[From Bill Powers (971122.0958 MST)]
Bruce Abbott (971122.1135 EST)--
Another, less pejorative term for prejudice is "generalization." All
sciences search for general rules that apply over a broad range of
individual cases, and for further refinements of these rules that permit one
to account for the apparent exceptions. If what you say is true of
psychological research, then it is true of all scientific research.
Thanks for the opportunity to expand on what I mean. This word
"generalization" can be taken in two entirely different senses. One sense
is the statistical sense: "men drive faster than women do." The other sense
is what I could call the "property" sense: "men do not generate ova; women
do not generate sperm." The difference between these two meanings of
generalization lies not just in our expectations, but in our understanding
of nature.
The generalization "men drive faster than women do" leads us to expect that
in any large enough random sample of men and women, we will find that the
average speed at which the men drive is greater than the average speed at
which the women drive. This does not, however, lead us to expect that every
man will drive faster than every woman, considering them in pairs. In other
words, this kind of generalization is not expected to hold true for every
observation.
The other generalization, however, is of the kind to which we expect no
exceptions at all. Not only would we be astonished to find a woman who
produced sperm, but such a discovery would call into question a very large
body of observations and theories, all the way down to our understanding of
chromosomes and genes. Any exception to this generalization would call for
immediate investigation and would require revision of all our understanding
that tells us that women not only DO NOT produce sperm, but they CAN NOT
produce sperm.
When we say that men drive faster than women, we do not mean that women
CAN'T drive faster than men -- only that in general they don't. There is no
basic reason that we couldn't have observed the opposite effect; it's
perfectly possible that we might have found that women drive faster than
men. Such a finding wouldn't upset any basic understanding of nature -- at
worst, someone's hypothesis would be shown incorrect. And along the same
lines, if one observed a women driving faster than all the men in
surrounding cars, nobody would be surprised or upset. This kind of
generalization is not disproven by any number of counterexamples, as long
as the population effect remains the same.
So what's the difference between these two kinds of generalization? Is it
just a matter of sigma? Is it just that some phenomena show a very small
scatter, while others show a large scatter? I don't think so.
The difference is in whether we base our expectations strictly on
observations, or whether we base them on an underlying -- imaginary --
model of how things work. When we use an underlying model, the model not
only predicts what we observe; it predicts what MUST happen and what CANNOT
happen. A factory-tuned Ford Escort cannot go faster than a factory-tuned
Ferrari can go. It's not that the top speed of an Escort is usually found
to be less than the top speed of a Ferarri; it's that the physical
properties of the Escort do not permit the same top speed that the physical
properties of the Ferarri permit, under the same conditions. This is, of
course, a generalization -- but it's one we expect to hold true in every
instance. And the reason we expect it is not based on observations of
populations of Escorts and Ferarris going as fast as they can; it's based
on knowledge of how these cars are constructed. For the Escort to beat a
Ferarri going at top speed, either the Escort must contain something other
than an Escort engine, or the laws of physics must be wrong.
So statistical generalizations tell us what is likely to happen, but
"property generalizations," or model-based generalizations, tell us what
must and must not happen if the model is correct. The difference in
usefulness of these two type of generalization is as great as the
difference between psychology and physics.
Now, "prejudice." What is wrong with prejudice, when the word is used
pejoratively? It is simply the risk of misjudging someone. Clearly, the
risk decreases as the scatter of the data decreases, but as in all
statistical generalizations there is still a nonzero risk involved in
judging a person according to the mean behavior of a group with which he or
she shares some common but irrelevant characteristics. To be reasonable, we
have to admit that when we have no underlying model, we must make
judgements based on experience rather than knowledge of properties. But
where do we set the threshold of reasonableness?
Consider the "welfare queen" who arrives in her new Cadillac at the welfare
office to pick up her 12 welfare checks under 12 names. This has clearly
been observed to happen, allowing for inflation of a good story. But how is
this observation to be brought to bear on another woman picking up her
welfare check from the same office, who has the same color of skin, who is
also ostensibly poor, who lives in the same neighborhood, who has the same
number of children by unknown fathers, and who speaks in the same patois?
Is this other woman also cheating the system, having children to pick up
support payments, and so on? If you listen to the people who want to do
away with welfare, the inference is clear: they're all the same, and they
all need to be taught a lesson.
That's obviously a case of prejudice. It happens that the conclusion is
also statistically wrong, but what of the cases where it's statistically
right? There can be just as much prejudice involved when the observations
support the generalization.
Suppose it were true that in a sample of 1000 welfare recipients who share
certain physical and circumstantial characteristics, 600 were found to be
cheating in some way. So the generalization, "Welfare applicants with these
characteristics are cheating the system" is statistically correct. How do
you use this generalization in dealing with the next applicant? Do you look
at the characteristics shared with welfare cheaters, conclude that this
person will probably also cheat, and deny the application? If you do that,
you will turn away 400 out of every thousand applicants who would NOT cheat
the system, and who actually need the aid just to live in poverty.
I'll leave it to Richard Kennaway to work out the threshold above which one
could say that using statistical generalizations to judge individuals does
not amount to prejudice. I'm sure you could work it out just as well, but
Richard is not biased in favor of the existing methods. And I'm sure you
will find that most psychological "findings" fall considerably below any
defensible threshold.
Best,
Bill P.
···
The general rules identified in scientific research allow one to make
predictions that will be generally true -- e.g., most normally-dressed
people will be comfortable at a room temperature near 72 degrees Fahrenheit.
To be more accurate, one needs to learn how other observable variables
affect the basic general relationship. For example, in the
temperature-comfort relationship, one might discover that, as the level of
muscular activity increases, the room temperature at which people are
comfortable tends to drop. To predict for an individual case, one needs to
assess the state of the relevant variables in that person.
The use of generalization is not restricted to scientific research. All of
us generalize, all the time. You expect the sidewalk beneath your feet to
support your weight, even though you've never walked on that particular
sidewalk before. You expect your car to start this morning, because cars
are supposed to start when you turn the key and because this one has done so
in the past. You expect that the people you pass at the mall to politely
keep a certain distance rather than deliberately shoving you aside. Most of
the time these assumptions serve us well, allowing us to adjust our behavior
in a way that is most likely to fit the circumstances. We do not normally
think of these expectations as prejudices.
It is only when we attribute undesirable characteristics universally to
_all_ individuals of a group, based solely on their group membership (e.g.,
all members of X are lazy), that this attribution would normally be called
"prejudice." In the worst cases, the generalization made is not even true
of the group as a whole (e.g., members of X in general are no lazier than
others).
When you state that most psychological research is built on prejudice, some
may infer something other than what is conveyed by your definition of
prejudice. The statement can be taken to suggest that the results of
psychological research offer nothing more than the researcher's preconceived
prejudices or biases rather than an evaluation based on objective evidence.
This idea is itself an expression of prejudice.
Regards,
Bruce