Professional Documents
Culture Documents
Using the selection paradigm, 120 ^S's were required to identify the relevant
attributes of cither a uaidimensional, conjunctive, or disjunctive concept,
working cither under standard conditions or with a hypothesis required after
each selection. The major result was that requiring hypotheses facilitated
solution of the conjunctive and disjunctive concepts while having no effect
on attainment of the unidimensional concept. A number of analyses were
made of the stimulus sequences selected by .S's, but in none of these was there
a difference between working conditions. Consideration was given to the
possibility that requiring hypotheses leads to more efficient use of the infor-
mation available, and the problems involved in obtaining evidence concerning
this interpretation were discussed.
In the standard selection paradigm for control over the sequence of stimuli. It is
studying concept attainment, each trial thus possible that required hypothesizing
consists of 5 selecting a stimulus from a results in 5s providing themselves with
given array and receiving information con- better, more informative stimulus sequen-
cerning its category membership (positive ces under the selection paradigm, whereas
or negative), with trials continuing until S there is no such opportunity under the
names the correct concept. Compared to reception paradigm. Byers and Davidson
performance under the standard condition, (1967) presented no analyses of the se-
Byers and Davidson (1967) found that- quences selected under their 2 conditions,
requiring 5s to hypothesize after each trial and thus this possibility remains untested.
resulted in more efficient concept attain- A second difference between the studies
ment. They suggested that requiring hy- concerns the type of concept employed.
potheses might have facilitated perfor- Byers and Davidson (1967) used a con-
mance by forcing 5s to (more regularly) junctive concept, while Karpf and Levine
consider the relevance or irrelevance of (1971) employed a simpler, unidimensional
the various stimulus attributes. concept. It has been reliably demon-
In constrast to this finding, Karpf and strated that unidimensional concepts are
Levine (1971) found no difference in per- easier to identify than conjunctive concepts
formance between a control condition and (e.g., Neisser & Weene, 1962). Thus, a
a hypothesis condition. However, there second possibility is that requiring hypo-
arc 2 salient differences between the studies theses will facilitate attainment of a con-
which might account for the difference in junctive concept, while having no effect on
results. First, Karpf and Levine used the the acquisition of unidimensional concepts.
reception paradigm, in which 5 has no This seems to be a plausible explanation.
1 If 5s under the control condition make
This research was supported by a grant from
the University of Illinois Research Board. Ap- rather efficient use of information when
preciation is expressed to Wendy Rothwell, Douglas solving for a unidimensional concept, then
M. Johnson, and Susan Zavala for their assistance a technique which, according to Byers and
in the collection and analysis of the data. The Davidson, leads to relatively efficient in-
paper was completed while the author was a re-
search fellow at the University of Aberdeen, formation utilization is not needed.
Scotland.
3
The present study was designed to pro-
Requests for reprints should be sent to Roger vide a partial examination of these issues.
IJ. Dominowski, Psychology Department, Univer-
sity of Illinois, Chicago Circle, Chicago, Illinois The selection paradigm was used, with 5s
60680. either working under standard conditions
387
388 ROGER L. DOMI\0\YSKI
or required to hypothesize after each trial. used first with the set of geometric forms and then
The design included a unidimensional con- with the set of drawings of men, with 5's assigned
to conditions in order of their appearance at the
cept and 2 bidimensional concepts, con- laboratory.
junctive and inclusive disjunctive. With Procedure. Each S was seen individually. \Yith
a disjunctive concept, members of the the stimulus display in front of .S, the stimulus
positive category do not have a common attributes were described, and .V was asked to
attribute combination, in contrast to a describe an arbitrarily chosen card. A card listing
the attributes was available to S throughout the
conjunctive concept. The study of the experiment. The type of concept and the number
effect of required hypothesizing was thus of attributes to be discovered were explained both
extended to a different type of concept, in general terms and by means of an example.
one which has regularly been found to be Using the example, 5 was asked to select 2 posi-
tive and 2 negative instances and to categorize
more difficult than conjunctive concepts an instance selected by B (chosen to represent, if
(e.g., Haygood & Bourne, 1965; Laughlin, needed, a truth table category not included in S's
1969). In addition to analyzing trials to selections). The instructions included 2 additional
criterion, various analyses of the stimulus reminders of the rule stipulating inclusion in the
sequences selected by Ss were performed positive category. It was stressed that S's task
was to discover, in as few card choices as possible,
in an attempt to determine if requiring the 2 attributes (or the single attribute, in the case
hypotheses produces differences in this ot unidimensional concepts) that E was using as
regard. the basis of membership in the positive category.
