Expert Performance: Its Structure and Acquisition

[archerships]

http://www.cebiz.org/cds/ericsson_charness.pdf

Expert Performance
Its Structure and Acquisition
K. Anders Ericsson
Department of Psychology Florida State University
Neil Charness
Department of Psychology University of Waterloo
ABSTRACT
Counter to the common belief that expert performance reflects innate abilities and capacities, recent
research in different domains of expertise has shown that expert performance is predominantly mediated by
acquired complex skills and physiological adaptations. For elite performers, supervised practice starts at
very young ages and is maintained at high daily levels for more than a decade. The effects of extended
deliberate practice are more far-reaching than is commonly believed. Performers can acquire skills that
circumvent basic limits on working memory capacity and sequential processing. Deliberate practice can
also lead to anatomical changes resulting from adaptations to intense physical activity. The study of expert
performance has important implications for our understanding of the structure and limits of human
adaptation and optimal learning.


In nearly every field of human endeavor, the performance of the best practitioners is so outstanding, so superior even to
the performance of other highly experienced individuals in the field, that most people believe a unique, qualitative
attribute, commonly called innate talent, must be invoked to account for this highest level of performance. Although these
differences in performance are by far the largest psychologists have been able to reliably measure among healthy adults,
exceptional performance has not, until recently, been extensively studied by scientists.
In the last decade, interest in outstanding and exceptional achievements and performance has increased dramatically.
Many books have been recently published on the topic of genius (for example, Gardner, 1993a ; Murray, 1989a ;
Simonton, 1984 , 1988b ; Weisberg, 1986 , 1993 ), exceptionally creative individuals ( D. B. Wallace & Gruber, 1989 ),
prodigies ( Feldman, 1986 ; A. Wallace, 1986 ), and exceptional performance and performers ( Howe, 1990 ; Radford,
1990 ; Smith, 1983 ). Of particular interest to the general public has been the remarkable ability of idiot savants or
savants, who in spite of a very low general intellectual functioning display superior performance in specific tasks and
domains, such as mental multiplication and recall of music ( Howe, 1990 ; Treffert, 1989 ). The pioneering research
comparing the performance of experts and beginners (novices) by de Groot (1946/1978) and Chase and Simon (1973) has
generated a great deal of research ( Chi, Glaser, & Farr, 1988 ; Ericsson & Smith, 1991b ). A parallel development in
computer science has sought to extract the knowledge of experts by interviews ( Hoffman, 1992 ) to build expert systems,
which are computer models that are designed to duplicate the performance of these experts and make their expertise
generally available. These efforts at artificial intelligence have been most successful in domains that have established
symbolic representations, such as mathematical calculation, chess, and music ( Barr & Feigenbaum, 1981—1982 : Cohen
& Feigenbaum, 1982 ), which incidentally are the main domains in which prodigies and savants have been able to display
clearly superior performance ( Feldman, 1980 , 1986 ). 1
The recent advances in our understanding of exceptional performance have had little impact on general theories in
psychology. The new knowledge has not fulfilled the humanistic goals of gaining insights from the lives of outstanding
people about how people might improve their lives. Maslow (1971) long ago eloquently expressed these goals:
If we want to know how fast a human being can run, then it is no use to average out the speed of a "good sample" of the
population; it is far better to collect Olympic gold medal winners and see how well they can do. If we want to know the
possibilities for spiritual growth, value growth, or moral development in human beings, then I maintain that we can learn
most by studying our moral, ethical, or saintly people....Even when "good specimens," the saints and sages and great
leaders of history, have been available for study, the temptation too often has been to consider them not human but
supernaturally endowed. (p. 7)
American Psychologist © 1994 by the American Psychological Association, Inc.
August 1994 Vol. 49, No. 8, 725-747 For personal use only--not for distribution.
The reasons for the lack of impact become clear when we consider the two most dominant approaches and their respective
goals. The human information-processing approach, or the skills approach, has attempted to explain exceptional
performance in terms of knowledge and skills acquired through experience. This approach, originally developed by
Newell and Simon (1972) , has tried to show that the basic information-processing system with its elementary information
processes and basic capacities remains intact during skill acquisition and that outstanding performance results from
incremental increases in knowledge and skill due to the extended effects of experience. By constraining the changes to
acquired knowledge and skill, this approach has been able to account for exceptional performance within existing general
theories of human cognition. According to this approach the mechanisms identified in laboratory studies of learning can
be extrapolated to account for expertise and expert performance by an incremental accumulation of knowledge and skill
over a decade of intense experience in the domain. The long duration of the necessary period of experience and the
presumed vast complexity of the accumulated knowledge has discouraged investigators from empirically studying the
acquisition of expert performance. Similarly, individual differences in expert performance, when the amount of experience
is controlled, have not been of major interest and have been typically assumed to reflect differences in the original
structure of basic processes, capacities, and abilities.
The other major approach focuses on the individual differences of exceptional performers that would allow them to
succeed in a specific domain. One of the most influential representatives of this approach is Howard Gardner, who in
1983 presented his theory of multiple intelligence in his book Frames of Mind: The Theory of Multiple Intelligences
(hereinafter referred to as Frames of Mind). Gardner (1983 , 1993a , 1993b) drew on the recent advances in biology and
brain physiology about neural mechanisms and localization of brain activity to propose an account of the achievements of
savants, prodigies, and geniuses in specific domains. He argued that exceptional performance results from a close match
between the individual's intelligence profile and the demands of the particular domain. A major concern in this approach
is the early identification and nurturing of children with high levels of the required intelligence for a specific domain.
Findings within this approach have limited implications for the lives of the vast majority of children and adults of average
abilities and talents.
In this article we propose a different approach to the study of exceptional performance and achievement, which we refer to
as the study of expert performance. Drawing on our earlier published research, we focus on reproducible, empirical
phenomena of superior performance. We will thus not seriously consider anecdotes or unique, events, including major
artistic and scientific innovations, because they cannot be repeatedly reproduced on demand and hence fall outside the
class of phenomena that can be studied by experimental methods. Our approach involves the identification of reproducible
superior performance in the everyday life of exceptional performers and the capture of this performance under laboratory
conditions. Later we show that the analysis of captured superior performance reveals that extended training alters the
cognitive and physiological processes of experts to a greater degree than is commonly believed possible. In the final
section of the article we review results from studying the lives of expert performers and identify the central role of large
amounts of focused training (deliberate practice), which we distinguish from other forms of experience in a domain. The
recent evidence for far-reaching effects of training leads us to start by reexamining the available evidence for innate talent
and specific gifts as necessary conditions for attaining the highest levels of performance in a domain.
