Professional Documents
Culture Documents
The Earlv Development of Numerical Reasonin :: Prentice Starkey University of Cu/ijornia
The Earlv Development of Numerical Reasonin :: Prentice Starkey University of Cu/ijornia
Recewed May 30, lY91; final revision accepted October 17, 1991
Abstract
Starkey, P.. 1992. The early development of numerical reasoning. Cognition. 43: 93-126
Children of age l-4 years were found capable of engaging in numerical reasoning.
Children were presented with a task in which they placed a set of objects one by one
into an opaque container. An experimenter then visibly performed either an
addition, a subtraction, or no transformation on the screened set. Children were
then instructed to remove all objects from the container. Across two experiments,
children searched for and removed the correct number of objects when set
numerosity was small. Knowledge of numerical identity and knowledge of the
effects of addition and subtraction transformations on numerosity were present even
in children who had not yet begun to count verbally. These findings provide
evidence that the emergence of numerical reasoning does not depend upon the prior
development of a verbal counting ability or upon cultural experience with numbers.
Introduction
Over the past two decades, investigators have attempted to discover and catalog
the cognitive capabilities of young children. Considerable progress has been made
in the domain of numerical cognition (e.g., Carpenter, Moser, & Romberg, 1982;
Gelman & Gallistel, 1978; Ginsburg, 1983; Piaget, Grize, Szeminska, & Vinh
Bang, 1977; Saxe & Gearhart, 1988; Siegler & Robinson, 1982). It has been
found that preschool children possess abilities to enumerate sets of concrete
objects (Gelman & Gallistel, 1978; Klahr & Wallace, 1976) and that even infants
can enumerate sets that are small in numerosity (Starkey & Cooper, 1980;
Correspondence fo: Prentice Starkey, Institute of Human Development and School of Education.
Tolman Hall, University of California, Berkeley, CA 94720, USA.
Starkey, Spelke, & Gelman, 1983,199O; Strauss & Curtis, 1981; van Loosbroek &
Smitsman, 1990). It has also been found that preschool children possess abilities
to reason about numerosities early in life, well before the point at which they will
receive explicit tutoring in mathematics in school (Cooper, 1984; Ginsburg, 1983;
Siegler & Robinson, 1982; Starkey & Gelman, 1982). Furthermore, several of
these early abilities have been found to be universally present across a great
variety of sociocultural contexts (e.g., Ginsburg & Russell, 1981; Klein &
Starkey, 1988; Saxe & Posner, 1983).
The existence of these early abilities raises questions about their origins and
developmental precursors. Our focus in this paper is on the origins and precursors
of numerical reasoning. We use the term numerical reasoning in much the same
way as Gelman and Gallistel (1978), who drew a distinction between young
children’s numerical abstraction capabilities and their numerical reasoning
capabilities. Numerical abstraction (or enumeration) comprises a set of abilities
that are used to form representations of numerosities of sets. An example is
verbal counting. Numerical reasoning comprises a set of abilities that are used to
operate upon or mentally manipulate representations of numerosity.
Gelman and Gallistel’s (1978) model of the development of numerical abstrac-
tion - in particular the development of counting - has received much attention
from investigators (e.g., Briars & Siegler, 1984; Gelman & Meek, 1983; Sophian,
1988). In contrast,. there have been fewer empirical tests of Gelman and Gallis-
tel’s (1978) model of the development of numerical reasoning, for example, tests
for the presence of knowledge of numerical equivalence (Michie, 1984; Sophian,
1988) and knowledge of addition and subtraction operators (Starkey and Gelman,
1982). Thus some basic assumptions of this model remain untested. For example,
the model proposes that one numerical abstraction ability, verbal counting, is the
earliest and in preschool children a prerequisite ability for children to engage in
numerical reasoning. It is assumed that in this age group the computational
routines that carry out numerical reasoning function only in situations in which
verbal counting has taken place. This is assumed, because only counting is
thought to provide children with the type of representation that accesses these
routines early in life, a representation of a specific numerosity (e.g., “set a
contains 2 objects”). These routines are not accessed by representations of
relative, unspecified numerosities (e.g., “set a is more numerous than set b”, with
no absolute values represented). Gelman and Gallistel (1978) describe another
numerical abstraction ability, subitizing, as a late (relative to the onset of verbal
counting) developing form of covert counting. Since subitizing is modeled as a
late development, it is assumed that children first engage in numerical reasoning
when they become able to form specific numerical representations of sets by
verbally counting them.
It has recently been discovered that verbal counting is not the earliest
numerical abstraction ability to appear in the child’s repertoire. At least one
Numerical reasoning 9s
‘One-to-one correspondence could be used both to sort sets and to order them by numerosity. For
example, if two sets, a and b, are compared, and it is found that for each element in a there is a
corresponding element in b and for each element in b there is a corresponding element in a. then the
two sets are equivalent in numerosity. If two sets, a and b, are compared, and it is found that for each
element in a there is a corresponding element in b but for some element in h there is no corresponding
element in a, then set b is larger in numerosity than set a.
96 P. Starkey
‘At the time we were developing this task (1981-1982), Hughes (1986) was also developmg a
similar task called the box task. Hughes’ task could be adapted for very young children by having them
search for objects without using number names; however, a methodological problem would arise,
because tactual cues indicating whether the box is empty could then be used by children.
98 P. Starkey
Experiment 1
Method
Subjects
Forty children from middle-class families participated as subjects. Four age
groups - nine 24-month-olds (median age, 24 months; range, 23.5-25 months),
sixteen 30-month-olds (median age, 30 months; range, 29.5-30.5 months), eight
36-month-olds (median age, 36 months; range, 35-38 months), and seven 42-
month-olds (median age, 42 months; range, 41-43 months) - were included.
Approximately equal numbers of boys and girls were included in each age group.
One of the 2Cmonth-olds was subsequently excluded for failure to differentiate
among problems. (Preliminary research had determined that a few children
apparently fail to differentiate among problems. These children search the same
number of times, usually one time, on every problem. The following criterion was
developed to exclude such children from subsequent analyses: children who
searched a fixed number of times on at least 80% of the problems were to be
excluded from data analyses.) One additional child, a 42-month-old, was excluded
for noncompliance with the experimenter’s instructions.
Stimuli
The stimuli were 28 numerical identity problems in which no numerical
transformations (e.g., addition) were performed. In each problem, a set of l-5
objects (table-tennis balls) was displayed. The composition of the set was either
Numerical reasoning 99
homogeneous (l-5 objects that were similar in form and color, for example a set
comprised of three red balls) or heterogeneous (3-4 objects that were similar in
form but dissimilar in color, for example a set comprised of one orange, one
green, and one blue ball). Balls were painted in one of five primary colors. Two
stimulus display conditions were included: a simultaneous display condition in
which all of the objects comprising a set were displayed simultaneously, and a
successive display condition in which all of the objects comprising a set were
displayed over time but only one object was visible at any given time.
