[From Bruce Abbott (951126.1325 EST)]
Bill Powers (951125.1605 MST) --
Bruce Abbott (951125.1510 EST)
An "independent variable" is not a variable that "assumes an
arbitrary value with no relationship to anything else." It is a
variable whose value (state) is manipulated by the experimenter.
Independent variables exist whenever an experimenter manipulates a
variable.Precisely the point I made later in the post:
The very concept of an experimentally-manipulated independent
variable, therefore, implies a control system in the background, a
system which can act to make a perceived pattern of change match a
reference pattern.
Yes. Your first-offered definition is a straw man--nobody defines
"independent variable" that way (or at least nobody I know of). You then
argued for what is actually the accepted definition, and note that this
latter definition "implies a control system in the background." I agreed.
But my point seems to have escaped you:
I am unclear about this distinction you make here. If an
independent variable is being manipulated for experimental
purposes, it cannot be changing naturally, as then one would not
have an experiment but only a type of correlational design.An environmental variable might change because other variables in the
environment change. The illumination in a cage might change because the
sun shines in the window. Or it might change in a way planned by an
experimenter, who adjusts the pattern of light and dark -- say, on a 27-
hour schedule. The first change is natural, and would be affected by the
season, by cloudiness, and by the state of the windowshades. The second
is under control by the experimenter, following a predetermined pattern.
There is no other essential differencd.
Your point did not escape me. Here is what you said that confused me:
When independent variables are manipulated for experimental purposes,
either they change naturally as other environmental variables change, or
the variables are deliberately made to change according to a
predetermined pattern.
By your own definition (i.e., the commonly accepted one), independent
variables cannot be "manipulated for experimental purposes" by allowing them
to change naturally: such variables are not independent variables at all. I
understand your point that independent variables imply a control system
working in the background to set their values; what I don't understand is
this contradiction.
The first change is natural, and would be affected by the
season, by cloudiness, and by the state of the windowshades. The second
is under control by the experimenter, following a predetermined pattern.
There is no other essential difference.
There is no essential difference from the point of view of the system under
study, but there is an absolutely HUGE difference from the point of view of
the experimenter, who does not have direct knowledge of the system's
structure. When variables change naturally, there is always the possibility
that other variables change with them. For example, smoking may increase
the liklihood of cancer by introducing irritants into the lungs, but the
mere association of quantity of cigarettes smoked and lung cancer rate does
not by itself establish smoking as the cause. There may be many other
variables differentiating smokers from nonsmokers (on average) that could
influence cancer rate. When a chick is observed to follow its mother, there
are many perceptual variables that all "come and go" with her in unison: her
size, general shape, the sounds she makes, her particular markings, her
smell, the way she moves. Which of these perceptions are key in allowing
the chick to recognize its mother and to differentiate her from all other
hens? You can't tell just by watching (although you _can_ generate some
good hypotheses that way). Instead, you must manipulate _one_ of these
variables while holding the others constant. If you change one variable
while holding the rest constant, any reproducible change in the dependent
variable must demonstrate an influence of IV on DV.
Now let us try a different experiment. I deliver a shock through
the grid floor of an operant chamber and the rat runs in a circle
around the chamber so long as the shock continues. Shock, the
independent variable, appears to be a cause of running, the
dependent variable. Taking a cue from physics, I then force the
rat to run (perhaps by reinforcing running behavior through
successive approximation). To my surprise, the rat's running
generates no measurable current in the grid. The reason, of
course, is that running and shock-level are not mutually-
influencing variables.Yes, but you're misidentifying the system in this case. The rat's
running is affecting the experienced shock, over time. The rat runs not
because of a single shock, but because of an effect experienced over
many instances of shock, and it runs because the running has an effect
on the experience. The experience of importance, therefore, is the joint
effect of administration of shocks and running, not the shocks alone or
the running alone. There is a relationship involving administered
shocks, running, and the experiences arising from shock.
The point I'm trying to argue with this example is that there are variables
out there that do not participate in mutual interaction. If the sun's rays
pass through my bedroom window onto my eyelids and therby awaken me, it does
not follow that if I awaken in the night, the sun's rays will then fall upon
my eyes. The rat's running or my awakening and turning my head away from
the window both affect the perception of those variables, it is true. I am
certainly not arguing against that position! We seem to be talking about
different things. I am discussing what the experimenter can do to learn
about the system under study and therefore am adopting the experimenter's
point of view, whereas you wish to discuss the variables involved from the
system's point of view.
The electric current is also a controlled variable. The experimenter
varies the schedule and strength of shock until some satisfactory
pattern of running is seen. So the administration of shock is the
experimenter's output, and the perceived pattern of running is the
controlled variable. If the rat does not run, the schedule is adjusted
or the strength of the current is raised until it does run. If it runs
too much -- continuously, or in some undesired pattern -- the
administration of shocks is against adjusted until the desired pattern
is seen (or the desired range of patterns). So the shocks affect the
running, and the running affects the shocks, when the entire
experimenter-animal system is considered.
Experiments are not usually run in the manner you describe, although there
may be some of that during pilot work to establish a set of effective
parameters for the experiment. The experimenter may decide to test the rat
using shock intensities ranging from 0.1 mA to 1.0 mA. Whether the rat runs
or not, and with what intensity, is up to the rat. The feedback path in
which the rat's running affects the experimenter's setting of the shock
intensity would not be present during the actual experiment and therefore
need not be taken into account when the data are analyzed and interpreted.
What follows in your post is a nice discussion of the organization of living
systems, showing that both one-way and mutual influences are possible, and
noting that in PCT it is assumed that all behavior is mediated by
closed-loop systems of mutually influencing variables. I don't disagree
with any of it.