Traditional View of the Role of Talent in Exceptional Performance
Since the emergence of civilization, philosophers have speculated about the origin of highly desirable individual
attributes, such as poetic ability, physical beauty, strength, wisdom, and skill in handiwork ( Murray, 1989b ). It was
generally believed that these attributes were gifts from the gods, and it was commonly recognized that "On the whole the
gods do not bestow more than one gift on a person" ( Murray, 1989b, p. 11 ). This view persisted in early Greek thought,
although direct divine intervention was replaced by natural causes. Ever since, there has been a bias toward attributing
high abilities to gifts rather than experience, as expressed by John Stuart Mill, there is "a common tendency among
mankind to consider all power which is not visibly the effect of practice, all skill which is not capable of being reduced to
mechanical rules, as the result of a particular gift" (quoted in Murray, 1989b, p. 12 ).
One important reason for this bias in attribution, we believe, is linked to immediate legitimatization of various activities
associated with the gifts. If the gods have bestowed a child with a special gift in a given art form, who would dare to
oppose its development, and who would not facilitate its expression so everyone could enjoy its wonderful creations? This
argument may appear strange today, but before the French Revolution the privileged status of kings and nobility and the
birthright of their children were primarily based on such claims.
The first systematic development of this argument for gaining social recognition to artists can be found in classic work on
The Lives of the Artist by Vasari ( Bull, 1987 ), originally published in 1568. This book provided the first major biography
of artists and is generally recognized as a major indirect influence on the layman's conceptions of artists even today
( Barolsky, 1991 ). Although Vasari's expressed goal was simply to provide a factual history of art, modern scholars argue
that "the Lives were partly designed to propagate ideas of the artist as someone providentially born with a vocation from
heaven, entitled to high recognition, remuneration and respect" ( Bull, 1987, Vol. 2, p. xxvi ). To support his claim,
Vasari tried to identify early signs of talent and ability in the lives of the artists he described. When facts were missing, he
is now known to have added or distorted material ( Barolsky, 1991 ). For example, Vasari dated his own first public
demonstration of high ability to the age of 9, although historians now know that he was 13 years old at that event ( Boase,
1979 ). His evaluations of specific pieces of art expressed his beliefs in divine gifts. Michelangelo's famous painting in the
Sistine Chapel, the Final Judgment, was described by Vasari as "the great example sent by God to men so that they can
perceive what can be done when intellects of the highest grade descend upon the earth" (quoted in Boase, 1979, pp. 251—
252 ). Vasari also tried to establish a link between the noble families and the families of outstanding artists by tracing the
heritage and family trees of the artists of his time to the great families of antiquity and to earlier great artists. However,
much of the reported evidence is now considered to have been invented by Vasari ( Barolsky, 1992 ). In the centuries
following Vasari, our civilization underwent major social changes leading to a greater social mobility through the
development of a skilled middle class and major progress in the accumulation of scientific knowledge. It became
increasingly clear that individuals could dramatically increase their performance though education and training, if they had
the necessary drive and motivation. Speculation on the nature of talent started to distinguish achievements due to innate
gifts from other achievements resulting from learning and training. In 1759 Edward Young published a famous book on
the origin of creative products, in which he argued that "An Original may be said to be of vegetable nature: it rises
spontaneously from the vital root of Genius; it grows, it is not made" (quoted with original italics in Murray, 1989b, p.
28 ). Hence, an important characteristic of genius and talent was the apparent absence of learning and training, and thus
talent and acquired skill became opposites ( Bate, 1989 ). A century later Galton (1869/1979) presented a comprehensive
scientific theory integrating talent and training that has continued to influence the conception of exceptional performance
among the general population.
Sir Francis Galton was the first scientist to investigate empirically the possibility that excellence in diverse fields and
domains has a common set of causes. On the basis of an analysis of eminent men in a wide range of domains and of their
relatives, Galton (1869/1979) argued that three factors had to be present: innate ability, eagerness to work, and "an
adequate power of doing a great deal of very laborious work" (p. 37). Because the importance of the last two factors–
motivation and effort–had already been recognized ( Ericsson, Krampe, & Heizmann, 1993 ), later investigators
concentrated primarily on showing that innate abilities and capacities are necessary to attain the highest levels of
performance.
Galton (1869/1979) acknowledged a necessary but not sufficient role for instruction and practice in achieving exceptional
performance. According to this view, performance increases monotonically as a function of practice toward an asymptote
representing a fixed upper bound on performance. Like Galton, contemporary researchers generally assume that training
can affect some of the components mediating performance but cannot affect others. If performance achieved after
extensive training is limited by components that cannot be modified, it is reasonable to assert that stable, genetically
determined factors determine the ultimate level of performance. If all possible changes in performance related to training
are attained after a fairly limited period of practice, this argument logically implies that individual differences in final
performance must reflect innate talents and natural abilities.
The view that talent or giftedness for a given activity is necessary to attain the highest levels of performance in that
activity is widely held among people in general. This view is particularly dominant in such domains of expertise as chess,
sports, music, and visual arts, where millions of individuals are active but only a very small number reach the highest
levels of performance.
