Dog Behavior Illustrates PCT, but you have to read between the lines

Such a clear-cut example. The Post invites commentary on the web page.

Ted Cloak

Dog Learning.pdf (175 KB)

···

+++++++++++++++++++++++++++++++++++
"It must be repeated loudly and often that just as there is no reasonable
relationship between a service performed and the annual payment of $12
million to an executive or manager, there need not be any direct correlation
between a service rendered and a guaranteed payment to a worker (or would-be
worker) of an amount calculated to provide the necessities of life and a
modicum of dignity." -- Stephen J. Fortunato Jr.

[From Rick Marken (2007.06.07.1450)]

Such a clear-cut example. The Post invites commentary on the web page.

When I first glanced at the article I thought "yes, cool, this is a
clearcut example of control, showing that dogs will vary their actions
(using their mouth or, alternatively, their paw if their mouth is
full) as necessary to produced an intended result (getting the bar
down)". But then I read it and saw that it was about imitation. I
suppose it shows that dogs control for imitating under certain
conditions (as in the "nothing in mouth of model" condition) . But not
all dogs imitated when they were supposed to and not all dogs didn't
imitate when they were not supposed to imitate (in the "something in
mouth of model" condition). So I guess it shows that some dogs control
for imitating some of the time under certain conditions, which is
interesting, I suppose, but not exactly a clear-cut example of
control, in my opinion, anyway.

When it comes to dog research, I think the most clear cut example of
control comes from research on dogs catching Frisbees. This research
is discussed briefly at the end of a paper of mine on fly ball
catching that is on the net at:

Best regards

Rick

···

On 6/7/07, Ted Cloak <tcloak@unm.edu> wrote:
--
Richard S. Marken PhD
Lecturer in Psychology UCLA
Statistical Analyst VHA
rsmarken@gmail.com

[From Rick Marken (2007.06.07.2220)]

Rick Marken (2007.06.07.1450)--

When it comes to dog research, I think the most clear cut example of
control comes from research on dogs catching Frisbees. This research
is discussed briefly at the end of a paper of mine on fly ball
catching that is on the net at:

http://www.mindreadings.com/OpticalTrajRM.pdf

Well, since no one seems to want to talk to me I guess I'll talk to myself.

I just read the Optical Trajectories paper (referenced above) and man
it seems awfully good to me. Am I nuts or is that one nifty piece of
work? I don't understand why the fellow who wrote it isn't famous;-)

Admiringly,

Rick
(Author of Optical Trajectories paper)

···

--
Richard S. Marken PhD
Lecturer in Psychology UCLA
Statistical Analyst VHA
rsmarken@gmail.com

I just read the Optical
Trajectories paper (referenced above) and man

it seems awfully good to me. Am I nuts or is that one nifty piece of

work? I don’t understand why the fellow who wrote it isn’t
famous;-)
[From Bill Powers (2007.06.08.1005 MDT)]

Rick Marken (2007.06.07.2220)

It’s always pleasant to see that you’ve written something good, although
there’s always the chance that you are more easily impressed by your own
work than by that of others. Got to watch out for that.

I think it works better, in the long run, to read your work for errors
and things that might have been done better. There are two advantages to
doing this. The first is that the next time you write a paper you will
probably do it better, and the second is that if someone else finds
mistakes, you will already know about them instead of being the last to
find out. That will keep you from being defensive after the errors have
proven to be real, which is much more embarrassing when you finally have
to give in.

All that is not to indicate that I found a lot of errors in the paper,
but is simply a turgid and wordy way of saying what my crowd used to say
back in the early school years: “Don’t break your
arm.” That was understood to be followed by " … patting
yourself on the back." Bragging, in my crowd, was not only
unsuccessful, but drew hoots of laughter and mockery, more so if there
was good reason to brag. I remember producing a 60-foot putt that bent
left, right, and then left again and came to rest six inches to the left
of the cup and even with it: my friend’s comment was “You pulled
it.” A response of “Nice try” would have been an
insult, a condescending pat on the head. Complementing another person for
an achievement was recognized to be a subtle way of claiming to be even
higher on the ladder and in a position to make judgements. That went for
criticisms, too. “You pulled it” was a subtle, or maybe twice
subtle, compliment.

I’d love to see Martin Taylor’s Layered-Protocol analysis of that
social-linguistic convention.

