Memory & Cognition

/999.27 (5). 79/-802

Division by multiplication
JAMIE I. D. CAMPBELL
University ofSaskatchewan, Saskatoon, Saskatchewan, Canada

In two experiments, item-specific transfer was examined in simple multiplication and division with
prime and probe problems separated by four to six trials. As was predicted by Rickard and Bourne's
(1996) identical-elements model, response time (RT) savings were larger with identical (e.g., prime
63 -7- 7, probe 63 -7- 7) than with inverted (63 -7- 9 and 63 -7- 7) division problems, whereas identical (7 X 9
and 7 X 9) and inverted (9 X 7 and 7 X 9) multiplication problems produced equivalent transfer.
Nonetheless, there was statistically significant transfer between inverted division problems. Further-
more, RT savings in the multiplication-to-division transfer conditions (e.g., prime 7 X 9, probe 63 -7- 7)
indicated that multiplication mediated large-number division problems. These latter effects are not
predicted by the identical-elements model but may be reconciled with the model by distinguishing as-
sociative transfer (facilitation owing to strengthening of a common problem node in memory) from me-
diated transfer (facilitation owing to mediation by a strengthened, related problem). Skilled adults can
exploit the conceptual correspondences between multiplication and division facts in a highly efficient
way to facilitate performance.

Over the last 20 years, research into the cognitive arith- fer of practice between multiplication and division prob-
metic skills of educated adults has focused almost exclu- lems. Specifically, they gave adults extensive practice on
sively on modeling the memory processes that subserve subsets ofsimple multiplication and division problems and
number fact retrieval (see Ashcraft, 1992, 1995, for ex- then tested performance on the same problems (identical
cellent reviews). This focus was motivated by the largely repetition), problems with operand order reversed (order
unquestioned assumption that skilled adults rely more or change), corresponding problems in the other operation
less exclusively on direct retrieval to solve single-digit (operation change), and "new" unpracticed control prob-
problems in the four operations (see Baroody, 1994). Re- lems. They observed transfer of practice (i.e., response
cently, however, researchers increasingly have found ev- time [RT] savings) in the order change condition for mul-
idence that educated adults employ a variety ofprocedural tiplication (e.g., 7 X 8 and 8 X 7) that was nearly as strong
strategies (e.g., counting, decomposition, transforma- as that with identical repetition. They found no transfer be-
tion, etc.), even for elementary problems such as 6 + 9 tween the two orders of division problems (e.g., 56 -:- 7 =
or 7 X 6 (Geary & Wiley, 1991; Lefevre, Bisanz, et al., 8 and 56 -i- 8 = 7), however, and little evidence that prac-
1996; Lefevre, Sadesky, & Bisanz, 1996). Adults' exten- tice transferred in either direction between correspond-
sive knowledge of conceptual and functional relation- ing division and multiplication problems. If such transfer
ships among problems and operations apparently can primarily measures strengthening of retrieval processes,
provide a relatively efficient procedural basis for arith- these results imply that people's performance of corre-
metic performance. sponding division and multiplication problems did not
This article examines the relation between adults' per- depend on a common retrievalprocess. Consistent with this,
formance of simple division and multiplication problems, Cipolotti and de Lacy Costello (1995) described an acal-
such as 6 X 8 = 48 and 48 -7- 6 = 8. The central issue is the culic patient whose multiplication memory was largely
extent to which the two operations are interdependent. intact, but who nonetheless presented severely impaired
One possibility is that corresponding problems in the two division performance.
operations are represented and retrieved as independent On the basis of their transfer findings, Rickard et al.
facts in memory. In support of this, Rickard and Bourne (1994; Rickard & Bourne, 1996) proposed an identical-
(1996; Rickard, Healy, & Bourne, 1994) found little trans- elements model of asymptotic memory for multiplication
and division facts. According to the model, each memo-
rized number fact (e.g., 6 X 8 = 48) is represented by a
This research was funded by a grant from the Natural Sciences and
memory node that consists of the two problem operands
Engineering Research Council of Canada. The author thanks Mark (e.g., 6 and 8), the answer (e.g., 48), and the arithmetic
Ashcraft, Morton Gernsbacher, Jo-Anne Lef-evre, and Jane Zbrodoff operation (e.g., multiply). A central theoretical claim of
for helpful comments on a previous version of this manuscript. The au- the model is that problems that have the same problem
thor also thanks Andrea Ludtke, who coordinated data collection for operands are represented by a common node. Hence, mul-
the present study. Correspondence concerning this article should be ad-
dressed to 1. I. D. Campbell, Department of Psychology, University of tiplication problems that differ only in operand order, such
Saskatchewan, 9 Campus Drive, Saskatoon S7N 5A5, Canada (e-mail: as 6 X 8 and 8 X 6, access the same node. In contrast,
[email protected]). problems that differ with respect to even one element ac-

791 Copyright 1999 Psychonomic Society, Inc.

cess different nodes. Thus, inverse division problems ticipants frequently used multiplication as a check after
(e.g., 48 -;- 6 and 48 -;- 8) are represented by different direct retrieval of the quotient. The multiplication check
nodes. Because practice strengthens only the node cor- on division trials would prime later multiplication errors,
responding to the practiced problem, transfer occurs be- as was observed (i.e., division to multiplication error
tween the two orders of multiplication problems, but priming). In contrast, multiplication trials would not be
there is no transfer between the two orders of division expected to prime division errors, because incorrect di-
problems and no transfer between corresponding multi- vision trials reflect errors of direct retrieval and division
plication and division facts. is not used to check multiplication.
Although the transfer data presented by Rickard et al. This explanation raises an obvious question, however.
(1994; Rickard & Bourne, 1996) clearly support the If adults often use multiplication to mediate or check di-
identical-elements model, the cognitive relation between vision, why did Rickard et al. (1994) and Rickard and
division and multiplication may often be more involved Bourne (1996) not observe transfer between division and
than the model implies. Typically,multiplication facts are multiplication? There were at least two salient method-
learned in school before division, and children apparently ological differences between Campbell's (1997) error-
rely initially on their knowledge of addition and multipli- priming experiment and Rickard et al.'s transfer studies
cation to solve division problems (Geary, 1994; Siegler, that could be important. First, in Campbell's (1997) study,
1988). These conditions could promote a system in which error priming was observed over a range of only about 10
division often is mediated by multiplication. This is con- trials, whereas Rickard et al. tested transfer after longer
sistent with the impaired multiplication and division per- delays (e.g., in separate test blocks at the end of practice
formance of a patient with brain injury, described by or after a I day layoff from practice). The error-priming
Hittmair-Delazer, Semenza, and Denes (1994). The 45- analysis may have measured short-term transfer effects
year-old patient presented parallel patterns of loss and that are not observed at longer delays. Second, the Rickard
preservation for corresponding multiplication and divi- et al. and Rickard and Bourne studies involved many
sion problems, and rehabilitation of multiplication trans- more repetitions ofproblems than did Campbell's (1997)
ferred to division. Hittmair-Delazer et al. concluded that study. During practice, participants received each problem
the patient's division performance depended on the avail- 40 times in the Rickard et al. study and 90 times in the
ability of the corresponding multiplication fact. Rickard and Bourne study. Each repetition presumably
Campbell (1997) also found evidence that adults' per- would increase the probability of direct retrieval the next
formance of simple division involved multiplication. In time the problem was encountered. Indeed, Rickard et al.
this study, university students were given division prob- (1994, p. 1149) suggested that the evidence of indepen-
lems that alternated with multiplication problems under dent processes for division and multiplication might have
instructions for speeded responses. RTs were highly cor- partially reflected learning within their experiments. In
related for corresponding division and multiplication the Campbell (1997) study, there were only three rep-
problems, and error characteristics indicated parallel re- etitions per item; consequently, that experiment poten-
trieval structures. Specifically, division errors were con- tially provided a better measure of the dependencies be-
strained by semantic distance in divisor units (i.e., the tween division and multiplication that existed prior to the
distance between the presented dividend and the product experiment.
implied by the error), rather than by distance from the cor-
rect quotient. This indicates that memory for division is EXPERIMENT 1
organized in terms of multiplicative relationships and is
consistent with the hypothesis that division was based on To pursue these issues, the following experiment mea-
memory for multiplication facts. Analysis of intertrial sured transfer between division and multiplication within
error priming (see Campbell & Clark, 1989), however, 72-trial blocks comprised of 36 prime-probe problem
complicated the picture. By comparing errors with the pairs. The prime-probe pairs represented identical repe-
answers or problems encountered during the 10 trials pre- tition, order change, and operation change conditions for
ceding an error, Campbell (1997) showed that multipli- both simple division and multiplication problems. The
cation errors (e.g., 7 X 9 = 56) were primed by previous two problems in each pair were separated by a number of
multiplication trials (8 X 7 = 56) and also by previous di- intervening trials. The experiment included short prime-
vision trials (56 -i- 7 = 8). In contrast, whereas division probe lags of 4, 5, or 6 trials and relatively long lags of
errors (56 -;- 7 = 9) were primed by previous division trials 50 trials, on average. Transfer is measured by comparing
(63 -i- 9 = 7), division errors were not primed by previous probe performance with performance in the appropriate
multiplication trials. prime condition. The following paragraphs outline pre-
To account for asymmetrical error priming (i.e., divi- dictions for transfer associated with different functional
sion to multiplication priming, but not multiplication to relations between division and multiplication.
division priming), Campbell suggested that the perfor-
mance of corresponding division and multiplication Simple Direct Retrieval
problems involved separate memory representations, as People may solve division by direct retrieval, with
is assumed in the identical-elements model, but that par- multiplication playing no role whatsoever. Ifwe assume,
 DIVISION BY MULTIPLICATION 793

