_' . : ':§’(‘. : _
' - f 1, .. f \3 '37-, .41. -.l .‘ WM“ ,_
"' ' 1

A“; . ‘ ' :2." .'ln':';.x “t; I .
.."v‘ 33?. ‘M f‘fw'C-i‘ 3a
3993.; 2.1"”? -'~’ w-Mv .3“ 1‘
” "7 r " My;

\._..

«3%

0
W1“ 3" ‘;

' \ , ‘ “W ~..g:‘ V i .

' . s ‘ - _ _ ‘ ' 3 .1 , ‘1 . ‘ 'J ' .u 4' .' ya
x ,7 9" 7.." '. . ' ,' . o ' ' , 'r « A «n.
. ,'
«'1‘

A HP»? ’
' I . ‘.‘ ’ Mb!
-. '.'.1.: V‘: .“w ‘R ' 4. - .> i" é”: .13.". -, . ';“‘.'fifr. Ego , 2. Eu? \‘z'l 1% '.
~ -...: - .-- Lu...” 5 . 7 a». .-
, ‘ "KI". M. - ( . L ., 1w ‘0 , ‘H ‘ . ~ A . ”1'32”“, ‘ W3.“ ‘9
.y ~ ‘ .. . at

vi
a ”33"
V
Q
L;

r ,
. ". . :‘3‘ " v. . .. 'J.’ >’ \ 'Iu‘-

u. ‘.l ‘1 ' ' .p . , . . .‘ . . - “5:"‘-.<“|‘;. .‘ y‘::;'-f~':n£ "1:. ' | ”M.
.. . -_ ' -; n ‘ ‘ ~ . * '. - :-;: - M. me. . ,s, :‘vf‘b'é, .;~.~:
J is)”: 1 f‘ f " -- ' ‘ ' u}: 1")",\ ’1' "-‘ $.29. “ gin-‘3“- , - fit;u‘a,"
'l.'_ K ., .. -.«'.‘~ .u
T", . .. _ .. MA A H-_ “at“. "..x.-
.‘ll ‘ - - -\ .n ,' "u‘_.I i.'-:‘ ~'
0) _AI -_‘ _. . ‘ 4.4 f.‘.’-- 31 0 ¢ 5
. l . . .

"hr: ..v -
; .. an an. .‘H:
MK”
. s ug- mm- 3' "~ g.
”3"?“ WWW“ 'I" “-“ "1» . gunk.
.\ 'v'. ViL‘f‘NE‘35'J” Gag?“ SWégW;Y:':L-:53." t
' "‘9 6' "r9333" ""Wpé'r.’n“.‘y‘§fi*fi'€:‘ ‘ ‘
- "I Ilufiv“ ' “L 'Q.‘ ‘
l ,_w - b 1, .

..v
'¢

:9. WED}

n .

. . ,

“awn“..r

'._ . "“4". L" .‘.L:o,',-; ‘-

.\-.';):.'C‘.-ss' ,‘ - - my.

\‘ t‘t.‘ . . n. ‘1'.
I“ ‘n‘i'..- 3° ‘ - ""IA
‘3 - -‘v -*

_, v.13

.1.-
‘.-

1")

f
;.~

\

‘11.
‘ 4

..;._,
5.'-"r'.J
‘ - . ‘

a

‘J

_o
-.A

“I

.v,

l
mu
un "' ‘fg '.
4‘.

i
- 9
J
~ P.
r'

. , . C;
' fl. L V ‘
1;. I, .‘.‘-. V4. 13" .~ ' ~ .\ --
1 h- achifx : ”:35; .
I x a V. Y. ‘ .
. . . N‘.‘

~‘f.

:14»:me mnemfik‘aa.

LR:

¢:.g

‘ ’VI"
I'+ A

 

mm

Date

TTTTTTI TTTTTTTTTTTTTTTTT

minihfl"‘““* a 31293 00638 9641

ann_lo “1"...‘4'.

   

I
T TTTTT TT

 

cw:
“W'thuiu‘wu b‘u“

I .. .4
:_. my qpflmf- 1!“
ll; ‘ ‘U-I val-

 

 

 

This is to certify that the
thesis entitled
Defining the Role of Automaticity in Skill
Acquisition: Interactions Between Working
Memory and Practice in Task Performance

presented by

Tracy L. Brown

has been accepted towards fulfillment
of the requirements for _ ,

Ph.D. Jegree in Psychology

7/7C6in/

T71 W7/

Major professor

July 23. 1985

 

MS U is an Affirmative Action/Equal Opportunity Institution

 

MSU

 

RETURNING MATERIALS:
Place in book drop to

 

 

 

“singles remove this checkout from
“ your record. FINES will
be charged if book is
returned after the date
stamped below.
r \ TH 08 h I 9:43
9*" 12.?» '
77:53 , 5::
‘5 (all; «3?
L5! ‘ ‘1: 1953‘: T?

 

 

 

 

 

 

DEFINING THE ROLE OF AUTOMATICITY IN SKILL ACQUISITION:
INTERACTIONS BETWEEN WORKING MEMORY AND PRACTICE IN TASK
PERFORMANCE

By

Tracy Lewis Brown

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Psychology

1985

ABSTRACT

DEFINING THE ROLE OF AUTOMATICITY IN SKILL ACQUISITION:
INTERACTIONS BETWEEN WORKING MEMORY AND PRACTICE IN TASK
PERFORMANCE

By

Tracy Lewis Brown

The existence and nature of automaticity is a hotly debated tOpic,
representing on one hand a central component in contemporary views of
skill acquisition, criticized on the other through its association with
a perspective on attention and processing resources that is falling into
disfavor. A review of the literature indicates that the problems
confronting the concept of automaticity result from the general
incoherence of current attentional theory and the failure to identify
the specific changes in underlying information processing operations
which correlate with the emergence of automatic performance. This paper
advances a formulation of automatic processes in skill acquisition based
on the assumption that task performance can be viewed as the ordered
execution of a number of task components linked together in the service
of an overall task goal. In this view, automaticity develops over
practice when task components become directly interassociated,
eliminating the need for their storage in working memory. The formation
of direct interassociations (unitization) is distinguished from another
logically possible class of mechanisms which may also mediate working
memory demands as a function of practice (optimization). To test this
framework, an experiment was conducted in which changes over practice in
a speeded sequential keypressing task were studied as a function of

concurrent task demands on working memory capacity, sequence length, and

frequency of sequence repetition. The results indicated that changes in
working memory demand over practice could plausibly be interpreted in
terms of unitization, which was manifested in the emergence of
integrated motor programs for frequently practiced sequences. There was
no evidence for the efficacy of optimization mechanisms in the task
studied. Whether such factors play an important role in tasks with less
determinate temporal structure remains an open question. It is concluded
that the concept of automaticity, if conceptually linked to specific
operational changes in the information processing system as practice
increases, should play a useful and important role in contemporary

accounts of skill acquisition.

TABLE OF CONTENTS
Introduction ....................

Chapter I: On the Status of Automaticity ..................

Chapter 2: Automaticity, Working Memory, and Skill

Chapter 3: Overview of Research Design and Rationale .

Coherence ......... . . .....

Systematicity and Completeness . ...........

Working Memory . . . _ ..........
Working Memory and Task Performance

Optimization ...............

Summary ................... . . .....

Shift Conditions . . . . . . ................

Manipulation of Sequence Length .......

Manipulation of Sequence Frequency . . . . . . .......

Analysis of Serial Position .........

Chapter 4: Review of Experimental Procedures and Design

Subjects . . . . ..................
Design . . . . . .............
Apparatus ....... . . . . .....
Procedure ...... . . . ............

Materials and Counterbalancing ......

Chapter 5: Description of Results ......... . .......

Overview of Variables and Data Analysis Procedures
Group and Practice Effects .........
Effects of Task Concurrence ........
Analyses with the NS-NDT Group

Serial Position Analyses ........

12

l7

l7

19

24

29

31

34

38

41

41

43

43

43

44

45

'49

52

52

54

55

59

64

TABLE OF CONTENTS (continued)

Effects Involving Sequence Frequency

Effects Involving Sequence Length . . . . .....
Chapter 6: Discussion and Conclusions .

Evidence of Unitization . .

Overview and Conclusions .
List of References ..................

Appendix A: Cell Means (ms) in the Overall Analysis From

T1 and T2 Including the No-Shift, Keyboard Shift, and

Sequence Shift Groups . ........

Appendix B: Cell Means (ms) in the Serial
Position Analyses for L=6 and L=4 Sequences

Appendix C: Cell Means in the T2 Analyses Including the

NS-NDT Group With the No-Shif t, Keyboard Shift, and

Sequence Shift Groups . .............

70
79
86
86
94

99

Al

A6

A10

LIST OF TABLES

Table l: Commonly Cited Features of Automatic Processes .........

Table 2: Cells Means in the Interaction of Group and
Practice of TTR, Fluency, and Keypressing Errors

Table 3: Cell Means (ms) for the Interaction of Group,
Practice and Task Concurrence in the Total Time
Analysis ....................

Table 4: Cell Means in the Interaction of Group and
Task Concurrence Contrasting the No-Shif t Group
with the NS-NDT Group at T2 for Total Time, TTR
and Keypressing Error Rates ...........

Table 5: Cell Means (ms) in the Interaction of Task
Concurrence and Serial Position in the L=6 Serial
Position Analysis ...............

Table 6: Cell Means (ms) for the Interaction of Group,
Practice, and Serial Positions for Sequences of
Length 6 ....................

Table 7: Cell Means (ms) for the Interaction of Group,
Task Concurrence, Frequency, and Serial Position in
the T2 Analysis Contrasting the No-Shif t Group with
the NS-NDT Group ...............

Table 8: Cell Means for the Interaction of Practice and
Sequence Length ...............

iv

14

56

58

61

66

67

75

82

INTRODUCTION

The concept of automaticity corresponds to a process that is widely
recognized in the phenomenology of everyday life. It refers to the
gradual emergence, with practice at a task, of performance that seems
effortless, fluent, and relatively free of concentration and conscious
attention. It connotes the liberation of attention during performance,
enabling one to concentrate on other aspects of the task at hand or to
attend to something else at the same time. It implies as well that some
form of change has occurred in how the task is handled; that some type
of learning mechanism has come into play which fuels the transition from
performance that requires effort and concentration to something that
seems easy, unconscious, and sometimes unintentional.

Despite the intuitive appeal of this sort of account as a component
of skill acquisition, the concept of automaticity has not been very well
developed in modern psychological formulations (Navon, 1985; Schneider,
1984). Although automaticity is mentioned a great deal, and in a variety
of different contexts, attempts to specify the underlying structural or
information processing changes associated with automaticity have been
rare (but include LaBerge, 1975; Logan, 1979; Anderson, 1982; Schneider,
1984). More often than not, treatments of automaticity have tended to
enumerate the features of automatic processes rather than to explain how
they got that way or why those features seem interrelated.

The purpose of the present research is to address the nature of the
transition from attended to automatic performance and the task variables
which make such a transition possible. In the first chapter I will

1

2

examine the strengths and shortcomings of existing conceptualizations of
automaticity with respect to their coherence, systematicity, and
completeness. The second chapter will build on the strengths identified
previously toward a general conceptual framework in which one form of
automaticity, called unitization, can be carefully defined. The third
and fourth chapters delineate, respectively, the rationale and method
for an experiment to test and refine the conceptualization. In chapters
five and six the results of the experiment are reviewed and discussed
with particular reference to the evidence for unitization and its

implications for the areas of attention and skill acquisition.

CHAPTER 1

22 the Status of Automaticity

 

While the concept of automaticity occupies a central position in
many contemporary views on skill acquisition, its status as a useful and
viable theoretical construct remains controversial. In this chapter I
review the empirical and theoretical background in which the issues
concerning automaticity are defined, concluding that the incoherence of
current attentional theorizing and the relative lack of explicit
formulation of the concept of automaticity lie at the heart of the
current ambivalence regarding its utility and viability. In addition,
some ideas are developed which may help to integrate, to some extent,
the diversity of views regarding attention as it concerns the concept of

automaticity.

Coherence.

The problem of coherence can be addressed by reference to the
diversity of views regarding attention and to the linkage between
attention and automaticity. Different views of attention have a tendency
to spawn different views of automaticity. For the so-called
"intraperceptual" theories of attention (Johnston & Dark, 1982; Kahneman
& Treisman, 1984; Broadbent, 1982), automaticity is seen as operating in
the perceptual domain and plays an important role in the early vs. late
selection issue (Deutsch & Deutsch, 1963; Treisman, 1964; Norman, 1968;
Treisman & Riley, 1969; Duncan, 1980; Johnston & Dark, 1982). In this

3

area, evidence concerning the automaticity or nonautomaticity of various
perceptual processes, including word and letter perception (Posner &
Snyder, 1975; Paap & Ogden, 1981; Kahneman & Treisman, 1984; Regan,
1981), perceptual selectivity (Treisman, 1964; Treisman & Geffen, 1967;
Treisman, Kahneman & Burkell, 1983), and visual search (Graboi, 1971;
Shiffrin & Schneider, 1977; Jonides & Gleitman, 1976), is taken as
informing the viability of the concept of automaticity in general.
Attention and automaticity interact somewhat differently in the area of
executive control (Carr, 1979), in which capacity demands are associated
with the coordination and execution of information processing operations
in task performance (Anderson, 1982; Logan, 1978, 1979, 1980; Shallice,
1972). In addition, the observation of capacity limitations in working
or short-term memory tasks (Miller, 1956; Waugh & Norman, 1965) has in
turn stimulated an association between attention and memory processes
(Atkinson & Shiffrin, 1968; Hasher & Zacks, 1979; Anderson, 1982). To
the extent that these domains differ in their conceptualization of
attention, so do they present diverse implications for the definition
and description of automaticity. A coherent approach toward the concept
of automaticity must address this multiplicity of contexts in which
automaticity is discussed, with the intention of establishing some
conceptual linkages between them.

Nevertheless, these varying perspectives on the problems of
attention, though they give central capacity limitations a different
locus in the information processing system and assign them different
functions, do allow the same general approach to automatic
processes-—specifically, that processes that constitute "attention" are

not involved in automatic performance. A more serious problem in terms

5

of coherence stems from the increasing popularity of multiple resource
views of attention (Allport, Antonis, & Reynolds, 1972; Navon & Gopher,
1979; Allport, 1980; Navon, 1984). Multiple resource models replace a
unitary concept of attention with the notion of an array of more
specialized resources. Thus, different tasks may require different
combinations of various resources, none of which is any more central, or
any more associated with the phenomenology of attention and
consciousness, than any other. This type of orientation has been
motivated by the finding that decrements in performance associated with
simultaneous performance of a secondary task (referred to as "task
concurrence costs" in dual-task performance) vary as a function of the
modality of two or more secondary tasks, or that two tasks equated for
difficulty produce different amounts of task concurrence costs when
paired with a third task (Allport, Antonis, & Reynolds, 1972; Kantowitz
& Knight, 1974, 1976). Findings of this sort indicate that at least part
of the interference effect arises from non-central processes specific to
modalities or "channels". Proponents of a more unitary concept of
attention have adopted the notion of structural interference (Kahneman,
1973) in response to these findings. Structural interference refers to
task concurrence costs arising from structural limitations of the
information processing architecture which are held to be conceptually
separate from interference caused by overloading a unitary pool of
central resources. As Allport (1980) has noted, the concept of
structural interference handles these findings nicely but fails to
provide a principled means for distinguishing structural interference
from capacity overload.

A major problem in evaluating multiple resource models has been

6

that none of the advocates of this view have attempted to specify the
nature, number, or organization of potential resource pools (Navon &
Gopher, 1979; Allport, 1980; Hirst, 1982). Arguments for the view have
tended to take one of two forms, either aggressively criticizing more
unitary conceptualizations of attentional resources (e.g., Allport,
1980), or in emphasizing the explanatory power of their own view (Navon
5 Gopher, 1979).

In the absence of information about the nature, number, and
organization of resource pools, the multiple resources orientation
serves rather dismally as a backdrop for thinking about automaticity.
One problem is that the multiple resources concept strips attentional
theory of its traditional linkage with phenomenology and consciousness
(Navon & Gopher, 1979; Posner & Snyder, 1975; Carr, 1979). If one wants
to establish a scientific explanation for the feeling that performing a
task requires less "effort" or "concentration" than at some earlier
point in practice, then one needs some means for relating attentional
resources with consciousness. In a view with innumerable, unidentified,
and unorganized resources (what might be called the "chronic
undifferentiated multiple resources" view) this phenomenon cannot be
addressed.

A second problem stems from the indeterminance of the function and
organization of resource pools. If attention is viewed as an assembly of
resource pools whose function and organization is unknown, then how is
automaticity to be represented? Assuming that each resource pool is
itself limited in capacity (as Navon & Gopher do) one might be led to
conclude that automaticity involves nothing more than non-overlapping

resource requirements. Tasks performed simultaneously may interfere

initially because they tap the same resource at the same point in time;
interference could decrease with practice by changing the resources
needed for the tasks or by adjusting their temporal coordination. From
this point of view, multiple resource theories have room for notions of
"restructuring" or strategy change (Chang, 1985) and for notions of
attention switching or time sharing (Broadbent, I982; Damos & Wickens,
1980), but not for automaticity as typically conceived.

The only other option--in the absence of an explicit organizational
scheme for resource pools-is the position that automatic processes
utilize no resources at all. The emergence of automaticity in skill
acquisition then reduces to the transition from resource-invested to
resource-free performance. The evaluation of this idea hinges
importantly on the definition of "resources". When "resources" are
broadly conceived, in the sense that any mechanism with an identifiable
function qualifies in principle as a resource, then the idea of
resource-free performance translates into a strange form of
dualism-automatic information processing becomes magical. This is
because resources as broadly conceived would include any part of the
information processing system. When resources are more exclusively
defined as mechanisms with identifiable functions that are themselves
limited in capacity, than to advance a formulation of automaticity
necessarily entails the postulation of information processing mechanisms
that are 225_ limited in capacity. This conceptualizaton has two
noteworthy implications. One is that it presupposes the same type of
distinction that is inherent in the more unitary views of attention that
the multiple resources view is intended to replace, namely that there is

some type of general or central capacity (or capacities). Second, it

8
implies the existence of information processing systems that are
unlimited in capacity, a claim which seems inconsistent with the
apparent limitations of the human nervous system.

The diversity of theoretical possibilities for conceptualizing the
structure and organization of attentional resources can be
mind-boggling, even in the domain of unitary views of attention, and it
becomes especially confusing given the added complications of multiple
resource views (see Navon, 1984). Moreover, some of the options coming
from multiple resource theory appear to be functionally equivalent to
the more traditional notions of central capacity limitations and
structural interference (Kahneman, 1973). Regardless of how one may
construe these possibilities, it is clear that the conceptual background
for automaticity provided by attention theory seems far from coherent;
confronted on the one hand by a lack of consensus regarding the domain
of phenomena to be addressed by the terms "attention" and
"automaticity", complicated on the other by fundamental
reconceptualizations of the nature and organization of attentional
resources.

Were this picture as grim as it seems, one tempting response would
be to discard the concept of automaticity as typically formulated in
lieu of more analytic constructs from multiple resource theory involving
interactions and tradeoffs among resource pools (see Navon & Gopher,
1979). Unfortunately, multiple resource theorists have been among the
first to point out that the conceivable patterns of trade-offs and
interactions between resources in a multiple resource framework are
woefully underdetermined by the experimental manipulations available in

dual-task methodology (Navon & Gopher, 1979; Navon, 1984). Navon (1984)

9

has recently argued that it is impossible to prove the necessity of a
concept of resources in accounting for dual-task performance and
apparent capacity limitations, and has gone so far as to take the
initial steps toward a model of "attention" that has no limited capacity
resources (Navon, 1985).

One can understand Navon’s pessimism when endeavoring to
disambiguate the findings of research on dual-task performance equipped
with a multiple resources framework that makes no assumptions about the
nature, number, or organization of the resources mediating performance.
There are indications, however, of a path leading out of this
theoretical morass (see Posner, 1982), which may not only clarify a
conceptual framework for automaticity but also provide a bridge between
the "chronic undifferentiated" multiple resources view and the more
traditional unitary, centrally-limited capacity model.