The 28 problems were presented across two experimental sessions as follows.
The simultaneous display condition, which was made up of two identical blocks of
seven problems, was presented in one session, and the successive display condi-
tion, which also was made up of two identical blocks of seven problems, was
presented in the other session. Problem order within a block was random.
Procedure
The children were tested individually in a private laboratory room with a
parent present. All children participated in two experimental sessions (median
intersession interval, 11 days). Children were escorted directly into the laboratory
room upon arrival. They initially spent approximately 5-10 minutes playing with
toys (Playmobile miniature people and playground equipment). The experimen-
ter initially watched the child’s play and then began to interact with the child by
joining in the play. The child’s parent sat approximately 135” from the child’s
midline, initially interacting with the experimenter and, if necessary, facilitating
interaction between the experimenter and the child. After the child was judged to
be reasonably comfortable in the room and was speaking or responding to verbal
directives of the experimenter, the searchbox task was introduced. This task will
first be described as it was presented in the simultaneous display condition, and it
will then be described as it was presented in the successive display condition.
Simultaneous display condition. The experimenter introduced the searchbox
task to the child as a game. The child or, if necessary, the experimenter then
labeled the main components of the task: a searchbox (“box”), the objects to be
hidden and searched for (“balls”), and receptacles in which the objects were
placed both before transferring them to the searchbox and after removing them
from the searchbox (“eggcups”). The experimenter placed a ball in an eggcup
and proceeded to demonstrate how to put the ball in the searchbox (“Look at
what 1 can do. I can put the ball in the box.“) and how to remove it (“Let’s set if
I can find it. I found it!“) and to place it back in the cup. The child was then
asked to put the ball in the box (“Would you like to put the ball in the box?“)
and then to remove it (“Can you get it? Can you find it?“) and to place the ball
into an eggcup (“You got it! And could you put it in one of the cups? Very
good.“) After the practice problem, the 14 experimental problems included in the
condition were administered.
100 P. Starkey
In each problem, the child was shown a set of l-5 objects. The entire set was
displayed in provoked correspondence with a subset of an array of six eggcups.
The child was instructed to attend to the initial display (“Look at these balls”,
pointing to the display) and was monitored for compliance. Next, the child was
told to transfer the set object-by-object from the eggcups to the searchbox (“Put
all these balls in the box. Put in one at a time.“) As the child placed balls into the
searchbox, a concealed assistant surreptitiously removed them through an open-
ing in the back of the searchbox (see description of apparatus, below). One to two
seconds after the child had placed the last ball into the searchbox, he or she was
told to remove the set from the searchbox (“Now, can you find them? Take out
all the balls.“). This phrasing was used on all problems (including problems in
which only one ball had been placed in the searchbox) in order to avoid verbally
cueing the child as to whether one versus multiple balls were in the box. As balls
were removed, the child was reminded to place them in eggcups (“Put it in a
cup.“), but the child was not required to place balls in the same subset of eggcups
that had held them before they were transferred to the searchbox. The concealed
assistant surreptitiously watched the child on a video monitor and inserted an
object into the searchbox just prior to the child’s search. This procedure was
followed until all objects had been removed by the child or until the child
prematurely ceased searching.
A temporal criterion was used to decide whether the child had completed his
or her search for objects: after the child released a retrieved ball, the experimen-
ter sat quietly and commenced timing mentally the duration of the elapsed
interval from the offset of the previous search (release of the retrieved ball from
the child’s hand) to the present. As soon as the duration was judged to be equal
to or greater than twice the average duration of the child’s preceding intersearch
intervals in the current problem, the experimenter removed the objects from the
eggcups and began the presentation of the next problem. Thus, for each child the
absolute duration required to meet the criterion could vary from problem to
problem. (The videotape record indicated that, in practice, the experimenter
erred on the side of waiting even longer than required by the criterion, approxi-
mately 2.5 rather than 2.0 times longer than the child’s intersearch interval, and
that the experimenter indeed sat quietly during the child’s search and thus did not
provide nonverbal cues to terminate the search.)
The child was allowed to search as few or as many times as he or she wished
but was not allowed to find more than the veridical number of objects. When the
child searched too many times, no ball was found on the final search. For
example, if a set of two objects was put into the searchbox and the child then
searched twice, removing the two objects, and then searched a third time, he or
she would find the searchbox to be empty. The problem would be terminated at
that point. After the child removed the final ball of the set from the searchbox,
the assistant briefly inserted and then removed an extra ball in order to exclude
Numerical reasoning 101
the possibility that the child’s decision to search again might be driven by some
cue that was associated with the assistant inserting the ball into the container.
When the child searched too few times and hence did not remove as many objects
as he or she had placed into the searchbox, the assistant retained the balls the
child erroneously failed to retrieve. This procedure was followed in order to avoid
giving the child visual feedback as to the correctness of his or her search (i.e.. by
seeing whether the experimenter removed any unretrieved balls from the box at
the end of the problem). No verbal feedback was given either.
In order to determine whether feedback would influence the child’s search
behavior, two groups of eight 30-month-olds were included in the experiment.
The no-feedback group was given no feedback when too few objects had been
retrieved from the searchbox. The feedback group saw the experimenter remove
the objects the child had erroneously failed to retrieve (“I think there might be
another one/some other ones. There was another one/some other ones.“). In all
other respects, the procedures followed for the no-feedback and the feedback
groups were identical.
Successive display condition. The searchbox task was introduced in the same
way as in the simultaneous display condition except that the experimenter
removed a ball from an opaque bag, held up the ball for the child to see (instead
of placing the ball in an eggcup), and placed the ball into the searchbox. When
the experimenter subsequently removed the ball from the searchbox, it was again
held up for the child to see and then was returned to the bag. The child was then
handed a ball from the bag and was instructed to place the ball into the
searchbox. After the child had done so, he or she was instructed to remove the
ball from the searchbox and to hand it to the experimenter. After one more
practice problem, again using one ball, the 14 experimental problems included in
the condition were administered.
In each of these problems, the child was handed a set of l-5 objects one by
one from the opaque bag. Immediately after the child placed an object into the
searchbox the next object in the set was removed from the bag and handed to him
or her. One to two seconds after the child had transferred all objects in the set to
the searchbox, he or she was asked to remove the set from the searchbox. The
child then removed a ball from the searchbox and handed it to the experimenter.
who immediately placed it into the bag. This continued until the child removed all
balls from the searchbox or ceased searching. In all other respects, the procedure
followed in this condition was identical to the procedure followed in the simulta-
neous display condition.
