Download Interactions in Psychology: Beyond Two Disciplines by Lee J. Cronbach and more Study notes Psychology in PDF only on Docsity!
Beyond the Two Disciplines of
Scientific Psychology
LEE J. CRONBACH Stanford University
The historic separation of experimental psychology from the study of individual differences impeded psychological research, So I argued when last I had occasion to address the APA audience (Cron- bach, 1957). It was time, I said, for the'manipu- lating and the correlating schools of research to crossbreed, to bring forth a science of Aptitude x Treatment interactions (ATIs). As that hybrid discipline is now flourishing, a progress report on ATI studies is the appropriate first business of this article. It is not practical to treat here the studies of ATIs in social behavior (e.g., Fiedler, 1973; McGuire, 1969), ATIs in re- sponse to drugs and therapy (e.g., Insel & Moos, 1974; Lasagna, 1972; Schildkraut, 1970), or ATIs in learning and motivation generally. I confine my- self to ATIs related to instruction, drawing on a comprehensive review Richard Snow and I have just completed (Cronbach & Snow, in press). In that field, several research programs have brought us a long way; particularly to be acknowledged are the sustained inquiries, of Bill McKeachie, Jack Atkinson, Russ Kropp and Fred King, George Stern, David Hunt, Victor Bunderson and Jack Dunham, and Snow and his graduate students. Important as ATIs are proving to be, the line of investigation I advocated in 1957 no longer seems sufficient. Interactions are not confined to the first order; the dimensions of the situation and of the person enter into complex interactions. This complexity forces us to ask once again, Should social science aspire to reduce behavior to laws?
This article was presented as a Distinguished Scientific Contribution Award address at the meeting of the American Psychological Association, New Orleans, September 2, 1974. I thank Denis Phillips, Robert Calfee, and Lee Shulman for critical comments. My thinking owes much to Richard Snow's enthusiastic collaboration over the years, but I do not speak for him in this article. Requests for reprints should be sent to Lee J. Cronbach, School of Education, Stanford University, Stanford, Cali- fornia 9430S. •
116 • FEBRUARY 1975 • AMERICAN PSYCHOLOGIST
Some 30 years ago, research in psychology became dedicated to the quest for nomothetic theory (Hil- gard & Lerner, 1951; Koch, 1959; Merton, 1949). Model building and hypothesis testing became the ruling ideal, and research problems were increas- ingly chosen to fit that mode. Taking stock today, I think most of us judge theoretical progress to have been disappointing. Many are uneasy with the intellectual style of psychological research (Gergen, 1973; Glass, 1972; Israel & Tajfel, 1972; McGuire, 1973; Newell, 1972). Here I shall cut short my comments on ATIs as such, in order to join in that discussion. I shall express some pessi- mism about our predominant norms and strategies and offer tentative thoughts about an alternative style of work. My sense of the importance of this discussion is heightened by Don Campbell's Lewin Memorial Award address (see Campbell, in press). I would not accuse him of agreeing with me, but if you put our articles side by side you have a binocular view of the scene. First, we can take a look at ATIs themselves. The typical ATI study is a two-group experiment. The measure of outcome is regressed onto a score recorded prior to treatment. If the regression lines in the two treatments differ in slope, that is evi- dence of Aptitude X Treatment interaction. Treat-, ment A in Figure 1, thoflgh best on the average, is not equally superior all along the aptitude scale. Snow and I give a general meaning to the term aptitude, letting it embrace any characteristic of the person that affects his response to the. treat- ment. To illustrate research designs and results, I shall describe work by Domino and by Majasan. Both of these have to do with stylistic variables rather than abilities.