as in the identical-elements model, that corresponding corresponding multiplication problems. In this case, we
division and multiplication facts are not mediated by a expect transfer in both directions, given that division is
common node and that multiplication is not used in any mediated by multiplication.
fashion on division trials, no transfer between division and Rickard and Bourne (1996, p. 1284) found equivalent
multiplication in either direction is expected. transfer effects for small and large problems; nonetheless,
in looking for evidence of mediation, it may be important
Direct Division Retrieval With to separately examine performance for small- and large-
a Postretrieval Check by Multiplication number problems. In general, small-number problems
Although division and multiplication facts may be (e.g.,2 X 3,6 -;- 2) are answered more quickly and ac-
represented independently and usually answered by direct curately than larger problems (6 X 8, 48 -;- 6), the so-
retrieval, people may have relatively low confidence in called problem-size effect. Small-number problems may
their division skill and use multiplication to check the re- be better learned because they are practiced more often
sult of direct division retrieval (i.e., the multiplication (Ashcraft, 1995; Ashcraft & Christy, 1995; Geary, 1996),
check occurs after the division response has been given). although it is likely that other factors contribute to the
In the experimental context, this would insert a covert effect (Campbell, 1995; LeFevre, Bisanz, et aI., 1996;
practice trial involving the corresponding multiplication LeFevre, Sadesky, & Bisanz, 1996; Zbrodoff, 1995). For
problem at that point in the trial sequence. The covert re- example, on the basis of retrospective verbal protocols,
trieval potentially would strengthen the multiplication LeFevre, Sadesky, and Bisanz (1996) proposed that
node and produce savings the next time that node was small-number problems are more likely to be solved by
accessed. Because multiplication is typically the better direct retrieval than are large-number problems, which
learned operation, multiplication would be used to check have a higher probability ofsolution by slower procedural
division, but division is not likely to be used to check mul- strategies (e.g., transformation, decomposition). Simi-
tiplication. Consequently, the covert transfer effect is ex- larly, small-number divisions may be more likely to be
pected to be asymmetrical: As Campbell (1997) observed solved by direct retrieval, whereas large-number problems
for intertrial error priming, if multiplication is used only may be relatively more likely to be mediated. To exam-
to check division, division trials should prime or produce ine this in the present experiment, transfer was analyzed
savings for multiplication, but multiplication trials are separately for small-number and large-number problems.
not expected to transfer to division. No multiplication-to-
division transfer would occur, because the multiplication Method
check is assumed to occur after direct retrieval of the di- Participants. Thirty volunteers (13 females, 17 males) were re-
vision fact and, therefore, multiplication would not con- cruited from the participant pool operated by the Department of
tribute directly to division performance. Psychology or through advertisements posted around the Univer-
sity of Saskatchewan. The advertisement described the experiment
as a study of simple arithmetic skills. The participants ranged in
Division Mediated by Multiplication age from 18 to 47 years (M = 26.2), reported normal or corrected-
Ifpeople sometimes use multiplication to check divi- to-normal vision, and were paid $6 for their participation.
sion, presumably it is because their division memory is Apparatus. Stimuli appeared on two high-resolution monitors
relatively weak. This suggests, however, that if division using an IBM-type personal computer, with one monitor being
memory was sufficiently weak, people would actually viewed by the experimenter and the other by the participant. Stim-
solve division problems by retrieving the corresponding uli appeared horizontally as white characters against a dark back-
ground. The participant sat approximately 50 cm from the monitor
multiplication fact (Rickard et aI., 1994, p. 1149). As an and wore a lapel microphone that activated a relay switch connected
example, 63 -i- 7 could be solved by using the division to the computer's serial port. The sound-activated relay controlled
problem as a cue to retrieve the multiplication fact that 7 a software clock that was accurate to ± I msec.
X 9 = 63 or 9 X 7 = 63. The nonmatching operand, 9, Stimuli and Design. Stimuli were the multiplication problems
would then be reported as the quotient. Because division composed of pairs of arabic digits between 2 and 9 (i.e., 2 X 2
performance would depend on the accessibility ofthe cor- through 9 x 9) and the corresponding division problems (4 -;- 2
responding multiplication fact, previous practice that fa- through 8\ -i- 9). The two operands in a problem were separated by
the operation sign ( -i- or X) with adjacent spaces. There are 36 pos-
cilitates retrieval ofthat multiplication fact (e.g., a recent sible pairings of the numbers 2 through 9 when commuted pairs
trial involving that multiplication problem) would transfer (e.g., 3 x 8 and 8 X 3) are counted as one problem. The set of 36
to division. Thus, in contrast to the checking hypothesis, includes 8 multiplication tie problems involving a single repeated
the mediation hypothesis predicts transfer from multi- operand (e.g., 2 x 2,5 X 5) and 28 multiplication nontie problems
plication to division. Whether or not transfer is observed (e.g.,2 X 3,5 X 7). There are 8 corresponding division ties (4 -;- 2,
in the other direction (i.e., from division to multiplication) 25 -i- 5) and 28 division nonties (6 -;- 2,35 -;- 7). Digits were 7 mm
high and 3 mm wide, and problems were approximately 15 mm in
depends on how the multiplication fact is processed dur- total width.
ing mediation. If the covert multiplication process medi- Each participant received 432 transfer tests involving prime-probe
ating division produces strengthening of the multiplica- problem pairs separated by some number of intervening trials. The
tion node, we would expect transfer from division to relation between the prime and probe defined six transfer conditions:
794 CAMPBELL