A brief counting task modeled after Gelman and Gallistel’s (1978) basic task
was administered in order to provide data on individual children’s ability to
produce verbal counts of sets like those presented in the searchbox task. In order
to avoid the possibility of biasing children toward the use of verbal counting
during administration of the searchbox task, the verbal counting task was
102 P. Starkey
administered at the end of the final session. The experimenter placed five balls in
eggcups that had been used in the simultaneous display condition and then told
children to count the set of balls. When children did not attempt to count, their
parents were instructed to try to elicit counting. Also, parents were asked whether
they had observed counting by their children at home.
Apparatus
The apparatus was a searchbox that was especially designed for the experi-
ment. It consisted of a lidded box with an opening in its top, a false floor, and a
hidden trap-door in its back. This opening in the top of the box was covered by
pieces of elastic fabric such that a person’s hand could be inserted through pieces
of the fabric and into the chamber of the searchbox without visually revealing the
chamber’s contents. The searchbox measured 20.3 x 25.4 x 30.5 cm. It housed a
roughly cubical (approximately 20 X 25 X 23 cm) chamber in which children
searched for objects in the searchbox. The searchbox was located on the floor
with its back surface attached to a 54-inch (1.37-m) high partition. One edge of
the partition was attached at a 90” angle to a second, similar partition. One
partition was adjacent to one wall of the laboratory room, and the other partition
was adjacent to another wall, thus forming an enclosure behind the searchbox.
An assistant sat concealed inside this enclosure. The back of the searchbox was
hinged, allowing the assistant access to the inside of the searchbox. A hinged false
floor in the searchbox could be lowered by the assistant so that balls could easily
be surreptitiously removed from or placed into the searchbox by the assistant.
The back and false floor of the searchbox made no audible sound when moved.
A color video camera and video cassette recorder (VCR) were used to obtain a
videotape record of the experimental sessions. A panel which was solid except for
a circular camera port concealed the camera and VCR from the child’s view. A
video monitor located behind the assistant’s enclosure enabled the assistant to see
the child’s actions. A microphone to the VCR was attached to the side of the
searchbox. The camera placement provided a clear view of the searchbox, a side
view of the experimenter’s face, and a frontal view of the child’s face and hands.
Design
There were three between-subjects factors in the design: 4 age levels (24, 30,
36, and 42 months), 2 sexes, and 2 orders of display conditions (problems
containing simultaneously displayed sets were solved first vs. problems containing
successively displayed sets were solved first). There were four within-subjects
factors: 2 display conditions (problems contained simultaneously displayed sets
vs. problems contained successively displayed sets), 5 set numerosities (l-5), 2 set
compositions (homogeneous vs. heterogeneous), and 2 problem blocks (block 1
and block 2, nested within display condition). For the 30-month-olds, there was
Numerical reasoning 103
Results
A JVC video editing control unit and color monitor were used in coding the data
from videotape in slow motion and in some instances frame by frame. Data were
reduced to (a) coded records of the number of searches children made for balls in
order to reconstruct the original set (inter-rater agreement on number of searches
2.95), (b) coded records of the durations of children’s intersearch intervals while
reconstructing the original set (inter-rater agreement on duration of search 2.95),
(c) transcriptions of children’s verbalizations and verbal counting behaviors.
In this section, we will first examine the data for general effects of age, sex,
and condition order and then will report analyses to determine whether children’s
levels of performance in the searchbox task exceeded the levels expected by
chance alone. We will then proceed to report data that bear upon the issue of
which numerical abstraction ability children used in solving the searchbox prob-
lems. This will include a tabulation of the rate of occurrence of verbal counting, a
comparison of performance in the two display conditions, and an analysis of set
numerosity effects. We then attempt to capture regularity in children’s errors
through an error analysis. Finally, set composition effects and findings from the
counting task will be examined.
Children’s solutions to the problems were scored as correct if the number of
times children searched for balls while reconstructing the original set corresponded
exactly to the original set’s numerosity. Solutions were scored as incorrect if
children searched too few or too many times for balls. We found that the solutions
of even the youngest children were often correct. On average, children correctly
solved about half (53%) of the set of 28 problems presented to them. A 4
(age) x 2 (sex) x 2 (or d er of display condition) analysis of variance on children’s
mean proportion of correct solutions across all problems revealed a significant
main effect of age, F(3/15) = 6.65, p < .005. Children’s proportions of correct
solutions across all problems at 24, 30, 36, and 42 months of age were 0.37, 0.52,
0.59, and 0.64, respectively. No other main effects or interactions were sig-
nificant.
One possible way to solve a searchbox problem (albeit with very limited
success) would be to attempt to guess the numerosity of the screened set. An
analysis was conducted to determine whether children relied on a simple guessing
strategy when attempting to reconstruct the set while it was screened from view.
In computing the proportion of correct solutions children can be expected to
produce by use of this strategy, it is important to note that the number of times a
child can possibly search during a searchbox problem changes with set numerosi-
104 P. Starkey
ty. For example, a child can search either once or twice when a set has a
numerosity of 1 (assuming that the child will attempt to search). To search once
and then to stop searching (i.e., to search the correct number of times) comprises
one possible type of response. To search twice and then to stop searching,
because nothing was found during the second search, comprises the other possible
type of response. For the purpose of our analysis, therefore, chance level on sets
of 1 object was considered to be l/2. Accordingly, chance levels on sets of 1,2, 3,
4, or 5 objects was considered to be 0.50, 0.33, 0.25, 0.20, and 0.17, respectively.
Thus, as numerosity increases, chance level decreases, and the proportion of
correct solutions needed in order to exceed chance decreases. These estimates of
chance-level performance are conservative in that they assume that children
would not guess that the numerosity of a screened set contains more than n + 1
items (e.g., that the numerosity of a screened l-item set is more than 2). If
children were to guess that numerosity is more than n + 1, then chance level
should be smaller than our estimate (e.g., smaller than l/2 for a l-item set).
Children’s proportions of correct solutions averaged across both display condi-
tions at each level of numerosity were calculated and were compared with the
above estimates of chance-level performance. As indicated in Table 1, the
observed level of performance exceeded the level expected by chance when set
numerosity was small (1, 2, or 3) but not when it was large (4 or 5).
