British J of Psychology - 2011 - Moeller - Three Processes Underlying The Carry Effect in Addition Evidence From Eye

20448295, 2011, 3, Downloaded from https://bpspsychub.onlinelibrary.wiley.com/doi/10.1111/j.2044-8295.2011.02034.x by The Open University, Wiley Online Library on [12/03/2023].
See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
623
The
British
Psychological
British Journal of Psychology (2011), 102, 623–645

C 2011 The British Psychological Society
Society
www.wileyonlinelibrary.com
Three processes underlying the carry effect

in addition – Evidence from eye tracking
Korbinian Moeller1,2 ∗ , Elise Klein2,3 and Hans-Christoph

Nuerk1,2
1
Knowledge Media Research Center, Tuebingen, Germany
2
Department of Psychology, Eberhard Karls University, Tuebingen, Germany
3
Department of Neurology, Section Neuropsychology, University Hospital
of RWTH Aachen University, Aachen, Germany
Recent research indicated that processes of unit-decade integration pose particular

difficulty on multi-digit addition. In fact, longer response latencies as well as higher
error rates have been observed for addition problems involving a carry operation (e.g.,
18 + 27) compared to problems not requiring a carry (e.g., 13 + 32). However, the
cognitive instantiation of this carry effect remained unknown. In the current study, this
question was pursued by recording participants’ eye fixation behaviour during addition
problem verification. Analyses of the eye fixation data suggested a prominent role of
the unit digits of the summands. The need for a carry seems to be recognized very
early during the encoding of the problem after initial unit sum calculation has established
the basis for the no carry/carry detection. Additionally, processes related to the actual
carry execution seemed to be associated with the processing of the decade digit of the
solution probe but were less unambiguous. Taken together, our findings indicate that
unit-based calculations and the associated recognition that a carry is needed as well as
its completion determine the difficulty of carry addition problems. On a more general
level, this study shows how the nature of numerical-cognitive processes can be further
differentiated by the evaluation of eye movement measures.
One of the probably most robust findings in multi-digit addition is the carry effect.
Whenever a carry is needed, both response latencies and/or error rates (ERs) increase
considerably (Ashcraft & Battaglia, 1978; Ashcraft & Stazyk, 1981; Deschuyteneer,
De Rammelaere, & Fias, 2005; Fürst & Hitch, 2000; Groen & Parkman, 1972; Imbo,
Vandierendonck, & De Rammelaere, 2007; Klein, Nuerk, Wood, Knops, & Willmes,
2009; Klein et al., 2010a, b; Kong et al., 2005; Moeller, Klein, & Nuerk, in press).
∗ Correspondence should be addressed to Korbinian Moeller, Department of Psychology, Eberhard Karls University, Tuebingen,
Friedrichstrasse 21, 72072 Tuebingen, Germany (e-mail: korbinian.moeller@uni-tuebingen.de).
DOI:10.1111/j.2044-8295.2011.02034.x
20448295, 2011, 3, Downloaded from https://bpspsychub.onlinelibrary.wiley.com/doi/10.1111/j.2044-8295.2011.02034.x by The Open University, Wiley Online Library on [12/03/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
624 Korbinian Moeller et al.
The carry effect and associated processes

Generally, whether or not a carry operation is needed is determined by the summands
of the addition problem: whenever the sum of the unit digits of the summands is ≥10,
a carry is necessary (e.g., 7 + 8 = 15 in 47 + 28 = 75), whereas no carry is needed
whenever the sum of the units is <10 (e.g., 2 + 3 = 5 in 52 + 23 = 75). For the above
example 47 + 28, the carry operation is executed by adding 1 (representing the decade
digit of the unit sum) to the sum of the decade digits of the summands. In this case,
the sum of the unit digits is 15, so the unit digit of the result (i.e., 75) is 5, and the decade
digit of the result is derived by updating the sum of the decade digits of the summands
by 1 (i.e., 4 + 2 + 1 = 7). However, our knowledge on the specificities determining this
higher difficulty of carry addition problems is still rather patchy.
On a theoretical/conceptual level, difficulties arising with the need for a carry have
been suggested to originate from different sources. On the one hand, Kalaman and
Lefevre (2007) argued that the carry effect represents an increased demand on working
memory resources, for example, by keeping track of intermediate results (see also
Imbo et al., 2007; Kazui, Kitagaki, & Mori, 2000; Zago et al., 2001). On the other
hand, Nuerk, Graf, and Willmes (2006) associated the increased difficulty of carry
addition problems with a higher workload for the correct classification (i.e., units, tens,
etc.) and manipulation of digits within the base-10 place-value structure of the Arabic
number system (i.e., carrying from one digit position to another, cf. Klein et al., 2010a;
Nuerk, Moeller, Klein, Willmes, & Fischer, in press for a review on multi-digit number
processing). Finally, Green, Lemaire, and Dufau (2007) suggested that the carry effect
may also reflect processes necessary to adapt solution strategies to the actual problem
(see also Torbeyns, Verschavel, & Ghesquiere, 2002). In sum, there seem to be different
sources of difficulty for the carry operation.
Eye tracking in numerical cognition research

At this point, evaluating eye fixation behaviour may supplement the interpretation
of response latencies and ERs when investigating the temporal dynamics of number
processing (cf. Brysbaert, 1995). According to the immediacy and eye mind assumption
the eyes tend to fixate those objects from which visual information is extracted to
support their cognitive evaluation (e.g., Just & Carpenter, 1980; Rayner & Pollatsek,
1989). Additionally, fixation durations are agreed to reflect a reliable measure of how
long processing of a particular stimulus takes (Rayner & Pollatsek, 1989; for a review
see Rayner, 1998). Thereby, evaluation of participants’ eye fixation behaviour may be a
valuable tool to dissociate cognitive processes in numerical cognition. There are different
eye-movement measures generally assumed to index different processing stages: For
instance, first fixation durations (FFDs) are thought to be sensitive to modulations
by early, mostly stimulus-driven, bottom-up processing (e.g., Deutsch, Frost, Pelleg,
Pollatsek, & Rayner, 2003; Rayner & Pollatsek, 1987; Wong & Chen, 1999), whereas
total reading times (TRTs) are suggested to be subject to modulations by later processing
stages involving top-down processes (e.g., Joseph et al., 2008; Traxler & Pickering,
1996; Rayner, Warren, Juhasz, & Liversidge, 2004; see also Calvo & Meseguer, 2002 for
a classification of eye-movement measures).
However, in contrast to other domains such as reading research, only few studies
in the domain of numerical cognition have tapped the potential of eye-tracking data
yet. Nevertheless, the eye-tracking methodology has already been employed to study
basic as well as more complex numerical processes. First evidence for the validity of eye
Three processes underlying the carry effect 625
fixation data for the evaluation of basic numerical processes has come from studies on
basic number related tasks such as number reading (Brysbaert, 1995) and magnitude
comparison (Moeller, Fischer, Nuerk, & Willmes, 2009a) as well as more complex
tasks such as the number bisection task (Moeller, Fischer, Nuerk, & Willmes, 2009b).
Furthermore, Moeller et al. (2009b) also applied above differentiation of early and later
processing as indicated by different measures to the field of numerical cognition. The
authors proposed a two-stage processing model of eye fixation behaviour for numerical
tasks. In an initial bottom-up processing stage the constituting digits of the numbers may
be identified largely automatic and assigned their stimulus-driven lexical values (e.g.,
place-value position, magnitude) by extracting their physical features and activating the
memorized visual number forms (cf. Henik & Tzelgov, 1982; Moeller et al., 2009b).
Moeller et al. (2009b) suggested that such early numerical processes should be reflected
by early processing measures such as FFD. The second processing stage reflecting a
subsequent top-down driven wrap-up stage is supposed to be initiated after all numbers
of the actual problem have been encoded. At this stage, the numbers of an arithmetical
problem may be integrated and put into relation with the other numbers by checking
top-down mediated plausibility and/or processing rules using the lexical attributes of
the single numbers. This process is supposed to be reflected by eye-movement measures
capturing rather late processing such as TRT.
However, as regards addition, a first study evaluating eye fixation behaviour by Green
and colleagues (2007) only employed eye tracking to validate the use of strategies,
which the participants were instructed to apply, but did not use the eye fixation data
to investigate the nature of the cognitive mechanisms underlying the difficulty of carry
operations in multi-digit addition. Against this background, the current study aimed at
dissociating different processes associated with solving carry addition problems. These
processes may help to better understand what makes carry addition problems more
difficult than their non-carry counterparts.
Three cognitive processes underlying the carry operation