(I) identical multiplication prime and probe (e.g., prime 6 X 9, the Arabic numbers I through 40, presented individually in a ran-
probe 6 X 9), (2) identical division prime and probe (e.g., prime 54 dom order. This provided an opportunity to adjust the sensitivity of
-7 6, probe 54 -7 6), (3) multiplication prime and probe with operand the sound-activated relay. Testing required about I h.
order changed (e.g., prime 6 X 9,probe9 x 6),(4) division prime and
probe with order changed (e.g., prime 54 -7 9, probe 54 -7 6), (5) op-
eration change with a division prime and multiplication probe (e.g., Results and Discussion
prime 54 -7 9, probe 6 X 9), and (6) operation change with a mul- The overall mean RT for correct trials in Experiment 1
tiplication prime and division probe (e.g., prime 6 X 9, probe 54 -7 was 912 msec, and the overall error rate was 11.9%. Sep-
6). Both short and long transfer distances were tested. Short trans- arate four-way repeated measures analyses of variance
fer distances were prime-probe lags of4,5, or 6 trials. Long trans- (ANOVAs) of transfer in the short- and long-lag condi-
fer distances were prime-probe lags of 50, 51, and between 30 and tions were performed. The factors included problem type
70 trials. The latter, variable long-lag condition corresponds to an
average lag of about 50 trials. Tie problems were included in all con- (prime or probe), transfer condition (identical repetition,
ditions to simplify programming but were excluded from analysis order change, or operation change), operation (division or
because the within-operation manipulation of operand order is irrel- multiplication), and problem size (small or large). Prob-
evant for ties. lems with a dividend or product greater than 25 were de-
The participants received 12 blocks of 72 problems, for a total of fined as large (Campbell, 1997). This split creates two
864 trials. Each 72-trial block included 36 multiplication trials and sets, each containing 14 nontie problems when commuted
36 division trials, with I prime-probe pair being based on each of
the 36 number pairs (e.g., 6 X 9,9 X 6, 54 -7 6, and 54 -7 9 are
pairs (e.g., 4 X 8 and 8 X 4 or 32 -;- 4 and 32 -;- 8) are
based on the pair 6,9). Across the 12 blocks, each of the 36 num- counted as one problem. A total of 155 RTs were spoiled
ber pairs was tested in each ofthe six transfer conditions, using both by failures of the voice-activated relay (0.8%), and an-
a short and a long transfer distance. Within a block, there were 6 other 289 (1.4%) were discarded as outliers more than
prime-probe pairs in each of the six transfer conditions, and three three SDs from each participant's grand mean for each
transfer conditions were tested with short and three with long dis- operation X size combination. Pairwise deletion of miss-
tances.In the following block, the assignment of transfer conditions
to short and long lags was reversed. Thus, all 12 transfer condition X
ing RT data (i.e., owing to errors, spoiled trials, or outliers)
lag combinations were tested six times in each pair of successive ensured for each participant that the mean prime RT and
odd- and even-numbered blocks. The frequencies of the six transfer mean probe RT in each operation X problem size X
conditions in odd- and even-numbered blocks was balanced across transfer cell were based on the same set of problems (a
participants, as was the assignment of the six transfer conditions to mean of 11.7 problems out of a maximum of 14 over all
specific short lags (i.e., 4, 5, or 6 trials) and long lags (i.e., 50, 51, conditions). Matched deletion of missing RTs could not
and between 30 to 70 trials). Over all trials, operand order (i.e., 3 X
be extended to include the operation or transfer condition
6 vs. 6 X 3; 18 -7 3 vs. 18 -7 6) was random with respect to trans-
fer condition, but across all conditions, all the problems were tested factors, because it resulted in a prohibitive depletion of
exactly six times in each order. To avoid repeating sequences of observations; consequently, different operation X transfer
problems or answers, the sequencing program avoided more than cells were not necessarily based on identical problem sets.
two successive transfer tests involving the same lag. Within these For tests ofoperation change transfer, the yoked prime and
constraints, the order of conditions across blocks for each problem probe problems involved different operations. Therefore,
and the order of problems in each block were independently ran- to obtain a baseline for operation change, each operation
domized for each participant.
Procedure. The participants were tested individually with an ex-
change probe trial was matched with the corresponding
perimenter present. Instructions described the experiment as a test operation change prime trial from the other operation.
ofspeed ofsimple numerical skills in which the task was to state the In all the following analyses, small problems produced
correct answer as quickly as possible. The participants were also significantly faster RTs and fewer errors than did large
advised that occasional errors are normal with speeded responding, problems, as was expected. Similarly, multiplication al-
and the experimenter occasionally reminded the participants to re- ways was significantly faster and less error prone than
spond as quickly as possible. The emphasis on speed was intended
to generate error rates sufficient to measure transfer in terms of
division, replicating previous research (e.g., Campbell,
error savings, in addition to RT savings. 1997; Rickard et al., 1994). Overall, small and large prob-
The experimenter initiated each block of trials. Each block con- lems produced RT means of 855 and 969 msec, respec-
sisted of a continuous sequence of trials with nothing to differenti- tively, and multiplication and division produced means of
ate primes and probes. For each trial, a fixation dot appeared at the 865 and 958 msec. Error rates for small and large prob-
center of the screen and flashed off-on-off over a l-sec interval. lems were 7.6% and 16.1%; for multiplication and divi-
The problem appeared on what would have been the next flash with
sion, they were 8.1% and 15.6%. Significance tests for
the operation sign (-7 or X) at fixation. Timing began when the
stimulus appeared and ended when the response triggered the main effects of problem size and operation are omitted
sound-activated relay. When a sound was detected, the displayed throughout. All other effects were significant at p < .001,
problem was replaced instantly by the fixation dot. This allowed the unless otherwise indicated, and only effects withp < .05
experimenter to detect premature or delayed triggering of the relay are reported.
and to mark such trials as spoiled. The I-sec fixation dot began to Short-lag probes versus short-lag primes: Re-
flash as soon as the experimenter entered the subject's answer; thus, sponse times. Table 1 presents mean RT for correct re-
the response-stimulus interval was usually slightly longer than
I sec. No feedback was provided regarding either speed or accu- sponses in the short-lag conditions as a function of op-
racy. The participants were offered a brief rest period between eration, problem size, problem type (prime or probe), and
blocks and were reminded about the importance of speed. Prior to transfer condition. The table also shows mean savings (i.e.,
the arithmetic task, the participants named as quickly as possible prime RT - probe RT) and the standard error of mean
 DIVISION BY MULTIPLICATION 795