One of the most prominent and influential scientists who draw on evidence from exceptional performance of artists,
scientists, and athletes for a biological theory of talent is Howard Gardner. In Framesof Mind, Gardner (1983) proposed
seven intelligences: linguistic, musical, spatial, logical—mathematical, bodily kinesthetic, and interpersonal and
intrapersonal intelligence–each an independent system with its own biological bases (p. 68). This theory is a refinement
and development of ideas expressed in an earlier book ( Gardner, 1973 ), in which the talent position was more explicitly
articulated, especially in the case of music. Gardner (1973) wrote,
Further evidence of the strong hereditary basis of musical talent comes from a number of sources. Most outstanding
musicians are discovered at an early age, usually before 6 and often as early as 2 or 3, even in households where relatively
little music is heard. Individual differences are tremendous among children, and training seems to have comparatively
little effect in reducing these differences. (p. 188)
He discussed possible mechanisms for talent in the context of music savants, who in spite of low intellectual functioning
display impressive music ability as children: "it seems possible that the children are reflecting a rhythmic and melodic
capacity that is primarily hereditary, and which needs as little external stimulation as does walking and talking in the
normal child" ( Gardner, 1973, p. 189 ). Although Gardner (1983) did not explicitly discuss his earlier positions, the
evidence from prodigies and savants remains central. Frames of Mind contains a careful review of the then available
research on the dramatic effects of training on performance. In particular, he reviewed the exceptional music performance
of young children trained with the Suzuki method and noted that many of these children who began training without
previous signs of musical talent attained levels comparable to music prodigies of earlier times and gained access to the
best music teachers in the world. The salient aspect of talent, according to Gardner (1983) , is no longer the innate
structure (gift) but rather the potential for achievement and the capacity to rapidly learn material relevant to one of the
intelligences. Gardner's (1983) view is consistent with Suzuki's rejection of inborn talent in music and Suzuki's
(1963/1981) early belief in individual differences in innate general ability to learn, although Suzuki's innate abilities were
not specific to a particular domain, such as music. However, in his later writings, Suzuki (1980/1981) argued that "every
child can be highly educated if he is given the proper training" (p. 233), and he blamed earlier training failures on
incorrect training methods and their inability to induce enthusiasm and motivation in the children. The clearest explication
of Gardner's (1983) view is found when he discussed his proposal for empirical assessments of individuals' profiles in
terms of the seven intelligences. He proposed a test in which "individuals were given the opportunity to learn to recognize
certain patterns [relevant to the particular domain] and were tested on their capacities to remember these from one day to
the next" (p. 385). On the basis of tests for each of the intelligences, "intellectual profiles could be drawn up in the first
year or two of life" (p. 386), although reliable assessments may have to wait until the preschool years because of "early
neural and functional plasticity" (p. 386). Gardner's own hunch about strong intellectual abilities was that "an individual
so blessed does not merely have an easy time learning new patterns; he learns them so readily that it is virtually impossible
for him to forget them" (pp. 385—386).
Our reading of Gardner's (1993a , 1993b) 2 most recent books leads us to conclude that his ideas on talent have not
fundamentally changed. According to Gardner's (1983) influential view, the evidence for the talent view is based on two
major sources of data on performance: the performance of prodigies and savants and the ability to predict future success
of individuals on the basis of early test results. Given that our knowledge about the exceptional performance of savants
and prodigies and the predictive validity of tests of basic abilities and talents have increased considerably in the past
decade, we briefly review the evidence or rather the lack of evidence for innate abilities and talent.
Performance of Prodigies and Savants
When the large collection of reports of amazing and inexplicable performance is surveyed, one finds that most of them
cannot even be firmly substantiated and can only rarely be replicated under controlled laboratory conditions. Probably the
best established phenomenon linked to talent in music is perfect pitch, or more accurately absolute pitch (AP). Only
approximately 0.01% of the general population have AP and are able to correctly name each of the 64 different tones,
whereas average musicians without AP can distinguish only approximately five or six categories of pitches when the
pitches are presented in isolation ( Takeuchi & Hulse, 1993 ). Many outstanding musicians display AP, and they first
reveal their ability in early childhood. With a few exceptions, adults appear to be unable to attain AP in spite of extended
efforts. Hence the characteristics of absolute pitch would seem to meet all of the criteria of innate talent, although there is
some controversy about how useful this ability is to the expert musicians. In a recent review of AP, Takeuchi and Hulse
(1993) concluded that the best account of the extensive and varied evidence points toward a theory that "states AP can be
acquired by anyone [italics added], but only during a limited period of development" (p. 355). They found that all
individuals with AP had started with music instruction early–nearly always before age five or six–and that several studies
had been successful in teaching AP to three- to six-year-old children. At older ages children perceive relations between
pitches, which leads to accurate relative pitch, something all skilled musicians have. "Young children prefer to process
absolute rather than the relative pitches of musical stimuli" (p. 356). Similar developmental trends from individual
features to relational attributes are found in other forms of perception during the same age period ( Takeuchi & Hulse,
1993 ). Rather than being a sign of innate talent, AP appears to be a natural consequence of appropriate instruction and of
ample opportunities to interact with a musical instrument, such as a piano, at very young ages.
Other proposed evidence for innate talent comes from studies of prodigies in music and chess who are able to attain high
levels of performance even as young children. In two influential books, Feldman (1980 , 1986) showed that acquisition of
skills in prodigies follows the same sequence of stages as in other individuals in the same domain. The primary difference
is that prodigies attain higher levels faster and at younger ages. For example, an analysis of Picasso's early drawings as a
child shows that he encountered and mastered problems in drawing in ways similar to less gifted individuals ( Pariser,
1987 ). Feldman (1986) also refuted the myth that prodigies acquire their skills irrespective of the environment. In fact, he
found evidence for the exact opposite, namely that "the more powerful and specific the gift, the more need for active,
sustained and specialized intervention" (p. 123) from skilled teachers and parents. He described the classic view of gifts,
in which parents are compelled to support their development, when he wrote, "When extreme talent shows itself it
demands nothing less than the willingness of one or both of the parents to give up almost everything else to make sure that
the talent is developed" (p. 122). A nice case in point is the child art prodigy Yani ( Ho, 1989 ), whose father gave up his
own painting career so as not to interfere with the novel style that his daughter was developing. Feldman (1980 , 1986)
argued that prodigious performance is rare because extreme talent for a specific activity in a particular child and the
necessary environmental support and instruction rarely coincide.
Contrary to common belief, most child prodigies never attain exceptional levels of performance as adults ( Barlow, 1952 ;
Feldman, 1986 ). When Scheinfeld (1939) examined the reported basis of the initial talent assessment by parents of
famous musicians, he found signs of interest in music rather than objective evidence of unusual capacity. For example,
Fritz Kreisler was "playing violin" (p. 239) with two sticks at age four, and Yehudi Menuhin had a "response to violins at
concerts" (p. 239) at the age of one and a half years. Very early start of music instruction would then lead to the
acquisition of absolute pitch. Furthermore, the vast majority of exceptional adult performers were never child prodigies,
but instead they started instruction early and increased their performance due to a sustained high level of training ( Bloom,
1985 ). The role of early instruction and maximal parental support appears to be much more important than innate talent,
and there are many examples of parents of exceptional performers who successfully designed optimal environments for
their children without any concern about innate talent (see Ericsson, Krampe, & Tech—Römer, 1993 , and Howe, 1990 ).
