WWWW“HIMIll/WWW!WHIUIWIWI

 

103
361
THS

 
 

    

// {Moos/MM

L 82 165! /

in WWW/III

//

9

LIBRARY
Michigan State

/
University

 

This is to certify that the
thesis entitled
ON THE INDEPENDENCE OF LOCATION AND IDENTITY
INFORMATION IN ICONIC MEMORY
presented by

Joseph Spencer Bmwn

has been accepted towards fulfillment
of the requirements for

M.S. degree in Psychologx

Mew

Major professor

Date 5/ ('7 3

0-7639 MS U is an Affirmative Action/Equal Opportunity Institution

 

 

 

IVIESI_J RETURNING MATERIALS:
Place in book drop to
LJBRARJES remove this checkout from
my... your record. FINES will

 

 

be charged if book is
returned after the date
stamped below.

 

 

 

 

 

 

ON THE INDEPENDENCE 0F LOCATION AND IDENTITY
INFORMATION IN ICONIC MEMORY

BY

Joseph Spencer Brown

A THESIS

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

MASTER OF SCIENCE

Department of Psychology

1987

ABSTRACT
ON THE INDEPENDENCE or LOCATION AND IDENTITY
INFORMATION IN ICONIC MEMORY
BY

Joseph Spencer Brown

Most contemporary models Of iconic memory assume that location and
identity information are coded separately rather than together. This
assumption is questionable given the small number Of contradictory studies
upon which it is based. Three experiments were conducted to examine this
contention. Experiments one and two used a recognition test to show that
location and identity information are stored as an integral unit. It was
suggested that the recognition memory test that was used is more sensitive
than previously employed measures to the actual contents of the icon. In
experiment three. pseudo—letter stimuli were used to investigate the nature
Of the representation of identity and location. It was found that the
representation more closely resembles an image than an abstract store.

Implications for models Of iconic memory are discussed.

ii

To my mother and father.

with all my love and gratitude

ACKNOWLEDGMENTS

I would like to express my grateful appreciation to the members of my
thesis com mitee: David Irwin, Thomas Carr. Rose lacks, and J. Kathryn Bock
for their advice and assistance. Also. my thanks to Beth Dettmer-Radtke for
assistance with data collection, and Jim Yeomans for designing the
pseudoletter stimuli used in experiment 3. Finally. special thanks to the
community Of graduate students in the experimental interest group at
Michigan State for the supportive environment so necessary for the
undertaking of projects Of this sort. This research was supported by the

National Science Foundation Grant BN5 85-19580 to D. E. Irwin.

iv

TABLE OF CONTENTS
Page number
List of Figures ......................... vi
lntrod uction .......................... l

A Test Of Identity and Location Information in Iconic

Memory ............................. 17
Experiment 1 ......................... l7
Experiment 2 ......................... 25
Experiment 3 ......................... 30
General Discussion ...................... 38
List of References ...................... 41

LIST OF FIGURES

Figure

l. Chow's predictions based on fast decay Of location

inf or mation .............................

2. Characterization Of Dixon's data .................

3. Experiment 1: Percent accuracy for arrow and no-arrow

trials ................................

4. Experiment 1; Percent accuracy for false present and

false absent trials .........................

5. Experiment 2: Percent accuracy for true. false present,

and false absent trials .......................

6. Pseudoletter stimuli used in Experiment 3 ...........

7. Ex riment 3: Percent accuracy for letter and

pseu oletter stimuli ........................

8. Experiment 3: Percent accuracy for false present and

false absent letter and pseudoletter stimuli ...........

vi

Page

10
15

21
24

27
32

34

36

INTRODUCTION

The systematic study Of iconic memory began with the work Of George
Sperling (1960). In these experiments, Sperling introduced the partial
report technique in hopes of examining the introspection of subjects in visual
experiments. that they saw more than they could report (Bridgin. 1933:
Wilcocks. 1925. reported in Sperling. 1960). The logic of these experiments
was that if subjects were asked to report only part of a stimulus array. but
didn‘t know what part they would be asked to report until the array was no
longer available. then an estimate of the number of items actually available
for report could be Obtained by multiplying the percent of the letters
correctly reported by the number of letters in the array.

To begin his examination of the phenomenon. Sperling first determined
the number of letters that subjects could report from an array if asked to
report the whole array. At exposure durations of up to 500 msecs. subjects
were able to report roughly 4.5 items (though there was considerable
individual variation). He then compared this to the number of items that
were "available“ for report. that is. the proportion of items that subjects
could report if only asked to report three or four items. Subjects had
considerably more items available than could be reported. for up to 500
msecs after stimulus offset. Further. Sperling found that the availability Of
these letters was disrupted by the presentation Of a bright post exposure
field. The final experiments of this seminal work demonstrated the
precategorical nature of the phenomenon that would later become known as
iconic memory. This was done by asking subjects to report either all the
letters or all the numbers in an array containing both. Subjects were unable
to report items based on the category of the item at levels higher than whole

1

report performance.

These experiments led Sperling to believe that there exists an image after
stimulus Off set that allows subjects to act on the image as if the array were
still present. This conclusion. along with the others derived from these
experiments. led to the formulation of what has been referred to as the
“classic" definition of iconic memory (Irwin and Yeomans.1986a: Yeomans
and Irwin. l 985). that is. as a precategorical. maskable. short-lived. visible
image.

Consistent with the notion that the icon is visible. Sperling. in the
discussion of these experiments. implied that experiments that measure
iconic memory by means Of estimating the subjective duration of flashes of
light are comparable to the partial report paradigm.

Following this logic. Sperling (1967) ran an experiment in which he asked
subjects to adjust an auditory "click" to correspond to the onset or Off set Of
visually presented stimuli. Though he reports little Of his data. this paradigm
was adopted by Ralph Haber and Lionel Standing (1970) to examine the
subjective duration of visual stimuli. In this experiment. subjects were asked
to adjust "clicks“ to correspond to the onset or offset of stimuli as the
duration of the stimuli and the type of pre and post exposure field were
varied. Defining visual persistence as the difference between the interval
between the two “clicks" and the actual exposure duration of the stimulus.
Haber and Standing found that visible persistence decreased as exposure
duration increased. Further. visible persistence could be ended with a noise
mask or a very bright post exposure field. Using essentially the same
procedure as that used by Haber and Standing. Robert Efron (1970)
determined that. in addition to being an inverse function of stimulus

duration. the duration of visible persistence is an inverse function of the

3

intensity of the stimulus when intensity was under 3.4 ft/lumens. Thus. by
means of similar experiments ( Bowen. Pola. and Matin. 1974 ). and more
Objective experiments utilizing integration Of pairs of stimuli (DiLOllO.
1978,1980; Eriksen and Collins, 1967) the consistent view of iconic storage
began to take shape.

Since virtually no finding in psychology goes unchallenged. it is not
surprising that the inverse relation between intensity and visible
persistence is not universally accepted. For instance. Long and Beaton
(1980,1982) found positive effects of stimulus intensity on the duration Of
visual persistence using techniques similar to those of Haber and Standing.
On closer examination. it becomes apparent that Long and Beaton's results
were caused by the very high intensities that they used. ranging from 1 to
30 f t/lu mens. It is not unreasonable to assume that these high intensities
lead to afterimages and thus are not measuring the same effects as other
investigators. Given this criticism. the traditional view of iconic memory
seems relatively safe from attack from this quarter (for a more detailed
review of Long’s work. see Irwin and Yeomans .l986b).

