were exposed to an enthusiastic
review of the first
edition [Corbin (1982)], which cited its summaries
of rigorous research while retaining
a conversational style and inclusion of ‘thought experiments’
_ using ‘the readers basic intuitive judgment as its
raw material’ (p. 219). In one example, Hogarth
gives the reader a false outcome to an event, later
demonstrating
how the reader became convinced
of the inevitability
of the (false) outcome. The
second edition retains these features and if anything, improves upon them.
The treatment
of judgmeent
and choice involves forecasting because choice is seen as reflecting two types of judgement:
predictive judgements
(forecasts) and evaluative judgements.
The quality
of choice depends
on the extent to which the
forecasts are accurate and the evaluations
faithfully reflect values. The forecasts of concern to
Prof. Hogarth
are subjective,
intuitive
forecasts.
His discussion
of choice is also relevant because
forecasters are frequently
called upon to predict
choices: how many consumers will choose to pay a
particular
price for a product
or service; how
many people will apply for a particular
job, or
choose an early retirement option; investor choices,
voter choices, etc.
The first three-chapters
of the book (and later,
Chapter 6) concentrate
on predictive judgements.
His review of the psychological
literature
reads
like a catalog of human judgmental
fallibilities,
including:
misattribution
of causes to random occurrences;
incorrect (causal) interpretations
of regression towards the mean, producing
overly extreme forecasts
for previously
extreme observations; failure to appropriately
incorporate
base
rate information
into forecasts: inferring the probability of a future event on the basis of how easy it
Randall J. Jones, Jr.
is
to imagine the event happening;
the lingering
Central State ~n~ve~~ity, Oklahoma, USA zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFED
impact
of whatever
initial
forecast
value was
considered, simply because it was the first considered, and adjustments
from it are insufficient;
use
Reference
of only a few cues for prediction;
use of redundant pieces of information
in a forecast as if these
Janis, Irving L., Victims of Groupthink
(Houghton
Mifflin
Company, Boston, 1972).
were independent
of other cues; and failure to
seek potentially
disconfirming
evidence. Not unRobin M. Hogarth,
Judgment and Choice: The
surprisingly,
the degree of confidence felt by peoPsychology ofDecision, Second edition (Wiley, New
ple in such forecasts is misplaced.
The reader’s
York, 1987) $24.95, 217.50, pp. 311.
resistance to accepting the validity of the research
findings is lessened through Prof. Hogarth’s inThis second edition of Judgment and Choice builds
volvement of the reader in mental exercises where
upon the success of its predecessor.
Forecasters
one finds oneself acting in just the way that rewhen forecasts are the basis of important
policy
decisions, and when forecasters are employees of
organizations
having vested interests in the results.
However, merely because the task is difficult is no
reason that the goal should be abandoned.
Perhaps it was the editors’ stance on this point
that precluded the logical sequel to the descriptive
case studies, namely an exploration
of ways to
improve the objectivity
of energy forecasts. The
authors of the country essays seem to have been
ideally suited for making such proposals,
given
their knowledge
of the country
environments
within which the forecasts are made.
The above criticism
does not di~nish
the
book’s principal achievement
of documenting
advocacy and bias in energy forecasting, which complex statistical models can easily mask. The editors were successful in attracting contributors
having intimate knowledge of forecasting practices in
the eight case-study countries and at IIASA. The
book is remarkably
well integrated
to have had
thirteen p~ti~ipating
writers, which undoubtedly
is a result of the continuing
guidance the editors
provided
to the contributors.
The editors’ two
introductory
chapters set the stage well, and their
two concluding chapters provide useful summaries
of the contributors’
main findings.
The book will be useful to a diverse audience,
including not only forecasters, but also students of
the energy industry,
of interest group behavior,
and of bureaucratic
politics. Many forecasters will
identify with the issues raised here and should be
reassured in knowing that similar dilemmas they
face are not unique, but are commonly shared by
colleagues.
136
Book reviews
search has demonstrated.
