Universität Potsdam

Reinhold Kliegl, Jacqui Smith, Paul B. Baltes

Testing-the-limits, expertise, and memory in

adulthood and old age

first published in:

Human memory and cognitive capabilities : mechanisms and performances;
symposium in memoriam Hermann Ebbinghaus -1885 - Berlin Humboldt
University - 1985- / ed. by F. Klix and H. Hagendorf. - Amsterdam [u.a.] :
North Holland, 1986, S. 395-407, ISBN 0-444-70069-2

Postprint published at the Institutional Repository of the Potsdam University:

In: Postprints der Universität Potsdam
Humanwissenschaftliche Reihe ; 147
http://opus.kobv.de/ubp/volltexte/2009/3906/
http://nbn-resolving.de/urn:nbn:de:kobv:517-opus-39063

Postprints der Universität Potsdam

Humanwissenschaftliche Reihe ; 147
 395

TESTING-THE-LIMITS, EXPERTISE, AND MEMORY IN ADULTHOOD AND

OLD AGE
i
Reinhold Kliegl, Jacqui Smith, and Paul B. Baltes
Max Planck Institute for Human Development
and Education
Berlin West

This research has three interrelated foci: (i) engi-

neering and testing a cognitive model of expert memory,
(ii) the study of intellectual reserve capacity and
(iii) the use of a testing-the-limits methodology to
magnify and delineate age differences in limits of
reserve capacity. The assumption is that age differences
are magnified if studied at high levels of expertise
or task difficulty. Results from age-comparative point
training studies in expert memory are reported. Both
young and elderly subjects reached high levels of skil-
led memory, confirming the model. However, despite this
sizeable reserve capacity, when compared to IQ-eguivalent
young adults, superior elderly showed decline in upper
limits of function.

INTRODUCTION
The present series of intensive single-subject studies offers
preliminary evidence. They represent a combination of pilot
projects aimed at establishing the basis for future research
on expertise and memory in a life-span developmental framework
(Kliegl and Baltes (1984)). However, the preliminary findings
are such that they warrant some tentative conclusions. Two
general strategies associated with cognitive psychology, on the
one hand, and developmental research methodology on the other,
guide our approach.
Our central cognitive-psychological orientiation is to synthe-
size or engineer a cognitive skill in a theory-guided manner
in order to: (1) better specify the components underlying the
skill, and (2) have control over the construction of the exper-
tise and over performance at high levels. We argue for a stra-
tegy in which expertise in memory is constructed in the labo-
ratory following an a priori specified multi-component model of
memory functioning. Rather than decomposing naturally acquired
skills involved in memory and cognitive processing (Hunt (1978),
Sternberg (1977)), we propose to construct systems and levels
of performance not usually naturally available in subjects'
repertoires. This permits control over the components involved.
When subjects have been taught to be experts according to a
model, their expert-level performances are assumed to only be
possible when, and if, they apply the strategies or processes
built into their system as a function of the acquisition pro-
cess.
396

The key argument of our developmental-methodological orientia-

tion is that age differences are more easily identified at per-
formance limits, that is, in test conditions that require sub-
jects to go beyond their normal range of functioning. In the
normal range of functioning, we contend that too many alterna-
tive interpretations cloud the picture of age comparison, such
as pre-experimental practice differences between age groups,
the use of substitutive or compensatory processes, or the
operation of task-extrinsic motivational factors. As a conse-
quence, evidence on the nature and explanatory origins of devel-
opmental change is often rather uncertain or contradictory.
Therefore, we propose to search for developmental and aging
functions under conditions of high difficulty and at high, ex-
pert-like levels of performance (Baltes, Dittmann-Kohle, and
Dixon (1984), Charness (1981), Kliegl and Baltes (1984), Salt-
house , and Somberg (1982), Salthouse, in press). At limits of
v

performance, we expect age differences to be magnified and the

relative importance of task or ability-extraneous performance
factors to be reduced. In the ideal case, then, non-overlapping
age distributions would result at limits of performance. The
methodological strategy to be used for such a study of high
levels of performance is called testing-the-limits (Baltes et
al. (1984), Baltes, M. and Kindermann.-, (1 985) , Kliegl and Baltes
(1984), Schmidt (1971), Wiedl (1984)).
In our present research program on memory development and memory
in old age we have combined both strategies. Theory-guided cog-
nitive synthesis or cognitive engineering is used to construct
and teach expert-like memory functioning using a model that
specifies the components and processes involved in an a priori
manner (Kliegl, Smith, Heckhausen, and Baltes (1985)). After
expert levels are acquired, subjects are tested at higher and
higher levels of performance, thus approximating their limits
or maximum reserve capacity. Using this dual-pronged approach,
we expect insights both into the range of reserve capacity as
well as into the processes characteristic of development and
aging.