A further question concerns which numerical abstraction ability children used
to solve searchbox problems. Children did not form numerical representations of
sets of objects by verbally counting them. They almost always solved the
searchbox problems in silence and without the motor behaviors that typically
accompany verbal counting by preschool children (e.g., sequential touching or
pointing to objects; see Gelman and Gallistel, 1978, for a description of speech
and motor behaviors preschool children produce while counting verbally). Spon-
taneous use of numerical (or other) terms to describe sets was rare. Only five
children (a 30-month-old, two 36-month-olds, and two 42-month-olds) verbally
counted a set or reported the numerosity of a set. Three of these five children did
so on just one problem, and two did so on five problems. In only one of these
cases did a child verbally count a set before retrieving it from the searchbox. In
the remaining cases, children verbally counted a set or simply reported its
numerosity after retrieving it from the searchbox.
We found that children correctly solved small-set problems in both display
conditions (Table 1). All age groups enumerated sets of l-3 items in both display
conditions, except for the 24-month-olds, who enumerated sets of l-3 items in the
simultaneous display condition but enumerated only sets of l-2 items in the
successive display condition. Relative difficulty of problem solving in the two
display conditions was examined in a planned comparison. Children’s proportions
of correct solutions to the small-set problems (specifically, problems containing
either homogeneous sets of l-3 items or heterogeneous sets of 3) in the
.Vumerical remming IO5
Numerosity of
Numerosity of heterogeneous
homogeneous sets sets
Age Display
condition 1 2 3 4 5 3 3
Note. Ages are given in months. nf and f refer to no-feedback and feedback groups. respectively.
“Means of the 30f group are not included in the overall means.
* p < .lO, two-tailed: **p < .05, two tailed.
efforts to reconstruct large sets usually failed. The typical data pattern of children
included relatively few errors on sets of l-3 objects but many errors on sets of 4
and 5. Children’s performance when homogeneous sets of 4 items were presented
was significantly lower than when sets of 3 items were presented, t(30) = 6.56,
p < .Ol. Furthermore, performance levels on sets of 4 and 5 were similar to one
another, t(30) = 1.54, p > .lO. Since children were not at ceiling on sets of
numerosity 1, 2, and 3, comparisons were also possible at these levels of
numerosity. Performance levels were lower on sets of 2 items than on sets of 1,
t(30) = 2.11, p < .05, and were lower on sets of 3 items than on sets of 2,
t(30) = 3.77, p < .Ol.
We next conducted an error analysis to establish whether children’s errors were
systematic and, if so, what their specific characteristics were. Since most children’s
errors occurred on large set problems, any patterns revealed by this analysis for
the most part reflect performance on the large set problems. One criterion that
was used in categorizing children’s error patterns required the child to have made
at least three errors, the majority of which were of one type. Two predominant
types of error pattern were identified: (a) a fixed numerosity search pattern that
resulted from children’s attempts to solve problems by searching a fixed number
of times (typically three times) regardless of the number of objects they had
placed in the searchbox; and (b) an exhaustive search pattern that resulted from
children’s attempts to solve problems by searching until they found no more
objects in the searchbox. The predominant error patterns of 75% of children’s
sessions fell into one of these two categories (Table 2). In the remaining sessions,
children’s errors are designated as other or U-2 errors (Table 2). Patterns were
scored as other if either a rare error pattern or no identifiable error pattern was
predominant in a session (e.g., some children produced a mixture of fixed
numerosity search errors and exhaustive search errors with neither type compris-
ing a majority across all error trials). If children made too few (O-2) errors for an
error analysis to be conducted with any degree of confidence, their session was
scored as a 0-2 error session. When children used a fixed numerosity strategy,
their typical error was to search three times (Table 3, top portion).
One child (who was subsequently excluded) searched only once on most
problems, apparently failing to differentiate among problems. One possible
explanation for this search pattern is that this child located the first ball in the
search chamber, but then searched for other balls before removing the first ball
from the search chamber. No other balls were found (because the other balls
were concealed in another chamber), so the child thought the searchbox was
empty and stopped searching. It is possible that other children upon discovering
that the search container was mysteriously empty came to view the task as
artificial and began to use ad hoc strategies, such as searching three times
regardless of the number of objects they had placed in the searchbox. These
explanations, however, must remain speculative, because the design of the
Numerical reasoning 107
Display condition
Fixed numerosity
Experiment Condition 1 2 3 4 5
1 Simultaneous 0 1 10 3 0
Successive 0 2 6 2 0
2 Addition 3 1 17 0 0
Subtraction 12 11 1 0 0
108 P. Starkey
searchbox did not allow us to observe children’s manual search behaviors inside
the search chamber.
Even though children’s search behaviors inside the search chamber were not
visible, it was evident from children’s occasional failures to find a ball that was in
the search chamber that at least some searches were not thorough. (When a ball
was searched for but was erroneously not found, the veridicality of the searchbox
was maintained by telling children to “feel all around in the box” and “in the
corners” until the ball was found.) These failures to find objects in the searchbox
led us to conduct a small pilot study in which a concealed assistant placed a ball in
a corner of the original cubical search chamber or in a cylindrical chamber. It was
found that children more frequently failed to find the ball when it was in a corner.
One possible explanation is that children did not sufficiently represent the internal
shape of the cubical chamber.
Our findings on children’s verbal counting ability, which were obtained at the
end of children’s final experimental session, indicated that the 24- and 30-month-
olds had not yet begun to count verbally. Our evidence for this was twofold. First,
the younger children in our sample did not count verbally when instructed to do
so by the experimenter. They exhibited no behaviors that would be scored as
counting in Gelman and Gallistel’s scoring system. Second, parental reports
confirmed that these children were not yet counting in any setting to the parents’
best knowledge. The older age groups were beginning to count; nevertheless
when presented with searchbox problems, they did not spontaneously do so.
Thus, the evidence clearly indicated that the 2-year-olds in our sample were not
yet counting verbally.
We now turn to children’s performance on the large-numerosity problems. The
most likely reason children used numerically undifferentiated strategies (the
exhaustive search and fixed numerosity strategies) on these problems was because
they were unable to represent the specific numerosities of the large sets. A further
question, however, is what led children to select the particular strategies they
used. The findings from one of our experimental conditions provide some relevant
information on this issue. In this condition, feedback was provided to children
when they searched too few times and thus failed to remove some of the objects
from the searchbox. This was done by having the experimenter visibly remove the
balls that children had failed to remove. The procedure was exploratory in nature,
so it was administered at only one intermediate.age level: 30 months. One group
of 30-month-olds was given feedback, and another group was not. Analyses of
variance of children’s mean proportions of correct solutions across all problems
revealed no significant effect of feedback. However, an examination of children’s
predominant error patterns revealed that the 30-month-old no-feedback children
searched exhaustively in only 19% of the experimental sessions and searched a
fixed number of times in 56%, whereas the feedback children searched exhaus-
tively in 56% of the sessions and searched a fixed number of times in 12%. Thus,
Numericul reasonin,q IOY
although the presence of feedback did not lead children to enumerate large sets
accurately, it affected children’s strategy selection on large-numerosity problems.
