New Operant Data

CSGnet Archive CSG1995
1

[From Rick Marken (950805.1800)]

Here are the results of an operant conditioning study that I just
found in the Journal of Jewish-Catholic Interaction (Tumarkin and
Westerschulte, 1995). The study was conducted on two tracking
deprived rats. S1 was a Rattus Yidishkite and S2 was a Rattus
Germanicus Catholicus. The researchers point out that these species
are particularly well suited for this kind of study; Rattus Yidishkite is
known for its love of Talmudic tracking tasks and, of course, Rattus
Germanicus Catholicus just follows orders. (Actually, S2 was 1/2 Yidishkite
and 1/2 Germanicus Catholicus; he followed orders but felt guilty about
it).

Both rats did a tracking task in which they were to keep a cursor aligned
with a stationary target by pressing the button on the mouse (not related
to either subject). S1 (the lover of tracking tasks) didn't have to be
told to do this; as soon as this rat saw the cursor he knew that he had to
get it under the target. S2 did this tracking task as soon as he was
ordered to "keep the cursor under the target" (he's such a mensch, that S2).

If the subjects had done nothing the target would have drifted away
from the target, to the left, at a constant rate; by pressing the mouse
button a sufficient number of times the subject could move the cursor
to the right. Each movement of the mouse to the right is a reinforcement.
The number of presses required to produce a reinforcement was varied in
the experiment; this number is called the "ratio". Three ratios were used
in the experiment: 2,5 and 10. When the ratio was 2 the rat had to press
twice to get a reinforcement; when the ratio was 5 the rat had to press
five times to get a reinforcement, etc.

The same ratio was in effect throughout an experimental run which
lasted 30 seconds (how long do you think you can keep these rats
doing this stuff?). The rats were tested under "normal" conditions
in which the cursor started near the target and reinforcements were
large enough to allow the rat to keep the cursor near the target
at all ratio requirements. The rats were also tested under "deprived"
conditions in which the cursor started many pixels to the left of
the target and reinforcements were too small to allow the rat to
move the cursor to the target.

The results (in terms of responses/sec and reinforcements/sec) are
presented for each ratio in each condition for the two subjects:

                           Normal Deprived

                        Resp/ Reinf/ Resp/ Reinf/
Ratio sec sec sec sec

2 S1 ND 1.92 .96 6.1 3.03
                D 1.4 1.06 5.5 4.1
        S2 ND 1.92 .96 5.6 2.8
                ND 1.32 1.08 5.6 4.3

5 S1 ND 4.62 .96 8.47 1.69
                D 3.29 .96 9.1 2.68
        S2 ND 4.74 .96 9.2 1.8
                D 3.42 .96 8.6 2.6

10 S1 ND 8.89 .90 9.8 .96
                D 6.56 1.0 10.0 1.39
        S2 ND 8.52 .78 10.1 .99
                D 7.26 .90 10.3 1.43

A nice feature of this experiment (a feature rarely seen in JEAB but,
apparently, more common in JJCI) was that the subjects were tested
with and without disturbances.

The rows labelled ND are data collected when there was no disturbance
present. The rows labelled D are data collected when there was
a disturbance present. The disturbance was a reinforcement that
was added on a random 1/2 of the trials after a ratio requirement
was completed. This added reinforcement was taken into account in
the calculation of reinforcements/sec. Both reinforcements/sec
and responses/second were calculated by counting the total number
of reinforcements or responses that occurred during a 30 second trial
and dividing by 30. So the total number of reinforcements (including
those randomly added in the D condition) and the total number of
responses in a 30 second session can be obtained by multiplying
the numbers in the table above by 30.

The people who did this research didn't know control theory; they were
just studying the effect of ratio and disturbance on behavior. There
seemed to be an effect of both in both normal and deprived conditions.

Would anyone like to interpret these results from a PCT perspective?

Best

Rick

2

[From Bruce Abbott (950806.1445 EST)]

Rick Marken (950805.1800) --

Here are the results of an operant conditioning study that I just
found in the Journal of Jewish-Catholic Interaction (Tumarkin and
Westerschulte, 1995). The study was conducted on two tracking
deprived rats. S1 was a Rattus Yidishkite and S2 was a Rattus
Germanicus Catholicus.

I was thinking about doing something like this myself. The question is, how
to implement it so as to reproduce conditions like those facing the rats.
I'm wondering at the validity of your reported study, since it failed to use
any Rattus Protestanti, which would have provided the necessary control for
work-ethic.

The same ratio was in effect throughout an experimental run which
lasted 30 seconds (how long do you think you can keep these rats
doing this stuff?). The rats were tested under "normal" conditions
in which the cursor started near the target and reinforcements were
large enough to allow the rat to keep the cursor near the target
at all ratio requirements. The rats were also tested under "deprived"
conditions in which the cursor started many pixels to the left of
the target and reinforcements were too small to allow the rat to
move the cursor to the target.

