[From Bill Powers (941026.1730 MDT)]
Bruce Abbott (941025.1250 EST) --
Sorry if I have maligned the EAB community. My fault for speaking of
that of which I do not know. If it's any comfort, one of Tom Bourbon's
students who is off getting a degree calls me now and then, and during
his last call he remarked that the EAB people seem to be the only ones
around who understand real experimentation. So you can see we aren't
spreading poisonous rumors.
I don't wish to continue this line of argument further than this as I
essentially agree with you.
Good. Let's get on with it.
Your derivation of my equation for VId is correct. In modeling the
environmental "feedback function," I would prefer it to yours if only
because it is easier to work with and understand.
Fine. The analytical approximation we use won't make much if any
difference in the PCT stimulations, because we're generating the
probabilistic aspects of the VI schedule from first principles anyway.
I think it would be necessary to begin at some level of deprivation in
order to speed up acquisition of the lever-pressing response and get
through the transient reorganization phase, at least on initial
studies.
I defer to your vastly greater experience. Is there any reason, though,
to make the deprivation period a separate phase? I understand that rats
are pretty dumb and they're going to get hungry before they learn the
trick of how to get food. Reorganization is driven by intrinsic error,
according to some theory I've heard of. Do you have to use shaping to
help speed up the process?
Regarding the 80% value, your analogy to a human being brought to 80%
free feeding weight does not provide an especially valid comparison, as
laboratory rats on ad libitum food tend to be overweight because food
can then be obtained with little expenditure of effort, in contrast to
conditions existing in the wild.
OK, my feelings about rats are not exactly Hinduesque, and I don't want
to dramatize them. Just a preference for not making things any more
difficult for them than necessary. Remember, though, my point about the
weight control loop. If you let the weight fall naturally while the rats
learn the tasks, we should see the natural effects of that control
system; if you _control_ the rats' weight, that control system can't
work. But that may be of concerned in later experiments.
The 80% value is used because higher values tend to produce less stable
performance.
This probably doesn't apply to continuous experiments; in the Timberlake
data I've seen, and in obesity experiments, the rats seem to control
weight very precisely (within 1%) when providing all their own food by
bar-pressing. We don't care so much about stable performance as stable
parameters in the models that describe performance. But we'll see. I
don't want to suggest a lot of deviations from standard practice.
Studies of meal patterning in rats indicate that they tend to feed
throughout the day in small "bouts," and so I would expect to observe
bouts of lever-pressing distributed in this way.
Right, and this is all the more reason to figure out a way to store the
data about every lever press and every reward. I think we will be able
to model the "bouts" of eating, in the way I suggested using the idea of
an "appetite" (short-term) control system and a "nutrition" (long-term)
system. It will be fun to try, anyhow.
A technical problem for me is how to gather data concerning what the
rat is doing throughout the study in addition to lever pressing. I may
have to resort to some kind of sampling procedure for some behaviors.
What do you think needs to be recorded?
The main thing is to know whether the rat is engaged in lever-pressing
or has wandered off to look for food elsewhere. We can't do this
perfectly, but it's better to have some indication of whether the rat is
at least in position to press the bar. How about a weight-sensing
treadle? In the Staddon/Motherall data, as I said, the scaling of the
two curves I mentioned suggests that the control system in question has
very nearly the same parameters under both body-weight conditions, but
that when the body weight is 98% of normal, the animal simply spends
less time at the lever, leading to proportional changes in both axes of
the plots (which I can't account for in any other way).
Incidentally, this brings up another reason for not externally
controlling the animals' weight during the experiment. When a rat is
starved and then given control of its food intake, it begins by pressing
the bar very fast and eating a lot, but as the final weight is
approached the rate of pressing and eating falls to a far smaller level.
This is true of hyothalamically-damaged rats which maintain a body
weight three times normal, as well as normal rats. The obese rats, when
their weight reaches the reference level, eat hardly any more than
normal rats. So if the rats were maintained at a low body weight
unrelated to how much they eat in the experimental conditions, we would
be looking at transient instead of steady-state behavior.
Determining amount of food consumed should not present a problem as the
feeder dispenses food in standard-size pellets. Although consuming
them involves a certain amount of waste, it is probably fairly constant
and could be determined with some precision by weighing what falls
beneath the grid floor.
That sounds like an excellent plan. It will remove one uncertainty from
the data.
But I'm going to have to build different accommodations for our
subjects to allow them to occupy the experimental chamber full time.
We'll have to provide water as well as food (and measure consumption of
both), which will certainly change the nature of the results (i.e.,
schedule-induced polydipsia).
Yeah, that might be a complication, but it's a natural one, considering
that water needs and food needs would tend to go together. Should we try
to model both systems, each disturbing the other? Actually I think that
just keeping track of water consumption would be enough for a pilot
study; we might think later of making _everything_ the rat wants
available through bar-pressing and try to model all the control systems
at once. But somehow I think we'd better keep it simple for this first
try (for me).
Phone number: (303) 247-7986.
Can you describe how the overall experiment would proceed?
···
----------------------------------
Different subject:
Fred Toates, of the Open University in England, attended our inaugural
European CSG meeting in Wales last spring. He is a devout Skinnerian,
but also a lapsed control theorist who has decided to get back into it.
I would like to invite him to participate in this project by email. OK
with you? Everybody else?
------------------------------------------------------------------------
Best,
Bill P.