708 BOOK REVIEWS
It is worthwhile to have this book. Selz has not gotten the credit one might
expect for his extensive work in thinking. As is clear from the book, and as is
noted by the editors, one of the reasons for Selz's relative neglect is his dif-
ficult style. Furthermore, Selz goes into considerable detail on the intro-
spections of individual subjects to individual stimuli. Not only is such
material hard to work through, but as has always been the case with such
material, one may disagree with the conclusions to be drawn from it
C.P.D.
Foundations of Information Integration Theory
By Norman H. Anderson. New York: Academic Press, 1981. 423 pp.
$36.00
Norman Anderson and his students have labored for two decades to create a
new approach and a coherent body of research addressing fundamental
issues in psychology. These contributions will be presented.as a book series
beginning with this volume, and followed by a companion Methods volume,
and further volumes covering “substantive areas, including social cognition,
decision making, developmental psychology, psychophysics, and psycholin-
guistics” (p. xiv). This volume presents a conceptual introduction to infor-
mation integration theory, discussing the major concepts and assumptions,
its origins and historical development, key phenomena that this approach
has successfully explained, and its relationship to other measurement
theories in psychology.
Information integration theory has been a rather inaccessible discipline
despite the rapid and regular stream of publications it has produced. The
bibliography in Foundations lists well over 100 papers by Anderson and a
dozen or more each by key students or associates such as Kaplan, Shanteau,
Birnbaum, Oden, and Hendrick. These include numerous Psychological
Review and Psychological Bulletin papers and dozens in other APA journals, in-
dicating the very high quality of this work. Yet, outside this group of highly
committed and productive associates, information integration theory is fre-
quently known of, but rarely known. A central goal of this book series then
must be to package and export their work to a broader audience.
‘The book consists of five rather long chapters. The first chapter presents
the basic structure and assumptions of information integration theory and
describes over a dozen domains in which research has been done. Essential-
ly, for any specific task requiring some response, multiple stimuli must each
be represented as psychological values (the Valuation function), they must
be combined into a single psychological response (the Integration function),
and they must be transformed into an observable response (the Response
function). A fundamental assumption is that valuation can be represented at
a molar level as a value or level on the response dimension for each stimulus
and a weight representing the salience or importance of this stimulus in the
overall response. Valuation itself is a result of integration processes at aBOOK REVIEWS 709
lower (molecular) level. For example, word recognition depends upon
lower-level cues of letters or sounds and context. Words are then integrated
into meanings. Information integration theory assumes that molecular
levels can be ignored in studying molar integration: “functional weights and
scale values are complete and exact summaries of the molecular processing”
(p. 9). This assumption allows Anderson to bypass the details of valuation
operations that information-processing psychologists may consider critical
and to focus on the integration of molar entities (hence the book’s title) such
as personality trait labels. This assumption also supports tasks in which
molar entities are presented to subjects under the implicit logic that in
naturalistic settings these molar entities would be the intermediate-level
summaries of the processing of lower-level behavioral cues.
The heart of information-integration theory is the discovery of rules by
which the values and weights of stimuli are integrated into an overall
response. Anderson calls this “cognitive algebra” and considers these rules
functionally similar to general-purpose adding, subtracting, multiplying,
and averaging procedures. In a typical task, subjects judge the likeableness
of a person described by a set of trait labels such as intelligent, warm, and
ambitious. The integration rules are uncovered through a method called
functional measurement, which stimultaneously validates the integration
rule, response scale, scale values and weights of stimuli. Stimulus scaling
thus becomes derived from integration theory instead of being a precursor to
substantive theory development as in traditional measurement theory.
The remarkable feats of functional measurement rest on providing multi-
ple stimuli manipulated in a factorial design and discovering interpretable
patterns of responses. For example, parallelism in a plot of responses to two
stimuli imply the joint necessity of a linear response scale and an adding or
equal-weight averaging integration rule. Nonparallelism, tested by exam-
ining interaction terms in analysis of variance, is harder to interpret since it
may reflect true stimulus interaction, a different integration rule, and/or a
nonlinear response scale. Fortunately, this “chain is as strong as its weakest
link” procedure frequently has revealed parallelism. Then, from a base of
known patterns, other phenomena have been explored. In_ particular,
multiplying models (such as the desirability of lottery tickets as a product of
probability of winning and amount to win) have been studied as reflected in
alinear fan pattern of data. Averaging models have been distinguished from
adding models by studying sets of different numbers of stimuli. A moderate-
value trait label combined with a positive trait lowers the resultant overall
impression. However, the same moderate trait raises the impression from a
negative trait. This result contradicts an adding formulation (unless sup-
plementary processes are hypothesized) but is explicable within an averag-
ing model.
