Modeling

CSGnet Archive CSG1998
1

From [ Marc Abrams (980817.1200) ]

In talking with Rick about converting his spreadsheet model in _Mind
Readings_ to a System Dynamics model in Stella/Vensim I came up with
an idea I think might be helpful for some of us on CSG. I would like
to propose a modeling tutorial on CSGnet. The purpose would be to
construct a simple control process model using _either_ a spreadsheet,
a System Dynamics software tool, or a regular programming language.
The purpose would be to get people to understand and possibly be able
to think in terms of model building rather then simply philosophical
discussions. It could provide a basis of understanding ( not unlike
the coercion spreadsheet model ). It would also bring to the front the
_non-Triviality_ in model building and provide a way for non model
builders to contribute to the modeling effort by understanding _what_
might and what might not work in a modeling environment.
My personal preferences are with a System Dynamics software tool.
Vensim PLE ( Vensim Personal learning Environment ) is free and is
available for both the Mac and PC. I also believe the SD software is
more extendable by non- experts. There _is_ a bit of a learning curve
involved but not any more or less then in the alternatives ( that is
spreadsheet, and Programming language ). I have instructional material
in both systems thinking ( i.e. feedback loops ) and SD modeling.

Bill, Rick, CSGers, what do you think?

Marc

2

From [ Marc Abrams (980818.1008) ]

For those that are interested the Vensim web site is:

http://www.std.com/vensim/venple.html

I would also recommend that you go to the ASU site ( there is a link
on the page for it. ) and download the tutorial for Vensim PLE.

I am not a mathematician, nor an expert modeler. But the graphical
interface and "systems" paradigm provides a basis for me to understand
how components of a system interact and the behavior that the
structure causes. Hopefully with master modelers Bill, Rick, and
others as yet unknown ( to me at least :slight_smile: ), Models will be ongoing
works in progress. ( At least thats a hope I have ) extendable and
understandable to all willing to put a _little_ time and effort into
learning some of the basics. If i can be of _any_ help to anyone with
regard to this you can reach me at 718-266-0540 anytime before 10:30
est. and at msa@panix.com

Marc

3

[from Jeff Vancouver 980824.1005 EST]

From [ Marc Abrams (980821.1640) ]

Jeff, what is surprising is that the paradigm ( SD ) is supposed to be
based on the notion of feedback ( both positive and negative ). But if
you look through the literature the behavioral models are all
_chained_ S->R models These are event based ( variables ), and through
a series of relationsips with other variables come back to form
"loops". These loops are positive ( growth and decay ) and negative
goal oriented ) based on the relationships between the variables. The
notion of cybernetics and control ( purposeful behavior ) is _not_
dealt with at all. Behavior has been treated as a linear input, output
process. _Implementing_ SD solutions have always been difficult and
problematic because ( in my opinion ) the behavioral models ( i.e.
reinforcement, reward & punishment, and assorted S-> R philosophies )
have dominated the thinking.

I am not sure this is exactly accurate. Richardson (one of Forrester's
students) has an excellent book on the relationship between SD and
cybernetics (FEEDBACK THOUGHT IN SOCIAL SCIENCE AND SYSTEMS THINKING,
University of Pennsylvania Press, 1991). It includes a section on PCT.
Highly recommended reading.

My understanding, which is always open to reorganization, is that control
systems (goal-based negative feedback loops) are merely one mechanism in
which "stability, dispite instability" arises. That is, Bill P. and von
Bertalanffy both start with the above quoted observation of much of nature,
however Bill (from Wiener) describes one mechanism that can account for the
observation, and used it to develop an elaborate model of human behavior.
Other mechanisms clearly exist. In this sense, control is merely a subset
explanation for systems that display dynamic homeostasis. Von Bertalanffy
and the subsequent SD movement was more interested in tracking (modeling)
the observations. They have been less interested (because it is more
difficult I suspect) to understand the structure that is responsible for
the observations. That said, I have seen attempts to understand; some
control based, some not. Sterman is an example. He has goals in his
models, but I am not sure they are operationalized in control systems.
Part of the reason though, is that these people are generally not modeling
humans, but physical or social systems.

I think that Bill now has the tools to
communicate in a language many systems people ( Systems Thinking is a
_very_ popular business fad right now :slight_smile: ) can understand. I am
excited about that capability. With the DSS software Bill ( if he
wants :slight_smile: ) can develop models. make animated user front ends ( model
ouputs that can be displayed in a variety of ways ) and compile and
distribute them. What his pendulum model might lack in a Vensim
version in programming elegance would be made up for in pedagogical
value.

On this I would agree.

I also think it is important to reiterate Bill P.'s point about the
difference between the modeling PCTers are use to and the SD modeling
paradigm. Among the other things I need to do today (this week) is to get
my APA talk on my web site. That uses an example of Bill's to illustrate
the point.

