MW?” ‘ 'w"

.£§r$33?}4xahgr
4&4 - 4

 

Q
"If 24%.

{In I' I '

E’E'fibv‘

HP% "

. , ~4

E 1‘5"

.:.

'3 $4? .

: L‘M: “c" u
:6 .3. \ 9g“ ”W ~
I}.

.4 6'?" .1».
y: a “7%

€ I

'IJ

_ gage/fix", m.- -4. )bnfiW'

. "fl
.. 2“":4'4f: t3£‘,‘.xu(,‘. ’Ex‘: I?” J
3,; 4“" a; " . ., - .1 T "W .1: .
XVFIE§§3~if :2". E‘H “ ‘J.’ M1,. 4 ' ~Nwrr-fvxg‘ . 5' 1 :.‘§I;‘-E:‘"\}““‘
g3~§m¢.n Mfififlx
.-Q'§‘.'
3 V. K .

HE‘S" ‘

«M
:::;::~:?’~f. a
.i'ei’ if.\1.‘.';“: 1-_
‘ *2

u
,.

\ J \
“VA”... . .C .7" 4"?"
n < - - ‘ 4 fi-
4 1'.
~‘ 73:14.6. “.2353. fig?!“

'01“; ’- fin‘ v?!"
'2; 3 ~,{ ~ 2. :C2: ‘5? ‘

' {E‘${ '\ 'I’f’r‘

s

4 40".

:uVig 26?; ”:45.
1J1}; -_

-' “$131.3.
'4'. ‘

. 'kgg‘ivt‘
« y

.k
I 5"“."hivh »
9

.n v;-
H‘

 

~.— ‘ "

A - .'. '.‘.‘.' '4 v;' #1 ' 4 /

' 1191::"5-HI .. ”'1' 37!;
‘6‘ , _~’~::’.«fl

fl 1
a‘, ' 4
k x». 4.».
. M w: w :x- ..
:5 H‘- l:‘a{,-/‘J' ‘7',“ x“ «‘Q' J! ‘Véu:

\é“ . . 4.1:" 4‘. - £3315; hr??? 15:?

e

I .

I“???
‘72:
flux 9.,

it". vf‘éJfKr: "“4",”

(-7-! ‘5! ”as!”
"n Jc‘;".4:,.",wl ‘
( ha" "7;; 05

4 i-

5;"

mm... «2 34.32135: v~ ;.

,(\‘ ‘ “5% u 4 ‘H

Wit: ”’2 51;]; Evifl‘.‘ "I?

. .7‘.‘ l . H‘ | 31* 3
c ‘ '

u; M ‘Hagj‘kffl :fzz‘

/
:fi.é;
T" 3;“) '\%:§q
'2" “
{3

‘ .8), -

' '\

. _'
15-.
.

view

3"!!!
‘ a
J» . .' ‘-'

‘Vmfla
Nfi

‘ . ‘ .' \, .. LI, ’T‘." '5? . i
. .r o - ." 1. "”‘ 4_.\. I
.- “,"'l‘l;l‘( .‘. , “1‘4" . . ‘ -' .1. r ) ‘i(‘ ‘3‘ L 4 9 _ '. ‘ 1'".er
"V ‘ . .711 . w z- , ,- _- ' "1“"! J ‘56."? "(9‘ ‘

 

A
\E .; '
24.2415

-;-)’ \~\“" (1,”

 

 

T'n'fig‘s

Date

0-7639

)goflQOEO

fl MICHIGAN STATE UNIVERSITY LIBRARIE

Illllflllli WW II! "lll I"§ll,|!'lllllllllll

1M 1!!

3 1293 00563 6414

LIESRARY
Michigan State
University

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

This is to certify that the

thesis entitled

STIMULUS COMPLEXITY AND THE DURATION
OF VISIBLE PERSISTENCE

presented by

James Michael Yeomans

has been accepted towards fulfillment
of the requirements for

M.A. degree in Psychology

QM E m

Major professor

[oi/(3,337

MS U is an Affirmative Action/Equal Opportunity Institution

 

 

MSU

 

 

RETURNING MATERIALS:
Place in book drop to
remove this checkout from

 

 

>37»

 

 

LIBRARIES
.—.:-—- your record. FINES will
be charged if book is
returned after the date
stamped below.
N.,~9~c~9£‘w
” fl*‘fia

 

STIMULUS COMPLEXITY AND THE DURATION

OF VISIBLE PERSISTENCE

BY

James Michael Yeomans

A THESIS

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

MASTER OF ARTS

Department of Psychology

1988

ABSTRACT

STIMULUS COMPLEXITY AND THE DURATION
OF VISIBLE PERSISTENCE

BY

James Michael Yeomans

Visible persistence refers to the ‘phenomenal
impression that a stimulus is still visibly present after
its offset. A dispute exists whether visible persistence
is under cognitive control ("process" hypothesis) or
merely reflects temporal sluggishness in the visual
pathway ("neural" hypothesis). This was investigated by
manipulating cognitive load (stimulus complexity) in 10
experiments. These experiments used one of two
techniques: fit) a temporal integration task, requiring
subjects ' to integrate perceptually two temporally
separated stimuli; and (2) an onset-offset method,
requiring subjects to make subjective judgments about the
onset and offset of stimuli. Complexity was found to
have no effect on visible persistence duration,
indicating that visible persistence is not under
cognitive control; these results support the ”neural"
hypothesis. A set of complexity criteria was developed,
and implications for models of visual processing were

discussed.

THIS THESIS IS

DEDICATED

TO MY MOM AND DAD

CLAIRE AND JACK YEOMANS

iii

h‘

54.

()1

.93

ACKNOWLEDGMENTS

This research was conducted in collaboration with
Professor David Irwin, and was supported by NSF Grant BNS
85-19580 to David Irwin.

I would like to thank the members of my graduate
committee for their help and support of this project. In
particular, I would like to thank my chairman, David
Irwin, for allowing me to move gradually into the
research domain, and for providing me with an example of
excellence in teaching and research. Professors Thomas
H. Carr and James L. Zacks have also been very
informative and supportive throughout this project.

Several of my fellow graduate students have also had
an influence on this project, as well as my life in
general at Michigan State University. In particular, I
would like to thank Bill Hildebrandt, Joseph Brown, and
G. A. Radvansky.

Most importantly, I would like to thank the many
close friends who I have made at Michigan State, as they
have helped me during some of the most wonderful and most
difficult times of my life. These friends are John

Geisler, Tom Klever, Stephanie Smilay, Becky Potts, Ted

iv

Brewer, Ron Lapporte, Jorge Gonzalez, Lisa Lipinski,
Steve Underwood, and Gary Wolfram.

Finally, I would like to thank .Andrea Dale for
teaching me the true meaning of love, and for being there

when I most needed a friend.

Table of Contents

LISTOFTABLESOOOOO 0000000000 .....OOOOOOOOOOOOOOOOViii

LIST OF FIGURES ....... . ....................... .......x
INTRODUCTION ......................................... l

EXPERIMENTS ......................................... 19

General Methods ................................. 19
Subjects.... ...................... ..... ....... 19
Stimuli.... ........ . ..... .....................19
Equipment.....................................21
Procedures....................................22
Conditions....................................23

Experiment 1....................................23
Materials and Methods.........................23
Results and Discussion........................ 25

Experiment 2....................................30
Materials and Methods............. ....... .....30
Results and Discussion........................31

Experiment 3............ ........................36
Materials and Methods. ........................36
Results and Discussion........................37

Experiment 4....................................41
Materials and Methods.........................42
Results and Discussion........................46

Experiment 5....................................51
Materials and Methods.........................52
Results and Discussion........................52

Experiment 6.......................... .......... 57
Materials and Methods.... ..... . ............... 58
Results and Discussion. ........ ........ .......62

Experiment 7 ............ . ....... . .............. 68
Materials and Methods ......................... 68
Results and Discussion. ...... ..... ...... ..... .69

Experiment 8...... ........ ... ........... ........76
Materials and Methods.........................77
Results and Discussion........................80

Experiment 9....................................87
Materials and Methods.........................88
Results and Discussion.. ..... ... ........... ... 89

vi

Experiment 10............. ........ . ......... ....93
Materials and Methods... ....... ...............94
Results and Discussion........ .............. ..95

GENERAL DISCUSSION. . . ................... . ...... . . . . 106

Symmetry..... ......... . .......... . ....... ......108
Subject Ratings... ............................. 109

Groupability..... ...... ....... ................. 109
Nameability....................................110
Distinctiveness................................110
Predictability.................................111

APPENDIX. 0 O O O O O .................................... 121

REFERENCESOOOO0.00......0.0.000.........OOOOOOOOOOO125

vii

Table

4a.