Unfortunately for the classical model of iconic memory. it is almost
certain that the assumption that the same underlying visual store is
measured by both tests of visual persistence and the partial report task
measure is not correct. This can be seen by contrasting the effects of
duration and intensity on the estimation of visual persistence with their
effects on performance on the partial report task. Adelson and Jonides
(1980) examined the effects of intensity on performance on a partial report
task. Using intensities ranging from 8.7 mL to 70 mL. they found that though
the overall level of accuracy could be affected by stimulus intensity. the
decay function of the partial report task was the same at all levels of

4

illumination. Further. the effect Of stimulus intensity on the overall level of
performance was to make performance worse as intensity decreased . Thus.
if variations of intensity affect the partial report task at all. it is not by
making the information last longer at lower lu minances. as might be
expected from the duration judgments Of the experiments discussed earlier.
but instead. by making the information less clear initially.

In addition to the discrepancy between the effects of luminance in partial
report tasks and duration judgments. the effect of stimulus duration in
partial report tasks and duration judgments indicates that the two measures
are. in fact measuring different underlying perceptual processes. Recall that
perceived duration measures find a negative effect of stimulus duration.
however. in a partial report task. either no effect (Sperling 1960) or a
positive effect is found. The best evidence for this contention is found in the
work of David Irwin and Jim Yeomans (Yeomans and Irwin. 1985: Irwin and
Yeomans.l986; Irwin and Brown. in press). For example. in a partial report
task in which the stimuli were three by three arrays of letters. and the
subject's task was to report a single row indicated by a probe. stimulus
durations ranging from 50 to 500 msecs had a positive effect on partial
report performance. This adds more support to the argument that iconic
memory is something more than simply a visible trace of the original
stimulus.

Max Coltheart (1980). upon reviewing this evidence and much more. in
what he referred to as an exhausting review. concluded that there are at
least three logically independent elements of what was being studied under
the rubric of iconic memory. The first of these elements is the neural
persistence in the visual pathway after stimulus offset. The second element
is visible persistence. defined by the subjective experience of the subject

5

that the stimulus is still present after Off set. Finally. there is informational
persistence. which is the availability of visual information after stimulus
offset. This is logically independent of visible persistence. since you need not
actually see the stimulus to remember what it looked like.

In addition to the logical independence of these three elements. there is
substantial evidence that at least the last two are substantially independent
in an empirical sense. That is. as described above. there are different effects
of duration and intensity on the two types of phenomena. The partial report
task only requires knowledge of the visual characteristics of the array. since
a location probe is used to cue report. and this task finds positive effects of
duration and intensity. while the duration judgment measures the visible
duration of the stimulus and is affected negatively by increasing duration
and intensity of the stimulus. Based on the original definition Of iconic
memory by Neisser (1967). Coltheart concludes that it is the measurement of
informational persistence as measured by partial report tasks that is the
true domain of the student of iconic memory (though this in no way should
be taken to denigrate the study of visible persistence). Accordingly. in this
discussion. I will use the term iconic memory to describe Coltheart's
informational persistence. I

Goltheart goes on to argue that. in addition to being non-visible. iconic
memory does not behave as you would expect an image to behave. That is.
instead of decaying slowly away like a photographic negative exposed to
light. it appears that different visual aspects of the information found in the
iconic store decay at different rates. Specifically. position information seems
to decay more rapidly than identity information. In support of this
contention. he cites work by Townsend (1973) (see also Dick.l969;

DiLollo. l 978 ).

6

Townsend performed a series Of experiments that explored the relation of
location and identity in the icon. The first Of these experiments. a variation
of the standard partial report paradigm. included an analysis of errors of
intrusion. that is. errors in which the subject responded by naming a letter
not contained in the stimulus array. Townsend pointed out that these errors
occurred at much lower frequencies than might be expected if iconic
memory were simply a decaying image. Though she doesn‘t point it out
directly. it is apparent that the remaining errors. those in which the letter
named by the subject was in the array. but was not in the position probed.
occurred at higher levels than predicted by a decaying image. This suggested
to Townsend that location might be stored separately from identity. and that
location information decays more quickly.

In a second experiment. Townsend asked subjects to indicate whether or
not a probe letter had appeared in the stimulus array. In this task. there
was no effect of cue delay. Townsend suggested that in tasks of this kind
subjects needed only identity inf ormation. and therefore no delay effect was
found because identity information did not decay in iconic memory. She
concluded that the decline in partial report performance was the result of
loss of spatial information about the elements Of the stimulus array.

While Townsend's results have been replicated by a number of people
(Yeomans and Irwin. 1985; Irwin and Yeomans.l986; Irwin and Brown. in
press; Mewhort. Campbell. Marchetti and Campbell. 1981; Mewhort.
Marchetti. Gurnsey and Campbell. 1984 ). there are at least two problems
that suggest caution. First. there is at least one alternative explanation for
the results of the first experiment; it is possible that when subjects did not
know what item had appeared at the probed location. they simply chose an

item at random from the set of items that they could remember. Such a

7

stategy would result in more location errors than intrusion errors. This is a
quite reasonable strategy given that there are far more possible items that
are not contained in the array than are contained in the array. It might seem
to the subject that the best chance of responding correctly is to assume that
they had mislocated an item. and therefore they should guess from the
subset Of letters that they were sure were in the array. This criticism might
seem quite weak were there converging evidence supporting Townsend's
conclusion. however. as we shall see. the literature is far from clear on this
point.

The second Of Townsend's experiments is also flawed. In this experiment.
in which subjects were asked to indicate whether a probe letter was in the
stimulus array. it seems clear that there should be no partial report
advantage and therefore no decline of accuracy with cue delay. regardless of
the nature of the icon. If the partial report advantage occurs because of
preferential processing of parts of the icon. then we would expect no decay
over time. because each item of the array would have to be processed in
order to compare it with the probe. Therefore. the response of the subject
would be based on the information that could be processed during the life Of
the icon. just like full report. Thus. the lack of decline in performance with
increasing cue delays only shows that if the subject‘s response requires the
same type of information as full report. it. like full report. does nOt decay
with time.

In addition to Townsend's investigation. there have been only a few
systematic attempts to investigate the decay of identity and location
specifically in iconic memory. The first of these. Ken Den Heyer (1972).
draws the conclusion that location and identity are coded together. and

decay at the same rate. To do this. den Heyer employed Garner and Morton's

B

(1969) procedure for testing perceptual independence. Garner and Morton's
procedure requires three tasks: processing both identity and location
simultaneously; processing identity or location alone in the presence of both
types of inf ormation; processing identity or location information without the
presence of the other type of information. To accomplish these conditions.
den Heyer used three different type of stimuli: the first consisted of three
2x4 matrices with letters occupying three of the eight positions in each
array; the second type consisted of three 2x4 matrices with a circle placed at
three of the eight possible positions in each matrix; the third type consisted
of a 3x3 matrix filled with lettters. Each type of stimulus was briefly
displayed and followed at various delays by a probe indicating which of the
three matrices to report (type one and two) or which column of letters to
report (type three) . The type of report varied: for the first type of stimuli. in
which both identity and location information were present. subjects were
asked to report: 1) only the identity of the three letters in the cued matrix:
2) only the locations of the three letters; or 3) both the location and identity
of the three letters. With stimuli of the second type. in which only location
information was present. subjects were asked to report the locations of the
three 0's. With the third type of stimuli. in which only identity was supposed
to be present, subjects were asked to name the three letters in the column
indicated by the probe in any order. Using this procedure. den Heyer
concluded that identity and location were coded together and decayed at the
same rate. He did. however. allow that the decay of location information
might have a greater effect on performance than decay of identity.