For example, as part of an educational experiment, Laslo and Klara Polgar ( Forbes, 1992 ) raised one of their daughters
to become the youngest international chess grand master ever–she was even younger than Bobby Fischer, who was the
youngest male achieving that exceptional level of chess-playing skill. In 1992 the three Polgar daughters were ranked first,
second, and sixth in the world among women chess players, respectively.
Although scientists and the popular press have been interested in the performance of prodigies, they have been especially
intrigued by so-called savants. Savants are individuals with a low level of general intellectual functioning who are able to
perform at high levels in some special tasks. In a few cases the parents have reported that these abilities made their
appearances suddenly, and they cited them as gifts from God ( Ericsson & Faivre, 1988 ; Feldman, 1986 ). More careful
study of the emergence of these and other cases shows that their detection may in some cases have been sudden, but the
opportunities, support, and encouragement for learning had preceded the original performance by years or even decades
( Ericsson & Faivre, 1988 ; Howe, 1990 ; Treffert, 1989 ). Subsequent laboratory studies of the performance of savants
have shown them to reflect acquired skills. For example, savants who can name the day of the week of an arbitrary date
(e.g., November 5, 1923) generate their answers using instructable methods that allow their performance to be reproduced
by a college student after a month of training (for a review see Ericsson & Faivre, 1988 ). The only ability that cannot be
reproduced after brief training concerns some savants' reputed ability to play a piece of music after a single hearing.
However, in a carefully controlled study of a music savant (J. L.), Charness, Clifton, and MacDonald (1988) showed that
reproduction of short (2- to 12-note) tonal sequences and recall of from two to four chords (4 notes each) depended on
whether the sequences or chords followed Western scale structure. Unfamiliar sequences that violated musical
conventions were poorly recalled past 6 notes. Short, familiar sequences of notes and chords were accurately recalled,
although recall dropped with length of sequence so that only 3 (of 24) 12-note familiar sequences were completely
correct. Attempts to train J. L. to learn temporally static 16-note melodies were unsuccessful. Even in the case of the
musical savant studied by Sloboda, Hermelin, and O'Connor (1985) , who was able to memorize a new piece of music,
there was a marked difference in success with a conventional versus a tonally unconventional piece. Thus, music savants,
like their normally intelligent expert counterparts, need access to stored patterns and retrieval structures to enable them to
retain long, unfamiliar musical patterns. Given that savants cannot read music–most of them are blind–they have to
acquire new music by listening, which would provide motivation and opportunities for the development of domainspecific
memory skills.
In summary, the evidence from systematic laboratory research on prodigies and savants provides no evidence for
giftedness or innate talent but shows that exceptional abilities are acquired often under optimal environmental conditions.
Prediction of Future Success Based on Innate Abilities and Talent
The importance of basic processes and capacities is central to many theorists in the human information-processing
tradition. In conceptual analogies with computers, investigators often distinguish between hardware (the physical
components of the computer) and software (computer programs and stored data). In models of human performance,
"software" corresponds to knowledge and strategies that can be readily changed as a function of training and learning, and
"hardware" refers to the basic elements that cannot be changed through training. Even theorists such as Chase and Simon
(1973) , who acknowledge that "practice is the major independent variable in the acquisition of skill" (p. 279), argue in
favor of individual differences in talent that predispose people to be successful in different domains: "Although there
clearly must be a set of specific aptitudes (e.g., aptitudes for handling spatial relations) that together comprise a talent for
chess, individual differences in such aptitudes are largely overshadowed by immense differences in chess experience" (p.
297). Bloom (1985) went through many different domains to point out some necessary qualities that are likely to be
mostly inborn, such as " motor coordination, speed of reflexes and hand—eye coordination" (p. 546). These views were
consistent with the available information at the time, such as high heritabilities for many of these characteristics. In their
review of sport psychology, Browne and Mahoney (1984) argued for the importance of fixed physiological traits for elite
performance of athletes and wrote that "there is good evidence that the limits of physiological capacity to become more
efficient with training is determined by genetics" (p. 609). They cited research reporting that percentage of muscle fibers
and aerobic capacity "are more than 90% determined by heredity for both male and female" (p. 609). However, more
recent reviews have shown that heritabilities in random samples of twins are much lower and range between zero and 40%
( Malina & Bouchard, 1991 ).
It is curious how little empirical evidence supports the talent view of expert and exceptional performance. Ever since
Galton, investigators have tried to measure individual differences in unmodifiable abilities and basic cognitive and
perceptual capacities. To minimize any influence from prior experience, they typically base their tests on simple tasks.
They measure simple reaction time and detection of sensory stimuli and present meaningless materials, such as nonsense
syllables and lists of digits, in tests of memory capacity. A recent review ( Ericsson, Krampe, & Tesch-Römer, 1993 )
showed that efforts to measure talent with objective tests for basic cognitive and perceptual motor abilities have been
remarkably unsuccessful in predicting final performance in specific domains. For example, elite athletes are able to react
much faster and make better perceptual discriminations to representative situations in their respective domains, but their
simple reaction times and perceptual acuity to simple stimuli during laboratory tests do not differ systematically from
those of other athletes or control subjects (for reviews see Regnier, Salmela, & Russell, 1993 , and Starkes & Deakin,
1985 ). Chess players' and other experts' superior memory for brief presentation of representative stimuli from their
domains compared with that of novices is eliminated when the elements of the same stimuli are presented in a randomly
arranged format ( Chase & Simon, 1973 ; see Ericsson & Smith, 1991a , for a review). The performance of elite chess
players on standard tests of spatial ability is not reliably different from control subjects ( Doll & Mayr, 1987 ). The
domain specificity of superior performance is striking and is observed in many differnt domains of expertise ( Ericsson,
Krampe, & Tesch-Römer, 1993 ).
