0% found this document useful (0 votes)
97 views29 pages

Smith1983 PDF

Uploaded by

Pavel Diana
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
97 views29 pages

Smith1983 PDF

Uploaded by

Pavel Diana
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 29

JOURNAL OF EXPERIMENTAL CHILD PSYCHOL.

OGY 36, 150-178 1183)

Development of Classification: The Use of Similarity and


Dimensional Relations
LINDA B. SMITH

Indiana University

Findings about perceptual development indicate that overall similarity is the


primary perceptual relation by which young children compare complex objects.
Traditional studies of classification, however, did not focus on children’s or-
ganizational use of holistic relations but rather on their ability to classify by
dimensions or criteria1 attributes. The results from such traditional studies suggest
that young children are deficient classifiers. The present research investigated
the possibility, contrary to the traditional view, that 4- to &year-old children are
competent and systematic classifiers at least by overall similarity. In three ex-
periments, preschoolers and kindergarteners classified various sets of multidi-
mensional stimuli that could be organized into categories by overall similarity
or by dimensional attributes. Consistent with the research in perceptual devel-
opment, the children were highly attentive to overall similarity. However, the
preschoolers in particular showed marked difficulty in using this relation to form
categories of more than two objects. The children’s difficulties were highly rem-
iniscent of traditional claims about early classification. Analyses of the classification
strategies used by the children, however, suggest that even the youngest children
understood the purpose of a classification. The developmental changes appear
to be in the ability to execute a classification. Importantly, type of classification
strategy was independent of type of category organization. Individual children
used the same strategies both when classifying by overall similarity and by
dimensional attributes. These results strongly suggest that it is the classification
skills themselves, and not just the ability to classify by particular relations, that
change with age.

Young children’s perceptions have often been characterized as holistic


(e.g.. Gibson, 1%9; Werner, l%l). One specific hypothesis about perceptual
development postulates that the young child’s holistic perceptions are
organized by overall similarity relations (Shepp & Swartz, 1976; L. Smith

This research was supported by the National Science Foundation (BNS 78-13019 and
BNS 81-09888). 1 thank Melanie Lockyear, Mary Martin, Rachel Ann Day, and Carol
McCord for their assistance and the children and staffs of the Heatherwood Day Care
Center, Penny Lane Nursery School, and Stinesville Elementary School for their participation.
The comments of Margaret Intons-Peterson, Rebecca Treiman, and the reviewers on earlier
versions of this manuscript are gratefully appreciated. Requests for reprints should be sent
to Linda B. Smith, Department of Psychology, Indiana University, Bloomington, IN 47405.

150
0022-0965/83 $3.00
Copyright 0 1983 by Academic Press. Inc.
All rights of reproduction in any form reserved.
DEVELOPMENT OF CLASSIFICATION 151

& Kemler, 1977). Evidence for this hypothesis is found in a variety of


tasks (see Kemler, 1982a), including free classification. In one study, L.
Smith & Kemler (1977) asked children to classify sets of four objects
varying in size and brightness. Figure 1A illustrates schematically the
structure of the sets given to the children. One possible classification of
these stimuli (Objects a and b versus Objects c and d) is a Similarity
classification and consists of grouping together the objects that are max-
imally similar on both dimensions and grouping apart the objects that
are maximally dissimilar. Across a variety of tetrads of stimuli, kinder-
garteners constructed Similarity classifications. With the same sets of
stimuli, fifth-grade children constructed Dimensional classifications (a
and c versus b and d)-classifications in which objects are grouped
together by identity on one dimension regardless of the degree of difference
on the other dimension. This finding of a developmental trend from
classification by overall similarity to classification by dimensions has
been replicated a number of times (e.g., Kemler, 1982b; L. Smith, 1979,
1981; Shepp, Burns, & McDonough, 1979; Ward, 1980). The results
support the view that overall similarity is a primary perceptual relation
by which young children compare objects.
Findings about the primacy of overall similarity in children’s perception
may be relevant to issues in the development of classification skills. The

3. a
l b(25) c(lS)* l b(25)
*a(141 l c(14)

.d (22) .d (22)
l C(lll efll). l f(21)

.,I
a
l b(25551 l b (2535)
o(lsb4)

.O (1444)
.C (1455)

l d (2213)

l d (2222)

.C(llll) e(1311). .f (2121)

FIG. 1. Schematic illustrations of stimulus sets employed in the three experiments: (A)
Sets 4, two dimensions; (B) Sets 6. two dimensions; (C) Sets 4, four dimensions; and (D)
Sets 6, four dimensions. Individual objects in the stimulus sets are represented as dots
such that the overall dissimilarity of objects within a set are approximated by the distances
of the dots. Examples of specific orderable values on the two (XY) or four (XYZW)
dimensions that combine to form unique objects are given in parentheses.
152 LINDA B. SMITH

young child traditionally has been viewed as unable to actively and


planfully construct systematic classifications (e.g., Bruner, Olver, &
Greenfield, 1966; Inhelder & Piaget, 1964; Vygotsky, 1962). This view
is based on one particular definition of a proper classification. According
to this definition, a proper classification is one in which objects are
partitioned into mutually exclusive groups according to some criteria1
property, for example, by identity on one dimension. Under this definition,
young children do classify poorly (e.g., Denney, 1972, 1974; Inhelder &
Piaget, 1964: Kofsky, 1966: Vygotsky. 1962). Given young children’s
proclivity for comparing objects holistically, this inability to construct
“proper,” that is, dimensional, classifications is not surprising. However,
if one employs another definition of a proper classification-one that
includes systematic overall similarity groupings-the young child’s clas-
sification skills might justifiably be viewed as substantial.
Several lines of argument favor the acceptance of overall-similarity
classifications as proper and nondeficient. First, overall-similarity clas-
sifications are structured classifications. Objects are grouped together
because they are. overall, very much alike and objects are grouped apart
because they are, overall, dissimilar. Such a classification seems no less
organized than one in which objects are grouped together because they
are exactly alike in one way. Second, overall-similarity classifications
are not restricted to the developmentally immature. Adults classify by
overall similarity when stimuli are complex and vary on more than two
dimensions (L. Smith, 1981). Finally, overall-similarity classifications
seem particularly reasonable in the context of current theories about the
structure of natural categories (e.g., Medin & Schaffer, 1978; Mervis &
Pani, 1980; Rosch, 1973; E. Smith & Medin, 1981). Many basic categories
such as dog and chair appear to be organized by global similarity relations.
Thus overall-similarity classifications appear to be a systematic and im-
portant way by which humans organize objects.
Arguments also may be mustered in favor of the contrary view that
overall-similarity groupings reflect limited classification skill. As Sugarman
(1982) has recently suggested (see also Johnston, in press), the young
child’s overall-similarity classifications may result from successive and
piecemeal comparisons of objects. Children may form what appear to
be strategic overall-similarity classifications by successively chaining highly
similar objects. The possibility that overall-similarity classifications result
from individual object matches without an overriding plan is compatible
with one interpretation of the developmental trend in perception. According
to this view, the trend from holistic to dimensional perception is a transition
from nonanalytic, stimulus-bound comparisons of objects to strategically
controlled comparisons (see Baron, 1978: Kemler, 1982a; L. Smith, 1979).
In other words, young children may perceptually group objects by overall
similarity because they are unable to strategically organize object com-
DEVELOPMENT OF CLASSiFICATION 153