Sincerely,

Jeff

4

[From Bruce Abbott (990825.1945 EST)]

Bill Powers (990824.09512 MDT) --

I'm particularly hoping that
Bruce Abbott will join in, not because I think his ideas are such hot
stuff, harumph, but because he is the most organized programmer among us
and may be able to teach us some tricks.

May I have a show of hands as to who is willing to participate actively in
the "Crowd" thread, and pledge not to get distracted by anything else on
CSGnet?

Actually I think I will join in, not because I think that Bill Powers' ideas
are such hot stuff, harumph, but because I think I might learn something of
value by so doing.

Do we have a quorum?

Bruce

5

[From Bill Powers (990826.1516 MDT)]

Bruce Abbott (990825.1945 EST)--

Actually I think I will join in, not because I think that Bill Powers' ideas
are such hot stuff, harumph, but because I think I might learn something of
value by so doing.

Do we have a quorum?

You bet. Welcome.

Best,

Bill P.

6

[From Bruce Abbott (2016.09.17.0910 EDT)]

Vyv, thanks for the link to the cognitive modeling paper. The advantages (and disadvantages) of computer modeling that the authors describe apply equally well to the modeling we do from a PCT perspective, and the authors offer a number of pieces of good advice concerning model construction and testing that are worth reading.

One section I found noteworthy is “3. Hints for a good modeling practice,” beginning on page 137. The first subsection covers model validation and discusses methods used to determine “goodness of fit”: how well the model’s behavior matches the empirical data. I know that this was an issue Bill Powers grappled with without resolving it entirely to his satisfaction. Correlations between the model’s behavior and that of the participant can be high even when there are obvious differences between the two. The two lines could differ by a constant amount, have different slopes, or both. Using root mean-squared error (RMSE) avoids this problem but can produce a large error when model and data differ only by some linear parameter (slope and/or intercept). Consequently Bill always included both a measure of correlation ® and RSME to assess goodness of fit. This practice is the first one recommended in the model validation section of the paper.

Bruce

···

From: Huddy, Vyv [mailto:v.huddy@ucl.ac.uk]
Sent: Saturday, September 17, 2016 5:37 AM
To: csgnet@lists.illinois.edu
Subject: Re: The straits of W.T.Magellan

Dear Dag,

You’ve said “descriptions rule” in psychology. I don’t know if you are aware but, to my knowledge, every psychology undergraduate gets core teaching in functional modelling research (e.g. connectionist models). Some courses provide classes where students build and test their own functional models of behaviour. This tradition emerged around the same time as Bill Powers papers in Byte and is now very mainstream. There is a summary here:

http://www2.units.it/delmisfa/papers/CognitiveModeling2007.pdf

The difficulty is that the majority of this work, from what I can tell, is from a linear causal point of view so is problmatic for understanding control.

Best wishes

Vyv

7

[Dag Forssell (2016 0917 09:30 PST)]

[Huddy Vyv (2016 0917 10:37
BST)] Datestamp
suggested

Huddy,

Thanks for this. I agree with Bruce’s ansver that the front discussion
applies to PCT very nicely. Regarding item 4. The papers in this issue,
it seems to me that each concerns a small bit of psychological
description and attempts to create a mathematical model to reproduce the
result.

You say this is mainstream. Why have these people not discovered PCT?

Tim Carey told me that when he took classes toward his doctorate, and
wanted to have some fun, he would ask the teacher making a claim:
“How does that work?”.

My impression from 26 years of reading CSGnet is that the answer will be:
“It is very complex.” Meaning: “I don’t have a
clue.”

When you say that this is mainstream, are you trying to tell me that the
vast majority of psychologists (and other life scientists) think in terms
of functional models–not meaning flow charts with boxes and arrows? If
undergrads get core teaching in functional modelling research, does this
mean that they get it, pay attention, and demand any kind of rigor in the
other courses?

That is not my impression. I see models and hear of models. It has never
been more than flowcharts filled with wishful thinking.

Seems to me that every study that hits the news correlates one
description with another, and at a modest percentage. I have never seen
news of a functional model hit the news. PCT has not hit the news either,
and that is too bad.

My interpretation is that the ten year old paper you linked is terrific,
pointing in the right direction, but a fringe effort, nothing that has a
major influence on psychologists of today.

On page iii of

http://www.livingcontrolsystems.com/download/pct_readings_ebook_2016.pdf
I wrote about my new cover illustration:

“Once you understand PCT, you realize that control is the
fundamental process of life. All living organisms control as long as they
live and when control ceases, life ceases. Control and living are
inexorably intertwined. Thus: I control�-therefore I live�-therefore I
control�-therefore I live…”

You say “is problmatic for understanding control”. As a
confirmed PCTer, to me that means “is problmatic for understanding
anything about life”.