Unfortunately. there is at least one major flaw in den Heyer's study. To
see this. consider the type of location information that is required in the

conditions in which only location information is to be reported. In this case.

9

rather than reporting location information about specific letters. subjects are
asked to report what is essentially the pattern Of the array. This is very
different than the kind of location information examined by Townsend
(1973). To see this we need only examine the work Of Phillips (1974) who
found that subjects can make same/different judgments about dot patterns
at levels above chance for delays of up to nine seconds. long after the icon
has decayed away. Therefore. it seems apparent that den Heyer's "location
only“ conditions are very different from locating a specific item within an
array.

A much more recent study Of the decay of location and identity in the
icon by Chow (1986) manages to further muddy the water by drawing the
only conclusion that could possibly disagree with both Of the previous
studies. That is. Chow concluded that. as Townsend contended. location and
identity are stored separately. but that identity decays mare quickly than
location information.

In Chow's experiments. he attempted to take advantage of the predictions
independence of identity and location make about the conditional probability
Of correct localization given correct identification and of correct identification
given correct localization. To do this. he took the strong position that the
decline in partial report performance is almost exclusively the result of the
loss of location information from iconic memory ( as do Mewhort. Campbell.
Marchetti and Campell;l98 l ). If this assumption is made. the conditional
probability of correct identification of an item of the array. given correct
localization of the item (P(I/L)). should be constant with increasing 181.
since identity information is not decaying as location information decays.
Further. the conditional probability of correct localization given correct
identification (ML! 1)) should decline with 151.

)0

 

1.07, 1.0}
l :l
i: :‘r‘

 

 

 

 

Figure 1. Chow's predictions based on fast decay Of location information.

Chow‘s task consisted Of presenting subjects with a seven letter array
followed at various intervals by a probe indicating what position in the array
the subject should report. The subject was asked to name the item indicated
I by the probe and its position. Chow found that the probability Of reporting
the location correctly. given the correct reporting of identity (PlL/ 1)) was
constant across 181. while the probability of reporting the correct identity
given correct reporting Of location (Pll/L)) declined with 151. This is in direct
opposition to the predictions based on the rapid and independent decay Of
location information. Chow concludes that the decline in partial report
performance with increasing cue delay is caused by the decay Of identity
information rather than location information.

As in the studies described above. Chow‘s conclusions are subject to
criticisms that suggest great caution in accepting this as the definitive
answer to the independence issue. The main criticism is that while Chow
assumed that the location judgments Of subjects were a direct assessment of
their knowledge about the location of the letters in the array. they were in
fact only a measure Of their ability tO localize the probe. That is. the subject
Obviously believes the letter they are reporting occupied the position

11

indicated by the probe. theref ore. all they need do is report the letter's
identity and then the position of the probe. Given this interpretation. it is not
surprising that P(L/ I) did not decline with 181. since the probe was
presented at the end of each 151. It is not surprising that. given a familiar
display, subjects could tell you the position of the probe no matter how long
the delay. Further. to correctly identify the letter in the array. you must
either have correctly localized the probe. or have mislocalized both the probe
and the letter. Since the mislocation of both probe and letter to the same
position should be quite rare. P(L/ I) must be quite high. since to correctly
identify the letter you must know the probe's position. In the same way. it is
not surprising that P(I/L) declined with 181. This decline could be the result
Of either a decline in identity information as Chow concludes. or a decline in
information about the location of the elements of the array that is
independent of ability to localize the probe. If you know the location of the
probe. but the positions of array items have decayed. you will not be able to
correctly name the item indicated by the probe. thus the decline.

It is obvious that the nature of the storage of identity and location in
iconic memory is far from clear. It is possible that this confusion exists
because the recall measures. such as partial report. which are the primary
source of data in these studies. are simply not suited for the task at hand. A
potential problem with these procedures is that instructions to report the
letter indicated by the probe may encourage subjects to devote more
attention to retaining the identity of the letters at the expense of their
location. Further. given that subjects undoubtedly have much more
experience with overlearned letter identity codes than with the location
codes associated with a particular experimental stimulus configuration. it

may not be too surprising that identity information of ten appears to be more

I2

durable than location information. Thus. the contradictory evidence found
above may be the result of encoding. retrieval or response biases which
differ across experiments. rather than differences in the amount of
information actually contained in iconic memory.

Even in studies especially designed to examine naming and locating in
briefly presented displays. the independence of location and identity
information is far from clear. For instance. Krumhansl and Thomas (1976)
found that accuracy for identifying a single letter was independent of
accuracy for locating that letter in absolute space. but the same investigators
found that identification and localization were not independent when
multiple letters were used (Krumhansl.l977. Krumhansl & Thomas. 1976;
1977). It is important to note that since these investigators were not
interested specifically in iconic memory. they presented a mask imediately
after stimulus offset and asked subjects to delay responses for 500-1000
msecs. thus it is quite possible that relatively little can be learned about
iconic memory. which decays considerably in the first 500 msecs and is
severely disrupted by masking in the first 150 to 200 msecs after offset
(Irwin and Yeomans. l986a). Logan (1975). in a visual search paradigm.
found that the presence of distractors interfered with locating a target. but
had no effect on naming it. suggesting that identification and localization are
independent. Here too. however. iconic memory was not specifically at issue.
so masks were presented immediately after stimulus Off set Thus. at best
this information bears on the extraction of information from the icon before
its destruction. rather than the nature of information storage within the icon
itself.

Given the vexed nature of the field. it is somewhat surprising that all

three of the most explicit models of iconic memory now in use all rely to

l3

varying degrees on the assumption that location information decays more
quickly than identity information in iconic memory. Before describing our
attempts to disentangle the problems associated with previous experiments.
let‘s first examine these theories and their reliance on this shaky
assumption.

The first of these theories is the dual buffer model of Mewhort. Campbell.
Marchetti and Campbell (1981; Mewhort. Marchetti. Gurnsey and Campbell.
1984 ). This theory argues that iconic memory consists of two distinct stages.
the first being a precategorical. duration dependent store of feature level
information about the stimulus: the second stage is a character buffer that
contains the identified elements of the array and their relative spatial
positions. In the second store. location information is more fragile than
identity information. that is. it decays faster and is more vulnerable to
masking. Thus. Mewhort et al argue that masking affects the two stores quite
differently. The disruption caused by masking in the first store results in the
disruption of feature level information. and thus errors in identification. the
identification errors discussed earlier. Masking during the duration of the
second buffer results only in errors of location. Among other things. this
distinction between feature and character buffers allows Mewhort et al to
account for the high number of location errors relative to intrusion errors
found in virtually all partial report tasks by arguing that these errors are the
result of decay of location information in the character buffer. Thus. the dual
buffer model "...makes a strong distinction between spatial and identity
information...“ (Mewhort et al. 1981. p. 52) .