The first indication lies in the observation that the experimental
data motivating the multiple resources view provides evidence only for
specific resources, not against the existence of general or central
resources existing in addition to specific resources (Allport, 1971;
Allport, Antonia, & Reynolds, 1972; Kantowitz & Knight, 1974, 1976;
Gopher, Brickner, & Navon, 1982). For example, in the investigation of
Allport et al (1972) variations in task concurrence costs as a function
of secondary task modality are taken as evidence for specific resources,
which is presented as evidence against the notion of a general purpose
central capacity. What the authors fail to consider is that there were
significant amounts of task concurrence costs for all primary-secondary
task combinations, which provides equally direct evidence for the

existence of general resources, in addition to modality-specific ones

10

(see Carr, 1979; and Posner, 1982; for similar arguments). The case is
further strengthened by the ubiquitiousness of dual-task interference
between tasks that are almost totally different in terms of their input
and output modalities (e.g., Keele, 1967; Johnston, Greenberg, Fisher, &
Martin, 1970; Trumbo & Noble, 1970; Kahneman & Wright, 1971; Kahneman,
1973; Logan, 1976, 1979; among others).

What is suggested by these considerations is a sort of hybrid model
of attention (Posner, 1982) consisting of both general and specific
resources. This seems a convenient formulation in a number of respects.
First, it serves as a bridge between the multiple resources view and the
central capacity limitation view, integrating the empirical findings
which have motivated the multiple resource notion with the
phenomenological and experimental basis for central capacity
limitations. In addition, it establishes a more coherent framework for
thinking about the transition from attended to automatic performance, to
the effect that the emergence of automaticity corresponds to a reduction
of demands made on central resources through the operation of learning
mechanisms enabled by increasing familiarity with the task. Thus,
automaticity may be viewed as a sort of task-specific learning which
produces a dedicated performance structure where central, general
purpose resources are not needed (Anderson, 1982; Pitts & Posner, 1964;
Keele, 1967).

As may be apparent, serious questions accompany this formulation of
automaticity. What is a "central resource"? What type of mechanism can
produce a decrease in the central resource commitment required for a
task? First, a "central resource" may be thought of as a general purpose

capacity-any limited capacity functional entity in the information

11

processing system that is likely to be required during initial practice
in a wide variety of tasks. Formulating central resources in this way
entails only the proposition that some resources are more general than
others; general in terms of their applicability to tasks of varying
nature. It does not necessarily imply that some resources are
intrinsically "central" because of their function (though this may be
the case), or that "central resources" are a well-defined set whose
members can be enumerated or described (though that would seem a
worthwhile enterprise). Interestingly, some good candidates for "central
resources" are suggested by the domains of attention described above:
perceptual processing and recognition (Broadbent, 1958, 1984; Posner &
Snyder, 1975; Johnston & Dark, 1982; Kahneman & Treisman, 1984; ),
memory processes including memory retrieval, working memory and
short-term memory (Waugh & Norman, 1965; Baddeley & Hitch, 1974;
Baddeley, 1976; Atkinson & Shiffrin, 1968), and executive control or
response planning and preparation (Logan, 1978, 1979, 1980; Carr, 1979).
The second question, involving the type of mechanisms which may mediate
demands on central resources as a function of practice, will be held for
a later section.

In summary, I have attempted to establish the claim that
formulations of automaticity are faced with the problem of incoherence
resulting from varying orientations toward "attention", including both
the domain of phenomena to be addressed by the term and the more
fundamental issue of its nature and function. Building on suggestions
from Posner (1982) and others, I have developed a basis for dealing with
the problem of incoherence, as regards the formulation of automaticity,

involving the notion of a "hybrid" model of attention which integrates

 

 

12
the features of the general purpose central limited capacity view with
the multiple resources view. It should be noted that a more specific
elaboration of the hybrid model is not required at this point given the
scope of the present research. The strategy is to identify a particular
function (or "resource") which for both logical and empirical
considerations can be confidently described as central, general-purpose,
limited in capacity, and linked to consciousness. Armed with a
description of this resource’s function, it is possible to explain how a
task may engage the resource and how, with practice, the commitment
required for the resource may be reduced. Whether this resource is the
only central resource, or if other central resources can be identified
independently, or if other functions can be shown to depend on this
resource independently of whether other resources exist, are questions
which can be pursued at a later time without compromising the basic
thrust of the present research. Thus, I am trying to establish a
conceptual framework and a concomitant experimental strategy for
identifying and eventually taxonomizing these functional resources that
qualify as central or general, and for explicating the mechanisms by
which performance that is initially limited by reliance on these
resources comes to rely less on them-~and hence becomes more

"automatic"-through practice.

Sytematicity and Completeness.

 

The problem of systematicity may be an indirect manifestation of a
predominant interest in the characteristics or features of automatic
processes rather than in how and why automatic processes become

automatic to begin with. In part, this may be because empirical

13

investigations involving automaticity have tended to be concerned with
whether a given process is or is not automatic (e.g., word perception),
rather than with a more fundamental description of the mechanisms which
underly the emergence of automaticity. This has led to a relative
abundance of what might be called "criterial" definitions of
automaticity, in which a list of characteristics, features, or criteria
is enumerated in the service of providing a decision algorithim for the
automaticity or non-automaticity of a given process (LaBerge, 1981;
Shiffrin & Schneider, 1977; Schneider & Shiffrin, 1977; Kahneman &
Treisman, 1984; Jonides, 1981; Hasher & Zacks, 1979).

Criterial definitions have their uses. Perhaps most importantly,
they have generated an impressive set of characteristics or features
with which a more complete treatment of automaticity must be concerned
(see Table 1). While serving adequately as a decision algorithim,
criterial definitions convey the impression that automaticity consists
of a variety of covarying, interassociated features whose underlying
dynamics are worthy of exploration. However, if one’s objective is to
focus on automaticity itself, rather than to use the concept of
automaticity as a supplementary feature in the description of some other
process, then something more is needed.

The problem of systematicity relates to this failing of criterial
definitions. In short, current formulations of automaticity are strong
in identifying the characteristics of automaticity, but weak in
explaining why those characteristics are interassociated and how they
deveIOp. What is needed might be called a "conceptual" definition--a
description of the underlying dynamics of automaticity which takes the

features identified by criterial defintions and integrates them into a

TABLE 1

Commonly Cited Features of Automatic Processes

Eeaturgslflgfjgitign

Cheap or free in terms of processing
resources, "cf f art", or drain on
general-purpose, limited capacity
central processing mechanism.

Absence of task concurrence costs;
automatic processes will not
interfere with simultaneously
performed tasks.

Obligatory but capacity demanding;
A fixed, predetermined, uninhibitable
allocation of resources in the service
of an obligatory process triggered

by specific stimuli.

Difficult to inhibit once started.

Difficult to inhibit before starting.

Increasing stimulus control.

Without intention.

Without awareness.

Cannot be improved upon by practice
or feedback.

WW

Most frequently mentioned
defining feature. See Shiffrin

& Schneider, 1977; LaBerge, 1981;
Hasher & Zacks, 1979; Jonides,
1981; Shiffrin, Dumais &
Schneider, 1981.

Exact claims depends on what is
meant by a ”simultaneously
performed task”. A corollary of
above. See Posner &

Snyder, 1975; Logan, 1979;

Griffith & Johnston, 1977;

Bahrick & Shelley, 1958.

Typically identified in connection
with Stroop and Stroop-like
phenomena. See also Shif f rin,
Dumais & Schneider, 1977; Navon
& Gopher, 1979; Paap & Ogden,
I981; Regan, 1978, 1981.

Shiffrin & Schneider, 1977;
LaBerge, 1981; Jonides, 1981.

A stronger version of above, also
conneced with Stroop-like effects.
See LaBerge, 1981, and alternative
interpretations of Stroop
phenomena, Francolina & Egeth,
1981; Kahneman & Treisman, 1984.

A derivative formulation of two
preceeding.

See Hasher & Zacks, 1979; Posner
& Snyder, 1975; Navon & Gopher,
1979, LaBerge, 1981.

See same references as one
preceeding.

See Hasher & Zacks, 1979.

TABLE 1 (continued)

Commonly Cited Features of Automatic Processes

Egaggrelegfinitign

Resistance to modifications,
inflexible.

Insensitive to changes in
expectancy.

Invariant across arousal level.

Characterized by (or limited to)
simplicity.

Resistant to introspective access.

Characterized by parallel processing.

Automatic processes are unlimited
in capacity themselves.

ngmgntflAssogiated Wgrk
See LaBerge, 1981.

Relates to two-process
formulation in work and letter
perception. See Posner & Snyder,
I975; Stanovich & West, 1981;
Neely, 1977. See also Jonides,
1981.

In connection with the
Yerkes-Dodson law. See Hasher
& Zacks, 1979.

See Hirst, Spelke, Reaves,
Caharack & Neisser, 1980;
Neissar, Hirst & Spelke,
1981; Lucas & Bub, 1981.

Reference to ”fingertip memory"
and difficulty in articulating
skilled performance. See
Anderson, 1976, I982.

Connected mainly to visual search
task. See Shif f rin & Schneider,
1977; Neissar, 1967; Graboi, 1971.

See Kahneman & Treisman, 1984;
Navon, 1975. There seems to be
some confusion over whether
”automatic” is taken as the
absence of capacity demand on a
central resource or if ”automatic"

is interpreted as unlimited in
capacity. These are very

different claims. See also

Logan, 1979.

16

conceptual framework in which the interrelationships and underlying
dynamics of these features can be explained and predicted.

The problem of completeness is closely related to the issue of
systematicity. It refers to the fact that, with rare exceptions
(identified below), contemporary treatments of automaticity have not
attempted to describe the underlying dynamics which mediate the
emergence of automaticity. While there are numerous descriptions of
"attended" performance (Norman, 1968; Shallice, 1972; Kahneman, 1973;
Posner, 1982), and numerous criterial descriptions of automatic
performance (e.g., LaBerge, 1981), the precise nature of the
transisition from attended to automatic performance seems to have been
avoided.

The rationale for the present research builds on the exceptions to
this description, most notably Anderson (1982) and Logan (1978, 1979,
1980). The objective is to merge these contributions with the
perspective on attention developed in the preceeding section in order to
develop a conceptual framework in which the issues concerning the
existence and nature of automaticity can be carefully framed and

investigated.

CHAPTER 2

Automaticity, Working Memory, and Skill.

The overall form of the following discussion involves the
develOpment of two general points. One is the identification of primary
memory (James, 1890; Waugh & Norman, 1965) or working memory (Baddeley &
Hitch, 1974; Baddeley, 1976) as a unitary central resource that is
limited in capacity. The second is the adoption of what might be called
a "skills assembly" view of task performance (Miller, Galanter, &
Pribram, 1960; Shallice, 1972; Logan, 1978, 1979, 1980; Reason, 1979;
Carr, 1979; Anderson, 1982; MacKay, 1982). Subsequent discussion will
focus on the interaction between working memory and task performance and

how the concept of automaticity can be addressed in this framework.

Working Memory.

In the preceeding section on the problem of coherence and the
challenges of the multiple resource views, a "hybrid" model of the
structure of attentional resources was outlined. The essential claim of
a hybrid model is that some limited capacity resources are more
"central" than others, where "central" reduces to the sense of a
"general purpose" function likely to be required across a wide spectrum
of conceivable tasks. The proposal I would like to advance is that a
temporary storage function, related to the existing concepts of primary
memory (James, 1890; Waugh & Norman, 1965), working memory (Baddeley &
Hitch, 1974; Baddeley, 1976), and short-term memory (Atkinson &

17

18

Shiffrin, 1968), can be identified as a "central", general purpose
limited capacity resource. (For convenience, I will refer to this
hypothetical resource as working memory.) This is not to say that
working memory is the only central resource, or that working memory
performs only the functions ascribed to it below, or that working memory
is not simply one manifestation of a more abstractly conceived central
resource. The logic of this development entails only that working
memory, as a temporary storage mechanism, is a limited capacity resource
that is engaged by the need to retain information for a period of time
during the performance of a task, and that this need is common to a wide
range of tasks.

Naturally, working memory can be (and has been) conceptualized in a
number of ways, as the plurality of labels with varying implications and
metaphors listed above indicates. These conceptualizations include
working memory as a sort of hard-wired assortment of "slots" into which
information is written for short-term storage (Atkinson & Shiffrin,
1968; Norman & Rumelhart, 1975), working memory as the temporary
activation of regions of long-term memory (Atkinson & Juola, 1974;
Shiffrin 8 Schneider, 1977), and working memory as a sort of "mental
blackboard" where processes write their output and read their input
(Allport, 1979). At a more fundamental level, working memory might be
described as a limited capacity process which establishes relational
properties or associations between concepts or items in long-term memory
that are not already established and stored there, and whose identity
and relationship is in some sense necessary for the successful
completion of a task.

With respect to the concept of automaticity being developed here,

19

the exact dynamics of working memory are not critical so long as it can
be agreed that working memory is a limited capacity entity which
represents information for temporary or short-term storage. There is
considerable evidence that working memory corresponds to a limited
capacity resource (see Lachman, Lachman, and Butterfield, 1979). There
are also good grounds for supposing that working memory is a central
resource, in the sense that it ought to be required across a wide
variety of tasks that might otherwise seem to have complementary
distributions of resource demands (see e.g., Keele, 1967; Kahneman,
1973; Logan, 1976, 1979). Finally, if granted the view that working
memory serves to represent relational information not already stored in
long-term memory, then it is easy to formulate some hypotheses about
what types of learning mechanisms may bring about a reduction in the

working memory commitment required for a task as practice increases.

Working Memory and Task Performance.

 

This view of the role of working memory in task performance builds
on work in the areas of skill acquisition (Anderson, 1982; MacKay, 1982;
Miller, Galanter, & Pribram, 1960), attention and consciousness
(Shallice, 1972; Carr, 1979), and on the role of attention in task
performance (Logan, 1978, 1979, 1980; Reason, 1979). As will be evident,
what these views amount to is a description of how different tasks
engage working memory resources during performance. This is taken to be
a prerequisite to proposing how working memory resources may 325 be
engaged by a task that has had more practice, which is the starting
point for defining automaticity.

We begin by assuming the view that any task performance situation

20

consists of a goal (Anderson, 1982), formulated in the mind of the
performer, that must be met in order for the task to be successfully
completed. The route to achieving that goal will probably depend on the
completion of a set of smaller goals, what might be called "sub-goals",
which lead from the initial state to the desired main goal state. Each
sub-goal may in turn entail more subordinate sub-goals, so that any task
can be described in terms of a hierarchy of goals ranging from the
overall or main goal at the top of the hierarchy to an assortment of
subordinate goals which must be assembled and met in order for the main
goal to be achieved. This description is highly similar to the TOTE unit
(Test - Operate - Test - Exit) formulation developed by Miller et a1
(1960). Each goal in the hierarchy might be viewed as a TOTE unit. This
type of framework is also inherent in recent treatments of skill
acquisition (Anderson, 1982; MacKay, 1982).

A key assumption in the view being developed here is that, in some
form or another, the information contained in this "goal-hierarchy" task
description must be represented in the mind of the performer in order
for the task to be successfully completed. Thus, in a novel task
situation, the performer is faced with the job of figuring out how the
task is done, which, in terms of this description, refers to formulating
and organizing the required goals and "sub-goals". This process of
assembling the "goal structure" of the task corresponds to
"interpretative" (Fitts, 1964) or "declarative" (Anderson, 1982)
performance, in which slow and hesitant performance is attributed to
uncertainty and errors in figuring out how to accomplish the task at
hand.

The second major assumption is that the goals in a goal structure,

21

when repeatedly assembled in the service of a particular task, become
directly and permanently interassociated with one another. This could be
formulated, for example, in terms of a network representation where task
goals are represented by nodes and their interassociations are links
between the nodes that increase in strength as a function of practice at
the task. The same principle can be expressed in terms of Anderson's
(1982) production system approach, in which a process called composition
combines two or more subordinate productions into a sort of
"macroproduction". At bottom, there appears to be a number of ways that
this process can be described (see also MacKay, 1982), each of which
carries along slightly different constraints and implications. For
present purposes, the terminology of the network representation will be
used with the implicit presupposition that this basic point can be
mapped into a more specialized framework when experimental evidence
gives us reasons for preferring one framework over another. For
convenience, the formation and strengthening of direct interassociations
among goals in a goal structure will be referred to as "unitization" of
the goal structure.

The concept of unitization, when combined with the preceeding
discussion of working memory, enables the generation of some hypotheses
about automaticity. These can be derived from the central assumption
that an important function of working memory is to assemble and maintain
the goal structure of a task during initial performance. After practice,
the process of unitization effectively reduces the amount of information
that needs to be represented in working memory during performance
through having established permanent associations between the relevant

goals. The process is analogous to the phenomenon of "chunking" as

22
identified in memory for chess positions (Chase & Simon, 1973) and in
digit span performance (Miller, 1956). The concept of working memory as
a limited capacity mechanism which represents new relational properties
between old ideas or concepts is consistent with this idea.

With regard to skill acquisition, the basic idea is that a novel
task situation is likely to require a new goal structure that is not
already unitized. On the other hand, many of the sub-goals for the task,
which could be described as more molecular components at lower levels in
the hierarchy of the goal structure, may be commonly performed entities
that are already strongly unitized. Thus, the involvement of working
memory may be restricted in general to the higher levels of the goal
structure required by the novel nature of the overall or main goal. This
suggests that novice performance involves selecting and organizing more
molecular, lower-level, unitized goals and assembling them into a form
that will satisfy the task’s main goal, while expert performanceT-at
least in routine skills (Hatano & Inagaki, 1983)-may involve retrieving
or activating a specialized, task-specific goal structure that is
already assembled, and for which the working memory requirements are
minimal. In this framework, the emergence of skill in a task can be
discussed in terms of the level of the goal hierarchy at which
unitization has occurred-having fewer pre-unitized goals in the goal
structure means greater working memory demands during performance.
Unitization at higher levels reduces working memory demands and speeds
performance through avoiding the need to represent or compute task
goals.

The process of unitization may be illustrated with a sequential

keypressing task. In this task, each of four different command labels

23
specifies a sequence of five keypressing responses represented as
digits. For example, command "set" is executed by pushing buttons 3, 2,
1, 5, and 4, in that order, where the buttons are labelled on a keyboard
in front of the subject. The instructions are to punch out the
keypressing sequence corresponding to the command label presented.

The goal structure for a keypressing trial in this task consists of
a main goal, "complete sequence X", which consists of sub-goals
designated by each digit in the sequence. Each subgoal consists of the
sensory and motor processes needed to locate and press the corresponding
key. In the present framework, the claim is that during initial practice
at this task the associations between the command label and the other
sub-goals in the goal structure must be represented in working memory
because those associations are not permanently learned. After
unitization, the established associations can be used to drive
keypressing performance without having to be represented in working
memory. The resulting savings in working memory commitment is meant to
correspond to the subjective feeling that less conscious attention is
required for the task, and that daydreaming or some other concurrent
activity which did not seem possible before seems possible now.

It should be noted that the development of unitization requires
what might be called "operational constancy". Specifically, the
sub-goals subsumed under any common dominating goal must be constant in
terms of their identity and sequence in order for stable associations to
form. This description of operational constancy corresponds to what has
been called "consistent mapping" in experiments on visual search
(Shiffrin & Schneider, 1977; Schneider & Shiffrin, 1977; Graboi, 1971;

Kristofferson, 1972, 1977; Neisser, Novick, & Lazar, 1963; Schneider &

24

Fisk, 1982) and on symbol-to-keypress mapping tasks (Logan, 1979). In
this research, the criteria for automaticity were met relatively quickly
when the correSpondence between specific stimuli (i.e., letters or
digits) and specific responses was held constant. Where
stimulus-response correspondence varied, however, no amount of practice
was sufficient to produce what could be considered as automatic
peformance.

This principle of operational constancy is taken to be the major
limitation on the development of unitization in task performance. If the
goal structure of a task varies continuously without repetition or
patterning, then the present framework claims that the task cannot in
principle be unitized. Variations in goal structure which follow a
systematic pattern translates into a more complex, though constant,
structure with context-dependent branching. More complex goal structures
are viewed here as requiring more practice for unitization, because more
associations must be formed and strengthened, but not as being

impossible to unitize in principle.