parisons. This view that overall-similarity classifications are nonstrategic


also is consistent with theories of natural concept formation and the
young child’s ease in forming such concepts. Several theories about
natural categories state that overall-similarity categories may be best
acquired tacitly (i.e., without explicit hypothesis testing) and reflect the
automatic or nonstrategic operation of perceptual and memorial processes
(Brooks, 1978; E. Smith & Medin, 1981; see also Kossan, 1981).
These contrasting views of the nature of children’s overall-similar clas-
sifications both presuppose that young children-by some means-form
systematic similarity classifications. This assumption may not be correct.
In all of the studies showing that young children classify by overall
similarity, the children classified sets of at most four objects. An “overall
similarity” classification of the four objects in Fig. IA could result from
the pairing of one object to a second highly similar object with the leftover
objects, by default, forming the second group. Such a one-object-to-one-
object match and a leftover pair hardly seems to justify the label “systematic
similarity classification.”
In sum, the degree of classificatory skill that underlies the young child’s
overall-similarity classifications is unclear. The following experiments
were designed to clarify the nature of children’s similarity classifications.
Preschoolers and kindergarteners were asked to classify relatively small
and large sets of objects. Following Sugarman (1982) the specific steps
through which the children progressed in constructing a classification as
well as the final structure of the classifications were analyzed.
EXPERIMENT 1
Children were asked to classify sets containing four objects and sets
containing six objects. The structure of the Sets 4 is shown in Fig. 1A.
The structure of the Sets 6 is represented schematically in Fig. 1B. The
Similarity classification of this set consists of forming two groups of three
objects (a, b, c, vs d, e, f). A dimensional classification of the set into
two groups of three objects by identity on Dimension X (a, c, e vs b,
d, f) is also possible. Similarity classifications rather than Dimensional
classifications were expected to be dominant. The question was how and
whether children would construct similarity classifications of the Sets 4
and 6.
Method
Subjects. Twelve children (6 male, 6 female) at each of two age levels
participated, preschoolers (M age, 4 years, 1 month; range 3-7 to 4-6)
and kindergarteners (M age, 5 years, 5 months; range, 5-2 to 5-l 1). All
children attended a preschool and kindergarten serving a predominantly
middle-class population.
Stimuli. The stimuli were circles varying in color and size. Color varied
154 LINDA B. SMITH

in five steps from light to dark green; in Coloraid notation the five colors
were YG-T,, YG-TI, GYG-T, , GYG-H, G-S. The five sizes or diameters
of the circles were 3.5, 4.0, 4.75, 5.75, and 6.75 cm. The specific values
on the two dimension were chosen from similarity scaling data generated
by four adults such that the range of similarities on the two dimensions
were comparable for the adults. Further, the discriminability of all one-
step differences on each dimension were tested via a same-different
oddity task in which four preschoolers participated. All one-step differences
were highly discriminable for the children.
Values of color and size were combined to form eight unique Sets 4
structured as shown in Fig. IA and eight unique Sets 6 structured as
shown in Fig. 1B. These unique sets were selected from various parts
of the two-dimensional space such that within each set all items within
a potential similarity group were judged by four adults to be more similar
overall to each other than all potential items within a Dimensional group.
For half the Sets 4 and half the Sets 6, the Dimensional classification
into two groups was possible by color (i.e., Dimension X in the figure
was color); for the other half of the sets such a Dimensional classification
was possible by size.
Procedure and design. Half the children at each age level classified
the 16 unique sets in one random order and half in a second random
order. Each child was tested individually in two sessions approximately
20 min each. To introduce the task, the child classified two demonstration
sets at the beginning of each session. The two demonstration sets each
contained five objects. One set was composed of triangles that varied
only in size, three small and two large triangles. The second demonstration
set consisted of squares that varied only in color, three yellow and two
red squares. The experimenter put the first demonstration set (for half
the children the color set, for the other half the size set) before the child
and told the child to “make groups, put the ones together that belong
together.” If the child did not spontaneously classify the identical objects
together (only two did not), the experimenter showed the child the correct
classification. All children correctly classified their second demonstration
set.
After the classification of the demonstration sets, the experimenter
told the child that there were more pictures and the child was again to
“make groups and put together the ones that go together.” The classification
sets were then, one at a time, given to the child to classify. The child
was allowed as much time as desired to classify a set. No specific feedback
was given to the child about his or her classification.
A second experimenter, the scorer, recorded the sequence of steps
through which the child progressed to complete a classification, that is,
the sequence of the behaviors of touching individual objects and of
moving one object close to another. The scorer drew schematic figures,
DEVELOPMENT OF CLASSIFICATION 155

making a new drawing each time an object was moved, added or taken
away from a group.

Results and Discussion


The major question of this experiment is the relative ease and manner
with which the children produced overall similarity classifications of the
Sets 4 and 6. Before considering this question, one preliminary issue
must be addressed. Given the structure of the classification sets, a fair
number of Similarity classifications could result not from the holistic
comparison of objects in terms of their overall similarity but rather from
the differentiated comparison of objects in terms of one and only one
dimension. For example, a child could attend selectively only to the
color of the circles. As examinations of Figs. IA; B reveal, such a child
might construct Dimensional classifications when Dimension X is color
(i.e., on half the trials) but Similarity classifications when Dimension X
is size (i.e., on the remaining trials). Such Similarity classifications would
then result not from attention to overall similarity but rather only to
similarities in color. To ensure that the analyses of group performances
in the construction of Similarity classifications accurately reflected attention
to overall similarity and not selective attention to one dimension, each
child’s pattern of classifications was examined separately for the Sets 4
and 6. A child was deemed a Preferrer if the child constructed Dimensional
classifications by one dimension (e.g., color) on at least three of the four
sets on which such a classification was possible and constructed no more
than one Dimensional classification by the other dimension (e.g., size).
Only three children, all kindergarteners, were Preferrers by this measure;
two of these children were Preferrers only when classifying Sets 4. Thus,
the proportions of Similarity classification reported below predominantly
reflect true overall similarity classifications.
One final preliminary result merits mention. Reorganizations of begun
classifications occurred rarely (less than 2% of the time). Specifically,
objects put together into a group were not subsequently separated by
the child. Thus in the following analyses of the steps through which the
children constructed classifications, the very few classifications that in-
cluded a removal of an object from a formed group were ignored.
Sets 4. The left-hand portion of Table 1 shows the proportion of Similarity
and Dimensional classifications of the Sets 4 by the preschoolers and
the kindergarteners. As is evident in the table, there are no developmental
differences in the classification of Sets 4. As expected, Similarity clas-
sifications were predominant, occurring reliably more often than Di-
mensional classifications (t(23) = 3.03, p < .Ol). Further, the manual
behaviors of the children suggest the explicit formation of two groups.
On 85 and 93% of all classifications at the two age levels, respectively,
the children formed two spatial groups by first putting two objects into
156 LINDA B. SMITH

TABLE I
MEAN PROWRTIONS OF SIMILARITY (S) AND DIMENSIONAL (D) CLASSIFICATIONS OF SETS 4
AND MEAN PROPORTIONS OF SIMILARITY 6). DIMENSIONAL (D), AND PAIR (P)
CLASSIFICATIONS OF SETS 6 IN EXPERIMENT I

Classification set

Sets 4 Sets 6

Age level s D s D” P

Preschool .66 .29 .08 .07 .73


C.16) C.12) C.07) C.07) C.12)

Kindergarten .66 .30 .2s .I8 .s4


C.19) C.13) c.13 f.19) C.29)
Note. Standard deviations are given in parentheses.
” Includes classifications into two groups of three by Dimension X and classifications
into four groups by Dimension Y (see Fig. 1B). The second kind of dimensional classification
occurred rarely, less than 2% of the time at both age levels.

spatial proximity and then moving the remaining two objects close together.
Consistent with previous studies (e.g., Kemler, 1982b; L. Smith & Kemler,
1977), young children construct overall similarity categories at least when
each category contains only two objects.
Sets 6. As shown in the right-hand portion of Table 1, Similarity
classifications were rarely constructed with Sets 6. The predominant
classification involved the construction of Pairs-three groups of two
objects each. In their formation of Pairs, the children proceeded quite
systematically, forming first one pair, then a second, and finishing a
classification by putting the remaining two objects together. The structure
of these Pair constructions and their implications concerning the children’s
level of classificatory skill is clarified by considering the kinds of pairs
that could be formed from Sets 6. Fifteen unique pairs of four types
were possible: (1) S + D pairs-four possible pairs in which objects
share a value on one dimension and are highly similar overall (e.g., Pairs
a, b and e, fin Fig. 1B); (2) S pairs-two possible pairs in which objects
do not share a value on either dimension but are highly similar on both
(Pairs b, c and d, e); (3) D pairs-four possible pairs in which the objects
share a value on one dimension but are not similar on the other (e.g.,
pairs a, e and b, f); and (4) Other pairs-five possible pairs in which the
objects neither share a value on one dimension nor are similar overall.
On 88% of all constructions, the first two pairs of objects formed by
the children were S + D or S pairs. Both S + D and S pairs are good
overall similarity pairings and thus the children’s Pair classifications appear
to be partial Similarity classifications. This interpretation of Pair clas-
sifications as incomplete Similarity classifications rather than incomplete
DEVELOPMENT OF CLASSIFICATION 157