The second major model of iconic memory which makes the assumption
of independence of location and identity infOrmation was proposed by Irwin

and Yeomans (1986a; Yeomans and Irwin. 1985). Like Mewhort et a1. they

I4

propose that iconic memory consists of two stages. The first stage consists Of
a visual storage which contains a visual analog of the stimulus and lasts 150
to 300 msecs after stimulus offset. independent of stimulus duration. The
visual store is vulnerable to masking. The second stage of the model is an
abstract coding of the identity of the elements of the array. paired with an
imprecise and/or quickly decaying coding of location information. This store
is held to be invulnerable to masking. As you can see. this model bears a
great deal of resemblance to Mewhort's model. especially in regards to the
treatment of identity and location information. The main differences lie in
predictions about the effects of stimulus duration on the duration of the first
stage of processing and the predictions about the second stage‘s vulnerability
to masking.

The final model of iconic memory that should be included was recently
proposed by Peter Dixon (1986). This model. also a two stage model. was
proposed to account for an unusual configuration of data found in a
verification task in which subjects were asked if a probed letter was in a
stimulus array. Very different results were found on present trials (in
which the probe was in the array ) than absent trials (the probe was not in
the array). Trials with short positive SOA's showed a decrement in
performance in the absent trials. but not in the present trials. relative to

negative and long positive SOA's. as shown in figure 2.

IS
CHARACTERIZATION OF DIXON DATA

— Present trials
— Absent trials

ACCURACY

-600 o 500
30A ’

Figure 2. Characterization of Dixon's data.

Though Dixon also proposed a dual buffer model. this model differed
substantially from the dual buffers proposed by Irwin and Yeomans and
Mewhort et al in that it contends that location information is lost in the first
buffer. That is. the first buffer consists of quickly decaying identity codes for
the items in the stimulus. which are processed in parallel. and even more
rapidly decaying location codes. The second buffer consists of a more
durable storage of stimulus identity with some location information. The
translation from the first buffer to the second is contended to be effortful
and time consuming. rather than the automatic process proposed by the
other two models.

This model explains Dixon's data by contending that. at short positive
SOA's. little or no information has been transfered from the first buffer to
the second before the probe appears for processing. Therefore. the
comparison is made between the durable second buffer representation of
the probe and the quickly decaying codes in the first buffer which includes

the original representation of the probe. Since location information has

16

decayed a great deal by this time. there is no way to tell whether the first
stage representation of the probe is the probe or a stimulus item. Therefore.
there is a high rate Of false positives. and thus low accuracy in the absent
conditions. If there is actually a dissociation of identity and location. as this
model assumes. this is indeed an adequate explanation of the data it intends
to explain. If. however. there is no such dissociation. then an alternative
model must be found.

As you can see. the assumption that identity and location are stored
separately with location information decaying more quickly is prevalent in
contemporary accounts of iconic memory. Therefore. it seems reasonable to

test this assumption in a set of experiments designed expressly for this

purpose.

A Test of Independence of Identity and Location Information in

Iconic Memory

The experiments described below were designed to investigate the coding
of location and identity information in a new way. using a recognition
memory procedure instead of a recall procedure. Recognition tests of
memory are generally regarded as more sensitive (e.g.. Baddeley. 1976). and
thus perhaps less subject to retrieval biases. If we find identity and location
to be independently coded. we will add support to the central assumption of
the several new models of iconic memory mentioned above. If. however. we
find that location and identity are not coded separately. these models will
have to be discarded or modified to a greater or lesser extent.

Experiment 1

In Experiment 1 we used a recognition memory procedure to investigate
performance in two conditions which differentially emphasized location and
identity coding. In one condition (the no war condition). subjects were
presented a circular array of 8 letters. This array was followed by a single
probe letter presented in the center of the array. the subjects‘ task in this
condition was to say whether the probe letter had appeared in the array.
This condition thus required subjects to respond on the basis of identity
information. with no regard for location information. The second condition
was quite diff erent. however. In this condition (the arrawcondition). the
probe consisted of a letter presented in the center of the array. along with a
line which pointed to one of the array locations; the subjects' task in this
condition was to say whether the probe letter had occupied the position
pointed to by the line. Thus. this condition forced subjects to respond on the
basis of both location and identity information. In trials in which the probe

17

18

letter did not appear at the probed location. we examined what effect having
the probe letter at another. unprobed location. had on performance. If
identity and location information are stored separately. with location
information decaying more rapidly. we would expect that the presence of the
probe letter at an unprobed position would result in many incorrect "yes“
responses as probe delay increased. because location information will have
decayed and the probe's identity code will match the identity code of one of
the elements in the array. An additional manipulation in both conditions
was the presentation of a masking stimulus immediately after array Off set
on half the trials. If there is a difference in the coding of location and
identity information. we thought it might be magnified under conditions of
masking (cf.. Chow. 1986: Irwin 8t Yeomans. 1986; Mewhort et al.. 1981).
Method

mm. Thirteen adults associated with the university participated as
subjects. All reported normal or corrected-to-normal vision. Each was paid
$3 for each of four l-hr sessions. All were naive as to the purposes of the
experiment. Three subjects failed to perform above 55% accuracy (chance -
50%). so their data were excluded from analysis.

WW1 Stimuli were presented on a Hewlett-Packard
l340A point-plotting scope equipped with P31 phosphor. A Digital
Equipment Corporation Micro-1 ll23+ computer controlled stimulus
presentation via digital-to-analog converters. The stimulus display consisted
of a circular array of 8 uppercase letters chosen randomly without
replacement from the set of all consonants excluding y. Each letter in the
display subtended 0.4 deg horizontally and 0.6 deg vertically. and the entire
display subtended 4.2 deg. The probe on no-arrow trials was a single letter

plotted in the center of the circular array: on arrow trials this letter was

19

accompanied by a line approximately 1.1 deg in length that pointed to one of
the locations in the circular array. The masking stimulus consisted of 8
individual dot matrices. one for each of the 8 array locations: each individual
dot matrix subtended 0.4 deg horizontally and 0.6 deg vertically. Subjects
indicated their responses by pressing keys on the keyboard of a Digital
Equipment Corporation VT-240 terminal attached to the computer. The
experimental chamber was illuminated in order to prevent subjects from
detecting phosphor decay.

Brogan. Subjects began each trial by hitting the return key on the
terminal keyboard. Five hundred ms later. the stimulus array was
presented for 50 ms. 0n half the trials. the masking stimulus was then
presented for 30 ms immediately after stimulus offset. while on the other
half nothing was presented for 30 ms. Either 50. 100. 150. 200. 250. 300.
350. or 500 ms after array offset the probe was presented for 50 ms in what
had been the center of the array. As described earlier. on half the trials (no
arrow trials). the probe consisted of a single letter: the subjects‘ task on
these trials was to indicate whether the probe appeared in the stimulus
array. 0n the other trials (arrow trials). the probe consisted of a single letter
and a line pointing to one of the array locations. On these trials. the subjects'
task was to indicate whether the probe letter had appeared in the indicated
position False arrow trials were of two types: The probe letter was present
in the stimulus display. but not in the probed position (false-pram”: or the
probe letter was not presented in the stimulus array at any position (false
resents). Subjects indicated their responses by pressing one key on the
terminal keyboard for "true" and a different key for “false." Feedback was
presented for 1 sec after this response.