THE ENGINEERING OF EXPERT MEMORY

General Approach. Experts are defined as persons whose perfor-
mance on some criterion task is far superior to that of normal
persons (Charness (1981), Chi, Glaser, and Rees (1982), Glaser
(1984), Hoyer (1985)). For the most part in extant research on
expertise, experts are identified from real-life settings (e.g.,
chess, typing, physics) and studied with regard to the cogniti-
ve systems and processes associated with expert functioning.
This has been true also for most work on expert memory (Erics-
son (1985)).
Our own approach is different. We join the smaller group of
researchers who focus on the controlled acquisition of complex
cognitive expertise under laboratory conditions (e.g., Anderson
(1981), Chase and Ericsson (1982), Salthouse and Somberg (1982)).
This approach is more tedious as it requires extensive longitu-
dinal, single-subject type work associated with the acquisition
of an expertise. At the same time, we believe that this approach
 397

permits better specification of the factors and processes invol

ved in an expertise.
We believe also that most individuals (and not only particular-
ly gifted ones) are quite capable of acquiring high levels of
expertise in select domains. There is nothing magical about
many forms of cognitive performances considered to be expertise
In many instances the necessary ingredients are but three.
First, there must be at least one explicit model of a cognitive
system that accounts for and simulates the acquisition of the
requisite knowledge and mental processes associated with the
expertise involved. Second, a given individual must have avai-
lable the normal set of cognitive resources characteristic of
most adults. Third, she/he must be sufficiently motivated to
engage in extensive practice. Certainly, individuals will dif-
fer in the "final" level of expertise achieved. Most, however,
will be successful in reaching high levels of performance.
Expert Memory in Digit and Word Span. To date, we have chosen
to focus on memory for strings of digits and words. Much in-
fluenced by the work of Chase and Ericsson (1982), we have con-
structed several explic it models and training programs for the
acquisition of skilled performance on digit and word span tasks
Each of these models is expected to be effective in principle,
although when applied, it is likely that subjects will differ
in their rate and level of acquisition as a function of the
particular method train ed.
Table 1 presents a taxonomy of basic cognitive processes and
declarative knowledge potentially involved in digit and word
span. The heuristic taxonomy presented in Table 1 organizes
skilled memory into three provinces:'working memory, long-term
memory encoding/retrieval strategies, and domains of knowledge.
Components from these provinces can be combined in various ways
to engineer expert-like performances in memory-span tasks.
Table 1
Taxonomy and Examples of Cognitive Components Potentially
Involved in Digit,Memory Tasks

Working LTM Encoding/ Domains of

Memory Retrieval Strate- Knowledge
gies
1 2 3 2 3
central executive ' ' - Method of Loci ' historical
2
knowledge
Figure Alphabet"
i
articulatory loop concrete
chaining 3
chunking^'^
visuo-spatial running times
scrach pad
Berli
Berlin land-
.2,3
marks
auditory loop
f inger loop phone numbers
398

Note
1
Primary components used by Chase and Ericsson's (1982) subjects.
2
Components used in the History-Dates Model.
3
Components used in the Digit-Noun Model.