Though it is widely agreed that the need for a carry determines difficulty in multi-
digit addition, the question what exactly causes the difficulty associated with the
carry operation is not yet resolved. We suggest that at least three processes can be
differentiated that may drive the increased difficulty when processing carry addition
problems.
(1) Before any carry procedure can be executed, it has to be recognized that a carry
is needed to compute the correct result. One way to determine whether a carry is
needed is to keep track of the sum of the unit digits of the summands: a carry is
required whenever the unit sum is equal or larger than 10. Thereby, the necessity
of a carry may be driven by early estimation processes upon the unit digits.
(2) After it has been decided on whether a carry is needed or not, the sum of the units
has to be calculated in either case. Consider, for instance, 47 + 28. Here, the sum
of the unit digits 7 + 8 equals 15 (analogue 52 + 23 → 2 + 3 = 5). Thereby, the
actual calculation of the unit sum is driven by calculation processes based upon
the unit digits.
(3) Finally, the carry procedure still has to be executed properly: this means that the
carry (i.e., the decade digit of the unit sum) has to be added to the sum of the
decade digits of the summands. In above mentioned example, the decade digit of
the unit sum is 1 (from 15). Adding this to the sum of the decade digits of the
summands, that is, 4 + 2 + 1 = 7 (italics indicating the carry) results in the decade
digit of the correct result (47 + 28 = 75).
Objectives of the current study

Applying above introduced opportunities of the eye-tracking methodology to an addition
verification task allows for deriving specific hypotheses for the influence of the
underlying cognitive processes on the eye fixation behaviour in addition problems with
and without carry. However, as already outlined above, the goal of the current study
was not a mere replication of reaction time (RT) data. Rather, following up on above
considerations on at least three possible origins of difficulty in carry addition problems,
more specific hypotheses for the eye fixation behaviour can be derived:
(1) The difficulty associated with the decision whether a carry operation is needed
should arise at the unit stage and should be an early effect. Thus, longer fixation
times should be observed on the unit digits of the summands for carry addition
problems. In particular, this should be already present in FFDs indicating bottom-
up processing of the magnitude information of the respective digits.
Following this rationale allows for an even more specific hypothesis: estimating
whether a carry is needed should be particularly difficult for problems with a unit
sum close to the critical value of 10 (e.g., 16 + 35 = 51 with a unit sum of 6 + 5 =
11), whereas the need for a carry should be easier to predict for problems with a
unit sum much smaller or larger than 10 (e.g., 18 + 29 = 47 with a unit sum of 17).
As the need for a carry may be evaluated primarily when processing the unit digit
of the second summand by early estimation processes, FFDs on the unit digit of the
second summand should increase as the unit sum gets closer to 10 (see Figure 1
Panel A for illustration).
(2) The difficulty associated with the calculation of the unit sum should also arise
at the unit calculation stage. However, this should be more prominently present
in longer TRTs on the unit digits of the summands for carry addition problems
than for non-carry problems. Additionally, regressions from the second to the first
summand should specifically involve the unit digits. This should be the case as
specific evaluation of the units (i.e., the unit sum being equal or larger than 10) is
required to explicitly compute the unit sum.
(3) The difficulty associated with the execution of the carry operation itself, reflecting
the application of a procedural rule, should thus be observed in longer TRTs on the
decade digit of the solution probe as the sum of the decade digits of the summands
has to be updated by the carry to obtain the correct result (e.g., adding 1 to the
sum of the decade digits). Accordingly, regressions from the solution probe back
to the summands should be prominently focused on the decade digits, probably
indicating a checking procedure.
However, please note that the sequence of process (2) following on process (1) may
not be mandatory. It is also conceivable that during the initial processing stages, unit
sum is not only estimated but actually calculated, and the result of this early computation
process is then used to derive the decision whether a carry is needed or not. In this case,
FFDs on the unit digit of the second summand should increase as the unit sum increases
per se (see Figure 1, Panel B for an illustration) – instead of increasing the closer the unit
sum gets to 10 (see Figure 1, Panel A, Panel C for a categorical influence of whether a
carry is needed or not). The present study pursued these questions.
A Distance to 10
0 5 10 15 20
Unit Sum
B Unit Sum
0 5 10 15 20
Unit Sum
0 5 10 15 20
Unit Sum
Figure 1. Hypothetical pattern of first fixation durations (FFDs) on the unit digit of the second
summand. Panel A illustrates the expected data pattern corroborating the notion of early approximate
estimation processes for determining whether a carry is needed or not. Panel B depicts the hypothetical
pattern associated with early exact calculation of the unit sum and finally the expected pattern for the
case of a categorical all-or-nothing carry detection can be read from Panel C.
Method
Participants
Twenty students (15 female) of the Paris-Lodron University of Salzburg participated in
the study as partial fulfilment of course requirements. Mean age was 22.7 years with
a standard deviation (SD) of 2.4 years (range: 20–31 years). All participants reported
normal or corrected to normal vision.
Apparatus
Eye fixation behaviour was recorded online using an EyeLink 1000 eye-tracking device
(SR Research, Mississauga, Ontario, Canada) and stored for offline analysis. The EyeLink
1000 provides a spatial resolution of less than 0.5◦ of visual angle at a sampling rate
of 1,000 Hz. Stimuli were presented on a 21 monitor with resolution set to 1,024 ×
768 pixels and driven at a refresh rate of 120 Hz. Participants’ heads were placed in
a chinrest throughout the experiment to keep viewing distance (50 cm) and viewing
angle constant.
Task, stimuli, and design

Participants had to evaluate whether addition problems were solved correctly by
simultaneously presented probes or not. The addition problems consisted of one- and
two-digit numbers in Arabic notation with a sum not exceeding 98. The stimulus set
comprised 96 addition problems. In a factorial 2 × 2 design, problem size (small vs.
large, i.e., sum < 40 vs. sum > 40) and the need for a carry operation (required vs.
not required) were manipulated orthogonally. Importantly, overall problem size was
matched between non-carry and carry addition problems (e.g., 43 + 32 = 75 vs.
48 + 27 = 75). Each problem was presented twice: once together with the correct
result and once with an incorrect probe. Incorrect probes deviated from the correct
result by either 2 or 10 to prevent parity-based solution strategies with average split kept
constant (i.e., zero) across item groups (for an comprehensive list of the stimuli and
stimulus properties see Appendices B and C in Klein et al., 2010a). To ensure that all
participants started to encode the addition problem at the first summand, a fixation point
to the left of this summand was used (x/y coordinates: 112/384). Addition problems were
presented in the form xx + xx = xx at a position slightly to the right of the centre of the
screen. Digits and arithmetic symbols were shown in white against a black background
using the un-proportional font New Courier (size: 48; style: bold). At this size, each digit
subtended 1.4◦ of visual angle in width and 1.9◦ in height. The distance between the
both summands as well as between the second summand and the solution probe equalled
10.4◦ of visual angle to keep influences of parafoveal pre-processing at a minimum.
Procedure
After being seated with their head stabilized by a chinrest, participants were instructed to
evaluate as fast and as accurately as possible whether the subsequently displayed addition
problems were solved correctly or not: when a problem was presented together with
its correct solution the right button of a response device had to be pressed by the index
finger of the right hand whereas a button press by the left index finger on a left button
indicated that a problem was presented with an incorrect solution. Then a nine-point
calibration of the eye-tracking system was conducted to maximize spatial resolution for
each participant. Thereafter, participants had to evaluate ten practice trials to familiarize
with display layout and task requirements. The experiment was set up in four blocks of
48 items each and lasted about 45 min.
Analysis
All subsequent analyses exclusively incorporated data from items presented with a
correct solution probe. This restriction was necessary as incorrect solution probes may
be rejected employing non-computational strategies (e.g., matching of the intermediate
unit sum). Additionally, evaluation processes of incorrect probes may be driven by factors
such as split between incorrect probe and correct result that do not play a role in the
evaluation of correct solution probes as they do not even exist for these (e.g., Klein
et al., 2009, 2010a). Furthermore, Menon et al. (2000; Menon, Mackenzie, Rivera, &
Reiss, 2002) were able to show that responses to correct and incorrect probes differ
even in their neural correlates: in incorrect addition problems, specific cortex areas were
additionally recruited in left dorsolateral and ventrolateral prefrontal cortices – areas
usually not associated with the processing of domain specific numerical information but
with monitoring of contextual information (e.g., Cabeza, Locantore, & Anderson, 2003;
Ranganath, Johnson, & D’Esposito, 2000) and the generation of alternative solutions to
a given problem (e.g., Donohue, Wendelken, Crone, & Bunge, 2005; Goel & Vartarian,
2005). As we were specifically interested in the processes necessary for deriving the
correct solution to an addition problem, we restricted analyses to those items involving
a correct solution probe.1
For the subsequent evaluation of eye fixation behaviour in terms of TRTs and FFDs,
areas of interest were defined for the decade as well as the unit digit of either the first
summand, the second summand, and the solution probe: each digit was centred in an
area of interest 75-pixel wide and 180-pixel high. Whenever a fixation fell within one
of these interest areas, it was considered as a fixation on the corresponding decade or
unit digit. Subsequently, FFD as well as TRT for each interest area was computed and
submitted to the respective analyses. Whenever necessary, post hoc comparisons were
conducted using the Games–Howell test at a significance level of p < .05 to account for
differing variances. Additionally, in cases where the sphericity assumption of the analysis
of variance (ANOVA) was violated, the Greenhouse–Geisser coefficient is given to allow
for an adjustment of the degrees of freedom.
Results
Participants who performed at or below chance level in at least one of the experimental
conditions where excluded from further analyses. This affected one participant. For
the other participants, ERs ranged from 1.0% to 13.3% (M = 7.0%; SD = 3.4%). Only
response latencies followed by a correct classification were considered for the RT
analysis. Additionally, a trimming procedure eliminated all latencies shorter than 200 ms
and longer than 5,000 ms in a first step. Subsequently, all RTs falling below or above three
SDs of an individual participant’s mean were excluded from the analyses in a second
step. For the analyses of the eye fixation data, only data from trials already included in
the RT analysis were considered. Additionally, ERs were arcsine transformed prior to the
analysis to approximate normal distribution.
RT and error data