Optimization.

 

Having defined the concept of unitization in terms of the formation
of interassociations between components in task performance, it is
important to consider that there are other logically possible ways in
which a task’s goal structure may change over practice-ways which may
also be capable of producing increases in speed and accuracy and
decreases in temporary storage requirements. These mechanisms, which I
will refer to collectively as optimization factors, correspond to

changes in the temporal and structural organization of the task other

25

than the formation and strengthening of interassociations among task
components that defines unitization. For example, one might learn after
practice that some elements of the task as performed initially are
actually irrelevant to the task’s goal, and can be dropped out without
harm. It is also possible that the task may be temporally reorganized,
changing the order in which task components are executed to maximize
efficiency. A third possibility is that the algorithm underlying
performance is fundamentally reorganized, as in Cheng’s (1985) analogy
contrasting addition and multiplication. If one learns to use a
multiplication algorithm instead of a serial addition strategy in mental
arithmetic, the resultant savings in temporary storage requirements come
from an algorithmic change rather than from the build-up of associations
in an otherwise unchanged task structure. In this example, the
contributions of unitization are limited to the working memory
representation of the sub-goals of the addition or multiplication
procedures themselves.

Another logically possible type of optimization could involve
changes in the ability to discriminate between task-relevant and
task-irrelevant information. During initial practice the performer might
take in and store information that, with practice, is learned to be
irrelevant and unnecessary to performance. As an example, consider a
sorting task where forms are separated into different stacks according
to a three-digit number. With practice, the performer may learn that
some numbers are uniquely identified by just the first digit, and that
the rest of the number need not be processed or stored. This would
reduce the working memory demands of the task by minimizing the amount

of information that needs to be handled during performance, a process

26

which capitalizes on learning about regularities in the structure of
task-relevant information. Such learning would seem to accrue gradually
as a result of experience in the task situation, but seems conceptually
quite different from the type of learning involved in unitization.

It is interesting to note that changes in the task algorithm
associated with optimization may interact with the process of
unitization. Dropping out, adding, or reorganizing task components may
produce discontinuities in performance efficiency over practice. The
speed of performance may dip downward, reflecting the transient demand
on working memory before unitization of the modified goal structure is
complete, and then increase towards a new and higher asymptote enabled
by increased efficiency. Such irregularities in performance gain over
practice could be related to the observation of plateaus in skill
acquisition (see Bryan & Harter, 1899; Logan, 1985). The probability of
errors might increase because of interference from old associations in
the goal structure, giving rise to branching errors (Reason, 1979;
Norman, 1981).

A third type of optimization relates to the temporal organization
of the task. This type of optimization stems from the observation that
many tasks may entail unavoidable working memory demands for information
that is constantly changing (and thus cannot be unitized). Examples
could include visual search under varied mapping conditions (Shiffrin &
Schneider, 1977; Graboi, 1971), or the working memory representation of
words or meanings in reading connected text. The point is that the
amount of demand on working memory for representing this sort of
information may vary during performance, being larger or smaller at one

time in performance than at another. Given this temporal profile of

27

working memory demands during performance, it is possible that these
demands may be spread out or redistributed in time enabling the
elimination of periods of excessive demand or the simultaneous
processing of other information at a particular point in time.

In the literature on dual-task performance (Bahrick & Shelley,
1958; Damos & Wickens, 1980; Gopher & North, 1977; Hirst, Spelke,
Reaves, Caharack, & Neisser, 1980; among many others) this temporal
adjustment process is called "attention switching" or "time sharing". In
some cases, it is offered as a theoretical alternative to automaticity
(Broadbent, 1982; Damos & Wickens, 1980), though in the present
framework both unitization and attention switching as described above
are considered as viable possibilities, the task being to isolate the
contributions of each in a particular task.

For present purposes, the most important point about attention
switching is that it does not require operational constancy. To
illustrate this, consider a dual-task variant of the sequential
keypressing task described earlier. In this version, a sequence of
digits is presented and the subject is instructed to press the
corresponding keys as labelled on a keyboard. The subject must retain
the digits in memory in order to complete the keypressing sequence after
the digits are removed from view. On dual-task trials, a separate series
of digits is presented before the keypressing sequence. The subject is
asked to recall these digits after completing the keypressing. Thus, the
digits to be recalled and the digits for the keypressing task must be
simultaneously represented in working memory. Because the recall digits
are constantly changing in identity from trial to trial they cannot be

chunked, and because the keypress sequences are also constantly

28

changing, they cannot be unitized.

Attention switching in this task would involve rapid alternation
between the recall digits and the keypressing digits. In a condition
where the keypressing digits were constantly changing and could not be
unitized, attention switching could develop because the overall temporal
organization of the task remains constant. Note, however, that
unitization could not develop if the keypressing digits changed from
trial to trial. This differential dependence on information constancy is
the principle basis for distinguishing between attentional switching and
unitization. It is worthwhile to note in addition that the development
of an attention switching strategy depends on practice under dual-task
conditions. If the subject is not given a chance to observe and exploit
the temporal organization of concurrent performance, then attention
switching would not have a chance to develop. Unitization, on the other
hand, can develop without the opportunity to practice both tasks
concurrently (see Logan, 1979). Thus, manipulating the amount of
dual-task practice in a task provides a means for distinguishing between
the contributions of attention switching (which is at heart a process
that operates on between-task temporal relationships) and unitization
(which is at heart a process that operates upon within-task sequential
relationships).

To summarize thus far, I have developed the concept of unitization,
based on previous work in skill acquisition (Bryan & Harter, 1897, 1899;
Miller, Galanter, & Pribram, 1960; Keele, 1967; Anderson, 1982; MacKay,
1982), and suggested how unitization may mediate the working memory
demands of a task as a function of practice. Unitization is viewed as

one of potentially many mechanisms which may influence working memory

29

demands. Several examples illustrating the additional mechanisms,
classed together for present purposes as optimization factors, have been
presented and discussed. While the present research focuses on the
mechanisms and conditions associated with unitization, this is viewed as
only the first step in distinguishing the potential variety of
mechanisms which may underlie contemporary use of the term
"automaticity".

One implication of this conceptual framework is that the phenomena
traditionally associated with automaticity may result from a variety of
interacting processes, variously potentiated by the nature of the task
and the conditions of practice. "Time sharing" and "automaticity",
typically viewed as competing theoretical options, may actually be
closely related processes working hand in hand to reduce capacity

demands in performance.

Summar .

This completes the sketch of the conceptual framework for
automaticity. The research described in the next chapter is designed to
determine whether automaticity occurs as defined here in reference to
the concepts of unitization and working memory. It seems important to
clarify, however, that this framework has some noteworthy limitations.
As noted above, the current view does not address the existence or
nature of "hard-wired" automatic processes (Hasher 8 Zacks, 1979). In
addition, the focus on working memory as a passive storage mechanism has
tended to obscure the potential role for automaticity in perceptual
processing-an area which relates directly to a significant portion of

existing work on automaticity (e.g., visual search and Stroop-like

3O

phenomena). Whether automaticity in perceptual processing is amenable to
the goal structure analysis developed in this framework is a question
for further development.

A more fundamental limitation is that the present formulation
serves only as a sort of minimal test case for automaticity, in the
sense that only one potential underlying mechanism associated with
automaticity-unitization-is conceptualized with respect to a
relatively narrowly defined central resource with a carefully prescribed
function. Good arguments could be made for other types of automaticity
that are defined with respect to other central resources or functions,
the development of which would complement the present framework. How one
construes these possibilities awaits a more refined and detailed model
of processing resources.

Despite this narrowness, the present framework seems to serve
several important purposes. One is that it may establish a more detailed
understanding of the relationship between attention and skill, providing
an important component in modern accounts of skill acquisition. In
addition, the distinction between unitization and optimization
highlights the possibility that there may be a number of ways that
attention demands may be reduced with practice. The current view
provides a basis for distinguishing between these possibilities.
Finally, if the framework developed here is supported in the data, it
may point the way toward a more fruitful conceptualization of the nature

of attention.

CHAPTER 3

Overview g£_Research Design and Rationale

 

The experiment described in the next two sections was designed to
lay the groundwork for a series of experiments focusing on the questions
and predictions of the preceeding formulation of automaticity. In the
present experiment, the principle objective was to document the
occurrence of unitization as it has been described and defined above.
Having established this, additional manipulations in the experiment were
used to provide further details about the nature of the mechanisms
underlying unitization and the conditions in which they may operate.

A sequential keypressing task, which was developed in the course of
a pilot study, was used in the experiment. In this task, subjects were
first asked to memorize several different digit strings (e.g., 6 - 5 - 1
- 4 - 2 - 3), and to associate each digit string with a sequence label,
consisting of a single letter. The subject was drilled to a criterion on
repeating the digit string when prompted with its corresponding sequence
label. After the sequences were memorized, the subject was presented
with a keyboard containing labelled pushbutton switches which
corresponded to the digits in the sequences. Subjects were instructed to
press the keys corresponding to the digit sequence when its sequence
label was visually presented on a computer monitor. The task can be
described as having two general components, accessing the digits in the
appropriate sequence and mapping the digits into the appropriate

keypressing sequence. Three dependent measures were taken from each

31

32

keypressing trial: the latency from the presentation of the sequence
label to the first keypress (called Time To Respond or TTR), the average
latency from one keypress to the next in the series (called Fluency),
and whether the sequence was completed correctly.

Two main variables in the experiment were practice and task
concurrence. The practice effect concerned changes in speed and accuracy
at two points in performance, one early (T1) and one late (T2) in the
experiment. These points were separated by about two and a half hours of
practice. Task concurrence involved the contrast between single-task
trials and dual-task trials. Single-task trials were exactly the same as
described above-responding to the sequence label with the appropriate
keypressing pattern was the subjects’ only concern. Dual-task trials
could be described as single-task trials carried out in the retention
interval of a digit span task. Eight digits were presented for about
four seconds, followed by the presentation of the sequence label. After
keypressing was completed, the subject was instructed to recall the
digits. Thus, the subject had to retain the digits in working memory
while processing the sequence label and completing the keypressing
sequence.

The key finding in the experiment was the interaction between
practice and task concurrence. Assuming that both tasks made initial
demands on working memory, the finding that task concurrence costs are
smaller later in practice (at T2) than earlier in practice (at T1)
indicates that the working memory demands of one or both tasks have
decreased. As such, the experiment was designed to isolate changes in
working memory demands relative to the keypressing task, with changes

attributable to practice at the digit span task being minor or,

33

preferably, nonexistent. For this reason, all of the practice
intervening between T1 and T2 consisted of single-task trials. In
addition, subjects were coached in the use of a verbal rehearsal
strategy to prevent the emergence of spontaneous rehearsal strategies
which could have produced discontinuities in digit span performance.
However, if practice related changes in digit span performance did
occur, the use of a control group which did not perform dual-task trials
until T2 would detect and isolate the effect. This control condition
will be described in greater detail below.

There are two noteworthy points about using this task as a way of
tracking the working memory demands of the keypressing task. First,
since the inference is based on observing changes in task concurrence,
alternative explanations involving initially non-overlapping resources,
or any other "static" formulations, can be ruled out. Second, the nature
of the digit span task allows a fair degree of confidence that initial
interference between the tasks is occurring in working memory and not in
some other resource, capacity, or function. This is because the digit
span task requires no input, output, or mental computation other than
that required to retain the digits during the interval in which the
keypressing task is performed. This makes the idea of more "peripheral"
interference between the tasks seem implausible. The logic of this
approach is described in more detail by Logan (1979).

It should also be noted that the use of two chronometric dependent
variables (TTR and fluency), could permit a further specification of
where the two tasks interfere. The TTR variable, which is the latency
from the presentation of the sequence label to the execution of the

first keypressing response, would include the time required for the

34

perceptual processing of the sequence label, the recall of the
associated digit sequence, and the preparatory or activational processes
needed to initiate keypressing. The fluency variable, which indicates
the rapidity of keypressing once started, would include the processes by
which motor control is transferred from one keypressing target to the
next and the processes driving the overt execution of motor responses.
Observing whether the practice by task concurrence interaction held for
TTR, fluency, or both, would provide additional information about the
aspects of the keypressing task which engage working memory and the
mechanisms which may mediate the changes in working memory requirements.

Having established that the working memory demands of the
keypressing task were reduced as a function of practice, the remaining
conditions in the experiment were designed to provide further
information about the nature of the mechanisms which mediated the
effect. The principle goal of these additional manipulations was to
determine if unitization, as formulated in the preceeding section,
played a role in the reduction of working memory demands. The
alternative to unitization is the class of factors referred to above as
optimization. The general strategy for distinguishing these
possibilities lies in the manipulation of operational constancy, which
would affect these mechanisms differently. Since unitization requires
Operational constancy, the removal of operational constancy would wipe
out the working memory savings attributable to unitization. Since
optimization factors are in principle independent of operational

constancy, these mechanisms would not necessarily be disrupted.

Shift Conditions.

 

35

The contrast conditions relating to operational constancy involved
three groups of subjects in the experiment. These three groups represent
two different manipulations of operational constancy and one control
group in which operational constancy was not varied. The groups will be
referred to as the Keyboard Shift group (KBS), the Sequence Shift group
(SES), and the No-Shift group (NS). All shifts were imposed at the same
point in practice, immediately before the second test period (T2).

Subjects in the Keyboard Shift (KBS) group were faced with a
seemingly random rearrangement of the keys on the keyboard. The
experimenter removed the switch caps from the pushbutton switches,
shuffled them, and replaced them on the keyboard in a new pattern. Thus,
the motoric patterns which map out the keypressing digits were changed,
while the digits themselves and their corresponding sequence labels
remained the same. This might be viewed as a "lowblevel" manipulation of
operational constancy since it affected the sub-goals under a digit
identity rather than the digit identities themselves (which would be
dominated by the overall trial goal in the goal structure). In general,
an increase in task concurrence costs following this type of shift would
support a unitization interpretation.

A particularly interesting way of interpreting the keyboard shift
situation lies in the possibility that performing the keypressing
sequences has developed into a unitary motor program (Keele & Summers,
1976), triggered by the sequence label. In this case, performance of the
keypressing sequences may no longer be mediated by the maintenance of
digit identities in working memory (which is consistent with the notion
of unitization). The keyboard shift manipulation would then force the

subject to fall back on the digit identities to mediate their new

36

mappings into motoric patterns. This would be consistent with Anderson’s
(1982) claim about the drop-out of verbal mediation in skill
acquisition. In the context of motor program theory, then, a substantial
increase in concurrence costs following the keyboard shift would be
strong evidence for unitization. However, it must be noted that motor
program theory assumes learning that is seguence specific. This
assumption causes an interpretational problem whose redress is one of
the functions of the sequence shift group.

Subjects in the Sequence Shift (SES) group were asked to learn an
entirely new set of sequences and sequence labels. The keyboard layout,
however, remained the same. This shift condition represents a
higher-level manipulation of operational constancy than the K38 group.
Since all task parameters were essentially the same with the exception
of the identity and order of operations, any detrimental effects
associated with the SES would be attributed to sequence-specific
learning. The degree to which performance was robust against the
sequence shift would indicate a role for such factors as attention
switching, keyboard knowledge, and fine-tuning of motor responding and
eye-hand coordination. If unitization did play a significant role in
mediating working memory demands, then task concurrence costs should
increase following the SES. There is, however, one version of an
attention switching hypothesis that could account for an increase in
task concurrence costs in this group. This is the argument that the SES
has disrupted the temporal organization of the task, and hence the
strategy for switching attention between tasks. The key question for
this formulation is why the temporal organization of the task was

disrupted in the first place. In order for the attention switching

37

explanation to replace a unitization interpretation, the locus of the
problem caused by the SES must lie outside of working memory processes.
Otherwise, one cannot explain the original cause of the temporal
disorganization without referring back to unitization, with a resulting
formulation which necessarily involves both unitization and attention
switching. For attention switching to completely explain the SES effect,
the source of the disruption must lie in some sort of peripheral
interference. There are a couple of considerations mitigating against
this sort of interpretation. First, it is difficult to envision a way in
which the SES could precipitate peripheral interference that was not
present before the shift. Second, since there was no Opportunity to
practice dual-task trials between T1 and T2, it seems unlikely that an
attention switching strategy would be able to develop. Additional
considerations concerning the emergence of time sharing strategies are
discussed below.

The SES group also provides information about optimization factors
that may have been affected by the keyboard shift condition. These
factors would support an explanation of the change in task concurrence
costs over practice in terms of knowledge of the keyboard rather than
sequence-specific knowledge. If, for example, the reduction of
interference is due to knowledge of key positions and/or the development
of a generalized set of keypressing motor programs, then the K88
condition would produce a recurrence of interference that would be
indistinguishable from the predicted effect of unitization. The SES
condition can disambiguate these possibilities. If all working memory
savings can be attributed to keyboard knowledge rather than sequence

knowledge, then the SES should not affect the load on working memory. If

38
there is any recurrence of task concurrence costs, then unitization must
play at least a partial role.

The No-Shift (NS) condition provided a baseline against which the
effects of the other shift conditions could be compared. The shift
effects could also be compared with each other, as well as against
initial performance at T1.

In addition to the three groups mentioned above, a control group
was used which did not experience dual-task trials until late in
practice at T2. This group, referred to as the no-shift/no dual-task
group (NS-NDT), served as a check on the possibility that the
interaction of practice and task concurrence could be explained by the
emergence of time-sharing strategies, since subjects in this group did
not have an opportunity to experience the demands of concurrent
performance.

It should be noted that all of these effects involving the shift
conditions could be present in the TTR measure, the fluency measure, or
both. Analysis of which shift conditions affected which components of
performance would provide additional analytic detail about the
mechanisms mediating reductions in the workig memory demands of

performance.

Manipulation 2£_Squence Length.

 

 

The variation of sequence length draws its motivation from research
by Henry & Rogers (1960), Sternberg, Monsell, Knoll, & Wright (1978),
and Rosenbaum, Saltzman, & Kingman (1984). The relevant finding in these
experiments has been that the latency to initiate a sequence of motor

movements, and the rapidity with which elements in the sequence are

39

completed, are an increasing function of the number of elements in the
sequence. This sequence length effect has been found in sequential arm
movements (Henry & Rogers, 1960), in speech and typewriting (Sternberg
et a1, 1978), and in simple finger movements (Rosenbaum et al, 1984). It
has held in both simple (Sternberg et al, 1978) and choice (Rosenbaum et
a1, 1984) tasks. It has also occurred across a variety of different
foreperiod conditions, ranging from the unavailability of any advance
information about the sequence to be performed (Rosenbaum et a1, 1984),
to as much as four seconds of advance time with countdown signals every
one second (Sternberg et al, 1978). Length effects have also been found
in conditions with variable foreperiods (Sternberg et a1, 1978).

The sequence length effect has uniformly been interpreted in terms
of a motor program buffer concept (Sternberg et a1, 1978; Rosenbaum et
a1, 1984). While there are a variety of different instantiations of the
concept, each yielding a slightly different model (Sternberg et a1
discuss six models, Rosenbaum et al entertain three models, one of which
has two versions), all of the explanations of the sequence length effect
have in common the idea that the elements of the sequence to be
performed are loaded as independent entities into the program buffer
which is then searched serially for each element to be executed. Thus,
the time taken to find the first element, as well as the time taken to
find the remaining elements when they are due to be performed, is a
function of the number of items in the motor buffer.

For present purposes, the relevant observation is that this sort of
formulation seems to be inconsistent with the sort of performance
structure that would result from unitization. The concept of unitization

would hold that practice with specific sequences would result in the

4o

formation of direct associations between the elements in a sequence. The
process of searching a motor buffer, indicating a piecemeal treatment of
sequence elements, would seem at odds with the idea that performance
consists of executing an integrated motor program.

This would be a problem for the concept of unitization if the
experiments reporting the sequence length effect used extensive practice
with specific sequences, bringing about the conditions where unitization
is supposed to occur. While the Sternberg et al experiments allowed
extensive practice, the sequences to be performed were changed after
every block of trials, which ranged from three to a maximum of
twenty-five sequence repititions. The Rosenbaum et al study had somewhat
lower, yet still fairly extensive, practice levels, but also varied the
sequences to be performed after every twenty-five trials. Thus, the
condition of operational constancy was not met in these studies and the
development of unitization would not be expected.

