Learning and Instruction 23 (2013) 125e135

Contents lists available at SciVerse ScienceDirect

Learning and Instruction

journal homepage: www.elsevier.com/locate/learninstruc

How training on exact or approximate mental representations of number can

enhance first-grade students’ basic number processing and arithmetic skills
Andreas Obersteiner a, *, Kristina Reiss a, Stefan Ufer b
a
Heinz Nixdorf-Stiftungslehrstuhl für Didaktik der Mathematik, TUM School of Education, Technische Universität München, Schellingstr. 33, D-80799 München, Germany
b
Lehrstuhl für Didaktik der Mathematik und Informatik, Mathematisches Institut, Ludwig-Maximilians-Universität München, Theresienstr. 39, D-80333 München, Germany

a r t i c l e i n f o a b s t r a c t

Article history: Theories of psychology and mathematics education recommend two instructional approaches to develop
Received 7 February 2012 students’ mental representations of number: The “exact” approach focuses on the development of exact
Received in revised form representations of organized dot patterns; the “approximate” approach focuses on the approximate
19 August 2012
representation of analogue magnitudes. This study provides for the first time empirical evidence for the
Accepted 21 August 2012
specific effects of these approaches by implementing them in a highly controlled learning environment.
147 first-graders were randomly assigned to one of three intervention groups that used an “exact”, an
Keywords:
“approximate”, or both versions of the same computer game, or to a control group. Performance on tasks
Mental number representations
Approximate number system
requiring exact or approximate number processing as well as achievement in arithmetic were measured
Computer-based intervention before and after the intervention. Results show that performance improved on tasks related to the exact
Numerical development or approximate number aspect trained, but there was no crossover effect. Achievement in arithmetic
Arithmetical competence increased for the experimental groups and tended to be higher after only exact or only approximate
training. Implications for teaching and learning in the classroom are discussed.
Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction 1.1. Cognitive systems for numerical representation

Children at a very young age seem to be able to mentally
represent small numerosities precisely and larger numerosities
approximately. Through instruction, children are supposed to
further develop mental representations of numbers that are
thought to be fundamental for higher-order arithmetical achieve-
ment. The two primary approaches to achieving this are (1) using
tasks and external representations that emphasize the exact
meaning of numbers and (2) using tasks and external representa-
tions that emphasize the approximate meaning of numbers. Both
approaches are well grounded in theory and supported by some
empirical evidence. However, to our knowledge, no study has
directly compared the differential effects of both approaches in
a controlled experimental setting. In this article, we present an
experimental study where we contrasted the two approaches at the
middle of grade 1, when children learn to deal with numbers up to
20. In the following, we elaborate on the mental representations of
numbers, how these representations are related to further arith-
metical development, and how they can be enhanced by
instruction.
* Corresponding author.
E-mail address: andreas.obersteiner@tum.de (A. Obersteiner).
Evidence from infant studies (e.g., Feigenson & Carey, 2003;
Feigenson, Carey, & Hauser, 2002; Feigenson & Halberda, 2004;
Wynn, 1992; Xu, 2003; see Barth, Baron, Spelke, & Carey, 2009),
brain imaging (e.g., Dehaene, Piazza, Pinel, & Cohen, 2003; Kucian,
Aster, von Loenneker, Dietrich, & Martin, 2008; Piazza, Mechelli,
Butterworth, & Price, 2002; Pinel, Dehaene, Riviere, & LeBihan,
2001), and experimental psychology (e.g., Dehaene & Cohen,
1994; Mandler & Shebo, 1982; Sekuler & Mierkiewicz, 1977;
Simon, Peterson, Patel, & Sathian, 1998) suggests that basic
numerical processing is achieved through two distinct cognitive
systems that represent numerical information (see Feigenson,
Dehaene, & Spelke, 2004, for a review). While one system repre-
sents small numerosities precisely, the other system represents
larger numerosities approximately. Representation in the “exact”
system is thought to be discrete, with each object in the external
world being represented by a mental entity. In the “approximate”
system, numerosities are represented analogously, and the repre-
sentation gets more imprecise with increasing magnitude.
During numerical development, these non-verbal representations
become linked to verbal representations, such as number words, and
to symbolic representations, such as Arabic numbers. According to
models developed by Dehaene (1992) and Von Aster (2000), the two
non-verbal representations of number are integrated in one module,