Response latencies and ERs were evaluated by a 2 × 2 ANOVA incorporating the factors
problem size (small vs. large) and carry (needed vs. not needed). The ANOVA showed
1 Please note that we are well aware that analyzing the data of the incorrectly solved problems may be informative on
possible strategy differences between processing these and correctly solved problems and thus, may further substantiate the
observations by Menon et al. (2002). However, for the sake of brevity and lucidity, the current study focused on evaluating
processes underlying the carry effect in correctly solved addition problems. See also Moeller et al. (in press) for strategy effects
on eye-fixation behaviour associated with the processing of incorrect solution probes.
A Response Latencies
RT in msec
3000
No-carry problems
Carry problems
2500
2000
1500
1000
Large Small
Problem Size
B Error Rates
% Errors
No-carry problems
Carry problems
10
0
Large Small
Problem Size
Figure 2. Response latencies (Panel A) and error rates (Panel B) separated for carry and non-carry
problems with either large or small problem size. Error bars depict 1 standard error of the mean (SEM).
that both response latencies as well as ERs were reliably influenced by problem size
[RT: F(1, 18) = 86.26, p < .001; ER: F(1, 18) = 16.09, p < .001; see Figure 2]: addition
problems with a relatively larger problem size resulted in longer latencies (2,772 ms) and
a higher ER (9.6% errors) as compared to problems with a small problem size (1,730 ms
and 4.3% errors). Moreover, carry also had a consistent influence on RTs and ERs [RT: F(1,
18) = 121.92, p < .001; ER: F(1, 18) = 5.13, p < .05; see Figure 1, Panel B] with increased
latencies and more errors being associated with carry addition problems (2,471 ms and
9.3% errors) as compared to non-carry problems (2,031 ms and 4.7% errors). Finally, a
significant interaction of problem size and carry was present for latencies [F(1, 18) =
8.35, p < .01] but not for ERs [F(1, 18) < 1], indicating that the carry effect for RTs
was more pronounced for addition problems with a large problem (540 ms) size than
for problems with a small problem size (341 ms).2
2 We wish to thank an anonymous reviewer for pointing out that for single digit addition, the need for a carry is inseparably
confounded with overall problem size (see Zbrodoff & Logan, 2005 for a review on the problem-size effect). However, in the
current stimulus set overall problem size was matched between non-carry and carry problems. Nevertheless, this leaves open
the possibility that the carry effect may be confounded with problem size of the unit sum. We agree that this might be the
case. However, as this is not at the heart of the current paper, we wish to refer the interested reader to the recent papers
by Klein and colleagues (2010a,b) who concluded that the carry effect contains both concomitant continuous and categorical
aspects. Evaluating this notion for the current data set indicated that the sum of the unit digits was indeed a reliable predictor
of item RT (b = .18, t = 2.14, p ⬍ .05) in addition to a categorical carry predictor (b = .27, t = 3.23, p ⬍ .01).
In sum, the current results replicated the standard problem size and carry effects as
well as the recent finding of a more pronounced carry effect for relatively larger problem
size (Klein et al., 2010a).
Eye fixation behaviour

In the following, the results of the analyses on the eye fixation data are reported. The
results directly addressing the influence of a carry operation on eye fixation behaviour
will be described following the chronological sequence of processing steps. (1) The
first fixation data are reported reflecting a measure of early processing. (2) The results
for the number of regressive eye-movements are given indicating the transition to later
processing stages. (3) The data on TRTs are described representing possible influences of
late top-down processing (cf. Calvo & Meseguer, 2002). (4) Finally, to evaluate whether
the recognition of the need for a carry operation is associated with either approximate
estimation or exact calculation of the unit sum, specific multiple regression analyses on
the processing of the second summand are reported (see our specific hypotheses above
for details).
For the following analyses, only problems with a large problem size were recruited.
This approach was chosen because the group of problems with a small problem size
involved several problems with a single-digit summand (e.g., 9 + 14 = 23). As in these
trials the number of decade digits varied, we focused on the problems with a large
problem size to ensure that the subsequent evaluation of the distribution of fixations
over tens and units was not confounded by the number of decade digits involved in the
addition problems.
For the sake of readability and to reduce complexity, the full F-, and p statistics of
the overall 2 × 3 × 2 ANOVA conducted for FFD and TRT discerning the factors
carry (needed vs. not needed), problem element (first summand vs. second summand
vs. solution probe), and digit position (decade vs. unit digit) are given in Table 1.
First fixation duration

The ANOVA revealed a main effect of carry [tested one-sided, t(18) = 1.91, p < .05; see
Table 1 for F statistics] and problem element (see Figure 3, Panels A and B). This indicated
that carry addition problems were associated with longer FFDs per digit as compared
to non-carry problems (163 ms vs. 152 ms, respectively). Additionally, fixations were
not distributed equally between the two summands and the solution probe. Post hoc
evaluations by the Games–Howell test indicated that FFD were longest on the second
summand (206 ms) followed by the solution probe and the first summand (141 ms vs.
127 ms, respectively). For the latter two, FFD did not differ reliably.
Moreover, the interaction of carry and problem element was also reliable. Post hoc
comparisons by the Games–Howell test showed that the carry effect for FFD was larger on
the second summand than on the first ( + 27 ms vs. −5 ms, respectively). The carry effect
on the solution probe was in between ( + 10 ms) and did not differ significantly from
that on either summand. Furthermore, problem element and digit position interacted
significantly consistently. Post hoc testing by the Games–Howell test indicated that the
effect of digit position (FFD on the unit digits – FFD on the decade digits) was largest
on the first summand but did not differ between the second summand and the solution
probe ( + 114 ms vs. −44 ms vs. −114 ms, respectively). Additional t-tests revealed
longer FFD on the units of the first summand than on its decade digits [t(18) = 10.33,
p < .001]. In contrast, the decade digits were fixated longer on the solution probe
Table 1. ANOVA outcome for factors incorporated in the analysis for specific carry effects on first
fixation durations (FFD) and total reading times (TRT). For cases in which the sphericity assumption
was violated, the Greenhouse–Geiser coefficient is given allowing for an adjustment of the degrees of
freedom.
FF TRT
Effects and interactions df F GG F GG
Carry 1,18 3.63∗ 0.83 42.29∗∗ -