Such experiences
help
lessen dismissive responses of the ‘I’m not like
that’ sort. Prof. Hogarth explains the psychological mechanisms
that produce these forecasts, and
includes a wide range of examples including sales
forecasts, forecasts for personnel decisions, medical decisions, and stock purchases.
Chapters 4 and 5 discuss choice: choice under
certainty (Chapter 4) and risky decision making
(Chapter 5). The latter represents a change from
the previous edition which only discussed risky
decision making briefly. In both chapters, rational
choice models are first presented, and the discussion proceeds to the processes through which intuitive choices are actually reached, emphasizing
how these differ from the rational choice models.
Hogarth explains the substantial
impacts of the
context in which the choices are presented and the
language used to describe identical choices, as well
as exploring the psychological
basis of such effects.
The title of Chapter 7, ‘The Role of Memory in
may appear forbiddingly
academic,
Judgement’,
and the reader may be tempted to skip it. That
would be a pity. In essence, this chapter looks at
the difficulties encountered
when one tries to improve one’s intuitive
forecasts and choices. In
order to improve them, one must look back into
one’s memory of past forecasts and choices, and
learn from any mistakes. Unfortunately,
memory
is a slippery thing: research shows that original
memory inputs are not retrieved
from human
memories in the way that they are from computers. Instead, memories
are more like reconstructions
than anything
else, and therefore are
only approximately
the same as the original. Reconstructed
(remembered)
forecasts are found to
be prone to the ‘hindsight bias’ such that one
‘remembers’
forecasting,
in hindsight,
whatever
transpired,
leading one to feel that ‘I knew it all
the time’. Such ‘memories’ inhibit identifying flaws
in ones forecasting
strategy, yielding little incentive to improve.
Congruent with one of Hogarth’s major themes,
that subjective
forecasts frequently
derive from
causal explanatory
models, Chapter 8 looks at the
question
of how to encourage
people to make
more accurate causal models and scenarios for the
future. A sort of ‘if you can’t beat them, join
them’ approach
seems to be taken here, which
accepts the pervasiveness
of causal thinking, even
when it is inappropriate.
The version of the ‘joining ‘em’ strategy involves creativity in the generation of scenarios and causal models. Techniques
for increasing
creativity are discussed. The difficulties of implementing
such techniques
- and
implementing
creative solutions - are also addressed. Resistance by others to such innovations
is
to be expected. Hogarth’s quote from zyxwvutsrqponmlk
The Prince
about such resistance
is apt, where Machiavelli
explains the difficulties in introducing
a new order
of things, since it ‘has for enemies, all those who
have done well under the old’ yet but ‘lukewarm
defenders
in those who may do well under the
new’ (p. 163).
In Chapter 9, Hogarth looks at decision aids
for improving
intuitive forecasts and choices. He
offers procedures and practices that could be implemented
to help address limitations
in human
judgement,
including
a particularly
useful table
listing eight key points for probabilistic
thinking
and prediction.
The final chapter (Chapter 10) offers an overview of human judgement,
which summarizes and
synthesizes much of the work presented earlier. In
addition, it includes what Hogarth refers to as a
‘speculative’
section on the origins of judgmental
bias. It is not only a discussion of origins. It also
argues that these biases were, and remain, adaptive in many situations.
It therefore
raises the
question of how problematic
these judgmental
fallibilities actually are. I am afraid that some readers may conclude
that the significance
of these
judgmental
liabilities
is more limited than was
demonstrated
earlier. My own reading of the literature is that these judgmental
failings are most apt
to spell trouble when there is little opportunity
for
trial and error learning,
and ‘midcourse corrections’ are difficult or costly: in short, when it is
important
to do it right the first time. These
situations are not just rare cases involving new or
complex technologies.
Anyone
who has experienced the effects of a disastrous hiring (or promotion) decision is personally familiar with the liabilities of a poor forecast of an individual’s
performance. Or, for a more homely instance,
one’s
experience with marital disruptions
among friends
and family give distressingly
familiar examples of
the limitations
of subjective forecasts in marital
choice, and emphasizes the importance
of the issues discussed. The book concludes with a quote
from Francis Bacon. ‘We do ill to exalt the powers
Book reviews
137
tion of multiple time series modeling procedures.
