Bulletin of the PsychonomicSociety
1978,Vot.tt (3),197-199
Perceptionas hypothesistesting
TIMOTHY A. SALTHOUSE
University of Missouri, Columbia, Missouri 65201
and
WARREN L. DANZIGER
WashingtonUniversity, St. Louis, Missouri 63103
Three experiments were conducted to investigate the previously reported finding that the
accuracy of perceptual recognition decreases as the amount of experience with degraded
versions of a visual stimulus increases. Since this result is apparently predictable only from the
view of perception as an active process of hypothesis generation, this negative-effect-of-priorexperience phenomenon has important implications for theoretical conceptualizations of
perception. None of the current experiments yielded any evidence of less accurate perceptual
identification with increased number of incomplete versions of the stimuli when accuracy was
assessedwith a cumulative measure of identification accuracy, but tlvo of the experiments did
provide such evidence when accuracy was assessed with a conditional measure of accuracy.
Consideration of the complete pattern of results led to the conclusion that there is no real
evidence that perception is impaired because of early experience with ambiguous versions of
a subject.
An apparently well-documentedbut largely ignored
phenomenonwhich is of great importance to the theoretical conceptualizationof the perceptualprocessis that
prior impoverished experience with a visual stimulus
impairs the subsequentidentification of that stimulus.
Independentstudies(i.e.,Bruner& Potter, 1964;Potter,
"if
1966; Wyatt & Campbell,l95l) havereportedthat
a subject is initially exposedto a blurred image that he
cannot recogrize, subsequentrecognition of the image
in clearer form is substantially delayed" (Bruner &
Potter, 1964. p. 424).
Although certainly not a detailed explanation, an
interpretation of this phenomenoncan be derived from
the conceptualizationof perceptionas an activeprocess
involving the generation, testin!, and verification of
hypotheses.PerhapsGregory (1970, 1974) is currently
the strongestproponent of this view, but the notion that
perception is somethingmore than the registeringand
catalogingof sensationsand that the perceiveris not
merely a passiverecipient of environmentalinformation
is at least as old as experimental psycholory (e.9.,
Boring, 1942), and has been reflected in many perceptual theories(e.g.,Allport, 1955).
One assumption of this conceptualizationis that
perception involves testing hypotheses,and that once
a hypothesis is establishedit may prevent or delay the
acceptanceof an alternativehypothesis.This interpretation thus attributes the negativeeffect of prior experThis research
wassupportedby a grantto the seniorauthor
from the Research
Councilof the Graduate
School,University
of Missouri,Columbia,and by NIA GrantAG40008,administered by J. Botwinick, WashingtonUniversity,St. Louis,
Missowi63103.
ienceto the formation of incorrect hypothesesabout the
identity of the stimulus that, once established,become
resistant to modification. Identification is therefore
impaired as a result of prior experiencebecausethe subject is seekingto confirm or disconfirm an inappropriate
hypothesis; without this prior experience,the subject
is not encumberedby an inappropriatehypothesisand
he need only evaluatethe currently availableinformation without any predispositions.
Such an interpretation of this phenomenon is far
from satisfactory since it provides no explanation of
how hypothesesare generatedin the first place,nor any
description of the mechanismof hypothesisgeneration
and verification. Nonetheless,it appearsthat only an
interpretation of this type can account, even if only
superficially, for the negative-effect-of-prior-experience
phenomenon.The existenceof this phenomenon,therefore, can be taken as evidence in support of the
conceptualizationof perception as a processof hypothesistesting.
Before fully accepting the phenomenon and its
important theoretical implications, it is desirable to
obtain assurancethat the result is reliable and not an
artifact of some proceduralor methodologicalpeculiarity. For this purpose, the current experiments were
designedand conducted.
In all three of the earlierstudieson this phenomenon,
the stimulusmaterialswere photographsor photographic
"an aerial
slidesof complex scenes(e.9.,
view of a street
light on the campusat night"). The procedureconsisted
of initially presentingthe photograph or slide in a very
blurred version and then gradually reducing the blur
to a point at which the subjectwas askedto identify the
197
I98
SALTHOUSE
AND DANZIGER
stimulus. The present experimentsutilized simple line alternative.The assignmentof correct alternativeto stimulus
drawingsof common objectsand animalsas the stimulus figure was balancedacrosssubjects.