Problem element 1,18 30.15∗∗ - 35.47∗∗ -
Digit position 1,18 1.19 - 1.40 -
Carry × problem element 2,36 6.64∗∗∗ 0.82 23.33∗∗ -
Problem element × place value stack 2,36 30.52∗∗ 0.84 26.36∗∗ 0.81
Carry × digit position 1,18 ⬍1 - 4.85† -
Carry × problem element × digit position 2,36 6.85† - 12.40∗∗ -
First summand:
Carry × digit position 1,18 ⬍1 - 13.58∗∗∗ -
Second summand:
Carry × digit position 1,18 7.08† - 4.46∗ -
Solution probe:
Carry × digit position 1,18 6.85† - 24.35∗∗ -
Note. ∗∗∗ p ⬍ .001; ∗∗ p ⬍ .01; ∗ p ⬍ .05; † p ⬍ .10.
[t(18) = 3.93, p < .001]. For the second summand, no significant difference was found
[t(18) = 2.03, p > .06].
Finally, the three-way interaction of carry, problem element, and digit position
was reliable. Breaking down this interaction revealed a significant interaction of carry
and digit position for the second summand and the solution probe, while it missed
significance for the first summand. This indicated that a needed carry operation
modulated FFD differentially on the unit digit of the second summand and on the decade
digit of the solution probe (second summand decade vs. unit digit + 2 ms vs. + 53 ms;
solution probe: + 30 ms vs. −8 ms, respectively; see Figure 3, Panel C).3
Taken together, these results are in line with our hypothesis that already very early
processes during the initial encoding of the problem are influenced by the need for a
carry operation. In particular, already at this early stage, the unit digit of the second
summand seemed to be of specific importance for the processing of a carry operation
[see also the results of (3) below]. This makes sense as the existence of a carry can
usually be determined after the unit digit of the second summand has been processed.
However, contrary to our expectations, we also observed early processing of the decade
digit of the solution probe and in particular so for carry trials.
So far, the analysis focused on early stimulus-driven processing of the addition
problem. In contrast, the following analyses on regressive saccades and TRTs will address
later processing stages associated with reprocessing of specific parts of the problem to
assist its successful solution.
3 Please note that the results for gaze durations were identical to those for FFDs and thus were not reported separately.
A First fixation duration for non-carry B First Fixation Duration for carry
Problems Problems
FFD in ms FFD in ms
250 250
200 200
150 150
100 100
50 50
1 2 + 5 1 = 6 3 1 8 + 4 5 = 6 3
C Carry effect on FFD (i.e., FFD B - FFD A)
ms
75
Decade digit
Unit digit
50
25
-25
st nd
1 Summand 2 Summand Solution probe
Figure 3. Distribution of FFD across decade and unit digits of the summands, and the solution probe
separated for non-carry (Panel A) carry problems (Panel B). Additionally, Panel C depicts the carry
effect on FFD in ms (i.e., FFD carry – FFD non-carry) for each digit of the problem separately. Error
bars indicate 1 SEM.
Analyses of regressive saccades

Origin of regressions. To investigate the origin of regressive saccades (i.e., leftward
saccades to a previous interest area), a 2 × 2 ANOVA was conducted separately for
regressions initiated from either tens or units of both the second summand as well as
the solution probe discerning the factors digit position (decade vs. unit digit) and carry
(needed vs. not needed).
The ANOVA evaluating the number of regressive saccades from the second summand
back to the first summand showed that carry addition problems were associated with
a reliably higher number of regressive eye movements [0.87 vs. 0.61 regressions; F(1,
18) = 27.58, p < .001; see Figure 4, Panel A]. The main effect of digit position did
not reach significance [F(1, 18) = 2.37, p = .14]. However, we observed a significant
interaction of carry and digit position [F(1, 18) = 4.26, p = .05] indicating that, in
particular, the number of regressions from the unit digit of the second summand
back to the first summand (as compared to regressions from the decade digit of
the second summand) was increased in carry addition problems ( + 0.37 vs. + 0.15
regressions).
Comparably, also the number of regressive saccades from the solution probe back
to the summands was moderated by the need for a carry [F(1, 18) = 8.68, p < .01;
A
Regressions starting from the second summand ... and landing on the first summand
Decade digit
1
Unit digit
0.75
0.5
0.25
0
Non-carry Carry Non-carry Carry
Problems Problems
B
Regressions starting from the solution probe ... and landing on the first or the second summand
1 Decade digit First summand Second summand

Unit digit
0.75
0.5
0.25
0
Non-carry Carry Non-carry Carry Non-carry Carry
Problems Problems Problems
Figure 4. Panel A (left chart) reflects the number of regressive (leftward) eye-movements elicited from
the second summand separated for decade and unit digit of the summand as well as for non-carry and
carry problems. Analogue with the results for FFD, a stronger increase of the number of regressions
due to the need for a carry can be observed on the unit digits. In the right chart of Panel A, the targets
(i.e., decade or unit digit) of regressive saccades from the second to the first summand are depicted
for non-carry and carry problems, respectively. In Panel B, similar information is given for regressions
from the solution probe (left chart) directed to either the first or the second summand (right chart).
Error bars indicate 1 SEM.
see Figure 4, Panel B]: in carry addition problems, participants looked back to the
summands significantly more often than in non-carry problems (0.54 vs. 0.46 regressions,
respectively). Additionally, the main effect of digit position indicated that more regressive
saccades to the summands started from the decade digit as compared to the unit digit
of the solution probe. This was further specified by the reliable two-way interaction
of carry and digit position [F(1, 18) = 17.24, p < .001] that was opposite to that of
the second summand above. The need for a carry specifically increased the number of
regressive saccades from the decade digit of the solution probe back to the summands
( + 0.15 vs. ± 0 regressions).
To sum up, these results corroborate the particular role of the unit digit of the second
summand and the decade digit of the solution probe for the processing of a carry. The
increase of the number of regressive saccades due to a needed carry was specifically
elevated on these digits.
Target of regressions. To evaluate the target of regressive saccades from the second
summand to the first summand, a 2 × 2 ANOVA with the factors digit position and carry
was conducted. Examining the regressive saccades from the second summand back to
the first summand, the ANOVA showed a reliable effect of carry [F(1, 18) = 11.72, p <
.01; see Figure 4, Panel A] with more regressions to the first summand for carry addition
problems as compared to non-carry problems (0.58 vs. 0.44 regressions, respectively).
Additionally, the main effect of digit position was significant as well [F(1, 18) = 42.20,
p < .001] indicating that the majority of regressive saccades were directed to the unit
digit, when looking back to the first summand (0.83 vs. 0.19 regressions). Furthermore,
the reliable interaction of these two factors [F(1, 18) = 6.55, p < .05] implied that the
increase of regressions towards the first summand due to the need for a carry was more
pronounced on the unit digit than on the decade digit of the first summand ( + 0.13
vs. + 0.05 regressions, respectively). In sum, again, the unit digit was the prominent
target of regressions and in particular so for carry trials.
Regressive saccades originating from the solution probe were analyzed by a 2 ×
2 × 2 ANOVA involving the factors digit position (unit vs. decade digits), carry (needed
vs. not needed), and problem element (first vs. second summand), because regressive
saccades from the solution probe could be targeted at either summand. This ANOVA
revealed a significant carry effect [F(1, 18) = 27.22, p < .001; see Figure 4, Panel B]: more
regressions from the solution probes to the summands were observed for carry than for
non-carry problems. Furthermore, also the two main effects of problem element [F(1,
18) = 18.42, p < .001] and digit position [F(1, 18) = 26.21, p < .001] were reliable. This
indicated that more regressive eye movements were directed to the second summand
as compared to the first summand (0.37 vs. 0.10 regressions, respectively). However, in
contrast to above analysis, regressive saccades from the solution probes were targeted
to the decade than to the unit digits (0.31 vs. 0.16 regressions, respectively) of both
summands.
Moreover, all possible two-way interactions were statistically reliable but not the
three-way interaction [F(1, 18) < 1]. The interaction of problem element and carry
[F(1, 18) = 21.27, p < .001] indicated that the increase of regressions to the summands
due to a carry was larger for the second summand than for the first summand ( + 0.17
vs. ± 0.00 regressions). Additionally, the interaction of summand and digit position [F(1,
18) = 12.71, p < .01] meant that the effect of digit position (i.e., more regressions to
the decade digit) was more pronounced on the second summand ( + 0.26 vs. + 0.03
regressions). Finally, the interaction of carry and digit position [F(1, 18) = 4.50, p < .05]
implied that the increase of regressions towards the two summands due to a required
carry was stronger for regressive eye movements towards the decade as compared to
the unit digit ( + 0.12 vs. + 0.05 regressions, respectively).
In summary, these observations about the target of regressive saccades confirmed the
previous analysis of the start of regressive saccades. The increase of regressive saccades
from the second back to the first summand due to a needed carry was primarily driven
by regressions directed to the unit digit. Regressive eye movements from the solution
probe back to the summands due to a carry were primarily directed to the decade digits
of the summands.
Thus, both origin and target of regressive saccades were unit-digit related for regres-
sions from the second summand and decade-digit related for the solution probe. Most
importantly, this was particularly prevalent when processing carry addition problems.
In this way, the analysis of regressive saccades clearly indicated that the carry operation
is associated with unit-related processes on the summand and decade-related processes