The question, then, is whether or not practice under conditions of
operational constancy would reduce or negate the sequence length effect.
If so, then the sequence length effect and the program buffer idea would
be characteristic of non-automatic performance, while the notion of
sequence-specific learning, involving the formation of a unitized goal
structure with direct associations between elements in the performance
sequence, would characterize of performance mediated by unitization.

In view of these considerations, the present experiment
incorporated the manipulation of sequence length. Each subject was asked
to learn six sequences, two sequences consisting of two keypresses, two
sequences with four keypresses, and two with six keypresses. The

measures of TTR and fluency should increase as sequence length increases

41

during initial practice (T1). At T2, the sequence length effect should
be significantly attenuated, according to predictions that would follow

from unitization.

Manipulation 2f Sequence Frequengy.

 

 

Another condition to be addressed involves the amount of practice
allowed on each sequence. Half of the sequences used (one sequence at
each level of length) were designated as high-frequency sequences, to be
practiced extensively in the practice period between T1 and T2. The
remaining sequences, the low-frequency sequences, were practiced only
occasionally in the practice period. During the test phases at T1 and T2
all sequences occurred equally often.

The potential significance of the manipulation of frequency is that
it can provide additional evidence regarding the role of optimization
and time sharing. The outcome of interest is if task concurrence costs
for low-frequency sequences are greater than for high-frequency
sequences. (This analysis would be based on T2 performance since the
frequency manipulation was not established at TI.) This would indicate
that low-frequency sequences have higher working memory requirements
than high-frequency sequences, despite the fact that both types of
sequences are performed with the same degree of overall task experience.
The operation of most optimization mechanisms could not play a role in
the effect, since these mechanisms should work equally well for both

types of sequences.

Analysis 2£_Serial Position.

In addition to the major independent variables described above,

42

another variable was derived by breaking down the fluency measure for
the six-element sequences (L=6). This involved the analysis of the
key-to-key transition latencies for each of the five transitions present
in an L-6 sequence, which permitted the comparison of keypressing rates
as subjects proceeded through the sequence. This analysis becomes
interesting in relation to MacKay’s (1982) model, which uses a concept
of anticipatory priming wherein performance of a sequence is supposed to
speed up later in the sequence due to forward-spreading activation in
the motor program. In regard to the concept of unitization, one might
expect forward-priming of the goal structure after unitization but not
before. Thus, seeing a trend toward speed-up later in practice but not
early in practice could provide a converging operation on the emergence
of unitization. The analysis could also reveal the presence of pauses or
hesitations in performance, perhaps indicative of on-line motor planning

processes.

CHAPTER 4

Review 2£_Experimental Procedures and Design

 

Subjects.

Sixty-four volunteers were recruited by advertisement in a local
student paper announcing the availability of payment for participation
in psychological research. Two of the 64 participants had to be dropped
because of their inability to memorize and execute the keypressing
sequences. These were replaced with two additional participants who were
associates of the experimenter. All subjects were naive regarding the
variables and predictions in the experiment, but were candidly informed
of the overall topic of the research.

Of the 64 participants whose data was used, 63 were in the college
age bracket between roughly 18 and 25 years of age. Approximately
two-thirds of the participants were female. Two of the participants were
fluent but nonnative speakers of English. All participants described
themselves as basically right-handed (and used their right hand only in
the keypressing task), with the exception of one left-hander that was
assigned to the sequence shift group. Informed consent was obtained
after the initial description by the experimenter of the purpose and

nature of the research.

Design.

The independent variables in the experiment were practice (T1 and
T2), task concurrence (single-task and dual-task), shift condition (KBS,

43

44

SES, NS, and NS-NDT), sequence length (2, 4, or 6 keypresses), and
sequence frequency (high or low). Shift condition was a between subjects
variable and the remaining conditions were varied within subjects.

The dependent variables were Time To Respond (TTR), Fluency, and
error rates. Fluency was calculated by taking the mean of the
inter-keypress latencies across a sequence. In addition, the recall
error rate in the digit span task on dual-task trials was recorded. The
digit span error rates were analyzed in terms of practice, shift

condition, sequence length, and sequence frequency.

AEparatus.

The experiment was conducted on an Apple II+ microcomputer equipped
with a custom-made interface package. The keypressing keyboard was
mounted on a wooden box with the face tilted about twenty degrees toward
the subject. The keyboard was placed on a platform in front of the CRT,
both of which were directly in front of the subject.

The switches were standard single-throw pushbutton switches with
about a 4 mm throw. The switch caps were square, approximately 1.6 cm a
side. Each switch was labelled with a number between 1 and 6, printed on
an address label affixed to the switch cap. In the K38 condition, the
switch caps, along with the affixed labels, were simply transferred from
one switch to another.

The switches were arranged in a circular pattern around a central
switch which was used as a starting point. The numeric identity given to
the switches was not orderly with respect to position. Instead, the
assignment of digit identities around the circle appeared random to the

subject. (Actually, the arrangement of digit identities was sharply

45

constrained by counterbalancing requirements.) The apparent randomness
of digit-to-switch assignment was necessary because of the KBS
condition, wherein rearranging the switches would ruin any orderly
layout, causing the KBS manipulation to be confounded with transitioning
from a systematic to an unsystematic layout. The distance from the
center of a key to its nearest neighbor on either side was constant

around the circle (approximately 4.5 cm).

Procedure.

The experiment began with a preliminary description of the purpose
and nature of the experiment prior to obtaining informed consent.
Subjects were then asked to memorize the digit sequences for the
keypressing task. This involved learning the sequence label (a single
consonant letter) and the two, four, or six digits associated with each
label. After initial presentation, the subject was drilled to a
criterion on producing the digits in correct order when provided with
the sequence label. The criterion was getting all sequences right five
times consecutively.

When the sequence learning criterion was reached, the subject was
presented with the keyboard and the mapping of the digit sequences into
keypressing sequences was explained. Subjects were instructed to make
their keypressing as rapid and accurate as possible, using the index and
middle fingers of the dominant hand held together to depress the
switches. On dual-task trials, subjects were told to protect performance
on the digit-span task at the cost of speed on the keypressing task, if
some interference between the tasks seemed unavoidable.

For the digit span task, subjects were coached to use a verbal

46

rehearsal strategy involving subvocal reptition of the span digits in
the retention interval. This approach to the problem of strategy
variance in digit span performance was motivated by the pilot project
and by theoretical considerations following from Chase & Ericcson
(1982). With regard to the pilot study, the sub-vocal repetition
strategy was (perhaps surprisingly) not immediately discovered by many
subjects. Despite the apparent fact that most people use such a means
for short-term retention of phone numbers and other lists, their initial
reaction to the dual-task trials was to drop the digits completely while
engaging the keypressing task. The attempt to reconstruct the span
digits after finishing was usually futile, resulting in dual-task costs
loading on digit span performance rather than on keypressing speed,
where the contrast between single-task and dual-task trials is needed to
fuel the task concurrence by practice interaction. Explicit coaching in
the maintenance of a verbal rehearsal "loop" during keypressing was
intended to ensure the discovery (and hopefully the use) of the same
strategy by all subjects at the beginning of the experiment, and to help
make task concurrence costs load on keypressing rather than digit span
performance. The continued use of the strategy was also encouraged at
T2.

The second consideration, from Chase & Ericcson (1982), is that the
use of a verbal rehearsal strategy tends to prevent the emergence of
more sophisticated strategies later in performance. The goal in this
respect was to ensure that subjects persisted in the use of the verbal
rehearsal strategy so that the working memory demands of the digit span
task would remain roughly constant over the amount of practice allowed

in the experiment (48 trials).

47

After these instructions were given, the keypressing phase of the
experiment was begun. The sequence of events on the single-task trials
was as follows: First, a message was printed on the CRT indicating that
the following trial was a single-task trial. This message was left on
the screen for the same amount of time as the digits would have remained
on the screen had it been a dual-task trial. The screen was then
cleared, followed about 1 second later by a fixation stimulus (a plus
sign), which was replaced by the sequence label after approximately 3
sec, which was left on the screen until the end of the timing interval
(6.4 seconds). After the timing interval had elapsed, a feedback screen
was presented which showed the latencies for each keypress in the
sequence and which indicated what the correct sequence should have been.
The computer then waited for the subject to press the ready key in the
center of the keyboard, indicating readiness for the next trial.

On dual-task trials, the sequence of events was the same as
described above except that digits were presented in lieu of the
"single-task" message and the subject was asked to recall the digits
prior to the presentation of the feedback screen. Recall was recorded by
having the subjects repeat the memory span digits out loud. Subjects
were instructed to give ordered recall. The feedback screen for
dual-task trials included all of the information given for single-task
trials, and included in addition information on recall accuracy in span
performance.

The central key on the keyboard (called the "ready" key) was not
used in the keypressing sequences themselves but served as a starting
point on each trial. Subjects were instructed to depress the key at the

beginning of a trial and to release it only after the sequence label

48

appeared on the screen. Its use was intended to ensure equal hand travel
distance to the beginning of any sequence. Trials in which the ready key
was released prematurely were flagged as missing data.

The initial test phase consisted of the first 48 trials in the
experiment, which has been referred to previously as T1. This consisted
of eight repetitions of each of the six sequences, with four single-task
trials and four dual-task trials for each. The order of sequence
presentations was randomized independently for each subject with the
constraint that each half of the test block (the first 24 and the second
24 trials) contained half of all trial types (i.e., each sequence was
seen four times with two trials each under single- and dual-task
conditions).

The next 300 trials in the experiment, which were all single task
trials, constituted the practice block. The trials in this block were
spread out over two days. The first day included the first test phase
(T1) followed by 132 trials of the practice phase. These 132 trials
consisted of 126 repetitions of the high frequency sequences (42 trials
per sequence), and 6 repetitions of the low frequency sequences (2
trials per sequence). The second day continued the practice phase with
156 single-task trials, 144 high frequency trials (48 trials per
sequence) and 12 low frequency trials (4 repetitions per low frequency
sequence). In total, the practice phase yielded 270 trials of practice
on the high frequency sequences (90 trials each), and 18 trials of
practice on low frequency sequences (6 trials each).

The shift conditions were imposed at the end of the practice phase,
immediately prior to the beginning of the final test phase (T2). At this

time, the sequence shift group stopped to learn new sequences and the

49

keyboard shift group had their keyboard rearranged. In sum, the keyboard
shift group did old sequences on a new keyboard layout, the sequence
shift group did new sequences on an old keyboard layout, and the
no-shift group did old sequences on an old keyboard layout. Like the
no-shift group, the NS-NDT group was not confronted with a shift
condition.

The final testing phase (T2) began at the conclusion of the
practice period. This was organized in the same manner as T1, with the

same number of trials and the same randomization algorithm.

Materials and Counterbalancing.

 

The digits for the memory span task on dual-task trials consisted
of the digits 0 through 9, excluding 7, and were selected and ordered
randomly for each dual-task trial independently for each subject.
Ordered recall instructions were given.

The design of the keypressing sequences was guided by six major
constraints: (1) No two-key combination was used twice in the set of six
sequences active for a subject at a given time. (2) All sequences within
a sequence set began on different keys. Since there were six sequences
and six keys, all the keys had one sequence which began on them. (3)
Sequences within each level of length were roughly equated in terms of
the total distance to be traversed by the hand during execution. (4) The
total distance to be traversed by the hand in the four-key sequences
(L-4) was exactly equal to the distance of the first four keypresses in
the L-6 sequences, allowing direct comparison of keypressing fluency at
equivalent serial positions between these lengths. (5) No sequence was

allowed to return to a key that was already pressed in that sequence.

50

(6) Sequences were designed to avoid highly characteristic, unique, or
symmetric mapping patterns that might cause some sequences to stand out.

In general, the counterbalancing of the sequences relative to the
independent variables in the design generated three categories of
control. The first category included the variables for which sequences
served as their own control. This applied for the variables of practice
and task concurrence, since each sequence appeared at each level of
these variables within a subject. The second category was where
sequences served as their own control through counterbalancing across
subjects rather than within subjects. This was the case for sequence
frequency, which was handled by varying sequence assignment within
groups. The third category was where the properties of sequences had to
be equated rather than counterbalanced, which held for the variables of
shift condition and sequence length.

The complication presented by the shift conditions was that
changing the keyboard layout radically altered the motor mapping pattern
of the sequences. The solution to this required the derivation of two
different keyboard layouts which, when a set of sequences was to be
transferred from one to the other, maintained the structural constraints
described above. In conjunction with this, two sets of sequences were
developed that could be mapped out on either keyboard within the same
constraints. Control for the properties of the sequences across the
shift conditions was then accomplished by varying the keyboard layout
and sequence set assignments within a group so that all of the possible
combinations occurred before and after the shifts.

The difficulty with the sequence length manipulation was in

ensuring that hand travel distances across different lengths were

51

approximately equal when averaged over the keypresses in each sequence.
To accomplish this, the use of short, medium, and long key-to-key
transitions was balanced across sequence length, keyboard layout,
sequence set and sequence frequency so that the average hand travel
distance between keypresses turned out about the same for all lengths.
The means were 6.50 cm for L=6, 6.49 cm for L-4, and 6.90 cm for L=2.

Finally, the sequences in the L-6 condition were designed so that
the key-to-key transition distance, whenaveraged over the sequences in
the two sequence sets, came out to be approximately equal across the
five different serial positions present in L-6 sequences. This was
completed to enable the analysis of fluency broken down by serial

position that was mentioned previously.

CHAPTER 5

Description 2f Results

 

Overview 2£_Variables and Data Analysis Procedures.

 

 

The principle chronometric dependent variables of the design, as
described previously, were Time To Respond (TTR), and Fluency. The TTR
measure was taken as the latency from the presentation of the sequence
label to the switch closure of the first keypress in the appropriate
keypressing sequence. The fluency measure was computed by taking the
average of the interkey latencies between all of the keypresses in the
sequence. In addition, a total time measure was computed by multiplying
the average fluency by five and then adding the corresponding TTR value.
This procedure adjusts the total time for sequences of length two and
four to what they would have been had six keypresses been performed at
the same rate. Thus, gross differences in total time across sequences of
different length are filtered out, allowing only the computationally
independent measures of TTR and fluency to influence variation of the
means across levels of length. The rationale behind the development of
the total time measure was to provide an overall performance metric
reflecting both TTR and fluency. Having established overall performance
differences through total time analyses, the data can then be broken
down into the more analytic measures of TTR and fluency to determine if
variation comes from one or both measures or, in the case of no total
time variation, if that is achieved through constancy in both TTR and

fluency or through tradeoffs between them.

52

53

Two additional chronometric analyses were performed by breaking
down the fluency measure into separate interkey latencies for each
key-to-key transition. In one analysis, the average interkey latency
across each serial position was examined for sequences of six keypresses
(L-6). In the other, the first three latencies in L=6 sequences were
compared with the three latencies of L-4 sequences.

In addition to the chronometric analyses, the error rates on
keypressing and the digit span task were recorded. These variables
provide checks for tradeoffs between speed and accuracy and between
primary and secondary tasks.

For the total time, TTR, fluency, and keypressing error rate
measures the independent variables were Group (no-shift, keyboard shift,
sequence shift), Practice (T1 or T2), Task Concurrence (single-task or
dual-task), Sequence Frequency (high-frequency or low-frequency), and
Sequence Length (2, 4, or 6 keypresses). The serial position analysis
for L-6 sequences included in addition five levels of serial position
(first through fifth interkey latency), but did not include sequence
length (since this analysis used only L=6 sequences). The contrast of
L=6 and L-4 included three levels of serial position (first through
third) and two levels of length (L-6 and L84). The analysis of digit
span error rates was carried out with the same variables as total time,
minus task concurrence (since the digit span task was performed on
dual-task trials only).

The analyses pertaining to the NS-NDT group (which performed
dual-task trials at T2 only) were planned comparisons carried out in
light of the issues concerning the effects of practice on the digit span

task and the contributions of practice on dual-task trials at T1. Thus,

54
the questions to be addressed with respect to this group were: (1)
whether this group was different from the other groups at T2 in digit
span performance, and (2) whether the task concurrence costs for this
group were any larger or smaller than the T2 task concurrence costs for
each of the other groups. To address these questions, the NS-NDT group
was compared individually against the no-shift, keyboard shift, and
sequence shift groups in the analysis of T2 performance alone for the
dependent variables of total time, TTR, fluency, keypressing error
rates, and digit span error rates.

The description of the results is broken down into six general
sections: (1) overall group and practice effects; (2) effects involving
task concurrence, practice, and group; (3) effects associated with the
NS-NDT group at T2; (4) effects associated with the serial position
analyses; (5) effects involving the manipulation of sequence frequency
in the practice period; and (6) effects involving sequence length,

practice, and group.

GrouE And Practice Effects.

 

Overall main effects for practice were large and robust for all
dependent variables with the exception of keypressing error rates. For
total time, TTR, fluency, digit-span performance, and in the serial
position analysis, the F-ratios ranged from 49.32 (for TTR) to 177.57
(for fluency). In the keypressing error data the practice effect was
marginal, F(1,45) - 3.25, p < .08, with the means showing a small
overall reduction in error rates from 13.52 at T1 to 10.72 at T2. The?

>

direction of the trend is inconsistent with a speed-accuracy tradeoffg

account 0

55

The interaction of group and practice was significant for total
time, TTR, fluency, and in the serial position analysis, with F-ratios
ranging from 6.30 to 14.21 (p < .01). The means (presented in Table 2)
showed a systematic decrease in the size of the practice effect across
the no-shift, keyboard shift, and sequence shift groups. In the TTR
data, the mean practice effect was 1629 ms for the no-shift group (from
2850 ms at T1 to 1221 ms at T2), 822 ms for the keyboard shift group
(from 2620 ms at T1 to 1798 ms at T2), and 250 ms for the sequence shift
group (from 2536 ms at T1 to 2286 ms at T2). In the fluency data, the
corresponding means were 190 ms (434 ms to 244 ms), 126 ms (434 ms to
308 ms), and 98 ms (459 ms to 361 ms).‘

The group by practice interaction was only marginal in the digit
span task, F(2,45) - 2.05, p < .15, and in the keypressing error rates,
F(2,45) - 2.70, p < .08. In both cases, however, the trend was in the
same direction as in the chronometric analyses.

Taken together, the group by practice interactions indicate that
the keyboard shift and sequence shift conditions served to attenuate the
size of the practice effect. The difference between the two shift
conditions suggests in addition that changing sequences was more

disruptive than changing keyboards.

Effects 2£_Task Concurrence.

 

In the overall analyses, using the no-shift, keyboard shift, and
sequence shift groups at both T1 and T2, the main effect of task
concurrence was significant at the .01 level for all dependent variables
as well as in the two serial position analyses. F-ratios ranged from

10.67 in the keypressing error rates to 65.70 in the total time

56

TABLE 2

Cell Means in the Interaction of Group and Practice
for TTR, Fluency, and Keyprcssing Errors.

__T_l__ _12_ Diffusing
TTR (ms)
No-Shif t Group 2850 1221 1629
Keyboard Shift Group 2620 1798 822
Sequence Shift Group 2536 2286 250
Fluency (ms)
No-Shif t Group 434 244 190
Keyboard Shift Group 434 308 126
Sequence Shift Group 459 361 98
Keyprcssing Errors
(percent error)
No-Shif t Group 13.0 5.6 7.4
Keyboard Shift Group 12.4 10.2 2.2

Sequence Shift Group 15.0 16.3 -l.3

57

analysis. For all measures, the means indicated better performance on
single-task trials than on dual-task trials. In total time, the
single-task mean was 3706 ms and the dual-task mean was 4463 ms,
yielding a difference of 757 ms. The corresponding means in the TTR data
were 1960 ms and 2477 ms, with a difference of 517 ms. For the fluency
data, the means of 349 ms and 397 ms yielded a 48 ms effect. The
keypressing error data revealed a 10.52 error rate on single-task trials
and a 13.62 error rate on dual-task trials.