Each subject completed 4 practice blocks and 20 experimental blocks of

20

128 trials each over four days. Each block contained 4 replications of the
factorial combination of probe type (arrow vs. no arrow). mask (present vs.
absent). and probe delay (50. 100. 150. 200. 250. 300. 350. or 500 ms): false
arrow trials were equally divided between false absents and false presents.
Probe type. mask presentation. and probe delay were sequenced randomly
within each block of trials.

B l | D' .

The analysis consisted of two major parts. The first part concerned
overall accuracy in each condition. The second part concerned accuracy on
the two types of false trial used on arrow trials. Figure 3 shows the results
for arrow and no-arrow trials under both masked and nonmasked conditions
as a function of probe delay. A 2 x 2 x 8 analysis of variance (collapsing
over type of arrow false trial) with factors of probe type. mask. and probe
delay revealed that accuracy was higher for arrow than for no-arrow trials
[F( 1.9) - 64.5, p < .001. MSe - 148.6) . higher for nonmasked than for masked
trials [F(l.9) - 57.1. p < .001. MSe - 41.1] . and higher at short as compared to
long probe delays [F(7.63) - 5.51. p < .001. MSe - 36.5] . In addition. there
was a slightly larger masking effect (7.1 II vs. 3.7%) for arrow trials than for
no-arrow trials [F( 1.9) - 4.9. p < .06. MSe - 47.8] . and the difference between
arrow and no-arrow trials. though significant at all probe delays. was greater
at probe delays of 200 ms and under than at probe delays greater than 200
ms [F(7.63) - 2.6. p < .02. MSe - 31.9]. This pattern arose because accuracy
for no-arrow trials was essentially constant across probe delay. whereas
accuracy for arrow trials declined as probe delay increased. The results of
this analysis reveal that our new recognition method of measuring very

short-term visual storage yields results very

‘00 I H ARRDw,No MASK

 

 

 

‘ H ARROW, MASK
as 1’ H N0 ARRDw,N0 MASK
<>—<> ND ARRDw,r1ASK
eo '
STANDARD ERROR
t) 75 v
LLI
SE .
o 70 I
I.)
E I»
u, 65
U
35
n. 60 0
55 i
so I

 

If ‘ L ‘ L 3 t i 4‘
so 100 150 200 250 300 350 500

PROBE DELAY (MSEC)

figure 3. Experiment 1: Percent accuracy for arrow and no arrow trials

21

22

similar to those obtained in partial-report experiments: Accuracy declined
as probe delay increased when both location and identity information were
required for responding. and mask presentation interfered with
performance. What is not known. however. is whether it is a loss of location
information which is responsible for the drop in accuracy as probe delay
increases. as previous theorists (e.g.. Mewhort et al.. 1981; Townsend. 1973)
have argued. In order to evaluate this possibility. we conducted a second
analysis which examined performance on false present and false absent
trials in the arrow condition.

To review the logic of this second analysis. if the drop in accuracy with
increasing probe delay is due to a more rapid decay of location information
over identity information. then we would expect accuracy to be lower on
false present trials (those in which the probe letter is in an unprobed
location) than on false absent trials (those in which the probe letter is not in
the array). This becomes Obvious if one considers the time at which location
information has decayed to zero but identity information is still available.
At this point. on false present trials subjects would know that the probe
letter was in the array but they would have no idea where it was. leading to
some incorrect ’yes” responses: on false absent trials. however. there would
be no such confusion so accuracy would be high. In order to test this idea.
we examined just the “false" data from the arrow trials of this experiment. A
2 x 2 x 8 analysis of variance with factors of false type (present vs. absent).
mask (present vs. absent). and probe delay revealed that accuracy was
higher for nonmasked than for masked arrays [F( 1.9) - 14.1. p < .005. MSe -
107.2) and that accuracy declined as probe delay increased [F(7.63) - 3.0. p <
.009. MSe - 165.6] . No effect involving type of false trial was found. The

data for false present and false absent trials are shown in figure 4. collapsed

23

over the masking variable because it did not interact with either probe delay
or false type. What figure 4 reveals is that at longer probe delays accuracy
was actually slightly higher on false present trials than on false absent trials.
directly opposite to the prediction of the independence argument. In other
words. at longer probe delays subjects actually performed better on ”false"
trials when the probe letter was in an unprobed position of the array than
when it was absent from the array. This result suggests that location and
identity are stored as a unit. rather than as separate codes with different
decay rates.

Another result of Experiment 1 also suggests that location and identity
information are nonindependent. Accuracy on arrow trials (which required
both location and identity information) was superior to accuracy on
no-arrow trials (which required only identity information) at all probe
delays; if identity and location are stored separately. then performance on
no-arrow trials should be superior to performance on arrow trials. because
only one source of information (identity) would be needed for a correct
response. This result not only shows that location and identity are stored
together. but that they are stored in such a form that retrieval is aided by
the addition of a location cue. The implications of this will be discussed after
Experiment 2.

To summarize. under recognition-memory conditions evidence was found
for the nonindependence of location and identity information. even though
under recall conditions (i.e.. partial report) location appears to decay before
identity. It may be. as mentioned earlier. that recall conditions induce
subjects to stress identification over localization at encoding or retrieval
time. producing an inaccurate portrayal of the way in which information is
stored in memory.

PERCENT CORRECT

100 I

\

H FALSE PRESENT
as «l H FALSE ABSENT

).

7s ‘*
70

65 T

50

 

so too 150 200 250 300 350 500

PROBE DELAY (I'ISEC)

Figure 4. Experiment 1: Percent accuracy for false present and
false absent trials.

24

25

However. although the results of Experiment 1 indicate that location and
identity are stored together. the longest probe delay used in this experiment
was 500 ms. Thus. it is possible that not enough time elapsed between
stimulus offset and probe presentation to allow any possible dissociation
between the two sources of information to occur. This seems unlikely.
because partial report experiments have found location errors to outnumber
intrusion errors at much shorter intervals: nonetheless. we thought it
necessary to replicate the arrow condition of Experiment 1 with longer probe
delays to see whether location and identity information become separated at
delays longer than 500 ms. This was the purpose of Experiment 2.

Experiment 2
MM

Subjects. Ten adults from the University community participated as
subjects. All reported normal or corrected-to-normal vision. All were naive
as to the purposes of the experiment. Each subject was paid 83 for a single
one-hour session.

W. The apparatus was the same as used in
Experiment I. Only arrow trials were used in this experiment. however. and
masks were never presented.

mtg. Each subject completed a one-hour session which consisted of
one practice block and five experimental blocks. Each block consisted of 128
trials of the arrow. no mask condition from Experiment 1. As in Experiment
1. there were equal numbers of true and false trials. and. within false trials.
an equal number of trials in which the probe letter was present at an
unprobed position and trials in which the probe letter was absent from the
array. Unlike Experiment 1. however. only four probe delays were used: 50.
350. 650. and 1000 ms. The longer probe delays were used to determine

26

whether identity and location become dissociated at longer intervals.
8 | I IT .

