Re[4]: My research on controlled variables

[From Chris Cherpas (980320.1935 PT)]

Bruce Abbott (980320.2005 EST)--

You conduct a typical tracking study, except that you collect data
continuously for 60 minutes rather than the more typical 1 minute.

Why would you collect data for 60 minutes? The kind of study you
describe _requires_ these long periods of time to get some results
given the apparatus and contingencies. This is not precision target
shooting, it's blasting at patterns of moving targets with a shot gun.
It's research, yes, but seems focused on a level of temporal integration
that would ideally presuppose we don't have to worry much about the very
low levels of control. How about just explaining how the rat presses
the bar first?

Your participant is nicely moving the cursor around via the mouse, keeping
the cursor close to the target. After a few minutes of this, he lets go of
the mouse and loosens his belt (which apparently has become uncomfortable) a
little, then grabs the mouse again and resumes controlling. Ten minutes
later, he gets up from the desk and disappears into the mens' room. A
minute or so after this, he returns to the desk and resumes controlling the
cursor. Then the phone rings and he answers it. As he talks with the voice
on the phone, his attention wanders and so does the cursor. He hangs up the
phone and good control returns.

I agree that this is a big problem (one of them) with the macro-behavioral
level at which the research is directed. The Test becomes much more
difficult with all this other stuff going on, does it not? The point is:
what is it that you want to study? I'm not saying these rat
studies shoot for something as grandiose as what proponents of the recent
"cultural" discussions seem to think is feasible as basic science in the
current state of PCT, but we're not taking advantage of the experimental
approach if we don't make every effort to build complex findings on
simpler ones.

You analyze the data and discover that there are periods when control is
excellent, periods when it is absent, and periods where it is there but at a
rather poor level. The average correlation of disturbance and output is
only 0.85.

OK, let's focus on those periods where control is excellent. Does the
experimental set-up and data collection allow you to do that? I see
big averages over hours and,from just that, I don't know when the rat
is loosening its belt and when its controlling for something about
a signaled shock condition.

But it is not so absurd to imagine little disruptions like this occurring
when your participant is a rat, who neither understands nor cares that you
want it to focus exclusively on the task you have set for
it during the period of the test.

Fact is, working with a rat isn't such an easy thing after all. I respect
that you (perhaps alone?) are trying to find ways of doing PCT studies with
non-verbal organisms. However, beating on that bar over and over again is
pretty messy as a basic datum for studying control other than in limited
domains (at least that's my current conclusion).

If you think you are going to get 99+ under these conditions, you better
hope for a miracle. A lot can happen in an hour.

I think we agree. So, why _set up_ a situation that requires lucky charms?

Under the most favorable schedule conditions, the rats in my changeover
studies typically spent between 85-95% of session time in the signaled
condition when given the means to do so, despite being disturbed back into
the unsignaled condition at the end of every minute they spent in the
signaled condition.

Despite...? But that's the experiment. What are we talking about in
terms of sessions length? Six hours? What's 5-15% of 360 minutes?
That would be 18-54 minutes hanging out in the unsignaled condition
-- have I got that right? All that time, much like the blanks
on the Nixon tapes, may be more trouble than it's worth to explain away.
Of course, I don't have an alternative at the moment and at least you're
talking about data, while I'm mostly talking about other talk. But
enough about me, let's get back to those critters...

Most often the rat was near the response lever when the houselight went
out, and ran immediately to the lever and pressed it.

You're saying near, but far enough to have to run, right?

An exception was during the first 5-20 minutes or so of the session, during
which time a phenomenon labeled "warmup" occurred, in which even a
well-practiced rat would tend to just stand there rather than approaching
and pressing the lever.

Surrre, I can understand that. I just don't think the rats had the means
to _ever_ warm up to the point where control might be studied in detail,
given the apparatus and contingencies.

Once the first lever-press occurred, however, the rats generally continued
to respond smoothly throughout the remainder of the session (as shown nicely
by our event-records). Each changeover was produced cleanly by a single press
of the lever, once the rat reached the from wherever it (the rat) happened to
be standing when the changeover period ended (usually at or near the lever but
occasionally not).

And did they stop pressing the lever after that single, condition-changing
lever press? Or did they keep banging away? Could a lever press, which is
not the single one required to get into the signaled condition, smell as
sweet (apologies to The Bard)? What if the rats had the opportunity to
lever-press _out_ of the signaled condition? How much would they do
that do you suppose?

There is nothing particularly difficult about the procedure and the rats
typically acquired the changeover response during the first changeover session,
although it usually took more sessions before it was being produced with
maximum efficiency.

I don't deny that it's interesting that they did this, or that the
studies you report do not in fact make this interesting thing observable.
I just wonder whether yet another case of something interesting can be
systematically broken down into an experimental analysis of...control.
I don't think the rats _ever_ reached maximum efficiency (to use your
words, perhaps unfairly) under these conditions.

Another personal motivation angle (mine): I suppose there aren't a lot of
physics experiments involving interesting phenomena you can actually run
literally in a vacuum, even though laws are stated as if you could -- so my
criteria for evaluating studies of control are probably not as severe
as I'm making them sound. But I'm personally questioning what I really
learned from the pigeon studies I ran (mostly matching law stuff) about
control systems, and I suppose your posts are catching some of the spill-over.
In any case, I would have to agree with the statement in Abbott & Badia (1979):
"Apparently the situation is more complex than either the preparation or
safety analyes suggest." But where do we go from here?

Your suggestion that it takes a PhD rat to handle the complexities of this
procedure does not stand up to the cold, hard evidence.

For me, it's not the rat's abilities I question (although, at some point,
it's clearly the fact we're trying to establish), it's the ability of
the experimental set-up to help us learn about the rat's abilities.

Under the circumstances, 85-95% of time in the signaled condition represents
excellent control.

Under the circumstances, 85-95% of time in the signaled condition
represents an interesting pattern, but I couldn't say how this is
explained in terms of control. I suspect these TEAB-style studies,
where powerful, uncontrollable disturbances are involved, may eventually
have an interpretation in terms of some parameters of intrinsic variables,
but there's too much noise in the data to say a lot about particular
control systems. I could study the speed of the car someone is driving
and relate it to stop lights, speed zones, and the like, but a lot of
detail would be missing about the control of perceptions involved in
driving. What do you think?

Best regards,