The role given to operational constancy in the development of
unitization suggests that task concurrence costs should decline with
practice for the no-shift group, but not for the keyboard shift and
sequence shift groups. Thus the theory predicts a three-factor
interaction between group, practice, and task concurrence. This
interaction produced a trend toward significance in the analysis of
total time, F(2,45) - 2.82, p < .07. As expected, the no-shift group
showed a significant reduction in task concurrence costs from T1 to T2,
F(1,15) - 7.26, p < .05 (see Table 3). This effect was due mainly to
reductions in TTR, since the same interaction occurred in the TTR data,
F(1,15) - 7.26, p < .05, but was not significant in fluency or
keypressing errors. The means showed a 591 ms reduction in task
concurrence costs with practice for total time and a 485 ms reduction in
TTR. In contrast to the no-shift group, neither the keyboard shift group
nor the sequence shift group produced a task concurrence by practice
interaction in any dependent measure. Thus Operational constancy, both
of key location as indicated by the keyboard shift group and of the
sequence of key identities as indicated by the sequence shift group,

does appear to be a major factor in the reduction of task concurrence

58
TABLE 3

Cell Means (ms) for the Interaction of Group, Practice,
and Task Concurrence in the Total Time Analysis.

 

 

 

 

T1 T2
ST DT ST DT
No-Shift Group 4443 5594 2160 2719
Keyboard Shift Group 4412 5164 2949 3730

Sequence Shift Group 4776 5189 3799 4385

59

costs with practice.

The total time means for the keyboard shift group showed a 752 ms
task concurrence cost at T1 and a 781 ms task concurrence cost at T2,
yielding a slight (nonsignificant) increase in task concurrence costs
over practice. Somewhat strikingly, this failure to reduce dual-task
interference occurred despite an overall practice effect of 1148 ms,
from 4788 ms at T1 to 3340 ms at T2 (in total time). In contrast, the
sequence shift group showed task concurrence costs of 713 ms at T1 and
588 ms at T2, yielding a slight (nonsignificant) 125 ms decrease in
dual-task interference, while at the same time showing a significantly
smaller overall practice effect of 740 ms, F(1,30) - 9.26, p < .01. This
outcome suggests that overall improvements in performance level are not
necessarily associated with concomitant changes in capacity demand.

In sum, the evidence concerning the role of the group manipulation
in mediating the reduction of task-concurrence costs over practice is
relatively clear cut. With respect to the keyboard shift group, it seems
quite safe to conclude that changing keyboards late in practice prevents
the savings in task concurrence costs that are possible without such a
change, since that group’s task concurrence effect actually increased a
bit at T2 relative to T1. Though the sequence shift group’s task
concurrence costs did go down at T2, the change was not significant,
suggesting that in this case, as well, an absence of operational

constancy interfered with practice-induced savings in task concurrence.

Analyses with the NS-NDT Group.

 

As described previously, the NS-NDT group experienced dual-task

trials at T2 only, having performed single-task trials up until the last

60

48 trials in the experiment. The importance of this group lies in the
observation that there was no opportunity to develop or refine the
temporal coordination of the keypressing and digit span tasks during
concurrent performance. Hence, this group can act as a control for the
emergence of time-sharing or resource management strategies which might
otherwise be advanced as explanations of the task concurrence reduction
effect shown by the no-shift group. In addition, this group did not have
the opportunity to practice the digit span task. Thus, changes in the
task concurrence effect attributable to practice effects in digit
performance can also be examined by comparing this group’s performance
to that of the other groups. Since the NS-NDT group did not produce
dual-task data at T1, the following analyses were carried out with the
T2 data only.

Turning first to the question of whether the NS-NDT group showed
greater task concurrence costs than the no-shift group-which would
indicate a role for time sharing or optimization factors-the
interaction of group and task concurrence was significant for total
time, F(1,30) - 5.89, p < .05, and for TTR, F(1,30) - 4.36, p < .05, but
was only marginal in the keypressing error rates, F(1,30) = 3.70, p <
.07 (see Table 4). Surprisingly, the means for total time and TTR
revealed that the task concurrence costs for the NS-NDT group were
smaller than for the no-shift group, which is the opposite from what
would be expected if experience on dual-task trials played a role in the
task concurrence reduction effect seen in the no-shift group. For the
NS-NDT group, the single-task mean was 2306 ms and the dual-task mean
was 2559 ms, yielding a 253 ms task concurrence effect. For the no-shift

group, the corresponding means of 2160 ms and 2720 ms yielded a 560 ms

61

TABLE 4

Cell Means in the Interaction of Group and Task

Total Time (ms)
No-Shif t Group
NS-NDT Group

TTR (ms)
No-Shif t Group
NS-NDT Group

Keyprcssing Errors

(percent error)
No-Shif t Group
NS-NDT Group

Concurrence Contrasting the No—Shif t Group
With the NS-NDT Group at T2 for Total
Time, TTR, and Keyprcssing Error Rates

ST

2160
2306

1044
1076

4.9
5.5

DT

2720
2559

1397
1252

6.2
10.9

Diff erence

560
253

353
176

62

effect. However, the marginal group by task concurrence interaction in
keypressing error rates suggested that the reversed interactions shown
in total time and TTR were offset by a tendency for the NS-NDT group to
be biased toward keypressing speed on dual-task trials. On single-task
trials, the error rates for both groups were exactly equal, while on
dual-task trials the NS-NDT group showed a 62 increase in error rates
against a 12 increase in the no-shift group.

This pattern was augmented by the comparison of digit span error
rates, which produced a marginal main effect of group, F(1,30) - 2.91, p
< .10, showing a 162 error rate for the no-shift group and a 252 error
rate for the NS-NDT group. The analysis also produced an interaction of
group and length, F(2,60) - 3.89, p < .05, showing a 152 difference
between groups at L-6, a 62 difference at L-4, and a 52 difference at
L-2.

In sum, the evidence indicates that the NS-NDT group was somewhat
worse at digit span performance but showed smaller task concurrence
costs than the no-shift group. This pattern suggests that practice at
the digit span task helped, but not by allowing the formation Of a task
control structure that maximized performance on both tasks. Instead, it
resulted in a greater emphasis on digit span performance at the expense
of keypressing speed on dual-task trials. Overall, the results do not
indicate that the net interference shown by the NS-NDT group was any
greater than that of the no-shift group. Thus, there is no evidence that
time sharing or optimization factors played a significant role in the
task concurrence reduction effect shown by the no-shift group.

The second question regarding the NS-NDT group is whether this

group’s task concurrence effect was any smaller than that of the

63

keyboard shift and sequence shift groups. Comparing the NS-NDT group
with the keyboard shift group produced a significant interaction of
group and task concurrence for total time, F(1,30) - 8.28, p < .01, and
in TTR, F(1,30) - 5.87, p < .05. The interaction was marginal in the
fluency data, F(1,30) - 3.17, p < .09, and nonsignificant in the
keypressing error rates. The means revealed a task concurrence effect of
781 ms for the keyboard shift group against the 253 ms task concurrence
effect for the NS-NDT group reported above. The means in TTR and fluency
followed the total time means, and the groups were not significantly
different in digit span performance. Thus with respect to the keyboard
shift group, there is straightforward evidence that task concurrence
costs were smaller for the NS-NDT group.

The results regarding the sequence shift group were not so clear
cut, however. The interaction of group and task concurrence was barely
marginal in the total time data, F(1,30) - 2.26, p < .15, and
nonsignificant in TTR, fluency, and keypressing errors. While the means
in all instances favored the NS-NDT group, analysis of the cell
deviations in the total time analysis suggested that greater error
variance associated with the sequence shift group may have prevented the
effects from reaching statistical significance. In addition, there were
no significant differences in digit span performance between the groups.
Given this outcome, it seems appropriate to conclude that the results of
the comparison with the sequence shift group are equivocal, failing to
inform the issue one way or the other.

Overall, the results Of the NS-NDT group help to establish two
claims relevant to the occurrence of unitization. The first is that the

task concurrence reduction effect shown by the no-shift group is

64

probably not due to time sharing or Optimization factors resulting from
experience with the capacity demand characteristics of dual-task trials
at TI. This is established by the contrast of the NS-NDT group with the
no-shift group and is indirectly supported by the interaction of group
and task concurrence with the keyboard shift group. The second is that
the maintenance of operational constancy produces evidence of
unitization apart from prior experience with dual-task trials, relative
to conditions where dual-task practice did occur but operational
constancy was not maintained. This is established by the interaction of
group and task concurrence with the keyboard shift group, but is
compromised somewhat by the failure of the corresponding interaction to
attain significance with the sequence shift group. Thus, the first point

seems solid but the second point must be taken with greater caution.

Serial Position Analyses.

 

Two sets of analyses were performed on the fluency data broken down
by serial position. The first, which compared the five interkey
latencies in L-6 sequences, bears on the prediction following from
MacKay (1982) that speed-up in keypressing rate at the later serial
positions may reflect forward-spreading activation in the motor program.
In addition, this analysis may show systematic patterns in interkey
latency over serial position that may reflect on-line planning
processes. The second serial position analysis contrasted the first
three serial positions in L=6 sequences with the three serial positions
in L-4 sequences. This analysis addressed the question Of whether
keypressing rates varied between lengths over equivalent serial

positions.

65

Moving first to the L=6 analysis, which included the no-shift,
keyboard shift, and sequence shift groups for both T1 and T2, the main
effect of serial position was significant, F(4,180) - 16.99, p < .01.
The means, from initial to final serial positions, were 445 ms, 434 ms,
507 ms, 356 ms, and 273 ms, suggesting that the effect consists of two
components: a tendency to produce a peak interkey latency at serial
position three, and a tendency to increase the keypressing rate at the
latter two serial positions.

The peak interkey latency at position three takes on added interest
in connection with the interaction of serial position and task
concurrence, F(4,180) - 4.11, p < .01. The mean task concurrence costs
taken across serial positions 1 through 5 were 63 ms, 54 ms, 122 ms, 43
ms, and 16 ms, reflecting a peak task concurrence effect at position
three and diminishing task concurrence effects at the two later
positions (see Table 5).

The pattern was further elaborated in the three-factor interaction
of group, practice, and serial postition, F(8,180) - 2.16 p < .05. The
means, presented in Table 6, indicate that the size of the position
three "bump" effect was reduced over practice by 262 ms for the no-shift
group, 179 ms for the keyboard shift group, and actually increased by 58
ms for the sequence shift group.

Taken together, these effects suggest that subjects may have been
using a "divide and conquer" strategy in the L=6 sequences, pausing
after the completion of three keypresses to load, prepare, or plan the
execution of the remaining keypresses. It is also possible that, on
dual-task trials, subjects were using the halfway point as a break to

reinforce the memory span digits. In either case, the interaction of

66
TABLE 5

Cell Means (ms) in the Interaction of
Task Concurrence and Serial Position
in the L=6 Serial Position Analysis.

Serial Position

 

 

1 2 3 4 5
Single Task 413 407 446 334 265
Dual Task 476 461 568 378 281

Difference 63 54 122 44 16

67
TABLE 6

Cell Means (ms) for the Interaction of Group, Practice
and Serial Positions for Sequences of Length 6.

Serial Position

 

 

__l_ 2 3 4 __.5_
N '5! '1: 5
Time 1 500 551 569 367 309
Time 2 248 311 307 237 209
r hi r u
Time 1 467 527 586 484 291
Time 2 389 363 407 324 208
n hif r

Time 1 593 471 557 379 344

Time 2 470 381 615 397 277

68

group, practice, and serial position, showing that the bump effect was
attenuated more for the no-shift group than for the other groups,
suggests that the tendency to "divide and conquer" lessens when
Operational constancy is maintained. The modest attenuation of the bump
effect shown by the keyboard shift group might be viewed as an
intermediate case where the constancy of the keypressing digits reduces
the bump effect to reflect only the costs associated with planning the
new motor mapping. In general, the results are consistent with the idea
that practice with constant sequences and motor mapping patterns gives
rise to the development of an integrated motor program with a
corresponding trend away from piecemeal treatment of sequence elements.
A final point to bring out in the L-6 analysis regards the question
of speed-up across serial position. Previously, it was noted that
MacKay’s (1982) model of skill acquisition advanced a concept Of
forward-spreading activation in the motor program. In the model, such
feed-forward activation is said to prime upcoming elements in a motor
sequence, producing a speed-up in the rate of responding as one
progresses through the sequence. As noted above, examination of the
means presented in Table 6 does reveal a tendency toward speeded
execution at the later serial positions, though the effect is
complicated by the ubiquitousness Of the "bump" effect at serial
position three. There is also a trend for task concurrence costs to get
smaller toward the end. It is interesting to consider whether the
construct of feed-forward activation in the motor program should be
advanced as an account of the effect, or if the trend can be more
readily interpreted as a cost associated with the planning of upcoming

motor movements. Examination of the means in Table 6 suggests that the

69

tendency toward speed-up holds generally in all conditions-at T1 as
well as T2 and in the keyboard shift and sequence shift conditions as
well as the no-shift and NS-NDT groups. Inasmuch as the effect holds in
conditions such as T1 and in the sequence shift group, where the
evidence suggests that performance was piecemeal in nature rather than
the result of a previously established motor program, it seems clear
that the more parsimonious option involving planning of upcoming motor
movements ought to be preferred.

The final serial position analysis to discuss is the contrast of
the first three keypresses of L=6 sequences with the three total
keypresses in L-4 keypresses. As described previously, this contrast
provides a test Of the length effect over equivalent serial positions.
The analysis, using two levels Of length (L=6 and L-4), and three levels
of serial position (first, second, and third) produced a main effect of
length, F(1,45) - 30.08, p < .01, an interaction between length and
serial position, F(2,90) - 10.14, p < .01, and a three-factor
interaction involving length, task concurrence, and serial position.
Looking at the means of the length by serial position interaction, the
values for the L-6 sequences of 445 ms, 434 ms, and 507 ms across
positions one through three reflect of course the same trend as noted
previously. The corresponding means for L-4 of 396 ms, 423 ms, and 337
ms reveal a negligible bump effect at serial position two, corresponding
to the halfway latency between keypresses two and three in L-4
sequences. Thus, at equivalent serial positions the keypressing rate is
faster for the L-4 sequences than for L=6 sequences. This is especially
true at position three, where the two lengths diverge sharply,

reflecting an increase to the bump at position three for L-6 sequences.

70

The means for the three-factor interaction again show the tendency
for L=6 sequences to show large task concurrence effects at position
three, going from 63 ms at position one to 54 ms at position two and 122
ms at position three. In contrast, the mean task concurrence effects
across serial positions one through three for L-4 sequences are 72 ms,
42 ms, and 51 ms. It seems reasonable to attribute the lack of a task
concurrence peak in the L-4 sequences to the difference in size,
producing less Of the tendency to "divide and conquer" that was seen in
the L-6 sequences. In sum, the contrast Of L=6 and L-4 performance over
equivalent serial positions indicated that keypressing rates for L-4
were faster at all serial positions held in common, and that no major

trends toward bump effects were found in the shorter sequences.

Effects Involving Sequence Freguency.

As described above, the sequence frequency variable involved
differential amounts Of practice across sequences in the practice period
intervening between T1 and T2. High frequency sequences were practiced
90 times in the practice period while low frequency sequences were
practiced only 6 times. To Offset a possible "surprise" effect the three
low frequency sequences were always presented once just prior to the
beginning of T2. Since there was no difference in practice levels across
sequences at T1, the frequency variable was meaningful only at T2.
Similarly, the total change in sequence labels and elements faced in the
sequence shift group made sequence frequency a dummy variable in this
condition.

The importance of the frequency manipulation resides in the ability

to test for the effects of practice on specific sequences with overall

71

exposure to the task controlled. Thus, it provides a converging
Operation on ethe role of time sharing and optimization factors in the
reduction of task concurrence costs over practice. This is because the
finding that task concurrence costs varied across levels of frequency
would indicate that working memory demands differed as a function of
practice with specific sequences irregardless of the degree of overall
exposure to the task, whereas most time sharing accounts would expect
time sharing strategies to be equally effective over all sequences.
Thus, the prediction following from the concept of unitization would be
that task concurrence costs are greater for lowarequency sequences than
for high-frequency sequences, particularly in the no-shift and NS-NDT
groups where operational constancy was maintained.

It is also important to consider how the test for frequency effects
ought to be conducted. Since frequency is meaningless at T1 for all
groups, and meaningless at T2 for the sequence shift group, it seems
appropriate to test for frequency effects at T2 only for the no-shift,
keyboard shift, and NS-NDT groups. The inclusion of data in which
frequency is a dummy variable could only result in a loss of power,
since random variation associated with the levels of frequency under
these conditions would be partitioned into the error term for the
corresponding F-test.

Significant main effects for frequency were found in total time,
F(1,45) - 8.75, p < .01, in TTR, F(1,45) - 5.44, p < .05, and in
fluency, F(1,45) - 5.62, p < .05, with the means indicating better
overall performance on high-frequency sequences than on lowarequency
sequences. In TTR, the means were 2641 ms for high-frequency sequences

and 2833 for low-frequency sequences, yielding a 192 ms overall

72

frequency effect. In addition, the interaction of group and frequency
was significant for total time, F(2,45) - 8.39, p < .01, and in TTR,
F(2,45) - 7.94, p < .01, but narrowly missed significance in the fluency
measure, F(2,45) - 3.45, p < .06. The total time means revealed a net
frequency effect of +442 ms for the no-shift group and +309 ms for the
NS-NDT group. Interestingly, the keyboard shift group showed a modest
reversal Of the frequency effect, indicating a 176 ms advantage for
low-frequency sequences over high-frequency sequences.

Thus far, the data indicate that the manipulation of frequency
produced a strongly facilitative effect for the no-shift and NS-NDT
groups, but actually produced some interference for the keyboard shift
group. The interference effect would be sensible, since subjects in this
group would have to inhibit the well-established mapping patterns
learned on the Old keyboard. If this were so, then inhibition might be
more difficult under dual-task conditions when the digit span task is
competing for attention that could be devoted to the new keyboard
mapping. For this group, then, the theory predicts that task concurrence
costs may be greater for high-frequency sequences, whereas for both the
no-shift and NS-NDT groups task concurrence costs should be greater for
the low-frequency sequences. An important question, then, is whether
frequency interacted with task concurrence and group. Unfortunately,
this interaction was not quite marginal in the total time measure,
F(2,45) - 2.42, p < .11, and nonsignificant in the other dependent
measures. While the means, as predicted, showed the reversed task
concurrence by frequency interaction for the keyboard shift group, the
simple effects test on the keyboard shift group alone failed to reach

significance. Thus, there is no support for the role of task concurrence

73
in the frequency reversal pattern of the keyboard shift group.

However, other evidence regarding the interaction of frequency and
task concurrence was obtained in the planned analyses including the
no-shift and NS-NDT groups that were discussed in connection with the
NS-NDT group findings. While these analyses were conducted to test for
interactions of group and task concurrence, it seems appropriate to
examine them for effects involving frequency--to look for frequency
effects where theory predicted they would be strongest. By combining
these groups, the frequency by task concurrence interaction can be
tested without the influence of the frequency by task concurrence
reversal pattern shown by the keyboard shift group.

These analyses produced significant interactions of frequency and
task concurrence in fluency, F(1,30) - 6.71, p < .05, and in the L=6
serial position data, F(1,30) - 7.94, p < .01. For the fluency measure,
examination of the means revealed task concurrence effects of 17 ms on
high-frequency sequences and 40 ms on low-frequency sequences. A similar
trend was shown in the L-6 means.

In the L-6 analysis, the variables of frequency and task
concurrence also participated in the four-factor interaction with group
and serial position, F(4,120) - 3.32, p < .05 (the means are provided in
Table 7). In approaching this interaction, it is useful to keep in mind
the general pattern of the serial position analysis for L-6, which
showed a "bump" effect at the midpoint of the sequence which was shown
to vary as a function of practice and task concurrence. In the present
interaction, it can be seen that the "bump" effect is larger for
lowarequency sequences than for high-frequency sequences. One

indication of this is that the position three latencies average out to

74

264 ms in the high-frequency sequences and 329 ms in the low-frequency
sequences. In view of the task concurrence by serial position
interaction from the overall L-6 analysis, it might also be expected
that task concurrence costs tend to be larger at position three, which
is also the case here. In addition, the means indicate that the task
concurrence costs at position three are greater for low-frequency
sequences than for high-frequency sequences, the means indicating a 50
ms task concurrence effect for high-frequency and a 122 ms effect for
low-frequency sequences. This establishes a relatively coherent pattern
for the three-factor interaction of task concurrence, frequency, and
serial position.