Table 1
Mean Correct Response Times (in Milliseconds)
in the Short-Lag Conditions of Experiment 1
Division Probes Multiplication Probes
Identical Order Operation Identical Order Operation
Repetition Change Change Repetition Change Change
Problem Type M SEM M SEM M SEM M SEM M SEM M SEM
Small Problems
Short-lag prime 922 945 906 815 816 828
Short-lag probe 840 907 925 773 770 807
Savings +82 14.7 +38 21.2 -19 17.1 +42 12.2 +46 13.4 +21 18.6
Large Problems
Short-lag prime 1,029 1,049 1,062 960 965 936
Short-lag probe 914 992 1,002 887 885 941
Savings +115 22.1 +57 32.3 +60 28.7 +73 17.2 +80 18.7 -5 20.2

savings, The average RT savings on short-lag probes, rel- the corresponding interaction, however, did not reach
ative to primes, was 49 msec [887 vs. 936 msec; F( I,29) = conventional significance levels [F(I,29) = 2.72, MSe =
49.85, MSe = 8,745]. Savings, however, varied across 8,774,p = .11]. Thus, in contrast to the predictions of the
transfer condition [F(2,58) = 11,50, MSe = 62,306], with identical-elements model, there was not clear statistical
mean savings of78, 55, and 14 msec for identical repe- evidence that transfer with order change differed for di-
tition, order change, and operation change conditions, vision and multiplication.
respectively. These effects were qualified, however, by a The analysis of operation change transfer demonstrated
significant four-way interaction [F(2,58) = 3.18, MSe = a three-way interaction of operation, problem type, and
5,143, p = ,049]. The four-way effect was decomposed problem size [F(I,29) = 6.33, MS e = 6,687, p = .018].
by performing separate operation X size X problem type Thus, the four-way interaction in the overall analysis oc-
ANOVAs at each level of the transfer factor (i.e., identi- curred mainly because there was a three-way interaction
cal repetition, order change, operation change). Only in the operation change condition but not in the identical-
significant main or interaction effects involving problem repetition or order change conditions. As is shown in
type (i.e. savings) are reported, Table I, the three-way interaction occurred because in-
The analysis of identical repetition indicated signifi- teroperation transfer was obtained only for the large di-
cant size X type [F( I,29) = 4.45, MSe = 3,325, p = .044] vision problems [mean savings of60 msec; t(29) = 2.09,
and operation X type [F(I,29) = 7.70, MSe = 3,300,p = p = .044]. The average savings in the other three size X
.0 I0] interactions. Specifically, savings from identical operation combinations was - I msec, and none of the
repetition were greater for large (94 msec) than for small combinations approached significance (smallest a= .26).
problems (62 msec) and greater for division (98 msec) The evidence of transfer from multiplication to division
than for multiplication (57 msec). The large-number is not expected according to the identical-elements model
problems and division problems were associated with rel- and is consistent with the hypothesis that large division
atively long RTs and, therefore, may simply offer more problems were mediated by multiplication in the short-
room for speed-up. lag transfer condition.
In the order change analysis, the average savings of Short-lag probes versus short-lag primes: Per-
55 msec was reliable [F(I,29) =24.16, MSe = 7,623], and centage of errors. Table 2 presents the mean percentage
there were no significant interactions involving problem of errors in each cell of the four-factor analysis. There
size or problem type (smallest a= .19). The failure to ob- were significantly fewer errors on probe (10.9%) than on
serve an interaction between operation and problem type prime (12.8%) problems, which confirmed savings on
is potentially at odds with the identical-elements model, errors [F(I,29) = 10.20, MSe = 67.93,p = .003], but there
which predicts transfer with order change for multipli- were no significant interactions involving problem type
cation, but not for division. Savings with an order change (smallest a=.14). Nonetheless, an examination of Table 2
were significant both for division [48 msec; t (29) = 2.17, indicates that error savings for division were due mainly
p = .039] and for multiplication [63 msec; t(29) = 5.23]. to large problems with identical repetition, whereas both
Nonetheless, for division, transfer with order-change small and large multiplication problems tended to pro-
(48 msec) appeared to be substantially smaller, relative duce fewer errors, given identical repetition, as did large
to identical repetition (98 msec), whereas transfer with multiplication problems with order change. The 12 error
order change in multiplication (63 msec) was essentially savings means in Table 2 and the 12 RT savings means in
equivalent to identical repetition (57 msec). The test of Table I were correlated [.53;p = .035, one-tailed]. Thus,
796 CAMPBELL

the pattern of error savings generally corresponded to tween the two orders of division problems and to pursue
the pattern of RT savings, although RT provided a more the operation X order change interaction predicted by the
sensitive measure of savings. identical-elements model. Experiment 2 also provided an
Long-lag conditions. Mean RTs for the long-lag opportunity to replicate the finding of asymmetrical in-
probes were significantly slower (917 msec), relative to teroperation transfer (i.e., from multiplication to divi-
their yoked primes [901 msec; F(l,29) = 13.91, MS. = sion, but not from division to multiplication).
3,201]. The apparent transfer costs in the long-lag RT data In Experiment 2, an attempt was also made to redress
indicate overall faster RTs at the beginning of blocks the baseline complications arising from long-lag primes
(i.e., where long-lag primes were most common), relative occurring near the beginning and their yoked primes oc-
to the end ofblocks (i.e., where long-lag probes were most curring toward the end of each block. For Experiment 2,
common). This might occur as a result of the experi- prime-probe distances in the long-lag conditions were
menter's reminding the participants between blocks to reduced to an average lag of 30 trials, as compared with
respond as quickly as possible. Fatigue and accumulating the average of 50 trials in Experiment I. This might re-
interference from preceding trials may also tend to slow duce the chances of different speed-accuracy criteria's
performance toward the end of each block. Regardless of affecting the long-lag primes and probes.
the cause, the generally slower RTs for the long-lag probes, Finally, multiplication RTs in Experiment I were quite
relative to the long-lag primes, compromises interpreta- fast, as compared with division RTs. If multiplication
tion of the long-lag data with respect to item-specific RTs were closer to their asymptote (i.e., their RT floor), it
transfer. Because the short-lag primes and probes were might have been more difficult to measure small transfer
distributed more evenly throughout each trial block and effects from division to multiplication RTs. This could
were separated only by four, five, or six trials, it is very contribute to the asymmetrical interoperation transfer
unlikely that the short-lag primes and probes involved (i.e., multiplication-to-division only). For Experiment 2,
systematically different speed-accuracy criteria. the experimenter reduced the emphasis on overall speed,
Discussion of Experiment 1.The results ofthe short- relative to Experiment I, by monitoring error rates and
lag conditions were quite straightforward. Large-number cautioning the participants if error rates substantially ex-
multiplication problems produced savings on corre- ceeded 10%. Reduced speed pressure ought to increase the
sponding division problems, but there was no transfer in sensitivity of the procedure for measuring RT transfer in
the reverse direction. This pattern is consistent with mul- both multiplication and division.
tiplication mediating large-number division. There also
was evidence of transfer between the two orders of divi- EXPERIMENT 2
sion problems. Neither ofthese effects is predicted by the
identical-elements model, but they do not necessarily Method
contradict the model. Although the model only predicts Thirty volunteers (17 males, 13 females) were recruited in the
transfer from the strengthening of a common memory same way as that in Experiment I and attended a single session last-
node, transfer would also be observed if performance on ing about I h. The participants ranged in age from 17 to 29 years
a given problem was mediated by retrieving a related (M = 21.9), reported normal or corrected-to-normal vision, and
problem and if recent practice has improved the accessi- were paid $6 for their participation. Except for reducing the long-
lag distances between primes and probes, the design of Experi-
bility of the mediating fact. As is discussed in more de- ment 2 was identical to that of Experiment I. In this experiment,
tail in the General Discussion section, such a mediation long-lag distances were trial lags of29, 31, or somewhere between
process may underlie both the multiplication-to-division 25 and 35 trials (i.e., an average of30), as compared with an aver-
transfer and the cross-order division-to-division transfer age of 50 trials in Experiment I. As in Experiment I, short transfer
observed. distances were prime-probe lags of 4,5, or 6 trials. The mean trial
Of course, we would expect especially strong transfer numbers for long-lag primes, long-lag probes, short-lag primes, and
short-lag probes were, respectively, 20.4, 50.5, 33, and 38. Out of
when the prime and the probe involve a common node. sight of the participant, the experimenter's video monitor displayed
Consequently, even allowing for mediated transfer, the a running total of errors per block. Initial instructions emphasized
identical-elements model predicts stronger transfer with speed, as in Experiment I, but the experimenter did not intervene
identical repetition than with an order change for division to comment about speed unless the participant's error rate in a block
(i.e. because the two orders of division problems are as- deviated substantially above or below 10%.
sumed to involve separate nodes), whereas multiplication-
to-multiplication transfer should be approximately equiv- Results and Discussioll
alent for identical-repetition and order change (i.e., because The data from Experiment 2 were analyzed in the same
commuted pairs of multiplication problems are assumed way as those in Experiment I. A total of 324RTs were
to access a common node). Although the data appeared to spoiled by failures of the voice-activated relay (1.6%),
present this pattern, the test of the corresponding inter- and another 314 (1.6%) were discarded as outliers more
action was not statistically significant. One purpose ofEx- than three SDs from the grand mean for each operation X
periment 2 was to replicate the evidence of transfer be- problem size condition. The mean number of items con-
 DIVISION BY MULTIPLICATION 797