This conclusion can be generalized with some qualifications to current tests of such general abilities as verbal and
quantitative intelligence. These tests typically measure acquired knowledge of mathematics, vocabulary, and grammar by
successful performance on items testing problem solving and comprehension. Performance during and immediately after
training is correlated with IQ, but the correlations between this type of ability test and performance in the domain many
months and years later is reduced (even after corrections for restriction of range) to such low values that Hulin, Henry, and
Noon (1990) questioned their usefulness and predictive validity. At the same time, the average IQ of expert performers,
especially in domains of expertise requiring thinking, such as chess, has been found to be higher than the average of the
normal population and corresponds roughly to that of college students. However, IQ does not reliably discriminate the
best adult performers from less accomplished adult performers in the same domain.
Even physiological and anatomical attributes can change dramatically in response to physical training. Almost everyone
recognizes that regular endurance and strength training uniformly improves aerobic endurance and strength, respectively.
As the amount and intensity or physical training is increased and maintained for long periods, far-reaching adaptations of
the body result (see Ericsson, Krampe, & Tesch-Römer, 1993 , for a review). For example, the sizes of hearts and lungs,
the flexibility of joints, and the strength of bones increase as the result of training, and the nature and extent of these
changes appear to be magnified when training overlaps with physical development during childhood and adolescence.
Furthermore, the number of capillaries supplying blood to trained muscles increases, and muscle fibers can change their
metabolic properties from fast twitch to slow twitch. With the clear exception of height, a surprisingly large number of
anatomical characteristics show specific changes and adaptations to the specific nature of extended intense training, which
we describe in more detail later in this article.
If one accepts the necessity of extended intense training for attaining expert performance–a claim that is empirically
supported later in this article–then it follows that currently available estimates of heritability of human characteristics do
not generalize to expert performance. An estimate of heritability is valid only for the range of environmental effects for
which the studied subjects have been exposed. With a few exceptions, studies of heritabilities have looked only at random
samples of subjects in the general population and have not restricted their analyses to individuals exposed to extended
training in a domain. The remaining data on exceptional and expert performers have not been able to demonstrate
systematic genetic influences. Explanations based on selective access to instruction and early training in a domain provide
as good or in some cases better accounts of familial relations of expert performers, such as the lineage of musicians in the
Bach family (see Ericsson, Krampe, & Tesch-Römer, 1993 , for a review).
In summary, we argue that the traditional assumptions of basic abilities and capacities (talent) that may remain stable in
studies of limited and short-term practice do not generalize to superior performance acquired over years and decades in a
specific domain. In addition, we will later review evidence showing that acquired skill can allow experts to circumvent
basic capacity limits of short-term memory and of the speed of basic reactions, making potential basic limits irrelevant.
Once the potential for change through practice is recognized, we believe that a search for individual differences that might
be predictive of exceptional and expert performance should refocus on the factors advocated by Charles Darwin (quoted
in Galton, 1908 ) in a letter to Galton after reading the first part of Galton's (1869/1979) book: "You have made a convert
of an opponent in one sense, for I have always maintained that excepting fools, men did not differ much in intellect, only
in zeal and hard work; I still think this is an eminently important difference" (p. 290). In commenting on Darwin's remark,
Galton (1908) agreed but argued that "character, including the aptitude for work, is heritable" (p. 291). On the basis of
their review, Ericsson, Krampe, and Tesch-Römer (1993) found that motivational factors are more likely to be the locus of
heritable influences than is innate talent. We explicate the connection between these "motivational" factors and the rate of
improving performance in a specific domain in the last section of this article.
There are two parts to the remaining portion of this article. First, we show that it is possible to study and analyze the
mechanisms that mediate expert performance. We also show that the critical mechanisms reflect complex, domain-specific
cognitive structures and skills that performers have acquired over extended periods of time. Hence, individuals do not
achieve expert performance by gradually refining and extrapolating the performance they exhibited before starting to
practice but instead by restructuring the performance and acquiring new methods and skills. In the final section, we show
that individuals improve their performance and attain an expert level, not as an automatic consequence of more experience
with an activity but rather through structured learning and effortful adaptation.
The Study of Expert Performance
The conceptions of expert performance as primarily an acquired skill versus a reflection of innate talents influence how
expert performance and expert performers are studied. When the goal is to identify critical talents and capacities,
investigators have located experts and then compared measurements of their abilities with those of control subjects on
standard laboratory tests. Tests involve simple stimuli and tasks in order to minimize any effects of previously acquired
knowledge and skill. Given the lack of success of this line of research, we advocate a different approach that identifies the
crucial aspects of experts' performance that these experts exhibit regularly at a superior level in their domain. If experts
have acquired their superior performance by extended adaptation to the specific constraints in their domains, we need to
identify representative tasks that incorporate these constraints to be able to reproduce the natural performance of experts
under controlled conditions in the laboratory. We illustrate this method of designing representative test situations with
several examples later in this section. Once the superior performance of experts can be reliably reproduced in a test
situation, this performance can then be analyzed to assess its mediating acquired mechanisms. Following Ericsson and
Smith (1991a) , we define expert performance as consistently superior performance on a specified set of representative
tasks for the domain that can be administered to any subject. The virtue of defining expert performance in this restricted
sense is that the definition both meets all the criteria of laboratory studies of performance and comes close to meeting
those for evaluating performance in many domains of expertise.
Perceived Experts Versus Consistent Expert Performance
In many domains, rules have evolved and standardized conditions, and fair methods have been designed for measuring
performance. The conditions of testing in many sports and other activities, such as typing competitions, are the same for
all participating individuals. In other domains, the criteria for expert performance cannot be easily translated into a set of
standardized tasks that captures and measures that performance. In some domains, expert performance is determined by
judges or by the results of competitive tournaments. Psychometric methods based on tournament results, most notably in
chess ( Elo, 1986 ), have successfully derived latent measures of performance on an interval scale. In the arts and
sciences, selected individuals are awarded prizes and honors by their peers, typically on the basis of significant
achievements such as published books and research articles and specific artistic performances.
Some type of metric is of course required to identify superior performance. The statistical term outlier may be a useful
heuristic for judging superior performance. Usually, if someone is performing at least two standard deviations above the
mean level in the population, that individual can be said to be performing at an expert level. In the domain of chess ( Elo,
1986 ), the term expert is defined as a range of chess ratings (2000—2199) approximately two to three standard deviations
(200 rating points) above the mean (1600 rating points) and five to six standard deviations above the mean of chess
players starting to play in chess tournaments.
