[From Bruce Abbott (951203.1645 EST)]
Dag Forssell (951128 1420) --
It seems to me that before we can determine whether anyone's work falls
within the realm of science, we have to develop criteria. We can then apply
those criteria to any given example and decide whether the work (or its
product) qualifies.
"Science" is sometimes defined as an organized body of knowledge obtained
through the use of the scientific method. This fits the literal meaning of
the term, which is "knowledge," while restricting the term to knowledge
obtained in a specific way.
"Science" is also defined as the process through which scientific knowledge
is accumulated. In this sense the terms "science" and "scientific method"
are synonomous. "Scientists" are people who use the scientific method to
build an organized body of knowledge about a subject.
The key, then, to whether a given activity qualifies as science or not is
whether it employs the scientific method to build an organized body of
knowledge. Thus, to develop our criteria, we must know what the scientific
method is. The scientific method comprises several interlocking elements.
The most basic element is _observation_. All scientific knowledge rests on
it. An _empirical_ question is one that can be answered through
observation; if a question is not empirical, it cannot be addressed through
science. This is what puts philosophical questions like "does God exist?"
or "how many angels can fit on the head of a pin?" out of bounds for science.
But not all observations would qualify as scientific. The observations must
be _systematic_ (carried out by adhering to a well-defined plan) and they
must be demonstrated to be _reproducable_. Systematic observations usually
involve measurement, so that quantitative relationships can be established
among variables, although simple classification is often used as well. The
reliability (reproducability) of observations is established through
systematic _replication_.
An observation is accepted as scientific if it is obtained through
systematic observation and can be replicated.
A second essential element of the scientific method is the development of
hypotheses and theories. Here I will use the term "hypothesis" to mean a
statement of a relationship among two or more variables and "theory" as a
set of assumptions about varibles and their relationships (including
variables not directly observable) from which hypotheses may be derived. To
be considered scientific, a new theory must be consistent with the
established "facts" of the discipline, or if not, give reason for believing
those facts to be in error (in the light of the theory).
A third essential element of the scientific method is testing. Withoug
testing, hypotheses and theories are just assertions, even though they may
appear reasonable in light of available evidence. Testing involves setting
up (or finding in nature) conditions under which the hypothesis (or specific
predictions from theory) can be assessed. If the relationship specified by
the hypothesis (or prediciton from theory) is confirmed by the test, then
the hypothesis or theory is supported; otherwise the hypothesis or
prediction is disconfirmed, and so, technically speaking, is the theory from
which the hypothesis or prediction was derived. [However, practical
problems such as incomplete control over extraneous variables, limits to the
precision of measurements, and perhaps ambiguity in the theory or in how to
apply it to a particular case, may limit the researcher's certainty as to
whether a given disconfirmation is actually the fault of the theory or a
result of experimental error or misapplication. In this regard, the more
"rigorous" the theory, the easier it is to determine that a given outcome
has actually disconfirmed a prediction of the theory. The most rigorous
theories are specified mathematically and yield quantitatively precise
predictions.]
Theories that fail to correctly predict some outcome (when the reliability
of this finding has been established) are revised in the light of this
result so as to be consistent with established scientific fact _and_ the new
finding (if the theorist can figure out a way to do this!) or discarded. As
Kuhn notes, even theories known to have problems (and thus known to be
wrong!) are usually retained if they have demonstrated usefulness in
organizing the data and making correct predictions in other areas, so long
as there is no better theory to replace them. Newton's conception of the
universe is fundamentally wrong, makes incorrect predictions under certain
conditions, and is _still_ in use for most practical work, even though, in
this case, better theories are available; in cases where there _are_ no
better theories, even poor theories (imprecise and/or known to make
incorrect predictions even within a limited domain) may continue to be used
while researchers look for something better, because they organize the data
and help to guide further research (heuristic value).