on the solution probe.
Carry effects on TRT

The 2 × 2 × 2 ANOVA with the factors carry, problem element, and digit position
revealed similar results for TRTs, which reflect processes expanding into later processing
stages, as previously observed for FFDs (see analysis ‘Origin of regressions’ above).
Accordingly, the main effects of carry and problem element were significant (see Figure 5,
Panels A and B; see Table 1 for statistical details). This indicated that carry addition
problems were associated with a higher average TRT per digit as compared to non-carry
problems (354 ms vs. 267 ms). Moreover, TRT were longest on the second summand
followed by the first summand and the solution probe (450 ms vs. 293 ms vs. 187 ms,
respectively) as revealed by post hoc evaluations by the Games–Howell test.
Moreover, the interaction of carry and problem element was also reliable. Post hoc
comparisons by the Games–Howell test showed that the increase in TRT due to a
needed carry was most pronounced on the second summand but did not differ between
A Total reading time for non-carry B Total reading time for carry problems
problems
TRT in ms TRT in ms
500 500
400 400
300 300
200 200
100 100
0 0
1 8 + 4 5 = 6 3 1 2 + 5 1 = 6 3
C Carry effect on TRT (i.e., TRT B - TRT A)
ms
Decade digit
200 Unit digit
150
100
50
1st Summand 2nd Summand Solution probe
Figure 5. Total reading times (TRT) on the decade and unit digits of the two summands and the
solution probe of either non-carry (Panel A) or carry problems (Panel B). Panel C indicates the carry
effect on TRT in ms (i.e., TRT carry – TRT non-carry) for each digit (tens and units) of both summands
and the probe. Please note that the carry effect for TRT on the decade digit of the first summand as
well as the unit digit of the solution probe is actually not zero (i.e., −1 ms and + 2 ms, respectively),
but too small to be depicted in the figure at this resolution. Error bars reflect 1 SEM.
the first summand and the solution probe ( + 156 ms vs. + 64 ms vs. + 44 ms,
respectively). Furthermore, problem element and digit position interacted significantly.
Post hoc testing by the Games–Howell test indicated that the effect of digit position
(TRT on the unit digits – TRT on the decade digits) was largest on the first summand
but did not differ between the second summand and the solution probe ( + 399 ms vs.
−114 ms vs. −170 ms). Additional t-tests revealed that this difference was significant
upon the first summand favouring the unit digits [t(18) > 7.22, p < .001], whereas
for the solution probe, the decade digits were fixated longer [t(18) > 3.93, p < .001].
However, for the second summand no significant difference was found [t(18) < 2.03,
p > .06]. In contrast to the ANOVA on FFDs, the two-way interaction of carry and digit
position was significant for TRT. This indicates that the need for a carry led to a reliably
stronger TRT increase, associated with rather later processing stages, on the unit digits
than on the decade digits of the problem ( + 111 ms vs. + 65 ms).
Finally, the three-way interaction of carry, problem element, and digit position was
reliable. Breaking down this three-way interaction into two-way interactions for the
different problem elements revealed a significant interaction of carry and digit position
for the first summand and the solution probe. In contrast, it was only marginally
significant for the second summand. However, the direction of these interactions
differed. Inspection of the marginal means showed that for the first summand as well as
for the second summand, a required carry operation resulted in a more pronounced
increase of TRT on the unit digits compared to the decade digits (first summand:
+ 119 ms vs. ± 0 ms; second summand: + 205 ms vs. + 101 ms; see Figure 4,
Panel C). Contrarily, for the solution probe the carry effect on TRT was stronger on
the decade digit than on the unit digit ( + 80 ms vs. + 2 ms fixations, respectively; see
Figure 5, Panel C).
In sum, the results on TRTs index that the specific and differential processing of the
unit digits of the summands and the decade digits of the solution probe as associated
with a needed carry operation expands from early into later processing stages. Thus,
this implies that during the entire course of processing a carry addition problem, the
unit digits of the summands as well as the decade digit of the result are of particular
importance.
Specific processing of the unit digit of the second summand

To differentiate whether the initial processing of the unit digits of the summands reflects
an approximate estimate of whether a carry is needed or rather involves the computation
of the unit sum, a stepwise multiple regression analysis was conducted. We were
interested which one of the three predictors unit sum, the categorical predictor carry
(needed vs. not needed), and the numerical distance from 10 accounted for variation of
FFD on the second summand per item in the most comprehensive way. As reliable effects
of overall problem size on eye-movement measures have been described in the literature
(e.g., Brysbaert, 1995) and have been replicated in the current study, we conducted the
present analysis on the residual FFD after controlling for overall problem size (observed
FFD – FFD predicted by overall problem size; analogous for the analyses on regressive
saccades and TRT, see below). The final regression model [R = .58, adjusted R2 = .33,
F(1, 46) = 23.63, p < .001] only included the predictor unit sum [constant = −92.49,
un-standardized B = 6.39, standardized b = 5.8, t(46) = 4.86, p < .001]. Inspection of
the beta weight indicated that the residuals of FFD increased with increasing unit sum.
From Figure 6 (Panel A), it can be taken that this means FFD on the second summand is
A Residual FFD in ms
100
50
-50
-100
-150
0 5 10 15 20
Unit Sum
B Residual Regressive Saccades
0.4
0.3
0.2
0.1
0.0
-0.1
-0.2
-0.3
-0.4
-0.5
0 5 10 15 20
Unit Sum
C Residual TRT in ms
400
300
200
100
0
-100
-200
-300
0 5 10 15 20
Unit Sum
Figure 6. Residual FFD (Panel A), residual number of regressions (Panel B), and residual TRT
(Panel C) after overall problem size has been partialled out by an item-based regression analysis
(e.g., overall FFD – FFD as predicted by overall problem size). Thereby, the graphs indicate in which
way residuals are influenced by unit sum. Regression lines indicate which predictor (i.e., continuous or
categorical) was identified to be the best predictor of residual FFD, number of regressions, and TRT,
respectively.
relatively underestimated by overall problem size for small unit sums, whereas it becomes
relatively overestimated for the largest unit sums. Basically, this means that FFD on the
second summand increased with unit sum even after controlling for influences of overall
problem size. Thus, this result suggests that already during the initial processing of the
unit digits of the summands their sum is computed – possibly to evaluate whether a
carry is needed or not.
An identical regression analysis on the residuals of the number of regressive saccades