Dimensional classifications is supported by two additional aspects of


these classifications. First, since there are twice as many possible S +
D pairs as S pairs, the proportions of first two formed pairs that were
S + D (.61) and S pairs (.31) are just what would be expected if the
children were equally likely to form each kind of overall similarity pair.
Second, children formed two S + D pairs on only 31% of all Set 6 trials
and on half of these (or 16% of all trials), the dimension (color or size)
that was constant in a pair was not the same in the two pairs. The
structure of children’s Pair classifications thus strongly suggest that the
children were attentive to overall similarity relations. Indeed, in their
constructions the children came very close to complete Similarity clas-
sifications. At the point at which a child had formed his or her first two
pairs in a construction, the child could have easily formed a complete
Similarity classification by (appropriately) adding the remaining stimuli
to the formed pairs. However, on 70% of all trials on which two similarity
pairs had been formed, the child manually moved the two remaining
objects together and formed a third group. Many of the children explicitly
stated that the two leftover objects “don’t go.”
Some developmental differences were observed (see Table 1). Spe-
cifically, the older children were less likely than the younger children to
construct Pairs (t(22) = 1.90, p < .05) and somewhat more likely to
construct Similarity and Dimensional classifications. The major result,
however, is the finding that young children do not spontaneously construct
overall similarity categories of more than two objects.
Conclusion. The assumption that young children easily form-by what-
ever means-similarity classifications does not appear to be correct. The
children classified, in a sense, systematically; they formed organized
pairs of objects but they did not extend the pairs to form a complete
overall similarity classification. The children’s performance is thus rem-
iniscent of traditional descriptions of early classifications as successive
comparisons of individual objects (e.g., Inhelder & Piaget, 1964). One
could argue that these pairings of objects merely show that the children
did not understand the task. The instructions “put together the ones that
go together,” are vague. However, all the children correctly classified
the second demonstration set used to introduce the task (and all but two
spontaneously classified correctly the first demonstration set). A correct
classification of the demonstration set required the formation of a category
containing three identical objects. Apparently, young children understand
the task if the organizing relation is compelling enough (i.e., total identity).
Further, the children were attentive to holistic similarity. Both when
given four objects to classify and when given six objects to classify, the
children formed pairs of holistically similar objects. What they did not
do was form complete overall-similarity classifications when such a clas-
sification required a group of three objects. Complete overall-similarity
158 LINDA B. SMITH

classifications apparently require more than attention to overall similarity


and more than the successive comparison of objects.
EXPERIMENT 2
The second experiment examined two hypotheses about children’s
difficulties in constructing groups containing more than two objects. First,
although young children might prefer to pair objects, they might also
readily divide six objects into two systematic groups of three if explicitly
instructed to form two groups. Thus, children’s free classifications were
compared with (instructionally) restricted classifications. Second, young
children might construct overall-similarity classifications rather than pairs,
if the overall similarity relations between the potential members of a
category were particularly strong. This possibility was investigated by
asking children to classify objects that varied on four dimensions as well
as objects that varied on only two dimensions. Previous work (L. Smith,
1981) has shown that even adults spontaneously classify stimuli varying
on four dimensions by overall similarity. Presumably, holistic similarity
is stronger, the more dimensions contribute to the relation,

Method
Subjects. Forty-eight preschoolers (M age, 4 years, 2 months: range,
3-8 to 4-6) and 48 kindergarteners (M age, 5 years, 7 months: range, 5-
3 to 6-O) who had not participated in Experiment I served as subjects.
The children attended one of several preschools or kindergartens serving
a middle-class population. Twelve children at each age level were randomly
assigned to one of four experimental conditions such that at least five
males and at least five females at each age level participated in each
condition.
Stimuli. Two types of stimuli were employed-stimuli which varied
on four dimensions and stimuli derived from the four-dimensional set
that varied on two of the four dimensions. Figure 2 shows examples of
the objects that comprised the four-dimension set. The four variable
dimensions were height of the rectangle, shape of the trapezoid on top
of the rectangle, color of the rectangle, and the orientation of an arrow
superimposed on the rectangle. As in Experiment 1 the specific values
on the dimensions were chosen from similarity-scaling data generated
by adults. The discriminability of specific values for the children was
confirmed via a same-different oddity task.
In the Four-Dimension conditions,‘the specific values were as follows:
(1) The Height of a rectangle of a constant l-in. width was one of the
following five values: 1.25, 1.50, 1.75, 2.00, or 2.25 in. (2) The Color of
the rectangle varied from a purplish-red to a bluish-violet, Coloraid notation:
RVR, RV, VRV, V, VBV. (3) The Shape which sat on the top of the
rectangle was a black trapezoid of constant height, 0.50 in., and constant
DEVELOPMENT OF CLASSIFICATION 159

FIG. 2. Examples of the stimulus objects varying on four dimensions (Height of rectangle,
Color of rectangle, Shape of trapezoid, Orientation of angle) employed in Experiments 2
and 3.

base width, 1.75 in. The shape was varied by varying the top width:
0.12, 0.38, 0.62, 1.00 and 1.38 in. (4) A white Arrow of constant 0.75
in. length was superimposed on the bottom left of the rectangle, and
rotated about a point 0.12 in. from the left side and 0.25 in. from the
base of the rectangle. The orientations of the arrow were 5, 25, 45, 65,
and 85” counterclockwise from horizontal.
The Two-Dimension condition was composed of six subconditions
resulting from all pairwise combinations of the four dimensions. In each
of these subconditions only two of the dimensions varied across the five
values listed above. The remaining two dimensions were “not present.”
When Height did not vary, it was constant at 1.25 in. When Color did
not vary, the rectangle was black. When Arrow or Shape did not vary,
they were literally not present.
For the Four-Dimension condition and each of the 6 Two-Dimension
subconditions, 20 unique Sets 6 were constructed. These unique sets
were selected from various areas in the multidimensional space such that
within a set all potential members of the similarity category were judged
more similar to each other by adults than potential members of separate
similarity categories. A sample Two-Dimension set is shown in Fig. 1B
and a sample Four-Dimension set is shown in Fig. ID. The relative
160 LINDA B. SMITH

dissimilarity of stimuli within the two sets is illustrated in the figure by


relative distance. Each set in both stimulus conditions afforded one Di-
mensional classification (by Dimension X in Fig. I) of the objects into
two groups of three objects each (a. c, e vs b, d. f) and an overall
similarity classification (a, b, c vs d, e, f). In each condition and sub-
condition, each specific dimension served equally often as Dimension X.
The 20 unique sets comprising each condition and subcondition were
arranged in two random orders for presentation.
Design and procedure. Half the children at each age level were given
the Four-Dimension stimuli to classify and half were given the Two-
Dimension stimuli. Equal numbers of children at each age level classified
the stimuli corresponding to the six subconditions of the Two-Dimension
condition. Within each of the two stimulus conditions, half the children
classified the stimuli under free-classification instructions. For these children
the procedure was identical to that employed in Experiment I. The
remaining children classified the stimuli under restricted-classification
instructions. These children classified the sets on a low table divided in
half by a black line. The child was initially given the first demonstration
set employed in Experiment I and told “Make two groups. Put the ones
together that go together. Some of the pictures go together over here
and some go together over here.” The child then classified the first and
second demonstration sets. These same instructions were given as needed
throughout the experiment. No child either under the free- or restricted-
classification instructions erred in classifying either demonstration set.
The children were tested individually in two sessions lasting approximately
20 min each.
Results and Discussion
A preliminary analysis of individual performances was conducted to
determine whether any of the children were selectively attending to one
dimension and thus inflating apparent Similarity classifications. A child
was deemed a Preferrer if the child constructed at least 80% of the
Dimensional classifications possible by one dimension and no more than
20% of the possible Dimensional classifications by any other dimension.
Only four of the 96 children, all kindergarteners, were Preferrers by this
measure. Thus, the proportions of Similarity classifications reported below
reflect primarily “true” overall-similarity classifications.
Final similarity class$cations. The major question of this experiment
was, again, how readily the children constructed overall-similarity clas-
sifications. Table 2 shows the mean proportions of Similarity classifications
in the various conditions as well as the mean proportions of other types
of classifications. The numbers of Similarity classifications per child were
submitted to an analysis of variance for a 2 (Age level) x 2 (Stimulus
condition) x 2 (Task) between-subjects design. The analysis revealed
DEVELOPMENT OF CLASSIFICATION 161