The task in all experimentswas self-pacedand the subjects
materials, and manipulated the impoverishmentof the
were allowed unlimited time to view the drawingsand write
stimuli by spatial masking of different percentagesof
their responses.
A responsewas requiredto eachstimulusfigure
the stimulus. If the negative-effect-of-prior-experienceevenif it was only a guess.
phenomenonis to be confirmed, one would expect the
accuracy of identifying these incomplete stimuli to be
RESULTS
inversely related to the amount of prior experience
(i.e., the number of previousincompleteversions)with
The data from each experiment were initially analyzthe stimuli.
ed by determining the percentageof figures correctly
identified on Version 7 (17.8% compldteness)for each
METHOD
subject and submitting thesepercentagesto an analysis
of variance. The analyseswere not significant for the
Subjects
data of ExperimentI [F(2,33) < 1.0] or Experiment2
participants
in
There were 36, 24, and 36 college-student
IF(2,46\ = 1.83, p > .151, but the analysisin ExperiExperiments| ,2, and 3, respectively.
ment 3 was significant[F(2,70) = 4.27, p ( .05]. The
Stimuli
mean percentagesare displayedin the top three rows of
Forty line drawings of common objects and animals similar
Table l: it can be seen that the differencesare in the
to those found in children's picture-word flash cards were used
as stimuli. For example,the first five stimuli were a key, a pipe, direction of the lowest amount of experienceproducing
the lowest level of accuracy,the opposite of the result
a saw, a bell, and a sailboat. All drawings occupied an area of
approximately4 x 6 in., centeredin a sheetof 8/z x l1 in. white
reported by earlierinvestigators.
paperin a looseleafnotebook.The impoverishedversionsof the
A second analysis of the data from Experiments2
drawings were produced by overlaying a mask consisting of
and
3 involved comparisonsof conditional percent
horizontal and vertical strips of paper on top of the figure and
values. These values were computed for each
correct
photocopying
then
the figure through the mask.Variouscombinationsof horizontaland verticalstrip widths from .125 to I in.
subject at each incomplete version number by dividing
were used to createdifferent degreesof impoverishment.The
the number of stimulusfiguresidentified on that version
percentageof the total paperareathat was not maskedby the
the number of figures not previously identified in
by
from
4.4%
with
values
of
strips ranged
to 25.lVo,
intermediate
versions.
earlier
6.3%,8.8%,10.1%,12.5%,15.6%,17.8%,and20.9Vo.
All masks
The conditional percent correct data from Versions7
were independentlyconstructedand applied; hence,although
each allowed increasingamounts of the total area of the figure
through 9 were analyzedin an analysisof variance,with
to be transmitted,the particular stimulus regionstransmitted
the three incomplete versions and the three levels of
were not necessarilythe sameas the percentagearea increased.
The incompleteversionswere always presentedin consecu- experienceas factors in each experiment. The level-ofexperience factor [Experiment 2, F(2,184) = 15.27,
tively increasing order, with all of the versions of one figure
presentedbefore any of the versionsof another figure were p ( . 0 0 0 1 ; E x p e r i m e n3t , F ( 2 , 2 8 0 )= 6 . 1 0 ,p < . 0 0 5 1 ,
presented.
the incompleteversionfactor [Experiment 2,F(2,184) =
11.40, p < .0001; Experiment 3, F(2,280) = 4.93,
Procedure
p < .01] , and the interaction of the two factors [ExperiExperiment l. Subjects were presented with one, four, or
ment 2, F(4,184) = 5.43, p ( .0005; Experiment3,
seven incomplete versions of the stimulus figures and were
asked to identify the figure after viewing the last incomplete
F(4,280) = 7 .52, p < .0001I were all statisticallysignifiversion. The incomplete versions presented were Version 7
cant. Separateanalyseson the data from each incom(178% completeness)for the subjectswith only one incomplete
plete version revealedthat only on Version 7 was the
version,Versions4 (10.1%completeness)
through7 for the subfactor statisticallysignificant IExperjects with four incompleteversions,and Versions| (4.47ocom- level-of-experience
pleteness) through 7 for the subjects with seven incomplete
iment 2, F(2,46) = 41.30, p < .0001; Experiment3,
versions.Thus, all subjectswere required to identify the stimulus
F(2,70) = 29.55, p ( .00011. The mean conditional
irgure after viewing the sameincompleteversion (i.e., Version 7),
percent correct valuesfor Version 7 are displayedin the
but they differed in the number of previousincomplete versions
bottom two rows of Table l
of the figurethey had seen.