4b.

LIST OF TABLES

Individual and mean results (% accuracy)
from experiment 1, as a function of

Page

interframe interval ...... . ....... ...... ....... 28

Individual and mean results (% accuracy)
from experiment 2, as a function of frame

1 duration .................................... 34

Individual and mean results (% accuracy)
from experiment 3, as a function of

interframe interval ........................... 40

Individual and mean results (% accuracy)
from experiment 4, with simple Ichikawa
patterns, as a function of interframe

interval .............. ...... ...... ...... ...... 49

Individual and mean results (8 accuracy)
from experiment 4, with complex Ichikawa
patterns, as a function of interframe

interval ................. ....... .............. 50

Individual and mean results (% accuracy)
from experiment 5, as a function of

interframe interval..................... ...... 55

Individual and mean persistence durations
(in ms) from experiment 6, as a function

of complexity and report condition.... ........ 65

Number and associated proportion of
trials in which subjects responded too
quickly, as a function of condition and

type of trial for experiment 6 ................ 67

Individual and mean persistence durations
(in m3) from experiment 7, as a function

of complexity and report condition ............ 73

viii

Table

10.

ll.

12.

13.

14.

Number and associated proportion of
trials in which subjects responded too
quickly, as a function of condition and

type of trial for experiment 7 ..............

Individual and mean persistence durations
(in ms) from experiment 8, as a function

of complexity and report condition ..........

Number and associated proportion of
trials in which subjects responded too
quickly, as a function of condition and

type of trial for experiment 8.... ..........

Individual and mean results (% accuracy)
from experiment 9, as a function of

interframe interval.... .....................

Individual and mean results (% accuracy)
from experiment 10, as a function of

interframe interval ......... .......... ......

Individual and mean results (% accuracy)
from the pooled data of experiments 9 and

Page

.075

..83

..84

..92

O. 98

10, as a function of interframe interval ..... 103

ix

LIST OF FIGURES

Figure Page

1. General procedure used in experiments

1-3 ......................................... 20
2. Percent correct for rectangles and

random letters as a function of

interframe interval in experiment 1 ......... 26
3. Percent correct for rectangles and

random letters as a function of frame

one duration in experiment 2........... ..... 32
4. Percent correct for X-arrays and random

letters as a function of interframe

interval in experiment 3 ......... .. ......... 38

5. Sample of low complexity stimuli
(Ichikawa, 1985) used in experiments 4
and 50.0000.0.00000.000000000000000. 00000000 43

6. Sample of high complexity stimuli
(Ichikawa, 1985) used in experiments 4

and 5 ....................................... 44
7. Percent correct for simple and complex

Ichikawa stimuli as a function of

interframe interval in experiment 4......... 47
8. Percent correct for simple and complex

Ichikawa stimuli as a function of
extended interframe intervals in

experiment 5 ................ . ............... 53
9. General procedure used for the onset-
offset experiments (i.e., 6, 7, and 8) ...... 60

10. Onset-offset persistence duration as a
function of complexity (simple versus
complex Ichikawa patterns) and report
conditions for experiment 6......... ........ 63

Figure Page

11. Onset-offset persistence duration as a
function of complexity (X-array versus
random letters) and report conditions
for experiment 7 ............................ 70

12. Pseudo—letters employed in experiments
8, 9, and 10. These pseudo-letters
were created by interchanging the line
segments of all of the consonants with
the exception of y...... .......... ..... ..... 78

13. Onset-offset persistence duration as a
function of complexity (random letters
versus pseudo-letters) and report
conditions for experiment 8 ....... . ......... 81

14. Percent correct for random letters and
pseudo~letters as a function of
interstimulus interval in experiment 9 ...... 90

15. Percent correct for random letters and
pseudo-letters when report was required
as a function of interframe interval in
experiment 10............ ...... .............96

16. Percent correct for random letters and
pseudo-letters as a function of
interstimulus interval for the pooled

data of experiments 9 and 10 ............... 102
17. Irwin and Yeomans' (1986) model of the

early stages of visual processing.. ........ 118
A.l. Simple Ichikawa Stimuli: 27—49. ........... 121
A.2. Simple Ichikawa Stimuli: 50-65.. .......... 122
A.3. Complex Ichikawa Stimuli: 101-124 ......... 123
A.4. Complex Ichikawa Stimuli: 125—140 ......... 124

xi

INTRODUCTION

Allen (1926) reported that the first reference to
visual persistence can be seen in the writings of
Aristotle (384-322 B.C.). Interest in this subject
matter was renewed in the late 1800's by several
researchers (e.g., Baxt, 1871; Cattell, 1886). Using
tachistoscopes, and presenting stimuli (i.e., arrays of
letters or numbers) at very short durations (usually 10-
50 ms), these studies demonstrated the existence of a
very short-term memory store. The general finding was
that only a small subset of information (i.e., 3—5 items)
could be reported. Subjects in these experiments
insisted that they saw more than they actually reported,
however, although. nothing came of ‘these introspective
reports for nearly 70 years.

Sperling (1960) performed some replications and
extensions of this pioneering work. In a typical
experiment, a 3 row by 4 column array of letters was
flashed tachistoscopically to subjects. As in the
previously noted studies, the subject's task was to
report as many letters as possible, and the finding was
that 3-5 letters were correctly reported. This method of

obtaining responses has come to be known as a full

IE

:4

h.

a:

report. Sperling, having replicated the previous
findings, then modified the procedure in the following
way: (1) the 3 x 4 letter array was presented; (2) a
variable length delay would occur; (3) a tone would be
sounded that indicated whidh row of the array to report
(i.e., a high frequency tone was used to cue the top row,
a medium frequency tone corresponded to the middle row,
and finally, a low frequency tone was used to cue the
bottom row) If the tone occurred immediately at offset,
subjects were able to report 3 out of 4 (on the average)
of the letters in the cued row. This suggests that at
least 9 of the letters were available for report (i.e., 3
letters from each of the 3 rows), since the subjects had
no knowledge of which row would be cued on each trial.
As the delay between the array offset and the cue onset
was increased, performance declined until it reached the
full report level (at about a 500 ms delay). This method
has come to be known as partial report.

At the same time, Averbach and Coriell (1961)
performed experiments using visual, rather than auditory
cues. Employing a visual pointer and 2 x 8 letter
arrays, they demonstrated that a partial report
superiority exists until a cue delay of about 500 ms.
This finding' was important because it removed output
interference as a possible explanation for the results.

In other words, although it could be argued that

Sperling's results were due to interference at the time
of reporting the letters (i.e., you have several letters
to report and they interfere with one another), Averbach
and Coriell only required their subjects to report one
letter per trial and found basically the same results.
In other experiments, Averbach and Coriell used a
different type of visual cue, an annulus that circled the
letter to be reported. At stimulus onset asynchronies
(i.e., the time from the onset of the array until the
onset of the cue) of 100 ms, they discovered an unusual
occurrence: circled letters seemed to be erased. This
phenomenon is often referred to as metacontrast.