TABLE 2
MEAN PROPORTIONS OF SIMILARITY (S), DIMENSIONAL (D), AND PAIR (P) CLASSIFICATIONS IN
EXPERIMENT 2

Task

Free Restricted

Age level Stimulus set S D” P S D P

Preschool 2 Dimensions .I0 .lO .71 .29 .I0 .42


4 Dimensions .40 .04 .47 .53 .08 .38
Kindergarten 2 Dimensions .29 .26 .41 .45 .30 .22
4 Dimensions .69 .I9 .I1 .78 .lO .lO

” Includes Dimensional classifications into two groups by Dimension X and into four or
five groups by Dimensions Y, Z, and W. The latter classifications occurred rarely, less
than 1% of the time.

significant main effects of Age (F(1, 88) = 31.61, p < .OOl), Stimulus
condition (F(1, 88) = 151.18, p < .OOl) and Task (F(1, 88) = 72.22,
p < .OOl). No interaction approached significance. As is evident in Table
2, the older children produced more Similarity classifications than the
younger children; more Similarity classifications were constructed when
four dimensions varied than when two dimensions varied, and Similarity
classifications were more frequent in the Restricted task in which the
children were explicitly instructed to form two groups than in the Free
task in which the children were simply told to “make groups.”
The results suggest that young children can, at least in some situations,
construct complete overall-similarity classifications. When children are
given the instructional clue to form two groups and when the stimuli
vary on relatively many dimensions such that the holistic similarity relations
are perceptually strong, the majority of all classifications are Similarity
classifications. Similarity classifications were produced with high con-
sistency by the kindergarteners when the stimuli varied on four dimensions.
The younger children, also, constructed Similarity classifications (and
not just pairs of objects) but the tendency to do so was considerably
weaker than that evident in the kindergarteners. Similarity classifications
constituted the majority of all preschoolers’ classifications (and just barely)
only when the perceptual relation was strong and when the children were
explicitly told to form two groups.
Pair classrjications. As in Experiment 1, the major alternative means
of classifying the sets was to construct Pairs. A substantial proportion
of the younger subjects’ classifications in all conditions were of this kind
(see Table 2).
The structure of the Pair classifications in the Free-classification task
was similar to those observed in Experiment 1. Pair classifications of
162 LINDA B. SMITH

the two-dimensional sets were begun with either S + D pairs, .61 of all
first two-formed pairs, or S pairs, . 27 of all first two-formed pairs. The
relative proportions of S + D and S pairs suggests, as in Experiment
1, the construction of partial overall similarity classifications. Also con-
sistent with this view is the relative infrequency of Pair classifications
that included two S + D pairs (.40 of all Pair classifications) that were
organized by the same dimension (.19 of all Pair classifications). The
structure of the Pair classifications of the four-dimensional sets is also
consistent with the characterization of these constructions as incomplete
Similarity classifications. All possible pairs within a Similarity group in
the four-dimensional condition shared a value on one of the four dimensions
and thus were S + D pairs. However, the children rarely formed two
S + D pairs by the same dimension (.08 of all Pair classifications).
The instructions of the Restricted-classification task were meant to
preclude the construction of three groups each containing two objects.
Nevertheless, the younger children in particular tried to construct such
pairings. These Pair classifications in the Restricted task were begun in
the same manner as those observed in Experiment 1 and in the Free-
classification condition of this experiment. On the two-dimensional sets
the two initial pairs were primarily S + D (.63) and S (.31) pairs and
constructions containing two S + D pairs organized by one dimension
occurred rarely (. 17 of all Pair constructions). On the four-dimensional
sets, the children began these classifications with two S + D pairs (the
only overall similarity pairs possible) on .95 of all trials; however, these
two pairs were organized by the same dimension on only .09 of all Pair
classifications. In other words, the children again formed two initial pairs,
the members of which were holistically similar to each other and proper
members of a, perhaps, ultimate overall similarity group. After these
beginnings, many children in the Restricted task protested that the remaining
two objects “don’t go.” When told that each object had to go on one
side of the line or the other, most children put the remaining two objects
on one side of the black line but spatially separated this “other” pair
from the earlier formed overall similarity pair. This type of 2 vs 2 + 2
classification was scored as a Pair classification.
The children did not appear to be happy with their “compromise”
Pair classifications in the Restricted classification task. Several children
throughout the experiment tried after first constructing two Similarity
pairs to hide the remaining two objects by sitting on them. When reminded
of the instructions, these children did complete the classification into a
2 vs 2 + 2 structure. One ingenious and diplomatic I-year-old in the
Restricted task constructed two similarity pairs, and then readily added
the remaining two objects to one pair. However, she ended every clas-
sification by laying her arm across the table such that the extra, unrelated,
two objects in the group of four were obscured. An equally ingenious
DEVELOPMENT OF CLASSIFICATION 163

preschooler added-n virtually every trial-the two leftover objects to


one formed pair by inserting the leftover objects under the other two so
that they were not visible. Apparently, these children understood the
instructions to make two groups and knew that the objects within a group
should be related to each other in some way. However, the solution of
finding a relation that held between more than two objects does not seem
to have occurred to them. For these children, attention to overall similarity
was apparently insufficient for the construction of a complete similarity
classification.
Steps in the construction of classijcutions. The claim that attention
to overall similarity does not necessarily result in the formation of complete
classifications is emphasized by examining one initial phase in the children’s
constructions-that in which two overall similarity groups of two objects
each had been formed. Each child’s number of classifications begun with
two similarity pairs (i.e., S + D pairs and/or S pairs) was submitted to
an analysis of variance for a 2 (Age) x 2 (Stimuli) x 2 (Task) between-
subjects design. The analysis yielded only a main effect of age (F(1, 88)
= 26.92, p < .OOl). The younger children began their constructions by
forming two similarity pairs on 80% of all trials whereas the older children
began their constructions in such a manner on 53% of the trials. (The
older children often began their classifications in a slightly different man-
ner-a point to which I will turn shortly.) Clearly, the younger children
were attending to overall similarity relations at least in the initial phase
of each classification. The fact that neither the main effect of stimuli nor
the main effect of task approach significance (F(1) 88) < 2.00 in both
cases) indicates that the effects of number of varying dimensions and
task instructions on final similarity classifications are not mediated in a
simple way by attention to overall similarity relations.
Table 3 shows the conditional probabilities of a final Similarity clas-
sification given a beginning classification consisting of two Similarity
pairs. These data were analyzed by submitting individual subjects’ con-
ditional probabilities to an analysis of variance for a 2 (Age) x 2 (Stimuli)
x 2 (Task) between-subjects design. All three main effects were significant.
Older children completed begun Similarity-Pair classifications by ap-
propriately adding the remaining two objects more often than did the
younger children (F(1, 88) = 40.16, p < .OOl); more classifications begun
in such a manner were completed as Similarity classifications under
restricted than under free instructions (F(1, 88) = 7.43, p < .Ol) and
when four dimensions varied than when two dimensions varied (F( 1, 88)
= 24.68, p < .OOOl). None of the interactions was statistically reliable.
Both older children and younger children, then, may attend to overall
similarity relations but the older children are better able to complete
begun Similarity classifications. The tendency to use overall similarity
to organize more than two objects into a group is also fostered by per-
164 LINDA B. SMITH

TABLE 3
THE CONDITIONAL PROBABILITY 01; A FINAL COMPLETE SIMILARITY CLASSIFICATION GIVEN
A CLASSIFICATION BEGUN WITH Two SIMILARITY PAIRS FOR PRESCHOOLERS
AND KINDERGARTENERS UNDER FREE AND RESTRICTED INSTRUCTIONS
AND WHEN 2 AND 4 DIMENSIONS VARIED
.~~__ ~,.
Task

Age level Stimulus set Free Restricted

Preschool 2 Dimensions .II .33


(.14) t.27)
4 Dimensions .39 .56
C.25) C.25)
Kindergarten 2 Dimensions .47 .61
C.32) C.27)
4 Dimensions .I9 .83
C-24) f.26)

Note. Standard deviations are in parentheses.