The central three chapters contain extended discussions of the basic phe-
nomena addressed by information-integration theory, some of which are
presented as tests of its explanatory power against competing theories, and710 BOOK REVIEWS
others represent phenomena revealed by this approach that must be further
explored. Most of the work is within the person-perception task of having
subjects judge a person (usually on likeableness) described by cues (usually
trait labels). The averaging model is compared to adding models and the set-
size effect is discussed. The assumption of meaning constancy (scale values
are independent of other cues) is contrasted against a Gestaltist position with
discussions of halo effects, primacy effects, and discounting of inconsistent
information. Additional research on memory, how initial impressions
average in with cue values, and the determinants of weight are presented,
along with suggestions for future work.
The final chapter discusses functional measurement as a measurement
theory in comparison to the measurement theories of Fechner, Thurstone,
Stevens, conjoint measurement, and so forth. Anderson argues forcefully
that these competing theories do not provide a true linear scale of measure-
ment, or they lack the broad applicability of his approach.
The impressions I received in reading this book are quite complex.
Information-integration theory has generated a large and amazingly
coherent body of research supporting Anderson's approach. Yet this is the
product of a very few people —the theory has not been successfully exported
beyond Anderson’s students and occasional outside colleagues. Nor will this
book (and book series) produce a ground swell of attention. This book is too
dry and advanced for a casual student yet lacks much of the critical detail of
the original papers. Anderson needs a popularizer to create visual and emo-
tional impact and a more palatable style, something akin to how Kelley
“translated” Heider’s attribution theory. On the other hand, this book is
clearly “the place” to go for a single discussion of information-integration
theory and the conceptual and empirical issues surrounding it. Researchers
who already know the work will buy the book as a handy summary and a
way of deepening their understanding. They will have their graduate
students read the book, and there will be a slow transmission of knowledge.
‘There is a more interesting issue — if information integration theory is so
important and carefully pursued, why has there been so little impact? Are
people blind, or is the theory not really useful? The answers I have produced
reflect more on the tactics of research than its scientific merit. Information-
integration theory seems to solve problems that people are not concerned
about. The bulk of research is on the integration of trait labels in person
perception. This task is an unintentional tactical error: it is too “soft” and
“social” for the experimental psychologists, mathematical psychologists, or
psychophysicists to appreciate, yet too “narrow” or “unrealistic” for social
psychologists. The truly interesting problems have shifted “deeper” to cogni-
tive psychological research on memory and problem solving and “broader”
to applied social psychological research with naturalistic content and direct
importance.
Information-integration theory offers its primary contribution in meth-
odology and measurement. Its arguments for general cognitive algebra pro-BOOK REVIEWS 711
cedures or separate valuation/integration/response functions have not
gained attention as a theoretical contribution. In part, this reflects the
general shrinkage of mathematical psychology and measurement as special-
ty areas. In part, information-processing psychology grew explosively dur-
ing this time and simply overshadowed Anderson’s work. Nor has the ap-
proach made a practical contribution to “solving a social problem.” The only
problems it solves are the problems it poses as experiments. This contributes
to its image as isolated, academic, and laboratory-oriented.
The criticism that information-integration theory works only with infor-
mation-integration tasks is quite important. The extreme care and precise
procedures of functional measurement create a highly favorable environ-
ment for judgment. Perhaps under more complex natural conditions the as-
sumptions of information-integration theory no longer hold. For example,
the discussion of inconsistency discounting (section 3.4.1) reveals that in-
structions establishing the degree of independence among cues alters the
weighting of inconsistent cues. An honest-gloomy-considerate target person
is evaluated with equal weights given to the three traits when told each trait
was supplied by an equally knowledgeable acquaintance of the target person
(standard Anderson instructions). However, when told that the acquain-
tances might not be equally good judges of personality (naturalistic instruc-
tions), subjects reduced the weight of the inconsistent trait. This is cavalierly
dismissed as a small effect because it is one-third the reduction when subjects
are told one trait does not belong. It illustrates, however, the problem of
turning natural events into integration tasks. Because this has not been in-
vestigated systematically, the broader relevance of information integration
theory remains moot. Anderson explicitly recognizes this problem: “Oddly
enough, the assessment of generality in the base area of person perception
by extensions to more complex and more realistic stimuli and settings has
been neglected" (p. 293).
In summary, this book is an accurate portrayal of an extensive body of
concepts and research within information integration theory. It is dry,
somewhat self-congratulatory, but clearly the single place to go to under-
stand Anderson's two decades of contribution. For those who disagree with
the theory, the impressive coherence of empirical results stands as a
challenge to reinterpret and revise in terms of more favored models. For
those who agree with the theory, the challenge is to extend it in a meaningful
way to contexts, settings, and processes of theoretical and social importance.
John S. Carroll, Loyola University of Chicago