Trials continued until 5 identified the correct con-
METHOD cept or until 64 card choices were made.
In both the control and hypothesis-required
Materials. Two stimulus sets were used. One (HR) conditions, S was given a supply of slips
set consisted of geometric forms varying on 6 ot paper listing the stimulus attributes. In con-
binary dimensions: circle or square, 1 or 2 figures, trol conditions, S was told that when he was fairly
black or red, solid or open figures, left or right sure that he knew the solution, he should check
side of card, and small or large. In the other the (1 or 2) attributes he thought were the answer
set, each stimulus was a simple cartoonlike draw- and give the slip to E. It was emphasized that
ing of a man having the following attributes: although .S could make one guess after each selec-
(smoking) cigar or pipe, (wearing) hat or beret, tion, he should not offer a solution unless he was
(wearing) bowtie or necktie, thin or fat, beard or fairly sure of it. In HR conditions, ,S was told
moustache, and tall or short. This second stimulus that he must complete one of the slips after each
set was included solely because we were considering selection. It was emphasized that the purpose of
their use in a subsequent experiment and wished completing the slips was to give E some idea of
to determine that they yielded results comparable how S was proceeding. Instructions included the
to the standard geometric forms. For both stimulus statement that some people preferred to make an
sets, stimuli were drawn on index cards and placed
initial guess about the solution and test it, making
in an array of 8 rows of 8 cards each, with the posi-
a new guess when it proved incorrect, whereas
tion of any card randomly determined. others preferred to start with a collection of at-
For each stimulus set, the order of preference tributes and try to eliminate irrelevant attributes
for the 6 dimensions had been previously deter- as they went along. It was stressed that E did
mined by having each of 30 .S's freely categorize not care what method 6' used, and that what was
the stimuli. The dimensions are listed above from requested was that S, after each selection, indicate
most preferred to least preferred, based on mean what he was currently considering—with the ob-
ranks. For purposes of the present experiment, vious exception that when .S thought he had iden-
the 2 highest ranking dimensions were defined as
tified the solution, he would check only those
"preferred" and the 2 lowest ranking as "not (1 or 2) attributes. In both conditions, it was
preferred." indicated that E would not respond to ..S's "hy-
Design and subjects. The basic design consisted potheses" except to inform S that the task was
of the factorial combination of 2 working condi- over when .S had checked the correct attributes.
tions (control vs. hypotheses required) with 3 con- Each .S worked at his own pace.
cept types (unidimensional, conjunctive, and dis-
junctive). The 2 stimulus sets and preferred and
nonpreferred dimensions (as solutions) were used RESULTS
equally often within each fundamental condition.
The 6's were 120 introductory psychology stu- The measure of task difficulty for each
dents who were fulfilling a course requirement. 51 was the number of trials he required
A schedule was constructed in which the 6 basic
conditions and the relevant dimensions (preferred before identifying the correct concept. It
or nonpreferred) alternated, until a total of 60 .V was possible for S to identify the correct
assignments was produced, This schedule was concept before all possible alternative an-
VRRUAUZIN'G HYPOTHESES IX CONCEPT IDKNTlFICATrON 389
swers had been logically eliminalccl on the propriate lorm were ofleral were 3.2 for
basis of the information S received from HR 5s and 2.4 for control 5s, a nonsignifi-
his stimulus selections. Byers and David- cant difference. These data indicate that,
son (1967) considered the possibility that in the present: study, HR 5s did not
HR 5s might be more likely than control obtain significantly more information about
& to identify the solution before it was the solution by proposing incorrect an-
the only logically remaining concept, and swers than did control 5s. Thus, the
they found that, at criterion, HR Ss had measure of adjusted trials to criterion can
more bits of information remaining (logo be used to compare solution efficiency in
number of concepts logically remaining) the HR and control conditions.
