Studying Control (was Help on Vensim model)

[From Rick Marken (981009.0940)]

Bill Powers (981008.1644 MDT) to Jeff Vancouver (981008.1422 EST)--

My point is that you can't really match a model to the data with
all these unknown factors influencing the results. You've
unwittingly chosen a task that is either trivially simple (if the
participant assumes nothing that isn't said), or extremely complex
and quite possibly beyond unaided human ability to carry out in
the best way.

My best counsel is to start SIMPLE and work up from there.

I like this "galilean" approach to research myself. But I think
that more "naturalistic" experiments like Jeff's can be useful
too, not because they demonstrate the accuracy of predictions of
the model but because they show how to think about and study
behavior as a control phenomenon.

Jeff's study let's Jeff _talk about_ behavior in control theory
terms. The subject's in Jeff's study are clearly doing a control
task; the subject's are asked to control the cost of the schedules
and they are given a means of controlling this variable. So
right off the bat, Jeff can talk about this research in terms
of _controlled variables_ instead of in terms of "variables
that influence scheduling behavior".

There are, of course, many variables each subject _might_ be
controlling. It's true that Jeff's experiment is not designed
to determine what variables each subject might be controlling,
but at least Jeff can _talk about_ the possibility that different
subjects are controlling different variables; he can mention what
some of the variables are that subjects _might_ be controlling.
So Jeff can talk about the fact that there are many possible
_controlled variables_ in the _same situation_; and he can mention
that this is true because controlled variables are _perceptual
representations_ of the subject's world.

Jeff can show that different people control different perceptual
aspects of the same situation because he tested _individual
subjects_. Since some subjects may simply be controlling schedule
cost and nothing else, a model controlling just this variable
should match the (admittedly trivially uninteresting) behavior of
these subjects quite well. If the model fails to fit the behavior
of most other subjects (as it surely will) then Jeff can point to
aspects of the behavior of these other subjects (including their own
verbal reports) which suggest that they were controlling variables
besides schedule cost alone.

This experiment can even be used to show why stimuli appear to
cause responses. The cost of the schedule appears to cause
subjects to delete shifts. But Jeff can show (at least for those
subject's controlling only cost) that these deletions are always
_exactly_ what are needed to get cost below the reference; if the
reference changes (the experimenter asks the subject to keep the
cost less than $20,000 instead of $14,000) than these same
"stimuli" no longer cause the same "response".

So I think the research Jeff is doing can be very useful and I
want to encourage Jeff to write it up and try to publish it. Few
psychologists seem to see the value of our "galilean" approach
to demonstrating a powerful principle (control of percpetion) in a
simple situation (various versions of tracking). Jeff's research
may not be clean and simple but it does demonstrate the principle
of control in a way that many psychologists would see as "relevant".

What I would like to see happen over the next few years is people
publishing papers that 1) _talk about_ the behavior of _individuals_
in control theory terms -- that talk about controlled variables,
disturbances, reference states, etc and 2) show how to _look at_
behavior in control theory terms, by "pushing" on potential
controlled variables and/or comparing the behavior of models that
control particular variables to the behavior of people who
might be controlling these variables.

Precise fit of the model to behavior is certainly important; the
galilean work is the real science. But I think work like Jeff's, if
it's written up properly, might be able to bring control theory,
in a non-contentionus, matter of fact (and fairly sneeky) way,
to a much larger audience.

Best

Rick

ยทยทยท

--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken

[from Jeff Vancouver 981008.1505 EST]

[From Rick Marken (981009.0940)]

Rick, if I wanted you to defend me, I would have asked. However, next time
I certainly will ask! Thanks for the support. You seem to actually
understand where I am coming from.

Precise fit of the model to behavior is certainly important; the
galilean work is the real science. But I think work like Jeff's, if
it's written up properly, might be able to bring control theory,
in a non-contentionus, matter of fact (and fairly sneeky) way,
to a much larger audience.

I will be sure to run a copy of it by you (and Bill and Mary). You all
were helpful with the self-regulation chapter.

Now all I need it more participants.

Sincerely,

Jeff

From [ Marc Abrams (981009.1956) ]

[From Rick Marken (981009.0940)]

Bill Powers (981008.1644 MDT) to Jeff Vancouver

(981008.1422 EST)--

My point is that you can't really match a model to the

data >>with all these unknown factors influencing the
results. >>You've unwittingly chosen a task that is either
trivially >>simple (if the participant assumes nothing that
isn't said), or >>extremely complex and quite possibly
beyond unaided >>human ability to carry out in the best way.

My best counsel is to start SIMPLE and work up from
there.