The finding that visual cues can create :masking
(i.e., erasures), as well as the introspective reports of
subjects in full report conditions (i.e., that
information faded before it could be reported) led to the
development of a "visual" notion of sensory persistence.
A great deal of evidence soon appeared that also seemed
to be consistent with this visual conception. In
general, it seemed that cueing by physical variables
(e.g., brightness) led to partial report superiority.
For example, color (Clark, 1969), brightness (von Wright,
1968), and shape (Turvey & Kravetz, 1970) have proved to
be effective cues. Also, the failure of informational
type cues (e.g., report only the letters that end in
"ee," Coltheart et al., 1974) lent strength to the idea

€21
1C0.

pure

iave

u. u.

of a decaying visual trace. Thus, these results led to
the consensual notion that visual information persists
after stimulus offset in a pre-categorical, high
capacity, quickly decaying memory store, which Neisser
(1967) called "iconic memory." This view shall
henceforth be referred to as the "traditional view" of
iconic memory.

The partial report technique has been called an
indirect measure of sensory persistence because it relies
on the retrieval of visual information about a stimulus
display, which does not necessarily have to be visible.
Similar results have been found using more direct
techniques. Direct methods are those that require more
purely visual judgments. For example, Sperling (1967)
invented a technique which measured the phenomenal
duration of a stimulus, in which subjects adjusted the
occurrence of a probe (tone) so that its onset and offset
appeared to be synchronous with stimulus onset and
offset. Another technique was introduced by Eriksen and
Collins (1967, 1968), in which two random dot patterns
were presented sequentially, separated by a variable
interframe interval (i.e., the time from the offset of
the first event to the onset of the second event). When
superimposed, these patterns formed a nonsense syllable,
and subjects were found to be able to integrate the two

displays for interframe intervals up to 100-300 ms.

I
*n
9..

Thus, the traditional view was further strengthened,
since the direct methods (i.e., highly visual in nature)
found evidence of a decaying trace lasting for time
periods similar to those of the indirect partial report
procedure.

There were some studies that seemed to conflict with
the traditional view, however. Dick (1969) presented
evidence which suggested that the physical aspects of the
display (e.g., location information, brightness, etc.)
seemed to decay faster than the stimulus identity
information. Townsend (1973) found data that were
consistent with Dick's results also. For example, most
of the errors in Townsend's partial report task were
mislocation errors (i.e., the incorrect letters in the
subject's response were present in the stimulus array,
but not at the cued locations), rather than intrusion
errors (i.e., the incorrect letters in the subjects'
report were nowhere in the array). This fact suggests
that items in the array were identified and remembered
quite well (provided the stimulus duration was not too
short), but their locations are forgotten. Mewhort,
Campbell, Marchetti, and Campbell (1981) found that
familiarity with the stimulus array (i.e., different
orders of approximation to English-~arrays in which the
letter combinations were varied as to how word-like they

were) reduced the number of intrusion errors but not

4»)-

bad

hfl‘l

mislocation errors. These findings of differential decay
of location and identity information suggest that perhaps
the indirect measure (partial report) is tapping more
than just raw stimulus information.

Recently, the unitary model of sensory persistence
(i.e., the traditional view) has come under heavy attack.
Coltheart (1980) has proposed a three component model to
explain the preceding failures of the traditional model.
Coltheart (1980) has stated that the "icon" should be
subdivided into the following types of persistence: (1)
neural persistence, due to residual activity in the
visual pathway; (2) visible persistence, or the
phenomenal impression that the stimulus is still visibly
present; and (3) informational persistence, or knowledge
about the visual properties of the stimulus. It is quite
possible that visible persistence is due to neural
persistence, since both processes are believed to last
for about 100 ms. The main division in this model occurs
between visible persistence and informational persistence
(i.e., what partial report is claimed to measure).
Evidence used to support this distinction is that the two
types of persistence appear to be differentially affected
by stimulus duration and intensity changes. Visible
gpersistence duration (measured using direct methodS)
decreases as stimulus duration or stimulus intensity

increases (e.g., Bowen, Pola, & Matin, 1974; DiLollo,

b a
3
Mu.

3.
.Q‘
\i‘

h\
Q.»
‘3

me.
u a
c \

 

1977, 1980; Efron, 1970a, 1970b, 1970c; Haber & Standing,
1969, 1970), while in partial report tasks (i.e.,
measuring informational persistence) increasing stimulus
intensity or stimulus duration results in no, or a
positive effect on performance (e.g., Adelson & Jonides,
1980; DiLollo, 1978; Long & Beaton, 1982; Sperling, 1960;
Yeomans & Irwin, 1985). Thus, to summarize, Coltheart
(1980) states that visible persistence and informational
persistence are separate phenomena, because they are
differentially affected by changes in either stimulus
duration or intensity.

This thesis will be concerned with exploring the
nature of visible persistence. The first question that
arises about visible persistence is "what exactly is it?"
Coltheart (1980) suggested that it resulted from residual
neural activity in the visual pathway. DiLollo (1980)
has taken a similar view, although he named it the
"recruiting phase" of stimulus processing. DiLollo
claims that this stage is really just temporal
sluggishness that is caused by the finite duration of a
message sent to the brain at stimulus onset. In other
words, neural messages do not travel at the speed of
light, so they appear to persist phenomenologically for a
period of time. DiLollo's main source of evidence comes

from a temporal integration paradigm, in which subjects

and

A

We

are required to integrate perceptually two temporally
separated stimuli in order to arrive at the correct
answer. A 5 x 5 array of dots was used, and the two
frames each consisted of 12 of the 25 dots. The subject
was required to determine which of the 25 locations had
been unoccupied. Performance was found to deteriorate as
the time from the onset of the first display was
increased (either by increasing the first frame's
duration or by increasing the delay between the 2
frames), up 'until an SOA. of about 100-130 ms (where
performance asymptoted). In fact, if the duration of the
first stimulus exceeded 130 ms, subjects performed at
about chance level even if there was no delay between the
two frames.

There have been other studies which also seem to
suggest that visible persistence is the result of a
residual neural trace. For example, Phillips (1974) had
subjects make same-different judgments on stimuli created
by randomly filling cells in a square matrix. The first
matrix varied in complexity, and this was manipulated by
changing the size of the matrix (i.e., 4 x 4 matrices
were the smallest used and 8 x 8 matrices were the
largest employed). The first matrix was presented for 1
second, then there ‘was a variable duration. interval,

after which the second matrix was presented until the

subject responded same or different. It was found that
pattern complexity had little effect when the intermatrix
interval was short (i.e., up to about 100 ms). This
suggested the presence of a complexity-independent neural
process operating shortly after the offset of the first
stimulus.

Finally, Long and his coworkers have provided a
considerable amount of evidence that visual persistence
is mediated by the photoreceptors (i.e., persistence
results are due to peripheral neural mechanisms). For
example, Long and Beaton (1982) employed a partial report
task and found that increases in stimulus luminance
resulted in a longer lasting icon. You will recall from
the above discussion that this is directly opposite to
the typical pattern found in visible persistence
experiments (i.e., as stimulus luminance increases,
visible persistence decreases), and also different from
the results of other partial report experiments (e.g.,
Sperling, 1960; Adelson & Jonides, 1980). Furthermore,
Long and McCarthy (1982b) had subjects respond to the
last fading trace of a stimulus (i.e., the asynchrony
judgment or probe matching task) and found increasing
persistence duration with increases in stimulus
luminance. Thus, Long demonstrated that both visible
persistence and partial report tasks showed positive

effects of stimulus luminance, which is entirely

10

consistent with a photoreceptor or "neural" locus of
persistence (i.e., persistence can be thought of as a
photoreceptor afterimage that has resulted from bleaching
(saturating) the rods or cones).