0959-4752/$ e see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.learninstruc.2012.08.004
126 A. Obersteiner et al. / Learning and Instruction 23 (2013) 125e135
which is responsible for the semantic understanding of numerical
information. Because they are developed prior to the other modules,
the two non-verbal systems are thought to be the “core systems of
number” (Feigenson et al., 2004, p. 307).
1.2. Mental number representations and basic number processing
The exact cognitive number system is thought to underlie the
quick recognition of small numerosities, for which the term subitizing
has been coined (e.g., Mandler & Shebo, 1982). When individuals are
asked to determine the numerosity of a set of objects, they have been
found to be almost equally fast and precise for numerosities up to
three or four, but response times and error rates increase rapidly for
larger sets. This holds true when objects are presented in random
order. For organized arrangements, performance is faster and more
precise, because groups of objects can be seen as one entity repre-
senting a larger quantity (“conceptual subitizing”, cf. Clements,
1999). Performance on enumeration tasks with small numbers of
dots or with organized dot patterns can therefore be used as
a measure of subitizing and conceptual subitizing skills, respectively.
While successfully performing these tasks is thought to rely on
the exact number system (see Section 1.1), number comparison
tasks have frequently been used as a measure of the approximate
number system (e.g., De Smedt, Verschaffel, & Ghesquière, 2009;
Sekuler & Mierkiewicz, 1977; Szücs & Goswami, 2007). In these
tasks, the larger of two numerosities (non-symbolic number
comparison or magnitude comparison) or the larger of two number
symbols (symbolic number comparison) has to be determined. As it
is not necessary (rather disadvantageous) to determine the exact
difference of the numbers (or numerosities, respectively), or even
to represent the numbers (resp. numerosities) exactly, number
comparison is regarded to rely on the approximate number system
(see Gilmore, Attridge, & Inglis, 2011). Another task relying on the
approximate number system is approximate calculation (e.g.,
Kucian et al., 2006). A typical task in this respect is to decide
whether 7 þ 8 is closer to 10 or 16.
Performance on both types of tasks has proven to be relevant for
further numerical development in children. For example, Landerl,
Bevan, and Butterworth (2004; see Koontz & Berch, 1996; Penner-
Wilger, Fast, LeFevre, & Smith-Chant, 2007) found that eight to
nine year old children with severe problems in mathematics also
showed deficits in subitizing. The same seems to hold true for
conceptual subitizing (Mulligan, Mitchelmore, & Prescott, 2006).
An even larger body of evidence has shown that approximate
number processing is related to arithmetical achievement. In
a longitudinal study, Halberda, Mazzocco, and Feigenson (2008)
found that performance on a non-symbolic number comparison
task at grade nine was retrospectively predictive for mathematics
achievement in each year from kindergarten to grade 6. For the
symbolic number comparison task, De Smedt et al. (2009) showed
that performance in grade 1 was substantially correlated with
performance in arithmetic in grade 2. Siegler and Opfer (2004)
asked children to indicate the position of a given number symbol
on an empty number line. Performance on this task was related to
achievement in mathematics in kindergartners as well as in
students in grade 1 and grade 2.
The assumption that performance on the abovementioned basic
number processing tasks is relevant for arithmetic development is
largely based on correlation or regression analysis.1 There are only
a small number of intervention studies, in which the development
1
The same holds true for the relevancy of other basic number processing tasks,
such as counting (Aunio & Niemivirta, 2010; Krajewski & Schneider, 2009) or
focusing on numerosity (Hannula & Lehtinen, 2005).
of mental number representations in children was enhanced and
the impact on achievement in arithmetic was evaluated. Those
studies differ in the instructional approach they pursue to enhance
either exact or approximate mental number representations.
1.3. Two main approaches to enhance mental number
representations
Following different theories about numerical development,
interventions differ in the types of problems children are asked to
solve and in the way numbers are externally represented. Some
researchers argue that the approximate number system is most
important for numerical development (e.g., Dehaene, 2009). This
would imply to prefer approximate external representations of
numerical information rather than exact representations. Approx-
imate numerical information can be represented using discrete
units like dots, when students are asked to estimate the numer-
osities of large sets of objects. A more frequently recommended
approach is to use analogue representations of numbers, such as an
empty number line (e.g., Beishuizen, 1999; Klein, Beishuizen, &
Treffers, 1998), where students are asked to indicate the approxi-
mate position of numbers. In both cases, the relative numerical
magnitude is more important than the exact one. Recognizing
relative number magnitudes quickly is important for calculation, as
it helps to find elaborated strategies. For instance, instead of adding
9, it may be easier to add 10 and take away 1. In order to recognize
that this strategy is helpful a quick representation of the relative
magnitudes of 9 and 10 is required.2
Other researchers argue that recognizing small numerosities
(based on subitizing) is the most fundamental skill for numerical
development (e.g., Clements, 1999; Butterworth, 2005). To make
use of subitizing, larger sets of dots should be represented in
organized patterns, so that numerosity judgements can be made
through conceptual subitizing (cf. Section 1.2). Moreover, using
structures of ten could motivate students to group dots in tens and
therefore prepare them to understand the decimal number system.
In German grade 1 classrooms, the ten-frame and the twenty-frame
are popular examples for such an instructional approach. In addition
to groups of ten, these external representations make use of groups
of five so that conceptual subitizing can be developed progressively.
The twenty-frame is considered particularly helpful when students
learn to add single-digit numbers with carries over ten. The theo-
retical assumption is that externally representing such additions
could result in mental representations with equivalent structures.
There is limited evidence for each of the abovementioned
instructional approaches. Kucian et al. (2011; cf. Siegler & Ramani,
2009; Ramani & Siegler, 2011) trained preschool children with low
numerical abilities using a computer game, in which the player had
to navigate a landing rocket onto the correct position on a horizontal
number line. The game aimed at developing a mental number line.
Indeed, Kucian et al. could show that after a training phase of 15 min,
five days a week, for five weeks, students performed better on
arithmetical tasks. Particularly high training gains were found for
low-achieving children on number line tasks. Training effects could
also be found on brain activation in parietal brain areas, hinting to
a qualitative change of mental processing after training and sup-
porting the existence of an approximate cognitive number system
(see Section 1.1). Wilson, Dehaene, Dubois, and Fayol (2009) also
aimed at the development of children’s approximate mental
number representation. In their game “The Number Race” (Wilson
et al., 2006), the player has to make number comparison
2
This is of course not sufficient to solve the task completely. To do so, exact
calculation is necessary as well.
 A. Obersteiner et al. / Learning and Instruction 23 (2013) 125e135
judgements with symbolic and non-symbolic numbers (sets of
dots), and move the game characters on a linear game board.
Kindergartners participated in six training sessions of 20 min each.
Compared to a control group, children improved significantly in
symbolic number comparison but not in non-symbolic comparison
or in addition, indicating that they learned to tighten the link
between symbolic and non-symbolic number representations.
While these studies showed that there were positive effects of
training focusing on the approximate number system, Frank and
Barner (2011) explored the effects of expertise in the use of exact
visual images for specific numbers (cf. Stigler, 1984). The abacus is
a physical device that represents numbers via beads that are
arranged in columns, with each column representing a place value.
After intensive training, visual images of specific abacus states may
help to mentally represent exact numbers. Frank and Barner found
that children who were experts in mental abacus showed specific
performance patterns, thus hinting at a mental representation of
exact abacus states.
These findings support the hypotheses that approximate and exact
instructional approaches have a positive impact on numerical devel-
opment compared to a control condition, but they do not reveal if one
approach is advantageous over the other. There is only one study that
has directly compared both approaches (Räsänen, Salminen, Wilson,
Aunio, & Dehaene, 2009): One intervention group used the
“Number Race” (Wilson et al., 2006; see above) to enhance approxi-
mate mental number representations, and the other group used the
game “Graphogame-Math”, which uses exact numerosities and
number symbols to enhance the exact representation of numbers. The
player’s task in the latter game is to select the correct visually pre-
sented number that corresponds to an auditory probe. Visual numbers
are represented in organized dot patterns, number symbols or addi-
tions and subtractions. In the study by Räsänen et al. (2009), 30
preschool children underwent daily trainings over a three weeks
period. The main result was that the Graphogame-Math group
showed particular improvement in comparing small numbers, while
the Number Race group showed particular improvement in
comparing larger numbers. The authors interpret their findings in
terms of different strategies that children used to perform comparison
tasks. While children of the Graphogame-Math group had learned to
represent exact numbers, children of the Number Race group had
activated an approximate number representation, thus out-
performing the other group with larger numbers. However, it can be
argued that the two groups played different games that differed also in
their overall design. Furthermore, it is an open question whether