ceptually strong similarity relations and by explicit instructions intended


to preclude the formation of only pairs.
The greater ability of older children to construct complete Similarity
classifications also is evident in the manner by which the older children
sometimes began their classifications. Whereas the younger children pre-
dominantly (80% of the time) began their classifications by first forming
two similarity pairs, the older children sometimes began their classifications
by first forming one similarity category of three objects and then completing
the classification by forming a second similarity category of three objects.
Across all conditions older children began classifications by first forming
one complete similarity group 25% of the time, whereas the younger
children did so only 6% of the time. Many more older than younger
children constructed at least one similarity classification in this manner
(x2(1) = 18.74, p < 301). This result, in conjunction with those concerning
the completion of begun pairs, again indicates an age-related increase in
the ability to compare and organize more than two objects at once.
Conclusion. This experiment more strongly than the first indicates a
developmental increase in the ability to construct systematic similarity
classifications. This developmental increase in the ability to classify is
not simply dependent on attention to the organizing relation. The next
experiment was designed to explore further the possibility of a devel-
opmental increase in classification skill that is independent of the organizing
relation.
EXPERIMENT 3
Although the results of the first two experiments suggest that young
children are not fully competent classifiers by overall similarity, they
DEVELOPMENT OF CLASSIFICATION 165

still might be “better” classifiers by this relation than by other possible


relations. Overall similarity is the primary perceptual relation by which
young children compare objects (e.g., Kemler, 1982b; L. Smith, 1979).
Thus, if young children’s level of classificatory skill is closely linked to
their perceptual skills, young children may be more able (though not
well able) to form systematic categories by overall similarity than by
component dimensions. A contrasting hypothesis is also possible. The
developmental trend in classification may reflect an increase in some
general ability that is at least partially independent of the particular
relation by which categories may be formed. A child who is able to
compare and organize together several objects at once may be able to
do so by a variety of relations, specifically by both overall similarity and
by identity on a dimension.
The first two experiments do not provide a test of these two contrasting
possibilities. The stimulus sets employed in the earlier experiments afforded
both overall similarity classifications and dimensional classifications but
the two possible classificatory organizations were pitted against each
other. Hence, one can only conclude that young children prefer to classify
by overall similarity. Therefore, this third experiment compared children’s
classifications of sets offering overall-similarity solutions and their clas-
sification of sets offering only dimensional solutions.
Method
Subjects. Twelve preschoolers (6 male, 6 female-M age, 4 years, 1
month; range 3-5 to 4-5) and twelve kindergarteners (6 male, 6 female-
M age, 5 years, 8 months; range 5-3 to 6-O) served as subjects. All
children attended a day-care center serving a middle-class population
and none had participated in the first two experiments.
Stimuli. The stimulus sets were assembled from the Four-Dimension
stimuli employed in Experiment 2. Five types of stimulus sets were
employed. The structure of two of these types is the same as the Sets
4 and Sets 6 employed in the earlier experiments and are illustrated in
Figs. lC, D. These sets afforded both a similarity and a dimensional
classification. For the sake of simplicity, however, and since overall
similarity is the relation preferred by young children, these two types
of sets will be called Similarity Sets 4 and 6. Eight unique Similarity
Sets 4 and eight unique Similarity Sets 6 were constructed as in Experiments
1 and 2.
Two of the other types of sets, Dimension Sets 4 and 6, afforded no
overall-similarity classification as all stimuli in the sets were approximately
of equal holistic similarity (as measured by adult similarity judgments).
The two types of Dimension sets did offer two distinct dimensional
classifications and are illustrated schematically in Fig. 3. The Dimension
Sets 4 consisted of four unique objects divisible into two groups of two
166 LINDA 6. SMITH

A. B,

00
00 bo (IIll) z4,
(1112) (4135)

do
X45) (5512)

de e. 10
c (7444, (4421) (1523) (5555)

FIG. 3. Schematic illustration of the Dimension Sets 4 (A) and Sets 6 (B) employed in
Experiment 3. Individual objects are represented as in Fig. I.

by identity on Dimension X and into two other groups of two by identity


on Dimension Y. The Dimension Sets 6 consisted of six unique objects
divisible into two groups of three by identity on Dimension X and into
two other groups of three by identity on Dimension Y. Eight unique sets
of each of these two types were employed. Within each type, each of
the four varying dimensions served equally often as Dimension X and
Y.
The fifth type of set was an Identity Set 6 and consisted of three
instances of each of two unique objects that differed on all varying
dimensions. Six unique Identity Sets 6 were constructed.
Thus, a total of 38 unique stimulus sets of five types was employed.
These sets were arranged in two unique orders for presentation. The
orders of sets were randomly determined with the constraint that at least
every sixth set was an Identity set.
Procedure. The procedure was identical to that of the Restricted clas-
sification condition of Experiment 2. In other words, the children were
explicitly instructed to classify the stimuli of each set into two groups.
The children were tested individually in three sessions lasting about 20
min each.
Results and Discussion
Preliminary issues. As in the first two experiments, a preliminary
analysis of individual performances was conducted to determine if some
children’s classifications were governed primarily by one dimension. A
child was deemed a Preferrer if the child constructed at least 75% of the
Dimensional classifications possible by one dimension and no more than
25% of the possible Dimensional classifications by any other dimension.
As in the first two experiments, very few of the children met this criterion
on the Similarity sets-no preschooler did on either the Sets 4 or the
Sets 6 of this type and only three kindergarteners did so on Sets 4 and
only one of these three did so on Sets 6. In contrast, about half the
DEVELOPMENT OF CLASSIFICATION 167

children met the criteria for being labeled a Preferrer on the Dimension
sets. On the Dimension Sets 4, six preschoolers and seven kindergarteners
classified these sets primarily by one dimension. On the Dimension Sets
6, three preschoolers and six kindergarteners classified by primarily one
of the four dimensions. Across children no one of the four dimensions
appears dominant; each of the four dimensions was “preferred” by at
least one child at each age level. These results alone suggest that young
children have some ability to organize objects into groups according to
identity on a dimension. Preference for one dimension over another
strongly implies the analysis of object comparisons into component di-
mensions (see, e.g., Cunningham & Odom, 1978; Garner, 1974).
The major question of this experiment is the relative ease by which
children construct overall-similarity classifications of the Similarity sets
and dimensional classifications of the Dimension sets. Therefore, in the
primary analyses, Similarity classifications of the Similarity sets are com-
pared with the Dimensional classifications of the Dimension sets. Di-
mensional classifications of the Similarity sets, which occurred relatively
infrequently (less than 12% of all classifications of Similarity sets) are
omitted from these analyses. Since only one of the two possible complete
classifications of the Similarity sets but both of the possible complete
classifications of the Dimension sets are included in these analyses. sup-
plementary analyses were also conducted. These secondary analyses
compared number of complete, coherent, classifications of the Similarity
sets (Similarity plus Dimensional classifications) with number of Dimen-
sional classifications of the Dimension sets.
Sets 4. The left-hand side of Table 4 shows the mean proportions of
final Similarity classifications of the Similarity Sets 4 and final Dimensional
classifications of the Dimension Sets 4. An analysis of variance for a
2 (Age) x 2 (Set) mixed design revealed no main effects or interactions
in these data (F < 1.00 in all cases). Evidently, both the younger and
the older children were well able to systematically classify objects into

TABLE 4
MEAN FROWRTIONS OF SIMILARITY CLASSIFICATIONS OF SIMILARITY SETS AND DIMENSIONAL
CLASSIFICATIONS OF DIMENSIONAL SETS AT THE Two AGE LEVELS IN EXPERIMENT 3

Stimulus set

Sets 4 Sets 6
Age level Similarity Dimension Similarity Dimension

Preschool .71 .I7 .42 .33


C.30) (.22) C.31) (20)
Kindergarten .77 .88 .68 .66
c.29) C.21) (28) C.19)

