Professional Documents
Culture Documents
Springer-Verlag 1999
ORIGINAL ARTICLE
Introduction
Mental arithmetic is an important everyday skill of
many adults. Solving simple arithmetic problems (such
as 7 6 ?) is also a key component of elementary
education. Research so far has revealed two major determinants of performance: the organization of simple
arithmetic facts in long-term memory and the processing
of the information in working memory. The rst topic
has attracted many researchers and has been documented fairly extensively (e.g., Anderson, 1983). The
role of working memory in arithmetic, however, has
received much less attention, and studies have mainly
focused on dicult problems, such as 435 287 or
13 18 13 21 13. The objective of the present
study was to further explore the role of working memory
in simple arithmetic.
Most researchers agree that arithmetic facts are
stored in an interrelated network in long-term memory
73
74
(1996) still hold when the sums are combined with the smallest split
possible (+1) and when they are combined with a bigger but not
``extreme'' split (+5). All the sums were controlled for the number
of carries: the c term was, for all three kind of sums, a number from
10 to 19 (carry 1). Because of this, no negative splits were included. If we had worked with both a positive split (e.g.,
8 4 17) and a negative split (e.g., 8 4 7), the carries would
have no longer been controlled. Just like in the study of Lemaire et
al., the numbers 0 and 1 were omitted because there is evidence that
these problems (e.g., x + 0 or x + 1) are not solved by retrieving
the solution directly from memory but instead by retrieving rules
(e.g., x + 0 x) that guide their solution (see Ashcraft, 1982;
Baroody, 1985).
The sums were further controlled as follows: (1) half of the
correct results of the sums were even, the other half were odd; (2)
for half of the sums, the rst number was bigger than the second
(a > b), for the other half the inverse was true (a < b), and (3) the
c term was never the product of a b, in other words, the associative-confusion/interference eect could not play a role in the
verication of the sums.
In this way, 24 combinations of the form a + b were formed. In
one condition, these were once presented with a split of +1, once
with a split of +5, and twice with the correct solution. As a result,
every series consisted of 96 (4 24) trials, namely 48 (2 24) true
and 48 (2 24) false single sums. Before the presentation of each
series, the sequence of all sums was randomized. Only in less than
0.5% of the trials was the sum presented the same as the one before.
Procedure and design. The stimuli were presented horizontally in
the center of a computer screen, in yellow with a black background.
The equations remained on until the participant responded, unless
there was still no response after 10 s. The participants were instructed to solve the sums as accurately and as fast as possible by
pressing the appropriate key. The left and the right button of the
mouse were designated as true and false. For half of the participants, the left button was designated as true, and for the other half,
it was designated as false. All participants were instructed to use
their forenger and their middle nger of the right hand to press
these keys. The inter-trial interval was 1 s.
Each participant participated under every condition, in contrast
to Lemaire et al. (1996), where the participants participated under
only one condition. There were four conditions: control, articulatory suppression, random letter generation, and random time interval generation.
In the control (CON) condition the participants solved the sums
without a secondary task. The articulatory suppression (AS) condition required the participants to say ``the'' (``de'' in Dutch) aloud
and quickly while they were solving the sums. This secondary task
was meant to load the phonological loop, and only the phonological loop. In contrast of Lemaire et al. (1996), the participants
did not say ``the'' every 2 s, but did so continuously and without
stopping. This modication was introduced because of the reason
mentioned before.
In the random letter generation (RLG) condition, the participants were required to say one random letter of the alphabet at a
rate of one letter per second while they were solving the sums. The
rate was indicated by a metronome that continued throughout the
series. The participants were instructed to avoid stereotypical sequences (e.g., ``a-b-c-d'' or ``o-p-q-r'') or spelling out words (e.g.,
``c-a-t''). Thus, the standard version of RLG was used in this study,
in contrast to the investigation of Lemaire et al. (1996), where a
modied version was used, namely one random letter of the series
``a-b-c-d-e-f'' at a rate of one per two seconds. Despite the already
mentioned drawback, this secondary task was introduced because
(a) the results of this condition should serve as a baseline for the
fourth condition, (b) it is one of the most frequently used tasks to
load the central executive, and (c) it would maintain the compatibility with the study of Lemaire et al.