The entry of the fourth factor-~group-may be approached by
recalling the task concurrence tradeoff pattern previously noted
concerning the no-shift and NS-NDT groups. That effect indicated that
the NS-NDT group seemed less inclined to allow keypressing speed to vary
in the face of the additional load imposed on dual-task trials. An
analogous effect in the present interaction would predict that the
NS-NDT group showed less variation across the cells of the three-factor
interaction of task concurrence, frequency, and serial position. An
examination of the means in Table 7 indicates that while a case could be
made that way, it is not quite that simple. To anticipate the eventual
conclusion, the pattern indicates that while the NS-NDT group was biased
toward constant keypressing rates across the levels of task concurrence,
they in fact did allow keypressing rates to vary more fully with respect
to sequence frequency.

One way of getting at this conclusion is to look at the total

amount of variation in keypressing latencies across serial position.

75

TABLE 7

Cell Means (ms) for the Interaction of Group, Task
Concurrence, Frequency, and Serial Position in
the T2 Analysis Contrasting the No-Shif t
Group With the NS-NDT Group.

Serial Position

 

l 2 3 4 5
NQ-Shin thgp
High Frequency
Single-Task 217 231 226 203 194
Dual-Task 259 282 310 222 208
Low Frequency
Single-Task 244 268 291 241 200
Dual-Task 270 464 403 281 233
N -N r
High Frequency
Single-Task 347 285 251 258 205
Dual-Task 247 277 267 261 205

Low Frequency
Single-Task 269 345 245 248 205
Dual-Task 322 339 378 248 222

76

This can be computed by summing the absolute values of the differences
between each serial position and the one following it. The result is a
measure of the overall fluctuation of keypressing rate over the whole
sequence. For the no-shift group at high-frequency, this measure comes
out to 51 ms on single-task trials and 153 ms on dual-task trials. The
higher variability on dual-task trials reflects in part differences in
the size of the "bump" effect. The corresponding variation measures for
the NS-NDT group on high-frequency sequences are 156 ms for single-task
trials and 102 ms on dual-task trials, yielding a difference of -54 ms.
The difference of +102 ms for the no-shift group against the -54 ms
difference for the NS-NDT group indicates that on high-frequency
sequences the NS-NDT group maintained a more nearly constant rate of
keypressing as they progressed through the sequence, with relative
indifference to task concurrence. Comparison of the same measures for
the low-frequency sequences reveals basically the same pattern,
indicating that the NS-NDT group also resisted variation in keypressing
rates over levels of task concurrence on low-frequency sequences,
relative to the no-shift group. However, looking at the same measures
contrasting variations in keypressing rates over frequency indicates
that the NS-NDT group did allow fluctuation with respect to frequency,
although still less than the no-shift group. Collapsing across task
concurrence, the no-shift group showed an average of 102 ms variation
for high frequency sequences and 282 ms of variation on low-frequency
sequences. The NS-NDT group showed 129 ms of variation on high-frequency
sequences and 217 ms on lowarequency sequences, yielding a difference
of 88 ms.

Additional support for this interpretation comes from the digit

77

span and keypressing error data contrasting these groups (these analyses
were carried out in the NS-NDT group tests described in a previous
section). In the digit data, the tradeoff of digit span performance for
keypressing speed on dual-task trials would lead one to predict a larger
difference between high-frequency and low-frequency trials for the
NS-NDT group than for the no-shift group. The means of the group by
frequency interaction in the digit span data indicate, instead, a
nonsignificant trend for the difference to be smaller for the NS-NDT
group (a 12 difference against a 42 difference). Similarly, as described
in the previous section addressing the NS-NDT group results, the
interaction of group and task concurrence in the keypressing error data
(which was marginal, p < .07) supports the tradeoff with respect to task
concurrence. However, the corresponding interaction between group and
frequency shows no trend toward a tradeoff, with the difference between
high-frequency and low-frequency sequences being 32 for the no-shift
group and only 12 for the NS-NDT group. Thus, the overall pattern seems
consistent with the proposition that the NS-NDT group resisted
variations in keypressing rate over levels of task concurrence but not
so much over the levels of frequency, where they seemed more willing to
allow variations in performance to show up in keypressing speed than in
keypressing errors. Given that the task concurrence manipulation is a
highly conspicuous variable from the subjects’ point of view, while the
frequency manipulation is relatively subtle and innocuous, this kind of
differential tradeoff seems a plausible outcome.

Further evidence suggestive of the effects associated with the
frequency manipulation was obtained in an analysis of keypressing error

patterns at T2 for the no-shift, keyboard shift, and NS-NDT groups. This

78

analysis focused on keypressing errors where part or all of one sequence
was substituted for the appropriate sequence to be performed. According
to the predictions that would follow from a unitization interpretation,
it would be expected that such intrusions would come from high-frequency
sequences more than from low-frequency sequences. Examination of the
occurrence of these "sequence-confusion" errors indicated that
high-frequency sequences tended to intrude on lowarequency sequences
twice as often as low-frequency sequences intruded on the performance of
high-frequency sequences. Of the thirty keypressing errors which
unambiguously indicated an intrusion from another sequence in the
sequence set, twenty were high-frequency sequences being substituted in
whole or part for low—frequency sequences. While the test for
significance was only marginal, F(1,45) - 3.36, p < .07, the effect is
consistent with the idea that the sequence elements in high-frequency
sequences were more strongly interassociated. In addition, it is
possible that high-frequency sequences were more "dominant" or readily
activated response patterns.

Taken overall, the distribution of effects involving frequency are
highly suggestive, but must be approached with caution. The overall
facilitation effect is consistent with the emergence of an integrated
motor program, as indicated in the serial position analysis described in
the previous section. The reversed frequency effect evident in the means
for the keyboard shift group is also consistent with the idea of an
integrated motor program. Otherwise, the tendency toward interference
with high-frequency sequences cannot be explained, since only the
keyboard mapping patterns were changed. If the interaction of frequency

and task concurrence in the analysis combining the no-shift and NS-NDT

79

groups is accepted, there is fairly direct evidence that task
concurrence costs varied over levels of frequency in spite of exactly
equal degrees of overall exposure to the task-—a result that would be
inconsistent with most time sharing or optimization accounts. The
four-factor interaction of frequency, task concurrence, group, and
serial position is also consistent with the idea that performance on
high-frequency sequences was mediated by the emergence of an integrated
motor program, indicative of a general trend away from piecemeal
treatment of sequence elements during keypressing. The overall pattern
is consistent with the idea that the task concurrence reduction effect
shown by the no-shift group is specific to the sequence and to the motor
pattern in which it is realized, not in the juggling and balancing of
resource demands over time made possible by experience with the demands

of concurrent processing.

Effects Involving_Squence Length.

 

The importance of the sequence length effect derives from the
research cited previously (Sternberg et al, 1978; Rosenbaum et al, 1984)
which indicated that sequence length effects generally failed to be
attenuated with practice. In the present experiment, in which the
concept of unitization is associated with the emergence of an integrated
motor program, it was anticipated that sequence length effects would be
attenuated given practice under conditions of operational constancy.
This would suggest a trend away from the piecemeal treatment of sequence
elements which is implied in the "motor program buffer" account of the
sequence length effect advanced by the Sternberg et al (1978) and

Rosenbaum et a1 (1984) studies, wherein operational constancy was not

80
present over practice. Thus, the present framework would predict that
sequence length effects should be attenuated over practice, showing less
dependence on the number of elements in the sequence in initiating
responding and in the rate of keypressing once started.

Overall, sequence length effects were robust and ubiquitous,
producing significant (all p < .01) F-ratios for all dependent variables
(except in the serial position analysis for L=6, in which length was not
a variable). The F-ratios ranged from 41.83 in the digits to 117.31 in
the total time analyses. In general, the shape of the length function
was very orderly, indicating better performance on shorter sequences.
For example, in the TTR data, the means for sequences of 6, 4, and 2
keypresses were 2628 ms, 2505 ms, and 1523 ms, respectively. The
corresponding means in the fluency data were 417 ms, 399 ms, and 302 ms.

It is also worth noting that the bulk of the variance across length
was between L-4 and L-2 sequences, with L=6 performance being typically
fairly close to L-4. This phenomenon may be largely attributable to the
"divide and conquer" strategy noted above in connection with the serial
position analysis for L=6. The use of the "divide and conquer" strategy
could mean that subjects were treating the L=6 sequences like two
three-element sequences, resulting in better performance on L=6
sequences than would be expected on the basis of a linear length
function. In addition, some subjects reported greater difficulty in
learning the L-4 sequences, and, while the reason for this seems
unclear, it is possible that L-4 performance may have been somewhat
poorer than would be expected from assuming a linear length function. In
some of the interactions involving length to be reported below, the

tendency toward non-linearity varied somewhat across the cells of the

81

interacting variable, resulting in the possibility that a component Of
the F-ratio was inflated by the variation in the differences between L-6
and L-4 performance, which would be partitioned into the treatment sums
of squares for the F-ratio. To safeguard against the possibility that
theoretically irrelevant variability associated with differences between
L=6 and L-4 performance contributed substantially to the interaction,
all of the length effect analyses were carried out twice, once in the
straightforward way where length is represented as a three-level
variable, and once with L-6 and L-4 performance averaged together,
making length a 'two-level variable (long vs. short). This latter
procedure provides a good estimation of the slope of the length
function, while filtering out the variance associated with
non-linearity. In the interactions to be reported below, all of the
effects were significant in both analyses. The F-ratios and means are
from the regular analyses with three levels of length.

The overall interaction of length and practice was significant for
total time, TTR, and in digit span performance (all p < .01). This was
nonsignificant in the keypressing error data, in the fluency measure,
and in the L-6 vs. L-4 contrast on serial position. The means, detailed
in Table 8, indicate that the length effect tended to diminish from T1
to T2.

The interaction of practice, length and group was significant in
the total time data, F(4,90) - 2.68, p < .01, and was marginal in the
serial position analysis contrasting L-6 with L-4, F(2,45) . 2.99, p <
.07, but did not attain significance in TTR, fluency, or keypressing
errors. Expressing the slope of the length function as the difference

between L=6 and L-2 performance in the total time data, the no-shift

£32

.- .- .. mom ~mp woe N»

 

a... .. -. .- 8m 0.: 5.. : 8.5 .E .2, .3
o: m: A ow m m~ N: Astoria acoutoac
8.x. 2 .cN s on : S : cam :30
n s a z o 2 2 .22... £883
.m.z A. q we mp o p~ F» chccm acmmmoca>ou
s: an no o2 on on». 2
.m... .c: :3 ~: 2.. ~ .3 : 85 >88:
8» 82 NS 82 so. a: 2
8.x. non. 22 22 88 S at» : 35 888“ 8 2.:
no: 83 8: 02... .3... no... 2
3.8 E: 33 an: :3 3 2mm : 8: .32
I «film js TN 4.24.4. ||.c_.| .3. .3. . To i 2.:

 

cameo; cocoauom uca Oumuoaca yo cowuomco»c_ 0:» Law acme: .dou

a wan<h

83

group showed a 1683 ms length effect at T1 and a 737 ms length effect at
T2, yielding a length attenuation effect of 946 ms. The keyboard shift
group presented a 488 ms length attenuation effect, with the length
effect decreasing from 1934 ms at T1 to 1446 ms at T2. The sequence
shift group showed a 2001 ms length effect at T1 which actually
increased to 2505 ms at T2, yielding a negative length attenuation
effect of -504 ms. This pattern indicates that operational constancy
played an important role in the attenuation of the length effect.

As would be expected, the analyses including the NS-NDT group at T2
also produced significant interactions of group and sequence length (p <
.01 for total time, TTR, fluency, and keypressing error rates). In the
total time means, the NS-NDT group presented a length effect of 808 ms,
from 2821 ms for L=6 to 2013 ms for L-2, against length effects of 737
ms for the no-shift group, 1446 ms for the keyboard shift group, and
2505 ms for the sequence shift group. Thus, the length effect at T2 for
the NS-NDT group was roughly comparable to that of the no-shift group,
while the keyboard shift and sequence shift groups showed much larger
length effects.

In the overall analyses, the length by task concurrence interaction
was significant for total time, TTR, fluency, and keypressing errors
(all p < .01, except for keypressing errors, where p < .05). In the
serial position analysis contrasting L=6 with L-4, the two-factor
interaction of length and task concurrence was not significant, but the
three factor interaction with serial position was significant at the .05
level. Task concurrence was not a variable in the digit span data.

In general, the shape of these interactions are about what one

would expect, showing greater task concurrence costs on the longer

84

sequences. The means also showed that task concurrence costs for L-6 and
L-4 sequences were approximately equal, with the bulk of the variance
arising from the difference in task concurrence costs between the two
longer sequences and the L-Z sequences. In the total time data, the
means were 4266 ms for single-task trials and 5175 ms for dual-task
trials, yielding a task concurrence effect of 909 ms. The corresponding
means for L-4 were 4008 ms and 4995 ms, yielding a slightly larger task
concurrence effect of 987 ms. For L-2, the means of 2846 ms and 3220 ms
yielded a task concurrence effect of 374 ms. The corresponding patterns
in the TTR and fluency data were almost exactly parallel. In the
keypressing error data, the size of the task concurrence cost--expressed
as the difference in error percentages-was 52 for L-6, 42 for L-4, and
02 for L-2.

Taken together, the effects involving sequence length are
consistent with the predictions that would follow from unitization-to
the extent that effects were found. 0n the one hand, there was
relatively solid evidence, based on the interaction of group, practice,
and length, that length effects are attenuated when operational
constancy is maintained. When motor patterns were changed, as in the
keyboard shift group, the degree of attenuation was reduced. When
sequence elements were changed, there appeared to be no attenuation at
all. This pattern would be expected if practice with unchanging
sequences and motor mapping patterns is associated with the emergence of
an integrated motor program and a trend away from piecemeal treatment of
sequence elements. 0n the other hand, however, sequence length effects
at T2 were still relatively large (464 ms for TTR, 54 ms for fluency,

overall), and there was no evidence that the task concurrence by length

85

interaction got smaller with practice, or that length interacted with
sequence frequency. These latter effects would also be expected on the
basis of unitization.

With respect to the size of the length effects at T2, it is
possible that further practice could enable a flat length function. This
remains an open question. It does seem clear, however, that the "motor
program buffer" models now need to incorporate features which can
account for the effects of motoric and compositional specificity.

~The lack of evidence for the three-factor interaction of practice,
task concurrence, and length, along with the failure for length to
interact with frequency, can be approached in two ways. One is to accept
the null result, with the subsequent suggestion that length effects have
two components, one that is sensitive to practice with operational
constancy, and one that is sensitive to task concurrence loads. The task
concurrence component is insensitive to practice, which is why length
does not interact with frequency. The other approach is to accept the
pattern as being basically uninformative, in view of the fact that null
results can be obtained for a variety of reasons (particularly the lack
of power), only one of which would be consistent with the first
approach. Given the high degree of variability apparent in the data, the

latter approach seems preferable.

CHAPTER 6

Discussion and Conclusions

 

Evidence of Unitization.

 

In review, the concept of unitization was developed in the
introduction as one of several logically possible ways in which the
capacity demands of a task may be reduced with practice. Unitization is
advanced as a specific and analytic formulation of one of the mechanisms
underlying the emergence of automaticity in skill acfiuisition, with the
goal in mind of being able to distinguish the contribution of
unitization from the contribution of other processes, such as
time-sharing (Broadbent, 1982) or "restructuring" (Cheng, 1985), which
are also capable of reducing the apparent capacity demands of task
performance. Given an unambiguous demonstration of the operation of
unitization in skill acquisition, it is hoped that the issues of
attention and automaticity in skill acquisition may be addressed with
more precision and systematicity than before, and that the operation of
other mechanisms mediating capacity demand may eventually be defined and
documented.

In the present experiment, unitization is conceptualized as the
emergence with practice of a unified motor program that can be called
and executed as a unit rather than constructed in a piecemeal fashion
from information held in working memory. Unitization is hypothesized to
be made possible by the formation of direct associations between
elements in the motor sequence. It is held to be applicable only when

there is fixity in the order and identity of elements in the sequence,
86

87

and produces the effect of reducing capacity demand by obviating the
need to store sequence elements in working memory during performance.

The evidence for unitization begins with the observation of the
practice by task concurrence interaction shown in the no-shift group.
This establishes, first, that there was interference between the tasks
during initial practice. That the locus of the interference lay in
working memory is supported by noting that the secondary task, digit
span performance, required no perceptual input or motoric output during
the time in which keypressing took place. Second, the amount of
interference between the tasks, as reflected in the difference in
keypressing latencies between single-task and dual-task trials, was
shown to decrease with practice. This indicates that there was some type
of change, occurring over practice, which enabled the tasks to be
carried out more independently of each other.

At this point, four hypotheses can be advanced to account for the
decrease in dual-task interference, so long as the locus of the initial
interference is accepted to be working memory. First, it is possible
that the capacity demands of the keypressing task decreased with
practice, allowing a speed-up in keypressing rates without trading off
secondary task performance. Second, it is possible that the capacity
demands of the digit span task decreased with practice. Third, it is
possible that the capacity demands of both tasks were reduced with
practice. Fourth, it may be that capacity demands were not changed for
either task, but merely redistributed over time to give only the
appearance of a reduction in capacity demand.

These possiblities can be largely disambiguated by considering the

results of the NS-NDT group, which did not practice the digit span task

88

or perform under dual-task conditions until the final test period late
in practice. First, this group showed no significant main effect
differences in digit span performance on dual-task trials at T2,
relative to the keyboard shift and sequence shift groups, despite the
fact that the NS-NDT group had no opportunity to practice the digit span
task prior to that point in the experiment. This suggests that practice
at the digit span task per se did not play a major role. The marginal
effect contrasting the NS-NDT group with the no-shift group in digit
performance, and the interaction between group and length, is offset by
the reversed interaction between group and task concurrence which
appeared in the total time and time-to-respond data. This showed a
smaller task concurrence effect for the NS-NDT group than for the
no-shift group. In sum, the data indicate that changes in digit span
performance alone cannot account for the task concurrence reduction
effect shown by the no-shift group, though such changes may have played
a role.

The hypothesis proposing the temporal redistribution of capacity
demands is also strained by the results of the NS-NDT group. This
hypothesis would seem to require that subjects have experience with the
capacity demand profile of concurrent performance before they would be
capable of formulating an efficient resource deployment strategy. In
fact, however, the NS-NDT group had no prior opportunity to experience
the demand profile of dual-task trials yet showed about the same amount .
of interference as the no-shift group, which did have experience with
dual-task trials. If the emergence of a time sharing strategy had played
a role in the task concurrence reduction effect for the no-shift group,

one would expect them to produce less interference than the NS-NDT

89

group. It could be suggested that a very quickly developing time sharing
strategy was possible, but this could not explain the strength of the
interference effect at T1 or the persistence of interference for the
keyboard shift and sequence shift groups, for whom the gross temporal
structure of the keypressing task did not change.

The data reviewed so far indicates that the reduction in the size
of the task concurrence effect shown by the no-shift group is primarily
due to changes in the capacity demands of the keypressing task. It is
worthwhile now to determine if the nature of the task concurrence
reduction effect is consistent with the concept of unitization as
addressed above. These considerations will also provide additional
evidence concerning the possible role of time sharing.

An important issue to consider is whether the appearance of
unitization is controlled by the constancy of sequence elements. In the
keyboard shift and sequence shift groups, the variation of sequence
elements and their motoric realizations should have prevented any
reliance on unitization that may have developed during practice. The
interaction of group, practice, and task concurrence showed that task
concurrence costs were reduced substantially for the no-shift group,
only slightly (and nonsignificantly) for the sequence shift group, and
not at all for the keyboard shift group. Thus, the effects of
operational constancy were consistent with a unitization account of the
findings. In connection with a time sharing view, it is apparent that if
a time sharing strategy was operating it did not survive the shift to a
new keyboard, as could reasonably be expected.