starting from the unit digit of the second summand (again controlled for overall problem
size) resulted in a final model [R = .68, adjusted R2 = .45, F(1, 46) = 39.01, p <
.001] incorporating the categorical carry predictor as the only significant predictor
[constant = −0.15, un-standardized B = 0.11, standardized b = .68, t(46) = 6.25, p <
.001]. Different from the results for FFD on the unit digit of the second summand, this
result indicated that the number of regressive saccades from the unit digit of the second
summand was not driven by computational processes evaluating the unit sum. Instead,
the influence was categorical with more regressive eye movements from the unit digit
of the second summand occurring for carry as compared to non-carry problems (see
Figure 6, Panel B).
A similar pattern of results was observed when predicting the residuals of TRT on the
unit digit of the second summand (again controlled for overall problem size) by the three
predictors unit sum, the categorical carry predictor and the numerical distance from 10.
As for the number of regressive saccades, the final model R = .70, adjusted R2 = .48, F(1,
46) = 44.21, p < .001] only included the categorical carry predictor [constant = −0.04,
un-standardized B = 87.70, standardized b = .70, t(46) = 6.65, p < .001]. The positive
beta weight synced with the pattern given in Figure 6 (Panel C) indicated that TRT on
the unit digit of the second summand was overestimated for non-carry problems and
underestimated for carry problems. This implied that TRT on the unit digit of the second
summand was higher for carry than for non-carry problems and not associated with the
processing of the unit sum.
Taken together, this pattern of results suggests that processes associated with the
computation of the unit sum occur already during the initial processing of the summands.
On the other hand, these results indicate that regressive eye movements from the unit
digit of the second summand as well as the larger TRTs on this digit seem to be associated
with the processing of the carry procedure itself as the latter variables were primarily
influenced by the categorical absence or need for a carry operation. Thereby, the latter
analyses were informative regarding both processing characteristics and sequencing of
the first two processing steps proposed in the introduction. As FFDs (controlled for
overall problem size) on the unit digit of the second summand increased with unit
sum per se rather than with unit sum approaching ten, the exact unit sum seems to
be computed already during the initial processing of the problem. This also means that
the decision of whether a carry is needed or not is not based upon an approximate
estimate of the unit sum. Consequently, the current data suggest that the sequence of
the processes (1) and (2) as postulated above should be reversed. Furthermore, the
results for regressions and TRT clearly indicate that at later processing stages, it is no
longer unit sum that drives difficulty but rather the categorical question whether or not
a carry is needed.
As an overall re-capitulation of the present analyses, it can be noted that the
results indicated clearly dissociable influences of the need for a carry operation on
the eye fixation patterns on both the summands and the solution probe in the addition-
verification paradigm. On the one hand, the need for a carry operation specifically
increased FFDs, regressive saccades, and TRT on the unit digits of the summands (in
particular on the unit digit of the second summand). On the other hand, identical results
were observed for FFDs, number of regressions, and TRT on the decade digit of the
solution probes in carry addition problems. Additionally, the results of the analyses of
origin and target of the observed regressions indicated that this primary processing of
the decade digits of the solution probe also led to specific refixations of the decade digits
of the summands. Finally, the analysis of whether continuously calculated unit sum or
the presence or absence of carry determined eye fixation behaviour revealed a two-fold
answer. Early processes seemed to be associated with continuous linear unit sum, while
later processes were rather associated with the presence or absence of a carry.
Discussion
The current study set off to investigate the underlying basic processes from which
the increased difficulty of carry addition problems arises. We hypothesized that the
difficulty of carry operations may be determined by at least three main processes: (1) it
has to be recognized whether or not a carry operation is required. (2) The unit digits
of the summands have to be added correctly. And (3), the actual carry procedure has
to be executed. As regards the processing of a carry operation, the present data were
meaningful on all of these three aspects. Taken together, the carry effect in multi-digit
addition seems to originate from (1) increased processing demands for the recognition
whether or not a carry is needed, (2) the actual addition of the unit digits of the summands
as well as (3) the final execution of the carry procedure – even though the exact order
of processes (1) and (2) did not seem mandatory. Results on each of these three aspects
and their implications will be discussed in turn.
Early unit-based computational processes and the necessity of a carry

In line with our hypothesis that recognizing whether a carry is needed may determine
problem difficulty, we observed that the need for a carry already increased FFDs on the
unit digits of the summands. This finding seems reasonable as it is the sum of exactly
these unit digits that indicates whether a carry is needed or not (i.e., unit sum of the
summands ≥10 vs. <10, respectively). Thus, the current data suggest that already during
the initial encoding of the problem (as indicated by FFDs), the need for a carry operation
is evaluated by unit-based processes as soon as it becomes apparent that the sum of
the unit digits becomes ≥10. As this effect was already present in FFDs, it seems to
reflect rather early bottom-up processing of highly specific magnitude information (i.e.,
focusing on unit digits) over different numbers. This is in line with the proposition of
a first integration of information from different numbers as formalized in the model
of eye fixation behaviour in numerical cognition by Moeller and colleagues (2009b).
However, different from our initial hypotheses, the current results suggest that it is
not an approximate estimation process that is employed to evaluate whether or not a
carry is needed. Instead, the finding that already FFDs on the unit digit of the second
summand increases continuously with the unit sum indicated that already during the
initial encoding of the problem unit sum seems to be computed exactly. Thus, the
current data suggest that the actual recognition that a carry is needed seems to be based
on the exact calculation of the unit sum of the summands. Yet, this means that the
sequencing of the two processes (i.e., evaluating whether a carry is needed followed by
the exact computation of the unit sum) as proposed above seems to be reversed actually.
Later unit-based processing of the summands

As expected, the need for a carry operation specifically increased regressive saccades
from and to the unit digits of the summands. In turn, this resulted in a specific inflation
of TRTs upon the unit digits of the summands. However, synced with above findings of
an early computation of the unit sum, it is likely that calculating the exact sum of the
unit digits is already begun during the initial processing of the summands. Additionally,
the results of the multiple regression analyses revealed a categorical carry predictor to
be the best predictor for the occurrence of such regressive eye movements as well as
an increase of TRTs. Thereby, specific regressive saccades from the unit digit of the
second summand seemed to be associated with the need for a carry in an all-or-nothing
manner. Thus, the current data indicate that unit-based processes expand into later
processing stages, probably involving more or less cognitively controlled applications
of calculation rules and/or procedures associated with the recognition that a carry is
needed and successive processes initiated only after the need for a carry operation has
been recognized.
This interpretation of specifically re-processing particular digits is corroborated by a
number of further observations. First, when assuming that participants performed the
addition task from left to right (as implicated by the initial fixation on the left of the first
summand), the unit digit of the second summand is of particular importance. Whenever
the unit sum becomes a two-digit number, a carry is needed and an according procedural
rule has to be applied. Additionally, it has to be kept in mind that the unit digit of this
two-digit unit sum nevertheless represents the unit digit of the correct result, whereas
the tens digit of the unit sum needs to be carried to update the decade digit of the result.
As these processes of carry recognition and unit sum evaluation go beyond the initial
early processing of magnitude information and are only elicited when a carry is needed,
they are initiated only after the exact unit sum has been calculated. Consequently, the
increase of TRTs on the unit digits of the summands should again be more pronounced
on the unit digit of the second summand than on the unit digit of the first summand. The
specific increase of TRT and the pattern of regressive saccades confirmed this hypothesis.
Taken together, this corroborates the notion that carry addition problems specifically
increase processing demands as regards the unit digits of the summands. Synced with
the findings regarding the first processing stage, this suggests that up to the point
where the carry operation is actually executed mental addition is vitally driven by unit-
based processing. However, even though our data provide evidence as regards the
order of early computing of the unit sum and then later deciding on whether a carry
is needed or not, our data do not necessarily suggest a strict sequential stage-based
model. It is well conceivable that these two processes operate in parallel in some kind
of horse race model or interact to a certain degree. In other words, it may be possible
that the processes associated with the categorical processing of the need for a carry
are already activated before the unit sum is calculated to the end entirely – thereby,
allowing for more adaptive processing and a more economical allocation of processing
resources.
Executing the carry – rule-based updating of the decade digit of the solution probe
In line with our expectations, TRTs on the decade digit of the solution probe were
specifically increased in carry addition problems. This finding seems to corroborate the
interpretation that the execution of the carry procedure may represent the application
of a procedural rule. Moreover, in line with the argument that the execution of the
carry may be specifically associated with processing the decade digit of the solution
probe (seeming to reflect processes of updating the decade sum of the summands by
the carry), it was also observed that most regressions from the solution probe back
to the summands started from its decade digit. Additionally, most of these regressions
were directed to the decade digits of the summands. On the one hand, this corroborates
the interpretation that processes concerning the execution of a carry in an addition
verification paradigm involved specific processing of the decade digit of the solution
probe and its constituents, that is, the decade digits of the summands. On the other
hand, the results of analyzing regressive saccades also indicate that the increase of TRT
may be driven by re-inspections of the decade digit of the second summand to come to
the correct decade digit of the solution probe in carry addition problems (see Figure 4,
Panel C) rather than being associated with initially evaluating whether a carry is needed
or not.
However, considering the results of the FFD, such procedural processes in executing
the carry operation may only be a part of the story. Evaluating the eye fixation behaviour
revealed that above described influence of a needed carry operation was already present
for FFDs. This complemented to our hypothesis of rule-based execution of the carry
operation as it implies influences of the carry effect already in very early processing
stages. This may suggest two different aspects: first, following our interpretation of the
effect on FFDs on the summands, this finding may indicate that even during the initial
encoding of the problem, first computational processes upon the appropriateness of the
solution probe may be carried out. The results of these early computations may serve as
the basis for related checking strategies such as excluding the solution probe by checking
the correctness of the decade digit (cf. Moeller et al., in press for empirical evidence
on such estimation/checking strategies). Second, on a broader level, integration of the
results for TRTs and FFDs indicated that the execution of the carry operation does not
seem to be a unitary rule-based process as initially hypothesized. Instead, the current
observations index that the actual execution of the carry operation seems to involve
a variety of cognitive processes, of which rule-based procedures and stimulus-driven
checking strategies may only be two aspects. Hence, it is important to note that the
processes involved in executing the carry operation seem to be much more complex
than we initially hypothesized. Therefore, no exhaustive conclusions on the differential
processing of the solution probe can be drawn.
Thus, further research on the issue of how a carry operation is actually executed is
needed. For instance, electroencephalographic methodologies would allow for a better
(temporal) differentiation between semantic memory and procedural processes (see
Klimesch, 1999; Klimesch, Schack, & Sauseng, 2005 for reviews). Additionally, it has to
be noted that the current results were observed for an addition verification paradigm with
the addition problem being presented horizontally in one line. Therefore, further studies
are desirable replicating the current results for other paradigms and stimulus layouts. In
particular, empirical data on the more common layout of having the summands above
each other as well as data from a production paradigms would be helpful to evaluate
whether the processes identified here are generally valid. However, it has to be kept
in mind that when using a production paradigm, there is actually no solution probe to
look at. Thus, such a study would only be capable of evaluating the current results as
regards the processing of the summands. Nevertheless, as regards the generalizability of
our results, it is promising that Moeller et al. (in press) already observed very similar
results for a choice reaction paradigm.
Conclusions and perspectives