Although the evidence described above supports the
_ view that visible persistence arises from residual neural
activity in the visual pathway, other investigators have
argued that visible persistence is really the
manifestation of ongoing visual information processing.
The reasoning behind this view is that the perceptual
system may maintain information long enough to allow for
various operations to be performed. According to this
view, then, visible persistence is not just a lingering
sensory trace of the stimulus, but is actually under
cognitive control. Several studies have been performed
in which complexity manipulations seemingly have had
effects on the duration of visible persistence (e.g.,
Avant, Lyman, & Antes, 1975; Avant & Lyman, 1975; Briggs
& Kinsbourne, 1972; Erwin, 1976). These studies have
indicated that more complex stimuli, which presumably
take longer to process, persist longer than less complex
stimuli. The existence of complexity effects on visible
persistence duration thus supports what might be called
the "process" view of visible persistence. New some of
these aforementioned studies shall be considered in more

detail.

ll

Avant, Lyman, and Antes (1975), using a two
alternative forced choice paradigm, showed an effect of
stimulus familiarity upon early visual processing. A
series of five experiments were performed in which two
variable length presentations of stimuli, separated by a
1 second interval, were compared. The task of the
subject was to say which of the two presentations was of
a longer duration. The types of stimuli used were
letters, words, and nonwords, and it was found that
nonwords were judged to persist longer than words, which,
in turn, were judged to persist longer than letters.
When a word was inverted, it was found to persist as long
as a nonword, thus suggesting that it was no longer a
familiar stimulus. These judged duration differences
were found to occur for stimulus durations ranging from
20-100 ms. Thus, these findings suggest that more
complex stimuli (i.e., nonwords) persist longer than less
complex stimuli (i.e., words or letters), and therefore
support a "process" model of visible persistence.
Complexity corresponds to familiarity in this study
(i.e., the less familiar, the more complex), and it can
be argued that the visual system tries to preserve
information that is unfamiliar to it, so that it can
perform more complete processing of its attributes.

The above findings led Avant and Lyman (1975) to

explore further the early visual processing of stimuli of

12

varying degrees of familiarity (complexity). The
previous experiments had shown that the apparent duration
differences were more substantial when a noise mask was
presented during the interframe interval. It was claimed
that the mask limited processing to its earliest stages
(i.e., before conscious recognition). This procedure
combined with their argument that more complex stimuli
require longer time periods to process led them to expect
a heightened effect of familiarity. In other words, the
fact that simple (or more familiar) stimuli are processed
faster leads to a shorter perceived persistence duration
than complex (i.e., less familiar) stimuli, and masking
serves to highlight these processing speed differences
since highly familiar stimuli are more fully analyzed
before the mask interferes. There is an assumption here
that all processing is curtailed when the mask appears,
however it has been argued extensively by many
researchers (e.g., Long, 1980) that this assumption is
not always true. Another interesting finding of Avant et
al. (1975) was that the greatest apparent duration
accompanied the least recognizable (i.e., to the subject)
stimulus. So the first experiment of Avant and Lyman
(1975) was an attempt to replicate the Avant et al.
(1975) noise mask procedure, and in fact, the findings

were quite comparable .

13

The second experiment attempted to increase
familiarity by presenting repetitions of the same
stimuli. It was found that the duration estimate
differences decreased as familiarity was induced.
Finally, experiment 3 compared duration judgments for
upright and inverted three letter words at various
exposure durations (i.e., 10, 20, and 30 ms). Judged
duration was found to be greater for the inverted words.

Therefore, Avant and his colleagues, using a novel
procedure, showed considerable support for the "process"
hypothesis. The short exposure durations and the
immediate masks that were employed make it conceivable
that they were studying the earliest stages of visual
processing. The number of experiments and the nature of
their manipulations at least make a complexity effect
tenable.

Another research project that showed evidence for an
effect of complexity on 'visual processing' is that of
Briggs and Kinsbourne (1972). An onset-offset technique
was used by these researchers in order to assess the
nature and duration of visible persistence. This
paradigm is based on Donder's method, and requires that
the mean of a series of reaction times to the onset of a
visual display be subtracted from the mean of a series of
reaction times to the offset of the display. It was also

necessary to subtract the exposure duration from

14

these reaction time values, since the stimulus durations
were not constant across blocks. These calculations led
to an estimate of persistence duration of the stimulus in
a particular condition, and these estimates were
determined for each subject. Catch trials were also
incorporated into the offset blocks to make sure subjects
were responding to stimulus offset. Three types of
stimulus presentations were used, namely letters
presented monoptically (i.e., presented to the dominant
eye), letters presented dichoptically (i.e., half of the
total stimulus presentation was given to each eye,
starting with the nondominant eye, on each trial), and
squares presented monoptically. Also, the duration of
presentation was varied (i.e., 100, 200, 300, 600, and
1000 ms).

The results showed that the persistence duration
estimates decreased as duration of the stimulus was
increased. It was also shown that dichoptic and monoptic
presentations showed the same results, thus suggesting a
central locus for these effects. More importantly,
squares were judged to persist for a shorter time than
letters at all exposure durations except 1000 ms. This
differential persistence trend is suggestive of a
complexity effect, although the differences were not
quite significant. This method, therefore, seems worthy

of further study, and in fact Erwin (1976; and Erwin &

15

Hershenson, 1974) has used this methodology in an attempt
to isolate the effects of stimulus complexity on visual
processing.

Erwin (1976) attempted to measure the phenomenal
(i.e., stimulus appears to be visible) and functional
(i.e., stimulus information is available for encoding)
durations of visual stimuli differing in informational
values, using a subtractive reaction time paradigm much
like that of Briggs and Kinsbourne (1972). The
complexity of stimuli was manipulated by using various
approximations to English (AE's). The more redundant the
letter string (7 letters in a row in these experiments),
the more predictable the array, and thus, the lower the
information content. A zero—order AE consists of 7
randomly sampled letters, and thus a letter at a certain
point in the array would have a 1/26 probability of being
a specific letter (e.g., a "B" in the second position).
Increases in redundancy (i.e., raising the approximation
to English), therefore make the stimuli more word-like.

Report of the presented array was required on some
trials, and this requirement. was found. to affect. the
reaction times to offset, but the reaction times to onset
were essentially unaffected. This result replicated
Erwin and Hershenson (1974). Random consonants, 0-, and

lst-order AE's persisted for a significantly longer

l6

duration when a report was obtained, than when a report
was not required, while 2nd—, and 3rd-order AE's did not
show this pattern of results. Thus, it was claimed that
"the phenomenal duration of a persisting representation
of a visual stimulus is a function of the amount of
encodable information in the stimulus if, and only if,
utilization of the target information is required"
(Erwin, 1976). Although Erwin claimed that the
information content of the stimulus does not matter when
encoding is not required, he stated that this variable
does have importance when more elaborative processing
must performed (i.e., the stimulus persistence is
maintained for a longer period of time so that more
processing operations can be performed).

The combined data from these experiments suggest
that complexity may, in fact, have some effect on the
duration of visible persistence, however, there are some
problems with the preceding studies. For example, most
of the positive complexity results have been found using
relatively subjective methods. More specifically, the
onset-offset technique relies upon judgments that are not
independently verifiable by the experimenter. The
forced-choice paradigm used by Avant and his colleagues
has a different problem, namely that it incorporates a
substantial memory component (i.e., there was a 1 second

delay between the 2 stimulus presentations). Thus, it is

l7

unclear whether Avant is studying visible persistence per
se.

There is eNen some evidence that both the "neural"
and "process" models can be combined to explain the
existing data. For example, Long and Wurst (1984)
performed two experiments (in which subjects depressed a
reaction time key when the stimulus had completely faded
from vieW) that attempted to test between the "neural"
and "process" models. Complexity was manipulated by
varying the number of interior angles in each polygon
shape, or by varying perimeterzlarea while maintaining
either perimeter or area constant. Long and Wurst found
an inverse complexity effect (i.e., estimated persistence
duration decreased as complexity was increased) when
solid black figures were used, but found a positive
complexity effect (i.e., supporting the ”process"
account) when the identical shapes were presented in
open, outline form. Thus, support was found for both the
"process" and "neural" models. These results also
suggest that stimulus attributes may conceivably have
played a role in determining which particular model was
correct in past experiments. It is, however, unclear
whether too much consideration should be given to these
experiments, since the task used is known to be highly

susceptible to criterion shifts.