I don't see any connection here between the "deprived" condition and what I
take to be its analog in food deprivation in more typical studies. Being
farther from the target is certainly a greater error in the cursor-position
system but is there any reason to expect that a greater error would lead to
a higher rate of responding? Wouldn't a constant rate be sufficient to
bring the cursor back, so long as this produced a sufficiently high rate of
cursor movement to the right? And by increasing the distance to target
while reducing size of the "reinforcement" you have confounded two
variables. If I buy into the analogy here you have increased "deprivation"
and decreased "pellet size" at the same time.

The results (in terms of responses/sec and reinforcements/sec) are
presented for each ratio in each condition for the two subjects:

. . .

Both reinforcements/sec
and responses/second were calculated by counting the total number
of reinforcements or responses that occurred during a 30 second trial
and dividing by 30. So the total number of reinforcements (including
those randomly added in the D condition) and the total number of
responses in a 30 second session can be obtained by multiplying
the numbers in the table above by 30.

Of course, I'd rather see the raw data, so I tried recovering them by
multiplying every number by 30. Strangely, I didn't always get an even
number of responses or reinforcers, even taking potential rounding error
into account.

What we see in the "normal" condition is about 29-30 "reinforcements" during
the 30 seconds of the session, or about 1 per second. I'd have to infer
that this is the equivalent "drift" rate of the cursor to the left of
target, and that the two "rats" responded at rates sufficient to counter the
drift and keep the cursor on or near target. Response rates increased as
required to keep the average error small as the ratio increased, except
perhaps for a small increase in error with ratio size as required to drive
the higher response output. When free "reinforcement" was added, rates were
reduced accordingly to keep the cursor on the target.

The "deprived" rats' data are more difficult to interpret since no
information was supplied about the "level of deprivation," rate at which
"deprivation" increased over time (was it still 1 "reinforcement" step/sec?)
or the "size" of the "reinforcement." This is unlike actual rat experiments
in that usually deprivation is held constant (it is not increasing rapidly
over time) and rewards are sufficiently large to at least somewhat reduce
deprivation over the course of a session (but not eliminate it). Response
rates increase with ratio but are similar whether or not free
"reinforcements" are administered. The increased rate is probably required
in order to compensate effectively for drift when "reinforcer" size is
smaller; the cursor never reaches target so the extra "reinforcement" does
result in quick elimination of error and thus a settling down to a rate just
large enough to compensate for drift. As the ratio goes up, higher rates
are necessary to oppose the drift, but the increased response requirement
drives down the number of "reinforcers" earned. By FR-10 I would guess that
your "rats" were responding at about maximum rate on the mouse button.

At a given ratio and "reinforcement" size, rate of responding in this analog
is driven mostly by the rate of drift; if this is analogous to rate of
increase in food deprivation then you are talking about a different
time-scale than that used in typical studies, and no supplemental feeding
between sessions. Your session figures are also affected by how long it
takes to reach target, after which the rates will decline to those necessary
to oppose drift; this would be analogous to letting the rats satiate on the
schedule, which again is not what is happening in typical studies.

Regards,

Bruce

3

[From Rick Marken (950806.1812)]

Bruce Abbott (950806.1445 EST) --

I don't see any connection here between the "deprived" condition and
what I take to be its analog in food deprivation in more typical
studies.

I'm not really trying to make my study precisely analogous to a typical
operant study. My point is to show that when you put control systems into
a situation similar to the typical operant study, something will
happen. What does happen is not always control so it's not always clear
what did happen.

Your analysis of what's going on in the "normal" condition is right on
target. That's because you can see the controlling done by these control
systems. Reinforcement rate is kept close to the same value (30) at all
ratios, whether or not a disturbance is present.

Your analysis of what's going on in the "deprived" condition is not as
perceptive -- not because you became less perceptive but because there
was less to perceive: there was no control. For example, you note:

Response rates increase with ratio but are similar whether or not free
"reinforcements" are administered. The increased rate is probably
required in order to compensate effectively for drift when "reinforcer"
size is smaller;

If you had done the "Abbott Test" of the response rates I think you
would have found that the increase in rates with ratio was an artifact
resulting from the fact that there was a "collection interval". So your
guess about the reason for increased response rate is irrelevant: the
increase in response rate with ratio is an artifact, as is the decrease in
reinforcement rate with ratio. These artifacts of non-control give the
impression that there is some control occurring in the "deprived" condition.
Here is a table of the number of reinforcements actually obtained in a
30 second period at each ratio in the "deprived" condition.