There are many other additions
throughout
the
book with plenty of references
to recent work
(including some from 1986). In fact, the references
alone are a very valuable part of this book. For
example, a tool in model selection proposed by
Hannan
and Rissanen in 1982 is included in the
sections on single time series and on multiple time
series, and estimation
techniques
that have been
Bernard J. Goitein
developed since the writing of the first edition are
Bradley University, Peoria, Illinois, USA zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJ
presented.
The approach of the entire book is to present,
in a coherent fashion, topics in forecasting
that
Reference
have practical
value to forecasters
of economic
and business data. The authors begin by introducCorbin, R.M., 1982, “Judgement
and choice: A review for
ing basic theoretical concepts for ARIMA models,
forecasters and futurists” [review of R.M. Hogarth, Judgesuch as the autocorrelation
structure. The second
ment and Choice: The Psychology of Decisions], Journal of
chapter is on spectral analysis. While this chapter
Forecasting, 1, 219-221.
contains material that a forecaster should know, it
is not necessary to read this chapter before readC.W.J. Granger and Paul Newbold,
Forecasting
ing the rest of the book. Chapter Three explains
Economic Time Series, Second edition (Harcourt
how to build univariate
ARIMA
models with
Brace Jovanovich,
New York, 1986) hard cover
several examples and practical suggestions for both
$49.50/paperback
$24.95, E37.50 pp. 337.
nonseasonal
and seasonal time series. Some theoretical aspects of forecasting
and presented
in
This small book contains an enormous amount of
Chapter Four. Chapter Five is aptly called ‘Practimaterial that is presented in an extremely inforcal Methods for Univariate
Time Series Forecastmative manner. In the preface to the first edition,
ing.’ Along with the Box-Jenkins
methods, one is
the authors state that the research projects which
shown
smoothing
methods
(including
simple,
resulted in that book were motivated by the lack
Holt-Winters,
and Brown’s general exponential
of connection
in the literature between the theosmoothing) and stepwise autoregression.
Forecastretical aspects of forecasting
and the applied
ing results are compared,
and some relationships
aspects. The second editions continues
to relate
these two aspects. The theoretical
concepts
of
between the methods are shown.
In Chapter Five, regression models for forecastforecasting are presented in precise mathematical
terms which are followed by intuitive explanations
ing are introduced.
The authors discuss extensively, with examples,
the problems
of autocorof the mathematical
formulas. The authors give
related error and model misspecification.
The preclear interpretations
and simple examples that only
sentation
in Chapter Five should strongly motia person with a thorough understanding
of the
vate the reader to want to understand
the material
underlying
theory could possibly give.
in the following two chapters, which are on the
Although this book does not contain detailed
theory with extensive proofs, it does require some
theory and applications
of multiple
time series
modeling and forecasting.
Modeling with transfer
mathematical
maturity on the part of the reader.
A minimum
level would be the ability that one
functions
is not explained in enough depth for a
achieves by completing
a sequence in mathematipractitioner
to understand
how to identify models.
cal statistics
or probability.
For example,
the
On the other hand, more space is devoted to
authors use generating
functions
(which they devector ARIMA models, and several examples are
fine and explain) very heavily. Furthermore,
a
given. In addition, three tests for causal direction
reader should be comfortable with matrix notation
are explained and compared.
and have a familiarity with regression analysis.
Chapter Nine contains an informative
presentaThe authors state that the major changes for
tion on the combination
and evaluation
of forethe second edition are in the theory and applicacasts. There are some very interesting points raised
of the human mind, when we should seek out its
proper helps’. Forecasters
especially
should be
concerned
with
such
‘proper
helps’.
The
temptation
to reject the unflattering
picture of
human reasoning painted by researchers is great.
It is better to seek means to cope with the problems than to deny them, and this book marks a
significant step toward that end.