Note. Standard deviations are in parentheses


168 LlNDA B. SMITH

groups of two by both overall similarity and, when there were no competing
overall similarity relations, by a component identity. Inclusion of Di-
mensional classifications of the Similarity set (mean proportions, .09 and
.I9 for the preschoolers and kindergarteners, respectively) does not alter
this conclusion. The numbers of complete coherent classifications of the
Similarity sets (Similarity plus Dimensional) and of the Dimension sets
(Dimensional) do not differ reliably at either age level (t( I I) < 1.OO in
both cases).
Sets 6. All the children classified the Identity sets into two groups of
three by absolute identity. The mean proportions of such classifications
were .97 for the preschoolers and 1.OOfor the kindergarteners. This result
corroborates the earlier suggestion that the children did understand the
instructions to “put together the ones that go together” and were able
to do so when the possible relation for classifying was completely
determined.
In contrast to the results on the Sets 4 and the Identity Sets 6, marked
developiental differences did emerge on the Similarity and Dimension
Sets bthe sets well classified only by forming two groups of three
nonidentical objects. The mean proportions of Similarity classifications
of the Similarity Sets 6 and the mean proportions of the Dimensional
classifications of Dimension Sets 6 are shown in the right-hand side of
Table 4.
Performances on the Similarity and Dimension Sets 6 were analyzed
first by submitting each child’s number of Similarity classifications on
the Similarity sets and each child’s number of Dimensional classifications
on the Dimension sets to an analysis of variance for a 2 (Age) by 2 (Set)
mixed design. The analysis yield only a main effect of age, F(1, 22) =
8.75, p < .Ol; older children formed Similarity and Dimensional categories
of three objects each more often than the younger children. Contrary to
possible predictions from the hypothesis that perception proceeds from
being organized by holistic similarity to being organized by dimensional
relations, coherent Similarity classifications were not more prevalent than
Dimensional ones. Neither the effect of Set (F(1, 22) = 2.09) nor the
interaction between Age and Set (RI, 22) = 1.06) was reliable. Analyses
comparing the number of coherent classifications (Similarity and Di-
mensional) of the Similarity sets with the number of Dimensional clas-
sifications of Dimension sets did yield a reliable Set effect for preschoolers.
Dimensional classifications of the Similarity sets occurred infrequently,
mean proportions .08 and .10 at the two age levels, respectively. None-
theless, when these infrequent dimensional classifications are taken into
account, the number of coherent classifications of the Similarity sets is
reliably greater than that of the Dimension sets for preschoolers (I( I I)
= 3.58, p < .Ol) but not for kindergarteners (HI I) = 1.57. p < .lO).
This correction for number of possible coherent classifications in the two
DEVELOPMENT OF CLASSIFICATION 169

sets is consistent with the view that the perceptual primacy of similarity
over dimensional relations decreases with age. Still, the developmental
increase in complete classifications does not appear to be dependent on
the particular kind of perceptual relation by which objects might be
classified. Older children are better able than younger children to form
coherent categories containing more than two objects regardless of the
relation.
The claim that there is a general ability to form groups of more than
two objects that is independent of kind of perceptual relation is bolstered
by considering the relation between an individual’s performance on the
Similarity Sets 6 and the Dimension Sets 6. The children who constructed
systematic categories by overall similarity tended to be the same ones
who constructed systematic categories by an identity on one dimension.
The correlation (Pearson Y) between Similarity classifications of the Sim-
ilarity Sets 6 and Dimensional classifications of the Dimension Sets 6
was .86 for the preschoolers and .71 for the kindergarteners. The strong
relationship between the use of these two distinct category organizations
suggests the importance of a third factor, the general ability to relate
more than two objects at once.
Again, the primary alternative classification of Sets 6 was the construction
of pairs. The structure of the Pairs formed on the Similarity sets was
identical to that observed under Restricted instructions in Experiment
2; .40 of the preschoolers’ and .20 of the kindergarteners’ classifications
of Similarity Sets 6 were three pairs of objects with two of those pairs
joined to form one group. Analogous pair classifications were formed
with the Dimension Sets bmean proportion .66 and .3 1 for preschoolers
and kindergarteners, respectively. The structure of these pairs is discussed
more fully in the next section.
Steps in the construction of classiJcations. As in the first two exper-
iments, the children rarely altered a begun classification by separating
objects that had been previously grouped together. Such alterations oc-
curred less than 2% of the time and are therefore omitted from the
following analyses. Table 5 summarizes the nature of classification be-
ginnings of the various Sets 6.
As in Experiment 2, a substantial proportion of classifications of the
Similarity Sets 6 began with the formation of two similarity pairs. Thus,
both the younger and the older children appear to have attended to the
overall similarity relations in these sets. However, the older children
were much more likely than the younger children to systematically use
this relation to add a third object to an initial similarity pair. The conditional
probability of a final similarity classification given a classification begun
with two similarity pairs was .31 for the preschoolers and .63 for the
kindergarteners (t(17) = 1.77, p < .05). The older children also tended
more than the younger children to begin a classification by forming one
170 LlNDA B. SMITH

TABLE 5
PROPORTIONS OF CLASSIFICATIONS OF SETS 6 IN
EXPERIMENT 3 BEGUN BY FIRST FORMING
Two PAIRS OR BY FIRST COMPLETING
ONE GROUP OF THREE

Age level

Stimulus set Preschool Kindergarten

Similarity
Similarity pairs .64 .50
Similarity
group of 3 .19 .34
Dimension
Dimensional
pairs .68 .58
Dimensional
group of 3 .I3 .39
Identity
Identity pairs .60 .41
Identity group
of 3 .39 .59

complete similarity group of three objects before starting a second group


(t(22) = 1.04, n.s.); 10 kindergarteners began at least one classification
in this manner whereas only five of the preschoolers did (x2(1) = 4.43,
p < .05).
The steps through which the children constructed classifications of the
Dimension sets are very much like those observed with the Similarity
sets. A substantial proportion of all classifications (see Table 5) began
with the formation of two pairs. The objects in these pairs were not
haphazardly related; objects were paired together that shared a value on
one of the dimensions. Again, both the younger and the older children
appear to have been attentive to the kind of relation by which a coherent
classification might be constructed. However, the older children again
appear more able to systematically use the relation to actually classify
the objects. First, many of the younger children’s first two pairs in the
classification of Dimension sets precluded a final Dimensional classification
as the two initial pairs were often governed by two distinct dimensions
(e.g., same-colored objects might form one pair and same-shaped objects
the second pair). On classifications begun by the formation of two di-
mensional pairs, the preschoolers’ pairs were organized by the same
dimension only 57% of the time. In contrast, kindergarteners’ initial
dimensional pairs were organized by the same dimension (and thus po-
tentially extendable to form a complete dimensional classification) 80%
DEVELOPMENT OF CLASSIFICATION 171

of the time. In other words, the older children more than the younger
children strategically formed complementary dimensional pairs (t(20) =
2.21, p < .02). A second area in which the older children’s use of
dimensional relations appears more systematic than that of the younger
children’s is in the extension of initial pairs to form the final classification.
Given two formed pairs governed by the same dimension, younger children
appropriately added the remaining objects to these pairs to form a final
dimensional classification only 45% of the time whereas the older children
did so 67% of the time (t(20) = 1.75, p < .05). Finally, and as in the
classification of the Similarity sets, the older children more than the
younger children began their classifications by completing a group of
three before starting a second group (t(22) = 3.40, p < .Ol ; see Table
5); 10 kindergarteners began a dimensional classification in this manner
at least once, whereas only five preschoolers did so. Interestingly, these
five preschoolers are precisely the same children who began at least one
Similarity classification with an initial group of three.
Similar differences in the younger and older children’s classifications
of the Identity Sets 6 were also observed. Although both the younger
and older children were well able to classify these sets into groups of
three objects each, children at the two age levels differed somewhat in
the procedures they used to achieve such a classification. The younger
children more than the older children tended to construct these classi-
fications by first forming two pairs and then adding the remaining objects
rather than by completing one group of three objects before beginning
the second group (r(22) = 2.48, p < .02; see Table 5).
Summary and conclusion. The developmental differences in the clas-
sification of the Dimension sets and the Similarity sets correspond closely.
Both the younger and the older children attend to and try to classify by
both holistic similarity and component dimensional relations. However,
the younger children seem generally unable to contemplate relations
between more than two objects at a time. Thus they often fail to form
a complete classification. The older children are more able to classify
strategically. Their ability to compare more than two objects at a time
is applied both when the primary relation is holistic and when it is a
component dimensional identity. The parallel results observed with the
Similarity and Dimension sets strongly suggests a developmental increase
in the general ability to classify that is not strictly dependent on advances
in perceptual development.
GENERAL DISCUSSION
By traditional accounts, the young child is a deficient classifier-unable
to impose an overriding organization and limited to piecemeal and suc-
cessive comparisons (e.g., Bruner & Olver, 1963; Inhelder & Piaget,
1964; Vygotsky, 1962). This traditional account has been challenged (Nel-
172 LINDA B. SMITH

