Hierlevels

[From Dick Robertson] 980315.2143CST
[From Rick Marken (980313.1000)]

Thanks for the paper, Dick. It's extremely well written and
_very_ interesting. I highly recommend it...
first step in showing how The Test can be used to study these
"high level" controlled perceptions. Just superb!

Thanks, Rick. Acknowledgement is a warm fuzzy

[from Jeff Vancouver 980313.1500 EST]

I read the Robertson, Goldstein, Mermel, & Musgrave paper on testing the
self as a control system....

they might find Steel, C.M. (1988) The psychology of self-affitmation:
Sustaining the integrity of the self. _Advances in Experimental Social
Psychology_ 21, 261-302.

Thanks, Jeff, I'll try to get a look at it.

And about higher level controlled perceptions--I wonder if any CSGnetters are
currently teaching a psych course using PCT? And if there are, I wonder if
there is any sentiment for another attempt at studying "controlling one's
grade" as an instance of high level control?

CSG members who attended the annual conf. a few years back might recall that I
had made such an attempt when teaching an intro course in psychology, using IMP
as the text in the early 90s. Bill Powers helped me plan out some of the
approach in hopes that we could model a student's progress through the course.
I had four alternative 20-item tests for each chapter, and student aides to
grade each student's performance right after taking each test, as well as
having him/her fill out a form asking what grade they were controlling for, &
next to compute the difference between their stated reference value and their
achieved score (pv) to provide an error measure, test by test. Since most
students needed around two attempts to pass each chapter (they couldn't do the
next until then) we had ca. 35-40 data points per student. Not very many
compared to the hundreds or even thousands you get from tracking experiments.
That was one of the problems.

The big problem, however, and the one that nixed the project was that most
students didn't consistently control for the grade aimed-at at the beginning of
the course. Using both the replies to the questionnaire to each test,
spontaneous comments and individual interviews, I discovered that many students
changed their reference-grade to meet their current running grade, instead of
increasing loop gain (by studying more, asking more questions, whatever) to
reduce the error signal.

This was, itself, a somewhat interesting finding. It seemed that some other cv
--in some cases they gave us the answer: fixed amount of study time per week--
was under tighter control than preferred grade.

A minority actually did seem to keep preferred grade under control. When
earlier running average did not >= 90% they studied_much_harder (& complained
about it). These were the ones who finally ended the course with A's.

If any one were motivated to update and improve this study, one way might be to
revive the old "programmed learning" method and prepare a pool of, say, 1000
questions that had to be 90% or better for A, etc, with ES computed after each
question, since computer testing is now so easy. Questions would not have to be
limited to showing how well the student recalled seeing the tested words in the
text. Practice-skill, such as writing a program for a tracking experiment,
accurate identification of real and false test-for-CV's and things like that,
might be included in the test items.

And finally, any attempt to model performance might be restricted only to those
students who ended with the reference-grade indicated at the beginning of the
course.

Any takers? Could this be a group-project of interest?