Table 2
Mean Errors (%) in the Short-Lag Conditions of Experiment 1
Division Probes Multiplication Probes
Identical Order Operation Identical Order Operation
Repetition Change Change Repetition Change Change
Problem Type M SEM M SEM M SEM M SEM M SEM M SEM
Small Problems
Short-lag prime 11.4 15.0 13.3 4.3 2.1 3.1
Short-lag probe 10.2 13.8 13.3 0.9 2.1 2.1
Savings + \.2 \.8 + \.2 \.7 0.0 2.2 +3.3 0.9 0.0 0.9 + \.0 1.1
Large Problems
Short-lag prime 18.8 20.2 19.3 15.2 16.4 15.0
Short-lag probe 13.6 2\.2 17.1 I\.4 12.1 12.6
Savings +5.2 2.1 -\.O 2.3 +2.2 2.8 +3.8 2.1 +4.3 2.0 +2.4 \.9

tributing to each RT mean was 12 problems out ofa max- primes, was 59 msec [994 vs. 1053 msec; F(I,29) = 63.62,
imum of 14 over all conditions. The overall mean RT for MSe = 9,749]. As in Experiment I, short-lag RT savings
correct trials in Experiment 2 was 1,024 msec, and the were greater for large problems (84 msec) than for small
error rate was 9.8%. Relative to Experiment I, the overall problems [33 msec; F(l ,29) = 15.37, MSe = 7,713]. Sav-
mean RT in Experiment 2 was 112 msec longer ( 12%), and ings also varied as a function oftransfer condition, as is in-
the overall error rate was reduced by about 20%. Over- dicated by the significant transfer X problem type inter-
all, in Experiment 2, small and large problems produced action [F(l,29) = 17.99, MSe = 5,937]. Specifically, the
RT means of941 and I, I06 msec, respectively, and mul- mean savings with identical repetition were 101 msec,
tiplication and division produced means of 980 and with order change it was 59 msec, and with operation
1,067 msec. Error rates for small and large problems were change it was 17 msec. There were no other significant ef-
5.4% and 14.2%, and for multiplication and division, fects (all Fs < I). To decompose the transfer X problem
error rates were 8.4% and 11.2%, respectively. type interaction and to present analyses that parallel those
As in Experiment I, the comparison of long-lag prime presented for Experiment I, separate operation X prob-
and probe RTs in Experiment 2 showed costs (i.e., prime lem size X problem type ANOVAs were performed at
RTs were faster then probe RTs). Thus, despite moving the each level of the transfer factor (i.e., identical repetition,
long-lag primes and probes closer together, the tendency order change, and operation change). Only significant ef-
for the participants to respond more quickly early in blocks fects involving problem type (i.e., savings) are reported.
than later in blocks apparently still contaminated the Analysis of the identical-repetition conditions con-
long-lag data. Because ofthis, there will be no further dis- firmed a main effect of problem type [mean savings of
cussion of the long-lag transfer conditions. 101 msec; F(l,29) = 129.07, MSe = 4,723]. Savings from
Short-lag probes versus short-lag primes: Re- identical repetition were greater for large (136 msec) than
sponse times. Table 3 presents mean RT in the short-lag for small problems [66 msec; F(I,29) = 1l.l5, MSe =
conditions as a function ofoperation, problem size, prob- 6,539, p = .002], but unlike Experiment I, there was no
lem type (i.e., primes vs. probes), and transfer condition. evidence that identical repetition benefited division more
The average RT savings on short-lag probes, relative to than multiplication (F < I). The reduced speed pressure

Table 3
Mean Correct Response Times (in Milliseconds)
in the Short-Lag Conditions of Experiment 2
Division Probes Multiplication Probes
Identical Order Operation Identical Order Operation
Repetition Change Change Repetition Change Change
Problem Type M SEM M SEM M SEM M SEM M SEM M SEM
Small Problems
Short-lag prime 1,010 1,011 990 906 921 907
Short-lag probe 926 986 1,009 858 863 906
Savings +84 18.0 +25 15.4 -19 25.0 +48 17.7 +58 15.8 +1 14.5
Large Problems
Short-lag prime 1,215 1.193 1,164 1,119 1,110 1,089
Short-lag probe 1,073 1,120 1,107 990 1,032 1,063
Savings +142 22.2 +73 29.0 +57 25.0 +129 2\.5 +78 27.4 +26 25.1
798 CAMPBELL

Table 4
Mean Errors (%) in the Short-Lag Conditions of Experiment 2
Division Probes Multiplication Probes
Identical Order Operation Identical Order Operation
Repetition Change Change Repetition Change Change
Problem Type M SEM M SEM M SEM M SEM M SEM M SEM
Small Problems
Short-lag prime 8.8 9.3 10.2 2.9 2.4 2.1
Short-lag probe 5.5 8.1 9.0 1.9 2.1 2.9
Savings +3.3 I.7 +1.2 1.9 +1.2 2.0 +1.0 0.9 0.3 0.9 -0.7 1.3
Large Problems
Short-lag prime 15.9 15.7 12.9 16.7 18.3 15.7
Short-lag probe 9.8 15.0 14.3 11.4 11.2 13.3
Savings +6.2 1.9 +0.7 2.6 -1.4 2.2 +5.3 1.6 +7.1 1.3 +2.4 2.6