son, 1974; Rosch. Mervis, Gray, Johnson & Boyes-Braem, 1976; Sugarman,
1982; L. Smith, 1979). A major criticism is that the traditional view
focused on only certain kinds of category organizations and ignored other
kinds of systems such as holistic similarity. A reasonable alternative
hypothesis to the traditional view thus emerged (see, L. Smith, 1979).
Young children might be competent classifiers by holistic relations; the
developmental trend in classification might not involve increasing clas-
sification skill per se but only an increase in the ability to use certain
kinds of category structure.
My results do not support the alternative hypothesis. There is remarkably
similar developmental growth in the ability to classify by overall similarities
and by dimensional identities. Apparently, classifying requires more than
attention to a potential organizing relation, Overall similarity comparisons
are clearly primary for young children; my results replicate numerous
others in this regard (e.g., Mervis & Crisafi, 1982; Kemler, 1982b; L.
Smith & Kemler, 1977). However, the greater perceptual primacy of
overall similarity over dimensional relations does not lead to more complete
similarity than dimensional classifications. This result, though somewhat
backhandedly, supports the notion that overall similarity classifications
are not a deficient, “nonstrategic,” form of classification. The construction
of overall similarity categories apparently involves some of the same
organizational abilities as the construction of dimensional classifications-
abilities that are not well established in the young child. Whatever the
precise specification of the critical ability, it is not simply attention to
potential organizing relations. All the children in the present experiments
attended to and indeed grouped objects (albeit incompletely) by overall
similarity and by dimensional attributes. Nevertheless, many of these
children often failed to construct systematic complete classifications.
Thus, the ability to classify systematically appears independent of the
kind of perceptual relation and perceptual development.
My results also are not fully compatible with the traditional claims
although aspects of them are reminiscent of earlier descriptions. The
younger children were well able to form pairs-to group objects on a
one-to-one basis-but were generally unable to impose an organization
that related three objects. Consistent with Piaget’s (1962; Inhelder &
Piaget, 1964) theorizing about the limits of preschool thought, it is as if
these children could conceptualize only one relation per object. Other
aspects of the younger children’s performances, for example, the fluctuating
dimensional criteria used by the preschoolers in classifying the Dimension
Sets 6 of Experiment 3 and the general tendency of all the children not
to break apart any already formed group, also fit with earlier descriptions
of the deficient nature of young children’s classifications (e.g., Denney,
1972; Inhelder & Piaget, 1964; Vygotsky, 1962).
The results also indicate, however, that even the youngest children
DEVELOPMENT OF CLASSIFICATION 173

possess some beginning classificatory abilities. As has been noted by


previous researchers (e.g., Bruner & Olver, 1963; Denney, 1972; Kofsky
& Osler, 1967), young children’s incomplete classifications are not com-
pletely haphazard. Moreover, and contrary to Piagetian claims (e.g.,
Inhelder & Piaget, 1964), the preschoolers partial classifications suggest,
that at very least, these children understood the purpose of a classification.
In particular, the children tended to search for and find pairs of objects
that might be properly grouped together. They did not, for example,
merely successively “chain” similar objects, one to the next-a “piece-
meal” procedure of “successive” comparisons that might be expected
to result, fortuitously, in complete overall similarity classifications (see
Sugarman, 1982). In a sense, even the younger children formed groups
of related objects; these groups, however, were often limited to two
objects each. Further, the children appeared quite unhappy when faced
with the possibility (as in Experiment 2) of putting unrelated objects into
a group and often used extreme measures (e.g., hiding objects) to avoid
doing so. The children seemed to know that classification involves putting
together objects that are related in some way. Their difficulty seems not
to be in knowing what classification is about but in executing an actual
classification.
Pilot data support this suggestion that young children know a bit about
what constitutes a good classification. Eight of the preschoolers in Ex-
periment 3 participated in a second task after the main experiment. In
this second task (following a procedure used earlier by Muller, 1978),
the children were asked to decide which of two puppets had produced
a better classification of a set of objects. The choice was between an
incomplete classification previously constructed by the subject and a
complete classification. On 98% of the six Similarity trials, the children
chose complete similarity classifications as better than their own incomplete
pair classifications and on 96% of the six Dimension trials, the children
chose complete dimensional classifications over their own pair classifi-
cations. These children recognized a well-structured classification when
they saw one. However, they apparently have difficulty in producing
such a classification themselves.
Developmental growth in the ability to execute a classification is evident
in the steps through which the children constructed spatial groupings in
the main experiments. At the lowest level of skill observed, the children
formed only pairs of objects. One pair would be formed and then the
next. Once a pair had been formed, it was essentially unalterable and
final; it would neither be broken apart nor added to. At the next level
of skill, the children formed complete classifications but did so via the
initial formation of pairs. A characteristic classification at this level would
consist of the formation of a pair, a shift to the second group and the
formation of the second pair, and then a shift back to the first group to
174 LINDA B. SMITH

add an object. The third and most advanced level of skill observed in
the experiments consisted of the formation of a complete classification
without shifting between incomplete groups. All potential members of
one group were found and grouped together before the second group
was formed.
This developmental trend, from pairs to successive pairs with addition
to the initial formation of complete groups, seems to contrast with the
developmental trend reported by Sugarman (1982) in her study of the
classification behavior of It- to 2&-year-olds. The I$-year-olds in Sugarman’s
study tended to form only one group; for example, in “classifying” a
particular set, these children might look for and group together only the
dolls, ignoring everything else. The 2-year-olds, in contrast, tended to
successively form complete groups; for example, they might look for
and then group together dolls and then look for and group together circles.
Finally, the 2+-year-olds classified by shifting back and forth between
groups-for example, putting together two dolls, then two circles, then
adding a third doll to the first group, and so on. On the surface, then,
the developmental trend reported by Sugarman seems to be the inverse
of that observed in the present experiments-Sugarman’s 2-year-olds
classified in a manner like that evident primarily in the kindergarteners
(forming one complete group before beginning the second) and Sugarman’s
2+-year-olds classified in a manner most like that of the 4-year-old pre-
schoolers in the present experiment (forming successive pairs and then
adding objects).
There are several methodological differences between Sugarman’s study
and the present one. These differences notwithstanding, the seemingly
opposite directions of the developmental trends may be compatible. Very
young children may form successive final (but perhaps incomplete) groups
because they are not (other than behaviorally) classifying objects into
two or more groups. As Sugarman suggests, very young children may
not be partitioning objects into categories when they select all the dolls
and leave other objects aside. These children may only be focusing on
and finding dolls. As Sugarman argues, shifts between two or more
incomplete groups in the construction of a classification implies an aware-
ness of the potential existence of several groups of related objects. All
the children in the present study appear to have possessed such an
awareness. The lowest level of classification observed-the construction
of pairs-consisted of the explicit formation of two groups of two related
objects and the children avoided grouping unrelated objects together.
The procedure sometimes used by the kindergarteners of forming complete
groups without shifting (the same behavior characteristic of Sugarman’s
youngest subjects) may indicate a further advancement in classification-
the use of a prior, global, plan. The formation of one complete group
followed by the formation of a second reluted and complete group suggests
DEVELOPMENT OF CLASSIFICATION 175