Best, Dick RobertsonFrom ???@??? Mon Mar 16 09:58:48 1998
Return-Path: owner-csgnet@POSTOFFICE.CSO.UIUC.EDU
Received: from beasley.cisco.com (mailgate-sj-2.cisco.com [171.69.2.135]) by pilgrim.cisco.com (8.8.5-Cisco.1/8.6.5) with ESMTP id WAA07699 for <bnevin@pilgrim.cisco.com>; Sun, 15 Mar 1998 22:54:03 -0500 (EST)
Received: from proxy1.cisco.com (proxy1.cisco.com [192.31.7.88]) by beasley.cisco.com (8.8.4-Cisco.1/CISCO.GATE.1.1) with ESMTP id TAA19536 for <bnevin@CISCO.COM>; Sun, 15 Mar 1998 19:54:02 -0800 (PST)
Received: (from smap@localhost)
  by proxy1.cisco.com (8.8.7/8.8.5) id TAA25036
  for <bnevin@CISCO.COM>; Sun, 15 Mar 1998 19:54:01 -0800 (PST)
Received: from postoffice.cso.uiuc.edu(128.174.5.11) by proxy1.cisco.com via smap (V2.0)
  id xma025030; Mon, 16 Mar 98 03:53:58 GMT
Received: from postoffice.cso.uiuc.edu (postoffice.cso.uiuc.edu [128.174.5.11])
  by postoffice.cso.uiuc.edu (8.8.8/8.8.8) with SMTP id VAA24732;
  Sun, 15 Mar 1998 21:51:18 -0600
Received: from POSTOFFICE.CSO.UIUC.EDU by POSTOFFICE.CSO.UIUC.EDU
          (LISTSERV-TCP/IP release 1.8b) with spool id 8173305 for
          CSGNET@POSTOFFICE.CSO.UIUC.EDU; Sun, 15 Mar 1998 21:51:16 -0600
Received: from orion.neiu.edu (orion2.neiu.edu [207.241.96.11]) by
          postoffice.cso.uiuc.edu (8.8.8/8.8.8) with ESMTP id VAA44678 for
          <csgnet@postoffice.cso.uiuc.edu>; Sun, 15 Mar 1998 21:48:34 -0600
Received: from robertson (ip97-56.neiu.edu [207.241.97.56]) by orion.neiu.edu
          (8.8.5/8.8.6) with SMTP id VAA10568 for
          <csgnet@postoffice.cso.uiuc.edu>; Sun, 15 Mar 1998 21:51:24 -0600
          (CST)
X-Mailer: Mozilla 3.01Gold (Win95; I)
MIME-Version: 1.0
Content-Type: multipart/mixed; boundary="------------27561434019"
Message-ID: <350CA086.7167@neiu.edu>
              <CSGNET@POSTOFFICE.CSO.UIUC.EDU>
Sender: "Control Systems Group Network (CSGnet)"
              <CSGNET@POSTOFFICE.CSO.UIUC.EDU>

ยทยทยท

Date: Sun, 15 Mar 1998 21:46:14 -0600
Reply-To: "Control Systems Group Network (CSGnet)"
From: R J Robertson <R-Robertson@NEIU.EDU>
Subject: Conditioning thread
To: Multiple recipients of list CSGNET
              <CSGNET@POSTOFFICE.CSO.UIUC.EDU>
X-UIDL: 1fbda2415677304dd25019571f82cd59

[From DickRobertson] 980315.2145CST

From Phil Runkel on 10 Mar 98.

Subject: Conditioning or the like

I never could make sense of conditioning, not when I was 20 years
old and not now when I am 80... write about conditioning in a way that suits
me even though I have read many of the postings on the CSGnet explaining how >
an episode called conditioning would look to PCT....Below...two episodes

Episode 1:

Will the control system for walking (getting one foot in front of
the other without falling down) suffice to get you to the
library? No. Walking can get you anywhere (so to speak). Will
the control system for seeing yourself within the library suffice
to guide your walking in the direction of the library? Yes. Now
what about a sign that you see part way there that reads, "This
way to the library"? That sign and a lot of other familiar
sights serve as subgoals and enable you to keep on the path.

Episode 2:

Will the control system for salivating suffice to bring you food
(meat powder)? No. Salivating can get you ready to chew on
various kinds of foods or none at all. Salivating is an early
portion of the program for eating, but salivating can be a part
of other programs, too: maybe cooling or chewing on a stick.
Will the control system for eating suffice to activate
salivation? Yes. Now what about a bell that you have come to
understand signals meat powder just around the corner? That
sound and a lot of other familiar experiences serve as subgoals
and enable you to keep on the path: seeing food, smelling it,
hearing it frying.

I'd like to try my hand (though maybe I'm as confused as you say you are)
start with Episode 2:
If we assume that salivating with food molecules in mouth/nose is hard wired--
part of the intrinsic system--then the CV might be something like, "presence of
digestive enzymes." However, I don't know whether there are sensors for
anything like that, in the mouth. What we do know is there are the taste and
smell sensors; could an increase from some base-line level of taste and smell
signals be the input of a control system for which salivary secretion is the
output? But, then just what is under control? Do we have any biologists who
know the answer?