In most domains it is easier to identify individuals who are socially recognized as experts than it is to specify observable
performance at which these individuals excel. The distinction between the perception of expertise and actual expert
performance becomes increasingly important as research has shown that the performance of some individuals who are
nominated as experts is not measurably superior. For example, studies have found that financial experts' stock investments
yield returns that are not consistently better than the average of the stock market, that is, financial experts' performance
does not differ from the result of essentially random selection of stocks. When successful investors are identified and their
subsequent investments are tracked, there is no evidence for sustained superiority. A large body of evidence has been
accumulated showing that experts frequently do not outperform other people in many relevant tasks in their domains of
expertise ( Camerer & Johnson, 1991 ). Experts may have much more knowledge and experience than others, yet their
performance on critical tasks may not be reliably better than that of nonexperts. In summary, researchers cannot seek out
experts and simply assume that their performance on relevant tasks is superior; they must instead demonstrate this
superior performance.
Identifying and Capturing Expert Performance
For most domains of expertise, people have at least an intuitive conception of the kind of activities at which an expert
should excel. In everyday life, however, these activities rarely have clearly defined starting and end points, nor do the
exact external conditions of a specific activity reoccur. The main challenge is thus to identify particular well-defined tasks
that frequently occur and that capture the essence of expert performance in a specific domain. It is then possible to
determine the contexts in which each task naturally occurs and to present these tasks in a controlled context to a larger
group of other experts.
De Groot's (1946/1978) research on expertise in chess is generally considered the pioneering effort to capture expert
performance. Ability in chess playing is determined by the outcomes of chess games between opponents competing in
tournaments. Each game is different and is rarely repeated exactly except for the case of moves in the opening phase of
the game. De Groot, who was himself a chess master, determined that the ability to play chess is best captured in the task
of selecting the next move for a given chess position taken from the middle of the game between two chess masters.
Consistently superior performance on this task for arbitrary chess positions logically implies a very high level of skill.
Researchers can therefore elicit experts' superiority in performing a critical task by presenting the same unfamiliar chess
position to any number of chess players and asking them to find the best next move. De Groot demonstrated that
performance on this task discriminates well between chess players at different levels of skill and thus captures the
essential phenomenon of ability to play this game.
In numerous subsequent studies, researchers have used a similar approach to study the highest levels of thinking in
accepted experts in various domains of expertise ( Chi et al., 1988 ; Ericsson & Smith, 1991b ). If expert performance
reflects extended adaptation to the demands of naturally occurring situations, it is important that researchers capture the
structure of these situations in order to elicit maximal performance from the experts. Furthermore, if the tasks designed for
research are sufficiently similar to normal situations, experts can rely on their existing skills, and no experiment-specific
changes are necessary. How similar these situations have to be to real-life situations is an empirical question. In general,
researchers should strive to define the simplest situation in which experts' superior performance can still be reliably
reproduced.
Description and Analysis of Expert Performance
The mere fact that it is possible to identify a set of representative tasks that can elicit superior performance from experts
under standardized conditions is important. It dramatically reduces the number of contextual factors that can logically be
essential for reproducing that superior performance. More important, it allows researchers to reproduce the phenomenon
of expert performance under controlled conditions and in a reliable fashion. Researchers can thus precisely describe the
tasks and stimuli and can theoretically determine which mechanisms are capable of reliably producing accurate
performance across the set of tasks. Part of the standard methodology in cognitive psychology is to analyze the possible
methods subjects could use to generate the correct response to a specific task, given their knowledge about procedures and
facts in the domain. The same methodology can be applied to tasks that capture expert performance. Because, however,
the knowledge experts may apply to a specific task is quite extensive and complex, it is virtually impossible for
nonexperts to understand an analysis of such a task. Instead of describing such a case, we illustrate the methodology and
related issues with a relatively simple skill, mental multiplication.
Mental Multiplication: An Illustration of Text Analysis
In a study of mental multiplication, the experimenter typically reads a problem to a subject: What is the result of
multiplying 24 by 36? The subject then reports the correct answer–864. It may be possible that highly experienced
subjects recognize that particular problem and retrieve the answer immediately from memory. That possibility is remote
for normal subjects, and one can surmise that they must calculate the answer by relying on their knowledge of the
multiplication table and familiar methods for complex multiplication. The most likely method is the paper-and-pencil
method taught in the schools, where 24 × 36 is broken down into 24 × 6 and 24 × 30 and the products are added together
(illustrated as Case B in Table 1 ). Often students are told to put the highest number first. By this rule, the first step in
solving 24 × 36 is to rearrange it as 36 × 24 and then to break it down as 36 × 4 and 36 × 20 (Case A). More sophisticated
subjects may recognize that 24 × 36 is equivalent to (30 — 6) × (30 + 6) and use the formula (a — b) × (a + b) = a 2 — b
2 , thus calculating 24 × 36 as 30 2 — 6 2 = 900 — 36 = 864 (Case C). Other subjects may recognize other shortcuts, such
as 24 × 36 = 2 × 12 × 3 × 12 = 6 × 12 2 = 6 × 144 (Case D). Skilled mental calculators often prefer to calculate the answer
in the reverse order, as is illustrated in Case E. Especially for more complex problems this procedure allows them to
report the first digit of the final result long before they have completed the calculation of the remaining digits. Because
most people expect that the entire answer has to be available before the first digit can be announced, the last method gives
the appearance of faster calculation speeds.
An investigator cannot determine on which of the methods in Table 1 a subject relied. However, if the subject was
instructed to think aloud (see Ericsson & Simon, 1993 , for the detailed procedure) while completing the mental
multiplication, the investigator could record in detail the mediating sequences of the subject's thoughts, as is illustrated in
the right panel of Table 1 . Although methodologically rigorous methods for encoding and evaluating think-aloud
protocols are available ( Ericsson & Simon, 1993 ), the visual match between Case B and the protocol in Table 1 is
sufficiently clear for the purposes of our illustration. Even with a less detailed record of the verbalized intermediate
products in the calculation, it is possible to reject most of the alternative methods as being inconsistent with a recorded
protocol.