In summary, the present results were meaningful in three respects: First, we were able to
identify the unit digits of the summands to be of specific importance for the processing
of (carry) addition problems. In this context, our data again indicated that – apart from
categorical (non-carry vs. carry) aspects – the carry effect also involves continuous
aspects reflected by the specific influence of the unit sum (see Klein et al., 2010a for a
detailed discussion). Second, the results of analyzing the regressive saccades indicated
that in the current paradigm processes associated with actually calculating the correct
result may not be finished entirely before participants have processed the solution
probe and have decided whether it was the correct solution to the problem at hand
or not. Finally, and most importantly for the research question at hand, the present
observations on the increased processing demands associated with a carry operation
indicated that the carry effect in multi-digit addition seems to originate from at least
three different sources: (1) Unit sum calculation: specific processing of the unit digits
of the summands to compute the sum of the unit digits. (2) Carry detection: further unit-
based processes on the summands associated with recognizing that a carry is needed. (3)
Carry execution: particular decade-based processes reflecting the execution of the carry
procedure. Executing the carry nevertheless seems to be mostly associated with the
processing of the decade digits (of the solution probe). The eye-movement data suggest
that process (1) seemed to precede process (2); although, a strict sequential stage-based
model cannot be derived from the data. Additionally, process (3) is present both in
early and late eye-movement processing data that may suggest that later processes are
already activated while early processes are still not completed entirely. Nevertheless,
our eye-movement study allowed for the first time to dissociate above three processes
underlying the carry operation by evaluating eye fixation behaviour.
Acknowledgements
This research was funded by the German Research Foundation (DFG) by means of a project
within the Research Group (Forschergruppe) Analyse und Förderung effektiver Lehr-Lern-
Prozesse (FOR 738/2/TP02) granted to Ulrike Cress and Hans-Christoph Nuerk supporting
Korbinian Moeller. Additionally, this research was in part funded by a project (Cluster 1, TP 1)
of the ScienceCampus (WissenschaftsCampus) Tübingen Bildung in Informations-umwelten
granted to Hans-Christoph Nuerk, Torsten Grust, Ulrich Trautwein, and Korbinian Moeller.
Finally, we wish to thank Marc Brysbaert for his helpful comments on an earlier version of
the manuscript. In particular, we are grateful to Marc Brysbaert for originally suggesting the
distinction between processes (1) and (2) that was not in the initial version of this manuscript.
His suggestion to distinguish these processes led to a number of additional analysis
(paragraph 4. in the result section) that confirmed the distinction – thereby, exploiting
the potential of eye-tracking data over common RT data.
References
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, 527–538. doi:10.1016/S0096-1515(07)60314-4
Ashcraft, M. H., & Stazyk, E. H. (1981). Mental addition: A test of three verification models. Memory
& Cognition, 9, 185–196. doi:10.3758/BF03202334
Brysbaert, M. (1995). Arabic number reading: On the nature of the numerical scale and the origin
of phonological recoding. Journal of Experimental Psychology: General, 124, 434–452.
Cabeza, R., Locantore, J. K., & Anderson, N. D. (2003). Lateralization of prefrontal activity during
episodic memory retrieval: Evidence for the production-monitoring hypothesis. Journal of
Cognitive Neuroscience, 15, 249–259. doi:10.1162/089892903321208187
Calvo, M. G., & Meseguer, E. (2002). Eye movements and processing stages in reading: Relative
contribution of visual, lexical, and contextual factors. The Spanish Journal of Psychology, 5,
66–77.
Deschuyteneer, M., De Rammelaere, S., & Fias, W. (2005). The addition of two-digit numbers:
Exploring carry versus no-carry problems. Psychology Science, 47, 74–83.
Deutsch, A., Frost, R., Pelleg, S., Pollatsek, A., & Rayner, K. (2003). Early morphological effects
in reading: Evidence from parafoveal preview benefit in Hebrew. Psychonomic Bulletin &
Review, 10, 415–422.
Donohue, S. E., Wendelken, C., Crone, E. A., & Bunge, S. A. (2005). Retrieving rules for behaviour
from long-term memory. NeuroImage, 26, 1140–1149. doi:10.1016/j.neuroimage.2005.03.019
Fürst, A. J., & Hitch, G. J. (2000). Separate roles for executive and phonological components of
working memory in mental arithmetic. Memory and Cognition, 28, 774–782. doi:10.3758/
BF03198412
Goel, V., & Vartarian, O. (2005). Dissociating the rules of the right ventral lateral and dorsal lateral
prefrontal cortex in generation and maintenance of hypotheses in set-shift problems. Cerebral
Cortex, 15, 1170–1177. doi:10.1093/cercor/bhh217
Green, H. J., Lemaire, P., & Dufau, S. (2007). Eye movement correlates of younger and older adults’
strategies for complex addition. Acta Psychologica, 125, 257–278. doi:10.1016/j.actpsy.2006.
08.001
Groen, G. J., & Parkman, J. M. (1972). A chronometric analysis of simple addition. Psychological
Review, 79, 329–343. doi:10.1037/h0032950
Henik, A., & Tzelgov, J. (1982). Is three greater than five: The relation between physical and seman-
tic size in comparison tasks. Memory and Cognition, 10, 389–395. doi:10.3758/BF03202431
Imbo, I., Vandierendonck, A., & De Rammelaere, S. (2007). The role of working memory in the
carry operation of mental arithmetic: Number and value of the carry. The Quarterly Journal
of Experimental Psychology, 60, 708–731. doi:10.1080/17470210600762447
Joseph, H. S. S. L., Liversedge, S. P., Blythe, H. I., White, S. J., Gathercole, S. E., & Rayner,
K. (2008). Children’s and adults’ processing of anomaly and implausibility during reading:
Evidence from eye movements. The Quarterly Journal of Experimental Psychology, 61, 708–
723. doi:10.1080/17470210701400657
Just, M. A., & Carpenter, P. A. (1980). A theory of reading: From eye fixations to comprehension.
Psychological Review, 87, 329–354.
Kalaman, D., & Lefevre, J. A. (2007). Working memory demands of exact and approximate addition.
European Journal of Cognitive Psychology, 19, 187–212. doi:10.1080/09541440600713445
Kazui, H., Kitagaki, H., & Mori, E. (2000). Cortical activation during retrieval of arithmetic facts and
actual calculation: A functional magnetic resonance imaging study. Psychiatry and Clinical
Neurosciences, 54, 479–485. doi:10.1046/j.1440-1819.2000.00739.x
Klein, E., Moeller, K., Dressel, K., Domahs, F., Wood, G., Willmes, K., & Nuerk, H.-C. (2010a). To
carry or not to carry – Is this the question? Disentangling the carry effect in multi-digit addition.
Acta Psychologica, 135, 67–76. doi:10.1016/j.actpsy.2010.06.002
Klein, E., Nuerk, H.-C., Wood, G., Knops, A., & Willmes, K. (2009). The exact vs. approximate
distinction in numerical cognition may not be exact, but only approximate: How different
processes work together in multi-digit addition. Brain and Cognition, 2, 369–381. doi:10.
1016/j.bandc.2008.08.031
Klein, E., Willmes, K., Dressel, K., Domahs, F., Wood, G., Nuerk, H.-C., & Moeller, K. (2010b).
Categorical and continuous – Disentangling the neural correlates of the carry effect in multi-
digit addition. Behavioral and Brain Functions, 6, 70, doi:10.1186/1744-9081-6-70
Klimesch, W. (1999). EEG alpha and theta oscillations reflect cognitive and memory performance:
A review and analysis. Brain Research Reviews, 29, 169–195.
Klimesch, W., Schack, B., & Sauseng, P. (2005). The functional significance of theta and upper
alpha oscillations. Experimental Psychology, 52, 1–10. doi:10.1016/S0165-0173(98)00056-3
Kong, J., Wang, C., Kwong, K., Vangel, M., Chuac, E., & Gollub, R. (2005). The neural substrate
of arithmetic operations and procedure complexity. Cognitive Brain Research, 22, 397–405.
doi:10.1016/j.cogbrainres.2004.09.011
Moeller, K., Fischer, M. H., Nuerk, H.-C., & Willmes, K. (2009a). Eye fixation behaviour in the
number bisection task: Evidence for temporal specificity. Acta Psychologica, 131, 209–220.
doi:10.1016/j.actpsy.2009.05.005
Moeller, K., Fischer, M. H., Nuerk, H.-C., & Willmes, K. (2009b). Sequential or parallel decomposed
processing of two-digit numbers? Evidence from eye-tracking. The Quarterly Journal of
Experimental Psychology, 62, 323–334. doi:10.1080/17470210801946740
Moeller, K., Klein, E., & Nuerk, H.-C. (in press). (No) small adults – children’s processing of carry
addition problems. Developmental Neuropsychology.
Menon, V., Rivera, S. M., White, C. D., Eliez, S., Glover, G. H., & Reiss, A. L. (2000). Functional
optimisation of arithmetic processing in perfect performers. Cognitive Brain Research, 9,
343–345. doi:10.1016/S0926-6410(00)00010-0
Menon, V., Mackenzie, K., Rivera, S. M., & Reiss, A. L. (2002). Prefrontal cortex involvement in
processing incorrect arithmetic equations: Evidence from event-related fMRI. Human Brain
Mapping, 16, 119–130. doi:10.1002/hbm.10035
Nuerk, H.-C., Graf, M., & Willmes, K. (2006). Grundlagen der Zahlenverarbeitung und des
Rechnens [Foundations of number processing and arithmetic]. Sprache, Stimme, Gehör:
Zeitschrift für Kommunikationsstörungen, Schwerpunktthema Dyskalkulie, 30, 147–153.
doi:10.1055/s-2006-951751
Nuerk, H.-C., Moeller, K., Klein, E., Willmes, K., & Fischer, M. H. (in press). Extending the mental
number line – A review of multi-digit number processing. Zeitschrift für Psychologie/Journal
of Psychology.
Ranganath, C., Johnson, M. K., & D’Esposito, M. (2000). Left anterior prefrontal activation increases
with demands to recall specific perceptual information. Journal of Neuroscience, 20, RC108.
doi:10.1016/j.neuroimage.2005.03.019
Rayner, K. (1998). Eye movements in reading and information processing: 20 years of research.
Psychological Bulletin, 124, 372–422. doi:10.1037/0033-2909.124.3.372
Rayner, K., & Pollatsek, A. (1987). Eye movements in reading: A tutorial review. In M. Coltheart
(Ed.), Attention and performance XII: The Psychology of reading (pp. 327–362). Hove, UK:
Lawrence Erlbaum Associates.
Rayner, K., & Pollatsek, A. (1989). The psychology of reading. Englewood Cliffs, NJ: Prentice Hall.
Rayner, K., Warren, T., Juhasz, B. J., & Liversidge, S. P. (2004). The effect of plausibility on
eye movements in reading. Journal of Experimental Psychology: Learning, Memory and
Cognition, 30, 1290–1301.
Torbeyns, J., Verschavel, L., & Ghesquiere, P. (2002). Strategic competence: Applying Siegler’s
theoretical and methodological framework to the domain of simple addition. European
Journal of Psychology of Education, 17, 275–292. doi:10.1007/BF03173537
Traxler, M. J., & Pickering, M. J. (1996). Plausibility and the processing of unbounded
dependencies: An eye-tracking study. Journal of Memory and Language, 35, 454–475.
doi:10.1006/jmla.1996.0025
Wong, K. F. E., & Chen, H.-C. (1999). Orthographic and phonological processing in reading
Chinese text: Evidence from eye fixations. Language and Cognitive Processes, 14, 461–480.
doi:10.1080/016909699386158
Zago, L., Pesenti, M., Mellet, E., Crivello, F., Mazoyer, B., & Tzourio-Mazoyer, N. (2001). Neural
correlates of simple and complex mental calculation. NeuroImage, 13, 314–327. doi:10.1006/
nimg.2000.0697
Zbrodoff, N. J., & Logan, G. D. (2005). What everyone finds: The problem size effect. In J. I. D.
Campbell (Ed.), Handbook of mathematical cognition (pp. 331–345). New York: Psychology
Press.
Received 24 February 2010; revised version received 7 February 2011