 

18

What is needed, therefore, is an objective
methodology to determine whether the "process" model
(i.e., processing demands determine persistence duration)
or the "neural" model (i.e., persistence occurs due to
temporal sluggishness in the visual system) is the
correct explanation of visible persistence. Thus, the
purpose of the following experiments will be to use an
objective paradigm to assess the effect of complexity

manipulations on visible persistence.

EXPERIMENTS

General Methods
The first five experiments used the same basic
procedures, and these are illustrated in Figure 1. This
section will highlight the general procedures used, and
any deviations from this general framework will be noted

within the individual experimental method sections.

Subjects

Subjects were drawn from the Michigan state
University community, and included undergraduates,
graduate students, and faculty. Most of these subjects
had little or no knowledge of what was being studied.
Subjects were paid for their participation. All subjects
had normal or corrected-to-normal vision. Six to twelve
subjects were used per experiment, and in several cases
(which shall be noted), subjects participated in more

than one experiment.

Stimuli

Half of all trials were designated as "high-
complexity" trials and the other half were designated as
"low-complexity" trials. The overall configuration on a

trial was a 5 x 5 (or 4 x 4 in some of the later

19

20

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

W-conp';um mommnmum
srmuu WW"
0 o o c 'u w
a 0
"MN! I D . D u c n f
n a r u
a o a a
INTIRI'RANE E
mum» g
a a v
u a o . w c J
mm: 2 a a 0
o a a n
o o o l x x
.L
303..th arseowos wrm
consonants or mm mm»
(m was usr, now 4 coumw 4).
runner . .

 

 

 

 

 

 

Figure . General procedure used in experiments 1-3.

21

experiments) array, which. was exposed in two frames.
These stimuli were always viewed as a distance of 13
inches. The visual angles for these experiments varied
with stimulus types, as well as with array size. The
first three experiments attempted to equate stimuli for
relative brightness, by constructing individual stimulus
elements out of approximately the same number of light

points on the oscilloscopic display (16 on the average).

Equipment

 

Stimuli were displayed on a Hewlett Packard, 1340A
display oscilloscope, driven by a Digital Equipment
Corporation Micro-11/23+ computer.

Preceding each of our experiments, the equipment
(particularly the display oscilloscope) was allowed to
run for 2 hours to warm up. It was previously discovered
that this period of warm up is necessary to prevent
intensity fluctuations. Also, just before each subject
was run, a program was carried out that set the minimum
intensity to the researcher's threshold value, to provide
more consistency of intensities across subjects.
Finally, in all of these experiments, the experimental
chamber was kept illuminated, so as to prevent the

subjects from detecting phosphor persistence.

22

Procedures

 

The paradigm used was a temporal integration task
that. was first employed by IHogben. and DiLollo (1974;
elaborating upon an idea by Eriksen & Collins, 1967), and
later used by DiLollo (1980). On each experimental trial
(after viewing a fixation point for 500 ms.), 24 of 25
elements from a 5 x 5 array were presented in two frames
of time, 12 elements per frame. The subject's task was
to identify the location in the 5 x 5 matrix (in row and
column coordinates), at which no element had been
presented. The subject triggered each trial by pressing
the return key on a keyboard, and was allowed to proceed
at his/her own rate. After entering the (row, column)
response on the keyboard, the subject was given feedback
p+ or -) as to the correctness of his/her response, and
this feedback also served as the fixation point for the
next trial.

Figure 1 summarizes the procedure that was employed.
On each trial, subjects pressed the return key to
initiate the trial, which led to the display of a
fixation point; this was followed on a randomly
determined basis, by either a frame of "high-complexity"
stimuli or a frame of "low—complexity" stimuli for some
duration; then an interframe interval of variable
duration elapsed; then frame 2 was presented for some

duration (always containing the same stimulus type as was

23

presented in frame 1); and finally the subject entered

his/her response and received feedback.

Conditions

 

Subjects generally performed these experiments over
a time period of 2-4 days, depending upon the number of
trials. Each of these blocks was constructed so that
subjects received an equal number of trials from each
complexity type (high and low) by timing parameter
(either frame 1 duration or interframe interval)

combination.

Experiment 1
In order to test between ”neural" and "process"
accounts of visible persistence, we varied the complexity
of the stimulus elements to see whether more complex
stimuli could be integrated over longer time periods than
less complex stimuli. It is proposed that if persistence
differences are found (i.e., a complexity effect), the

"process" hypothesis will be supported.

Materials and Methods

Subjects. Six subjects with normal or corrected-to-
normal vision were run in this experiment. Each subject
was paid $3 for each of the 4 one—hour sessions. Each of

these sessions included a block from experiment 1 and a

24

block from experiment 2, and these blocks were run in a

counterbalanced order.

Stimuli. Half of the trials were designated as
"high-complexity" trials, and the stimuli in this case
were randomly—chosen letters (i.e., all letters were used
except a, e, i, o, u, and y). The overall configuration
on a trial was a 5 x 5 array, which was exposed in two
frames. At a 13 inch viewing distance, the array
subtended 3.1 degrees of visual angle horizontally and
3.6 degrees vertically. Each letter within the array
subtended .3 degrees horizontally and .4 degrees
vertically, while each space between letters subtended .4
degrees both vertically and horizontally.

The other half of the trials were designated "low-
complexity" trials, and the array elements were always
rectangular shapes in these cases. These rectangles took
up the same amount of visual angle as the letters and
were constructed using 16 light points (i.e., the average
number used for constructing letters). Random letters
were assumed to be more complex than rectangles both
cognitively and featurally. Since this experiment was
exploratory, we decided to use a rather gross measure of
complexity, keeping in mind that there were many

confounds we could consider later if a complexity effect

was found.

25

Equipment. See General Methods.

Procedures. See General Methods.

 

Conditions. Frame 1 and frame 2 both had durations

 

of 20 ms. Interframe intervals were varied randomly from
trial to trial. The values used were: 10, 30, 50, 70,
90, and 120 ms. Thus, this experiment was concerned with
how interframe intervals (i.e., the time from the offset
of frame 1 to the onset of frame 2) affect temporal
integration with stimuli of varying degrees of
complexity. Each subject completed 1200 trials over a 4—
day period. There were thus 100trials for each of the
two complexity levels at each of' the six interframe
intervals. All of the conditions were randomly
intermixed, such that each subject received 25 trials in

each condition per session.

Results and Discussion

The results from experiment 1 are shown in Figure 2.
All main effects were significant: block, F(3,15) =
4.25, p . .0232; stimulus type, F(l,5) = 8.10, p s .0360;
and delay, F(5,25) -= 59.97, p < .0001. The block x
delay, Fw15,75) .. 1.87, p . .0395; and stimulus type x
delay, F(5,25) - 2.67, p - .0459, interactions were also
found to be significant. Due to the significance of the

interactions, the main effects shall not be considered in

26

1 00‘

90‘

XF—fix Rectangles

0---o Random Letters

80‘

70~

60

50

40‘

30‘

IPIZIQCZISI¢1P (:(3121213C21P

 

 

 

o T W S ‘J I 1

0 1 0 30 50 7o 90 150

I NTERFRAME :I: NTERVAL. ( MS EC )

Figure 2. Percent correct for rectangles and random
letters as a function of interframe interval

in experiment 1.

27

further detail. Post-hoc comparisons on. the block x
delay interaction (Scheffe 95% confidence interval
halfwidth - 17.38) showed that accuracy decreased as the
delay was lengthened, but blocks 1-4 did not differ at
any of the delays in any systematic manner.

