BOOK REVIEWS I57

tive, but merely to indicate where supplementation from an instructor

or additional reading is needed.
JAMES C. REED
Wayne State University
A Study of Thinking, by Jerome S. Bruner, Jacqueline J. Goodnow,
and George A. Austin. New York: Wiley, I956. Pp. xi + 330.
IN 4 Study of Thinking Jerome Bruner and his associates of the
Harvard Laboratory of Social Relations present the outcome of a
five year project on concept attainment or categorizing. Concept
formation has been the subject of numerous investigations, and one
might wonder why these Harvard researches merit publication in
book form. In a review of concept formation experiments (6), W. E.
Viancke opined that concept formation is poorly understood despite
the considerable amount of research time devoted to the topic, and
he suggests that the methods of investigation have been better
adapted to showing the results of categorizing than to revealing its
nature. The studies of Bruner and his coworkers have been directed
toward this neglected aspect: the process of concept formation. Their
approach has been chiefly descriptive. One does not find discussions
in terms of systematic positions such as those of learning or personal-
ity theories.
The general nature of categorizing is considered in the first three
chapters. Some points of this discussion follow. Understanding of
categorizing illuminates a wide range of problems in psychology, not
only in the areas of concept formation and perception; judgment,
memory, problem-solving, inventive thinking, and esthetics involve
categorizing. We base our categorizing on attributes which vary in
their preferential value and validity, but preference need not be based
on validity. For example, under pressure of time we rely upon the
most readily available cues, and these may not be the most valid.
Categories defined by the joint presence of attributes (conjunctive)
i.e. large red triangle, seem the most &dquo;natural&dquo; type of category, but
classes may be defined on an either-or basis (disjunctive), i.e. either
large figure or triangle. A review of the experimental literature is not
presented. This decision was &dquo;made much easier by the recent appear-
ance of reviews&dquo; by Viancke, Johnson, Leeper, and Humphrey.
The remaining chapters present the concept attainment experi-
ments. All subjects were students at Harvard or nearby colleges.
With the exception of studies on categories based on probabilistic
cues, investigations followed the procedure described below. The
nature of the task was carefully explained: to attain one concept of
a certain kind (conjunctive or disjunctive). The subject was shown
which attributes should be considered, and he was given examples of
possible concepts. These procedures were guided by the aim of show-
I58 EDUCATIONAL AND PSYCHOLOGICAL MEASUREMENT

ing concept attainment at its best. In some cases the experimenter

presented a predetermined series of instances, while in other studies
the subject selected instances to test. The investigators considered
it important that the decisions of the subject during concept attain-
ment be externalized. Sometimes the subject was required to guess
the concept after each instance, otherwise the subject was told to
&dquo;feel free&dquo; to state an hypothesis. The authors argue that the classical
rote learning method of Hull (4) obscures the process of concept
attainment; their method shows concept formation with the per-
ceptual-abstraction phase by-passed. E announced after each instance
whether it was positive or negative.
The behavior of subjects during concept attainment is described
in reference to ideal strategies.
An ideal strategy is basically an analytic device used as a yardstick
against which to compare the performance of human operators in
the situations we set them. It is our conception of the &dquo;pure form&dquo;
of the kind of behavior we have been observing: the way we would
set a computer to do what the subject appears to be doing (i, p. 241).
The ideal strategies are sets of rules for acquiring, retaining and
utilizing information in a series of instances. These strategies meet
three objectives with varying degrees of success. One objective is to
gain from each instance the maximum amount of information con-
cerning the concept. In concept attainment tasks the mental book-
keeping may be considerable, so an objective of a strategy is to
manage cognitive strain. The third aim is the regulation of risk;
some strategies involve the slow working out of the concept with
sure attainment after a certain number of trials, while others permit
one to gamble, so with luck the category is attained very quickly.
For an example of an ideal strategy, consider the case where the
experimenter presents a predetermined series of instances. The rules
of one of the strategies-the wholist strategy-are: (a) begin with a
positive instance in toto as the hypothesis; (b) when a positive
instance is encountered which the hypothesis did not correctly
categorize, take as your new hypothesis what the old hypothesis and
the instance have in common; (c) ignore all other instances. Of the
subjects who gave an initial hypothesis according to the rules of the
wholist strategy, half followed the wholist rules perfectly on subse-
quent choices; the rest appeared to use the wholist approach despite
occasional deviations.
The central problem of the research was to describe how various
conditions affected the use of strategies. In one study, S had reference
to a display of all possible instances, positive and negative. S was
required to arrive at the concept by selecting instances. After two
problems had been solved with the display, S was required to solve
a third concept without it. The purpose of this study was to see how
different strategies fared under increased cognitive strain. In another
 BOOK REVIEWS I59
investigation, subjects were drastically limited in the number of
instances they could test, and an increased use of risk taking strategies
was observed. When thematic material (parent and child scenes) was
substituted for geometric designs, a considerable change in strategies
was noted. With the thematic material certain hypotheses were