To be considered scientific, a theory must be "testable" in this way. That
is, it must be capable of disproof. There must be some empirical test of
the theory to which the theory can be submitted, which if failed will
disprove (disconfirm) the theory. Looser theories (those that do not make
precise quantitative predicitons) are more difficult to disconfirm than
tighter theories. Even those that do make precise quantitative predictions
may be difficult to test if the variables involved cannot be controlled
and/or measured with precision, as then the experimental error is too wide
to assess the accuracy of the prediction with any confidence. Even so, it
may be possible to reject a prediction if the observed function has the
wrong _form_. For example, if the theory predicts that two observable
variables will be linearly related and what is observed is a parabola, the
theoretical prediction is incorrect.
A Dispute about Criteria
Dag, it seems to me that our dispute is a dispute over criteria. I would
hold that anyone engaged in any part of the activities I have listed above
is doing science. This would include making systematic observations under
carefully defined and noted conditions (and validating those observations
through replication), systematizing the observations by developing
well-defined categories (and perhaps hierarchies of categories) and sorting
the observations into them, developing theories and hypotheses consistent
with established scientific knowledge on the subject, and systematically
testing those theories and hypotheses through properly designed
manipulations and further observations. I hold that most of what goes on in
conventional psychological research falls into one or more of these
activities and therefore qualifies as scientific work.
Your criteria apparently are different. For you, science seems to involve
the following set of activities:
1. Developing a structural model based on physical elements whose reality
can be determined (at least theoretically) by looking for them (e.g.,
signals, transducers, comparators within the physical system being
studied), and whose functional properties are mathematically specified
in the model. By specifying these properties and the linkages among
these elements, one defines a system whose behavior can be deduced
by appropriate mathematical techniques or by simulation. The elements
and their connections must be consistent with what is known about the
layout of the physical system being modeled.
2. Testing the model against data collected under conditions under which
the model would be expected to apply.
3. Modifying the model in the light of discrepancies between the observed
behavior and that predicted by the model, until model and observations
agree within a small margin of experimental error.
4. Exploring the physical structure to determine whether the elements and
connections of the modified model correspond to elements and connections
of the physical structure.
I hope I haven't left out anything, or misstated your criteria. Assuming
that this is correct, I agree that this is certainly "high science," the way
science should be done whenever the goal is to identify the mechanism
beneath the observations. A good mechanism (one consistent with both the
observations of physical structure and with observable behavior) provides a
powerful means of prediction and (possibly) control, not to mention
"understanding." Any science that can achieve this clearly deserves the
name "science." My only question for you is whether it is possible that
something less than this should qualify as science, if it follows the
methods I outlined earlier.
Before going off to graduate school to become an experimental psychologist,
I worked for five years at the Owens-Illinois Glass Company's Technical
Center as a technician. I worked in Glass Science, a department within
Corporate Research that employed about a dozen engineers and Ph.D. chemists.
Their job was to develop new knowledge about glass which the company could
exploit to produce new products. I don't know for sure, but I suspect you
count physical chemistry among the "hard" sciences. Do you know how I spent
my time? Doing emprical studies of the properties of various mixtures of
chemical compounds which, when heated sufficiently, tend to produce a glass.
So far as I am aware, not once did any of these Ph.D. researchers begin with
a model of the physical structures (molecules, in this case), hypothesize
linkages among them in the mixture, and come up with a precise quantitative
prediction about the properties (behavior) of the resulting glass. The
situation may have improved since 1973 (when I left O-I), but at that time
an expert in physical chemistry could not even predict whether a given
mixture would _form_ a glass, based on its composition and the temperatures
to which the mixture would be exposed. In fact, one could say that the
whole point of the research was to find a model from which such predictions
could be made. So we did empirical studies, varying the composition and the
specific temperatures to which the mixture would be exposed in a systematic
way and then measuring the physical characteristics of the resulting glass
(if indeed you got a glass) along a large number of dimensions.
Now I still think we were doing science, there in Glass Science, back in
1973. By your criteria we were doing mush. But if this is mush, where does
the knowledge come from that eventually allows one to construct the precise
mathematical models that are the sine qua non of an advanced science? It
would seem to me that there have to be quite a few "mushists" at work
providing the foundation so that the "real" scientists can come along and do
the real science.