a combination of both trainings adds up to even better performance.
To summarize, there is limited empirical evidence showing that
specific training based on either an exact or an approximate
instructional approach has positive effects compared to no specific
training. However, there is no evidence so far contrasting the effects
of an exact and an approximate instructional approach within
a rigorously controlled learning environment. Moreover, the impact
on achievement in arithmetic in a broader sense is still unclear.
2. Research questions and hypotheses
The aim of our study was to implement an “exact” (using
organized dot patterns to enhance exact mental number repre-
sentations) and an “approximate” (using random dot patterns and
analogue representations to enhance approximate mental number
representations) instructional approach in a highly controlled
manner, and to compare the effects on achievement in arithmetic.
We addressed two major research questions:
a) Are there differential effects of exact or approximate training
on basic number processing skills in children at grade 1?
127
Table 1
Study design and sample sizes.
“Approximate” training
Yes No
“Exact” training Yes AE E
Approximate & exact Exact
n ¼ 39 n ¼ 39
No A CG
Approximate Control group
n ¼ 35 n ¼ 34
For both kinds of training, learning gains were expected to be
higher than in a control condition without training (Hypothesis 1).
Considering the theoretical distinction between an exact and an
approximate cognitive system for representing numbers, differen-
tial effects were expected, that means, the effects are restricted to
the type of numerical processing (exact or approximate) that is
trained (Hypothesis 2).
b) Are there differential effects on achievement in arithmetic?
Both instructional approaches were expected to have a positive
impact on achievement in arithmetic (Hypothesis 3). Due to the
lack of empirical evidence we were not able to predict whether one
approach had a larger effect than the other.
3. Methods
3.1. Sample and study design
To investigate differential effects of exact and approximate
training, we chose a 2  2-design with pre and post measures. The
exact and the approximate treatments were realized in half of the
sample, resulting in four intervention groups (see Table 1): Group A
(“approximate”) received approximate training only, group E (“exact”)
received exact training only. Group AE (“approximate þ exact”) was
treated with both trainings, alternating in each session, and group CG
(“control group”) did not have any numerical training. Instead, this
group received language training.
The original sample consisted of N ¼ 204 first-grade students from
eleven classrooms at four German primary schools. The students were
randomly assigned to one of the four groups with the restriction that
each group was represented equally within each classroom. Due to
illness during the intervention or test sessions, there were 57 drop-
outs, resulting in a final sample of 147 children (mean age: 6.91 years,
SD ¼ 0.39, 75 female). The final numbers of students in each group are
displayed in Table 1. There were no significant group differences with
respect to age, F(3, 143) ¼ 1.33, p ¼ .267, or the number of female
students, c2 (3, N ¼ 147) ¼ 4.39, p ¼ .223.
3.2. Computer games
To implement the trainings in a highly controlled manner we
opted for a computer-based intervention. We designed two
versions of the same computer game to make sure that training
effects were due to the differences in the relevant aspects rather
than in the overall game design. “The Number Race” is a computer
game that has been developed by Wilson et al. (2006; see Section
1.3) with the aim of remediating dyscalculia. We used this open
source software3 as a basis and developed two new game versions.
The overall design of the original game is described in detail by
3
The software is freely available under http://www.unicog.org/main/pages.php?
page¼NumberRace.
128 A. Obersteiner et al. / Learning and Instruction 23 (2013) 125e135
Fig. 1. First (left) and second (right) screen of the approximate version.
Wilson et al. (2006). The game that is introduced in the following
contains elements of the original game as well as newly developed
ones. Section 3.2.1 refers to both the approximate and the exact
game version and Sections 3.2.2 and 3.2.3 refer to the specific
elements of each game version separately.
3.2.1. Overview of the game4
Basically, the game consists of two screens. On the first screen the
player can collect diamonds by solving a numerical task. Tasks are
represented in symbolic format using Arabic symbols, in non-
symbolic format (diamonds), or in verbal format (spoken
language). Solving a task means to choose the correct of two possible
answers that appear on the left and on the right side of the screen,
respectively. The player has to respond quickly, because otherwise
the computer-driven competitor character will collect the diamonds.
Once the player has made his/her choice, the second screen appears,
where he/she can use his/her diamonds to move his/her and the
competitor’s character on a game board, consisting of 40 squares.
Each player is allowed to move the number of squares corresponding
to the number of diamonds he/she has collected. When the player
reaches the end of the board first, he/she wins the game and collects
a reward. With enough rewards collected the player can unlock an
additional character to be used in the next games.
There is an adaptive algorithm, adjusting to the player’s
performance in three dimensions. The first dimension increases the
difficulty of the tasks (see Sections 3.2.2 and 3.2.3). The second
dimension is speed. The player is forced to give faster responses due
to the increasing speed of the competitor. The third dimension,
conceptual complexity, increases task difficulty by specific features
that are defined by 44 game levels. For example, with increasing
game level, quantities are represented more frequently in symbolic
format rather than in non-symbolic format. The contents of the
levels were designed to be as highly comparable as possible
between the two versions of the game (see Appendix, for detailed
descriptions). The game level progressively increases by 1 when
performance is high. The number range for the tasks is 1e10 for the
first 22 game levels, and is extended to 20 for further levels. In
game level 23, the board on the second screen is being extended to
60 squares, so that the duration of a game is sufficiently long. The
algorithm adapts to the player’s performance on the last 20 tasks,
and targets at an average performance of 75% correct.
As the aim of the intervention was to foster students’ non-verbal
exact or approximate mental number representations, the player is
encouraged not to use counting strategies for moving the charac-
ters on the board. The player is forced to click directly on the correct
4
Further development of the new game versions is still ongoing. The software
(German version only) will be available on request by the end of 2013.
end square rather than on each square the character would pass.
Only then the character would move.
To make sure that students understood how to play the game,
there were tutorial video-clips that were presented at the begin-
ning and at specific moments throughout the game. For instance,
when a player used counting strategies to move his/her character,
a video-clip was presented that explained more elaborated strate-
gies (depending on the game version: estimating the distances on
the game board or using the game board’s structure; see Sections
3.2.2 and 3.2.3) to find the correct position on the board.
Several process data such as the number of tasks performed,
response times, and accuracy are saved in log-files (see Section 4.1).
3.2.2. Approximate version
The approximate game version is similar to the original game
developed by Wilson et al. (2006). It involves tasks for which an
approximate representation of numbers and numerosities is
advantageous over the use of exact representations. There are
number comparison tasks, in which the player has to choose the
larger of two quantities. Depending on the level, there are estima-
tion and approximate calculation tasks. With these tasks, an addi-
tional display panel (hereafter referred to as “upper panel”) appears
on the first screen and two fish appear on the left and on the right
side of the screen (see Fig. 1). In estimation tasks, a quantity appears
in the upper panel. The two fish at the bottom, which are introduced
to be very bad at maths, make guesses on the number of the
quantity presented. The player has to choose the better guess. While
one of these guesses differs only 1 from the exact number, the other
guess differs more than 2. Considering the time pressure posed by
the computer-driven competitor, the player has to rely on an
approximate strategy to make a fast and correct choice. In approx-
imate calculation tasks, the amount of diamonds is presented as an
addition or subtraction problem. Difficulty is increased by
decreasing the ratio between the numbers that need to be compared
and by choosing larger numbers. In the approximate version, sets of
diamonds are always presented in random arrangements.
On the second screen (see Fig. 1) the game board appears. Unlike
in the exact game version (see Section 3.2.3), the squares of the
board are not visible, so that the board is more similar to a contin-
uous line and the player has to rely on approximate strategies in
order to move his/her character to the correct position. Symbolic
numbers on the right of the board indicate the number of squares
after each line.
3.2.3. Exact version
The exact version of the game involves tasks that cannot be
solved reliably by only activating approximate mental representa-
tions, but require an exact processing of numbers and numerosities.
The upper panel is present in all the tasks, and a quantity is presented
 A. Obersteiner et al. / Learning and Instruction 23 (2013) 125e135
Fig. 2. First (left) and second (right) screen of the exact version.
(see Fig. 2). The player’s task is to identify the same quantity on the
left or on the right. While on one side the exact quantity appears, the
quantity on the other side is by 1 larger or smaller. Thus, the player
cannot rely on an approximate strategy but has to identify the exact
amount. Difficulty is increased by increasing the numbers. In the
exact version, sets of diamonds are always presented in grouped
patterns with structures of five and ten (twenty-frame).
Like in the original version there are squares on the game board
of the second screen (see Fig. 2). In addition, the board is divided
into sections of five squares. The player can use the structure in