Table 2 illustrates two concrete implementations: (1) the

History-Dates model, and (2) the Digit-Noun model. The two mo-
dels are identical in the use of the Method of Loci (see below).
They differ mainly in the type of permanent knowledge they re-
quire. In the History-Dates model, knowledge about historical
dates is acquired in terms of the last triplet of each histori-
cal unit (1945 = 9 4 5 , 1492 = 492). One thousand historical da-
tes are necessary to have a full account of all possible vari-
ations of random digit sequences (000 - 999). In the Digit-Noun
model, digit doublets are receded into concrete nouns (e.g.,
40 = RoSe) using a digit-consonant scheme known as Figure-Alpha-
bet. One hundred words are necessary to have a full account of
all digit doublets possible (00, 01 ... 9 9 ) . Thus, the two models
differ in the kind and size of the knowledge necessary to gene-
rate all possible combinations of random digit sequences. Table
2 illustrates how a random string of digits can be recoded into
a random sequence of historical events or concrete nouns, respec-
tively.
In both models, sequences of recoded digit triplets (i.e., hi-
storical events) or of recoded digit doublets (i.e., nouns) are
committed to long-term memory by means of the mnemonic technique
known as the Method of Loci (cf. Bower (197o), Spence (1984),
Volkmann (1929), Yates (1966)). Our subjects work with a list of
30 or 40 locations representing landmarks in Berlin (West) as
indicated in the middle column of Table 2. During encoding sub-
jects form funny, bizarre or dynamic images or thoughts linking
the to-be-remembered items (i.e., historical dates or concrete
nouns) with the Berlin locations. Because landmarks are mental-
ly visited in an invariant sequence during encoding, they can
serve as retrieval cues when the subject "re-visits" them at
time of recall. The Berlin landmarks, their sequence and the
Method of Loci procedure thus constitute a highly overlearned
part of permanent knowledge. Which particular historical date
or concrete noun must be remembered at which location is, of
course, trial-dependent.
The sequential chunking of digits into historical events (or
concrete nouns) and the formation of images or thought associ-
ations between these to-be-remembered items and their correspon-
ding landmark does not put a heavy burden on working-memory
subsystems such as rehearsal loop or visuospatial scratch-pad
(cf. Baddeley (1983)). Only central executive functions which
serve to integrate relevant knowledge elements are required.
Therefore, in the present models, expertise in memory span is
not constituted by increasing short-term or working memory
(as in the research of Reisberg, Rappaport, and 0'Shaughnessy
(1984)), but by invoking long-term memory encoding processes
and permanent knowledge during encoding (as proposed by Chase
 399

Table 2
Illustration of History-Dates and Digit-Noun
Models

Knowledge System Encoding/ Knowledge System

(History-Dates Model) Retrieval (Digit-Noun
Sequence Model)

Digits to be coded: Digits to be

coded:
492789945 ... 618
407800 ... 86
Historical Dates Method of Loci Digit-Noun Pairs
30 Berlin landmarks
(1) 492=1492=Columbus (1) Botanical Garden (1) 40=R S=RoSe

(2) 789=1789=French Rev (2) Museum (2) 78=C F=CoFfee

(3) 945=1945=End WW II (3) Church (3) 00=S S=SuSy

(30) 618=1618=Beginning (30) Fountain (30) 86=F SH=FiSH

3O-Year War

Note
In each model the Method of Loci is combined with one of two
knowledge systems. In the History-Dates model, using digit
triplets, 1000 dates (000-999) would be necessary to encode
all digit triplets. The Digit-Noun model, using digit doublets,
provides a match to all random sequences based on 100 nouns.

and Ericsson (1982)). Once the memory expertise has been acqui-
red, capacity limitations in working memory functioning (e.g.,
in the capacity of the central executive to integrate STM and
LTM processes) can be probed, for example, by manipulating the
rate at which items to-be-remembered are presented (e.g., re-
ducing presentation rates). Such a probe technique illustrates
one theory-guided implementation of our testing-the-limits stra-
tegy.
Do the Models Work? Preliminary data suggest that the two models
outlined here do operate as predicted. So far in our training
program all adult subjects have been quite capable of acquiring
the Method of Loci and using this mnemonic strategy to remember
long series of words. Some subjects have reached a level at
which they are able to combine this method with their acquired
knowledge system about digits to remember long digit sequences.
As expected, however, subjects differ in the rate and level of
expert memory achieved. At present, not all subjects in our
400

laboratory have been exposed to all components of the system.