(On small numerosity problems children continued to enumerate accurately.) It
can not be concluded that the feedback children simply learned the exhaustive
search strategy from the experimenter, because the experimenter did not model
this strategy. The experimenter stopped searching when the final ball was
removed. To summarize, the findings were that the strategy used on large-
numerosity problems (but not the strategy used on small-numerosity problems)
differed when feedback was provided as opposed to when it was not.
One objective of Experiment 1 was to determine whether the ability of
children to solve searchbox problems is robust or whether it is affected by a set’s
numerosity and composition and by how the set was displayed. A significant effect
of set numerosity was evident at all age levels, but an effect of display condition
(simultaneous vs. successive) was not. Another set property that was varied in the
experiment was set composition (homogeneous vs. heterogeneous composition).
Children were presented with problems containing either homogeneous sets of
l-5 objects or heterogeneous sets of 3-4. Children’s mean proportions of correct
solutions to the heterogeneous-set problems, averaged over display condition, are
given in Table 1. We compared children’s proportions of correct solutions on
heterogeneous sets with their performance on homogeneous sets of the same
numerosity. Planned comparisons revealed that children solved the homogeneous-
set problems more often than the heterogeneous-set problems when set numerosi-
ty was 3, t(30) = 2.20, p < .05, but not when it was 4. t(30) = 1.28. p > 10.
Discussion
Children did not simply attempt to guess the numerosity of a small set while it was
screened from view inside the searchbox. Instead, they formed a representation of
the set’s numerosity, retained the representation for a period of seconds (at least),
and utilized the representation in reconstructing the screened set. Children
formed and used numerical representations spontaneously, that is, without any
direct instruction to enumerate the original set or to use the product of their
numerical abstraction activity to guide their subsequent search for the set of
objects. A finding from one of our earlier searchbox experiments (Starkey, 1983)
provides additional evidence that children expected numerosity to be unchanged
by spatial displacement. In this experiment, 2-year-olds were presented with a
numerical identity problem in which they transferred a 2-object set to the
searchbox and then searched for it. We surreptitiously removed one object,
however, so after finding one object children found nothing when they reached
into the searchbox a second time. When children failed to find a second object
inside, they engaged in social referencing, as if they were seeking an explanation
110 P. Starkey
Experiment 2
Method
Subjects
Fifty-six children from middle-class families participated as subjects. Six age
groups, sixteen l&month-olds (median age, 18 months; range, 17.5-18.5
months) - eight 24-month-olds (median age, 24 months; range, 23-24.5 months),
eight 30-month-olds (median age, 30 months; range, 29.5-30.5 months), eight
36-month-olds (median age, 36 months; range, 35-36.5 months). eight 42-month-
‘An effect of heterogeneity does not imply a total inability to enumerate heterogeneous sets.
Heterogeneity may only make accurate enumeration more difficult and thus depress performance.
Even infants can enumerate small heterogeneous sets (e.g., Starkey et al., 1990; Strauss & Curtis.
1981; van Loosbroek & Smitsman, 1990), but the relative difficulty of enumerating heterogeneous
versus homogeneous sets has yet to be investigated systematically in infants.
112 P. Starkey
olds (median age, 42 months; range, 41.5-43.5 months), and eight 48-month-olds
(median age, 48 months; range, 47-49 months) - were included. Approximately
equal numbers of boys and girls were included in each age group. Of these, four
children (one from each of the four younger age groups) were subsequently
excluded for failure to differentiate among problems. All four searched a fixed
number of times, once, on at least 80% of the problems. Seven additional
children, all l&month-olds, were excluded for noncompliance and were replaced
by other subjects.
Stimuli
The stimuli were 30 addition and subtraction problems, in which l-3 objects
were either added to or subtracted from a set, and 4 numerical identity problems
in which no numerical transformations were performed (Table 4). In each
problem, a set of l-5 objects (table-tennis balls) was displayed. The composition
of the set was homogeneous (objects were similar in form and color, for example
a set comprised of two red balls). Balls were painted in a primary color. Only the
simultaneous display condition was included in the experiment.
The 34 problems were presented across two experimental sessions as follows.
An addition condition, which was made up of two identical blocks of eight
Problem type”
Augend or Type of Addend or
Transformation minuend numerical subtrahend Proportion
condition numerositv transformation numerositv correctb
Addition 1 None (identity) .91* .12
0 + .86” .25
1 + .I7 4.21
1 + .52? .40
1 + .13 + .31
2 + .55 2 .44
2 + .18 ? .32
2 + .08 ” .21
3 + .15 + .26
addition problems and one numerical identity problem, was presented in one
session, and a subtraction condition, which was made up of two identical blocks of
nine subtraction problems and one numerical identity problem, was presented in
the other session. Order of addition and subtraction problems within a block was
random.
Procedure
The experimental procedure was identical to the procedure followed in the
simultaneous display condition of Experiment 1 except for the following changes.
In all of the problems, the child initially placed objects into the searchbox one by
one as in the numerical identity problems. Then, in the addition problems, the
experimenter removed a ball from an opaque bag and placed it into the searchbox
(“Now watch what I’m doing. I’m putting another ball in.“). If more than one
object was to be added, this procedure was repeated the requisite number of
times to achieve a sum of the prescribed magnitude. In the subtraction problems,
the experimenter removed a ball from the searchbox and placed it into the
opaque bag (“Now watch what I’m doing. I’m taking a ball out.“). If more than
one object was to be subtracted, this procedure was repeated as in the addition
problems. After the addition or subtraction transformation had been performed,
the child was instructed to remove all of the balls from the searchbox. As in
Experiment 1, only one ball was in the chamber of the searchbox while the child
was executing a search. Feedback was not given when children failed to remove
the entire set from the searchbox. Children participated in two experimental
sessions (median intersession interval, 7 days).
Apparatus
After a considerable amount of pilot testing, a new searchbox was designed in
order to eliminate the need for a concealed assistant to remove objects from or to
insert objects into the searchbox surreptitiously. The final product was a mechani-
cal box that was operated by a remote control unit placed out of the child’s view,
next to the experimenter (Figure 1). One button on the control unit controlled
the positioning of a 27-cm diameter revolving panel that was located inside near
the top of the searchbox. A 12-cm diameter hole in one section of this panel
provided manual access to a chamber in the searchbox when the panel was
rotated into a particular position. This hole was covered by pieces of elastic fabric
such that a hand could be inserted through pieces of the fabric and into the
chamber without visually revealing the chamber’s contents.