The random time interval generation task (Vandierendonck et
al., 1998) was used as a fourth condition; the participants were
asked to tap an unpredictable rhythm on the zero key of the
numeric keypad while they were solving the sums. They were
instructed to use their left forenger and were told that the rhythm
had to be as random and unpredictable as possible. This task was
meant to load the central executive and not to interfere with one of
the slave systems, as in the previous secondary task. In order to
obtain a measure of randomness, the tap sequences of the participants were registered.
Four conditions imply 24 possible sequences of these conditions
(4! 24). The rst 24 participants each participated in one of these
sequences. The sequences for the next 16 participants were selected
at random, but with the restriction that no sequence was executed
by more than two participants. As a result, each sequence was
executed by (approximately) the same number of participants, and
possible eects of learning or boredom were eliminated. The participants were permitted a 3-min rest period between the conditions.
At the start of the experiment, the participants solved 16 random practice trials (8 true sums, 4 split 1, 4 split 5) in order to
familiarize themselves with the apparatus, the procedure, the
stimulus display, and the response keys. After each practice trial,
the participants received feedback: according to the answer given,
the text ``correct answer'' or ``wrong answer'' was given at the
bottom of the screen for 1 s. After these practice trials, no more
feedback was given, and the already mentioned sequence of conditions was followed. Each series started with a xation point (``!'')
in the middle of the screen that remained for 500 ms. In the
experimental conditions, the participants were rst required to
practise the concurrent task until they felt comfortable with it
and showed no apparent problems. They were told that is was
important not to stop performing the secondary task while they
were solving the sums. If they stopped, which happened only once,
this condition was rerun all over. The participants were tested
individually in a quiet room. Each experimental session lasted
approximately 2535 min.
A 4 (Load: CON, AS, RLG, RIG) 3 (Sum: true, split +1,
split +5) within-participants design was used. This design was
preferred to a between-participants design in order to eliminate
dierences in arithmetic skills between the participants. Thus, every
participant solved three kinds of sums under each condition. As a
result, all participants solved exactly 400 sums [16 practice trials + (4 96)] experimental trials.
Results
The data were analyzed by means of a multivariate
analysis with contrasts between the 12 dependent variables: the mean latencies per participant under the
combinations of condition and kind of sum, or the
proportion of correct responses per participant under
the combinations of condition and kind of sum. This
analysis conforms to the suggestions of McCall and
Appelbaum (1973) for a correct analysis of repeatedmeasures designs.
Randomness analyses
Every sequence of random taps1 was analyzed by means
of the method described by Vandierendonck et al. (1998)
and more in detail in Vandierendonck (1998) in order to
1
75
nd out whether participants complied with task instructions. In essence, random time intervals can be
converted into a series of binary events. To that end, the
complete time course is subdivided into a sequence of
xed intervals, each of which either contains an event (a
keypress) or does not. By means of appropriate statistics, the degree of statistical independence and the tendency to alternate (or to perseverate, if such arose) can
be estimated. Six participants deviated from randomness. The data analyses reported below were performed,
with and without these six participants. As there were no
dierences in the pattern of results, the analyses on the
complete data are reported here. By means of these
analyses, a median-split was also introduced: the participants were divided into a group with the 20 most
random sequences and a group with the 20 least random
sequences. The latencies and the proportions of correct
answers of these two groups were analyzed and compared. As there were no dierences in the pattern of
results between the two groups, this analysis will not be
reported.
Latencies
The trials in which the participants made errors were
dropped from the analysis. The mean latencies and their
standard deviations are presented in Table 1. The main
eect of condition was signicant, F(3,37) 89.44,
p < .001, as was the main eect of kind of sum,
F(2,38) 41.61, p < .001, but their interaction was
not: F(6,34) 1.64, p > .15. In contrast to the control
condition, AS had no eect, F(1,39) 1.38, p > .20, but
RLG and RIG did have an eect: F(1,39) 206.07,
p < .001 and F(1,39) 34.21, p < .001, respectively.
