Rejoinder to PCT Critique

[from Gary Cziko 920909.1335]

I have written a response to a critique by Amundson et al. to my article
"Purposeful Behavior as the Control as Perception: Implications for
Educational Research" which will appear in _Educational Researcher_ which I
have appended below.

Amundson et al. don't do much to actually critique PCT, but rather attack
me on my non-PCT-related arguments. Since I see the PCT part much more
important, I concentrate on this in my rejoinder.

I realize that this might not make a whole lot of sense without seeing
Amundson et al.'s paper, but I have sent it anyway in case there are some
obvious things I can do to improve my reply.

Also, the emphasized words don't show up as such in this ASCII version of
the my paper.

I will need to send this off in a few days, so if anyone has comments,
please send them as soon as possible.--Gary

ยทยทยท

===========================================================
On The Threats to Educational Research
Not Faced by Amundson, Serlin, and Lehrer

Gary A. Cziko
University of Illinois at Urbana-Champaign

Amundson et al. (1992) charge that I misunderstand or misrepresent the
views of the authors I cite in support of my claims as well as the views of
Lehrer et al. (1990). They also find that there is nothing in my
discussion of statistical methods and perceptual control theory which
justifies my criticism of the methods of mainstream quantitative
educational and psychological research. I will consider each of these
charges in turn.

Concerning my "misunderstanding or misrepresentation" of authors such as
Cronbach and Snow on educational research, Lakatos and Einstein on the
philosophy and methods of science, and Gould on biological evolution, I can
easily provide additional evidence to support my portrayals. For example,
while Amundson et al. chide me for using "the narrowest of readings and
most selective of quotations" . . . "to depict Lakatos as an advocate of
prediction and control" (p. 5), it did not take me long to find an even
stronger Lakatosian endorsement of the role of prediction in the very same
essay quoted repeatedly by Amundson et al.:

"For the sophisticated falsificationist, learning about a theory is
primarily learning which new facts it anticipated: indeed, for the
sort of Popperian empiricism I advocate, the only relevant evidence is the
evidence anticipated by a theory, and empiricalness (or scientific
character) and theoretical progress are inseparably connected." (Lakatos,
1970, p. 123, first two emphases added)

I could provide similar support for my understanding of Einstein (that he
ascribed great importance to predictive tests of his theory) and Gould
(that he considers unpredictable contingency to a play major role in the
evolution of species and that he is not anti-selectionist). But since
these issues are secondary to the purpose of my paper, I will not discuss
them further here.

On the other hand, it seems that I did unwittingly (and not intentionally)
misrepresent the views of Lehrer et al. on educational research,
particularly in stating that " . . . Lehrer et al. appear convinced that
progress in educational research is a matter of adding more and more
variables and sophisticated statistical analyses to a one-way,
input-output, external-causation model of human behavior" (p. 28) and they
protest that "in fact, quite the opposite is true" (p. 16). But what does
"quite the opposite" mean? Does this mean they reject a one-way causal
model of human behavior? Or does it mean they maintain hope that we will
eventually hit upon a small number of the "right" variables which will
account for human behavior in educational settings? Unfortunately,
Amundson et al. do not make their views very clear in this respect.

The primary purpose of my paper was to introduce perceptual control theory
(PCT) to the educational research community and discuss its implications
for educational research. My very brief discussion of group-based
statistics was included to show that the typical results achieved using
these methods are not very impressive by any reasonable standards,
particularly when one attempts to make predictions concerning individuals.
My discussion of PCT was then intended to provide an explanation of why
this is the case. The main point is this: If PCT is correct in positing a
closed-loop, negative-feedback relationship between individuals and their
environment, then the basic model (and the assumptions) underlying
group-based statistical analyses as used in educational and psychological
research is simply inadequate to the task of providing explanations for
purposeful behavior. This is the major point of my article and a point
which Amundson et al. do not address.

In an attempt to make this argument clearer, let us consider the basic
building block of statistical analysis as used in educational and
psychological research. This can be expressed in the simple formula y = bx
+ a, where x refers to an independent variable (usually some environmental
condition or stimulus), y to some dependent variable (usually some response
or behavioral outcome), and b and a are the regression weight and
y-intercept (predicted value of y when x is zero), respectively.
Correlations, t-tests, analysis of variance (univariate and multivariate,
with or without covariates) multiple regression, canonical correlation,
discriminant analyses, and path analysis all built upon this basic
equation. For all these analyses it is assumed that y depends on x, i.e.,
that there is a one-way direction of causality from x to y1. And in this
important sense, despite Amundson et al.'s protest on page 6, the great
majority of the group-statistical methods used in educational and
psychological research are monolithic.2

But research on PCT has shown that such a model is incomplete and therefore
inadequate for even the most mundane of purposeful behaviors. In a simple
tracking task (in which a subject seated before a computer screen
manipulates a mouse or joystick in order to keep a cursor positioned at a
certain goal location in spite of constant and unpredictable disturbances
applied to the position of the cursor), the subject's response ("dependent"
variable y) does indeed depend on the position of the cursor on the screen
("independent" variable x), but x also depends on y at the same time.
Thus, there are no observable "independent" or "dependent" variables in the
usual sense of these words. In fact, the only truly independent variable
is the internal goal or reference level of the subject, a perceptual
standard which is not determined by any environmental stimulus. While
this reference level is not directly observable by outsiders, it can
nonetheless be inferred by observing that the subject maintains the
cursor's position at or near a certain location in spite of the
disturbances influencing the cursor. That is, we infer purposive behavior
and perceptual control when there is a low correlation between what are
typically considered "stimulus" and "response" since while the subject's
actions must continually vary in order to compensate for disturbances, the
position of the cursor changes relatively little since it (actually its
perception) is being controlled by the subject. I argue that if a one-way
model of causality cannot account for something as simple as this tracking
behavior, it is likewise inadequate to account for more complex purposeful
behaviors such as those involved in teaching history or learning calculus.
I read the inconsistent and noncumulative nature of the existing
quantitative research base in education as supportive of this argument.