The search chamber was cylindrical in shape (diameter, 10.2 cm; depth, 19 cm)
and comprised the shape within which children searched for objects in the
searchbox. This shape replaced the earlier cubical one which had occasionally
resulted in children’s failure to locate an object that had rolled into a corner. The
chamber’s small diameter and shallowness ensured that children would easily
114 P. Starkey
(b) (4
Figure 1.
locate an object when they searched. A mechanical trap-door in the floor of the
search chamber was controlled from the remote control unit. This made it
possible to remove objects from the search chamber surreptitiously and to place
them in a lower concealed chamber. The searchbox also housed concealed feeder
chutes, operated by remote control, which held a concealed bank of objects for
use in the experiment. This made it possible to place objects into the search
chamber surreptitiously. A combination of sound-treatment measures (interior
padding and background noise from a fan) rendered inaudible the movement of
concealed objects inside the searchbox. It was necessary to reload the searchbox’s
Numericul reasoning 11s
concealed bank of objects after the child had completed the first block of
problems in a session. This was accomplished by having the child take a brief
(2-5-minute) break outside the room with a parent while the experimenter
prepared the searchbox for the next block of problems.
Sessions were videotaped. The camera placement provided a clear view of the
searchbox, a profile of the experimenter’s face and hands, and a frontal view of
the child’s face and hands.
Design
There were three between-subjects factors in the design: 6 age levels (18. 24,
30, 36, 42, and 48 months), 2 sexes, and 2 orders of transformation conditions
(problems containing addition transformations were solved first vs. problems
containing subtraction transformations were solved first). There were three
within-subjects factors: 2 transformation conditions (problems containing addition
transformations vs. problems containing subtraction transformations), 6 augend
or minuend numerosities (O-5), 3 addend or subtrahend numerosities (l-3), and
2 trial blocks (nested within transformation condition). Transformation condition
was a between-subjects factor at the l&month age level.
Results
We will first report our findings on whether young children knew that addition
and subtraction transformations change the numerosities of the screened sets and,
if so, whether they knew the ordinal effects of the transformations. We will also
describe our findings on young children’s ability to calculate the specific effects of
transformations (i.e., precise sums or remainders). Subsequent analyses will
examine the relative difficulty of addition and subtraction problems, and young
children’s typical error patterns will be described.
The first set of analyses was conducted to determine whether children knew
that addition and subtraction transformations changed the numerosity of the
original (i.e., augend or minuend) sets, with addition increasing the numerosity
and subtraction decreasing it. On the vast majority of the problems (84% of the
addition and 91% of the subtraction problems; 88% overall), the children
searched some number of times that differed from the number of objects in the
original set. On the remaining 12% of the problems, children erroneously
searched some number of times that corresponded to the numerosity of the
original set. (These errors may reflect lapses in children’s attention that occurred
while transformations were being performed by the experimenter rather than lack
of knowledge that addition and subtraction alter the numerosity of a set.) We next
calculated the percentage of solutions to addition problems in which children’s
sums (i.e., the numbers of times they searched in these problems) were less than.
116 P. Starkey
equal to, or greater than the numerosity of the augend set. Likewise, we
calculated the percentage of solutions to subtraction problems in which children’s
remainders were less than, equal to, or greater than the numerosity of the
minuend set. If children have some understanding of the ordinal effects of
addition and subtraction, we can expect to find that children’s sums (regardless of
whether they were correct) will be greater than the augend sets and that their
remainder sets will be less than the minuend sets. Fully 81% of children’s sums
and remainders were ordinally correct (i.e., sums were greater than augends and
remainders were smaller than minuends), 13% of children’s sums were equal to
the augend or minuend numerosity, and 6% were ordinally incorrect.
The percentage of ordinally correct solutions, however, may be inflated due to
a feature of the searchbox task, namely that on subtraction problems children
could not search for a number of times that exceeded the minuend numerosity.
The searchbox always contained the arithmetically correct number of objects, so
on a 3 - 1 problem, for example, children could search no more than three times,
the minuend numerosity, removing 2 objects and then searching a third time in
which no object is found. Furthermore, on addition problems in which the augend
numerosity was 1, children would have to search zero times in order to search for
a number of times that is less than the augend numerosity. Thus, the addition
problems whose augend numerosity is greater than 1 (i.e., 2 + 1, 2 + 2, 2 + 3, and
3 + 1) constitute the most conservative test of whether children’s searches were
ordinally correct. It was estimated that by chance alone 68% of children’s
searches on these four problems would exceed the augend numerosity (2 + 1:
possible answers are 1, 3, 4, and chance is 2/3 or 0.67; 2 + 2: possible answers are
1, 3, 4, 5, and chance is 314 or 0.75; 2 + 3: possible answers are 1, 3, 4, 5, 6, and
chance is 4/5 or 0.80; 3 + 1: possible answers are 1, 2, 4, 5, and chance is 214 or
0.50; overall, chance is 0.68). Children’s proportions of ordinally correct solutions
were compared with the proportions expected by chance. It was found that 91%
of children’s solutions on these problems were ordinally correct. This percentage
was significantly higher than expected by chance, t(44) = 7.04, p < .Ol. Percen-
tages for the individual age groups were as follows: 64% (18 months), 100% (24
and 30 months), 98% (36 months), 93% (42 months), and 98% (48 months).
These percentages significantly exceeded chance at all ages except 18 months. The
18-month-olds’ data formed a bimodal distribution: the percentages of two
children were O%, and the remainder ranged from 60% to 100% (mean, 80%).
Thus, even in this conservative test, the data indicated that children, including
some of the youngest in our sample, knew the ordinal effect of the trans-
formation.
We next turned to the question of whether children were able to arrive at the
precise sum or remainder when presented with an addition or subtraction
problem. As in Experiment 1, we expected set size effects due to the numerical
abstraction ability used by young children to form numerical representations.
Numerical reasoning 117
‘Values expected by chance on the subtraction problems were calculated as follows: problems with
a minuend of 2: 0.50 (2 - 1 problem)/1 = 0.50 (chance value); minuend of 3: 0.33 (3 ~ 1 problem) +
0.50 (3 - 2 problem)/2 = 0.42; minuend of 4: 0.25 (4 - 1 problem) + 0.33 (4 - 2 problem) + 0.50
(4 - 3 problem)/3 = 0.36; minuend of 5: 0.20 (5 - 1 problem)/1 = 0.20.
118 P.Starkey
tion problems (mean proportion correct: 0.53 2 0.13). Subtraction problems were
solved correctly significantly more often than addition problems, t(36) = 3.83.
p < .Ol.