than control Ss. In the present study, By comparing the number of precrite-
the mean numbers of bits remaining at cri- rion solution guesses to the number of
terion were .63 for HR and .30 for control. trials needed to reach criterion, some idea
While these means and their difference are can be obtained of the functional difference
lower than those reported by Byers and between the working conditions. Overall,
Davidson, the difference is in the same control 5s required a mean of 20.2 trials
direction and suggests a slight bias in the to identify the correct concept; the com-
trials-to-criterion measure favoring HR Ss. paratively small number of precriterion
Therefore, the analyses to be presented solution guesses (2.4) seems consistent
were based on an adjusted trials-to-critc- with their instructions to offer a solution
rion measure, determined according to the only if they were fairly sure of it. The
formula used by Byers and Davidson: overall mean number of trials to criterion
Adjusted trials = Trials + (Bits Remain- for HR 5s was 13,9. Since these 5s com-
ing) (Trials/Bit), where the last term is pleted a hypothesis slip after each trial
based on the rate at which alternative but can be considered to have made an
answers were logically eliminated prior to average of only 3.2 precriterion solution
criterion. For the present data, the ad- guesses, it can be seen that most of their
justment was slight, and the outcome of hypotheses involved some number of at-
the analysis was the same for both ad- tributes other than that required for the
justed and unadjusted trials to criterion. solution. Other investigators, although
It might be argued that any trials-to- limiting their analyses to 5s' initial hy-
criterion measure is biased in favor of HR potheses, have reported a tendency for the
6s because it fails to take into account size of 5s' hypotheses to vary from that re-
the greater opportunity allowed HR 5s to quired for solution when 5s are left free to
eliminate an incorrect answer by offering choose their method of finding the solution
it to E and learning that it. is wrong. (Bourne, 1963, 1965; Siegel, 1969). In the
However, this argument is not supported present study, HR 5s completed hypothesis
by analysis of Ss' hypotheses. Recall slips in a variety of ways, and there was
that while control 5s used hypothesis slips very little indication in their hypotheses
only to offer solutions, HR 5s were in- that they followed obvious methods like one
structed to use the slips simply to indicate hypothesis (possible answer) at a time or
what they were currently considering and foctising. In the absence of additional in-
told that they could use any method they formation, all that might safely be assumed
chose. Further, all 5s were carefully in- is that HR 5s were, as required, more ac-
structed regarding the number of relevant tively involved in dealing with the informa-
attributes in the correct concept (see Mil- tion they received than were control 5s.
ler, 1971). Only if 5 checked the correct The initial analysis of adjusted trials to
number of attributes (1 for unidimensional, criterion included as orthogonal factors the
2 for bidimensional) was it possible for 2 working conditions, 3 concept types,
him to eliminate an incorrect answer on 2 stimulus sets, and 2 levels of preference
the basis of E's nonresponsc (= wrong). for relevant dimensions. The major re-
The mean numbers of precriterion trials sults are depicted in Table 1. The overall
on which incorrect hypotheses of the ap- HR mean was considerably lower than the
390 ROGER L. UOMINOYVSKI
cusing if it met either of 2 criteria: («) the Focusing scores for the disjunctive con-
chosen instance belonged to the same cate- cept were lower than those for the con-
gory as the focus instance; or (b) the junctive concept, F (1, 107) = 16.25,
chosen instance belonged to the opposite p < .001 (Table 4). This result is con-
category of the focus instance, but dif- sistent with the report by Conant and
fered from the focus on only one dimen- Trabasso (1964) that 5s solving for a con-
sion, whose relevance was u n k n o w n (see junctive concept adopt a positive focus
Byers, 1967). The second criterion means sooner than 5s solving for a disjunctive
that if S selected a negative instance concept adopt a negative focus (as noted
which differed on 2 dimensions from a posi- above, it does not establish that any 5
tive focus, the choice was counted toward was in fact focusing).