I like this "galilean" approach to research myself. But I

think

that more "naturalistic" experiments like Jeff's can be

useful

too, not because they demonstrate the accuracy of
predictions of the model but because they show how to think
about and study behavior as a control phenomenon.

In what way does this model do that?

Jeff's study let's Jeff _talk about_ behavior in control

theory

terms. The subject's in Jeff's study are clearly doing a
control task; the subject's are asked to control the cost

of >the schedules and they are given a means of controlling
this >variable. So right off the bat, Jeff can talk about
this research >>in terms of _controlled variables_ instead
of in terms of >>"variables that influence scheduling
behavior".

He can also talk about it in terms of compliance. Exactly
how does this model talk about how we control variables
rather then "variables that influence scheduling"? You can
"talk" about it. I just don't see this model supporting that
argument.

There are, of course, many variables each subject _might_
be controlling.

Each subject _is_ controlling for any number of variables.
Can we identify one or more at any one point in time?

It's true that Jeff's experiment is not designed
to determine what variables each subject might be
controlling, but at least Jeff can _talk about_ the

possibility >that different subjects are controlling
different variables; he >can mention what some of the
variables are that subjects _might_ be controlling.

Fine, What's the pupose of the model? It sounds like a case
study might be more appropriate. What is the model supposed
to show? What insights are to be gained?

So Jeff can talk about the fact that there are many

possible

_controlled variables_ in the _same situation_; and he can
mention that this is true because controlled variables are
_perceptual representations_ of the subject's world.

Sure can, and he doesn't need a complex SD model to show
that. Two rubber bands will do

Jeff can show that different people control different
perceptual aspects of the same situation because he tested
_individual subjects_. Since some subjects may simply be
controlling schedule cost and nothing else, a model
controlling just this variable should match the (admittedly
trivially uninteresting) behavior of these subjects quite

well.

This is what they were asked to do. When you do the rubber
band experiment and someone refuses to listen to your
instructions how do you determine what else they might be
controlling for? Does it matter? No, because the success of
the experiment ( showing that control exists ) is based on
that cooperation.

If the model fails to fit the behavior of most other

subjects

(as it surely will) then Jeff can point to aspects of the
behavior of these other subjects (including their own
verbal reports) which suggest that they were controlling
variables besides schedule cost alone.

Has the data supported this? What other variables were being
controlled instead? Why the non-compliance?

This experiment can even be used to show why stimuli
appear to cause responses. The cost of the schedule
appears to cause subjects to delete shifts.

That's how the model is formulated. Mangers delete shifts to
get the cost of schedules under budget. Do they have an
alternative method? What do you call it when an outside
agent sets a reference level you willingly comply with?

But Jeff can show (at least for those subject's controlling
only cost) that these deletions are always _exactly_ what

are >needed to get cost below the reference; if the

reference changes (the experimenter asks the subject to
keep the cost less than $20,000 instead of $14,000) than
these same "stimuli" no longer cause the same "response".

Unless of course the subject for other reasons refuses to
adopt the experimenter's reference level and adopts one of
his own.

So I think the research Jeff is doing can be very useful

and I

want to encourage Jeff to write it up and try to publish

it.

I would also like to encourage Jeff to continue, but i think
he needs to lower his horizons a bit to fit his modeling
skills. As his skills increase, so will the messages he will
be able to communicate. Jeff, it really took a set of balls
to jump in with both feet like that. For that I really
applaud your effort. Do't be so hard on yourself. Given
some thought I think you'll come up with a more comfortable
fit for yourself.

Few psychologists seem to see the value of our "galilean"
approach to demonstrating a powerful principle (control of
percpetion) in a simple situation (various versions of
tracking). Jeff's research may not be clean and simple but

it >does demonstrate the principle of control in a way that
many >psychologists would see as "relevant".

In my opinion, not with *this* particular model.

What I would like to see happen over the next few years is
people publishing papers that 1) _talk about_ the behavior

of >_individuals_ in control theory terms -- that talk about

controlled variables, disturbances, reference states, etc

and

A big Amen to that.

2) show how to _look at_ behavior in control theory terms,
by "pushing" on potential controlled variables and/or
comparing the behavior of models that control particular
variables to the behavior of people who might be

controlling >these variables.

So, who's stopping you. I thought that's what this whole
modeling effort with Vensim was about.

Precise fit of the model to behavior is certainly

important; the

galilean work is the real science. But I think work like

Jeff's, >if it's written up properly, might be able to bring
control >theory, in a non-contentionus, matter of fact (and
fairly >sneeky) way, to a much larger audience.

Another Big Amen to this one. :slight_smile: If Bill and Bob E, can
come up with a "PCT agent" that could be used and modified
for each model I think that would help us all spread the
word a lot easier.

Marc