To begin, our training program required participation in only
30 sessions. For some subjects this was not sufficient to achie-
ve a sufficient level of performance in the Method of Loci and
the requisite digit knowledge. There is no indication, however,
that the majority of subjects in the program would not achieve
some level of expertise if they continued to participate in
additional sessions.
Figures 1 and 2 are presented as illustrations of the perfor-
mance of subjects who acquired high performance levels relati-
vely quickly. They represent data from two young subjects with

3 Sec / Digit* 2 Sec / Digit* 1 Sec / Digit*

Figure 1
Subjects Sp's acquisition of skilled digit memory
using the History-Dates model: Number of digits,
recalled in correct positions as a function of ex-
perimental trial and presentation rate per digit.

above average Hamburg-Wechsler-IQ (Kliegl, Smith, Heckhausen,

and Baltes (1985)). At present, these subjects have achieved
(after some 40-60 sessions) levels of performance approximating
more and more those attained by so-called expert mnemonists
(Ericsson (1985)) and by the two subjects of Chase and Ericsson
(1982). The best performances known to us of Chase and Ericsson's
two subjects were digit spans of 82 and 68. Note, however, that
these levels were achieved following 264 and 286 practice ses-
sions, respectively. The best performances on digit span tasks
reported prior to the Chase and Ericsson work were around 20
digits and were typically displayed by "mental calculators" or
professional mnemonists.
One of our subjects (SP) (see Figure 1) uses the History-Dates
model with 30 locations (maximum possible span = 90 digits).
Currently, this 23-year old subject is able to recall strings
of 90 digits at presentation rates of one to two seconds per
 401

o
00
II.
a Practice Self-Paced 10 Sec / Digit 5 Sec / Digit 2 Sec / Digit

5 5 10 15 20 25 30 35 40 45 50
Z
Trial
Figure 2
Subject BB's acquisition of skilled digit memory using
the Digit-Noun model: Number of digits recalled in cor-
rect position as a function of experimental trial and
presentation rate per digit.

digit. Recall is restricted, however., to digit sequences con-

sistent with his knowledge of historical dates (currently, SP
has 100 historical dates available). The other young subject
(BB) (Figure 2 ) , age 19, use the Digit-Noun model with 40 lo-
cations, permitting him to store and retrieve potentially any
random sequence of 80 digits (40 locations x 2 digits). This
subject is currently able to recall 80 digits at a presentation
rate of five seconds. Performance at a 3-second interval is
currently being practiced.
The cognitive engineering of memory expertise works also with
older subjects. Currently, ten elderly persons (ages 67-78 years)
are participating in a long-term intensive training program. To
date, all have been taught to use the Method of Loci with 40
locations to remember strings of nouns. After 30 sessions,
eight of these ten elderly persons were able to use the Method
of Loci under self-paced conditions to recall consistently clo-
se to 40 words in correct serial position. The remaining two
elderly subjects—while not fulfilling the 90% criterion—are
able to recall correctly between 80% and 90%. Four elderly have
progressed from self-paced to faster presentation rates. Three
of them are able to recall 40-word lists at criterion (36 in
correct position of 40) when nouns are presented in fixed inter-
vals of 20 seconds. One of our older subjects, the best, is
able to recall 30 of 40 words that were presented at a rate of
four seconds per word.
Can older adults also be trained to use the History-Dates model
to encode and retrieve strings of digits? Of four elderly per-
402

sons who, so far, have participated in a training program based

on the History-Dates model, all were able to recall digit se-
quences of 60 (i.e., using 20 locations) under self-paced con-
ditions. The following is the best performance we have so far
observed in our laboratory: One elderly woman (age 69)—after
13 practice trials of the History-Dates model and 3 8 sessions
of memory training—is able to recall strings of 120 digits
(40 locations x 30 triplets) presented once at a fixed rate of
eight seconds per digit. Note that presentation of such a long
digit string takes 16 minutes.
In sum, these finding indicate that the models of expert memory
for digit- and word-span work, in principle, for all partici-
pating subjects. All subjects have the reserve capacity to reach
fairly high levels of expertise. Note again, that the expertise
acquired is not based on an unknown system of functioning.
Rather, level of expertise is based, at least primarily, on the
components specified a priori by the investigators. Unless the
subjects use the model engineered, they would not be able to
produce expert-level functioning.
We have also designed investigative strategies by which the
components involved in the models are (a) measured separately,
and (b) systematically removed from the system. For example,
presenting subjects with digit sequences incompatible with their
knowledge system (History-Dates model) and requiring that the
sequence of locations constituting the map for the Method of
Loci be scrambled are two techniques we have used to investiga-
te the susceptibility of the expert system to componential de-
construction. Although these additional findings on deconstruc-
ting the system are not part of this report, it is fair to sta-
te that the results are consistent with the interpretation offe-
red here. When components are removed, the expertise—as predic-
ted—breaks down.