Personal Style or Belief as an
Interacting Variable
Domino (1968) investigated the relation of college success to the Ai (Achievement via Independence)
and Ac (Achievement via Conformance) scores of Cough's California Personality Inventory. A stu- dent scores high on Ai if he says, in effect, "I do good work when I can set tasks for myself." Domino anticipated that instructors who dominated the class—who pressed for conformity—would get poorer results from high-Ai students than instruc- tors who pressed for independent work. A natural- istic review of student grades confirmed this. And it confirmed the reverse relation for high-Ac stu- dents, the ones who say "I do well in meeting requirements others set for me." Domino (1971) added to the evidence with a manipulative experi- ment. He assembled four classes, filling two with High Ai's and two with High Ac's. The same instructor taught introductory psychology to all four sections, pressing for conformity in two and for independence in the other two. Outcomes were indeed better when the student's style of learning matched the instructor's press (Table 1). All but one of Domino's dependent variables showed the ATI he had predicted. (The exceptional variable was a measure of originality. On this, the inde- pendent students had an advantage no matter how the class was operated.) Others have found rather similar results (Dowaliby & Schumer, 1973; Mc- Keachie, Isaacson, & Milholland, 1964, Sec. VI- A-3). And, according to a personal communica- tion, Goldberg's (1972) large experiment turned up such an interaction in only one of two courses. A quite different hunch about instructor-student match was pursued by Majasan (1972). He sus- pected that an instructor communicates better to students whose beliefs on key matters concur with his. In introductory psychology, beliefs about the intellectual character of psychology would be perti- nent. Majasan developed a short bipolar scale, each item offering a "behavioristic" (B) and a
LU § O 0
B
A P T I T U D E
Figure 1. Regressions within two treatments.
"humanistic" (H) alternative. For example: The central focus of the study of human behavior should be: (a) the specific principles that apply to unique individuals. (H) (b) the general principles that apply to all individuals. (B) (a) People's observable actions capable of ob- jective interpretation should be the primary concern of psychology. (B) (b) Psychologists should be primarily concerned with the subjective experiences underlying people's actions. (H) The instructor and the students filled out the scale at the start of the course. Majasan predicted that students who responded much like the in- structor would do best. The criterion was the stu- dent's total score over all the course examinations; these examinations were usually assembled from multiple-choice items provided by the textbook publisher.
TABLE 1
Outcomes of a Psychology Course under Four Combinations of Personality and Treatment
Mean outcome (rescaled)^8 Student pattern
Independent (High Ai, Low Ac)
Conforming (Low Ai, High Ac)
Instructor press
Independence Conformity Independence Conformity
Exam
98 87 78 100
Course grade
100 83 66 89
Originality of thought
99 100 65 59
Student satisfaction
100 88 82 94
Note. Data are from Domino (1971).
- The value of 100 was assigned to the average score of the highest ranking group, and other averages were scaled proportionately. Here, Exam combines a factual multiple-choice test and a quality score for the final essay exam that Domino (1971) reported separately. Likewise, two satis- faction measures have been pooled.
AMERICAN PSYCHOLOGIST • FEBRUARY 1975 • 117
istics and the college student's need for power,
need for affiliation, etc. But results were strangely
inconsistent from year to year and from course to
course. Some effects were significantly moderated
by sex or ability of the student. Insofar as a gen-
eralization can be glimpsed through the tangle of
evidence, it is this: The constructively motivated
student (who seeks challenges and takes responsi-
bility) is at his best when an instructor challenges
him and then leaves him to pursue his own thoughts
and projects. In contrast, the defensive student
tends to profit when the instructor lays out the
work in detail.
As for abilities, the interactions did not turn out
as we had anticipated. If we are to have warrant
for instructing some students one way and some
another, regression slopes must differ from treat-
ment to treatment. When Goldine Gleser and I
first came to realize this, we forecast that tradi-
tional scholastic aptitude would not be a source
of interactions (Cronbach & Gleser, 1957, pp. 125-
127). Just because it is general, we expected it to
have the same predictive validity for almost any
kind of instruction. Only specialized aptitude mea-
sures could be expected to forecast differential
success, we thought.