Think-Aloud Protocols and Task Analysis in Research on Expert Performance
Since the demise of introspective analysis of consciousness around the turn of the century, investigators have been
reluctant to consider any type of verbal report as valid data on subject's cognitive processes. More recently investigators
have been particularly concerned that having subjects generate verbal reports changes the underlying processes. In a
recent review of more than 40 experimental studies comparing performance with and without verbalization, Ericsson and
Simon (1993) showed that the structure of cognitive processes can change if subjects are required to explain their
cognitive processes. In contrast, if subjects were asked simply to verbalize the thoughts that come to their attention (think
aloud), Ericsson and Simon found no reliable evidence that structural changes to cognitive processing occurred. Thinking
aloud appears only to require additional time for subjects to complete verbalization and therefore leads to somewhat
longer solution times in some cases.
A critical concern in applying this methodology to expert performance is how much information the thinkaloud protocols
of experts contain about the mediating cognitive processes. Obviously many forms of skilled perceptual—motor
performance are so rapid that concurrent verbalization of though would seem impossible. We later consider alternative
methodologies for such cases; but for a wide range of expert performance, think-aloud protocols have provided a rich
source of information on expert performance. In his work on chess masters, de Groot (1946/1978) instructed his subjects
to think aloud as they identified the best move for chess positions. From an analysis of the verbal reports, de Groot was
able to describe how his subjects selected their moves. First they familiarized themselves with the position and extracted
the strengths and weaknesses of its structure. Then they systematically explored the consequences of promising moves and
the opponent's likely countermoves by planning several moves ahead. From subjects' verbalizations, de Groot and
subsequent investigators ( Charness, 1981a ) have been able to represent the sequences of moves subjects explored as
search trees and to measure the amount and depth of planning for chess players at different levels of expertise (see Figure
1 ). The results of these analyses show that the amount and depth of search increase as a function of chess expertise to a
given point (the level of chess experts); thereafter, no further systematic differences were found ( Charness, 1989 ). That
the very best chess players still differ in their ability to find and selectively explore the most promising moves suggests
that the structure of their internal representation of chess positions differs.
The central importance of experts' representation of solutions is revealed by verbal reports in other domains such as
physics and medical diagnosis. When novices in physics solve a problem, they typically start with the question that asks
for, say, a velocity; then they try to recall formulas for calculating velocities and then construct step by step a sequence of
formulas by reasoning backward from the goal to the information given in the problem. In contrast, more experienced
subjects proceed by forward reasoning. As they read the description of the problem situation, an integrated representation
is generated and updated, so when they finally encounter the question in the problem text, they simply retrieve a solution
plan from memory ( Larkin, McDermott, Simon, & Simon, 1980 ). This finding suggests that experts form an immediate
representation of the problem that systematically cues their knowledge, whereas novices do not have this kind of orderly
and efficient access to their knowledge. Similarly, medical experts comprehend and integrate the information they receive
about patients to find the correct diagnosis by reasoning forward, whereas less accomplished practitioners tend to generate
plausible diagnoses that aid their search for confirming and disconfirming evidence ( Patel & Groen, 1991 ).
Experts' internal representation of the relevant information about the situation is critical to their ability to reason, to plan
out, and to evaluate consequences of possible actions. Approximately 100 years ago Binet was intrigued by some chess
players' claims that they could visualize chess positions clearly when they played chess games without a visible
chessboard (blindfold chess). Binet (1894) and subsequently Luria (1968) studied individuals with exceptional memory
abilities, who claimed to visualize as a mental image the information presented to them. These claims, if substantiated,
would imply that some individuals have a sensory-based memory akin to a photographic memory, making them
qualitatively different from the vast majority of human adults. To gain understanding of these processes and capacities,
investigators have turned to tests of perception and memory.
Immediate Memory of Perceived Situations
To study subjects' immediate perception of chess positions, de Groot (1946/1978) restricted the presentation to 2—15
seconds and then removed the chess position from view. Even after such a brief exposure, the best chess players were able
to describe the structure of the chess position and could reproduce the locations of all the chess pieces almost perfectly.
Weaker chess players' memory was much worse, and generally the amount of information chess players could recall was
found to be a function of skill. In a classic study Chase and Simon (1973) studied subjects' memory for briefly presented
chess positions and replicated de Groot's findings under controlled conditions. To the same subjects Chase and Simon also
presented chess positions with randomly rearranged chess pieces. Memory for these scrambled positions was uniformly
poor and did not differ as a function of skill. This finding has been frequently replicated and shows that the superior
memory for briefly presented chess positions in not due to any general memory ability, such as photographic memory, but
depends critically on subjects' ability to perceive meaningful patterns and relations between chess pieces. Originally Chase
and Simon proposed that experts' superior short-term memory for chess positions was due to their ability to recognize
configurations of chess pieces on the basis of their knowledge of vast numbers of specific patterns of pieces. With greater
knowledge of more complex and larger configurations of chess pieces (chunks), an expert could recall more individual
chess pieces with the same number of chunks. Hence Chase and Simon could account for very large individual differences
in memory for chess positions within the limits of the capacity of normal short-term memory (STM), which is
approximately seven chunks ( Miller, 1956 ).
The Chase—Simon theory has been very influential. It gives an elegant account of experts' superior memory only for
representative stimuli from their domain, and not even for randomly rearranged versions of the same stimuli (see Ericsson
& J. Smith, 1991a , for a summary of the various domains of expertise in which this finding has been demonstrated). At
that time Chase and Simon (1973) believed that storage of new information in long-term memory (LTM) was quite time
consuming and that memory for briefly presented information could be maintained only in STM for experts and
nonexperts alike. However, subsequent research by Chase and Ericsson (1982) on the effects of practice on a specific task
measuring the capacity of STM has shown that through extended practice (more than 200 hours), it is possible for subjects
to improve performance by more than 1,000%. These improvements are not mediated by increasingly larger chunks in
STM but reflect the acquisition of memory skills that enable subjects to store information in LTM and thereby circumvent
the capacity constraint of STM. Hence with extensive practice it is possible to attain skills that lead to qualitative, not
simply quantitative, differences in memory performance for a specific type of presented information.