especially salient. The preference for testing &dquo;reasonable&dquo; hypothesis

with thematic material resulted in utilization of less effective strategies
than for abstract materials. Disjunctive concepts proved relatively
difficult for subject to handle, despite the fact that the nature of
either-or concepts was carefully explained. Effective strategies
required that S lean heavily on information from negative instances-
what the category is not. S’s frequently proved either unwilling or
unable to utilize this indirect information. Many persisted in employ-
ing inappropriate strategies based on positive instances.
Several studies are reported on categorizing with probabilistic cues.
R. E. Goodnow required subjects to assign aircraft silhouettes to one
of two classes. Examples of each type were demonstrated and the
relevant cues were pointed out. E determined that some cues would
have only partial validity, e.g., E placed silhouettes with straight
tail 67 per cent of the trials in one category and the remaining 33 per
cent in the other. That preference for cues differs from validity of
cues was apparent from the data, e.g. a number of subjects treated
the completely valid cue as if it were less than certain, and the cue
with 67 per cent validity was actually used to predict the category
80 per cent of the trials. For one group, after a number of trials E
discontinued validation (i.e. stopped reporting which category the
silhouette belonged to). Without the validation S tended to use
partially valid cues on an all-or-none basis.
A number of previously published experiments on categorizing
with event frequency as the only cue are discussed, e.g. Jarvik’s
experiment (5) where S predicted whether E would say &dquo;check&dquo; or
&dquo;plus,&dquo; or J. J. Goodnow’s study (3) in which S decided whether to
bet on the right or the left key of a gambling device. In some of these
two-choice problems S shows &dquo;event matching&dquo;: the proportion he
predicts one of the two alternatives will occur is the same as the pro-
portion of times the event occurs. The explanation offered for event
matching is that S is using the over-all relative frequency as a general
guide, and that S has a problem solving orientation. Conspicuous by
its absence is the fact that a mathematical learning model with
certain trial-by-trial effects will produce event matching (2): the
simple model requires no assumption that S is perceiving over-all
relative frequency.
An appendix on language and categories has been written by Roger
Brown for the psychologist unacquainted with psycholinguistics.
The chapter contains a discussion of units of linguistic analysis such
as phoneme and morpheme, the problem of linguistic meaning, and
I60 EDUCATIONAL AND PSYCHOLOGICAL MEASUREMENT

the relation between language and culture. Some five experiments on

psycholinguistics are discussed, one previously unpublished.
The fact that the authors present a small number of tables adds to
the readibility, but we would have appreciated some additional tables,
perhaps in an appendix. An appendix might also have included the
raw data from one or two of the studies. We would have liked to try
our hand at classifying subjects according to strategy to determine
whether the description is as objective as the authors maintain.
Some readers will find the lack of discussion of the experimental
results in terms of theoretical positions to be an important omission.
Others will look in vain for discussion of non-human categorizing.
In this reviewer’s opinion, A Study of Thinking rates high in read-
ability and stimulation. The material is well organized, with frequent
overviews and summaries; the general discussions are enlivened by
well-chosen examples. For those interested in experimenting on
concept formation, the volume offers many leads; the model of ideal
strategies should prove a useful descriptive tool, and much food for
experimental thought is provided in the general discussions.
THURLOW R. WILSON
University of New Mexico
REFERENCES
i. Bruner, J. S., Goodnow, Jacqueline J., and Austin, G. A.
A Study of Thinking. New
York: John Wiley & Sons, I956.
2. Bush, R. R. and Mosteller, F. Stochastic Models for Learning. New York: John

Wiley & Sons, I955.

3. Goodnow, Jacqueline J. "Determinants of Choice-Distribution in Two-Choice
Situations." American Journal of Psychology, LXVIII (I955), I06-II6.
4. Hull, C. L. "Quantitative Aspect of the Evolution of Concepts." Psychological
Monographs, XXVIII (I920), No. I (Whole Number I23).
5. Jarvik, M. E. "Probability Learning and a Negative Recency Effect in the Serial
Anticipation of Alternative Symbols." Journal of Experimental Psychology,
XLI (I955), 29I-297.
6. Viancke, W. E. "The Investigation of Concept Formation." Psychological Bulletin,
XLVIII (I95I), I-3I.

The Object Relations Technique by Herbert Phillipson. Glencoe, Illi-

nois, The Free Press, I955. Pp. x+ 224, $6.00.
THIs book introduces an impressive new projective method of
personality assessment by a clinical psychologist at the Tavistock
Clinic, London.
The rationale of the approach is expressed in terms of unconscious
object relations, constructs used by British psychoanalysts in psycho-
therapy. According to this approach &dquo;the personality seeks to explain
its main features in terms of the individual’s need to impose on the
persons and things of his external world those unconscious relation-
ships between himself and figures of his internal world which have
resulted from his early experience of the objects upon which he
depended for the satisfaction of his primitive needs.&dquo;