(and generally longer RTs) in Experiment 2 than in Ex- for large than for small problems [mean savings of3.4%
periment 1 may have provided more equal opportunities vs. 1.0%; F(l ,29) = 5.39, MSe = 45.75, P = .027], but as
for the two operations to demonstrate RT savings. Table 4 shows, this effect arose mainly from multiplica-
In the order change condition, the average savings of tion trials [F(1,58) = 9.72, MS e = 27.15,p = .004). Error
59 msec was significant [F(l,29) = 18.27, MSe = 11,395], savings also varied as a function oftransfer condition: For
and, as in Experiment 1, there were no significant inter- identical repetition, order change, and operation change,
actions involving problem type or operation (smallest respectively, the mean error savings were 3.9%, 2.3%, and
a = .16). RT savings with an order change occurred both 0.3% [F(2,58) = 4.00, MS e = 48.04, P = .024]. Table 4
for division [49 msec, t(29) = 2.80, p = .009] and for shows that error savings for division occurred for both
multiplication [68 msec, t(29) = 3.91, P = .001). Thus, small and large problems with identical repetition, whereas
Experiment 2 confirmed transfer between the two orders there was no evidence of error savings in other division
of division problems over relatively short lags, an effect conditions. For multiplication, error savings were ob-
not predicted by the identical-elements model. The iden- served for large problems, both with identical-repetition
tical-elements model also predicts stronger transfer for and order change primes. The 12 error savings means in
identical repetition than for order change for division, but Table 4 and the 12 RT savings means in Table 3 were cor-
not for multiplication. An ANOVA that included both related [.69; p = .006, one-tailed). Thus, as in the short-
identical-repetition and order change conditions (but ex- lag conditions of Experiment 1, the pattern of error sav-
cluded operation change) did not confirm the expected ings generally corresponded to the pattern ofRT savings,
operation X problem type X transfer interaction although, again, RTs apparently provided a more sensitive
[F(l,29) = 1.78, MSe = 8,210,p = .19). measure.
The operation change analysis indicated a two-way
interaction between problem type and problem size EXPERIMENTS 1 AND 2 COMBINED
[F(l,29) = 6.55, MS e = 5,908, p = .016). As Table 3
shows, this represents larger savings for large problems The short-lag conditions in the two experiments were
(42 msec) than for small problems ( -9 msec). The three- practically identical; consequently, the two data sets were
way interaction among operation, problem size, and combined into a single set, to provide a maximally sen-
problem type was not significant (F < 1). Nonetheless, sitive analysis oftwo theoretically important issues: trans-
Table 3 indicates that, as in Experiment 1, only the large fer with order change and transfer with operation change.
division problems showed significant savings with an As the pattern of error savings parallelled the pattern of
operation change [57 msec; t(29) = 2.28,p = .015, one- RT savings in both experiments, the combined analysis
tailed]. The average savings in the other three size X op- was restricted to the RT data, which appeared to provide
eration combinations were 4 msec, and none approached a more sensitive measure of transfer. Figure 1 presents
significance (smallest a = .31). Thus, Experiment 2 repli- mean RT savings as a function of operation, problem
cated both the evidence of multiplication-to-division size, and transfer condition for the 60 participants in Ex-
transfer and the absence of division-to-multiplication periments 1 and 2. The error bars represent one standard
transfer found in Experiment 1. error of the mean.
Short-lag probes versus short-lag primes: Percent- The combined analysis clearly demonstrated transfer
age of errors. Table 4 presents the mean percentage of across the two orders of division problems [F(1,59) =
errors in each condition. Probe trials produced fewer er- 12.02, MSe = 11,734], both for small [32 msec, t(59) =
rors than prime trials, overall [8.7% vs. 10.9%; F( 1,29) = 2.48,p = .016] and for large problems [65 msec; t(59) =
13.26, MS e = 65.75], but problem type interacted with 3.00, p = .004]. The critical effect for the identical-
other variables. Specifically, error savings were greater elements model, however, concerns the prediction that
 DIVISION BY MULTIPLICATION 799

140 140
Division Probes Multiplication Probes

120
o Small f1IIjLarge
120
o Small f1IIjLarge

-en
100
- 100

-E
en
C)
C
oS;
80
en
-E
en
C)
C
oS;
80

to to
en 60 en 60
~ t-
o::
C 40 C 40
to to
(I) (I)
~ ~
20 20

0 0

-20 -20
Identical Order Operation Identical Order Operation
Repetition Change Change Repetition Change Change
Transfer Condition Transfer Condition
Figure I. Mean short-lag response time savings as a function of operation, problem size, and transfer condition (Experiments 1 and
2 combined).

transfer should be stronger with identical repetition than quite efficient, inasmuch as the operation effect was
with order change for division, but not for multiplication. smaller for the problems that apparently were more likely
Using the combined data, an ANOVAthat included both to be mediated.
identical-repetition and order change conditions (but ex-
cluded operation change) confirmed the expected oper- GENERAL DISCUSSION
ation X problem type X transfer interaction [F(l,59) =
4.53, MSe = 5,370, P = .037]. Specifically, transfer in di- The General Discussion is divided into four sections.
vision was smaller with order change (48 msec) than with The first section discusses the evidence that large-number
identical repetition (106 msec), whereas RT savings in division problems were mediated by multiplication. The
multiplication were essentially the same with order change second section considers explanations for asymmetrical
(66 msec) or identical repetition (73 msec). transfer-that is, the finding of transfer from multiplica-
With respect to operation change transfer, the combined tion to division, but not from division to multiplication.
analysis confirmed the three-way interaction of operation, The third section addresses the use of multiplication to
problem size, and problem type [F(l,59) = 4.73, MSe = check rather than to mediate division, and the final section
9,671,p = .034]. As Figure I shows, the interaction oc- discusses implications of the results for the identical-
curred because the only significant interoperation sav- elements model (Rickard & Bourne, 1996).
ings involved transfer from multiplication to division for
large problems [58 msec; t(59) = 3.11, p = .003].1 This Mediation of Division by Multiplication
effect suggests that large-number division problems were The short-lag conditions of Experiments 1 and 2
mediated by multiplication in the operation change con- showed that large-number division problems (e.g., 63 -i-
dition. Whereas division RTs were generally longer than 7 = ?) benefited from retrieval ofthe corresponding mul-
multiplication RTs, this difference was smaller for large tiplication fact (e.g., 7 X 9 = 63) during the preceding
(53 msec) than for small problems [l l O msec; F(l,59) = four to six trials. This effect is expected if multiplication
5.72, MS e = 8,722, p = .020; see Tables 1 and 3]. This mediated division, because division performance ought
suggests that mediated division for large problems was to improve with the accessibility of the mediating multi-
800 CAMPBELL