Figure 5 shows the results for true. false absent. and false present trials
as a function of probe delay. An analysis of variance of overall accuracy
collapsed across true/false showed that accuracy declined as probe delay
increased [F(3.27) - 6.8. p < .002. MSe - 25.6) . A second analysis of just the
false trials with factors of false type (present vs. absent) and probe delay
(50. 350. 650. 1000 ms) revealed a higher level of accuracy (65.9% vs.
62.4%) for false present trials than for false absent trials [F(l.9) - 8.7. p < .02.
MSe - 28.3] and a decline in accuracy with increasing probe delay [F(3.27) -
2.8. p < .06. MSe - 110.7] . but no interaction between false type and probe
delay [F < I) . In other words. we found that accuracy on false trials was
significantly higher when the probe letter was present in the array at an
unprobed position than when the probe letter was absent from the array. In
Experiment 1 we had found a nonsignificant trend in this direction; with the
greater power of this experiment. the difference achieved significance. This
finding is incompatible with the independence hypothesis. because if location
and identity information are stored separately. one would expect lower
accuracy on false present trials because of confusion generated by the probe
letter matching the identity code of one of the array elements. Thus. even at
probe delays as long as 1 sec no evidence was found for a dissociation
between location and identity information.

A question which arises from this experiment is why false present trials
are summer to false absent trials. This finding may actually constitute
additional support for the hypothesis that location and identity information

100 T 9—9 TRUE
H FALSE PRESENT

5t
\

 

 

 

BS i H FALSE ABSENT
BO 0
:3 7s "
LU
g 0
o 70
u
'7: I.
w 65
t.)
33 .
(L 60 i
55
SO 1?
f 3 ‘f 1* 3
50 350 650 1000

PROBE DELAY (MSEC)

figure 5. Experiment 2: Percent accuracy for true. false present. and false
absent trials.

27

2B

are stored together. based on the following argument. If location and
identity are linked. then on false present trials the subject actually has two
ways of reaching the correct response: the subject may note that the probe
letter does not match the letter in the probed position. or the subject may
note that the probe letter actually appears in an unprobed position and thus
couldn‘t possibly also appear at the probed position. 0n false absent trials.
however. the subject can make a correct response only by comparing the
probe letter with the letter occupying the probed position. This state of
affairs also holds for ”true" trials. of course. and figure 5 shows that accuracy
for false present trials was generally higher than accuracy for true trials.

In sum. the results of Experiment 2 replicate and extend the arrow
condition results of Experiment 1. and further strengthen the argument that
location and identity information are stored together. rather than separately.
These results are problematical for the several models of iconic memory
mentioned earlier which assume that location and identity are independent.
with separate rates of decay. They also raise the question of the nature of
the memory representation underlying performance in iconic memory
experiments. That is. in what kind of memory representation are location
and identity linked as a unit?

Two possibilities come to mind. The first. and perhaps most obvious. is
that the memory representation is image-like in form. An image-like
conception of iconic memory can easily explain both the nonindependence of
location and identity information found in Experiments 1 and 2 (images. by
nature. store form and location as a unit). and the superiority of arrow trials
over no-arrow trials found in Experiment 1; if the storage were in the form
of an image. then the arrow would aid in scanning the image by directing
attention to the only pertinent position in the array. whereas in the

29

no-arrow condition the whole image would have to be scanned.

But these two findings can also be explained by an abstract.
non-image-like conception of iconic memory. Such a representation might
consist of abstract identity codes with associated location coordinates: for
example. subjects might code the contents of a circular array such as those
used in the present experiments as "B-l. M-2. G-3“ and so on. and compare
the probe against this list This conception of iconic memory could account
for the superiority of arrow trials over no-arrow trials in either of two ways.
First. if the representation is location-addressable. subjects could use the
arrow to access the identity code linked to the probed location. then compare
it with the probe letter's identity code; on no-arrow trials. however. the
probe letter would have to be compared with all the array elements. Second.
even if the representation is not location-addressable. fewer comparisons
would need to be made on arrow trials than on no-arrow trials. No-arrow
probes would require 8 comparisons (given our 8 item arrays) on negative
trials. and 4.5 comparisons (on the average) on positive trials. But arrow
probes contain two pieces of information. an identity code and a location
code. and this reduces the number of comparisons that would have to be
made on negative trials because a mismatch of either location or identity is
sufficient to specify a negative response. Thus. on positive trials. 4.5
comparisons (on the average) would still be needed to detect a match
between the probe and an array element. but on negative trials. only 4.5
comparisons (on the average) would be required on false absent trials and
only 2.25 (on the average) would be required on false present trials for a
mismatch to be found. Based on the assumption that fewer comparisons
leads to superior performance. either of these nonvisual conceptions of iconic

memory could account for the finding of higher accuracy on arrow as

30

opposed to no-arrow trials.

Perhaps the most important point about these alternative conceptions of
iconic memory is that they all hold that location and identity are stored as a
unit. in contrast to the other models of iconic memory described earlier.
Nonetheless. it would be satisfying and informative to discriminate between
these alternatives. Unfortunately. the results of the first 2 experiments are
insufficient for this purpose; so. we conducted a third experiment in which
we varied the codability of the elements in the stimulus array in an attempt

to test between the image and non-image conceptions of iconic memory.

Experiment 3

In order to discriminate between the image and non-image conceptions of
iconic memory. in Experiment 3 we used stimuli that would presumably be
easier to store in one form than another. Specifically. we used pseudoletters
constructed by rearranging the lines of our letter stimuli. Since these
pseudoletters are novel visual stimuli. they presumably lack abstract
identity codes. Thus. if iconic memory is abstract in nature. recognition
memory performance with these stimuli should be quite poor. If iconic
memory is image-like in nature. however. recognition memory performance
with these stimuli should be similar to performance with letter stimuli.
Overall accuracy for the pseudoletter stimuli may be less than for letter
stimuli. simply for reasons of f amiliarity. but in other respects performance
for the two types of stimuli should be very similar.

Another. intermediate pattern of results might be obtained as well.
however. Several models of iconic memory assume that persistence consists
of two components. a raw. image-like. precategorical representation and an

abstract. non-image-like. postcategorical representation (e.g.. Irwin &

31

Yeomans. 1986; Mewhort et al.. 1981). For brief exposures such as those
used in the present experiments. these models assume that the image-like
representation is present immediately after stimulus Offset. but is then
replaced by the postcategorical representation. Based on these models of
iconic memory. we would expect letter and pseudoletter stimuli to exhibit
similar performance at short probe delays. but then at some critical probe
delay pseudoletter performance should drop dramatically as the transition
from an image-like to an abstract representation is made.

Method

Subjegs. Ten adults associated with the University participated as
subjects. All reported normal or corrected-to-normal vision. and all were
naive as to the hypotheses under study. Each subject was paid 33 for each of
two one-hour sessions.

W. The equipment used was the same as in the
previous experiments. Two types of stimuli were used. however. The first
type was simply the set of 20 consonants used in the previousexperiments.
The second type was a set of 20 letter-like stimuli (pseudoletters)
constructed by rearranging the lines composing the letter stimuli. These
pseudoletters had the same dimensions as the letter stimuli and they were
designed to not resemble any letter or any common symbol; these stimuli
are shown in Figure 6. The stimulus array thus consisted of a circular array
of 8 letters or 8 pseudoletters; the probe was always of the arrow variety.

containing a letter or a pseudoletter and a line pointing to one of the array

Figure 6. Pseudoletter stimuli used in Experiment 3.

32

‘1? (I) 3 5.5

33

positions.