A simpler version of the dog salivating has been reported on by Eric Kandel,
studying aplysia, which I reported on pp. 113-115 in IMP, mainly because he had
worked out the nerve circuits in enough detail to show (to me, anyway) the
presence of a(n at least) two-order control system. His "gill cover reflex" is
a simple on-off in which skin sensors in the area of the opening supply the
input. He too, of course, doesn't consider what the CV might be, since he
wasn't working with a PCT basis, but it's not too wild a speculation that--in
its natural environment--what would most often stimulate the skin sensors would
be sand and bottom goo, which would not be good to breathe-in, hence evolution
might well have selected for skin sensors that go off when touched by anything.
Next, Kandell added one more feature that brought his study into line with
Pavlov. He shocked the beast on its head and found that, after that, the gill
cover closing no longer accomodated--i.e. it didn't reduce -> 0 with repeated
touching of the skin, but maintained high level output. Thus shock was
parallel with Pavlov's bell. Kandell even had doped out the chemistry of
transmitter packets being-used-up--or: level of transmitter packets at 1st
order comparator--as what the second order output was controlling.

It is easier to see shock as triggering the reorg. system with the resulting
building of the more complex three-order ( or more ) system, than to see the
bell as a trigger for reorganizing. But, that whole issue leads to the larger
question of good old Locke's association of ideas--something not much developed
in PCT or HPCT, but a rich mine for a young investigator, I would think.

While we wouldn't at first call the dog's awareness of the bell as an idea, nor
his awareness of the smell/taste of food as another idea, I think that is
because Locke was only thinking about human learning. From the vantage point
of PCT I think what he was talking about applies to learning in any animal,
broadly speaking. Thus, Pavlov either didn't know about "association of ideas"
or he wanted to coin his own terms to emphasize the originality of his
discoveries (often the case), or he genuinely thought that the triggering of
something as fixed as a "reflex" could properly be described as becoming
conditional--via association with a "neutral" signal. We know now that most
"reflexes" aren't all that fixed under certain conditions, but that was beyond
his time or interest.

Now as to episode 1 -- the whole series you describe seems to me to involve the
question of how the hierarchy gets built with the subquestions of the formation
of associations between previously ineffective PVs, resulting in the
incorporation of lower level control systems into more and more higher order
systems, as implied in your examples. I think Franz Plooij's work has probably
gone into more detail about the development of the hierarchy than anything else
I've seen, but again, isn't this a pretty open area? If you have a (to start
with) rough principle for doing things as expeditiously as possible, and
another for valuing the library, wouldn't continuing error signals in the first
be keeping enough reorganization going on the way to the library--each time you
go--so that PVs that are at first random to the library-finding program (like
the sight of signs, memories of the appearance of different streets, etc)
become built into more an more circuit complexity?

I don't know whether this speaks to your questions very much, but it strikes me
as an important thread and worth working on.