The special-ability hypothesis got off on the
wrong foot. Between 1960 and 1970, many of us
searched fruitlessly for interactions of abilities in
the Thurstone or Guilford systems. One hypothesis
Snow and I pursued ran like this: "High spatial
ability makes for success when the instruction uses
diagrams as much as possible, and minimizes
words." No interaction of this sort was found, in
our shop (Markle, 1968) or elsewhere. With
hindsight, we can see that the low-spatial student
ought to profit from diagrams if they display rela-
tions that the high-spatial student can visualize
without help. Conversely, the diagrams can put
the high-spatial student ahead when the instruc-
tional diagram is complex and has to be trans-
formed mentally to be understood. This more
closely reasoned hypothesis has not been tested.
We expect a close scrutiny of cognitive processes to
be a profitable next phase of work on ATIs.
Contrary to our original view, conventional tests
of mental ability or level of educational develop-
ment do interact. They predict how much is
learned from most instruction of fixed duration;
but whether the regression slope is steep or shallow
depends on the instructional procedure (Cronbach
& Snow, in press, chaps. 5-11). One way to reduce
the effect of general ability is to bring in pictures
or diagrams. Another is to make lessons more
didactic, less inductive. On the whole, the regres-
sion of outcome onto general ability tends to be
relatively steep when the instruction requires the
learner to actively transform information, and it
tends to be shallow when the demands are less.
But the generalization is weak, with many studies
running counter to the trend.
Inconsistencies as Higher Order Interactions
In attempting to generalize from the literature,
Snow and I have been thwarted by the inconsistent
findings coming from roughly similar inquiries.
Successive studies employing the same treatment
variable find different outcome-on-aptitude slopes.
Some fraction of this inconsistency arises from
statistical sampling error, but the remainder is
evidence of unidentified interactions (McGuire,
When ATIs are present, a general statement
about a treatment effect is misleading because the
effect will come or go depending on the kind of
person treated. When ATIs are present, a gen-
eralization about aptitude is an uncertain basis
for prediction because the regression slope will
depend on the treatment chosen. Having said this
much in 1957, I was shortsighted not to apply the
same argument to interaction effects themselves.
An ATI result can be taken as a general conclusion
only if it is not in turn moderated by further
variables. If Aptitude X Treatment X Sex inter-
act, for example, then the Aptitude X Treatment
effect does not tell the story. Once we attend to
interactions, we enter a hall of mirrors that extends
to infinity. However far we carry our analysis—to
third order or fifth order or any other—untested
interactions of a still higher order can be envisioned.
When I say something like that, some colleague
is likely to reply: "In my experience, interaction
effects are not large." To check that out, let us
look at the magnitude of various effects in one
ecology. The last four volumes of the Journal of
Personality and Social Psychology contain 17
studies with the same design: A X B X C, persons
AMERICAN PSYCHOLOGIST • FEBRUARY 1975 • 119
TABLE 2
Magniiudes of Main Effects and Interaction Effects in 17 Social-Psychological Studies '
Cumulative percentage of components Standardized variance component
2.00-4. 1.00-1. .50-. .30-. .10-. .00-.
Largest
12 ( 7) 29 ( 20) 47 ( 33) 65 ( 53) 94 ( 93) 100 (100)
Main effect Second largest
6 ( 7) 12 ( 13) 41 ( 40) 100 (100)
Interaction effect Smallest
6 ( 0) 100 (100)
First order
19 ( 26) 35 ( 31) 54 ( 49) 100 (100)
Second order
12 ( 22) 12 ( 22) 47 ( 55) 100 (100)
Note. Components are scaled so that the two smaller main effects and the four interaction effects sum to 1.00 within any study. Each study had an A X B X C design. Cumulative proportions in parentheses take into account only those effects having a single degree of freedom.
nested within cells.^1 Each effect size can be de-
scribed by the estimated variance component. For
Table 2 I brought the studies to a common scale;
interactions, we see, were in the same range of
magnitude as the second-largest main effect. If
these main effects were worth attention, so were
the interactions.