plication fact. It is unlikely that the multiplication-to- for large division problems, relative to small division prob-
division transfer could be due to a simple priming mech- lems. More generally, mediation occurred despite a mod-
anism, given the combined effects of interference from erately high level of speed pressure, which ought to re-
several intervening problems and the approximately 15- duce the opportunity for mediation and encourage people
sec elapsed time, on average, between the prime and the to rely on direct retrieval. Furthermore, the multiplication-
probe. Furthermore, if the observed transfer reflected a to-division mediation process was sufficiently robust to
simple priming process, we would also expect transfer be detected (as RT savings), despite six repetitions of each
from multiplication to small division problems and from problem in each operand order. Although repetition of an
division to multiplication, but these effects were not ob- individual item presumably will lead to direct retrieval
served. eventually, the present data suggest that mediation can be
Experiments 1 and 2 apparently demonstrated media- efficient enough to contribute to performance even after
tion of division by multiplication, but to what extent can several recent repetitions of the probe item. The capac-
the results be generalized beyond the conditions of these ity to incorporate mediation under these conditions sug-
experiments? In the short-lag conditions, episodic or se- gests that skilled adults can exploit the conceptual cor-
mantic information associated with the recent prime trial respondence between multiplication and division in a
might be easily reactivated by the probe. Mediating facts highly efficient way to facilitate performance.
would not necessarily be as accessible under different
conditions. Nonetheless, the experiment did not promote Asymmetrical Interoperation Transfer
mediation in haphazard or improbable ways. For example, Whereas there was clear evidence of multiplication-
the small-number problems in the basic operations are to-division transfer for large division problems, there was
usually well memorized and, therefore, would be more no evidence of transfer in the opposite direction (i.e.,
likely to be solved by direct retrieval, rather than by medi- from division to multiplication). This asymmetry is rather
ation (LeFevre, Bisanz, et aI., 1996; LeFevre, Sadesky, & surprising: If division were mediated consistently by re-
Bisanz, 1996). Consistent with this, interoperation transfer trieval ofthe corresponding multiplication fact, we might
occurred for large-number division problems, but there expect the multiplication node to be strengthened and to
was no evidence of multiplication-to-division transfer produce transfer from division primes to corresponding
for small-number division problems. This cannot be at- multiplication probes. The failure to observe division-to-
tributed to a general difficulty in measuring transfer for multiplication transfer cannot reflect a general difficulty
the small division problems, because these items demon- in measuring transfer to multiplication in these experi-
strated substantial transfer in the identical-repetition ments, because there was robust transfer for multiplica-
condition. Rather, the multiplication-to-division transfer tion problems in the within-operation transfer conditions.
results suggest that large division problems were more Two explanations for the asymmetrical transfer sug-
likely to be mediated by multiplication than were small di- gest themselves. First, mediation of division by multi-
vision problems. plication may depend on the accessibility of the mediat-
What might be the nature of the mediation process? ing multiplication fact. For example, it is possible that the
Mediation implies knowledge that multiplication and di- division probes were mediated by multiplication in the
vision are inverse operations and, more specifically, that multiplication-to-division condition because the corre-
there is a logical correspondence between the numerical sponding multiplication primes were recently retrieved
components of division and multiplication facts. For ex- (i.e., in the preceding four to six trials) and were, there-
ample, the division problem 63 -i- 7 would serve as a re- fore, highly accessible. In contrast, division primes may
trieval cue for the multiplication fact that 7 X 9 = 63 or have been less likely to be solved by mediation, because
9 X 7 = 63. The number that is present in the multipli- they were not preceded recently by the corresponding mul-
cation fact and not present in the division stimulus (i.e., tiplication problem. In this case, we would not observe
9) must be the quotient. In this view, mediation entails a transfer from division to multiplication, because division
match-and-select process that identifies the matching primes in the division-to-multiplication transfer condition
numbers in corresponding division stimuli and multipli- were relatively unlikely to be mediated by multiplication.
cation facts and selects the nonmatching component ofthe Alternatively, the asymmetrical transfer (i.e., multiplica-
multiplication fact as the quotient. tion-to-division but not division-to-multiplication) could
Although it is often assumed that strategies involving indicate that the process by which multiplication medi-
transformations or mediation would be inefficient, rela- ated division did not produce strong priming or strength-
tive to direct retrieval, the present results suggest that di- ening of the multiplication node. For example, there may
vision mediated by multiplication may be very efficient. be no intentional review or consolidation ofthe mediating
Although larger division problems apparently were me- multiplication fact. Rather, the quotient generated may
diated by multiplication more than were smaller division simply be associated with the presented division problem,
problems, the mean RT difference between small divi- increasing only the strength ofthe division fact in memory.
sion and multiplication problems was actually larger than The present data demonstrated multiplication-to-
the difference between large division and multiplication division transfer, whereas Campbell (1997) did not ob-
problems. Thus, mediation did not lead to greater RT costs serve multiplication-to-division error priming. This may
 DIVISION BY MULTIPLICATION 801

reflect basic differences between the two phenomena. trieval or guessing. The increased errors owing to speed
Transfer is facilitation measured between two correspond- pressure would also promote checking after the fact. Thus,
ing (i.e., identical or closely related) problems (e.g., 6 X speed pressure is likely both to reduce mediation and to
8, 8 X 6, 48 -i- 6, or 48 -:- 8), whereas error priming is an increase checking. However, there must be adequate op-
interference effect measured between two different prob- portunity for checking to occur. Consequently, with very
lems (e.g., 6 X 8 may prime the error 7 X 6 = 48; or 42 -:- short intertrial intervals, as in the present experiment, or
7 may prime the error 8 X 6 = 42). With respect to if intertrial intervals are filled with a distractor task,
multiplication-to-division transfer, a division probe would checking is likely to be curtailed.
be an effective cue to retrieve the mediating multiplica-
tion fact, because corresponding division and multipli- Implications for the Identical-Elements Model
cation problems share all their numerical components in Methodological differences probably explain why
common. With respect to error priming, in contrast, di- Rickard and Bourne (1996; Rickard et aI., 1994) found
vision problems presumably do not strongly activate non- little or no transfer between division and multiplication
corresponding multiplication nodes; consequently, non- in either direction, whereas the present study clearly
corresponding nodes would be unlikely to become demonstrated transfer from multiplication-to-division
erroneously involved in mediation and to produce error for large problems. First, we measured transfer over a
priming. This may explain why we observe multiplication- range of only 4-6 trials here, whereas Rickard et al.
to-division transfer but not multiplication-to-division tested transfer after longer delays (e.g., in separate test
error priming. blocks at the end of practice or after a l-day layoff from
practice). Mediation may be promoted by recent retrieval
Mediation Versus Checking of a potential mediator, favoring mediation in the present
of Division by Multiplication experiments more than in the Rickard et al. studies. Sec-
As was discussed in the preceding section, mediation ond, in the present studies, each problem in each operation
might not produce division-to-multiplication transfer, but was tested six times across blocks, and transfer tests were
the multiplication check hypothesized by Campbell (1997) distributed across the 12 blocks. In contrast, Rickard et al.
to account for division-to-multiplication error priming presented a large number ofpractice blocks (40 or 90 rep-
presumably would produce division-to-multiplication etitions of practice items) prior to testing transfer to un-
transfer, if checking occurred in the present study. Ingen- practiced items. It is possible that division performance
eral, the division-to-multiplication error priming ob- was mediated over early practice blocks but was based
served by Campbell (1997) is consistent with the view on direct retrieval for the majority ofpractice trials. In this
that people often access multiplication facts to either case, the effects of mediated transfer at test might have
mediate or check their division performance, but the dif- been weak, relative to those from associative transfer.
ferent asymmetries for short-lag transfer and error prim- Despite the different methodologies and some differ-
ing suggest either that transfer and error priming are ences in the pattern of transfer results, the present ex-
governed by different factors (see the preceding section), periments confirmed an important prediction of the
or that using multiplication to check division was less identical-elements model. Division-to-division transfer
likely in the present experiments. One good reason why was larger with identical repetition than with an order
checking would be less likely here is that the response- change, whereas multiplication-to-multiplication transfer
stimulus interval was only about 1-1.5 sec, as compared was equivalent for these two conditions. This operation
with 2.5 sec in Campbell's (1997) study. Given that check- X order change interaction is predicted by the identical-
ing occurred between trials, the shorter response-stimulus elements model, according to which the two orders of di-
interval in the present study would curtail checking. vision problems involve separate memory nodes, whereas
More generally, however, when would people check the two orders of a multiplication problem access a com-
division with multiplication, rather than mediate division mon node. Associative transfer arises in the latter case,
by multiplication? The purpose ofchecking division with owing to strengthening of the common node. There was
multiplication is, presumably, to detect errors. This im- evidence of some transfer, however, between the two or-
plies that people know or assume that memory for mul- ders of division problems (e.g., 54 -i- 6 = 9 and 54 -i- 9 =
tiplication is reliable, or at least more reliable than direct 6). This is not predicted by the identical-elements model,
retrieval of division facts. In this case, however, we might but it does not necessarily contradict the model, if we
expect people routinely to mediate division by multipli- distinguish between associative transfer (i.e., facilitation
cation, rather than only to check for an error after the di- owing to strengthening of a common problem node) and
vision answer is produced. Nonetheless, checking (rather mediated transfer (i.e., facilitation owing to mediation
than mediation) would be expected to occur when per- by a related problem). In the case of mediated transfer,
formance demands or conditions of the task environment anything that increases accessibility of the mediator
reduce opportunities for mediation but afford opportu- (e.g., a recent practice trial) ought to yield facilitation for
nities for checking. For example, conditions that empha- the mediated problem. Given independent nodes for the
size speed over accuracy would reduce opportunities for two orders of division problems, cross-order transfer in
mediation and promote increased reliance on direct re- division would arise if both orders were mediated by a
802 CAMPBELL