AGE DIFFERENCES
The second main research strategy that characterizes our general
approach is the use of expertise as a tool to implement a tes-
ting-the-limits strategy directed toward the magnification of
developmental differences. As has been mentioned already, same-
aged subjects differ markedly in their rate and level of acqui-
sition. We also expect age differences to be enlarged and to
be robust (perhaps even irreversible) if studied at experts le-
vels and near limits of performance.
At present, our age-comparative design is incomplete, in part
because the intensity of research forced us to work with small
numbers of participants. As our general expectation for the
tasks involved is one of fairly robust aging loss at expert
levels of functioning, we decided to proceed in our pilot pro-
gram of research in a manner that favored falsification of the
expectation: (1) We selected only well-functioning and a larger
number of older adults (N = 10) than younger adults, thereby
increasing the likelihood that some older adult may outperform
the young; (2) We selected among the young not only above-aver-
age (N = 2) but also below-average subjects (N = 3 ) . The select-
ion criterion was estimated Hamburg-Wechsler IQ. Our older su-
 403

perior and younger superior adults have an IQ higher than 115.

The below average young adults have an IQ at or below 100.
Figure 3 presents relevant data for the Method of Loci (40 lo-
cations, nouns as stimuli) from all subjects who had appropria-
te data (superior old: N = 10; below average young: N = 3;
superior young: N = 1 ) . The outcome is clear. Note first that
the subjects did not differ markedly at baseline. Second, all
subjects demonstrate a substantial amount of reserve capacity.
For example, under self-paced conditions and following 30 ses-
sions of training (about one hour each) the lowest performance
in a test trial is displayed by one older subject. This elderly
person, however, was still able to recall (immediately following
presentation) 33 of 40 words in the correct order.
Third, as testing conditions for recall were made more and more
diffictult by shortening the presentation rate (from self-paced
to fixed rates of 20, 15, 10, 4, and 2 seconds), individual dif-
ferences incrased. Individual differences—not very pronounced
at baseline—are magnified at expert levels.
Fourth, none of the ten superior older adults reached the level
of performance displayed by their young superior counterpart.
The above-average young adult was able to repeat a list of 40
words correctly up to the fast presentation rate of 4 seconds
per word.

+
Pre-Training Post-Training ( 3 0 Sessions)

lUscc 4scc Scif-Puccd 2Uscc 15scc lüscc 4scc 2sec

Presentation Rate

Figure 3
Expertise in Method of Loci: Mean and range of recall
for 40 words in three subject groups of young and old
adults.
404

Fifth, the performance of above-average older adults is at

about the same average level as that for below-average young
adults. The largest share of the existing difference between
well-functioning elderly and below-average young subjects in
favor of the elderly is produced by one elderly who functions
at a higher level than her age peers.
Sixth, the digit span performance of this best older adult,
however, does not yet approach the speed of processing (8 sec.
vs. 4 sec.) displayed by our one superior young adult (SP, see
Figure 1 ) . The second young above-average subject (BB) working
with us using the Digit-Noun condition is also working at a
faster presentation rate than our best elderly person (see Fi-
gure 2 ) . Thus, at limits the two young IQ-equivalent subjects
outperform all elderly persons studied so far.
Can the robust age difference suggested by the comparison bet-
ween superior elderly and the two young adults (who are compa-
rable in IQ to the elderly) be specified further? At present,
we are conducting further experiments to identify the origin(s)
of this age difference at limits of performance. What strikes
us most is that the origin(s) may reside primarily in the for-
mation of thoughts leading to images as required by the Method
of Loci mnemonic. All subjects appear to be able to accuracy.
What seems to distinguish subjects is their ability to generate
and maintain "good" images or thoughts for encoding and retrie-
val. Good images in the Method of Loci are assumed to be unique,
vivid, and dynamic. Superior older adults like below-average
young adults seem to have more difficulty in generating such
images at a consistently fast rate. However, further work is
required to substantiate this hypothesis.