Experiment 2. Subjectswere required to write an identification responseto every incomplete version in order to monitor
conditional probabilities of correct identification. Incomplete
Versions 8 (20.9% completeness)and 9 (25.O7ocompleteness)
and the complete version of the figure were presented after
Version 7 to provide further assessments
of conditional identification accuracy.
Experiment 3. Subjectswere provided with a verbal response
alternative for each figure and instructed to indicate whether
that response was corect (YES) or incorrect (NO) on each
incompleteversion.The incorrect alternativeswere selectedfrom
the most frequent wrong responsesto the figures in Experiment 2. Each subject receivedone-halfof the figureswith the
correct responsealternative and one-half with the incorrect
cumulativeandconditional [11ti,rl*r,"r
Amount of Prior Experience
-
E^Peii-
Cumulative
PercentCorrect
-
ment
High
Moderate
I
2
68.3
64;l
82.0
67.1
66.0
83.1
26.8
32.3
J
Conditional
PercentCorrect
nccuracyon version 7
2
J
I1.4
29.3
Low
64.0
54.2
7s.6
54.2
75 . 4
PERCEPTION
AND HYPOTHESES
DISCUSSION
I99
specific conclusion, however, the phenomenon that greater
amounts of experience with an ambiguous stimulus impair
subsequent identification of the stimulus can no tonger be interpreted as supporting the active hypothesis-testing view of perception. Of course, this does not mean that active theodes of
perception are not viable, but only that this particular phenomenon cannot be considered as evidence for that position.
None of the three experimentsreported in this project provided any indication that the cumulativeaccuracyof identifying
a degradedvisualstimuluswasinverselyrelatedto the amountof
experiencewith more severelydegradedversionsof the stimulus.
However,Experiments2 and 3 indicatedthat subjectswith no
prior experiencewith incompleteversionsof the stimuli were
REFERENCES
more accuratewith the measureof conditionalpercentcorrect
than subjectswith prior experience.Sincethe advantageof no
prior experienceholds only for the first exposureof an incom- Ar,rronr, F. H. Theories of perception and the concept of
structure. New York: Wiley, 1955.
plete versionof the stimulus,and is not apparentin the cumulative percent correct measureof accuracy,a simpleinterpreta- Bonrxc, E. G. Sensation and perception in the history of experimmtal psychology. New York: Appleton-Century-Crofts, 1942.
tion of this finding seemspossible:The subjectswith prior
Bnuven, J. S., & Porrrn, M. C. Interference in visual recognition.
experiencewith the stimuli identified some of the stimuli on
Science. 1964. 144. 424-425.
incompleteversionsprior to the critical version,so the condiGnroony, R. L. The intelligent sye. New York: McGraw-Hill,
tional percent correct measurefor those subjectsincludesa
1970.
higher proportion of difficult-to-identify stimuli than that for
the subjectsfor whom the critical versionwasthe first exposure Gnrcony, R. L. Choosing a paradigm for perception. In E. C.
Carterette & M. P. Friedman (Eds.), Handbook of perception:
to the stimulus.The important point is that the total numberof
Historical and philosophical roots of perception (Vol. 1).
figuresidentified on or beforethe criticalversionis the samefor
New York: Academic Press, 1974.
all subjectsregardless
of the amount of prior experiencewith the
Porrrn, M. C. On perceptual recognition. In J. S. Bruner,
stimuli.
R. R. Olver, & P. M. Greenfield (Eds.), Jtzdies in cognitive
The experimentsthereforeleadto one or more of the following conclusions:( I ) that the negative+ffect-of-prior-experience gr?wt,. New York: Wiley, 1966.
Wylrr, D. F., & ClMpsprI,, D. T. On the liability of stereotype
phenomenonis unreliableand cannot be replicated;(2) that the
or hlpothesis..Iournal of Abnormal and Social Prychology, 1951,
phenomenonis specificto a particulartype of stimuli and does
46, 496-500.
not have general applicability; (3) that the phenomenonis
demonstrableonly with the conditional measureof accuracy
which leads to artifactual differenceswhen there are no differencesin the cumulativemeasureof accuracy.Regardless
(Receivedfor publication Dectmber 27,1977.)
of the