focusing if one of the 2 differences had Focusing scores were higher following
been shown to be irrelevant by an earlier an initial positive instance for the uni-
stimulus selection. For each S, the per- dimensional concept, whereas the type of
centage of focusing choices was calculated initial instance made no difference with
for the 6 trials (or less, if criterion was bidimensional concept, F (1, 107) = 6.70,
reached earlier) following the first oppor- p < .01. This interaction must be care-
tunity to focus. Thus, focusing was mea- fully inspected. Only for the unidimen-
sured from Trial 1 for the unidimensional sional concept did the type of initial in-
concept (for either a positive or negative stance determine the type of instance from
initial instance), from the first positive which focusing was measured ; for the bi-
instance for conjunctive groups, and from dimensional concepts, it affected only the
the first negative instance for disjunctive trial on which the first focus instance was
groups (see Bruner, Goodnow, & Austin, encountered. The data therefore indicate
1956). In the conjunctive - negative ini- that a delay of a few trials in encountering
tial instance groups, the first positive in- a focus instance did not affect focusing
stance was encountered, on the average, scores (bidimensional), but that focusing
on Trial 4, while the average trial of the scores based on a positive instance of a
first negative instance was Trial 5 for unidimensional concept were higher than
the disjunctive - positive initial instance those based on a negative instance. It
groups. In presenting this analysis, no should be noted that while type of initial
claim is made that a high focusing score instance affected focusing score for the
indicates that S was in fact following a unidimensional concept, it did not affect
focusing strategy. It has been convinc- trials to criterion.
ingly argued that one cannot distinguish Several additional comments seem in
focusing from several other strategies on order. An estimate of "chance" focusing
the basis of card choices alone (Taplin & scorcs was obtained by assuming that the
Jeeves, 1972). Rather, the focusing score pool of possible stimulus selections con-
is viewed simply as one of a number of sisted of 63 equally likely alternatives after
possible indexes of the potential usefulness 5 encountered the focus instance, 62
of 5s' choices. equally likely alternatives after the first
choice, 51 after the second, and so on for
6 choices. The probability of a focusing
TABLE 4
choice at any step was equal to the pro-
PKKCKXTAGK OF "FOCUSING CHOICES" FOLLOWING portion of the remaining stimulus popula-
FIRST POSSIBLE Focus INSTAXCI;
tion which either belonged to the same
I inslanc category as the focus instance or which
Coiicfinual Ovt all
rule ; im- tn belonged to the opposite category- but
: Po.si live [ Negative j
differed from the focus on only one dimen-
Unidimensional ] 68 .5 ; 37 5 53 0 sion (note that this is more stringent than
Conjunctive 52 .3 i 53 2 52 8
Disjunctive 28 .8 i 24 2 , 26 5 the second criterion which was applied
to 5s' actual choices). This procedure
V K R B A I . I Z I N G HYPOTHESES IX CONCEPT I D E N T I F I C A T I O N 393
A l t h o u g h no data arc presently available tainment. Journal of Verbal Learning and Verbal
which demonstrate directly that MR 5s do Behavior, 1967, 6, 595-600.
process information more efficiently, this in- CHLEREK, J., & DOMINOWSKI, R. L. The effect
terpretation has been advocated. Bycrs and of practice on utilization of information from
positive and negative instances in identifying
Davidson (1967) proposed that HR 5s more
disjunctive concepts. Canadian Journal of Psy-
regularly consider the relevance or irrelevance chology, 1970, 24, 64-69.
of stimulus attributes. Williams (1971), al- CLAYTON, K. N., MERRYMAN, C. T., & LEONARD,
though making no direct comparison of HR T. B., HI. Rate of concept identification and
and control conditions, suggested that HR 5s the noticeability of the relevant dimension.
are more consistent in f o r m u l a t i n g and chang- Psychonomic Science, 1969, 15, 109-110.
ing hypotheses. The question remains as to CONANT, M. B., & TRAHASSO, T. Conjunctive and
how to obtain the needed information regard- disjunctive concept formation under equal-infor-
ing 5s' processing of information. The only mation conditions. Journal of Experimental
uncontaminating method would seem to be Psychology, 1964, 67, 250-255.