DISCUSSION AND SUMMARY

The research program described here examines two questions:
(1) the effectiveness of a cognitive engineering approach for
the acquisition of skilled (expert) memory for digit and word
span; and (2) the usefulness of a testing-the-limits approach
for the specification and magnification of subject differences
particularly as these relate to aging.
Two findings, preliminary as they are, are of particular signi-
ficance. First, it has been demonstrated that cognitive engi-
neering toward expert memory is possible for many adults inclu-
ding the elderly. The cognitive apparatus of most healthy adult
and elderly individuals seems to hold the necessary reserve ca-
pacity. Second, expertise-like models can be used for the mag-
nification of individual differences in general and of age dif-
ferences specificly.The usefulness of expertise models for the
study of subject differences is further enhanced when combined
with a testing-the-limits approach, and especially when that
approach is aimed at testing performance in conditions that re-
quire all the reserves available to a given subject (e.g.,
stress testing). In the present study, this was accomplished by
increasing presentation rate. Thus, the research technique has
successfully magnified individual differences—differences not
apparent at baseline.
 405

Aside from the study of additional subjects and extending the

practice offered, the next steps in our research program are
two-fold. First, we would like to specify more precisely the
processual nature of the components responsible for the obtained
aging effect. In general, since our engineering of maximum func-
tioning is model- or theory-guided, we expect to be in a better
position, than is true for past gerontological work, to deter-
mine for each individual the particular cognitive components
that lead to expertise but also to breakdowns in performance.
Second, by further use of the testing-the-limits methodology
we expect to gain insight not only into aspects of reserve ca-
pacity (plasticity) but also into the constraints on development.
Life-span research on intellectual development over the last
decade has opened our vista for cognitive plasticity in the
sense of not yet realized (unknown) possibilities for adult
growth in intelligence (Baltes (1984), Dixon and Baltes (1985),
Labouvie-Vief (1982)). The present work on limits—while at the
same time continuing the search for the unknown potential—of-
fers the necessary counterpart: A systematic search for con-
straints (Keil (1981)). Plasticity of and constraint on devel-
opment are the cornerstones for a comprehensive view of what
individuals can do, may be able to do, and may not be able to
do as they develop and age (Baltes et al. (1984), Lerner (1984),
Salthouse and Samberg (1982)).Of course, knowledge about the
limiting boundaries is always incomplete. There most likely are
no definite fixed limits in the absolute sense. New conditions
may emerge that add to the resources available. However, aside
from futuristic thinking, we tend to believe that information
about peak performance is better information about plasticity
and constraints than assuming everything is possible or impos-
sible in principle.

FOOTNOTES
1
The research program on Expertise and Life-span Development
of Memory is co-directed by Paul B. Baltes and Reinhold Kliegl.
Additional scientists in the research program are Jutta Heck-
hausen (research fellow) and Jacqui Smith (research fellow).
We thank especially Roger A. Dixon, Daniel P. Keating, Timothy
A. Salthouse, and Alexander von Eye for helpful discussions.
For assistence in data collection and data analysis we would
like to acknowledge W. Assmann, S. Lempert, E. Pichler, A.
Rentz, and W. Scholtysik.
2
We do not assume that processes of aging are always changes
toward decline. On the contrary, similar to Cattell and Horn's
model of fluid vs. crystallized intelligence, we distinguish
between two domains evincing distinct life-span trajectories:
the mechanics and the pragmatics of intelligence(Baltes et al.
(1984)). The content-free mechanics are assumed to show decli-
ne with aging at limits of functioning; select aspects of the
content-rich pragmatics, contrariwise/ are expected to possib-
ly show advances into old age in healthy elderly^The present
research on expert memory deals more with the mechanics than the
pragmatics. Thus, we expect decline with aging near limits of
functioning.
406