The results of the frequency manipulation provide additional

converging support of the unitization account. In the two groups where

90

frequency effects were predicted to be the strongest, the no-shift and
NS-NDT groups, task concurrence costs were greater for low-frequency
sequences than for high-frequency sequences. Thus, task concurrence
costs were seen to vary over specific sequences within a fixed level of
overall exposure to the task. In the serial position analysis, changes
in the rate of keypressing over practice for the L-6 sequences were
consistent with a trend toward the formation of an integrated motor
program. The overall analysis also revealed a trend for the keyboard
shift group to experience interference on high-frequency sequences,
which is also suggestive of the emergence of an integrated motor
program.

Taken together, the frequency data support the unitization account
in two ways. The first is that better performance on lowarequency
sequences in the keyboard shift group reflects difficulty in inhibiting
the more established motor patterns for high-frequency sequences. If
performance were not being mediated by motor programs prior to the shift
in keyboards, why should changing to new keyboards produce interference,
since only the motor patterns have changed? Secondly, the effect of
frequency facilitation is suggestive of the incremental build-up of
interassociations between elements of the keypressing sequence.

Additional evidence can be garnered from the L-6 serial position
analysis. The interaction of task concurrence and serial position showed
a much larger task concurrence cost at the position 3 "bump" than at the
other positions, suggesting that subjects may have been dividing the
sequences into two three-element halves, pausing at the completion of
the first half to prepare for the second half. The interactions with

practice and group indicated that the size of the position 3 "bump" was

91

markedly attenuated after practice for the no-shift group and less so
for the keyboard shift and sequence shift groups. These results imply
that the no-shift group transitioned from what could be called a
piecemeal performance strategy to performance more in line with the idea
of an integrated motor program. The overall increases in keypressing
speed, the smoothing out of variability in keypressing rate across
serial position, and the decrease of task concurrence costs at the
midpoint are all consistent with such an interpretation. Interestingly,
the data permit the speculation that time sharing may indeed have played
a role in performance at T1, with subjects using the break at position 3
to reinforce memory span digits on dual-task trials. This would be
consonant with the increased size of the task concurrence costs at
position 3. The pattern shown by the no-shift group over practice might
then reflect the abandonment of this demonstrably ineffective strategy
when the sequence elements were bound into a unified motor program.

Less direct evidence for unitization is presented by the sequence
length manipulation. The effects of sequence length were attenuated more
for the no-shift group than for the keyboard shift and sequence shift
groups, suggesting that practice with unchanging sequences and motor
mapping patterns is a necessary condition in the reduction of length
effects. The attenuation shown by the no-shift group indicates that
response times and keypressing rates came to be less dependent on the
number of elements in the sequences, again suggesting a trend away from
piecemeal treatment of sequence elements in performance. This pattern
must be regarded with more caution, however, because length effects at
T2 for the no-shift group were still relatively large, leaving open the

question of whether the length function would eventually flatten out. In

92

addition, there was no evidence that the reduction of the length effect
also related to reduction of task concurrence costs. While length
interacted with task concurrence in the manner that would be expected,
the higher order interaction with practice was not statistically
significant. While it could be argued that length effect reduction
reflects changes in the internal dynamics of the motor program that are
independent of capacity reduction, it is also possible that the
nonsignificance of the higher-order interaction is merely a type II
error. Pending a more powerful test, it should probably be regarded as
uninformative.

Taken as a whole, the data appear to converge on the proposition
that the nature of the task concurrence reduction effect shown in the
no-shift group is a result of changes in the memory demands of the
keypressing task. In turn, the changes in memory demand appear related
to the emergence of an integrated or "compiled" motor program whose
elements no longer need to be represented in working memory during
performance. This pattern is consonant with the conceptualization of
unitization.

With regard to time sharing and the more general class of factors
referred to previously as "optimization", there is little evidence that
such processes played a substantial role in the findings. The only
exception to this might be the "divide and conquer" strategy indicated
in the serial position analyses, which the data suggest was replaced by
unitization whenever possible. It should be noted, however, that a
suitably committed and imaginative theorist could still probably
construct an account of the results in terms of time sharing, though the

path might be tortuous. To do so would be to miss the point, however,

93

that the data are strikingly consistent with unitization at every turn.

It seems worthwhile to question, on the other hand, why the present
experiment failed to produce evidence of time sharing or optimization.
One consideration is to note the possibility of time sharing in the
serial position "bump" effect noted above. The tendency for the task
concurrence peak at the midpoint of the sequence to drop out later in
practice would suggest that the emergence of an integrated motor program
obviated the need to "take a break" from keypressing to reinforce the
memory span digits. Thus, the mechanism by which unitization was
demonstrated in this experiment may have worked against the continued
use of time sharing strategies. Such an interpretation suggests that in
cases where unitization is possible it may replace rather than augment
time sharing adaptations.

It is also useful to note that the present task is somewhat austere
relative to the more complex task environments (such as typing, playing
a musical instrument, or operating a control system) which are typically
viewed as a promising arena for the operation of time sharing factors
(Logan, 1985). In the present task, allowing only very restricted
practice at the digit span task with explicit instruction to use a
verbal rehearsal strategy, it is possible that spending any time at all
away from active rehearsal of the memory span digits may have produced
immediate detrimental effects in span performance. Thus, the task may be
viewed as being highly constrained in its temporal organization,
relative to more continuous and multifaceted tasks such as those
mentioned above. This sort of interpretation would be consistent with
the outcome from Logan’s (1979) experiments which also used digit span

performance as a secondary task, and which also failed to find any

94

evidence for time sharing. These considerations suggest that the
demonstration of time sharing effects might be best approached through
the study of a more temporally indeterminate task where the options for
perceptual processing, memory and control functions, and output
processes are more subject to strategic variation. At the same time, it
would be necessary for the task to be amenable to the same type of
analytic scrutiny possible in the present experiment to disentangle the
effects of time sharing from the effects of unitization or

automaticity.

Overview and Conclusions

 

Granted it has been established that the concept of unitization
provides a sensible and useful account of the results, what can be said
about the issues of attention and automaticity in skill acquisition? At
bottom, the answer lies in having pinpointed with confidence and
precision one particular way in which capacity demands in skilled
performance may be mediated with practice. Having identified one
plausible capacity reduction mechanism, with an accompanying model of
how, when, and why it works, we seem better prepared to advance a
general conceptual framework in which the plethora of remaining
questions can be framed and resolved.

The characteristics of the conceptual framework to be advanced flow
logically from the nature of the capacity reduction mechanism identified
in this research. In the present case, the formation of a unitized
representation of the elements in a performance sequence changed the
working memory requirements of performance, suggesting that the working

memory representation of elements in a performance sequence can be made

95

unnecessary. This appears to be made possible by a type of learning,
presumably involving the formation of direct interassociations between
the elements of the sequence, that enables the performance sequence to
be treated as a single unit. The process is analogous to the concept of
chunking (Miller, 1956), and could aptly be described as the chunking of
performance sequences. It is likely that this type of learning is
dependent on constancy in the order and identity of the components in
the sequence.

The underlying point in this development is that the concept of
unitization ought to be extended beyond the domain of motor
programming-to include the idea that mental operations which are not
directly tied to output processes may also be bound into unified
"programs" (Anderson, 1982; Carr, 1979; Reason, 1979). While it seems
plausible that central processing mechanisms should be organized in the
same general ways as motor planning and responding, it would also seem
worthwhile to examine current models of information processing in light
of the relationship between capacity demand and operational constancy
that has been revealed in the present research. At the same time, it
seems important to move beyond the overarching binary classification of
mental processes into "automatic" and "controlled" processes (Shiffrin &
Schneider, 1977; Schneider & Shiffrin, 1977), to a finer—grained
analysis of the mechanisms which allow "automatic" processes to occur
(see Logan, 1985).

One suggestion produced by this interpretation is that the
acquisition of skill in other tasks can be addressed in terms of the
effects of unitization and examined for the type of sequence constancy

needed for unitization to occur. Equipped with well founded hypotheses

96
regarding how and why capacity demands may be reduced, we are better
able to distinguish between the possibilities.

In a more general vein, the interpretation suggests that the
concept of limited capacity, along with the idea that different
processes may mediate the demands made upon it, ought not to be
abandoned in lieu of the more vaguely formulated alternatives coming
from multiple resource theory (Navon and Gopher, 1979; Allport, 1980),
and from the work of Neisser, Hirst, and Spelke (1981; Hirst, Spelke,
Reaves, Caharack, and Neisser, 1980). As outlined previously, assuming a
hybrid model of processing resources with both central and non-central
resources can accommodate the empirical challenges to the central
capacity view while providing at the same time a large advantage in
theoretical utility over the "Neisserian" approach, in which "skill" is
an unanalyzed explanatory construct rather than a phenomenon that is
itself to be explained. In the present case it has been proposed that
working memory, serving a temporary storage function for task-relevant
data, is a central resource likely to be required by tasks of diverse
natures. Having documented the Operation of a process which mediates
demand on this resource, it can be argued that the limited central
capacity concept still has considerable utility, so long as it is
augmented with a hybrid model of processing resources and is not too
simplistically conceived. This approach has the additional advantage of
preserving the link with our phenomenologically-based perceptions of
skill acquisition.

It is abundantly clear, however, that even if the preceeding points
are established and agreed upon, numerous difficult issues and problems

remain. Despite their apparent diversity, I would like to propose that

97

these problems boil down to a profound lack of resolution in two areas.
One lies in the nature of "resources" or "capacity"; in the inability to
specify or agree upon a functional basis for deciding what "resouces" do
and how "capacity" relates to the execution of the mental operations
which we presume to process information (Navon, 1984). The second, which
is closely related, is the lack of generality, coherence, and
completeness in our contemporary descriptions of mental operations
themselves.

An excellent illustration lies in Cheng’s (1985) criticism of the
"automatic vs. controlled processing" distinction of Shiffrin and
Schneider (1977; Schneider and Shiffrin, 1977). Cheng offers a
reinterpretation of the results presented in Shiffrin and Schneider (and
in related research) which they interpret as evidence of automatic
processing. In Cheng’s view, the improvements in performance seen under
consistent mapping conditions can be better explained by the concept of
"restructuring"-—a change in the task algorithm-~rather than in the
emergence of "automatic" processing. Such an argument implies a rather
strange definition of automaticity-a process which enables
capacity-free processing without changes in the information processing
algorithm by which performance is achieved. Finding evidence for that
type of automaticity would seem a losing proposition considering the
difficulty in imagining a process which changes capacity demands without
altering, at some level of description, the mental operations underlying
performance. The point is that arguments of this nature can proceed
indefinitely if our descriptions of information processing algorithms or
mental operations are not linked to the concepts of attention and

capacity. It is precisely in the interplay of the mental operations or

98

algorithms used to achieve performance and their respective demands for
temporary storage, perceptual processing, control, and output that
constitutes the area which most deserves systematic scrutiny in regard
to skill acquisition.

The present research has taken its cue from these considerations.
The effort has been to identify a particular function traditionally
associated with the concept of central resources and to track changes in
the demand for this function as a result of changes in the information
processing algorithm underlying performance. Additional research with
this strategy in mind (see also Anderson, 1982), geared toward the
identification of particular functions and specific mental Operations
which relate to them, may eventually disambiguate the confusion of

process and content which plagues this area.

LIST OF REFERENCES

Allport, D. A. Parallel encoding within and between elementary stimulus
dimensions. Perception and Psychophysics, 1971, 12, 104-108.

Allport, D. A. Word recognition in reading. In P. A. Kolers, M. E.
Wrolstad, & H. Bouma (Eds.), Processing_2£_Visible Language (Vol 1).
New York: Plenum Press, 1979. f

 

 

Allport, D. A. Patterns and actions: Cognitive mechanisms are content

specific. In G. Claxton (Ed.). Cognitive Psychology. London:
Routledge & Kegan Paul, 1980.

Allport, D. A., Antonia, B., & Reynolds, P. On the disproof of the
single channel hypothesis. ggarterly Journal 2£_Experimental
Psychology, 1972, 23, 225-235.

 

 

Anderson, J. R. Language, Memory, and Thought. Hillsdale, N. J.:
Lawrence Erlbaum Associates, 1976.

Anderson, J. R. Acquisition of cognitive skills. Psychological Review,
1982, 82, 369-406.

Atkinson, R. C. & Juola, J. F. Search and decision processes in
recognition memory. In D. H. Luce and P. Suppes (Eds.), Contemporary
Developments $3_Mathematical Psychology (Vol 1). San Francisco:
Freeman, 1974.

 

Atkinson, R. C. & Shiffrin, R. M. Human memory: A proposed system and
its control processes. In K. W. Spence and J. T. Spence (Eds.),
Advances $2 the Psychology 2£_Learning and Motivation: Research.2§d_
Theory (Vol 2): New York: Academic Press, 1968.

 

Baddeley, A. D. The Psychology 2f Memory. New York: Basic Books, 1976.

Baddeley, A. D. and Hitch, G. Working memory. In G. Bower (Ed.),
Recent Advances 32 Learning and Motivation (Vol 8). New York:
Academic Press, 1974.

 

Bahrick, H. P. and Shelly, C. Time sharing as an index of
automatization. Journal 2£_Experimental Psychology, 1958, 26,
288-293.

 

Broadbent, D. E. Perception and Communication. Pergamon Press, 1958.

Broadbent, D. E. Task combination and the selective intake of
information. Acta Psychologica, 1982, 59, 253-290.

Broadbent, D. E. The maltese cross: A new simplistic model for memory.
Behavioral and Brain Sciences, 1984, 8, 198-214.

 

99

100

Bryan, W. L. and Harter, N. Studies in the physiology and psychology of
the telegraphic language. Psychological Review, 1897, 4, 27-53.

 

Bryan, W. L. and Harter, N. Studies on the telegraphic language: The
acquisition of a hierarchy of habits. Psychological Review, 1899, 6,
344-375.

 

Carr, T. H. Consciousness in models of human information processing:
Primary memory, executive control, and input regulation. In G.
Underwood and R. Stevens (Eds.), Aspects of Consciousness (Vol 1).
London: Academic Press, 1979.

 

Chase, W. G. and Ericcson, K. A. Skill and working memory. In G. H.
Bower (Ed.), The Psychology of Learniog_and Motivation (Vol 16). New
York: Academic Press, 1982.

 

 

Chase, W. G. and Simon, H. A. The mind’s eye in chess. In W. G. Chase
(Ed.), Visual Information Processing, New York: Academic Press,
1973.

 

Cheng, P. W. Restructuring versus automaticity: Alternative accounts of
skill acquisition. Psychological Review, 1985, 22, 414-423.

Damos, D. L. and Wickens, C. D. The identification and transfer of time
sharing skills. Acta Psychologica, 1980, 32, 15-39.

 

Deutsch, J. A. and Deutsch, D. Attention: Some theoretical
considerations. Psychological Review, 1963, 19, 80-90.

 

Fitts, P. M. Perceptual-motor skill learning. In A. W. Melton (Ed.),
Categories 2£_Human Learning. New York: Academic Press, 1964.

 

Francolini, C. M. and Egeth, H. E. On the nonautomaticity of
’automatic’ activation: Evidence of selective seeing. Perception §_
Psychophysics, 1980, 21, 331-342.

 

 

Gopher, D., Brickner, M., and Navon, D. Different difficulty
manipulations interact differently with task emphasis: Evidence for
multiple resources. Journal 2£_Experimentsl Psychology: Human
Perception 222_Performance, 1982, 8, 146-157.

 

 

 

 

Gopher, D. and North, R. A. Manipulating the conditions of training in
time-sharing performance. Human Factors, 1977, 12, 583-593.

 

Graboi, D. Searching for targets: The effects of specific practice.
Perception é Psychophysics, 1971, 19, 300-304.

 

 

Griffith, D. and Johnston, W. A. Stage 2 processing and the divided
attention effect. Memory §_Cognition, 1977, 2, 597-601.

 

Hasher, L. & Zacks, R. T. Automatic and effortful processes in memory.
Journal o£_Bxperimental Psychology: General, 1979, 108, 356-388.

 

Hatano, G. & Inagaki, K. Two courses of expertise. Paper presented at

101

the Conference on Child Development in Japan and the United States.
April, 1983, Stanford Ca.

Henry, F. M. & Rogers, E. E. Increased response latency for complicated
movements and a "memory drum" theory of neuromotor reaction.
Research Quarterly of the American Association for Health, Physical
education, and Recreation, 1960, 31, 448-458. '———

 

 

 

 

 

Hirst, W. Is there a fixed limit on attentional capacity? Paper
presented at the American Psychological Association, Washington, D.
C., 1982.

Hirst, W., Spelke, E., Reaves, C., Caharack, G. & Neisser, U. Dividing
attention without alternation or automaticity. Journal of.
Experimental Psychology: General, 1980, 109, 98-117.

 

James, W. The Principles of Psychology (2 Vols). Dover Publications,
1950. (Originally published by H. Holt & Co., 1890)

 

Johnston, W. A. & Dark, V. J. In defense of intraperceptual theories of
attention. Journal of Experimental Psychology: Human Perception and
Performance, 1982, 8, _407-421.

 

 

 

Johnston, W. A., Greenberg, S. N., Fisher, R. P. & Martin, D. W.
Divided attention: A vehicle for monitoring memory processes.
Journal 2£_Experimental Psychology, 1970, 82, 164-171.

 

Jonides, J. Voluntary versus automatic control over the mind’s eye’s
movement. In J. Long & A. Baddeley (Eds.), Attention and Performance

 

IE, Hillsdale, N. J.: Lawrence Erlbaum Associates, 1981.

Jonides, J. & Gleitman, H. The benefit of categorization in visual
search. Perception §_Psychophysics, 1976, 29, 289-298.

 

 

Kahneman, D. Attention and Effort. Englewood Cliffs, N. J.:
Prentice-Hall, 1973.

 

Kahneman, D. & Wright, P. Changes of pupil size and rehearsal
strategies in a short-term memory task. anrterly Journal of
Experimental Psychology, 1971, 22, 187-196.

 

Kahneman, D. & Treisman, A. M. Changing views of attention and
automaticity. In R. Parasuraman & D. R. Davies (Eds.), Varieties of
Attention. New York: Academic Press, 1984.

Kantowitz, B. H. & Knight, J. L. Testing tapping timesharing. Journal
of Experimental Psychology, 1974, 103, 331-336.

 

Kantowitz, B. H. 8 Knight, J. L. Testing tapping timesharing II:
Auditory secondary task. Acta Psychologica, 1976, 29, 343-362.

 

Keele, S. W. Compatibility and time-sharing in serial reaction time.
Journal 2£_Experimental Psychology, 1967, 12, 529-539.

 

102

Keele, S. W. 8 Summers, J. J. The structure of motor programs. In G.
Stelmach (Ed.), Motor Control: Issues and Trends. New York: Academic
Press, 1976.

Kristofferson, M. W. Effects of practice on character classification
performance. Canadian Journal 2£_Psychology, 1972, 26, 54-60.

 

Kristofferson, M. W. The effects of practice with positive sets in a
memory scanning task can be completely transferred to a different
positive set. Memory 8 Cognition, 1977, 5, 177-186.

LaBerge, D. Acquisition of automatic processing in perceptual and
associative learning. In P. M. A. Rabbit 8 S. Dornic (Eds.),
Attention and Performance 2, New York: Academic Press, 1975.

 

LaBerge, D. Automatic information processing: A review. In J. Long 8
A. Baddeley (Eds.), Attention and Performance VI. Hillsdale, N. J.:
Lawrence Erlbaum Associates, 1977.

 

Lachman, R., Lachman, J. L., 8 Butterfield, E. C. Cognitive Psychology_
and Information Processipg: An Introduction. Hillsdale, N. J.:
Lawrence Erlbaum Associates, 1979.

 

 

 

Logan, G. D. Converging evidence for automatic perceptual processing in
visual search. Canadian Journal of Psychology, 1976, 32, 193-200.

 

Logan, G. D. Attention in character-classification tasks: Evidence for
the automaticity of component stages. Journal 2£_Experimental
Psychology: General, 1978, 107, 32-63.

 

 

Logan, G. D. On the use of a concurrent memory load to measure
attention and automaticity. Journal of Experimental Psychology:
Human Perception and Performance, 1979,5,189-207.