Goals of Science
Philosophers of science usually say that there are three goals of the
scientific enterprise: understanding, prediction, and control. As I have
already indicated, a model that specifies a physical mechanism and has been
validated empirically can be said to provide a deep understanding of those
phenomena of relevance to the theory, allows precise prediction, and may
make excellent control of the (real) system's behavior possible. Who would
want less, if this is possible? Yet even when there is as yet hardly a clue
as to how a model can be constructed of a given system, it is still possible
to go a fair way toward reaching those scientific goals, although
necessarily the results will be much more limited. For many decades local
meterologists predicted the weather based on a knowledge of relationships
among locally-measured variables -- temperature, barometric pressure,
relative humidity, wind direction and speed, type and amount of cloud cover
-- and the dynamic changes in those variables. The empirical relationships
among these variables did not permit quantitatively precise predictions of
the weather, even over so short a term as six hours, but the predictions
made were certainly better than what had been available before these
variables had been systematically measured and the relationships among them
discovered. One can also say that having knowledge of such relationships
provided a certain level of understanding. Why is it raining? Well, the
wind swung around to the east, the barometric pressure dropped, the humidity
was rising, and the cloud cover was increasing. It's raining because rain
usually follows such changes. Today we might offer a "better" explanation:
it's raining because a cold front is passing through, or because a "low"
bearing considerable moisture is passing over the state. Such knowledge
even permits a degree of control. If it's going to rain, I'll reschedule my
picnic. Just discovering empirical relationships (especially the ones
labeled "causal") and developing a quantitative model of them can be of
practical value even though one does not yet possess a model of the
mechanism, but only of the behavior of the mechanism as seen under certain
specifiec conditions.
The laws of physics which lie beneath the ebb and flow of the weather are
well-understood, yet even the best computer models cannot predict the
weather in, say, Fort Wayne, Indiana, with a high degree of accuracy, even
over the short term. Physics is science, but meterology, evidently, is mush.
Much of what goes on in psychology is aimed at discovering stable
relationships that might permit some degree of prediction and (perhaps)
control. The models developed for these situations are limited and may not
generalize very far beyond the conditions within which the model was
originally designed to function. Researchers get interested in addressing
certain phenomena, and try to identify the conditions under which those
phenomena occur (and do not occur) as part of the effort to understand what
is going on. People get angry, as indicated by a certain constellation of
changes in their actions toward other people, themselves, or in some cases,
inanimate objects. What are the necessary conditions that bring about this
state? In the most general terms, how does behavior change? What are the
consequences of the resulting actions? Can we predict what will make a
person angry? Can we predict what the person will probably do when in this
state? To try to answer these questions, researchers develop hypotheses,
manipulate variables, assess whether the hypotheses are supported by the
results, revise their thinking in light of the outcome. The physical
mechanism beneath these relationships remains mostly unknown and
unspecified, but some useful information about the variables involved and
their relationships has been developed, and this information may prove very
useful for guiding the devopment of a physical model. Meanwhile, more of
practical value is known (scientifically established) than was available
before the research was done (which may have been nothing more than myths
and aphorisms). Even when "common sense" notions are confirmed, what was
just common sense is now part of scientifically established knowledge.
Science, Mushy Science, and Mush
Perhaps what we need is to recognize that there is a continuum between what
you call science and the pure mush of unconfirmed speculation, untestested
(and unstestable) theory, and anecdotal evidence. Much of what goes on in
normal, everyday scientific work is not the testing and refinement of
quantitative physical models, but it is still science in that it employs the
methods of science to establish a body of knowledge. During the early
stages of a new science, most of the work will necessarily be directed
toward establishing a body empirical phenomena, identifying the variables at
work in those phenomena and the relationships among the observables, and
developing and testing hypotheses about how the system functions. Only when
sufficient knowledge has been accumulated can one begin to develop and test
a quantitative physical model, and even then one has to understand what is
possible.
In this regard, there is a reason why the model known as control theory
developed first within electrical engineering and not psychology, in
addition to the one proposed by Bill Powers (use of IV-DV methods).