REFERENCES
Anderson, J.R. (ed.), Cognitive Skills and Their Acquisi-
tion (Erlbaum, Hillsdale, N.J., 1981).
Baddeley, A.D., Working memory, Philosophical Transactions
of the Royal Society of London, B302 (1983) 311-324.
Baltes, M.M., and Kindermann, T., Die Bedeutung der Plasti-
zität für die klinische Beurteilung des Leistungsverhaltens
im Alter, in: Bente, D., Coper, H., and Konowski, F. (eds.),
Hirnorganisches Psychosyndrom: Methoden zur Objektivierung
der therapeutischen Wirksamkeit (Springer, Heidelberg,
1985).
Baltes, P.B., Intelligenz im Alter, Spektrum der Wissen-
schaft 5 (1984) 46-60.
Baltes, P.B., Dittmann-Kohli, F., and Dixon, R.A., New
perspectives on the development of intelligence in adult-
hood: Toward a dual-process conception and a model of se-
lective optimization with compensation, in: Baltes, P.B.,
and Brim, Jr., O.G. (eds.), Life-Span Development and Be-
havior 6 (Academic Press, New York, 1984).
Bower, G.H., Analysis of a mnemonic device, American Scien-
tist 58 (197o) 496-51o.
Charness, N., Aging and skilled problem solving, Journal
of Experimental Psychology 110 (1981) 21-38.
Chase, W.G., and Ericsson, K.A., Skill and working memory,
in: Bower, G.W. (ed.), The Psychology of Learning and
Motivation 16 (Academic Press, New York, 1982).
Chi, M.T.H., Glaser, R., and Rees, E., Expertise in problem
solving, in: Sternberg, R.J. (ed.), Advances in the Psych-
ology of Human Intelligence .(Erlbaum, Hillsdale, N.J. 1982).
Dixon, R.A., and Baltes, P.B., Toward life-span research
on the functions and pragmatics of intelligence, in: Stern-
berg, R.J. and Wagner R.K. (eds.), Practical Intelligence:
Origins of Competence in the Everyday World (Cambridge Uni-
versity Press, New York, in press).
Ericsson, K.A., Memory skill, Canadian Journal of Psych-
ology 39 (1985) 188-231.
Glaser, R., Education and thinking: The role of knowledge,
American Psychologist 39 (1984) 93-104.
Hoyer, W.J., Aging and the development of expert cognition,
in: Schlechter, T.M., and Toglia, M.P. (eds.), New Direc-
tions in Cognitive Science (Ablex, Norwood, N.J., 19 85 i .
Hunt, E., Mechanics of Verbal Ability, Psychological Review
85 (1978) 109-130.
Keil, F.C., Constraints on knowledge and cognitive devel-
opment, Psychological Review 88 (1981) 197-227.
Kliegl, R., and Baltes, P.B., Cognitive reserve capacity,
expertise, and aging, Unpublished manuscript (Max Planck
Institute for Human Development and Education, Berlin West,
1984).
Kliegl, R., Smith, J., Heckhausen, J., and Baltes, P.B.,
Construction of expertise: Engineering skilled memory, Un-
published manuscript (Max Planck Institute for Human Devel-
opment and Education, Berlin West, 1985).
Labouvie-Vief, G., Dynamic development and mature autonomy:
A theoretical prologue, Human Development 25 (1982) 161-191.
 407

(19) Lerner, R.M., On the Nature of Human Plasticity (Cambridge

University Press, New York, 1984).
(20) Reisberg, D., Rappaport, I., and 0 Shaughnessy, M., Limits
1

of working memory: The digit-span, Journal of Experimental

Psychology: Learning, Memory, and Cognition 10 (1984) 203-
221 .
(21) Salthouse, T.A., Theory of Cognitive Aging (Elsevier,
Amsterdam, in press ) .
(22) Salthouse, T.A., and Somberg, B.L., Skilled performance:
Effects of adult age and experience on elementary proces-
ses, Journal of Experimental Psychology: General 111 (1982)
176-207.
(23) Schmidt, L.R., Testing the Limits im Leistungsverhalten:
Möglichkeiten und Grenzen, in: Duhm, E. (ed.), Praxis der
klinischen Psychologie 2 (Hogrefe, Göttingen, 1971).
(24) Spence, J.D., The Memory Palace of Matteo Ricci (Viking
Penguin, New York, 1984).
(25) Sternberg, R.J., Intelligence, Information Processing, and
Analogical Reasoning: The Componential Analysis of Human
Abilities (Erlbaum, Hillsdale, N.J., 1977).
(26) Volkman, L., Ars memorativa, Jahrbuch der Kunsthistorischen
Sammlungen in Wien 3 (1928) 111-200.
(27) Wiedl, K.H., Lerntests: Nur Forschungsmittel und Forschungs-
gegenstand? Zeitschrift für Entwicklungspsychologie und
Pädagogische Psychologie 16 (1984) 245-281.
(28) Yates, F.A., The Art of Memory (Routledge and Kegan Paul,
London, 1966) .