common, strengthened multiplication fact or ifrecent re- ASHCRAFT, M. H. (1995). Cognitive psychology and simple arithmetic:
trieval and strengthening of one order ofa division prob- A review and summary of new directions. Mathematical Cognition,
1,3-34.
lem was exploited to mediate performance of the other ASHCRAFT, M. H., & CHRISTY, K. S. (1995). The frequency of arithmetic
order. Greater savings with identical repetition than with facts in elementary texts: Addition and multiplication in grades 1-6.
an operand change in division might arise because more Journalfor Research in Mathematics Education.S, 396-421.
processing components are repeated in the case of iden- BAROODY, A. J. (1994). An evaluation of evidence supporting fact-
tical repetition (i.e., encoding of stimulus information, retrieval models. Learning & Individual Differences, 6, 1-36.
CAMPBELL, J. I. D. (1995). Mechanisms of simple addition and multi-
retrieval processes, response production) and, therefore, plication: A modified network-interference theory and simulation.
more stages of processing may contribute savings. Mathematical Cognition, I, 121-164.
Finally,the multiplication-to-division transfer observed CAMPBELL, J. I. D. (1997). On the relation between skilled performance
here, although not predicted by the identical-elements of simple division and multiplication. Journal ofExperimental Psy-
chology: Learning. Memory. & Cognition, 23, 1140-1159.
model, is not inherently at odds with the model's repre-
CAMPBELL, J. I. D., & CLARK, J. M. (1989). Time course oferror-priming
sentational assumptions. We found evidence that large in number fact retrieval: Evidence for excitatory and inhibitory
division problems were mediated by multiplication, but mechanisms. Journal ofExperimental Psychology: Learning. Mem-
there seems to be no reason in principle why people can- ory. & Cognition, 15,920-929.
not develop memory processes for division facts that do C!POLOTTI, L., & DE LACY COSTELLO, A. (1995). Selective impairment
for simple division. Cortex, 31, 433-449.
not involve multiplication. Indeed, the absence of trans- s
GEARY, D. C. (1994). Children mathematical development. Washing-
fer from small multiplication to small division problems ton, DC: American Psychological Association.
supports this and is consistent with the identical- GEARY, D. C. (1996). The problem size effect in mental addition: De-
elements model. Nonetheless, the present data suggest that velopmental and cross-national trends. Mathematical Cognition, 2,
63-93.
skilled adults apparently do utilize multiplication knowl-
GEARY, D. C; & WILEY, 1. G. (1991). Cognitive addition: Strategy
edge in their processing of large-number division facts, choice and speed-of-processing differences in young and elderly
at least under conditions in which potential mediators are adults. Psychology & Aging, 6, 474-483.
readily accessible. HITTMAIR-DELAZER, M., SEMENZA, C; & DENES, G. (1994). Concepts
and facts in calculation. Brain, 117, 715-728.
LEFEVRE, 1., BISANZ, J., DALEY, K. E., BUFFONE, L., GREEN HAM, S. L.,
CONCLUDING COMMENTS & SADESKY, G. S. (1996). Multiple routes to solution of single-digit
multiplication problems. Journal ofExperimental Psychology: Gen-
Recall that Hittmair-Delazer et al. (1994) described an eral, 125, 284-306.
acalculic patient whose division performance apparently LEFEVRE, J., SADESKY, G. S., & BISANZ, J. (1996). Selection ofproce-
dures in mental addition: Reassessing the problem size effect in
depended on multiplication facts, whereas Cipolotti and
adults. Journal ofExperimental Psychology: Learning. Memory. &
de Lacy Costello (1995) studied an acalculic patient with Cognition, 22, 216-230.
largely intact multiplication performance but severely RICKARD, T. c., & BOURNE, L. E., JR. (1996). Some tests of an identical
impaired division. Mediation ofdivision by multiplication elements model of basic arithmetic skills. Journal of Experimental
undoubtedly depends on a variety of factors, and individ- Psychology: Learning. Memory. & Cognition, 22,1281-1295.
RICKARD, T. c., HEALY, A. E, & BOURNE, L. E., JR. (1994). On the rep-
ual differences in these factors may explain the appar- resentation ofarithmetic facts: Operand order, symbol, and operation
ently inconsistent neuropsychological evidence. Individ- transfer effects. Journal of Experimental Psychology: Learning.
uals with very well developed memory for division facts Memory. & Cognition, 20,1139-1153.
presumably rely very little on mediation. Similarly, indi- SIEGLER, R. S. (1988). Strategy choice procedures and the development
of multiplication skill. Journal ofExperimental Psychology: General,
viduals who have poorly developed memory for multipli-
117,258-275.
cation are unlikely to rely on multiplication to perform ZORODOFF, N. J. (1995). Why is 9 + 7 harder than 2 + 3? Strength and
division. It would not be surprising for mediation to be interference as explanations for the problem-size effect. Memory &
common, however.LeFevre, Bisanz, et al. (1996; LeFevre, Cognition,23,689-700.
Sadesky, & Bisanz, 1996) concluded that well-educated
NOTE
adults often use a variety of nonretrieval strategies to
perform simple arithmetic problems, especially large- I. Among the 14 small problems, half involve repeated products that
number problems, which typically are not as well mem- are correct for two combinations of operands (i.e., 12, 16, 18, or 24),
orized. The present results suggest similarly that media- whereas there is only one repeated product (i.e., 36) among the large
problems. Because they are not associated with a unique problem, items
tion of division by multiplication may be a persistent
with repeated product/dividends might produce weak transfer. In this
strategy or process used by skilled adults. Although simple case, the high proportion of small problems with repeated product/div-
direct retrieval is probably the modal process used by ed- idends could contribute to the finding that small-problem multiplication
ucated adults for basic arithmetic, the emerging evidence primes did not produce transfer to division. To check this, a four-factor
suggests that adults also possess a repertoire ofefficient, ANOVA (operation, transfer condition, lag, and repeated vs. unique
products/dividends) of the combined data that included only small-
conceptually based procedures for simple arithmetic that number problems was performed separately for RT and errors. There
researchers are only beginning to recognize and study. was no evidence ofdifferential transfer for items with unique versus re-
peated products/dividends, either for RT or for errors.
REFERENCES
ASHCRAFT, M. H. (1992). Cognitive arithmetic: A review of theory and (Manuscript received December 18, 1997;
data. Cognition, 44, 75-106. revision accepted for publication October 5, 1998.)