order to move his/her character and does not need to rely on
approximation or counting strategies.
3.2.4. Control group
Students of the control group used the software “Oriolus”
(Oriolus Lernprogramme, 2010), a learning environment on German
language for grades 1e4. Children can choose tasks from different
units (e.g., word spelling, reading) with increasing difficulty.
3.3. Pre and post measures
Pre and post measures were designed to tap performance on
exact or approximate basic number processing tasks that were
closely related to the intervention, and performance on arithmet-
ical tasks that were not directly trained. Basic number processing
was assessed by computerized tests; achievement in arithmetic
was assessed by a paper-pencil achievement test.
3.3.1. Tests of basic number processing
We distinguished between “exact” and “approximate” basic
number processing tasks. The former type of tasks required deter-
mining numerosities precisely, whereas for the latter type of tasks
approximate number processing was more efficient (see Section 1.2).
Exact number processing tasks included subitizing and conceptual
subitizing, approximate number processing tasks included magni-
tude comparison, number comparison, and approximate calculation.
All tasks were administered in groups of 12e18 children. Each
student sat in front of one computer and used headphones. Video-
based instruction was given before each task. After the instruction,
practice trials were presented and immediate feedback was given.
During the test phase, no feedback was given. Response times and
accuracy were recorded by the computer, using E-Prime 2.0 software
(Schneider, Eschman, & Zuccolotto, 2002).
3.3.1.1. Subitizing (3 items). A fixation cross (2 s) was followed by
a number of randomly arranged same-sized dots in black colour
against a white background. Students were instructed to press the
space bar as soon as they knew the correct number of dots. The
129
time limit was 15 s.5 After pressing the space bar, the students had
to type in their response. Relevant keys were labelled on the
keyboard. Each numerosity from 1 to 20 appeared exactly once in
random order. To assess subitizing, only items with 1e3 dots were
considered for the analysis, as larger numerosities would reflect
counting skills.6
3.3.1.2. Conceptual subitizing (20 items). The presentation was
nearly the same as in the subitizing task. However, dots were given
in either red or blue colour (randomized across items) and in
structures of five and ten, just as they usually appear on a twenty-
frame. The time limit was 10 s.
3.3.1.3. Magnitude comparison (20 items). After a fixation cross,
sets of randomly arranged dots appeared on the left and on the
right side of the screen for a maximum of 4 s. Students had to
indicate the larger amount of dots by pressing a corresponding key.
The difference between the magnitudes varied from 1 to 12. There
were two trials for each of the differences 1e8 and one trial for each
of the differences 9e12. Order of presentation was randomized. In
one half of the trials the total area of both sets of dots was equal, in
the other half dot sizes were equal. The side of the correct answer
(left or right) was counterbalanced across trials.
3.3.1.4. Number comparison (20 items). This task was nearly the
same as the magnitude comparison task, with the only difference
that number symbols (sized approximately 8  4 cm) were pre-
sented instead of dots. The selected numbers were equivalent in
their magnitudes to those of the magnitude comparison tasks.
3.3.1.5. Approximate calculation (10 items). After a fixation cross, an
addition or subtraction task appeared on the screen for 3 s,
followed by two numbers, one on the left and one on the right side
of the screen. Students had to estimate the result of the calculation
and to choose the nearest number. The alternative numbers
differed 1 or more than 2 from the exact result and were presented
on the left or on the right equally often.
3.3.2. Test of achievement in arithmetic
Items for the paper-pencil arithmetic test were taken from the
“Hamburger Rechentest” (Lorenz, 2007). The test consisted of 45
5
Time limits for this and the following tasks were adjusted according to the
results of a pilot study.
6
Dividing dot enumeration into a subitizing range and a counting range is
a commonly applied method to assess subitizing (e.g., Landerl et al., 2004). As we
did not aim at enhancing counting abilities, these data are not considered for
further analysis.
130 A. Obersteiner et al. / Learning and Instruction 23 (2013) 125e135
Table 2
Log-file data for the three experimental groups.
Group A Group E
M SD M SD
Level max 40.14 5.42 41.03
Tasks
Total number 206.89 31.92 222.72
Accuracy rate (%) 83.82 6.31 78.10
Response time (msec) 3296.00 1198.14 4051.28
Clicks on game board
Total number 1879.31 912.44 846.00
Accuracy rate (%) 22.69 10.90 50.50
items distributed over the following five subscales: On the task
“preceding/succeeding numbers” (10 items), the students were pre-
sented one of three consecutive numbers (up to 20) and asked to find
the remaining two numbers, on “number sequences” (7 items), the
students had to recognize the pattern of a number sequence made of
four numbers in order to continue the sequence by two more
numbers, on “number ordering” (6 items), the students had to place
four randomly arranged numbers in the correct order, on “addition/
subtraction” (16 items), the students had to solve addition and
subtraction tasks, on “number line” (6 items), the students were
asked to identify the correct numbers that corresponded to two
marked positions on a number line. There were 0-to-10 number lines
on which the locations of 0, 5 and 10 were labelled, and 0-to-20
number lines on which the locations of 0, 5, 10, 15, and 20 were
labelled. For each subscale, there was a time limit between 1 and
3 min. The reliability of the test was high (Cronbach-a ¼ .86 for pre-
test and .87 for post-test).
3.4. Procedure
Students whose parents had given written informed consent were
randomly assigned to one of the four intervention groups (see Section
3.1). Half of the students (about 12) of one classroom at a time
participated in each intervention session in a separate room. For all
groups, there were 10 training sessions of 30 min each, over a period
of 4 weeks. The students also participated in regular classroom
instruction. During the intervention sessions, two trained student
assistants supervised the children and were prepared to give poten-
tial advice. They were instructed to watch students’ performance and
to start tutorial video-clips (see Section 3.2.2) when required.
3.5. Statistical analysis
As the computerized tasks are easy to solve for first-grade
students, high accuracy rates with ceiling effects were expected.7
Thus, only response time data were considered to evaluate the
treatment effects.8 Error trials and trials with response times that
deviated more than two standard deviations from the individual
mean were excluded from further analysis. If all items of one task
had maximum response time (indicating inattention or problems
with task procedure), this student’s data were excluded from
analyses of the respective task. Furthermore, two outliers were
excluded for number comparison (post measurement), and six
outliers for subitizing (one at pre and five at post measurement).
7
For example, high accuracy rates for number comparison tasks in students at
the beginning of first-grade are reported by De Smedt et al. (2009).
8
Accuracy data was only considered to analyze students’ success on training
tasks, but not on pre and post-test measures.
Group AE All
M SD M SD
5.84 39.00 5.92 40.05 5.75
34.45 208.05 37.45 212.75 35.23
9.88 79.94 6.06 80.50 7.95
1053.91 3650.39 864.85 3678.98 1077.41
479.20 1324.67 600.98 1331.26 793.14
21.83 28.86 12.07 34.42 19.77
To analyze the treatment effects we conducted a two-way
analysis of covariance for each task, including the factors Approxi-
mate (yes/no) and Exact (yes/no), with pre-test performance as
covariate and post-test performance as the dependent variable. A-
priori-contrasts (with one-tailed t-tests) were used to compare
performance of the three experimental groups (A, E, AE) to the
control group (CG).
4. Results
We first checked whether there were any significant group
differences in pre-test measures. As expected, this was the case
neither for any of the basic number processing tasks nor for the
arithmetic achievement test, Fs < 2.15, ps > .097. To check the
success of the intervention we first report on the results of process
data analyses. Then we present the intervention effects on basic
number processing and on arithmetical achievement.
4.1. Manipulation check
Table 2 shows descriptive results of the log-files for the three
experimental groups (A, E, AE). One-way analyses of variance with
the factor Group (A, E, AE) revealed that there were no significant
differences in the maximum game level the students achieved, F(2,
110) ¼ 1.22, p ¼ .299, or in the number of tasks the students played,
F(2, 110) ¼ 2.46, p ¼ .091. Therefore, students of both experimental
groups had a comparable exposure to the intervention tasks.
With respect to mean accuracy rates for the tasks in the game,
there were significant differences between groups, F(2, 110) ¼ 5.30,
p ¼ .006, partial h2 ¼ 0.088. Post-hoc Scheffé-tests revealed that
only the difference between groups A and E was significant
(p ¼ .007), but not between the other groups (ps > .098), with
higher accuracy rates for group A. With respect to the mean
response times there was also a significant effect of Group, F(2,
110) ¼ 4.87, p ¼ .009, partial h2 ¼ 0.081. Students of group A were
faster in solving the tasks than students of group E (p ¼ .010),
without significant differences between the other groups
(ps > .240). The tasks in the approximate game version seemed to
be slightly easier than the tasks in the exact version.
There was a significant effect of Group on the total number of
clicks on the game board, F(2, 110) ¼ 21.34, p < .001, partial
h2 ¼ 0.280, with higher numbers of clicks for group A, followed by
group AE and group E (all group differences significant, ps < .05).
There was also a significant effect of Group on the relative number
of times the students clicked onto the correct square, F(2,
110) ¼ 31.99, p < .001, partial h2 ¼ 0.368, with higher accuracy rates
for group E (ps < .001) but no difference between groups A and AE
(p ¼ .265). These data indicate that for students in the exact version
it was easier to find the correct square on the board, whereas
students in the approximate version found it hard to find the
 A. Obersteiner et al. / Learning and Instruction 23 (2013) 125e135 131