DENNY, J. P. The elimination of empiricist biases
some form of posttask interview, a method in experimental psychology as exemplified in the
which has systematically been avoided by most discovery of strategies of concept learning having
researchers but which seems worth trying, as deductive, concrete, and inductive components.
others have suggested (Denny, 1969; Taplin I n , Proceedings of the Nineteenth Congress of Psy-
& Jeeves, 1972). An alternative method, that chology. London: British Psychological Society,
of interrogating 5s during the task, can present 1971. (Abstract)
a dilemma to the researcher. The problem is DOMINOWSKI, R. L. Concept attainment as a func-
t h a t i n t e r r u p t i n g the task to ask 5, for example, tion of instance contiguity and number of irrel-
w h a t he learned f r o m a particular trial opera- evant dimensions. Journal of Experimental
Psychology, 1969, 82, 573-574.
tionally destroys the "control" condition. It DOWNING, B. Discriminability and preference in
is conceivable that any of a number of inter- concept identification. Psychonoinic Science,
ruptive techniques, like requiring hypothesiz- 1969, 14, 85-86.
ing or asking 5 what he learned from a trial, HAYGOOD, R. C., & BOURNE, L. E., Ju. Attributc-
might alter the manner in which i n f o r m a t i o n and rule-learning aspects of conceptual behavior.
is processed. Should this be the case, the evi- Psychological Review, 1965, 72, 175-195.
dence obtained through an interruptive tech- KARI'F, D., & LEVINE, M. Blank-trial probes
nique could not be generalized to the control and introtacts in human discrimination learning.
condition. It is possible that researchers Journal of Experimental Psychology, 1971, 90,
51-55.
might have to decide if they wish to study
LAUGHLIN, P. R. Selection versus reception con-
conceptual behavior under the simpler control cept-attainment paradigms as a function of
condition, using indirect methods, or whether memory, concept rule, and concept universe.
it would be more profitable to adopt a con- Journal of Educational Psvchology, 1969, 60,
siderably modified task in order to gain more 267-273.
direct information about, the behavior involved. MILLER, L. A. Hypothesis analysis of conjunctive
concept-learning situations. Psychological Review,
1971, 78, 262-271.
REFERENCES MYERS, J. L. Fundamentals of experimental design.
BOURNE, L. E., JR. Factors affecting strategies Boston: Allyn & Bacon, 1966.
used in problems of concept-formation. American N KISSER, I"., & \YliKNK, P. Hierarchies in concept
Journal of Psychology, 1963, 76, 229-238. attainment. Journal o! Experimental Psychology,
BOURNE, L. E., JR. Hypotheses and hypothesis 1962, 64, 640-645.
shifts in classification learning. Journal of Gen- SIEGKL, L. S. Concept attainment as a function
eral Psychology, 1965, 72, 251-261. of amount and form of information. Journal of
BOURNE, L. E., JR., EKSTUAND, B. R., & MONT- Experimental Psychology, 1969, 81, 464-468.
TAPLIN, J. E. Effect of initial instance on at-
GOMERY, B. Concept learning as a function of
the conceptual rule and the availability of posi- tribute identification of concepts using a selec-
tion procedure. Journal of Experimental Psy-
tive and negative instances. Journal of Experi-
mental Psychology, 1969, 82, 538-544. chology, 1971, 88, 177-181.
TAPLIN, J. E., & JEEVES, M. A. Strategies in
BUL'NER, J. S., GOODNO\V, J. J., & AUSTIN, G. A.
concept learning. (Program on Concept Learning
.4 study of thinking. New York: \Yiley, 1956. Report No. 19) Boulder: University of Colorado,
BYERS, J. L. A note on the calculation of stra- 1972.
tegies in concept attainment. American Educa- WILLIAMS, G. E. A model of memory in concept
tional Research Journal, 1967, 4, 361-366. learning. Cognitive Psychology, 1971, 2, 158-184.
BYERS, J. L., & DAVIDSON, R. E. The role of
hypothesizing in the facilitation of concept at- (Received December 4, 1972)