anyone with statistical background will not be fooled.

I recommend this book. It is easy to read and not too large. Maybe
equivalent to Strunk and White in the writing arena.

Edward J. Gallaher, Ph.D.
VA Medical Center Research Service (151W)
Portland OR 97201
(503) 220-8262 x6677
Assoc Prof Physiology/Pharmacology
and Behavioral Neuroscience
Oregon Health Sciences Univ.
gallaher@teleport.com

Jim,

I pondered your question about the data being true to scale and finally
decided the question was in the same category as:

"Have you stopped beating your children yet?"

Of course its true to scale, yet are the scales equivalent. I admit that I
havent had time to spend with Vensim as of yet (hopefully soon) yet I can
site from my presentation experience with ithink. Because ithink allows you
to plot up to 5 sets of data on one graph, and they can all have their own
scales, unless one if very carefull in defining the limits for the scales the
plots that you look at can be very deceiving relative to each other.

If I plot something on a scale from 0 to .2, like a percentage and something
else on a scale of 0 to 25, it can appear that a small change in one relates
to a major change in the other variable, and it may not be the case.

And if you set everything to the same sclae some of the changes may be so
small as to appear imperceptible...

I guess it just takes carefull thought and understanding.

Hope this makes some sense,
Gene Bellinger...
CrbnBlu@aol.com

Far better than How to Lie with Statistics is "The visual display of
quantitative information" by Tufte. It is the undisputed definitive
work on the topic, and a visual masterpiece in its own right.

John Sterman
JSterman@mit.edu

In demonstrating a system dynamics model in a client meeting, a fellow asked if
all of the data in a plot was "true to scale". I hadnt thought much about that
topic before and now wonder what an appropriate response might be.

The simulation output referred to was displayed in a Vensim Custom Graph. The
graph was used to illustrate inventory (cu.ft), processing capacity
(cu.ft/month), and changes in processing capacity (cu.ft/(month*month)). I
matched the scaling of variables with common units of measure, i.e., varaibles
with units "cu.ft/month" were displayed on a common scale. So, I guess the
clients question related to the scaling of the variables with non-common units.

A friend suggested that I acquire the book, Lying with Statistics; I presume he
meant that it would help me to avoid prevarication. Another colleague mentioned
that software, perhaps DYNAMO, had some "truth in scaling" capability. But Im
curious whether there is a rule or rule-of-thumb that one should employ under
the circumstances I described?

Jim Thompson
Gemini Consulting, Inc.
55 Reservoir Road
Farmington, CT 06032
phone 860/676-8152
fax 860/676-8052
e-mail 73424.1506@compuserve.com