The stimulus type x delay interaction was also
examined with planned comparisons (Bonferroni 95%
confidence interval halfwidth - 6.38). Table 1 shows the
mean and individual results for the ‘typestini x. delay
interaction. Within the rectangle condition, each
increase in delay resulted in a significant decrement in
accuracy through the 90 ms delay. For the letter
condition, accuracy progressively decreased throughout
the 120 ms delay, at which point it appeared to asymptote
at a low level. All of the delays in the letter
condition showed significant differences from each other,
except that 70 ms was not significantly different from 90
ms. In sum, performance with random letters was
significantly better than with rectangles overall, and
the rectangles reached asymptote at an earlier interframe
interval than did the random letters. Thus, random
letters were found to persist longer than rectangles in
experiment 1. It must be noted, however, that these
differences are small at each of the delays. In fact, at

no delay is the difference between random letters and

28

Table 1. Individual and mean results (% accuracy) from
experiment 1, as a function of interframe

 

 

 

 

 

 

 

interval
Stimuli: Rectangles
Interframe Interval
Subject # 10 30 50 70 90 120
1 97 63 20 19 08 12
2 96 91 64 36 19 24
3 78 33 18 18 08 ll
4 83 50 22 15 13 14
5 77 71 49.45 38 25 10.85
6 95 66 20 19 07 08
MEAN 87.67 62.33 32.24 24.17 13.33 13.31
Stimuli: Random Letters
1 94 74 25 09 12 07
2 96 94 71 41 - 33 16
3 72 40 17 ll 14 11
4 89 46 27 15 ll 16
5 79 74 55.50 39.25 26.13 16
6 89 69 29 17 15 08

MEAN 86.50 66.17 37.42 22.04 18.52 12.33

 

29

rectangles significant. These findings are strong enough
to promote further experimentation, however.

To summarize the results of experiment 1: (1)
performance deteriorated as the interframe interval (or
delay) was increased, for both types of stimuli; (2)
performance for random letters was superior to that for
rectangles; and (3) rectangles reached an asymptote
earlier than random letters (i.e., the letters persisted
longer).

These findings provide support for the "process"
model of visible persistence, since the random letters
were found to persist longer than the rectangles. The
fact that the methodology used for these experiments was
more objective than previous complexity studies
strengthens the support for the ”process" view. .As
described earlier, the stimuli that were used were
equated for brightness (i.e., the number of light points
used to construct them), so it is unlikely that
differences in retinal excitation accounted for the
results. This set of findings is in line with the
results of other researchers (e.g., Erwin, 1976; Briggs
and Kinsbourne, 1972). As expected, performance got
worse the longer the interval between the two frames.
This finding replicated the results of several
researchers using variants of this technique (e.g.,

Eriksen 8. Collins, 1967; DiLollo, 1980). There was one

3O

concern with our findings, however, and that was the
small absolute size of the differences. It was,

therefore, necessary to perform additional experiments.

Experiment 2

In experiment 2, we examined the complexity question
in a slightly different way. The interval between two
frames can be varied in alternative ways. The purpose of
this experiment was to rerun experiment 1, varying the
stimulus onset asynchrony (i.e., the time from the onset
of frame 1 to the onset of frame 2) to see what effect
this manipulation has on persistence with stimuli of
varying degrees of complexity. This experiment is not
without precedent, since DiLollo (1980) has suggested
(and provided some evidence showing) that visible
persistence duration is actually dependent on stimulus
onset asynchrony (SOA) rather than interframe interval
(IFI). If complexity affects visible persistence
duration, we ‘would expect high complexity stimuli to

persist over longer frame 1 durations.

Materials and Methods
Subjects. The same six subjects served in this
experiment, as both of these studies were run

concurrently, in a counterbalanced order.

31

Stimuli. The same stimuli were used in this
experiment as were used in experiment 1.

Equipment. See General Methods.

 

Procedures. See General Methods.

 

Conditions. In this experiment, frame 2 and the

 

interframe interval were maintained constant at 20 ms.
The duration of the first frame ‘was varied, and the
values of this variable were: 20, 40, 60, 80, 100, and
150 ms. All of the conditions were randomly intermixed,
such that each subject received 25 trials in each

condition per session.

Results and Discussion

The results of this experiment are shown in Figure
3. All main effects were significant: block, F(3,15) -
15.04, p - .0001; stimulus type, F(1,5) - 13.73, p =
.0139; and duration, F(5,25) =- 35.55, p < .0001. No
interactions were found to be significant in this
experiment, however, as can be seen from the following F
values and probabilities: block x stimulus type, F(3,15)
= .25, p - .8624; block x duration, F(15,75) = .99, p =
.4724; and block x stimulus type x duration, F(15,75) =
1.36, p - .1925.

Follow—up analyses were, therefore, performed on the
main effects. For the block term, block 1 was found to

have significantly lower accuracy than blocks 2, 3, and 4

32

100‘
H ' Rectangles

o— —-0 Random Letters

40‘

 

30-

20‘

PERCENT CORRECT

10‘

 

 

o - 4 1 1 4 .
0 20 40 60 80 100 150

FRAME ONE DURATION (MSEC)

Figure 3. Percent correct for rectangles and random
letters as a function of frame one duration

in experiment 2.

33

(Scheffe 95% confidence interval halfwidth = 7.57),
although these last three blocks were not significantly
different from one another. For duration, 20 ms was
found to have significantly higher accuracy than the 60-
150 ms durations; 40 ms was significantly different from
80-150 ms; 60 ms was significantly different from 100-150
ms; 80 ms was significantly different from 150 ms; and
finally 100 ms was significantly different from 150 ms
(Bonferroni 95% confidence interval halfwidth - 10.61).
For stimulus types, letters resulted in significantly
higher accuracy than rectangles. Although the stimulus
type x duration interaction was not significant, F(5,25)
. .51, p - .7665; planned comparisons were performed to
ascertain whether the letters persisted for a longer
duration than the rectangles (Bonferroni halfwidth a
8.03), and no significant differences were found to be
present. Mean and individual results for this
interaction can be seen in Table 2.

In sum, performance was found to decline as the
frame 1 duration was increased for both types of stimuli.
Performance was found to improve slightly as the subject
became more practiced. Finally, performance with random
letters was found to be better than performance with
rectangles, however, there were no significant
interactions. It seems instead that performance with

random letters was better than performance with

34

Table 2. Individual and mean results (% accuracy) from
experiment 2, as a function of frame 1 duration

 

 

Stimuli: Rectangles

 

Frame One Duration

 

 

Subject # 20 40 60 80 100 150
1 96 75 65 64 52 37
2 89 89 88 90 82 62
3 56 41 34 35 24 28
4 62 53 44 38 39 32
5 83.15 70.15 71.70 59.45 60.23 43
6 75 68 56 47 37 24

MEAN 76.86 66.03 59.78 55.58 49.04 37.67

 

Stimuli: Random Letters

 

1 90 87 75 68 57 29
2 94 90 92 88 77 73
3 66 54 46 35 39 26
4 77 54 59 37 42 31
5 78 73 67.48 71 62 42.45
6 71 67 61 51 41 35

MEAN 79.33 70.83 66.75 58.33 53.00 39.41

 

35

rectangles at each frame 1 duration, and this suggests
that there may be another process occurring. What might
explain why random letters were successfully integrated
significantly more often, but were not found to persist
longer? As stated earlier, we did control for the number
of light points in the display, but there may be some
alternative explanation. Therefore, this experiment does
not provide conclusive support for the "process" view,
since only accuracy, and not persistence duration was
higher for random letters.