So I have a few comments/questions.

p. 631. “the model … always looked straight ahead at the
ball.” What does “straight ahead” mean, especially in
conjunction with “at the ball”? The text indicates that the
plotted points are directions within the visual field with the gaze fixed
on home plate. But that contradicts the statement that the model looked
“at the ball.” If the plots showed the direction of the ball
with the fielder always looking in the lateral direction of the ball, the
angle beta would always be zero. If the fielder were actually looking at
the ball in the vertical direction as well, alpha would be zero, too. I
think that if you look at real fielders chasing a ball, most of the time
they are in fact looking at the ball, not at home plate.

p. 631. While you speak of differences between the behavior of your model
and the behavior of the Shaffer and McBeath model, you don’t show the
behavior of the Shaffer and McBeath model, or of the real fielder. It
troubles me that the critique of the S&M model is not made on the
basis of fits to data, but on its predictions relative to the behavior of
your model. Since there’s no indication of how well your model fits the
real data, perhaps the deviation of the S&M predictions from your
model’s behavior represents an improvement, not a defect. There is
nothing in the article to show that is not the case.

p. 631. I’m also bothered by calling the trajectories of Fig. 1
“linear”. They are curved, not linear, as shown by comparison
with the straight lines you drew in the figure. For the two outer curves,
the slope at the ends appears to be about half of what it is at the start
of the trajectories, at the bottom.

p. 631. The replotting of the angle alpha as tan(alpha) shows the
right-hand plot starting at an time of about negative 1/2 second instead
of at zero where the other two plots begin. The upward curvature of these
plots is an artifact of the tangent function, which goes to infinity at
an angle of 90 degrees, and becomes -infinity for slightly larger angles.
There’s no explanation of why the tangent function is introduced, or what
meaning would be given to the plot for angles great than 90
degrees.

p. 632. While the TCV is illustrated using your model, it doesn’t appear
to be applied in the paper to any real data. There is a qualitative
comparison with some frisbee-catching data, but with no figures or
numbers (but with a reference to a figure in another paper which the
reader then has to obtain, somehow). You seem to be using your model as
the “real data” against which the other models are
judged.

p. 632. Fig 3B appears to show that the angle gamma remains well below
0.01 degree (could you have meant radians?) whether there is a
disturbance or not. Plot B is described as “lateral displacement
(gamma) of the ball image from the line of gaze plotted over time for the
trajectories in panel A.” Panel A is described as the
“optical trajectory” of the ball with and without disturbances
of its path.

These two panels are contradictory for the same reasons that applied to
Fig. 1. The “direction of gaze” would have to track the
changing lateral angle of the ball shown in panel A within 0.01 degrees
in order produce the plot in panel B. But if it did that, then the
“optical trajectory” of the ball would remain within 0.01
degrees of the centerline of panel A. Clearly, the plotted variables
cannot both be what you say they are.

The main problem I have with this paper is its avoidance of any
meaningful reference to real data – all the plots seem to be
theoretical. This would be appropriate if the point were simply to
illustrate how to do the TCV, but there is frequent reference to the
details of the model (like the curvatures of the plots) which bring in
the question of which model fits the data better. The TCV depends
critically on comparing model performance to real performance; it would
have been useful to show how the various models compare in their fits to
real data, since that is what the TCV is all about.

The other problem is simply that the descriptions of the variables are
not consistent with the plots. I suspect that the referees did not read
this paper very carefully.

Best,

Bill P.

We need a way to change the heading when we change the subject.

Ted

···

From: Control Systems
Group Network (CSGnet) [mailto:CSGNET@LISTSERV.UIUC.EDU] On Behalf Of Bill Powers
Sent: Friday, June 08, 2007 12:03
PM
To: CSGNET@LISTSERV.UIUC.EDU
Subject: Re: Dog Behavior
Illustrates PCT, but you have to read between the lines

[From Bill Powers (2007.06.08.1005 MDT)]

Rick Marken (2007.06.07.2220)

I just read the Optical Trajectories paper (referenced above) and man

it seems awfully good to me. Am I nuts or is that one nifty piece of

work? I don’t understand why the fellow who wrote it isn’t famous;-)

It’s always pleasant to see that you’ve written something good, although
there’s always the chance that you are more easily impressed by your own work
than by that of others. Got to watch out for that.

I think it works better, in the long run, to read your work for errors and
things that might have been done better. There are two advantages to doing
this. The first is that the next time you write a paper you will probably do it
better, and the second is that if someone else finds mistakes, you will already
know about them instead of being the last to find out. That will keep you from
being defensive after the errors have proven to be real, which is much more
embarrassing when you finally have to give in.