Most branches of psychology have to be con-
cerned with higher order interactions.' According
to McGuire's (1968, 1969) reviews, response to
a persuasive communication, for example, is de-
termined by interaction among characteristics of
persuader, listener, message, and setting.
In the personality field, it is neglect of inter-
actions that has kept alive the battle between the
"situationists" and the trait theorists (Bowers,
1973). Propositions about traits are actuarial
statements, valid over, situations in the aggregate.
Thus, a total score on a religious attitude ques-
tionnaire turns out to be a fine predictor of response
in real life, when the criterion is an average over
100 kinds of relevant activity—saying grace, voic-
ing conscientious objection, etc. (Fishbein & Azjen,
(^1) The interactions were not all of the ATI type. Some were interactions of situational characteristics, and rarely there was an analysis of two characteristics of the subject. I passed over studies having "replicates" as a factor, and two multivariate analyses of variance. Where a con- tinuous characteristic of subjects had been blocked in two levels, I made a correction for continuity (Cronbach & Snow, chap. 4) to increase the component of variance for that factor and its interactions. In each analysis I assumed factors to be fixed. In standardizing, I set 100% equal to the sum of the two lesser main effects and the four interactions. The largest main effect is often little more than a demonstration of the obvious, and I therefore did not include it in the scaling unit. (The residual com- ponent, which includes person main effects and numerous interactions, was typically many times the size of the other effects combined.)
1974). The trait measure, however, has negligible
power to forecast what the high scorer is likely to
do in any one situation. The contention that be-
havior is determined by the situation alone is
equally wrong. In studies Bowers reviewed, the
Person X Situation interaction usually accounted
for more variance than the situation effect.
Mischel (1973) argued that research in per-
sonality cannot but become the study of higher
order interactions:
For example, to predict a subject's voluntary delay of gratification, one may have to know how old he is, his sex, the experimenter's sex, the particular objects for which he is waiting, the consequences of not waiting, the models to whom he was just exposed, his immediately prior ex- perience—the list gets almost endless, (p. 256)
Those seven variables can give rise to 120 inter-
actions, a number beyond the practical reach of a
direct experiment. In his work on these variables,
Mischel has directly observed only interactions of
the lower orders. It is by inference from incon-
sistencies across experiments that he makes a case
for some relations at about the fourth order
(Mischel & Moore, 1973). If reactions are so
complexly conditioned, it is not even faintly sur-
prising that we get contradictory conclusions from
experiments taking only two or three factors into
account.
The problem is as pressing for cognitive psy-
chology as for personality. Newell (1972) la-
mented the current fragmentation in the study of
information processing. He tallied 59 colonies of
investigators, each collecting data on its own nar-
row task. Because the fine structure of the task
and the person's characteristics influence outcomes,
results obtained under such disparate conditions
cannot be linked up. Newell doubted that the
usual experimental strategy of narrowing condi-
120 • FEBRUARY 1975 • AMERICAN PSYCHOLOGIST
of maple shavings, but they will be orderly. I shall
argue later in this article for close observation of
effects the hypothesis under test says nothing about.
One of the strong points of the laboratory sciences
is that investigators find out about these matters
before they "run a study." They bring the subject
into a somewhat standard condition, they fine-tune
their apparatus, and they check out details of the
stimulus—all this before any subjects are run for
the record. In educational, social, and develop-
mental experiments, I fear that the common prac-
tice is to allow one casual pilot run to make sure
the procedures can be carried out at all, and then
to move at once into a formal experiment in which
only the variables that enter the hypothesis are
observed.
The second asset of the animal experimenter is
that the system he investigates can usually be iso-
lated. Effects are rarely sensitive to what is hap-
pening outside the laboratory room. What happens
to one animal is not usually allowed to influence
the behavior of the others. But the human sub-
ject's reaction in the experiment is influenced by
his past and recent experiences elsewhere, and by
what he has heard about psychologists (Freedman,
Cohen, & Hennessy, 1973; Gergen, 1973).