Table 3
Estimated marginal means and standard errors of the mean for response times (in
msec) in basic number processing tasks.

Group A Group E Group AE Group CG

Subitizing
Conceptual subitizing
Magnitude comparison
Number comparison
Approximate calculation
N M SE M SE M SE M SE
135 1568 84 1643 78 1649 82 1789 85
144 2450 125 2152 118 2365 118 2618 125
147 795 24 893 23 802 23 883 24
145 888 25 921 24 892 24 965 26
147 1768 99 1877 99 1797 99 2114 108

control group revealed significantly higher performance for the

Fig. 3. Mean level as a function of the number of tasks played, for the three experi-
mental groups. Number of tasks is restricted to 227, as less than 25% of students played
even more tasks. Thus, the data is not reliable for larger numbers of tasks. Note that in
this figure levels are numbered from 0 to 43.
correct square, so that they clicked more often onto incorrect
squares. This finding can be explained by the differences in the
board design (see Section 3.2). In the approximate version, the
squares were not visible, so that students were forced to use an
approximate strategy in order to find the correct position. In the
exact version, students could use the structure of the board. All in
all, the strategies used on the game board seem to be as intended.
The content of tasks differed between the versions of the game.
To compare students’ success throughout the game, we plotted the
game level that a student had reached as a function of the number
of tasks he/she had solved. There was some deviation between
individual students. However, when we compared the group means
(see Fig. 3), the graphs were nearly identical. Thus, the students of
each group proceeded in the game in a highly comparable way.
In summary, though there were slight differences in difficulty
with respect to the tasks and the proceeding on the game board, the
overall success in the game versions was highly comparable
between the experimental groups.
4.2. Effects on basic number processing
As expected, there was a decrease in reaction times from pre to
post measurement for the whole sample for all tasks, indicating
a general improvement after a period of five weeks.9 The estimated
marginal means of the analyses of covariance for all groups are
displayed in Table 3.
For subitizing, there was no significant main effect of Exact, F(1,
129) ¼ 0.15, p ¼ .696, partial h2 ¼ 0.001, or Approximate, F(1,
129) ¼ 1.70, p ¼ .195, partial h2 ¼ 0.013, and also no interaction
effect, F(1, 129) ¼ 1.89, p ¼ .171, partial h2 ¼ 0.014, indicating no
specific effects of the approximate or the exact intervention. Only
the a-priori-comparison between the experimental groups and the
9
These data are not presented here, as we were interested in group differences
after the training rather than in the overall improvements from pre to post-test.
experimental groups as compared to the control group,
t(131) ¼ 1.73, p ¼ .044, d ¼ 0.43.
For the conceptual subitizing task, there was a significant main
effect for Exact, F(1, 139) ¼ 5.10, p ¼ .025, partial h2 ¼ 0.035, with
higher performance for groups with exact training as compared to
groups without exact training. No significant effects were found for
Approximate, F(1, 139) ¼ 0.03, p ¼ .855, partial h2 < 0.001, or the
interaction, F(1, 139) ¼ 2.45, p ¼ .120, partial h2 ¼ 0.017. The a-
priori-contrast between experimental and control groups showed
an advantage for the experimental groups, t(141) ¼ 2.09, p ¼ .019,
d ¼ 0.56.
For magnitude comparison, there was a main effect of Approx-
imate, F(1, 142) ¼ 14.15, p < .001, partial h2 ¼ 0.091, indicating
higher performance for groups with approximate training, but no
significant effect of Exact, F(1, 142) ¼ 0.12, p ¼ .730, partial
h2 ¼ 0.001, and also no interaction effect, F(1, 142) < 0.01, p ¼ .953,
partial h2 < 0.001. Again, experimental groups showed higher
performance than the control group, t(144) ¼ 1.77, p ¼ .039,
d ¼ 0.44.
The same picture appeared for the number comparison task,
namely a significant main effect of Approximate, F(1, 140) ¼ 4.51,
p ¼ .036, partial h2 ¼ 0.031, with higher performance for groups
with approximate training. The effect of Exact was not significant,
F(1, 140) ¼ 0.68, p ¼ .411, partial h2 ¼ 0.005, and neither was the
interaction effect, F(1, 140) ¼ 0.97, p ¼ .325, partial h2 ¼ 0.007.
Contrasting experimental and control groups confirmed higher
performance for experimental groups, t(142) ¼ 2.19, p ¼ .015,
d ¼ 0.30.
For the approximate calculation task the analysis also revealed
a significant effect of Approximate, F(1, 141) ¼ 4.37, p ¼ .038, partial
h2 ¼ 0.030, indicating higher performance for groups with
approximate training as compared to groups without approximate
training, but no effect of Exact, F(1, 141) ¼ 1.03, p ¼ .313, partial
h2 ¼ 0.007, and no interaction effect, F(1, 141) ¼ 1.68, p ¼ .198,
partial h2 ¼ 0.012. The experimental groups performed significantly
better than the control group, t(143) ¼ 2.46, p ¼ .008, d ¼ 1.65.
To summarize, the approximate training had positive effects
only on tasks that were supposed to rely on the approximate
mental representation of number. The exact training had positive
effects only on the conceptual subitizing task, which was supposed
to rely on the exact number representation, but not on subitizing.
Contrasting the experimental groups and the control group
revealed significantly higher performance for the experimental
groups in all tasks.
4.3. Effects on arithmetical achievement
Comparing the mean scores at pre (M ¼ 24.28, SD ¼ 6.13) and
post (M ¼ 27.86, SD ¼ 6.50) measurement revealed the expected
overall improvement in arithmetical achievement for the whole
sample. The estimated marginal means of the analysis of covariance
are displayed in Table 4.
132 A. Obersteiner et al. / Learning and Instruction 23 (2013) 125e135

Table 4
Estimated marginal means of arithmetical performance (test score).

Group A Group E Group AE Group CG

M SE M SE M SE M SE
Arithmetical achievement 28.72 0.85 28.68 0.80 27.51 0.80 26.43 0.86
Preceding/succeeding numbers 10.03 0.23 9.42 0.21 9.41 0.21 9.14 0.23
Number sequences 3.87 0.33 4.26 0.32 3.96 0.32 3.66 0.34
Number ordering 4.68 0.21 4.74 0.20 4.64 0.20 4.52 0.21
Addition/subtraction 7.32 0.43 7.95 0.41 7.34 0.41 7.04 0.44
Number line 2.36 0.25 2.50 0.24 2.50 0.24 1.96 0.26

The analysis of covariance revealed no significant effect of Exact,
F(1, 142) ¼ 0.39, p ¼ .533, partial h2 ¼ 0.003, and no significant
effect of Approximate, F(1, 142) ¼ 0.46, p ¼ .499, partial h2 ¼ 0.003.
However, there was a significant interaction effect, F(1, 142) ¼ 4.39,
p ¼ .038, partial h2 ¼ 0.030.
Fig. 4 illustrates this effect. While group A and group E showed
equal performance on a high level, group AE showed lower
performance than both groups, but still better performance than
the control group. The exact and the approximate training seemed
to be equally effective and to be more effective when they were not
combined. Post-hoc analyses revealed that only the comparisons of
group A versus control group and group E versus control group
were marginally significant (p ¼ .059 and p ¼ .057 respectively),
while all other group differences were not significant (ps > .300).
However, the a-priori-contrast comparing the experimental groups
to the control groups was significant, t(144) ¼ 1.91, p ¼ .029,
d ¼ 0.40, revealing higher performance for the experimental groups
as compared to the control group.
To analyze potential training effects on the various aspects of
arithmetical achievement in more detail, we analyzed each
subscale of the arithmetical achievement test separately.10 The
descriptive data (see Table 4) suggest that for almost all subscales,
students of group E had highest performance and students of the
control group had lowest performance. However, there were no
significant effects of Exact, F(1, 142) < 1.86, p > .174, or Approximate,
F(1, 142) < 0.66, p > .422, on any of these subscales, and there were
no interaction effects either, F(1, 142) < 1.16, p > .287.
which relies on the exact number representation, only the exact but
not the approximate approach improved performance. This means
that there was no crossover effect between enhanced approximate
and exact number processing, corroborating the theory that
approximate and exact number processing rely on distinct cogni-
tive systems (see Section 1.1; Hypothesis 2). Hence, neither the
approximate nor the exact instructional approach in isolation was
sufficient to enhance all kinds of numerical skills (cf. Butterworth,
2005; Dehaene, 2009), so that both aspects of number processing
need to be addressed by instruction.
Students of all experimental groups together improved their
subitizing skills more than students of the control group, but there
were no main effects for exact or approximate training. Thus,
numerical training had a general positive effect on the skill to
quickly determine small numerosities. In view of the relatively
large response times for subitizing and the method of assessment,
it could be discussed if we indeed assessed what is usually labelled
“subitizing”, or if improvements in performance could also be due
to enhanced counting or a general improvement in giving fast
responses (see below). Considering that the subitizing range
increases only from about 3 in young children to about 5 in adults,
the trainability of this skill may be very limited and for the most
part restricted to response time measures.
We found a main effect of exact training on conceptual subitizing.
In the conceptual subitizing task, exactly the same organized dot
patterns were used as in the exact game version. Thus, the effect
could be restricted to the specific representation rather than