Electrical engineers had not only the proper formal mathematical training,
but were dealing with a system of elements of known structure and function,
which they could manipulate at will simply by wiring them up in whatever way
interested them. Already possessing some basic laws (e.g., Ohm's), they
could write the proper equations describing the functional relationships
expected (within experimental error) of each component (reisistor, inductor,
etc.), work out what is to be expected when one feeds back a portion of a
differential amplifier's output to one of its inputs, then build the
physical circuit and compare prediction to the system's actual behavior.
Piece of cake! All it took (besides the right training) to come up with the
answer was a practical problem that got the right guy thinking about how to
solve it.
Meanwhile, here are all these researchers in psychology, facing a system of
enormous complexity and lacking a means to anything more than manipulate
variables and observe the result. Nevertheless, they succeed in building an
empirical model (reinforcement theory) that seems to work well in a lot of
situations under investigation, based on a selectionist principle borrowed
from evolutionary theory. Research keeps on turning up difficulties for the
theory, and soon there are numerous proposals on modifications to, or
elaborations of the theory in order to account for these anomalies. I don't
know about you, but this doesn't sound like an untestable theory to me. If
it's untestable, why does it keep getting into trouble? But it remains
useful in that it has provided a guide to further research (heuristic value)
and has had success in certain applications, and the hope remains that the
right set of additional assumptions will be found to make it work
successfully in all cases. (Not MY hope, by the way; I have a better theory
to go on.)
Probably the single most important reason why psychology has not attempted
to develop a purely mechanistic (as opposed to descriptive) account of
behavior is the belief that such an account is not possible given the
current state of our knowledge regarding the structure of the brain.
Psychologists have from time to time attempted to construct such models, but
they have been based on an erroneous conception of the nervous system (S-R
reflex; Pitts-McCulloch neurons; digital computer) or have involved
structures presumed (from theoretical considerations) to exist (perhaps in
distributed fashion) which carry out inferred functions (e.g., memory
storage and retrieval) in ways for which there is insufficient knowledge to
posit a physical organzation to do the job. What results is speculative
enough for you to define as mush, and unsatisfactory enough for most
psychologists to avoid them. As a result, psychologists have learned to
keep their speculations restricted within narrow areas of research and to
depend mainly on discovery of empirical relationships in the belief that
eventually there will be enough known to begin building the foundation from
which the descriptive science's relationships can be deduced.
Please note that I am not defending this state of affairs, although I think
I understand how it came about, nor am I arguing that descriptive theories
(like Kepler's) are on a par with mechanistic ones (like Newton's). There
has been a whole series of events in the history of psychology which has
prevented it from moving from descriptive and quasi-mechanistic theory to
physically-grounded mechanistic theory, even though the information needed
to move from the former to the latter has been around for a long time. But
I am arguing that much of what scientists do in most fields is not
restricted to building and testing such models; where you see this kind of
activity most in evidence is within the applied fields of well-developed
sciences for which the required foundation of basic theory already exists.
In those areas the question is not so much trying to understand new
phenomena for which existing theory fails as in seeing how to apply existing
theory to specific cases. For engineers used to doing only the latter,
anything less may seem like mush. What you may not understand is the amount
of effort by scientists doing the kind of work you would not define as
scientific that had to be completed before the magnificant theories you see
as the only product to true science could be conceived and tested.
Finally, I want to correct an impression you left of my view in your post.
I do not accord the term "science" to a person's work because that person
"tried hard." A person's work is "science" if it followed the methods of
science. Whether the person's theoretical analysis stands the test of time,
only time will tell, and in many cases what seemed on the best evidence
available at the time to be a valid explanation has turned out to be
fundamentally wrong. Even Newton was wrong (even though under most
conditions the results given by his analysis are correct to twenty decimal
places), but that fact does not convert Newton's work from science to mush.
I will hold Thorndike's work as science even though his explanation did not
provide a physical mechanism, and even though it has proved to be wrong,
simply because it was based on systematic observation, was consistent with
the data and the then-current theories of physiology, and yielded testable
predictions. The fact that, given the information available, Thorndike
could not construct a physical model that might be found to have a physical
reality within the brain, does not disqualify Thorndike's work from the ream
of the scientific, in my view.
Regards,
Bruce
) can handle data, it is only