Some social scientists nowadays are eager to
establish rigorous generalizations about social
policy by conducting experiments in the field. We
have already seen mammoth federal experiments on
performance contracting in education, on alternative
rules for making "negative income tax" payments,
and on alternative practices in compensatory edu-
cation. As these experiments have moved toward
completion, their advocates have become increas-
ingly pensive. Alice Rivlin, a leader among those
advocates, has just reiterated her belief that formal
social experiments are worth their cost. But she
also (Rivlin, 1973) entertains the thought that
their proper use is to compare alternative rules,
rules so formal that the winning competitor can be
embodied in an act of Congress and enforced uni-
formly over the nation. The welfare alternatives
are of this sort. Rivlin doubts that gross experi-
mental comparison can produce useful rules for
schooling, where a treatment is multifaceted, can-
not be standardized, and interacts with pupil back-
ground. Under these circumstances, the between-
school variation in practices swamps out any
generalized effect of the specified treatment vari-
able.
The Times as a Source of Interactions
Not even the income experiments establish a secure
generalization. They tell us what rule produced
the best results during the three tryout years. This
is an observation about recent history, not an en-
during conclusion. A decade from now, with
changes in the economy, in social morale, in the
family structure, and in aspirations, community
attitudes will be different. The rules tested in 1970
might compare very differently if tested again in
1985. We tend to speak of a scientific conclusion
as if it were eternal, but in every field empirical
relations change (Gergen, 1973; Schlenker, 1974).
It was once a sound generalization that DDT kills
mosquitos, until the evolutionary mechanism acted
rapidly to make mosquitos resistant to DDT.
Further along the time scale, consider star maps
for navigation. Those maps have not changed
perceptibly during recorded history. And yet the
stars are shifting in their courses, and it is only a
matter of time—a lot of time—before the map
changes beyond recognition. All that can be con-
sidered eternal is the laws. The gravitational
"constant" will change from today's value, but we
can assume that s will continue to equal £ gt^2.
In psychology, Ghiselli (1974) suggested that
even such a reliable finding as the superiority of
distributed practice over massed practice may not
remain valid from one generation to another. Simi-
larly, J. W. Atkinson (1974) pointed out that when
a substantial relation is found between personality
variables, it describes only "the modal personality
of a particular society at a particular time in his-
tory" (p. 408). He went on to say:
I believe that the early success of Lewin et al. (1944) in the study of level of aspiration can be attributed largely to the fact that their subject samples, drawn from in and around German, and later American universities in the decades prior to World War II, were homogeneously high in achievement and low in anxiety, (p. 409)
Of a piece with this observation is the recognition
that the California F Scale is obsolescent (Ghiselli,
1974; Lake, Miles, & Earle, 1973). The 25-year-
old research supporting its construct validity gives
us little warrant for interpreting scores today be-
cause with new times the items carry new implica-
tions. Perhaps the best example of all is Bronfen-
brenner's (1958) backward look at research com-
paring middle-class and lower-class parenting. Class
differences observed in the 1950s were sometimes
just the reverse of what had been observed in 1930.