I still think descriptions rule throughout this field.

Please show us more :slight_smile:

Best, Dag

···

At 02:37 AM 9/17/2016, Huddy Vyv > wrote:

Dear Dag,

You’ve said “descriptions rule” in psychology. I don’t know if
you are aware but, to my knowledge, every psychology undergraduate gets
core teaching in functional modelling research (e.g. connectionist
models). Some courses provide classes where students build and test their
own functional models of behaviour. This tradition emerged around the
same time as Bill Powers papers in Byte and is now very mainstream. There
is a summary here:


http://www2.units.it/delmisfa/papers/CognitiveModeling2007.pdf

The difficulty is that the majority of this work, from what I can tell,
is from a linear causal point of view so is problmatic for understanding
control.

8

[From Rick Marken (106.09.17.1150)]

···

Dag Forssell (2016 0917 09:30 PST)

Huddy Vyv (2016 0917 10:37
BST)]

I don’t know if
you are aware but, to my knowledge, every psychology undergraduate gets
core teaching in functional modelling research (e.g. connectionist
models).

The difficulty is that the majority of this work, from what I can tell,
is from a linear causal point of view so is problmatic for understanding
control.

DF: You say this is mainstream. Why have these people not discovered PCT?

RM: Psychologists who understand modeling (and, for that matter, control theory or physics or mathematics) don’t necessarily discover PCT and even if they do they don’t necessarily apply their knowledge of these things to PCT correctly. I think that you do need to be reasonably skilled at modeling and math in order to be able to understand PCT correctly; but the sine qua none for understanding PCT is to be willing to approach it without preconceptions about how behavior works or how to study it.

Best

Rick

–
Richard S. Marken

“The childhood of the human race is far from over. We
have a long way to go before most people will understand that what they do for
others is just as important to their well-being as what they do for
themselves.” – William T. Powers

9

Vyv Huddy (2016 0917 21:01 BST)]

[Dag Forssell (2016 0917 09:30 PST)]

DF: When you say that this is mainstream, are you trying to tell me that the vast majority of psychologists (and other life scientists) think in terms of functional models–not meaning flow charts with boxes and arrows?

VH: No I’m not saying that.

DF: If undergrads get core teaching in functional modelling research, does this mean that they get it, pay attention, and demand any kind of rigor in the other courses?

VH: I think it is a big enough part of the curriculum for almost all to get some exposure to it and have the opportunity to explore further. For example, the Eyesenck and
Keane textbook Cognitive Psychology: A Student Handbook (recomended by many courses for their Cognitive psychology module) has a secton in the opening chapter on approaches to cognition subtitled “computational cognitive science”.
Modeling isn’t, however, even close to being the dominant paradigm. I felt some balance was warranted to the earlier comments.

···

From: Dag Forssell csgarchive@pctresources.com
Sent: 17 September 2016 17:57
To: csgnet@lists.illinois.edu
Subject: Modeling

At 02:37 AM 9/17/2016, Huddy Vyv wrote:

Dear Dag,

You’ve said “descriptions rule” in psychology. I don’t know if you are aware but, to my knowledge, every psychology undergraduate gets core teaching in functional modelling research (e.g. connectionist models). Some courses provide classes where students build
and test their own functional models of behaviour. This tradition emerged around the same time as Bill Powers papers in Byte and is now very mainstream. There is a summary here:

http://www2.units.it/delmisfa/papers/CognitiveModeling2007.pdf

The difficulty is that the majority of this work, from what I can tell, is from a linear causal point of view so is problmatic for understanding control.
[Dag Forssell (2016 0917 09:30 PST)]

[Huddy Vyv (2016 0917 10:37 BST)] Datestamp suggested

Huddy,

Thanks for this. I agree with Bruce’s ansver that the front discussion applies to PCT very nicely. Regarding item 4. The papers in this issue, it seems to me that each concerns a small bit of psychological description and attempts to create a mathematical model
to reproduce the result.

You say this is mainstream. Why have these people not discovered PCT?

Tim Carey told me that when he took classes toward his doctorate, and wanted to have some fun, he would ask the teacher making a claim: “How does that work?”.

My impression from 26 years of reading CSGnet is that the answer will be: “It is very complex.” Meaning: “I don’t have a clue.”

When you say that this is mainstream, are you trying to tell me that the vast majority of psychologists (and other life scientists) think in terms of functional models–not meaning flow charts with boxes and arrows? If undergrads get core teaching in functional
modelling research, does this mean that they get it, pay attention, and demand any kind of rigor in the other courses?