All that is not to indicate that I found a lot of errors in the paper, but is
simply a turgid and wordy way of saying what my crowd used to say back
in the early school years: “Don’t break your arm.” That
was understood to be followed by " … patting yourself on the back."
Bragging, in my crowd, was not only unsuccessful, but drew hoots of laughter
and mockery, more so if there was good reason to brag. I remember producing a
60-foot putt that bent left, right, and then left again and came to rest six
inches to the left of the cup and even with it: my friend’s comment was
“You pulled it.” A response of “Nice try” would have
been an insult, a condescending pat on the head. Complementing another person
for an achievement was recognized to be a subtle way of claiming to be even
higher on the ladder and in a position to make judgements. That went for
criticisms, too. “You pulled it” was a subtle, or maybe twice subtle,
compliment.

I’d love to see Martin Taylor’s Layered-Protocol analysis of that
social-linguistic convention.

So I have a few comments/questions.

p. 631. “the model … always looked straight ahead at the ball.”
What does “straight ahead” mean, especially in conjunction with
“at the ball”? The text indicates that the plotted points are
directions within the visual field with the gaze fixed on home plate. But that
contradicts the statement that the model looked “at the ball.” If the
plots showed the direction of the ball with the fielder always looking in the
lateral direction of the ball, the angle beta would always be zero. If the
fielder were actually looking at the ball in the vertical direction as well,
alpha would be zero, too. I think that if you look at real fielders chasing a
ball, most of the time they are in fact looking at the ball, not at home plate.

p. 631. While you speak of differences between the behavior of your model and
the behavior of the Shaffer and McBeath model, you don’t show the behavior of
the Shaffer and McBeath model, or of the real fielder. It troubles me that the
critique of the S&M model is not made on the basis of fits to data, but on
its predictions relative to the behavior of your model. Since there’s no
indication of how well your model fits the real data, perhaps the deviation of
the S&M predictions from your model’s behavior represents an improvement,
not a defect. There is nothing in the article to show that is not the case.

p. 631. I’m also bothered by calling the trajectories of Fig. 1
“linear”. They are curved, not linear, as shown by comparison with
the straight lines you drew in the figure. For the two outer curves, the slope
at the ends appears to be about half of what it is at the start of the
trajectories, at the bottom.

p. 631. The replotting of the angle alpha as tan(alpha) shows the right-hand
plot starting at an time of about negative 1/2 second instead of at zero where
the other two plots begin. The upward curvature of these plots is an artifact
of the tangent function, which goes to infinity at an angle of 90 degrees, and
becomes -infinity for slightly larger angles. There’s no explanation of why the
tangent function is introduced, or what meaning would be given to the plot for
angles great than 90 degrees.

p. 632. While the TCV is illustrated using your model, it doesn’t appear to be
applied in the paper to any real data. There is a qualitative comparison with
some frisbee-catching data, but with no figures or numbers (but with a
reference to a figure in another paper which the reader then has to obtain,
somehow). You seem to be using your model as the “real data” against
which the other models are judged.

p. 632. Fig 3B appears to show that the angle gamma remains well below 0.01
degree (could you have meant radians?) whether there is a disturbance or not.
Plot B is described as “lateral displacement (gamma) of the ball image
from the line of gaze plotted over time for the trajectories in panel A.”
Panel A is described as the “optical trajectory” of the ball
with and without disturbances of its path.

These two panels are contradictory for the same reasons that applied to Fig. 1.
The “direction of gaze” would have to track the changing lateral
angle of the ball shown in panel A within 0.01 degrees in order produce the
plot in panel B. But if it did that, then the “optical trajectory” of
the ball would remain within 0.01 degrees of the centerline of panel A.
Clearly, the plotted variables cannot both be what you say they are.

The main problem I have with this paper is its avoidance of any meaningful
reference to real data – all the plots seem to be theoretical. This would be
appropriate if the point were simply to illustrate how to do the TCV, but there
is frequent reference to the details of the model (like the curvatures of the
plots) which bring in the question of which model fits the data better. The TCV
depends critically on comparing model performance to real performance; it would
have been useful to show how the various models compare in their fits to real
data, since that is what the TCV is all about.

The other problem is simply that the descriptions of the variables are not
consistent with the plots. I suspect that the referees did not read this paper
very carefully.

Best,

Bill P.