Generalizations decay. At one time a conclusion
122 • FEBRUARY 1975 • AMERICAN PSYCHOLOGIST
describes the existing situation well, at a later time it accounts for rather little variance, and ultimately it is valid only as history. The half-life of an empirical proposition may be great or small. The more open a system, the shorter the half-life of relations within it are likely to be. This puts construct validation (Cronbach, 1971; Cronbach & Meehl, 1955) in a new light. Because Meehl and I were importing into psychology a rationale developed out of physical science, we spoke as if a fixed reality is to be accounted for. Events are accounted for—and predicted—by a network of propositions connecting abstract con- structs. The network is patiently revised until it gives a good account of the original data, and of new data as they come in. Propositions describing atoms and electrons have a long half-life, and the physical theorist can regard the processes in his world as steady. Rarely is a social or behavioral phenomenon isolated enough to have this steady- process property. Hence the explanations we live by will perhaps always remain partial, and distant from real events (Scriven, 1956, 19S9b), and rather short lived. The atheoretical regularities of the actuary are even more time bound. An actuarial table describing human affairs changes from science into history before it can be set in type. Our troubles do not arise because human events are in principle unlawful; man and his creations are part of the natural world. The trouble, as I see it, is that we cannot store up generalizations and constructs for ultimate assembly into a network. It is as if we needed a gross of dry cells to power an engine and could only make one a month. The energy would leak out of the first cells before we had half the battery completed. So it is with the potency of our generalizations. If the effect of a treatment changes over a few decades, that incon- sistency is an effect, a Treatment X Decade inter- action that must itself be regulated by whatever laws there be. Such interactions frustrate any would-be theorist who mixes data from several decades indiscriminately into the phenomenal pic- ture he tries to explain. The obvious example of success in coming to explanatory grips with interactions involving time is evolutionary theory in biology (Scriven, 19S9a).
Darwin considered observations on species against the background of ecologies and viewed his data in Galapagos as only the latest snapshot of an ever-
changing ecology. The positivistic'strategy of fix- ing conditions in order to reach strong generaliza-
tions (Allport, 1964, p. 550) fits with the concept that processes are steady and can be fragmented into nearly independent systems. Psychologists toward the physiological end of our investigative range probably can live with that as their principal strategy. Those of us toward the social end of the range cannot.
Interpretation in Context,
Contrasted with Generalization
Social science has been dedicated to formal testing of nomothetic propositions. Given the difficulties interactions create for social science, what might a better strategy be? This has recently been dis- cussed by Gergen (1973),^3 Glass (1972), Newell (1972), McGuire (1973), Snow (1974), and Camp- bell (in press); all of them, in one way or another, propose to break away from the preoccupation with fixed-condition experiments that seek generaliza- tions. I endorse many points of these several authors that I lack time to echo here. My only major divergence is that most of the others expect more progress toward enduring theoretical structures than I do. Advocates of "theory" mean many things. I am as prepared as anyone to endorse,the value of such model building as we see McGuire or Jack Atkinson doing. But I cannot go so far as Suppes (1974), who exhorted us that theorizing is our principal duty and that in the fullness of time our successors will erect "theoretical palaces." Suppes and I both subscribe to the position that our abstract concepts perform a great service in altering the prevailing view of man (Cronbach & Suppes, 1969, pp. 122-134; cf. Gergen, 1973, on "sensitization"). But a point of view is not a theory, capable of sharp predictions to new condi- tions. The experimental strategy dominant in psychol- ogy since 1950 has only limited ability to detect interactions. Typically, the investigator delimits the range of situations considered in his research program by fixing many aspects of the conditions under which the subject is observed. The inter- actions of any fixed aspect are thereby concealed, being pulled into the main effect or into the inter- actions of other variables. The concealed inter- action may even wipe out a real main effect of the
(^3) Note Schlenker's (1974) dissent, especially on points on which I echo Gergen.
