[From Rick Marken (951127.2000)]
Bruce Abbott (9527.1650 EST) --
The whole point of an experiment is _control_. If you allow the independent
variable to vary on its own, you have no idea what other variables are
varying with it.
I disagree, though it's really not an important issue since the kind of
experimentation where this is a concern is the wrong approach to studying
control systems anyway. I disagree because you know just as much (or as
little) about what other variables might be varying when you manipulate
a variable as when you don't. When you manipulate a variable there are always
confounding variables. Indeed, that's the game that's played in conventional
psychological research. Since psychologists aren't studying models, they get
their research published by doing experiments that eliminate confounding
variables present in other people's experiments; there are always confounding
variables, so there are alway new issues of JEAB, JEP, etc.
Me:
Please post a reinforcement model that does not attribute to a reinforcer
an ability to strengthen that is completely independent of the state
of the organism.
Bruce:
How about the classic (and defunct) Hullian model
Ok. I suppose Hull's model does include organismic variables. But, as I
recall, it also attributes characteristics to reinforcers (like incentive
value) that are independent of the organism.
Me:
why are there no studies done by reinforcement theorists that can even be
construed as being about determining what organisms control?
Bruce:
If (as EABers believe) the appearance of control emerges from differential
reinforcement, then there is nothing to study except differential
reinforcement.
I don't think EABers believe ANYTHING about the appearance of control because
they don't know what control (purpose) is. If they did, they would know what
controlled variables are and they would be testing to determine what variables
ARE controlled.
You continue to cling to the belief that reinforcement theorists have a
problem because they have the wrong THEORY of control (the theory that
control emerges from differential reinforcement). In fact, reinforcement
theorists have a problem because they don't know WHAT control is; since they
don't know what control is, they don't have a theory of control (or purpose
or whatever they want to call it). It's not a definitional probleml it's a
FACTUAL problem (remember).
Anyway, you didn't answer my question: why are there no studies done by
reinforcement theorists that can even be construed as being about
determining what organisms control?
Hint: The answer has to do with "controlled variables". You may feel
free to include this phrase ("controlled variable") in your answer;
the more times you use it (correctly) the more credit you get.
Me:
Please state as clearly and simply as possible what you think control
theory has to offer reinforcement theorists?
Bruce:
A whole new way to view behavior, reinforcement, and purpose.
Right: behavior is controlled perceptual input; "reinforcement" is an
inapproaprite metaphor that has been used to describe disturbances
to controlled variables and purpose is the goal toward which action
coaxes a controlled variable.
Do you want to be the one to tell the reinforcement theorists about this
or should I.
I would like to know how you would go about telling them that they are
not studying behavior properly. I would also like to know how you
would tell them that it's not their theories (such as they are) that are
wrong but, rather, their concept of the nature of behavior.
Bill Powers (951127.0750 MST) --
varying the environmental variable is equivalent to varying d; measuring
the response is equivalent to measuring o. But between d and o lies
controlled variable, i, which is held by the feedback action at a value
r which is determined inside the behaving system. The result is (for the
system assumed above) that
o = (r - k2*d)/k1
In other words, the observed dependence of o on d is of a form determined by
environmental constants, and does not reflect the actual input-output
function of the behaving system, even when r happens to be constant.
standard methodology makes it unlikely that i would be discovered as an
important variable...
This, I think, accounts for why control theory was not discovered by
psychologists.
Me too.
And here's another attempt to illustrate the point. Below are data from
an experiment where the IV is the length of a horizontal line and the DV is
the length of a vertical line (assume cm measures). Note that there is a
nice inverse relationship between values of the IV and DV; this is confirmed
by regression analysis (the slope and intercept of the regression equation
are in the lower right). The R squared (RSQ) value of the relationship
between IV and DV is also in the lower right; the IV accounts for over 80% of
the variance in the DV; this is a STRONG relationship by conventional
standards and it would be interpreted as a causal relationship because we
manipulated the IV under controlled conditions: the conclusion would be
that increases in horizontal line length cause decreases in vertical line
length. There would be explanations of this effect in terms of the
"intimidating" properties of horizontal lines; some might say that
horizontal lines afford vertical opposition. In other words, we get
conventinoal psychology.
IV (d) 1.00 2.00 3.00 4.00 5.00
DV (o) 1.00 0.50 0.33 0.25 0.20
Pred DV 0.83 0.64 0.46 0.27 0.09
Dev 0.17 0.14 0.12 0.02 0.11
Slope -0.19
Intercept 1.0
RSQ (IV-DV) 0.81
In fact, the relationship between IV and DV exists because the subject is
controlling the perceived area of a rectangle (the controlled variable, CV
or controlled input, i) trying to keep it at 1 cm2 (the value of r). The
actual values of the controlled variable in each disturbance (IV) condition
were as follows:
CV (i) 1.00 1.00 1.00 1.00 1.00
The RSQ of the DV (o) regressed on i (taken as the IV) was 0.11. So i (the
controlled variable) would be rejected as an important variable by the
conventional analysis (indeed, the RSQ value was as high as it was -- 0.11
- - only because control was not perfect; it is a result of random variation
in the controlled variable that happened, by accident, to be correlated with
variation in output (DV)). In fact, the controlled variable is what this
behavior is about, and the conventional analysis misses it or rules it out.
The observed relationship between IV and DV exists, not because the IV
causes variations in the DV but because the subject is keeping a controlled
variable (area of a rectangle) in a reference state.
In the range of IV values selected for this experiment, the relationship
between IV and DV appears linear. It is actually non-linear (the relationship
is DV = 1/IV). But the important point is that the nature of the relationship
between IV and DV depends on the geometry of the situation, not on the nature
of the subject. This is the Behavioral Illusion; the observed relationship
between IV and DV in this experiment depends on the nature of the environment
(actually, the nature of the environmental relationship between IV and DV ),
NOT on the nature of the subject (which these experiments are design to
explore). The nature of the relationship between IV and DV can be easily
changed by changing the feedback relationship between output (DV) and input
(CV). Ways to do this are left as an exercise. But when you do it,
you get a whole new relationship between IV and DV -- with no change in
the subject. Scientific psychologists who "get" the Behavioral Illusion
either 1) stop getting it (it just COULDN'T mean THAT) 2) jump out of 10th
story windows or 3) stop paying attention to conventional research and
start doing PCT research. The first option is by far the most common.
Bill Powers (951127.1350 MST) --
Martin and Rick discussing inference of causality:
Is causality really something that can be "inferred?"
No. But conventional psychologists say things like "correlation doesn't imply
causality" and I was playing along. In fact, conventional psychologists
should be testing models. Actually, they are testing a model (a version of a
lineal causal model); the model keeps failing but that doesn's seem to phase
them; there's always statistics.
Best
Rick