From experimental analyses of their trained subjects and from a review of data on other individuals with exceptional
memory, Chase and Ericsson (1982 ; Ericsson, 1985) extracted several general findings of skilled memory that apply to
all subjects. Exceptional memory is nearly always restricted to one type of material, frequently random sequences of
digits. The convergence of acquired memory skills and alleged exceptional memory was demonstrated when the trained
subjects performed tasks given previously to "exceptional" subjects. Figure 2 (middle panel) shows a matrix that Binet
presented visually to his subjects. Below the matrix are several orders in which the same subjects were asked to recall the
numbers from the matrix that they memorized. Ericsson and Chase (1982) found that their subjects matched or surpassed
the exceptional subjects both in the speed of initial memorization and in the speed of subsequent recall. A detailed
analysis contrasting the speed for different orders of recall showed the same pattern in trained and exceptional subjects,
both of whom recalled by rows faster than by columns. Consistent with their acquired memory skill, the trained subjects
encoded each row of the matrix as a group by relying on their extensive knowledge of facts relevant to numbers. They
then associated a cue corresponding to the spatial location of each row with a retrieval structure illustrated in the top panel
of Figure 2 . To recall numbers in flexible order, subjects retrieved the relevant row using the corresponding retrieval cue
and then extracted the desired next digit or digits. The high correlation between the recall times predicted from this
method and the recall times observed for both exceptional and trained subjects imply that these groups have a similar
memory representation. When the biographical background of individuals exhibiting exceptional memory performance
was examined, Ericsson (1985 , 1988) found evidence for extended experience and practice with related memory tasks.
Hence, these exceptional individuals and the trained college students should be viewed as expert performers on these
laboratory tasks, where the same type of memory skills has been acquired during extended prior experience.
Acquired memory skill (skilled memory theory, Ericsson & Staszewski, 1989 ; and long-term working memory, Ericsson
& Kintsch, 1994 ) accounts well even for the superior memory of experts. In many types of expert performance, research
has shown that working memory is essentially unaffected by interruptions, during which the experts are forced to engage
in an unrelated activity designed to eliminate any continued storage of information in STM. After the interruption and
after a brief delay involving recall and reactivation of relevant information stored in LTM, experts can resume activity
without decrements in performance. Storage in LTM is further evidenced by experts' ability to recall relevant information
about the task even when they are unexpectedly asked for recall after the task has been completed. The amount recalled is
found to increase as a function of the level of expert performance in chess ( Charness, 1991 ).
The critical aspect of experts' working memory is not the amount of information stored per se but rather how the
information is stored and indexed in LTM. In support of this claim, several cases have been reported in which nonexperts
have been able to match the amount of domain-specific information recalled by experts, but without attaining the expert's
sophisticated representation of the information. After 50 hours of training on memory for presented chess positions, a
college student with minimal knowledge of chess was able to match the performance of chess masters ( Ericcson & Harris,
1990 ). However, an analysis of how the chess position was encoded revealed that the trained subject focused on
perceptually salient patterns in the periphery of the chessboard, whereas the chess master attended to the central aspects
critical to the selection of the next moves ( Ericsson & Harris, 1990 ). When told explicitly to memorize presented
medical information, medical students match or even surpass medical experts ( Patel & Groen, 1991 ; Schmidt &
Boshuizen, 1993 ). However, the medical experts are more able than medical students to identify and recall the important
pieces of presented information. Medical experts also encode more general clinical findings, which are sufficient for
reasoning about the case but not specific enough to recall or reconstruct the detailed facts presented about the medical
patient ( Boshuizen & Schmidt, 1992 ; Groen & Patel, 1988 ).
Experts acquire skill in memory to meet specific demands of encoding and accessibility in specific activities in a given
domain. For this reason their skill does not transfer from one domain to another. The demands for storage of intermediate
products in mental calculation differ from the demands of blindfold chess, wherein the chess master must be able not
simply to access the current position but also to plan and accurately select the best chess moves. The acquisition of
memory skill in a domain is integrated with the acquisition of skill in organizing acquired knowledge and refining of
procedures and strategies, and it allows experts to circumvent limits on working memory imposed by the limited capacity
of STM.
Perceptual—Motor Skill in Expert Performance
In many domains it is critical that experts respond not just accurately but also rapidly in dynamically changing situations.
A skilled performer needs to be able to perceive and encode the current situation as well as to select and execute an action
or a series of actions rapidly. In laboratory studies of skill acquisition, investigators have been able to demonstrate an
increase in the speed of perceptual—motor reactions as a direct function of practice. With extensive amounts of practice,
subjects are able to evoke automatically the correct reaction to familiar stimulus situations. This analysis of perceived
situations and automatically evoked responses is central to our understanding of skilled performance, yet it seems to be
insufficient to account for the speeds observed in many types of expert performance. The time it takes to respond to a
stimulus even after extensive training is often between 0.5 and 1.0 seconds, which is too slow to account for a return of a
hard tennis serve, a goalie's catching a hockey puck, and fluent motor activities in typing and music.
The standard paradigm in laboratory psychology relies on independent trials in which the occurrence of the presented
stimulus, which the subject does not control, defines the beginning of a trial. In contrast, in the perceptual environment in
everyday life, expert performance is continuous and changing, and experts must be able to recognize if and when a
particular action is required. Most important, it is possible for the expert to analyze the current situation and thereby
anticipate future events. Research on the return of a tennis serve shows that experts do not wait unitl they can see the ball
approaching them. Instead they carefully study the action of the server's racquet and are able to predict approximately
where in the service area the tennis ball will land even before the server has hit the ball. Abernethy (1991) has recently
reviewed the critical role of anticipation in expert performance in many racquet sports. Similarly, expert typists are
looking well ahead at the text they are typing in any particular instant. The difference between the text visually fixated and
the letters typed in a given instant (eye—hand span) increases with the typists' typing speed. High-speed filming of the
movements of expert typists' fingers shows that their fingers are simultaneously moved toward the relevant keys well
ahead of when they are actually struck. The largest differences in speed between expert and novice typists are found for
successive keystrokes made with fingers of different hands because the corresponding movements can overlap completely
after extended typing practice. When the typing situation is artificially changed to eliminate looking ahead at the text to be
typed, the speed advantage of expert typists is virtually eliminated ( Salthouse, 1991a ). Similar findings relating the
amount of looking ahead

Comments

[lettucethink.livejournal.com]

Thank you for posting this!

(You even prompted a new memory category!)

[crasch.livejournal.com]

You're welcome! I'm glad you found it useful.

Thanks!

[inpetto.livejournal.com]

Thanks. Great article. I printed the PDF and read it during and after dinner. I posted some thoughts and questions the article raised in my own journal.