Also, the eect of RIG was signicantly dierent from
AS, F(1,39) 41.65, p < .001, and from RLG,
F(1,39) 195.77, p < .001.
Across all conditions, there was a signicant split
eect, i.e., the sum with split 5 (weighted M=1628
ms) were solved signicantly faster than sums with split
=1 (weighted M=1833 ms): F(1,39) 68.67, p < .001.
The latencies of true sums (weighted M 1588 ms) were
signicantly shorter than the latencies of sums with
split 1, F(1,39) 66.95, p < .001, but did not dier
Table 1 Mean latencies (in ms) per condition and per kind of
sum.a Standard deviations are given between parentheses.
AS articulatory suppression; RLG random letter generation;
RIG random time interval generation
True
Split 1
Split 5
a
Control
AS
RLG
RIG
1203
(315)
1455
(349)
1183
(281)
1260
(363)
1481
(422)
1239
(356)
2552a
(695)
2778a
(787)
2670a
(726)
1415a
(396)
1659a
(412)
1466a
(426)
0.94
(0.05)
0.87
(0.08)
0.98
(0.03)
AS
RLG
a
0.92
(0.06)
0.87
(0.10)
0.97
(0.03)
0.86
(0.11)
0.83a
(0.11)
0.93a
(0.07)
RIG
0.93
(0.06)
0.86
(0.09)
0.95a
(0.08)
Discussion
The aim of the present study was to explore whether the
conclusions of Lemaire et al. (1996) still hold if some
important modications are introduced. Other stimuli
76
References
Anderson, J. R. (1983). A spreading activation theory of memory. Journal of Verbal Learning and Verbal Behavior, 2, 261
295.
Ashcraft, M. H. (1982). The development of mental arithmetic: A
chronometric approach. Developmental Review, 3, 213236.
Ashcraft, M. H., & Battaglia, J. (1978). Cognitive arithmetic: Evidence for retrieval and decision processes in mental addition.
Journal of Experimental Psychology: Human Learning and
Memory, 4, 527538.
77
Ashcraft, M. H., & Fierman, B. A. (1982). Mental addition in
third, fourth, and sixth graders. Journal of Experimental Child
Psychology, 3, 216234.
Baddeley, A. D. (1986). Working memory. Oxford: Oxford University Press.
Baddeley, A. D. (1990). Human memory: Theory and practice.
Hove: Erlbaum.
Baddeley, A. D., & Hitch, G. J. (1974). Working memory. In G.
Bower (Ed.), Recent advances in learning and motivation (Vol.
6). Hillsdale, NJ: Erlbaum.
Baroody, A. J. (1985). Mastery of the basic number combinations:
Internalization of relationships or facts? Journal for Research in
Mathematics Education, 16, 8398.
Campbell, J. I. D., & Graham, D. J. (1985). Mental multiplication
skill: Structure, process, and acquisition. Canadian Journal of
Psychology, 39, 338366.
Evans, F. J. (1978). Monitoring attention deployment by random
number generation: An index to measure subjective randomness. Bulletin of the Psychonomic Society, 12, 3538.
Groen, G. J., & Parkman, J. M. (1972). A chronometric analysis of
simple addition. Psychological Review, 79, 329343.
Hitch, G. J. (1978). The role of short-term working memory in
mental arithmetic. Cognitive Psychology, 10, 302323.
Krueger, L. E. (1986). Why 2 2 5 looks so wrong: On the oddeven rule in product verication. Memory & Cognition, 14, 141
149.
LeFevre, J., & Kulak, A. G. (1994). Individual dierences in the
obligatory activation of addition facts. Memory & Cognition,
22, 188200.
Lemaire, P., & Siegler, R. S. (1995). Four aspects of strategic
change: Contribution to children's learning multiplication.
Journal of Experimental Psychology: General, 124, 8397.
Lemaire, P., Barrett, S. E., Fayol, M., & Abdi, H. (1994). Automatic activation of addition and multiplication facts in ele-