As in Experiment 1, an error analysis was conducted to determine whether
errors were systematic. The analysis revealed that the predominant types of error
pattern were a fixed numerosity search pattern and an exhaustive search pattern
(Table 6). When children searched a fixed number of times, they typically
searched 3 times when presented with addition problems but searched 1 or 2 times
when presented with subtraction problems (Table 3). The l&month-olds pro-
Problem type
Age Error pattern Addition Subtraction Mean
18 Exhaustive search .oo .38 .lY
Fixed numerosity .40 .38 .3Y
Other .60 .25 .42
O-2 errors .oo 00 .oo
duced one type of error that older children rarely produced. Several such errors
led to the exclusion of the child for noncompliance. The error seemed to stem
from children’s confusion as to whether balls should be removed from or placed
into the searchbox. Specifically, the younger children prematurely stopped putting
balls into the searchbox and then began removing balls, or they prematurely
stopped removing balls from the searchbox and then began reinserting them.
Some of these children also put a single ball from a larger set into and out of the
searchbox repeatedly. The remaining 18-month-olds typically made the same
types of errors as the older children (Table 6).
Discussion
In the addition and subtraction problems of this experiment, sets were displaced
from the array of eggcups to the searchbox. Sets, while screened from view inside
the searchbox, underwent an addition or subtraction transformation. On fully
88% of the addition and subtraction problems, children searched a number of
times that was different from the numerosity of the set before it was placed in the
searchbox and transformed. numerically. In contrast, on 53% of the numerical
identity problems of Experiment 1 (on 75% of the problems with sets of 1-3
items), children correctly searched a number of times that was identical to the
numerosity of the set before it was placed in the searchbox. These findings clearly
show that children treated the addition and subtraction problems differently from
the numerical identity problems. Even the very young children, the 18- and
24-month-olds who were at the transition between infancy and early childhood,
knew that the addition and subtraction transformations changed the numerosity of
the sets, and they knew this despite their inability to see the sets during the act of
transforming them. Furthermore, the children were often able to arrive at the
precise numerosity of the sums or remainders that were inside the searchbox.
When set numerosity exceeded children’s ability to determine the precise
numerosity of the sum or remainder, their erroneous sums were nonetheless
greater in numerosity than the actual augends and their erroneous remainders
were less in numerosity than the actual minuends. A common type of error
pattern was one in which children searched a fixed number of times during the
addition session or the subtraction session. This pattern occurred more often in
Experiment 2 (in 59% of the 24- to 42-month-olds’ sessions) than in Experiment 1
(in 37% of the same age groups’ sessions). The fixed number of times children
searched differed across experiments and conditions: on the numerical identity
problems of Experiment 1 and on the addition problems of Experiment 2,
children usually searched 3 times, but on the subtraction problems they searched
only 1 or 2 times (Table 3). Given that only one addition problem (3 + 1)
contained an augend whose numerosity was 3 or greater, and only one subtraction
Numerical reasoning 12I
General discussion
subtraction was performed, they marked a number of items in the image that
corresponded to the numerosity of the subtrahend; when an addition was
performed, they formed a second image that corresponded to the numerosity of
the addend set. In both cases, as children removed objects from the searchbox,
they marked items in the image(s) until no unmarked items remained. Scanning
the image and marking objects are also kinetic in character, but different from the
subprocesses in the first kinetic imagery explanation. An advantage of the second
kinetic imagery explanation is that it readily accounts for why addition, which
involved the maintenance of two images, was more difficult than subtraction,
which involved only one. Errors can be accounted for if addition triggers
additional image formation and subtraction triggers the marking of objects.
Some plausibility is lent to the imagery hypothesis by the findings of a series of
experiments by Siegler and his collaborators (e.g., Siegler & Robinson, 1982;
Siegler & Shrager, 1984). To illustrate, in one experiment (Siegler and Robinson,
1982) 3- to 5-year-olds were presented with a task in which single sets of
imaginary objects underwent an addition transformation. The type and number of
objects comprising the augend set and the type and magnitude of the transforma-
tion were provided verbally in a word problem. At no point in the presentation of
the problems were sets physically present. Thus, children did not directly
enumerate a physical set. The children’s task was to state the numerosity of the
sum set. It was found that 3-year-olds usually solved only the small-set problems
correctly, and they typically did so without the use of verbal counting. Overt
strategies, including counting strategies, were evident in the older children.
Specifically, children sometimes held up subsets of their fingers which were
substituted for the imaginary objects and then they either verbally counted these
subsets and stated the answer or they did not verbally count but stated the answer
anyway. Also, children sometimes engaged in counting without using external
substitutes for the imagined objects. Overt strategies, however, were used by the
4- and 5-year-olds on only 38% of the problems, the majority of which were
large-set problems. No visible strategy was used on the remaining 620/o, including
many of the small-set problems.
In order to determine how closely the set numerosity effect in Siegler and
Robinson’s data resembled the one in our searchbox experiment, we requested
Siegler and Robinson’s raw data. The data on 4- and 5-year-olds’ solutions were
graciously provided. (The 3-year-olds’ data were not readily located.) We reanal-
yzed the 4-year-olds’ data, the age group closest to those included in our
experiment. The reanalysis revealed that in general, as sum numerosity increased,
the proportion of correct solutions decreased. As sum numerosity increased from
2 to 3 or from 4 to 5, the proportion of correct solutions decreased by lo%, or
less, but when sum numerosity increased from 3 to 4 the proportion of correct
solutions decreased by more than 20%. Thus, there was a discontinuity in Siegler
and Robinson’s data at the same point as the discontinuity in our data. This
124 P. Starkey
present and used by young children in their numerical reasoning, and that at least
one numerical abstraction ability is used in numerical reasoning before verbal
counting is used for this purpose.
References
Beckmann. H. (1924). Die Entwickhmg der Zahlleistung bei 2-6 juhrigen kindern. Zeitschrif/ fur
angewandte Psychologie, 22, l-72.
Beilin, H. (1968). Cognitive capacities of young children: A replication. Science, 162, 920-921.
Blevins-Knabe, B., Cooper, R.G., Starkey, P., Mace, P.G., & Leitner, E. (1987). Preschoolers
sometimes know less than we think: The use of quantifiers to solve addition and subtraction
tasks. Bulletin of the Psychonomic Society, 25, 31-34.
Briars. D., & Siegler, R.S. (1984). A featural analysis of preschoolers’ counting knowledge.
Development Psychology, 20, 607-618.
Bullock. M., & Gehnan, R. (1977). Numerical reasoning in young children: The ordering principle
Child Development, 48, 427-434.
Carpenter, T.P., Moser, J.M., & Romberg, T.A. (Eds.) (1982). Addition and subtraction: A cogntrive
perspective (pp. 157-192). Hillsdale, NJ: Erlbaum.