5. Discussion

We contrasted an “exact” and an “approximate” instructional

approach to enhance mental representations of numbers in first-
grade students. To realize these approaches we used two
different versions of one computer game (based on “The Number
Race” by Wilson et al., 2006), that differed in the relevant aspects
with respect to an exact or an approximate instructional approach
but were identical in the overall design. The log-file data confirmed
that performance throughout the game was highly comparable
across the experimental groups. We were interested in the differ-
ential effects of each approach.
As we had hypothesized, the increase of performance on basic
numerical processing from pre-test to post-test was more
pronounced for students who had received mathematical training
compared to students who had received language training
(Hypothesis 1). This result showed that basic number processing
skills could be enhanced within ten training sessions.
For each task relying on the approximate number representa-
tion, only the approximate but not the exact instructional approach
Fig. 4. Interaction effect between the “approximate” and the “exact” teaching
improved performance. Analogously, for conceptual subitizing,
approach with respect to arithmetical achievement: The diagram shows the estimated
marginal means on the post-test of arithmetical achievement, corrected for pre-test
performance. Performance was highest for students who received only approximate
10
Due to ceiling effects in the subscale “preceding/succeeding numbers” in pre- but not exact or only exact but not approximate training, followed by combined
test and post-test, this subscale was not taken into account. training and the control group (neither exact nor approximate training).
 A. Obersteiner et al. / Learning and Instruction 23 (2013) 125e135
revealing a general improvement in conceptual subitizing. Never-
theless, we may argue that this finding is of particular interest to
mathematics education. The use of organized dot patterns in math-
ematics classroom instruction is considered to support numerical
learning (e.g., McGuire, Kinzie, & Berch, 2012). There are strong
theoretical assumptions that the use of such representations
enhances the development of equivalent mental representations in
students (cf. Chao, Stigler, & Woodward, 2000). However, to our
knowledge, there is no empirical evidence so far that students
understand the concept of the twenty-frame. On the contrary, there
is even some evidence that low-achieving children had problems in
understanding the concept of manipulatives with organized patterns
even after they had used them in the classroom (Lorenz, 1998). Our
findings provide some evidence that students e at least after specific
training e improve their skills in recognizing the correct number on
the twenty-frame. As the students received no further instruction on
the concept of the twenty-frame, we can be confident that during
regular classroom instruction, most students will understand this
concept.
With respect to the effects of approximate training on comparison
tasks, we found even higher effect sizes for magnitude comparison
than for number comparison. In contrast, Wilson et al. (2009) found
positive effects on number comparison tasks but not on magnitude
comparison tasks in low-achieving preschool children. Our diverging
findings may be due to the more intensive training (300 min versus
120 min) or to the method of measurement. While in the Wilson et al.
study children had to respond to cards with dots that were presented
in individual settings and the number of correctly answered items
was used as a measure, we used computerized tests and response
time measures. It is possible that improvement in performance could
only be measurable by means of response times but not accuracy
rates. This raises the more general question on how reliable we can
measure very basic numerical skills. In a recent study, Gilmore et al.
(2011) showed that performance on different tasks involving
approximate number processing were not correlated. Therefore, we
should be cautious about interpreting performance on single tasks as
a measure of the related mental representation. Further research is
certainly needed to clarify this issue.
With regard to arithmetical achievement, we found a significant
positive effect of numerical training as compared to language
training (Hypothesis 3). This finding supports the idea that training
on very basic number processing can have a positive impact on
arithmetical achievement in a broader sense. This finding adds to
the literature, where the relationship between basic number pro-
cessing and arithmetical achievement has mostly been analyzed by
regression or correlation analyses (Aunio & Niemivirta, 2010; De
Smedt et al., 2009; Halberda et al., 2008; Hannula & Lehtinen,
2005; Krajewski & Schneider, 2009; Siegler & Opfer, 2004), but
only rarely by intervention studies.
In spite of an overall advantage of numerical training, we found
no main effects for exact or approximate training. There was
a significant interaction effect, indicating that performance for the
groups that had received only exact or only approximate training
tended to be higher than for group AE that had received both
trainings, followed by the control group. However, group differ-
ences were not or only marginally significant. This finding makes it
difficult to clearly answer the question which instructional
approach was the most effective one. We would like to point out
two tendencies that our data reveal: First, a combination of both
trainings did not lead to higher learning gains than language
training. Students of the combined group had a reduced amount of
exact training and also a reduced amount of approximate training.
Thus, their lower performance could be due to lower exposure to
each training. On the other hand, students may have been confused
because they had to switch between different types of tasks
133
between the sessions. From a mathematics education perspective,
a combination of both approaches is certainly necessary to develop
flexible numerical understanding in the long run. Our results do not
support this assumption at first sight. However, we should be
cautious when generalizing our findings. We trained very specific
skills, using a very restricted method, and connections between the
two skills were not part of the training. Moreover, compared to the
long-lasting development of number concepts, our intervention
was very short. Second, none of the two teaching approaches was
advantageous over the other. Instead, both approximate and exact
training led to equal performance in arithmetic. The descriptive
results pattern suggests that both enhanced approximate number
processing (as highlighted by Dehaene, 2009) and enhanced exact
number processing (as highlighted by Butterworth, 2005) can
contribute to performance in arithmetic. Further research studying
the effects of basic number processing on different facets of arith-
metic skills is certainly needed (see below).
All in all, the effects of the training programmes on arithmetical
achievement were quite small. This is in line with other interven-
tion studies (see Räsänen et al., 2009) and it reveals once more that
though basic number processing is certainly an important prereq-
uisite, arithmetical achievement is a complex construct involving
other important facets. For example, several studies have shown
that counting abilities are an important prerequisite for higher-
order arithmetical achievement (e.g., Aunio & Niemivirta, 2010;
Krajewski & Schneider, 2009). In our study, counting abilities
were considered neither in the training nor in the tests. Thus, there
could be other relevant basic skills that were not taken into
account. From a theoretical perspective, exploring the specific
relationship between basic number processing and further arith-
metical development is still an important challenge, as it may
indeed not be as clear as it has been suggested by the above-
mentioned studies that used correlation or regression analyses. For
example, Schneider, Grabner, and Paetsch (2009) found that the use
of the internal mental number line was virtually unrelated to
mathematical achievement in 5th and 6th graders, when several
other measures were taken into account.
In which way approximate or exact number processing skills
contribute to different facets of arithmetical achievement (see
Sections 1.2 and 1.3) has to remain unanswered by our data. This
counts even more, as we did not find any significant effects of
training on the subscales of the arithmetical achievement test, not
even for tasks that seem to be strongly related to one of the training
programme (e.g., “number line” should be related to the approxi-
mate training). This could also hint to the fact that working on an
arithmetic paper-pencil test is a much more complex task than the
very basic number processing that was required in the computer-
based tests.
Considering the methodology of our study, we could overcome
several challenges that might appear in classroom studies. First, all
the groups spent exactly the same period of time on the inter-
vention. In particular, also the control group used the computer for
the same time periods. Second, the students were randomly
assigned to the intervention groups. This is particularly important
to rule out effects of classroom instruction or other differences of
the environment. Third, even if the intervention groups in the study
by Räsänen et al. (2009) both used a computer game, these games
were not identical in the overall conception. By using different
versions of the same game, we succeeded in implementing
different treatments in a highly controlled way.
It may be regarded as a limitation of our study that students of
the control group used a computer game that was very different
from the game versions of the experimental groups. In particular,
no fast responses were required in the control group game. Thus,
the effects could partly be due to a general improvement in giving
134 A. Obersteiner et al. / Learning and Instruction 23 (2013) 125e135

fast responses to visually presented stimuli. However, we also
measured general speed in giving fast responses in a computerized
task (data not reported here). When we used this measure as an
additional covariate, the effects were only marginally reduced.
6. Conclusions
To our knowledge, this is the first study that compared the exact
and the approximate instructional approach to enhance mental
representations of numbers in a highly controlled learning environ-
ment. The finding that directly trained skills were improved separately
and that each approach led to equal achievement in arithmetic
contributes to the educational literature in two ways. First, it supports
the theory of distinct cognitive systems for the representation of
approximate or exact numerical information. This can also affect
classroom instruction where approximate tasks such as estimation or
approximate calculation play a minor role as compared to exact
calculation. Second, our data are in line with the idea that the use of
external representations (and related problems) can lead to
improvement in students’ understanding of such representations. This
is an important empirical finding that supports theories from
psychology and mathematics education but also experience from
classroom practice. More empirical data could help to corroborate
theories about the effects of using external representations of numbers
in the classroom and its impact on arithmetical achievement.
Acknowledgements
This work was supported by research grant 01 JG 0922 from the
German Federal Ministry of Education and Research (BMBF). The
authors wish to thank the student assistants that supervised the
students and helped with data collection. We also would like to
thank the teachers, students and their parents for participating and
for the excellent collaboration.