In fact, the pattern. of' results does suggest an
alternative explanation for our findings. Recently,
DiLollo and Hogben (1985) have suggested that visible
persistence duration is affected by suppressive
interactions among array elements that are close together
in space and/or time. Their experiments consisted of
displays in which a point stepped around a circular path
on an oscilloscope. Observers estimated the number of
points seen simultaneously. Several variables were
manipulated including spatial proximity of successive
points and eccentricity in the visual field. It was
found that the duration of visible persistence was highly
dependent on interpoint separation, at least up to
distances of about .5 degrees. Also, both visible
persistence duration and the degree of suppression were
found to increase with retinal eccentricity. These

suppressive forces, therefore, act somewhat like the

36

metacontrast effects that are seen throughout the visual
masking literature. It was suggested by DiLollo and
Hogben (1985; also Hogben & DiLollo, 1985) that
suppression is an active inhibitory force that serves to
reduce the smear that arises from objects in motion.

Thus, before we can conclude that more complex
stimuli persist longer than less complex stimuli, the
possibility of suppressive interactions must be ruled
out, especially since our stimuli were separated by
distances of less than .5 degrees in several cases. In
other words, it could be that the results of the first
two experiments were obtained due to intercontour
suppression, since the contours of the rectangles are
closer together than are the contours of the random

letters.

Experiment 3
The purpose of experiment 3 was to compare random
letters with stimuli which. had intercontour distances
more like those of the random letters. The suppression
explanation predicts that with intercontour distance

roughly equated, no complexity effect should be found.

Materials and Methods
Subjects. Six subjects were used in this
experiment. Two of these subjects had taken part in

experiments 1 and 2. Subjects all had normal or

37

corrected-to-normal vision, and were each paid $1.50 for

each of 4 half-hour sessions.

Stimuli. The same stimuli were employed, except
that X's were used instead of rectangles. These X's were
composed of 16 light points just as the previous stimuli
had been (i.e., controlled for retinal excitation).
Thus, the low complexity stimuli were X's in this case,
and the high complexity stimuli were once again the

random letters.

Equipment. See General Methods.

 

Procedures. See General Methods.

 

Conditions. As in experiment 1, frame 1 and frame 2

 

had durations of 20 ms. Interframe intervals were varied
randomly on each trial. The values used were: 10, 30,
50, 70, 90, and 120 ms. Again, each subject received
1200 trials over the four sessions, comprised of 100
trials at each of the two complexity levels at each of

the six interframe intervals.

Results and Discussion

The results for experiment 3 are shown in Figure 4.
Block, F(3,15) = 7.88, p = .0022; and delay, F(5,25) =
37.14, p < .0001 were significant, but stimulus type,

F(1,5) = .10, p = .7654 was far from significant. No

1 00
90
50
70-
60-
50
40«
30

I
20

I
10‘

IPIZIQCZIEIW1P (ZCDIQIRIZCZWP

 

O a

38

H X-Array

.— — ... Random Letters

 

JCII1PIZIEIrlilkhdlB JEIUUPIZIIKZZKI. {ED453IBCZI’

Figure 4.

Percent correct for X-arrays and random
letters as a function of interframe

.interval in experiment 3.

39

interactions were found to be significant, as can be seen
by the following F values and probabilities: block x
stimulus type, F(3,125) - .31, p - .8192; block x delay,
. F(15,75) = 1.09, p - .3677; stimulus type x delay;F(5,25)
= .74, p --= .6036; and block x stimulus type x delay,
F(15,65) a .69, p = .7873.

Post-hoc comparisons were performed on the block
term, and it was found that block 1 showed significantly
lower accuracy than both blocks 3 and. 4. :No other
comparisons among blocks were significant (Scheffe
halfwidth = 4.49). Planned comparisons on the delay
condition showed that 10 ms differed significantly from
30 ms which differed significantly from all of the other
delays (i.e., 50,70, 90, and 120 ms), however these
delays did not differ significantly from. one another
(Bonferroni halfwidth - 20.80).

In summary, performance deteriorated as the
interframe interval was increased. Again, there was
slight evidence of a practice effect, since block 1 was
inferior to the other blocks. Finally, there was no
difference in performance from the X's and random
letters. Also, both types of stimuli reached asymptote
at the same point, so neither stimulus type persisted
longer than the other. Mean and individual results for

the typestim x delay interaction can be seen in Table 3.

40

Table 3. Individual and mean results (% accuracy) from
experiment 3, as a function of interframe

 

 

 

 

 

 

 

interval
Stimuli: X-Array
Interframe Interval
Subject # 10 30 50 7O 90 120
1 98 96 79 40 28 18
2 59 35 l3 13 11 12
3 94 83 41 22 20 16
4 94 56 27 ll 10 10
5 74 42 22 11 17 09
6 81 23 10 15 10 09
MEAN 83.33 55.83 32.00 18.67 16.00 12.33
Stimuli: Random Letters
1 98 90 73 43 22 30
2 58 26 . 25 12 16 ' 09
3 94 84 45 21 15 12
4 92 59 28 13 08 13
5 80 44 22 15 10 07
6 74 22 16 ll 13 12

MEAN 82.67 54.17 34.83 19.17 14.00 13.83

 

41

Thus experiment 3 suggests that stimulus complexity
is not affecting the duration of visible persistence.
Furthermore, the results of the first two experiments,
which showed some evidence of a complexity effect, were
shown to probably be due to an artifact, namely the
operation of suppressive forces caused by the proximity
of array elements. These results provide support for the
"neural" hypothesis (i.e., visible jpersistence results
from residual neural activity in the visual pathway and
is unaffected by variations of stimulus complexity)
rather than the "process" hypothesis (i.e., stimulus
complexity affects the duration of visible persistence).

It is also possible, however, that we have simply
not manipulated complexity effectively. The following
two experiments will manipulate complexity in a different
way, in the hope of generating some type of complexity
finding. 0n the other hand, if no complexity effects are
found, then substantial credence will be given to the

"neural" hypothesis of visible persistence.

Experiment 4
In this experiment, instead of varying the
complexity of the array elements, we varied the
complexity of the overall pattern formed by the elements.

In order to do this, we used a set of stimuli which had

42

been independently rated for complexity. These ratings
were reported in a recent study by Ichikawa (1985).
Ichikawa's subjects rated the complexity of various
arrangements of 8 dots in 4 x 4 ihatrices. In our
experiment, we used some of his rated stimuli as first
frames in our temporal integration procedure (details to
follOW). If’ stimulus complexity affects visible
persistence duration, then subjects should be able to
integrate the high complexity patterns over longer

interframe intervals than the low complexity patterns.

Materials and Methods

Subjects. Twelve subjects with normal or corrected-
to—normal vision were run in this experiment. All but
three of these subjects had participated in at least one
of the previous experiments. Each subject was paid $3

for each of 2 one-hour sessions.

Stimuli. The stimuli used for this experiment were
similar to those used previously, but they also differed
in several respects, which should be elaborated upon. We
used 40 patterns which had been rated low in complexity
in the Ichikawa (1985) study (i.e., pattern numbers 26-
65), and 40 patterns which had been rated high in
complexity (i.e., pattern numbers 101-140). The complete
set of stimuli is illustrated in Appendix A. Samples of

these patterns are illustrated in Figures 5 and 6,

43

_ .
u _ O u 0 u

. .
w---4---q---l---n
_ . _ .

_ . _

_ o _ o _ _ _
_ . _ _ _
......1..5..+...:4...i
_ o . o _ _ _
_ _ _ _ _

rIILIIIFIILIIL

......IJI ......IIIJIIIJ
.
_

Sample of low complexity stimuli

1985) used in experiments

(Ichikawa,
4 and 5.

Pi ure 5.

44

TIIJIIIJIIIAIIIJ
_ _ _ _

_
O _ _

m... _ . I I

_ ..."..IIJIIIIWII...

. o _ o _ . o _

. _ _ u _

.,-.---4---.--l

_ _ . o _ _

TIILIII_IIIF _
_ . III.