2129995111112. Each subject completed 12 blocks of 128 trials each over a
2-day period. In half the blocks letters were used as stimuli. and in the
other half pseudoletters were used as stimuli. This assignment alternated
from block to block; half the subjects started with a letter block and half
with a pseudoletter block. The first block with each type of stimulus was
discarded as practice.

Subjects began each trial by hitting the return key on the terminal
keyboard. Five hundred ms later. the circular array of 8 letters or 8
pseudoletters was presented for 50 ms. Then. either 50. 100. 150. 200. 250.
300. 350. or 500 ms later. the probe was presented. The probe consisted of
a letter (in letter blocks) or a pseudoletter (in pseudoletter blocks) and a line
pointing to one of the array positions; the subjects' task was to indicate
whether the probe letter (or pseudoletter) had appeared in the indicated
position. "False" trials were of two types: The probe letter (or pseudoletter)
was present in the stimulus display but not in the indicated position (false
presents). or the probe letter (or pseudoletter) had not appeared in the
display (false absents). Feedback was presented after each response. Each
block of 128 trials consisted of 8 ”true” trials and 8 ”false" trials at each of
the 8 probe delays; probe delay and trial type were sequenced randomly
within each block of trials.

8 II I D' .

The analysis consisted of two parts. The first part concerned overall
accuracy in each condition: Figure 7 shows the accuracy results for letter and
pseudoletter stimuli as a function of probe delay. A 2 x 8 analysis of
variance with factors of stimulus type (letter vs. pseudoletter) and probe
delay (50. 100. 150. 200.250. 300. 350. 500) revealed that accuracy

100
1} H LETTERS

85 r D—D PSEUDOLETTERS

\

80 NP

75 1‘

PERCENT CORRECT

60 4» \3

SS "

50

 

l A L A A A A

f

50 100 150 200 250 300 350 500
PROBE DELAY (I‘ISEC)

figure 7. Experiment 3: Percent accuracy for letter and psuedoletter stimuli.

34

35

declined as probe delay increased [F(7.63) - 5.8. p < .001. MSe - 29.6] and
that accuracy was higher (74.6% vs. 62.9%) on letter trials than on
pseudoletter trials [F( 1.9) - 256.0. p < .001. MSe - 21.2] . No interaction was
found between these two variables [F(7. 63) - 1.2. p > .3. MSe - 22.4] .
however.

The second analysis concerned accuracy on the two types of false trials:
Figure 8 shows the results for false presents and false absents as a function
of probe delay for both letter and pseudoletter stimuli. A 2 x 2 x 8 analysis
of variance with factors of false type. stimulus type. and probe delay
revealed that accuracy was higher (79.2% vs. 59.2%) for letter trials than for
pseudoletter trials [F( 1.9) - 39.1. p < .001. MSe - 816.4] . Note that although
accuracy on pseudoletter false trials was near 50 %. this doesn't represent
near-chance performance; performance would be near chance only when the
total accuracy of true and false trials together is 50%. There was also a
marginal main effect of false type, due to slightly higher accuracy (69.9% vs.
68.5%) on false present trials than on false absent trials [F(l.9) - 2.9. p < .13.
MSe - 59.4] . No other main effects or interactions were significant An
additional analysis was conducted to deter mine whether the non-significant
trend toward an advantage for false absent trials found in the pseudoletter
condition might be significant in early blocks of trials when name codes are
less likely to occur. To this end an analysis was conducted with block
number as a factor. Though there was a main effect for block number. it did
not interact with any other factor and so the entire analysis is not included.

The results of the second analysis again provide no evidence that location
information decays before identity information. Accuracy for both letters
and pseudoletters was slightly higher for false present trials than for false
absent trials. directly counter to the predictions of the independence

H LETTER, FALSE PRESENT

 

 

 

 

 

 

100 I H LETTER, FALSE ABSENT
4f D—D PSEUDOLETTER. FALSE PRESENT
85 f <>-—-<> PSEUDOLETTER, FALSE ABSENT
it ,
8° \~‘ A
5 75 l
LLJ
E .t
o 70
I.)
5 .
m 65
U
5 JI-
o. 60 <>
55 i? g
50 0
f s s c s .L c e :
50 100 ISO 200 250 300 350 500

PROBE DELAY (I‘lSEC)

figure 8. Experiment 3: Percent accuracy for false present and false present
letter and pseudoletter stimuli.

36

37

argument. This result supports our earlier finding that location and identity
information are stored as a unit. and further shows that this is true for
pseudoletter stimuli as well as letter stimuli.

This result. along with the results of the first analysis. also allows us to
tentatively discriminate between the image and non-image conceptions of
iconic memory. The results of the first analysis show that accuracy for
pseudoletters was significantly above chance at each probe delay. and
furthermore showed the same decay over probe delay as did the letter
stimuli; these results indicate that abstract identity codes (which
pseudoletters presumably lack) do not form the basis of iconic memory. In
other words. these findings are more consistent with an image-like
conception of iconic me mory. rather than an abstract conception. Nor was
there any evidence for a dual-b uffer conception of persistence. since the
difference in accuracy between letters and pseudoletters did not change as
probe delay increased.

In summary. the results of the third experiment also indicate that
location and identity are nonindependent. and furthermore suggest that
form and location information are stored in an image-like representation. Of
course. this final conclusion is based on the assumption that it takes longer
than two experimental sessions to develop identity codes for pseudoletter
stimuli; furthermore. it is possible to conceive of an abstract representation
which is based not on name or identity codes but on feature lists or some
such stimulus description. Unfortunately. this possibility is not testable
using the present paradigm. What we can say. however. is that if iconic
memory is abstract in nature. it must not rely on name codes as its unit of

storage. or it develops these codes very rapidly.

General Discussion

The purpose of the present research was to test the hypothesis that
location and identity information about the contents of briefly presented
visual displays is stored separately. with different rates of decay. Although
previous research using recall measures of memory (e.g.. partial report) has
shown that location errors are much more common than intrusion errors. the
results of our three experiments using a recognition test of memory indicate
that location and identity are stored as a unit. There was no evidence that
location information decays more quickly than identity information.

The partial report technique is an indirect method of assessing the
storage characteristics of iconic memory. because some transformation of the
information stored in memory may be necessary in order for a response to
be produced. Thus. it is possible that the preponderance of location errors
over intrusion errors found in partial report tasks is due to retrieval or
response biases rather than to differences in the way in which location and
identity information are stored in memory. Partial report experiments
almost always use letters or numbers as stimuli; since subjects have had
much more experience with the overlearned identity codes associated with
these stimuli than with the location coordinates associated with any
particular experimental display. it may not be too surprising that under
conditions of uncertainty subjects prefer to report a character they know
was present in the stimulus array (thereby making a location error) rather
than reporting a character that may have been absent. Our recognition test
of iconic memory eliminates this kind of response bias and yields a more
direct measure of memory storage per se. revealing that in fact location and

identity are stored as a unit

38

39

These results have negative implications for the several models of iconic
memory which have assumed that location and identity information decay at
different rates (eg. Irwin 8: Yeomans. 1986; Mewhort. et a1. 1981; Dixon.
1986). Furthermore. the results of Experiment 3 showed that pseudoletter
stimuli. which presumably lack abstract identity codes. are stored in a
fashion similar to letter stimuli. This finding suggests that iconic memory is
image-like in nature. rather than some abstract representation of location
and identity information. This. too. is inconsistent with the models of iconic
memory mentioned above.