Appendix

Table 5
Levels of the difficulty dimension “conceptual complexity” for the approximate game version.

Level Format Format Fade upper Task Instructional goal

upper panel response panel
1e2 ns ns No Number comparison Attention to and approximate processing of numerosities
3 ns s No Estimation Approximate processing; link symbolic representation to numerosities
4 ns s Yes (2 s) Estimation Increase reliance on symbolic representation
5 sþv ns No Estimation Approximate processing; link numerosities to symbolic representation
6 sþv ns Yes (2 s) Estimation Increase attention to symbolic representation
7 s þ ns þ v s þ ns Yes (4 s) Number comparison Use quantity information of symbolic and non-symbolic number
representations to compare quantities
8e9 s þ ns þ v s þ ns Yes (2 s) Number comparison Increase reliance on symbolic representation
10 s s No Estimation Complete reliance on symbolic representation
11 ns s No Approximate addition Approximate quantity representation in addition
12 s s No Approximate addition Approximate addition with complete reliance on symbolic representation
13 s s Yes (2 s) Approximate addition Increase fluency in symbolic approximate addition
14e15 s s No Number comparison, Fluency in representing quantity of additions
addition problems
16 ns s No Approximate subtraction Approximate quantity representation in subtraction
17 s s No Approximate subtraction Approximate subtraction with complete reliance on symbolic representation
18 s s Yes (2 s) Approximate subtraction Increase fluency in symbolic approximate subtraction
19e20 s s No Number comparison, Fluency in representing quantity of subtractions
subtraction problems
21e22 s s No Number comparison, Fluency and flexibility in representing quantity of additions and subtractions
addition/subtraction problems
23e44 Same contents as for levels 1e22 in the number range 10e20

Note. ns ¼ Non-symbolic; s ¼ symbolic; v ¼ verbal.

Table 6
Levels of the difficulty dimension “conceptual complexity” for the exact game version.

Level Format upper panel Format response Fade Task Instructional goal
1 ns ns No Number identification Attention to and identification of organized dot patterns
2e3 ns ns Yes (2 s) Number identification Increased attention to and identification of organized dot patterns
4 ns s No Number identification Link symbolic representation to organized dot pattern
5e6 ns s Yes (2 s) Number identification Increase reliance on symbolic representation
7 sþv ns No Number identification Link organized dot pattern to symbolic representation
8e9 sþv ns Yes (2 s) Number identification Increase attention to symbolic representation
10e11 ns ns No Addition Identify result of addition of organized dot patterns
12 ns s No Addition Link result of addition of organized dot patterns to symbolic representation
13 ns s Yes (2 s) Addition Increase fluency in linking results to symbolic representation
14 s ns No Addition Link result of symbolic addition to organized dot pattern
15 s ns Yes (2 s) Addition Increase fluency in linking result to organized dot pattern
16 ns þ s ns No Subtraction Identify result of subtraction of number symbol from organized dot pattern
17 ns þ s s No Subtraction Identify result of subtraction of number symbol from organized dot pattern
18 ns þ s s Yes (2 s) Subtraction Increase fluency in identifying result of subtraction
19 s ns No Subtraction Identify organized dot pattern as result of symbolic subtraction
20 s ns Yes (2 s) Subtraction Increase fluency in symbolic subtraction
21e22 s ns Yes (4 s) Addition/subtraction Fluency and flexibility in symbolic addition/subtraction and
identifying corresponding organized dot pattern
23e44 Same contents as for levels 1e22 in the number range 10e20

Note. ns ¼ Non-symbolic; s ¼ symbolic; v ¼ verbal.