British J of Psychology - 2011 - Moeller - Three Processes Underlying The Carry Effect in Addition Evidence From Eye

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

British J of Psychology - 2011 - Moeller - Three Processes Underlying The Carry Effect in Addition Evidence From Eye

Uploaded by

Copyright:

Available Formats

20448295, 2011, 3, Downloaded from https://bpspsychub.onlinelibrary.wiley.com/doi/10.1111/j.2044-8295.2011.02034.x by The Open University, Wiley Online Library on [12/03/2023].

Three processes underlying the carry effect

Korbinian Moeller1,2 ∗ , Elise Klein2,3 and Hans-Christoph

Recent research indicated that processes of unit-decade integration pose particular

The carry effect and associated processes

Eye tracking in numerical cognition research

Three cognitive processes underlying the carry operation

Objectives of the current study

Task, stimuli, and design

RT and error data

Eye fixation behaviour

First fixation duration

Effects and interactions df F GG F GG

Carry 1,18 3.63∗ 0.83 42.29∗∗ -

Note. ∗∗∗ p ⬍ .001; ∗∗ p ⬍ .01; ∗ p ⬍ .05; † p ⬍ .10.

Analyses of regressive saccades

1 Decade digit First summand Second summand

is associated with unit-related processes on the summand and decade-related processes

Carry effects on TRT

1st Summand 2nd Summand Solution probe

Specific processing of the unit digit of the second summand

An identical regression analysis on the residuals of the number of regressive saccades

Early unit-based computational processes and the necessity of a carry

Later unit-based processing of the summands

Conclusions and perspectives

Received 24 February 2010; revised version received 7 February 2011

You might also like