Best, Dick RobertsonFrom ???@??? Mon Mar 16 09:59:00 1998
Return-Path: owner-csgnet@POSTOFFICE.CSO.UIUC.EDU
Received: from beasley.cisco.com (mailgate-sj-2.cisco.com [171.69.2.135]) by pilgrim.cisco.com (8.8.5-Cisco.1/8.6.5) with ESMTP id IAA19240 for <bnevin@pilgrim.cisco.com>; Mon, 16 Mar 1998 08:29:09 -0500 (EST)
Received: from proxy2.cisco.com (proxy2.cisco.com [192.31.7.89]) by beasley.cisco.com (8.8.4-Cisco.1/CISCO.GATE.1.1) with ESMTP id FAA25778 for <bnevin@CISCO.COM>; Mon, 16 Mar 1998 05:29:07 -0800 (PST)
Received: (from smap@localhost)
  by proxy2.cisco.com (8.8.7/8.8.5) id FAA29505
  for <bnevin@CISCO.COM>; Mon, 16 Mar 1998 05:29:06 -0800 (PST)
Received: from postoffice.cso.uiuc.edu(128.174.5.11) by proxy2.cisco.com via smap (V2.0)
  id xma029496; Mon, 16 Mar 98 13:29:03 GMT
Received: from postoffice.cso.uiuc.edu (postoffice.cso.uiuc.edu [128.174.5.11])
  by postoffice.cso.uiuc.edu (8.8.8/8.8.8) with SMTP id HAA17578;
  Mon, 16 Mar 1998 07:26:54 -0600
Received: from POSTOFFICE.CSO.UIUC.EDU by POSTOFFICE.CSO.UIUC.EDU
          (LISTSERV-TCP/IP release 1.8b) with spool id 8175956 for
          CSGNET@POSTOFFICE.CSO.UIUC.EDU; Mon, 16 Mar 1998 07:26:53 -0600
Received: from animas.frontier.net (root@frontier.net [199.45.141.1]) by
          postoffice.cso.uiuc.edu (8.8.8/8.8.8) with ESMTP id HAA34460 for
          <CSGNET@POSTOFFICE.CSO.UIUC.EDU>; Mon, 16 Mar 1998 07:26:51 -0600
Received: from pm3-1-50.frontier.net (pm3-1-50.frontier.net [199.45.211.50]) by
          animas.frontier.net (8.8.8/8.8.8) with SMTP id GAA13762 for
          <CSGNET@POSTOFFICE.CSO.UIUC.EDU>; Mon, 16 Mar 1998 06:26:43 -0700
X-Sender: powers_w@mail.frontier.net
X-Mailer: Windows Eudora Light Version 3.0.1 (32)
References: <3.0.3.32.19980313151310.00fc1154@oak.cats.ohiou.edu>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Message-ID: <3.0.1.32.19980316015442.0068eb98@mail.frontier.net>
Date: Mon, 16 Mar 1998 01:54:42 -0700
Reply-To: "Control Systems Group Network (CSGnet)"
              <CSGNET@POSTOFFICE.CSO.UIUC.EDU>
Sender: "Control Systems Group Network (CSGnet)"
              <CSGNET@POSTOFFICE.CSO.UIUC.EDU>
From: Bill Powers <powers_w@FRONTIER.NET>
Subject: Re: Testing "Self"
To: Multiple recipients of list CSGNET
              <CSGNET@POSTOFFICE.CSO.UIUC.EDU>
In-Reply-To: <350C7789.7080@earthlink.net>
X-UIDL: 0b9ac562b5db30a8894b08a3c835d5b4

[From Bill Powers (980316.0111 MST)]

Rick Marken (980315.1650)--

I agree with your evaluation of

The Robertson, Goldstein, Mermel, & Musgrave paper on
testing the self as a control system

This was just a beginning. As you say,

We don't know how to test to determine a controlled "self"
perception right now any more than people knew how to measure
the gravitational constant right after Newton proposed it's
existence. We just have to try stuff to see what works (and we
know it _doesn't_ work to the extent that the results are not
perfect).

The point I want to make is that when you know so little, if you want good
reliable results you have to keep your hypotheses simple.

Bruce Gregory (980315.2020 EST) says

My reservations stem from my lack of conviction that "self perception" is a
truly useful concept. Surely some people control a perception of self. And
some cultures emphasize the control of this perception more than other
cultures. But it seems highly unlikely to me that we will ever be able to
say any more about self perception than it is a higher level perception that
some people control. It would be very interesting to find such high level
perceptions that _everyone_ without exception controls, but this seems
highly unlikely.

I really don't agree with this. The problem here may be that "self" is
often talked about as if it's something mysterious, hidden, complex, or
even mystical. But it doesn't have to be that way. For example, one aspect
of my own self concept is that I am male, rather than female. If anyone
referred to me as "she," I would perceive this as a simple mistake, and if
it seemed necessary, would correct it. It would be quite easy for someone
to do the Test with me, and discover this aspect of my self-concept. The
results of experiments would be clear-cut and highly reliable. Of course
nobody would do back-flips at discovering this basic fact about me, but the
point of this sort of exploration isn't to produce fireworks; it's to
establish scientifically justifiable statements of fact.