The notion of a controlled perceptual variable is so foreign to mainstream
quantitative researchers in education and psychology that I cannot pretend
to be surprised if Amundson et al. provide no evidence of making much sense
of my description of it and its implications for research. But
fortunately, they and other readers need not rely on my brief introduction.
Runkel's 1990 book Casting Nets and Testing Specimens provides a clear and
(I find) convincing argument for how group-statistical analyses make
unwarranted one-way causal assumptions concerning human behavior. Runkel
also explains how such analyses are actually analyses of relative
frequencies and as such cannot provide us with the explanations we seek
concerning the psychological functioning and behavior of living organisms
as autonomous systems (my two quotes from Taylor made the same important
point, a point also not approached by Amundson et al.). And Powers has
developed two quite remarkable computer programs which demonstrate,
statistically analyze, and simulate the phenomenon of perceptual control.3

Amundson et al. state that "there is nothing in Cziko's presentation of
perceptual control theory that precludes the utility of 'clinical trials'
(Schrag, 1992)." This is not the case if they mean to endorse the utility
of educational trials in which different treatment conditions
("independent" variables) are compared for their effects on educational
outcomes ("dependent" variables). Again, such a research methodology
assumes a one-way model of causation which cannot account for either
teaching or learning as purposeful behaviors. As absurd as it may
initially seem (although probably no more absurd than the initial proposals
that the earth is round and that species evolve), PCT explains why it is
that external environmental factors are not the cause of behavior and as a
consequence educational treatments do not cause educational outcomes.
Indeed, without the understanding of purposeful behavior that PCT offers,
Schrag is correct in stating that "the positivist paradigm is hard to
avoid" (p. 7). But once it is realized that Schrag is also apparently
committed to a one-way view of causation for which there is now a more
adequate alternative, his positivist paradigm becomes quite avoidable
indeed.

Amundson et el.. conclude their critique with the prediction: "Important
results will come from actual research, not from a priori armchair
declarations." I couldn't agree more. What Amundson et al. (and the rest
of the mainstream educational and psychological research community) do not
realize, however, is that research results already exist which convincingly
demonstrate that the one-way causation model which underlies virtually all
group-statistics-based educational and psychological research is inadequate
to account for even the simplest of purposeful behaviors. Examples of such
research can be found in Powers (1989, 1992), Marken (1992) and in many of
the papers included in the September/October 1990 issue of American
Behavioral Scientist (Vol. 34, No. 1) and interested readers can easily
replicate the phenomenon and findings of perceptual control by using the
computer programs designed for this purpose by Powers. In addition to the
empirical and predictive adequacy of PCT in those areas in which it has
been tested, it also receives high marks for Amundson et al.'s criteria of
"simplicity of foundation" and "logical coherence."

There are undoubtedly many educational researchers who are either content
with the current methods of mainstream quantitative educational research or
believe that if problems exist it is because these existing methods are not
being properly used. For these individuals, my article and the work of
Powers and other PCT researchers constitutes disturbances to be resisted,
ignored, or ridiculed. This is exactly what PCT would predict. But there
are undoubtedly many others who are dissatisfied with the results that
quantitative educational research methods have delivered to date. I can
only hope that some individuals in this latter group will find my article
of interest and the PCT ideas it introduces worth pursuing. And I look
forward to hearing from them (and hopefully Amundson et al. as well) on
CSGnet4 where we can continue this conversation.

References

Marken, R. S. (1992). Mind readings: Experimental studies of purpose.
Gravel Switch, KY: Control Systems Group.

Powers, W. T. (1989). Living control systems: Selected papers of William T.
Powers. Gravel Switch, KY: Control Systems Group.

Powers, W. T. (1992). Living control systems II: Selected papers of William
T. Powers. Gravel Switch, KY: Control Systems Group.

Runkel, P. J. (1990). Casting nets and testing specimens: Two grand methods
of psychology. New York: Praeger.

Shrag, F. (1992). In defense of positivist research paradigms. Educational
Researcher, 21(5), 5-8.

Notes

1This is not technically true for simple bivariate relationships derived
from correlational (non-experimental) studies where it is recognized that
the direction of causality is unknown and might be from y to x instead of
from x to y, or that some other variable z might be the independent
variable determining both x and y. Nonetheless, in all of these cases it
is still assumed that there is a one-way direction of causality, whichever
way and from whatever source it may be.

2Amundson et al. ask "who are these people" (p. 7), namely, the people who
use the research methods I criticize. As a introduction to this community
of researchers, I suggest they begin by looking through the last four
issues of American Educational Research Journal (fall 1991 through summer
1992) where in the section on "Teaching, Learning, and Human Development"
two-thirds (16) of the 24 published studies employed statistical analysis
of group-data in an attempt to find relationships between one or more
independent variables and one or more dependent variables.

3Powers's computer programs can be run on an IBM-compatible personal
computer with a mouse or joystick. These programs are available via
anonymous FTP from biome.bio.ns.ca as pub/csg/programs/msdos/dem1a.exe and
pub/csg/programs/msdos/dem2a.exe.

4CSGnet is an electronic forum on PCT with the address
CSG-L@VMD.CSO.UIUC.EDU. It is also accessible as a Usenet (NetNews) group
as bit.listserv.csg-l. Individuals can subscribe to CSGnet via
LISTSERV@VMD.CSO.UIUC.EDU or can contact the author (G-CZIKO@UIUC.EDU) for
more information.