 

 

Logan, G. D. Short-term memory demands of reaction-time tasks that
differ in complexity. Journal of Experimental Psychology: Human
Perception and Performance, 1980, 6, 375-389.

 

Logan, G. D. Skill and automaticity: Relations, implications, and
future directions. Canadian Journal 2£_Psychology, 1985, 32,
367-386.

 

 

Lucas, M. 8 Bub, D. Can practice result in the ability to divide
attention between two complex tasks? Comment on Hirst et al. Journal
of Experimental Psychology: General, 1981, 110, 495-498.

 

MacKay, D. G. The problems of flexibility, fluency, and speed-accuracy
tradeoff in skilled behavior. Psychological Review, 1982, 82,
483-506.

 

Miller, G. A. The magical number seven, plus or minus two: Some limits
on our capacity for processing information. Psychological Review,
1956, 23, 81-97.

 

103

Miller, G. A., Galanter, E., 8 Pribram, R. H. Plans and the Structure
2£_Behavior. New York: Henry Holt 8 Co., 1960.

Navon, D. Resources-A theoretical soupstone? Psychological Review,
1984, 22, 216-234.

Navon, D. Attention division or attention sharing? Laboratory for

Information Processing and Decision Making, Report No. 20, January,
1985.

Navon, D. 8 GOpher, D. On the economy of the human processing system.
Psychological Review, 1979, 22, 214-255.

 

Neely, J. H. Semantic priming and retrieval from lexical memory: Roles
of inhibitionless spreading activation and limited-capacity
attention. Journal 22 Experimental Psychology: General, 1977, 106,
226-254.

 

Neisser, U. Cognitive Psychology. New York: Appleton, 1967.

 

Neisser, U., Hirst, W. 8 Spelke, E. S. Limited capacity theories and
the notion of automaticity: Reply to Lucas and Bub. Journal 22
Experimental Psychology: General, 1981, 110, 499-500.

 

Neisser, U., Novick, R., 8 Lazar, R. Searching for ten targets
simultaneously. Perceptual and Motor Skills, 1963, 21, 955-961.

 

 

Norman, D. A. Toward a theory of memory and attention. Psychological
Review, 1968, 12, 522-536.

Norman, D. A. 8 Rumelhart, D. E. Explorations 22_Cognition. New York:
W. H. Freeman 8 Co., 1975.

 

Paap, K. R. 8 Ogden, W. Letter encoding is an obligatory but
capacity-demanding operation. Journal 22 Experimental Psychology:
Human Perception and Performance, 1981, 1, 518-528.

 

 

 

Posner, M. I. Cumulative development of attentional theory. American
Psychologist, 1982, 21, 168-179.

 

Posner, M. I. 8 Snyder, C. R. R. Attention and cognitive control. In
R. L. 80130 (Ed.), Information Processing and Cognition. Hillsdale,
N. J.: Lawrence Erlbaum Associates, 1975.

Reason, J. Actions not as planned: The price of automatization. In G.
Underwood 8 R. Stevens (Eds.), Aspects of Consciousness (Vol. 1).
London: Academic Press, 1979.

 

Regan, J. E. Involuntary automatic processingin color-naming tasks.
Perception §_Psychophysics, 1978, 25, 130-136.

 

 

Regan, J. E. Automaticity and learning: Effects of familiarity on
naming letters. Journal o£_Experimental Psychology: Human Perception
and Performance, 1981, 1, 180-195.

 

 

104

Rosenbaum, D. A., Saltzman, E., 8 Kingman, A. Choosing between movement
sequences. In S. Kornblum 8 J. Requin (Eds.), Preparatory States and
Processes. Hillsdale, N. J.: Lawrence Erlbaum associates, 1984.

Schneider, W. Attention, learning, and the transition from controlled
to automatic processing. Paper presented to the annual conference of
the Psychonomic Society, San Antonio, Texas, November, 1984.

Schneider, W. 8 Fisk, A. D. Concurrent automatic and controlled visual
search: Can processing occur without resource cost? Journal 22
Experimental Psychology: Learning, Memory, and Cognition, 1982, 2,

261-278.

Schneider, W. 8 Shiffrin, R. M. Controlled and automatic human
information processing I: Detection, search, and attention.
Psychological Review, 1977, 23, 1-66.

Shallice, T. Dual functions of consciousness. Psychological Review,
1972, 12, 383-393.

Shiffrin, R. M., Dumais, S. T., 8 Schneider, W. Characteristics of
automatism. In J. Long 8 A. Baddeley (Eds.), Attention and
Performance 12, Hillsdale, N. J.: Lawrence Erlbaum Associates, 1981.

Shiffrin, R. M. 8 Schneider, W. Controlled and automatic human
information processing II: Perceptual learning, automatic attending,
and a general theory. Psychological Review, 1977, 22, 127-190.

Stanovich, K. E. 8 West, R. F. The effect of sentence context on
ongoing word recognition: Tests of a two-process theory. Journal 22
Experimental Psychology: Human Perception and Performance, 1981, 1,

658-672.

Sternberg, S., Monsell, S., Knoll, R. L. 8 Wright, G. E. The latency
and duration of rapid movement sequences: comparisons of speech and
typewriting. In G. E. Stelmach (Ed.), Information Processing 22_
Motor Control 222 Learning. New York: Academic Press, 1978.

Treisman, A. M. Verbal cues, language, and meaning in selective
attention. American Journal 22 Psychology, 1964, 11, 206-219.

Treisman, A. M. 8 Riley, J. G. A. Is selective attention selective
perception or selective response? A further test. Journal 22
Experimental Psychology, 1969, 12, 27-34.

Trumbo, d. 8 Noble, M. Secondary task effects on serial verbal
learning. Journal 22 Experimental Psychology, 1970, 22, 418-424.

Waugh, N. C. 8 Norman, D. A. Primary memory. Psychological Review,
1965, 12, 89-104.

APPENDICES

A1
APPENDIX A
Cell Means (ms) in the Overall Analysis From

T1 and T2 Including the No-Shif t, Keyboard
Shift, and Sequence Shift Groups.

 

 

TOTAL TIME
Single-Task Dual-Task
High Low High Low
Ereggengy ‘ Erggggngg Erggggngy Ergguengy
We
Time 1
Length=6 4772 4984 6417 5771
Length-4 4929 5113 6624 6399
Length-2 3412 3448 4065 4286
Time 2
Length=6 2149 2598 2754 3388
Length=4 2021 2574 2603 3249
Length-2 1745 1871 2040 2285
r hif
Time 1
Length-=6 4874 5147 6054 5789
Length-4 4646 5014 5907 5898
Length-2 3336 3456 3602 3732
Time 2
Length-6 3377 3344 4554 4148
Length-4 3388 3139 4572 3916
Length-2 2178 2270 2497 2694
h' u
Time 1
Length-=6 5242 4955 6090 6129
Length-4 4938 4766 5432 6027
Length-2 3643 3313 3929 3526
Time 2
Length=6 4805 4943 5090 5920
Length-4 3894 3676 4547 4766
Length-2 2727 2747 3027 2959

A2

APPENDIX A

Cell Means (ms) in the Overall Analysis From
T1 and T2 Including the No-Shif t, Keyboard
Shift, and Sequence Shift Groups.

TIME TO RESPOND

 

 

Single-Task Dual-Task
High Low High Low
Frequency Frggggncy Frggpgngy Ereggengy
- h' r
Time 1
Length-6 2607 2847 3900 3422
Length-4 2774 3063 4095 3697
Length-2 1551 1746 2132 2367
Time 2
Length-6 1071 1302 1450 1699
Length-4 975 1300 1411 1772
Length-2 733 884 939 1109
KW
Time 1
Length=6 2529 2823 3511 3124
Length-4 2582 2784 3487 3478
Length-2 1631 1734 1762 1998
Time 2
Length-6 1660 1744 2475 2302
Length-4 1846 1594 2735 2121
Length-2 1150 1167 1347 1434
W
Time 1
Length-=6 2826 2780 3413 3577
Length-4 2512 2454 2821 3236
Length-2 1689 1559 1945 1612
Time 2
Length-6 2621 3056 2858 3464
Length-4 2141 1963 2671 2602
Length=2 1386 1462 1649 1566

A3

APPENDIX A

Cell Means (ms) in the Overall Analysis From
T1 and T2 Including the No-Shif t, Keyboard
Shift, and Sequence Shift Groups.

FLUENCY
_SinslsiasL— w—
High Low High Low
Emem Bream manna! W
Egg-Shift Group
Time 1
Length-6 433 428 503 470
Length-4 431 410 506 540
Length-2 . 372 340 387 384
Time 2
Length-6 216 259 261 338
Length-4 209 255 238 295
Length-2 202 197 220 235
WW2
Time 1
Length-6 469 465 508 533
Length-4 413 446 484 484
Length-2 341 344 368 347
Time 2
Length-6 344 320 416 369
Length-4 308 309 367 359
Length-2 206 221 230 252
WW2
Time 1
Length-6 483 435 535 510
Length-4 485 462 522 558
Length-2 391 351 397 383
Time 2
Length-6 437 377 446 491
Length-4 350 343 375 433

Length-2 268 257 276 279

A4

APPENDIX A

Cell Means (ms) in the Overall Analysis From
T1 and T2 Including the No-Shif t, Keyboard
Shift, and Sequence Shift Groups.

KEYPRESSING ERROR RATES
(percent error)

1 1 -T Qual-Iask
High Low High Low
15W n Em n mam Emma
- ° r u
Time 1
Length=6 25.0 15.6 29.7 15.6
Length-4 12.5 9.4 14.1 14.1
Length-2 4.7 3.1 4.7 7.8
Time 2
Length-6 6.3 7.8 6.2 12.0
Length-4 3.1 4.7 6.2 10.9
Length-2 1.6 6.2 0.0 1.6
W
Time 1
Length-6 12.5 12.5 12.5 26.6
Length-4 14.1 9.4 25.0 20.3
Length-2 3.1 7.8 1.6 3.1
Time 2
Length-6 12.5 12.5 15.6 12.5
Length-4 9.4 9.4 18.8 17.2
Length-2 3.1 1.6 9.4 0.0
Magnesium
Time 1
Length-6 23.4 18.8 26.6 37.5
Length-4 23.4 12.5 10.9 20.3
Length-2 0.0 1.6 3.1 1.6
Time 2
Length-6 26.6 25.0 25.0 40.6
Length-4 7.8 20.3 14.0 14.0

Length-2 7.8 4.7 6.2 3.1

A5

APPENDIX A

Cell Means (ms) in the Overall Analysis From
T1 and T2 Including the No-Shift, Keyboard
Shift, and Sequence Shift Groups.

DIGIT SPAN PERFORMANCE
(percent error)

 

Single-Task Qggl-jljask
High Low High Low
Erma W mama imam
Egg-Shift Group
Time 1
Length=6 -- -- 42.4 35.5
Length=4 -- -- 33.4 32.4
Length=2 -- -- 23.6 24.6
Time 2
Length-6 -- -- 13.1 17.6
Length-4 -- -- 15.2 24.2
Length-=2 -- -- 15.0 1 1.3
hi r
Time 1
Length-6 -- -- 34.2 36.9
Length-4 -- -- 27.5 27.1
Length=2 -- -- 25.4 22.5
Time 2
Length-=6 -- -- 22.3 22.5
Length-4 -- -- 15.6 15.8
Length-2 -- -- 12.1 14.5
W
Time 1
Length=6 -- -- 44.9 49.8
Length-4 -- -- 27.3 33.0
Length=2 -- -- 24.8 24.6
Time 2 .
Length-6 -- -- 34.0 31.2
Length-4 -- -- 23.8 26.2

Length-2 -- -- 13.3 15.2

A6

 

 

 

 

 

-. .. mm~ F-m oo~ .. .. o- com -~
nnw Fw~ moe eoe osw —o~ pe~ Po~ new ee~
.. -. on~ noN oo~ -. .. m—~ op~ now
oo~ -~ awn paw om~ no: no~ own «mm npw
.. .. Fne oeo woe .. .. o~n use see
cmm com com one epm epn ooe eon nhe eon
.. .. owe onm —-e .. .. can one one
own Nmn ~00 one oeo h- omn eae mom eme
m e n m P m e w Ila. F
comumooa domcom co_u_aom domcom
.mmmmrawam xmm».o.ummm
oumwccaocoulmmm»

nonconuom end men on; gov momxduc< co_u_moa .o_eom one c_ “use «coo: ..ou

a x-ozmma<

ESE

ouguucom
>ocoavoem :04
euguucom
ousuucoa
>ocoacocm gum:
~ m:.:
euguucoa
ouguucoa
xucoacocm so;
enzymes;
essence;
>ucoaaocm sum:
P m:_h
amuse uwmcm.mz

A7

 

 

 

.. .. e8 8.. an
E in 2... 2m 8m
.. .. 2... 2n an
.2 2n 2.. a... 03
.. .. Sn a: 83
t... 2.. .8 as one
.. .. .5 a. 83
88 3. .2 2m 3..
m .e n lul 11.
smuwsm dleOm
.. . .
0

 

88.260» en. a:- 0.... .3. 3.31.2 5.33.. .380» 2.» c. ARV 83: ...eu

 

 

 

wo~
Em

OON
mom

~me
«me

Nme
Noe

 

 

. . . . 8... on...
8. o2. 2... Sn
. . . . 3... an
2... 8n can 2...
. . . . 8... n3
«8 «2 Sn tn
. . . . Sn 2.
EN .3 2e 2..
m e A J
S p u mat a. — L8
L3 no...

u gonna“?

enzyme“...
01398..
>§cm so.

.1523
02:98..
>828... 5:.
N u:.:

.7598.
9.580.
>829: so.

.1298.
8585..
sucoauotm gu_=
. wx:

mmmumlmuflmwuuummmxmm

118

.. .. Nmn 0mm ewe .. .. eo~ nun chm euguuco.

 

 

 

 

 

 

om~ oen ooo onn mom —m~ nn~ eoe nnn ode ongum:o.
5:260...— 30..
.. .. can hon son .. .. n- as» oen ensumco.
o- mne own ose «on o~n pan ~eo nun woe ouguo:o.
6:26»... :3:
N u:.h
.. .. m~m soe ouo -. .. nne pom ome enguoco.
osm Noe o~m nee mph nwn 5mm Nae ~ee Fpm onguuco.
5:269... 33
.. .. epe omm eem .. .. Non o~o -e eazucco.
wen ~oe oen epm own s~n eon eoe nae F~o ouguu:o.
5:269: .32.
F wt.»
mmmco uo_gm «mammmmm
n e n N — w e n ~ 9
cavemaom .omcom comupooa .auuom
.mmHuqqaauuunnuuuuun muqfluuqqaqmnuuuuuuulu
oommuumeMMImmm»

acne-zoom en. we. on. 88. .o.>.ac< no.8..oa .a...» «a. c. .8-. .coox ..ou

a x_oxmam<

A9

.. .. neN o~n ~n~ .. .. m- oNN n- eunumco.

 

 

 

 

 

 

 

-~ wem nan onn -n mo~ oe~ me~ men oo~ onzumco.
>ocoaoocm mo.
.. .. oe~ onN oom .. .. an o- om~ enzum:o.
mo~ Pow uo~ NAN se~ mo~ om~ PmN mom sen ouguu:o.
>ucoacoem sum:
~ wt:
-. .. .. -. .. .. .. .. .. .. 1588..
.. .. .. -. .. .. .. .. .. .. 8583
xucoscocm so.
.. .. .. .. .. .. .- .. .. .. 1583
.. .. .. .. .. .. .. .. .. .. oufiuco.
>u:o:uo:: cum:
9 us.»
.— 52.9.
m e n M. P m e n. N F
:o.u_aom .umcom :o_u_aom .o.:om
8112. 8.41.1: .8738:

 

 

o ccaoco an»

moocoavom en. u:o on. :0. uon>.o:< :o_ummom .a_com ecu :_ Ammo 0:00: ..ou

a x—azmaa<

A10
APPENDIX C
Cell Means in the T2 Analyses Including the

NS-NDT Group With the No-Shift, Keyboard
Shift, and Sequence Shift Groups.

TOTAL TIME
(m8)
Single-1351; Dual-lash
High Low High Low
Emma Emuma Emma Emana
N - i
Length-6 2149 2598 2754 3388
Length-4 2021 2574 2603 3249
Length-2 1745 1871 2040 2285
Kahcardihiftfitan
Length-6 3377 3345 4554 4148
Length-4 3388 3139 4572 3916
Length-2 2178 2270 2497 2694
n hi r
Length-6 4805 4943 5090 5920
Length-4 3894 3676 4547 4677
Length-2 2727 2747 3027 2959
W
Length-6 2494 2877 2666 3245
Length-4 2186 2398 2519 2750

Length-2 1803 2076 1998 2174

A11

APPENDIX C

Cell Means in the T2 Analyses Including the
NS-NDT Group With the No-Shif t, Keyboard

TIME TO RESPOND
(m8)

hktfihflliinuuz

Length-6
Length-4
Length-2

liexhganiIEUILIEuuuz
Length-6

Length-4
Length-2

Sequencejfluftiircun
Length-6

Length-=4
Length-2

NS-EDI lem
Length-6

Length=4
Length-2

Shift, and Sequence Shift Groups.

1 Jr
High Low
Elenueasx. 11:92:29!

1071 1302
975 1301
733 884
1660 1744
1846 1594
1150 1167
2621 3056
2141 1963
1386 1462
1152 1523
980 1185
732 883

_____IMHUiIaflt_____
High Low
125992298 Elenueacx
1450 1699
1412 1772
939 1110
2475 2302
2735 2121
1347 1434
2858 3464
2671 2602
1649 1566
1379 1639
1270 1401
880 944

A12
APPENDIX C
Cell Means in the T2 Analyses Including the

NS-NDT Group With the No-Shif t, Keyboard
Shift, and Sequence Shift Groups.

FLUENCY
(mS)
511181942515 Dual-113515
High Low High Low
mpg! 11993290931 W91 Exam
- i r
Length-6 216 259 261 338
Length-4 209 255 238 296
Length-2 202 197 220 235
W
Length-6 344 320 416 369
Length-4 308 309 367 359
Length-2 206 221 230 252
MW
Length=6 437 377 446 491
Length-4 350 343 375 433
Length-2 268 257 276 279
W
Length-6 268 271 257 321
Length-4 241 243 250 270

Length-2 214 238 224 246

A13

APPENDIX C

Cell Means in the T2 Analyses Including the
NS-NDT Group With the No-Shif t, Keyboard

Shift, and Sequence Shift Groups.

KEYPRESSING ERROR RATES
(percent error)

-h' r

Length-6
Length-4
Length-2

hif r
Length-6
Length-4
Length=2
hif r u
Length=6
Length-4
Length-2
N -N r
Length-6

Length-4
Length-2

__Sinalcfiasls—_
High Low
11991150931 W n
6.2 7.8
3.1 4.7
1.6 6.2
12.5 12.5
9.4 9.4
3.1 1.6
26.6 25.0
7.8 20.3
7.8 4.7
14.0 7.8
4.7 4.7
0.0 1.6

21131-13 SK
High Low
Mm 15299151191

6.2 12.0
6.2 10.9
0.0 1.6
15.6 12.5
18.8 17.2
9.4 0.0
25.0 40.6
14.0 14.0
6.2 3.1
17.2 20.3
10.9 17.2
0.0 0.0

A14
APPENDIX C

Cell Means in the T2 Analyses Including the
NS-NDT Group With the No-Shif t, Keyboard
Shift, and Sequence Shift Groups.

DIGIT SPAN PERFORMANCE
(percent error)

Single-leek Deal-Task
High Low High Low
Fregeeney Eregeeney Fregeeney Fregeeney
N - h' r
Lengths-6 -- -- 13.1 17.6
Length=4 -- -- 15.2 24.2
Length=2 -- -- 15.0 11.3

K r hif r

Length-6 -- -- 22.3 22.4

Length=4 -- -- 15.6 15.8

Length-:2 -- -- 12.1 14.5
W

Length-=6 -- -- 34.0 31.2

Length-4 -- -- 23.8 26.2

Length-2 -- -- 13.3 15.2
354112]: Green

Length-6 -- -- 32.6 28.3

Length-4 -- -- 24.2 28.3

Length-2 -- -- 17.8 18.8