                        2 5 10

S1 ND 83 55 30
        D 132 (125) 77 (80) 43 (45)

S2 ND 91 51 29
        D 123 (135) 81 (75) 42 (42)

As the ratio goes lower it looks like the number of reinforcements is
increasing toward a reference level of over one hundred per 30 second
period. But the disturbance (D) condition shows that there was no control
at all. The actual number of reinforcements received was nearly the same
as the number of reinforcements that would have been expected
(in parentheses) if there were no resistance to the disturbance.

This little study shows some of the problems that result from trying
to interpret the results of conventional research in PCT terms. Before
we start developing models of control it is best to know that control
is, in fact, occuring. There may be some conventional studies that provide
clear evidence of control (the Collier, Hursh, & Hamlin study may be one)
but I suspect that they are few and far between. That's why I keep harping
on the importance of doing control studies from scratch -- based on an
understanding of the nature of control systems.

I suppose that it is useful to learn why most conventional studies are
a waste of time from a PCT perspective. So maybe the time spent debunking
conventional studies is not a waste of time. We may not be learning anything
about control from these studies -- but at least we'll learn why we're not.

Best

Rick

4

<[Bill Leach 950806.23:59 U.S. Eastern Time Zone]

[From Rick Marken (950806.1812)]

I suppose that it is useful to learn why most conventional studies are
a waste of time from a PCT perspective. So maybe the time spent
debunking conventional studies is not a waste of time. We may not be
learning anything about control from these studies -- but at least we'll
learn why we're not.

Yes, it should be obvious that just demonstrating that living beings are
control systems and asserting that any study efforts not based upon
control theory are a waste of time has not been generally effective.

OTOH changes in Operant studies based upon both a specific knowledge of
why the studies are not providing the knowledge sought and how employing
PCT can produce meaningful data could make a difference.

Hopefully Bill and Bruce can produce experiments that "look enough" like
standard EAB work to be considered as a "direction" change within that
community. With any luck, this will be perceived as a gentle shift
rather than the profound blow between the eyes that it actually is!

-bill

5

[From Bruce Abbott (950807.1100 EST)]

Rick Marken (950806.1812) --

Your analysis of what's going on in the "deprived" condition is not as
perceptive -- not because you became less perceptive but because there
was less to perceive: there was no control.

I could see that response rate was not being controlled as there was no
effect on response rate of the added disturbance. However, I'm still
curious as to what actually WAS going on in this experiment. (I assume you
actually ran this experiment as opposed to making up the data.) You tell us
what was NOT going on, but fail to offer an analysis that I can make sense
of. How about it?

Regards,

Bruce

6

[From Rick Marken (950807.1100)]

Bruce Abbott (950807.1100 EST) --

I could see that response rate was not being controlled as there was no
effect on response rate of the added disturbance. However, I'm still
curious as to what actually WAS going on in this experiment. (I assume you
actually ran this experiment as opposed to making up the data.) You tell us
what was NOT going on, but fail to offer an analysis that I can make sense
of. How about it?

Yes, I actually ran the experiment and it can now be revealed: I am (S1) and
my son Ari is (S2).

You know what was going on in the "normal" condition; the subjects kept the
cursor on target by pressing the mouse button as necessary (depending on
ratio and disturbance).

I'm not sure what was going on in the "deprived" condition. Both subjects
ended up pressing the button at a very high (and fairly regular) rate in all
conditions. The cursor was clearly not under control (as the disturbance data
shows); it is not clear what was under control, if anything.

I pressed as fast as I could in the deprived condition because I tried to
maintain the goal of moving the cursor to the target. The cursor never really
moved toward the target much; it just stayed about 80 pixels to the left of
the target. If I hadn't pressed at all it would have drifted slowly to the
left so maybe I was controlling for keeping the target from moving to the
left. This would have to be tested, of course, with disturbances.

I don't know why my son pressed as fast as he could in the deprived
condition; I didn't ask him to. He might have thought that he could move the
cursor to the target if he could just press fast enough. Or he might have
wanted to see how fast he could press and the cursor be damned (he's very
athletic). I'll ask him if you like.

Of course, it's very likely that another subject would have responded quite
differently in the deprived condition; I'm pretty sure that my daughter
Lise would have stopped pressing pretty early in the game; she is not a fan
of hopeless causes (but she's in Paris at the moment). However, all subjects
would behaved exactly like S1 and S2 if they were controlling the cursor.

All we know about what happened in the "deprived" condition of this
experiment is that the subjects did NOT control reinforcement rate (and,
hence, the cursor). That is, all we can tell about the "deprived" condition
is what is NOT going on (control). When we try to make sense of the results
obtained in this condition we are doing exactly what I believe research
psychologists have been doing for the last century -- making sense of
of data that make no sense.

Best

Rick