That is not my impression. I see models and hear of models. It has never been more than flowcharts filled with wishful thinking.

Seems to me that every study that hits the news correlates one description with another, and at a modest percentage. I have never seen news of a functional model hit the news. PCT has not hit the news either, and that is too bad.

My interpretation is that the ten year old paper you linked is terrific, pointing in the right direction, but a fringe effort, nothing that has a major influence on psychologists of today.

On page iii of
http://www.livingcontrolsystems.com/download/pct_readings_ebook_2016.pdf I wrote about my new cover illustration:

“Once you understand PCT, you realize that control is the fundamental process of life. All living organisms control as long as they live and when control ceases, life ceases. Control and living are inexorably intertwined. Thus: I control­-therefore I live­-therefore
I control­-therefore I live…”

You say “is problmatic for understanding control”. As a confirmed PCTer, to me that means “is problmatic for understanding anything about life”.

I still think descriptions rule throughout this field.

Please show us more :slight_smile:

Best, Dag

10

Vyv Huddy (2016 0917 21:18 BST)]

···

[From Bruce Abbott (2016.09.17.0910 EDT)]

BA: Bill always included both a measure of correlation ® and RSME to assess goodness of fit. This practice is the first one recommended in the model validation section of the
paper.

VH: This is really helpful Bruce. Is there anything in Bill Powers writings about assessing fit? Or elsewhere in PCT approaches to modelling?

Bruce

From: Huddy, Vyv [mailto:v.huddy@ucl.ac.uk]
Sent: Saturday, September 17, 2016 5:37 AM
To: csgnet@lists.illinois.edu
Subject: Re: The straits of W.T.Magellan

Dear Dag,

You’ve said “descriptions rule” in psychology. I don’t know if you are aware but, to my knowledge, every psychology undergraduate gets core teaching in functional modelling research
(e.g. connectionist models). Some courses provide classes where students build and test their own functional models of behaviour. This tradition emerged around the same time as Bill Powers papers in Byte and is now very mainstream. There is a summary here:

http://www2.units.it/delmisfa/papers/CognitiveModeling2007.pdf

The difficulty is that the majority of this work, from what I can tell, is from a linear causal point of view so is problmatic for understanding
control.

Best wishes

Vyv

11

[From Rick Marken (2016.09.17.1340)]

···

Vyv Huddy (2016 0917 21:18 BST)]

Bruce Abbott (2016.09.17.0910 EDT)

BA: Bill always included both a measure of correlation (R) and RSME to assess goodness of fit. This practice is the first one recommended in the model validation section of the
paper.

VH: This is really helpful Bruce. Is there anything in Bill Powers writings about assessing fit? Or elsewhere in PCT approaches to modelling?

RM: I discuss this a bit in the paper entitled “Testing for Controlled Variables” which is reprinted as Ch. 4. in “Doing Research on Purpose”. In that research, measuring goodness of fit in terms of RMS error rather than R^2 was crucial; both models fit the data equally well in terms of R^2 but there was a clear advantage of one model over the other (the difference between the models being in the perception controlled) when fit was measured as RMS error.

Best

Rick

Bruce

From: Huddy, Vyv [mailto:v.huddy@ucl.ac.uk]
Sent: Saturday, September 17, 2016 5:37 AM
To: csgnet@lists.illinois.edu
Subject: Re: The straits of W.T.Magellan

Dear Dag,

You’ve said “descriptions rule” in psychology. I don’t know if you are aware but, to my knowledge, every psychology undergraduate gets core teaching in functional modelling research
(e.g. connectionist models). Some courses provide classes where students build and test their own functional models of behaviour. This tradition emerged around the same time as Bill Powers papers in Byte and is now very mainstream. There is a summary here:

http://www2.units.it/delmisfa/papers/CognitiveModeling2007.pdf

The difficulty is that the majority of this work, from what I can tell, is from a linear causal point of view so is problmatic for understanding
control.

Best wishes

Vyv

–

Richard S. Marken

“The childhood of the human race is far from over. We
have a long way to go before most people will understand that what they do for
others is just as important to their well-being as what they do for
themselves.” – William T. Powers

12

[Martin Taylor 2016.09.17.18.54]

See attached for one.

Martin

CSG2005bFittingData.ppt (314 KB)

···

On 2016/09/17 4:20 PM, Huddy, Vyv
wrote:

        Vyv Huddy (2016 0917 21:18

BST)]

              [From Bruce

Abbott (2016.09.17.0910 EDT)]

              BA: Bill always

included both a measure of correlation (R) and RSME to
assess goodness of fit. This practice is the first
one recommended in the model validation section of the
paper.

              VH: This is

really helpful Bruce. Is there anything in Bill Powers
writings about assessing fit? Or elsewhere in PCT
approaches to modelling?