AMERICAN PSYCHOLOGIST • FEBRUARY 1975 • 123
curred during treatment and measurement. As he goes from situation to situation, his first task is to describe and interpret the effect anew in each locale, perhaps taking into account factors unique to that locale of series of events (cf. Geertz, 1973, chap. 1, on "thick description"). As results accumulate, a person who seeks understanding will do his best to-trace how the uncontrolled factors could have caused local departures from the modal effect. That is, generalization comes late, and the exception is taken as seriously as the rule, Majasan reached a fine statistical generalization, but he had no way to go behind his pretest and posttest to learn what mediated the effect. To carry the work further^ Katharine Baker at Stan- ford has gathered dissertation data on a wide range of variables. She replicated the Majasan procedure (in a smaller number of classes), adding classroom observations, and collected information on course content and assignments. Whatever regressions appear in her several classes, it is my hope that her data will enable her to give a plausible account of the mediating events that generated them. When we give proper weight to local conditions, any generalization is a working hypothesis, not a conclusion. The personnel tester, for example, long ago discovered the hazard in generalizing about predictive validity, because test validity varies with the labor pool, the conditions of the job, and the criterion. To select salesmen using a test found valid in other firms is, indefensible in the absence of solid knowledge about how aptitudes interact with the parameters along which sales jobs vary. Hence, personnel testers are taught to collect local data before putting a selection scheme into opera- tion, and periodically thereafter. Likewise, positive results obtained with a new procedure for early education in one community warrant another com- munity trying it. But instead of trusting that those results generalize, the next community needs its own local evaluation. These examples are from applied work, but the^1 same style can be used when one's motives are pure. Mischel's (1974; Mischel & Moore, 1973) work on delay of gratification shows how gen- eralized thinking can be enriched by local observa- tion. He considered data from Trinidad or Uganda or the United States against the community back-
ground. He found out what his subjects were say- ing to themselves during the delay interval and used that to explain their scores. He varied his procedures and used the inconsistencies from ex-
periment to experiment to make the case for specific interactions, The two scientific disciplines, experimental con- trol and systematic correlation, answer formal ques- tions stated in advance. Intensive local observation goes beyond discipline to an open-eyed, open- minded appreciation of the surprises nature deposits in the investigative net. This kind of interpreta- tion is historical more than scientific. I suspect that if the psychologist were to read more widely in history, ethnology, and the centuries of human- istic writings on man and society, he would be better prepared for this part of his work.
Realizable Aspirations for Social Inquiry
Social scientists generally, and psychologists in particular, have modeled their work on physical science, aspiring to amass empirical generalizations, to restructure them into more general laws, and to weld scattered laws into coherent theory. That lofty aspiration is far from realization. A nomo- thetic theory would ideally tell us the necessary and sufficient conditions for a particular result. Supplied the situational parameters A, B, and C, a theory would forecast outcome Y with a modest margin of error. But parameters D, E, F, and so on, also influence results, and hence a prediction from A, B, and C alone cannot be strong when D, E, and F vary freely. Theorists are reminded from time to time that the person who states a principle must also state the boundary conditions that limit its application. The psychologist can describe the conditions under which his generalizations have held, or the domain of which they provide an actuarial summary. He cannot often state the boundaries defining how far they will hold (Dona- gan, 1962). How could Weber, or anyone down to 1940, have told us whether the pitch "law" would apply to sonar echoes? No one had heard such echoes until the sonar apparatus itself was invented, and there was no reason for theorists to consider in advance the variables characteristic of the sonar reverberation. The forecast of Y from A, B, and C will be valid enough, if conditions D, E, F, etc., are held con- stant in establishing and in applying the law. It will be actuarially valid, valid on the average, if it was established in a representative sample from a universe of^1 situations, as long as the universe re- mains constant. When the universe changes, we have to go beyond our actuarial rule. As Meehl
AMERICAN PSYCHOLOGIST • FEBRUARY 197S • 125
(1957) has said, when we step outside the range
of our experience, we have to use our heads.
Though enduring systematic theories about man
in society are not likely to be achieved, systematic
inquiry can realistically hope to make two contri-
butions. One reasonable aspiration is to assess
local events accurately, to improve short-run con-
trol (Glass, 1972). The other reasonable aspira-
tion is to develop explanatory concepts, concepts
that will help people use their heads.
Short-run empiricism takes soundings as one
proceeds into unfamiliar waters. Only by testing
pitch discrimination on sea echoes could Ford and
I learn that a SOO-cycle- base has no advantage.