Once you can predict my actions when my perception of the fact of my gender
is disturbed, you can start asking for more details. For example, would I
be willing to undergo a sex-change operation, say next Tuesday? Now my
self-concept takes on another dimension, the idea that my sex might change.
I would definitely not agree to the operation, but some other male might be
grateful for the chance. This makes the reference level more visible.

And of course we can go on from there, in little steps, asking questions
about simple obvious things, Testing as we go. This is what lawyers are
taught to do: never ask a question (of nature in this case) to which you
are not pretty sure of the answer. Unlike the lawyer, we are delighted when
we get an unanticipated answer, because that means we're about to learn
something we didn't know before. It's the shock of violating confident
expectations that tells us we're truly testing some hypothesis, and that
the one we're testing is clearly, unequivocally, wrong.

Bruce G. makes an assumption that isn't correct: that when we investigate
the self, we're looking for particular self-concepts that are common to all
people. This isn't true. A self-concept is just a class of concepts that
happen to be about oneself. If you think of yourself as having a
profession, then what you call yourself as a member of that profession is a
self-concept, as in "I'm a psychologist". If I introduce you to someone as
a physiologist, you will correct me. If you don't have a profession, your
self-concept may include the idea "unemployed" or "uncommitted" or
"unfettered." A self-concept is just whatever concepts you have that relate
to yourself rather than someone else, or some thing.

What about aspects of the self like "introverted?" I thing that such terms
may refer -- indirectly -- to real dimensions of self-perception, but that
we're a long way from being able to sort out what they are. The proof of
that is that when we ask about such characteristics, we get results of the
usual kind, where correlations of 0.5 with anything else are considered
high. If you want high correlations, ask a man how he would feel about
wearing a miniskirt and high heels to work tomorrow, not whether he's
introverted.

I suppose it's asking too much to hope that psychologists will scale back
their ambitions and start asking questions that are simple enough to have
clear answers. But that's what they have to do if they want to build a
science.

Best,

Bill P.

[from Jeff Vancouver 980316.0930 EST]

[From Dick Robertson] 980315.2143CST
And about higher level controlled perceptions--I wonder if any CSGnetters are
currently teaching a psych course using PCT? And if there are, I wonder if
there is any sentiment for another attempt at studying "controlling one's
grade" as an instance of high level control?

I am conducting a study regarding the control of grade next term. However,
because of the effect you found in your earlier study (about the changing
reference condition and not much change in output) when faced with a
discrepancy, I am using correlational data not as much to provide evidence
of PCT, but to call to question some treasured "conventional psychology" -
namely social cognitive theory (Bandura) and goal setting theory (Locke).
I have found a similar effect in my mastermind game paradigm. You also
might want to check out:

Kernan, M.C., & Lord, R.G. (1990). The effect of valence, expectancies,
and goal-performance discrepancies in single and multiple goal
environments. Journal of Applied Psychology, 75, 194-203.

They look at the control of grade as a hierarchical processes (though their
description of the hierarchy is somewhat different then HPCT).

Your conjecture that the lack of control of grade is due to a tighter
control of other factors (time to study) seems to be right on track (i.e.,
along my line of thinking). Testing that hypothesis is not easy though (a
colleague has attempted to collect data on that idea, we have not analyzed
the data yet; it is based on self-report and therefore sloppy). Meanwhile,
another hypothesis that I harbor is that self-report of grade goal is a
poor measure of the actual reference condition. That hypothesis has two
correlates. One, that grade is a reasonable construct to describe r or p,
but the self-report is not a very accurate measure of r; or two, that grade
is only tangentially related to what r and p are all about (i.e., they are
not really controlling for grade, but something like grade). Clearly,
ethical issues make the test, which would be useful for testing these
hypotheses, very difficult to do in this context.

Given your conjecture, however, I am not sure I see the merit of the
computer testing study. It might show that students _can_ control grade,
but if we know the simultaneous control of other perceptions overwhelms the
control of grades, how useful would the results be? Likewise, the study
would not eliminate the measurement hypotheses because the context is very
different.

Later

Jeff

Sincerely,

Jeff