The model of iconic memory proposed by Irwin and Yeomans (1986) was
based on several partial report experiments which showed that presentation
of a masking stimulus 150-300 ms after stimulus offset interfered with
report. regardless of stimulus duration. although accuracy did increase as
stimulus duration increased. These results were interpreted as evidence for
a two-store model of iconic memory. the first a fast-decaying.
duration-independent. maskable representation. and the second a durable.
nonmaskable. abstract identity representation. The results of the present
research suggest that this second memory store is in fact image-er in
nature. rather than abstract It is important to note that the only remaning
evidence for a two-store model of iconic memory is the masking data of
Irwin and Yeomans. It is possible that the recognition paradigm used in
these experiments might find different results than those found in the
masking experiments of Irwin and Yeomans and might allow us to dispense
with a two stage model altogether.

Although the concept of a nonmaskable image seems counterintuitive.
there is other experimental evidence for the existence of such a

representation. For example. Posner. Boles. Eichelman. 8: Taylor (1969).

40

using the Posner and Keele (1967) letter-matching procedure. found that a
reaction time advantage for physical matches (AA) over name matches (Aa)
is found even when a noise mask is presented between the first and second
letters. Similarly. Phillips (1974) found that accurate same/different
judgments of random dot patterns could be made even if a mask was
presented between the two patterns being compared. These results provide
strong support for the existence of nonmaskable. image-like representations
in memory.

In conclusion. the results of the present research indicate that location
and identity information about the contents of a brief 1y-presented visual
display is stored in a unified. image-like form. rather than as separate

abstract codes with different rates of decay.

LIST OF REFERENCES

Adelson. E. H.. & Jonides. J. (.1980) The ozsychohysics of Iconic storage.
. 486-493.1.

I” II‘I._ QOOL 'e II._I 0100018.
Baddeley. A. D. (1976). W. New York: Basic Books.

Bowen. R. W.. Pola. J.. 8: Matin. L. (1974). Visual rsistence: Effectsof
flash luminance. duration. and energy. L4. 295-303.

Coltheart. M. (1980). Iconic memory and visible persistence. mm
mm 21 183- 228.

Coltheart. M.. Lea. C. D.. 8: Thompson. K. (1974). In defence of Iconic
memory Wm 6-33 641

Chow. S. L. (1986).. Iconic memo .locationinformation. and artial report.

I l1__0,_u‘ II‘I1_.‘}IOOJ‘QIIZI‘ o I man an I lli‘l‘

4-4.

Den Heyer. K. (1972). The processing of multidimensional information from
iconic storage: Pergegtgzall In‘d‘ependence and comparative rates of decay.
mammal.

Dick. A O. (1969). Relations between the sensory register and short- -term
stor gage In 81.2chIstoscopIc recognition.

Dick. A. O. (1974). Iconic memory and its relation to perceptual grocesses
and other mechanisms. Was. 16. 575- 96.

Di Lollo. V. (1978). On the Epatio-temporal interactions of brief visual
dispI s. lnR. H. Day 8: V Stanley (Edam) Wlpp
39-55. Perth: University of Western Australia Press.

Di Lollo. V. (1980). Temporal integration in visual mem . journal of
Wm 1.0.2. 75-97. W

Dixon. P. (1986). Attention and interference In the perception of brief visual
displays 0 I ‘ II‘I.__ ") son"; I IIQI" oI OJICII
Performance. 13 - 48.

Efron. R. (1970). Effect of stimulus duration on perceptual onset and offset
latencies We. 8. 23123.4

Eriksen ..C W. 8: Collins. JF. (1968). Initial traces versus the p chological
moment in the tem alorganization of form. Joumalflmnmemal
25191191981. 12. 37- 32.8

41

42

Garner. W. R.. 8: Morton. J. (1969). Perceptual independence: Definitions.
gaffezlgband experImental paradigms. WWII], 12.

Haber. R. N.. 8: Standin . L. (1969). Direct measures of short-term visual
storage. Wm 21. 43-54.

Haber. RN. 8: Standing. L. (1970). Direct estimates of the apparent
duration of a flash. Whom 24. 216-229.

Irwin. D. E.. 8: Brown. 5.. S. (in press). Tests of a model of informational
persIstence. .

Irwin. D. E.. 8: Yeomans. J. M. (l986a). Sensory registration and

informational persistence. 4 0

Irwin. DE. 8: Yeomans. J.M. (l986b). Persisting arguments about visual
persistence : Reply to Long. W 3.2. 225-230.

Kru mhansl._ C. L. (1977). Naming and locating simultaneously and
sequentially presented letters. Wm 22. 293-302.

Krumhansl. C. L.. 8:Thomas. E. A. C. (1976). Extract' identity and location
information f roamgbtéigg 1y2 ggesented letter mraysm

Krumhansl, C. L.. 8: Thomas. E. A. C. (1977). Effect of level of confusability
on reporting letéelrszlg'gnggiefly presented visual displays. Eempflgnj:
Weiss. -

Logan. G. D. (1975). On the independence of naming and locat masked
targets in visual search. MW 22. 1-58.

Long}. G. M.. 8: Beaton. R. J. (1980). The contribution of visual persistence to
t e perceived duration of brief targets. W
28.. 422-430.

Lonhgr. G. M.. 8: Beaton. R. g (1982). The case for peripheral persistence:
fects of target and ackground luminance on a partial-report task.

(I In IIaI 3 "3 II 0,3' g II ._I ' o“: m ,I0 ' IJ II 51.;

-3391.
Mewhort. D. J. K. Camfilzell. A. J.. Marchetti. F. M.. 8: Campbell. J. I. D. (1981).

Identification. loca ation. and "iconic" me mor_/y: An evaluation of the
bar-probe task. ' ' 9_. 50-6 .

MewhortD.J.K.. Marchetti. F.M.. Gurnse .R.. 8: Cam bell. AJ (1984). .
InformatIon persistence: A dual-bu fer model or InItIa VIsual processIng.
In H. Bouma 8: D.G.Bouwhu18 (Eds). WW
(ppZO7-298). London: Erlbaum.

Neisser.U. (1967). Wining. New York: Appleton-Century-Crof ts.

43
Phillips. W. A. (1974). On the distinction between sensory storage and
W115.

short- -term visual memory. 83-290.

Posner. M. 1., Boles. S J.. Eichelman. W. H.. 8:Ta lor. R. L. (1969). Retention
of visual and name codes-if ($11 1e 2letters. W
W7 t

Posner. M 1.. 8:Keele. S. W. (1967) Decay of visualinformation from a
single letter. Science. 118. 137- 139.

Sperling. G. (1960). The information available In brief visual presentations.
14(11. Whole No. 498).

Sperling.G. (1967). Succesive apggaximations to a model for short-term
Wise. .

memory. 285-292
Townsend. V. M. (1973). Loss of s atial and identity information following a
tlalcgisitlegoopic exposure MW 28.

Yeomans. J. M.. 8: Irwin. D. E. (1985). Stimulus duration and partial report
performance W32. 163-169.

V

HICHIGAN STATE UNI
II1|WI)“IIIWIWINIIIUII
31293006

I'lllljllljl“

E3121