Any evaluator is engaged in monitoring an opera-
tion in context. Though from persistent work in
many contexts he may reach an actuarial gen-
eralization of some power, this will rarely be a
basis for direct control of any single operation,
Given Domino's (1971) impressive generalization,
a college counselor might classify instructors as
pressing for conformity or individuality, and might
then advise the high-Ac student as to which course
section to enroll in. But the generalization gives
no guarantee of his individual success because it
ignores additional variables such as those of Maja-
san. Hence the Domino interaction ought not
to control irreversible assignments; rather, it
offers the student a hypothesis about choice of
section, one that he can confirm or reverse after
he is two weeks into the term. Short-run empiri-
cism is "response sensitive" (R. Atkinson & Paul-
son, 1971); one monitors responses to the treat-
ment and adjusts it, instead of prescribing a fixed
treatment on the basis of a generalization from
prior experience with other persons or in other
locales.
In order to give a wide reach to our explanations,
we make experience cumulative by abstracting from
it. The explanatory constructs that we find fruit-
ful combine into a view of man, his institutions,
and his behavior. The informed public projects
each new circumstance against that background,
and so is able to react more intelligently. As time
passes, the prevailing view will not necessarily
progress from hazy vision to crude sketch to ar-
ticulate blueprint. A general statement can be
highly accurate only if it specifies interactive effects
that it takes a large amount of data to pin down.
Some effects in the network will change in form,
in a span of one or two generations, even before
enough qualifying clauses have been added to de-
scribe the effect accurately. Though our sketch
of man may become more elaborate, it will remain
a sketch.
Social scientists are rightly proud of the disci-
pline we draw from the natural-science side of our
ancestry. Scientific discipline is what we uniquely
add to the time-honored ways of studying man.
Too narrow an identification with science, however,
has fixed our eyes upon an inappropriate goal. The
goal of our work, I have argued here, is not to
amass generalizations atop which a theoretical
tower can someday be erected (cf. Scriven, 19S9b,
p. 471). The special task of the social scientist
in each generation is to pin down the contemporary
facts. Beyond that, he shares with the humanistic
scholar and the artist in the effort to gain insight
into contemporary relationships, and to realign the
culture's view of man with present realities. To
know man as he is is no mean aspiration.
REFERENCES
Allport, G. W. Pattern and growth in personality. New York: Holt, Rinehart & Winston, 1964. Atkinson, J. W. Motivational determinants of intellective performance and cumulative achievement. In .J. W. Atkinson & J. O. Raynor (Eds.), Motivation and achieve- ment. Washington, D.C.: Winston, 1974. Atkinson, R. C. Teaching children to read using a com- puter. American Psychologist, 1974, 29, 169-178. Atkinson, R. C., & Paulson, J. A. An approach to the psychology of instruction. Psychological Bulletin, 1972, 78, 49-61. Bowers, K. S. Situationism in psychology: An analysis and a critique. Psychological Review, 1973, 80, 307-336. Bronfenbrenner, U. Socialization and social class through time and space. In E. E. Maccoby, T. M. Newcomb, & E. L. Hartley (Eds.), Readings in social psychology (3rd ed). New York: Holt, Rinehart & Winston, 19S8. Brophy, J. E., & Evertson, C. M. Low-inference observa- tional coding measures and teacher effectiveness. Austin: University of Texas, 1973. (ERIC Document Reproduc- tion Service No. ED 077879) Campbell, D. T. The social scientist as methodological servant of the experimenting society. Policy Studies Journal, 1973, 2, 72-7S. Campbell, D. T. Qualitative knowing in action research. Journal of Social Issues, in press. Cronbach, L. J. Report on a psychometric mission to Clinicia. Psychometrika, 1954, 19, 263-270. Cronbach, L. J. The two disciplines of scientific psychology. American Psychologist, 19S7, 12, 671-684. Cronbach, L. J. Test validation. In R. L. Thorndike (Ed.), Educational measurement. Washington, D.C.: American Council on Education, 1971. Cronbach, L. J., & Gleser, G. C. Psychological tests and personnel decisions. Urbana: University of Illinois Press, 19S7. Cronbach, L. J., & Meehl, P. E. Construct validation in psychological tests. Psychological Bulletin, 1955, 52, 281-302.
126 • FEBRUARY 1975 • AMERICAN PSYCHOLOGIST