 A. Obersteiner et al. / Learning and Instruction 23 (2013) 125e135 135

References Landerl, K., Bevan, A., & Butterworth, B. (2004). Developmental dyscalculia and
basic numerical capacities: a study of 8e9-year-old students. Cognition, 93, 99e
125. http://dx.doi.org/10.1016/j.cognition.2003.11.004.
Aunio, P., & Niemivirta, M. (2010). Predicting children’s mathematical performance
Lorenz, J. H. (1998). Anschauung und Veranschaulichungsmittel im Mathemati-
in grade one by early numeracy. Learning and Individual Differences, 20, 427e
kunterricht. Mentales visuelles Operieren und Rechenleistung (2nd ed.). Göttin-
435. http://dx.doi.org/10.1016/j.lindif.2010.06.003.
gen: Hogrefe.
Barth, H., Baron, A., Spelke, E., & Carey, S. (2009). Children’s multiplicative trans-
Lorenz, J. H. (2007). Hamburger Rechentest. Manual. Hamburg: Freie und Hansestadt
formations of discrete and continuous quantities. Journal of Experimental Child
Hamburg, Behörde für Bildung und Sport.
Psychology, 103(4), 441e454. http://dx.doi.org/10.1016/j.jecp.2009.01.014.
Mandler, G., & Shebo, B. J. (1982). Subitizing: an analysis of its component processes.
Beishuizen, M. (1999). The empty number line as a new model. In I. Thompson (Ed.),
Journal of Experimental Psychology: General, 11, 1e22. http://dx.doi.org/10.1037/
Issues in teaching numeracy in primary schools (pp. 157e168). Buckingham:
0096-3445.111.1.1.
Open University Press.
McGuire, P., Kinzie, M. B., & Berch, D. B. (2012). Developing number sense in pre-k
Butterworth, B. (2005). The development of arithmetical abilities. Journal of Child
with five-frames. Early Childhood Education Journal, 40, 213e222. http://
Psychology and Psychiatry, 46(1), 3e18. http://dx.doi.org/10.1111/j.1469-
dx.doi.org/10.1007/s10643-011-0479-4.
7610.2005.00374.x.
Mulligan, J. T., Mitchelmore, M. C., & Prescott, A. (2006). Integrating concepts and
Chao, S., Stigler, J. W., & Woodward, J. A. (2000). The effects of physical materials on
processes in early mathematics: the Australian pattern and structure mathematics
kindergartners’ learning of number concepts. Cognition and Instruction, 18,
awareness project (PAS-MAP). In J. Novotná, H. Moraová, M. Krátká, & N. Stehlíková
285e316. http://dx.doi.org/10.1207/S1532690XCI1803_1.
(Eds.). Proceedings of the 30th annual conference of the International Group for the
Clements, D. H. (1999). Subitizing: what is it? Why teach it? Teaching Children
Psychology of Mathematics Education, Vol. 4 (pp. 209e216). Prague: PME.
Mathematics, 5(7), 400e405.
Oriolus Lernprogramme. (2010). Oriolus-Lernprogramme. Deutsch 1e4. Kleinsen-
De Smedt, B., Verschaffel, L., & Ghesquière, P. (2009). The predictive value of
delbach: Oriolus Lernprogramme.
numerical magnitude comparison for individual differences in mathematics
Penner-Wilger, M., Fast, L., LeFevre, J., & Smith-Chant, B. L. (2007). The foundations
achievement. Journal of Experimental Child Psychology, 103(4), 469e479. http://
of numeracy: subitizing, finger gnosia, and fine-motor ability. In
dx.doi.org/10.1016/j.jecp.2009.01.010.
D. S. McNamara, & J. G. Trafton (Eds.), Proceedings of the 29th annual conference
Dehaene, S. (1992). Varieties of numerical abilities. Cognition, 44, 1e42. http://
of the Cognitive Science Society (pp. 1385e1390). Austin, TX: Erlbaum.
dx.doi.org/10.1016/0010-0277(92)90049-N.
Piazza, M., Mechelli, A., Butterworth, B., & Price, C. J. (2002). Are subitizing and
Dehaene, S. (2009). Origins of mathematical intuitions: the case of arithmetic.
counting implemented as separate or functionally overlapping processes?
Annals of the New York Academy of Sciences, 1156, 232e259. http://dx.doi.org/
NeuroImage, 15, 435e446. http://dx.doi.org/10.1006/nimg.2001.0980.
10.1111/j.1749-6632.2009.04469.x.
Pinel, P., Dehaene, S., RiviËre, D., & LeBihan, D. (2001). Modulation of parietal
Dehaene, S., & Cohen, L. (1994). Dissociable mechanisms of subitizing and counting:
activation by semantic distance in a number comparison task. NeuroImage,
neuropsychological evidence from simultanagnosic patients. Journal of Experi-
14(5), 1013e1026. http://dx.doi.org/10.1006/nimg.2001.0913.
mental Psychology: Human Perception and Performance, 20(5), 958e975. http://
Ramani, G. B., & Siegler, R. S. (2011). Reducing the gap in numerical knowledge
dx.doi.org/10.1037/0096-1523.20.5.958.
between low- and middle-income preschoolers. Journal of Applied Develop-
Dehaene, S., Piazza, M., Pinel, P., & Cohen, L. (2003). Three parietal circuits for
mental Psychology, 32, 146e159. http://dx.doi.org/10.1016/j.appdev.2011.02.005.
number processing. Cognitive Neuropsychology, 20, 487e506. http://dx.doi.org/
Räsänen, P., Salminen, J., Wilson, A. J., Aunio, P., & Dehaene, S. (2009). Computer-
10.1080/02643290244000239.
assisted intervention for children with low numeracy skills. Cognitive Devel-
Feigenson, L., & Carey, S. (2003). Tracking individuals via object-files: evidence from
opment, 24, 450e472. http://dx.doi.org/10.1016/j.cogdev.2009.09.003.
infants’ manual search. Developmental Science, 6, 568e584. http://dx.doi.org/
Schneider, M., Grabner, R. H., & Paetsch, J. (2009). Mental number line, number line
10.1111/1467-7687.00313.
estimation, and mathematical achievement: their interrelations in grades 5 and
Feigenson, L., Carey, S., & Hauser, M. (2002). The representations underlying infants’
6. Journal of Educational Psychology, 101(2), 359e372. http://dx.doi.org/10.1037/
choice of more: object files versus analogue magnitudes. Psychological Science,
a0013840.
13(2), 150e156. http://dx.doi.org/10.1111/1467-9280.00427.
Schneider, W., Eschman, A., & Zuccolotto, A. (2002). E-Prime user’s guide. Pittsburgh:
Feigenson, L., Dehaene, S., & Spelke, E. (2004). Core systems of number. Trends in
Psychology Software Tools Inc.
Cognitive Sciences, 87, 307e314. http://dx.doi.org/10.1016/j.tics.2004.05.002.
Sekuler, R., & Mierkiewicz, D. (1977). Children’s judgments of numerical inequality.
Feigenson, L., & Halberda, J. (2004). Infants chunk object arrays into sets of individuals.
Child Development, 48, 630e633.
Cognition, 91, 173e190. http://dx.doi.org/10.1016/j.cognition.2003.09.003.
Siegler, R. S., & Opfer, J. E. (2004). Development of numerical estimation in young
Frank, M. C., & Barner, D. (2011). Representing exact number visually using mental
children. Child Development, 75, 428e444. http://dx.doi.org/10.1111/j.1467-
abacus. Journal of Experimental Psychology: General, 141(1), 134e149. http://
8624.2004.00684.x.
dx.doi.org/10.1037/a0024427.
Siegler, R. S., & Ramani, G. B. (2009). Playing linear number board games e but not
Gilmore, C., Attridge, N., & Inglis, M. (2011). Measuring the approximate number
circular ones e improves low-income preschoolers’ numerical understanding.
system. Quarterly Journal of Experimental Psychology, 64(11), 2099e2109. http://
Journal of Educational Psychology, 101(3), 545e560. http://dx.doi.org/10.1037/
dx.doi.org/10.1080/17470218.2011.574710.
a0014239.
Halberda, J., Mazzocco, M. M. M., & Feigenson, L. (2008). Individual differences in
Simon, T. J., Peterson, S., Patel, G., & Sathian, K. (1998). Do the magnocellular and
non-verbal number acuity correlate with maths achievement. Nature, 455,
parvocellular visual pathways contribute differentially to subitizing and
665e669. http://dx.doi.org/10.1038/nature07246.
counting? Perception and Psychophysics, 60(3), 451e464. http://dx.doi.org/
Hannula, M. M., & Lehtinen, E. (2005). Spontaneous focusing on numerosity and
10.3758/BF03206866.
mathematical skills of young children. Learning and Instruction, 15(3), 237e256.
Stigler, J. W. (1984). “Mental abacus”: the effect of abacus training on Chinese
http://dx.doi.org/10.1016/j.jecp.2010.06.004.
children’s mental calculation. Cognitive Psychology, 16(2), 145e176. http://
Klein, A., Beishuizen, M., & Treffers, A. (1998). The empty number line in Dutch
dx.doi.org/10.1016/0010-0285(84)90006-9.
second grades. Realistic versus gradual program design. Journal for Research in
Szücs, D., & Goswami, U. (2007). Educational neuroscience: defining a discipline for
Mathematics Education, 29, 443e464.
the study of mental representations. Mind, Brain, and Education, 1, 114e127.
Koontz, K. L., & Berch, D. B. (1996). Identifying simple numerical stimuli: processing
http://dx.doi.org/10.1111/j.1751-228X.2007.00012.x.
inefficiencies exhibited by arithmetic learning disabled children. Mathematical
Von Aster, M. (2000). Developmental cognitive neuropsychology of number pro-
Cognition, 2, 1e24. http://dx.doi.org/10.1080/135467996387525.
cessing and calculation: varieties of developmental dyscalculia. European Child
Krajewski, K., & Schneider, W. (2009). Early development of quantity to numbere
& Adolescent Psychiatry, 9, II/41eII/57.
word linkage as a precursor of mathematical school achievement and mathe-
Wilson, A. J., Dehaene, S., Dubois, O., & Fayol, M. (2009). Effects of an adaptive game
matical difficulties: findings from a four-year longitudinal study. Learning and
intervention on accessing number sense in low-socioeconomic-status kinder-
Instruction, 19, 513e526. http://dx.doi.org/10.1016/j.learninstruc.2008.10.002.
garten children. Mind, Brain, and Education, 3(4), 224e234. http://dx.doi.org/
Kucian, K., Aster, M., von Loenneker, T., Dietrich, T., & Martin, E. (2008). Develop-
10.1111/j.1751-228X.2009.01075.x.
ment of neural networks for exact and approximate calculation: a fMRI study.
Wilson, A. J., Dehaene, S., Pinel, P., Revkin, S. K., Cohen, L., & Cohen, D. (2006).
Developmental Neuropsychology, 33, 447e473. http://dx.doi.org/10.1080/
Principles underlying the design of “the number race”, an adaptive computer
87565640802101474.
game for remediation of dyscalculia. Behavioral and Brain Functions, 2, 19. http://
Kucian, K., Grond, U., Rotzer, S., Henzi, B., Schönmann, C., Plangger, F., et al. (2011).
dx.doi.org/10.1186/1744-9081-2-19.
Mental number line training in children with developmental dyscalculia. Neu-
Wynn, K. (1992). Addition and subtraction by human infants. Nature, 358, 749e750.
roImage, 57(3), 782e795. http://dx.doi.org/10.1016/j.neuroimage.2011.01.070.
http://dx.doi.org/10.1038/358749a0.
Kucian, K., Loenneker, T., Dietrich, T., Dosch, M., Martin, E., & von Aster, M. (2006).
Xu, F. (2003). Numerosity discrimination in infants: evidence for two systems of
Impaired neural networks for approximate calculation in dyscalculic children:
representations. Cognition, 89(1), B15eB25. http://dx.doi.org/10.1016/S0010-
a functional MRI study. Behavioral and Brain Functions, 2, 1e17. http://dx.doi.org/
0277(03)00050-7.
10.1186/1744-9081-2-31.