an ability to envision the final structure of a classification while in the


midst of constructing it. Thus very young children and older children
may form successive groups without shifting for very different reasons.
Very young children may form successive final groups because they focus
on only one kind of object at a time. Older children may form successive
final groups without shifting because they have a prior plan that allows
them to form complete interrelated groups, one complete group at a time.
The above suggestions are not meant to imply that Sugarman’s 2$-
year-olds and the preschoolers in the present study possess similar abilities.
The level of classification skill shown by a child is surely dependent
on the specific stimulus and task situations, which were quite different
the two studies. The importance of stimulus factors is demonstrated in
the present results. Preschoolers constructed complete identity classifi-
cations but often formed only partial, pair, classifications when the stimulus
objects were discriminably different. Further, at both age levels, the
frequency of complete similarity classifications increased when the number
of varying dimensions and presumably the perceptual strength of holistic
similarity increased. Task factors are also likely to be important. In the
present study, explicit instructions as to the number of groups to be
formed resulted in increased complete classifications. Previous studies
have also shown that young children benefit from a variety of task supports
(e.g., Denney & Acito, 1974; Markman, Cox, & Machida, 1981; L. Smith,
1979). Strategic classification is probably not an all-or-none ability. With
development, classificatory skills may become less fragile and less de-
pendent on stimulus or contextual support and more flexible and generally
applied.
One final issue deserves discussion-the hypothesized developmental
trend from holistic to dimensionally differentiated perception. There is
ample evidence that young children often do not analyze objects into
their dimensional components but rather tend to compare objects holistically
(e.g., Kemler, 1982b; Kendler, 1979; Miller, 1979; Shepp & Swartz, 1976;
L. Smith, 1980; Tighe & Tighe, 1978; Ward, 1980). However, it is unclear
whether the young child lacks some basic dimensional analysis ability
or whether the young child’s proclivity for nondimensional comparisons
reflects performance limitations (see Kemler, 1982~; Miller, 1979; Tighe
& Tighe, 1978, for further discussions of this issue). The current results
favor the latter view. In the classification sets offering both strong overall
similarity categories and potential dimensional categories, the children
focused their attention on overall similarity-clearly preferring to compare
objects holistically rather than analytically. However, when strong com-
peting holistic-similarity relations were removed from the classification
sets, the children analytically compared and grouped objects by their
dimensional constituents. Moreover, the level of classification skill by
dimensions was as high or as low as the level of skill by overall similarity.
176 LINDA B. SMITH

Thus, young children are able to analyze objects dimensionally and indeed
are able to strategically use these relations in classifications tasks as well
as they use holistic ones.
In conclusion, classification by overall similarity appears to depend
on the development of general classification skills that are also prerequisite
for dimensional classifications. Developmental differences in classification
do not reduce simply to the differential use of overall similarity by
younger subjects and dimensional relations by older subjects. There are
classification skills-independent of the organizing relation-that them-
selves change with age. The specific limits observed on young children’s
ability to classify by overall similarity are both reminiscent of traditional
accounts of the growth of classification and distinct from those accounts.
In their execution of a classification, young children often do appear to
have difficulty going beyond one-to-one object comparisons. However,
even the youngest children-and those with the least classification skill-
seemed to know that the purpose of a classification is to organize related
objects into separate groups.
REFERENCES
Baron, J. Intelligence and general strategies. In G. Underwood (Ed.). Sfrutegies in informution
processing. New York/London: Academic Press. 1978.
Brooks, L. Nonanalytic concept formation and memory for instances. In E. Rosch &
B. B. Lloyd (Eds.), Cognition and categorizufion. Hillsdale, N.J.: Erlbaum. 1979.
Bruner, J. S., & Olver, R. R. Development of equivalence transformations in children.
Monograph of the Society for Research in Child Development, 1963. No. 28.
Bruner. J. S., Olver, R. R., & Greenfield. P. M. Studies in cognitive growth. New York:
Wiley, 1966.
Cunningham, J. G.. & Odom. R. D. The role of perceptual salience in the development
of analysis and synthesis process. Child De\Belopment, 1978. 49, 815-823.
Denney. N. W. A developmental study of free classification in children. Child De~~elopment,
1972. 43, 221-232.
Denney, N. W.. & Acito, M. A. Classification training in two and three year old children.
Journal of Experimental Child Psychology, 1974, 17, 37-48.
Garner, W. R. The processing of itzformation and structure. Potomac. Md.: Erlbaum, 1974.
Gibson, E. J. Principles qf perceptual learning and developmenr. New York: Appleton-
Century-Crofts, 1969.
Inhelder, B., & Piaget, J. The ear/y growth of logic in fhe child. New York: Norton, 1964.
Johnston, J. R. Cognitive prerequisites: The evidence from children learning English. In
D. Slobin (Ed.), Universals of language acquisition. Hillsdale, N.J.: Erlbaum. in press.
Kemler, D. G. Wholistic and analytic modes in perceptual and cognitive development. In
T. Tighe & B. G. Shepp (Eds.), Inferactions: Perception, cognition, und development.
A second Dartmouth-multiperspective conference. Hillsdale. N.J.: Erlbaum, 1982. (a)
Kemler, D. G. Classification in young and retarded children: The primacy of overall
similarity relations. Child Development. 1982. 53, 768-779. (b)
Kemler, D. G. The ability for dimensional analysis in preschool and retarded children:
Evidence from comparison, conservation and prediction tasks. Journal of Experimental
Child Psychology, 1982, 34, 469-489. (c)
DEVELOPMENT OF CLASSIFICATION 177

Kendler, T. S. Cross-sectional research, longitudinal theory, and discriminative transfer


ontogeny. Human Developmenf. 1979, 22, 235-254.
Kofsky, E. A scalogram study of classificatory development. Child Development. 1966.
37, 191-204.
Kofsky, E., & Osler. S. F. Free classification in children. Child Development, 1967. 38,
921-931.
Kossan, N. E. Developmental differences in concept acquisition strategies. Child Devel-
opment. 1981, 52, 290-298.
Markman, G., Cox, B.. & Machida, S. The standard object-sorting task as a measure of
conceptual organization. Developmental Psychology. 1981, 17, 1IS-1 17.
Medin. D. L.. & Schaffer. M. M. Context theory of classification learning. Psychological
Revienl, 1978, 85, 207-238.
Mervis, C. B., & Crisafi, M. A. Order of acquisition of subordinate-, basic-, and superordinate-
level categories. Child Development, 1982. 53, 258-266.
Mervis, C. B.. & Pani, J. R. Acquisition of basic object categories. Cognitive Psycho/ogy,
1980. 12, 496-522.
Miller, P. H. Stimulus dimension, problem solving. and Piaget. In G. A. Hale & M. Lewis
(Eds.). Attention and cognitive development. New York: Plenum, 1979.
Muller, A. A. The subjective structure of a “good” classijcation for adults and children.
Unpublished dissertation, Indiana University. 1978.
Nelson, K. Concept, word and sentence: Interrelations in acquisition and development.
Psychological Re\*ieMl, 1974, 81, 267-285.
Piaget, J. Play. dreams und imitation in childhood. New York: Norton, 1962.
Rosch, E. H. On the internal structure of perceptual and semantic categories. In T. E.
Moore (Ed.), Cognitive development and the acquisition of language. New York:
Academic Press, 1973.
Rosch, E., Mervis, C. B., Gray, W. D.. Johnson, D. M., & Boyes-Braem, P. Basic objects
in natural categories. Cognitive Psychology, 1976, 8, 382-439.
Shepp. B. E.. & Swartz, K. B. Selective attention and the processing of integral and
nonintegral dimensions: A developmental study. Journal of Experimental Child Psy-
chology, 1976, 22, 73-85.
Shepp, B., Burns, B., & McDonough. D. The relation of stimulus structure to perceptual
and cognitive development: Further tests of a separability hypothesis. In J. Becker
& F. Wilkening (Eds.), The integration qf kformation by children. Hillsdale. N.J.:
Erlbaum, 1980.
Smith, E. E., & Medin, D. L. Categories and concepts. Cambridge, Mass.: Harvard Univ.
Press, 1981.
Smith. L. B. Perceptual development and category generalization. Chi/d Development,
1979, 50, 705-715.
Smith, L. B. Development and the continuum of separability. Perception & Psychophysics,
1980, 28, 164-179.
Smith, L. B. The importance of the overall similarity of objects for adults’ and children’s
classifications. Journal of Experimental Psychology: Human Perception and Performunce,
1981, 1, 811-824.
Smith, L. B., & Kemler, D. G. Developmental trends in free classification: Evidence for
a new conceptualization of perceptual development. Journal of Experimental Child
Psychology, 1977, 24, 279-298.
Sugarman, S. Developmental change in early representational intelligence: Evidence from
spatial classification strategies and related verbal expressions. Cognitive Psychology,
1982, 14, 410-449.
Tighe. T. J., & Tighe, L. S. A perceptual view of conceptual development. In R. D. Walk
& H. L. Pick (Eds.), Perception and experience. New York: Plenum, 1978.
178 LINDA B. SMITH

Vygotsky, L. S. Though/ und lunguuge. Cambridge, Mass.: MIT Press. 1962.


Ward, T. B. Separable and integral responding by children and adults to the dimensions
of length and density. Child Development. 1980. 51, 676-684.
Werner, H. Comparafive psychology of mental development. (Rev. ed.). New York: Scientific
Editions, 1961.

RECEIVED:July 8, 1982: REVISED:November 4, 1982.

You might also like