Cooper, R.G.. Jr. (1984). Early number development: Discovering number space with addition and
subtraction. In C. Sophian (Ed.), The origins of cognitive skills. Hillsdale. NJ: Erlbaum.
Cornell, E.H.. & Heth, C.D. (1983). Spatial cognition: Gathering strategies used by preschool
children. Journal of Experimental Child Psychology, 35. 93-1 IO.
Fischer, J.-P. (1991). Le subitizing et la discontinuite apres 3. In J. Bideand, C. Meljac, & J.-P.
Fischer (Eds.), Les chemins du nombre (pp. 235-258). Paris: Presses universit aires de Iille.
Fuson, K.C. (1988). Children’s counting and concepts of number. New York: Springer-Verlag.
Gast, H. (1957). Der umgang mit zahlen und zahlgebilden in der fruhen kindheit. Zeitschr$ fur
Psychologie. 161, l-90.
Gallistel, C.R. & Gelman, R. (in press). Subitizing: Rapid preverbal counting. In W. Kesscn. A.
Ortony, & F. Craik, Essays in honor of George Mandler. Hillsdale. NJ: Erlbaum.
Gelman. R., & Gallistel. C.R. (1978). The child’s understanding of number. Cambridge. MA:
Harvard University Press.
Gelman, R., & Meek. E. (1983). Preschoolers counting: Principles before skill. Cognition. 1.i.
343-359.
Gelman. R., Sr Tucker, M.F. (1975). Further investigations of the young child’s conception 01
number. Child Development, 46, 167-175.
Ginsburg, H.P. (Ed.) (1983). The development of mathematicul thinking. New York: Academic Press.
Ginsburg, H.P., & Russell, R.L. (1981). Social class and racial influences on early mathematical
thinking. Monographs of the Society for Research in Child Development, 46 (6, serial no. 193).
Harris P.L. (1985). The origins of search and number skills. In H.M. Wellman (Ed.). Children‘s
searching: The development of search skill and spatial representation (pp. 1055122). Hills-
dale, NJ: Erlbaum.
Hughes. M. (1986). Children and number: Difficulties in learning mathemutics Oxford: Basis
Blackwell.
Klahr. D.. & Wallace, J.G. (1976). Cognitive development: An informution processing view Hillsdale.
NJ: Erlbaum.
Klein. A. (1984). The Early development of arithmetic reasoning: Numerative activities and logical
operations. Dissertation Abstracts International, 4-5, 3?5B-376B.
Klein, A.. 6t Starkey. P. (1988). Universals in the development of early arithmetic cognition. In G.B.
Saxe & M. Gearhart (Eds.), Children’s mathematics. New Directions for Child Developmeni. I1
(pp. S-26). (W. Damon, series editor). San Francisco: Jossey-Bass.
126 P. Starkey
Kosslyn, S.M., Margolis, J.A., Barrett, A.M., Goldknopf, E.J., & Daly, P.F. (1990). Age differences
in imagery abilities. Child Development, 61, 995-1010.
Mehler, J., & Bever, T.G. (1967). Cognitive capacity of very young children. Science, 158, 141-142.
Michie, S. (1984). Why preschoolers are reluctant to count spontaneously. British Journal of
Developmental Psychology, 2, 347-358.
Piaget, J. (1954). The construction of reality in the child. New York: Basic Books.
Piaget, J. (1968). Quantification, conservation and nativism. Science, 162, 976-979.
Piaget, J., Grize, J.B., Szeminska, A., & Bang, V. (1977). Epistemology and psychology offunctions.
New York: Norton.
Piaget, J., & Inhelder, B. (1971). Mental imagery in the child. New York: Basic Books.
Saxe, G.B. (1977). A developmental analysis of notational counting. Child Development, 48,
1512-1520.
Saxe, G.B., & Posner, J. (1983). The development of numerical cognition: Cross cultural perspec-
tives. In H. Ginsburg (Ed.), The development of mathematical thinking (pp. 291-317). New
York: Academic Press.
Saxe, G.B., & Gearhart, M. (Eds.) (1988). Child ren’s mathematics. New Directions for Child
Development, 41 (W. Damon, series editor). San Francisco: Jossey-Bass.
Siegel, L. (1973). The role of spatial arrangement and heterogeneity in the development of numerical
equivalence. Canadian Journal of Psychology, 10, 907-912.
Siegler, R.S., & Robinson, M. (1982). The development of numerical understandings. Advances in
Child Development and Behavior, 16, 242-312.
Siegler, R.S., & Shrager, J. (1984). Strategy choices in addition and subtraction: How do children
know what to do? In C. Sophian (Ed.), Origins of cognitive skills (pp. 229-293). Hillsdale, NJ:
Erlbaum.
Silverman, I.W., & Rose, A.P. (1980). Subitizing and counting skills in 3-year-olds. Developmental
Psychology, 16, 539-540.
Sophian, C. (1988). Early developments in children’s understanding of number: Inferences about
numerosity and one-to-one correspondence. Child Development, 59, 1397-1414.
Sophian, C. (1987). Early developments in children’s use of counting to solve quantitative problems.
Cognition and Instruction, 4, 61-90.
Sophian, C., & Adams, N. (1987). Infants’ understanding of numerical transformations. British
Journal of Developmental Psychology, 5, 257-264.
Starkey, P. (1983). Some prescursors of early arithmetic competencies.Paper presented at the meeting
of the Society for Research in Child Development, Detroit, MI.
Starkey, P., & Cooper, R.G., Jr. (1980). Perception of numbers by human infants. Science, 210,
1033-1035.
Starkey, P., & Cooper, R.G. (199.1). Th e nature and development of subitizing in young children.
Manuscript submitted for publication.
Starkey, P., & Gelman, R. (1982). The development of addition and subtraction abilities prior to
formal schooling in arithmetic. In T.P. Carpenter, J.M. Moser, & T.A. Romberg (Eds.),
Addition and subtraction: A cognitive perspective (pp. 99-115). Hillsdale, NJ: Erlbaum.
Starkey, P., Spelke, E.S., & Gelman, R. (1990). Numerical abstraction by human infants. Cognition,
36, 97-127.
Starkey, P., Spelke, E.S., & Gelman, R. (1991). Toward a comparative psychology of number.
Cognition 39, 171-172.
Starkey, P., Spelke, E.S., & Gelman, R. (1983). Detection of intermodal numerical correspondences
by human infants. Science, 222, 179-181.
Strauss, M.S., & Curtis, L.E. (1981). Infant perception of numerosity. Child Development, 52,
1145-1152.
van Loosbroek, E., & Smitsman, A.W. (1990). Visual perception of numerosity in infancy. De-
velopmental Psychology, 26, 916-922.