Yet no manipulation of any of these variables will change either r
or d, although changing either of the latter will force a
simultaneous change in p, e, o, and i as they approach new
equilibrium values. A set of simple experiments designed to
identify the causal relationships would show that both r and d
affect p, e, o, and i, that changes in r do not affect p, and
changes in p do not affect r.This is true, and both r and d have been identified in PCT as true
independent variables (relative to the loop). But they are causal only
in a _joint_ manner; neither one alone is a cause of any unique
behavior. And the effect is an effect on the entire loop, not just one
variable in it such as o.
When we control the value of, say, i, then during the experiment it has the
status of an independent variable and we can see how varying i changes the
other variables around the loop. When you say that r and d are "true"
independent variables, I take you to mean that the influence between r or d
and the other variables is one-way, whereas normally i would both influence
and be influenced if we did not "clamp" its value for the experiment.
That's fine; I just want to clarify this as, in the broader definition, ANY
variable manipulated by the experimenter is a "true" independent variable.
In establishing this distinction between "true" and "false" (?) independent
variables, you return to an isomorphism between IV-DV and cause-effect which
I had avoided in my broader definition of the term. If independent
variables are only those having a one-way influence during normal system
operation, then an independent variable is a cause and a dependent variable
(if it varies with the independent variable while all other variables are
held constant) is an effect, as we normally use these terms. But there is
nothing in experimental logic that prevents one from manipulating variables
that normally participate in a set of mutual influences. For example, in a
control system I can hold the reference level and disturbance constant while
varying the perceptual input, and watch what happens to the error level and
output. If I'm clever, I can establish the perceptual input function, the
comparator function, the output function, and the environment function. The
fact that I do this by manipulating an independent variable does not mean
that I buy into a system of lineal causality in which causes and effects can
be traced sequentially around the loop.
This is true, and both r and d have been identified in PCT as true
independent variables (relative to the loop). But they are causal only
in a _joint_ manner; neither one alone is a cause of any unique
behavior. And the effect is an effect on the entire loop, not just one
variable in it such as o.
The key word in your second sentence is "unique." If you hold the reference
constant and vary d, you get predictable output. If you hold the
disturbance constant and manipulate r, you get predictable output. There is
nothing in the method that prevents one from determining that both variables
affect the output (or any of the other within-loop variables), or from
assessing their interaction (joint influence). Nor is there anything that
prevents one from determining that this influence is on the entire loop and
not just one variable.
There is nothing wrong with IV-DV in itself: it's simply a method of
dealing with apparent causality. But I emphasize the "apparent." One can
get entirely the wrong impression of how a system works by using the IV-
DV approach naively. This is the main point that PCTers have been trying
to make.
I disagree with your statement that IV-DV is "simply a method of dealing
with apparent causality," unless you include circular causality within that
category. More generally, IV-DV, properly applied, establishes which
variables influence which, whether that influence is unidirectional or
bidirectional, simultaneous or apparently sequential. The problem is not
that IV-DV implies lineal cause-effect, but that it has often been
interpreted that way by those who employ it.
But it seems to me that you have _not_ been arguing against a "naive" use of
IV-DV methods, otherwise I don't know what we _have_ been arguing about. My
understanding is that you were arguing against IV-DV methods, period, for
the reason that, in your view, they _are_ only methods for studying apparent
causality, in an era when notions of cause-effect have been banished from
other sciences. When Rick says that IV-DV methods cannot tell you anything
about control, and you seem to agree (at least, I don't recall your
correcting him on that score), I don't know what other conclusion to draw.
But perhaps I have not perceived your position correctly. At any rate, it
does seem that we do agree that such methods can be used fruitfully in the
study of living control systems, so long as one does not interpret the
results naively in terms of lineal cause and effect.
In your vocabulary of present-time relationship, do you distinguish
relationships in which there exists a direct influence of one
variable upon another (or even mutual influences) from
relationships in which two variables change together over time, not
because they directly influence each other, but because each is
responding to the influence of the same third variable? In the
traditional vocabulary one would say that the former relationship
is causal whereas the latter is only correlational. If you do make
this distinction, what terms convey it? It seems to me that the
term "causal" has continued to enjoy acceptance in psychology
precisely because it so nicely conveys this distinction.In a system analysis, where component input-output relationships are
expressed mathematically and then the whole system of equations is
solved, there is no need for this artificial distinction. All such
distinctions fall out of the solution.
Sorry, I'm talking about a system whose structure remains to be determined,
not one for which the system equations are known. If a third (unknown)
variable influences both of two variables under observation, variations in
the third variable will make it appear as though the two observed variables
are directly linked. Mere observation of the two variables during normal
system operation cannot determine whether this direct link is present or
merely an illusion, a byproduct of their common link to this third,
unobserved variable. This is an analytic problem that requires an IV-DV
approach to resolve.
You have made the valid point that there are, indeed, one-way causal
relations in nature. They are seen whenever you draw a boundary around
some part of the environment to create inputs and outputs. Whether you
intended to or not, you also pointed out that there are one-way
relationships which depend on power amplification, and these, piecewise,
are also true cause-effect relationships. In such cases, IV-DV analysis
is entirely appropriate. But by the same token, IV-DV can give a false
impression of what is going on between input and output -- it can give
the impression that there is a simple causal chain connecting input to
output, when in fact the interior of the system can be organized in ways
that are not describable that way -- a control organization being one,
but not the only one.
That false impression can be avoided if one begins testing with a knowledge
that close-loop systems not only exist, but are common, and with a firm
understanding of how closed-loop systems work and can be analyzed. One is
less likely to commit errors of interpretation if one is aware of alternate
possibilities and knows how to test for them.
As to cause-effect and IV-DV not being the same thing, I'm not sure what
you mean. Explain again, please.
I think I just did!
Regards,
Bruce