_

.IIIFIIILIIILIIIL

Sample of high complexity stimuli

(Ichikawa, 1985) used in experiments

4 and 5.

Fi ure 6.

45

respectively. The first frame consisted of one of the
Ichikawa 8-dot patterns, either a simple or a complex
one, and the second frame was formed by randomly filling
in 7 of the 8 remaining locations. Thus, the two frames
formed 4 x 4 arrays as opposed to the 5 x 5 arrays used
in the preceding experiments. The stimuli in this
experiment were solitary dots at each location, as
opposed to rectangles or letters. The entire array of
dots subtended 2.45 degrees of visual angle horizontally
and 2.72 degrees of visual angle vertically. Each dot
occupied .04 degrees of ‘visual angle, and the Spaces
between dots occupied .79 degrees of visual angle

horizontally and .88 degrees of visual angle vertically.
Equipment. See General Methods.

Procedures. The procedures used in this experiment
were for the most part identical to those used in the
previous experiments, except that 4 x 4 arrays were

employed. Again, the task was to identify the unoccupied

location.

Conditions. Both frames 1 and 2 were 10 ms in
duration. The interframe interval was varied, but

different values were used in this study, namely 10, 30,

46

50, 70, 85, and 100 ms (i.e., the longer intervals were
excluded since performance dropped so low in the previous
experiments). Each subject received 40 trials in each of
the 2 complexity type by 6 .interframe interval
conditions. Although this experiment was run over two
days, it was not broken into two blocks, since the two
sessions were not identical in their trial composition
(however, every subject did receive each of the patterns
once for each of the six interframe intervals by the end

of the experiment).

Results and Discussion

The results for experiment 4 are shown in Figure 7.
The main effect for delay was significant, F(5,55) =
52.62, p < .0001, but stimulus type was not significant,
F(1,11) - .33, p - .575. No interactions were found to
be significant, since the stimulus type x delay term
resulted in the following values: F(5,55) = .81, p =
.5442.

Planned comparisons were performed on the delay
condition (Bonferroni halfwidth = 15.91). Delays of 10,
30, and 50 ms were found to differ from 70, 85, and 100
ms significantly, but not from each other. Also, 70 as
differed significantly from 100 ms.

In summary, performance deteriorated as the

interframe interval increased, but there was no effect of

1 00 -
90-
80-
70-
60.

sol

40J

30'

20'

EPISIQCZIBIW1P (ZCDIQIRIBCZ1F

104

 

47

 

H Simple Ichikawa Stimuli

o—--o Complex Ichikawa Stimuli

 

I V ‘

0 1 0 30 50 70 8'5 160

JEI'fiPIBIZI’IiIthIB JEI¢1P1312\IZ\IE {CDJESIBCZI’

Figure

7. Percent correct for simple and complex
Ichikawa stimuli as a function of inter-
frame interval in experiment 4.

48

complexity on overall accuracy (or persistence duration,
since neither curve dropped to asymptote before the
other). This can be seen by looking at the mean and
individual results for the typestim x delay interaction,
which are given in Tables 4a and 4b. Thus, this
experiment also suggests that the "neural" model, rather
than the "process" model of visible persistence, is
correct.

There may be a problem with experiment 4 though.
The results suggest that no complexity effect is
operating, however, the accuracy of subjects was quite
high, even at the 100 ms delay (probably due to the fact
that 4 x 4 (smaller) arrays were used).

This raises an important issue: How might one
determine the level of subj ect's performance expected if
persistence was not operating (e.g., at very long
delays)? In other words, what is the best performance
that a subject could achieve relying solely on chance
performance? In the case of a 5 x 5 array, the best
performance allowable (at long delays) by chance is
probably about 16.7%. This is determined in the
following way: When the first frame of 12 elements is
presented, some of these elements may enter into short-
term memory; we will assume that 7 (see Miller, 1956) are
held in short—term memory. When the second frame is

presented, all 12 elements 'will be available. Thus, at

49

 

 

 

 

 

Table 4a. Individual and mean results (% accuracy) from
experiment 4, with simple Ichikawa patterns,
as a function of interframe interval ‘

Stimuli: Simple Ichikawa Patterns
Interframe Interval
Subject # 10 30 50 70 85 100
1 100 95 92.5 57.5 40 25
2 97.5 97.5 97.5 82.5 77.5 82.5
3 95 100 97.5 100 95 90
4 92.5 95 82 5 72.5 50 47.5
5 82.5 90 77.5 70 47.5 55
6 100 92.5 72.5 50 40 42.5
7 92.5 92.5 92.5 67.5 75 67.5
8 90 90 78.5 50.0 50.0 27.5
9 95 90 77.5 67.5 47.5 45
10 95 87.5 82.5 77.5 60 45
11 82.5 90 67.5 50 45 27.5
12 90 87.5 80 47.5 52.5 65
MEAN 92.7 92.3 83.8 66.0 56.7

51.7

 

50

 

 

 

 

 

Table 4b. Individual and mean results (% accuracy) from
experiment 4, with complex Ichikawa patterns,
as a function of interframe interval

Stimuli: Complex Ichikawa Patterns
Interframe Interval
Subject # 10 30 50 70 85 100
1 97.5 100 92.5 67.5 30 35
2 95 97.5 97.5 92.5 85 65
3 95 100 97.5 97.5 90 85
4 95 97.5 87.5 80 65 52.5
5 97.5 92.5 87.5 67.5 35 45
6 95 100 82.5 55 45 57.5
7 90 85 87.5 82.5 60 6O
8 87.5 95 75 52.5 37.5 45
9 92.5 92.5 95 72.5 62.5 47.5
10 100 95 87.5 67.5 35 37.5
11 87.5 85 60 42.5 52.5 37.5
12 82.5 80 80 60 45 52.5
MEAN 92.9 93.3 85.8 69.8 53.5 51.7

 

51

best (with long delays), the subject should have 7
elements from the first frame (in STM), and 12 elements
from the second frame on which to base his/her answer.
Therefore, chances become 1 out of 6 (i.e., 16.7%) since
there are 25 total locations.

Similar computations can be performed for the 4 x 4
array, and chance performance turns out to be about 50%,
since the subject has 7 elements from frame 1 available
in short-term memory, and 7 elements from frame 2 still
visible. There are 16 possible locations in this case.

Inspection of Figures 2, 3, 4, and 7 shows that
these are approximately the baselines reached when long
interframe intervals are considered. Thus, it seems
likely that we used long enough delays.

However, it still seems possible that we might have
found a complexity effect in experiment 4 if we had
looked at longer interframe intervals, as predicted by
the "process" hypothesis. Experiment 5 considered this

possibility.

Experiment 5
The present experiment is an attempt to replicate
and extend the results of experiment 4. Interframe
intervals were varied over a wider range, and fewer

subjects with more trials per subject were run (i.e., to

52

decrease the high between-subjects variance that occurred

in experiment 4).

Materials and Methods

 

Subjects. Four subjects (who had participated in at

 

least one of the pmevious experimentS) were run in this
experiment. Subjects were paid $3 for each of 2 one-hour

sessions.

Stimuli. The same stimuli that had been employed in

experiment 4 were used in this experiment.

Equipment. See General Methods.

 

Procedures. See General Methods.

 

Conditions. Frames 1 and 2 were again 10 ms in

 

duration. The interframe intervals that were used were
10, 40, 70, 100, 130, and 160 ms. Also, each subject
received 80 trials in each of the 2 complexity level by 6
interframe interval conditions, and these 80 trials were

split into 40 trials per session.

Results and Discussion

The results for experiment 5 are shown in Figure 8.

Only the main effect of delay was significant, F(5,15)

26.95, p < .0001. The main effects of block, F(1,3)

.03, p = .8806; and stimulus type, F(1,3) = 1.19, p =

.3546 did not approach. significance. 1k) interactions

53

1 00‘
‘90.
80«
704
604

50-

 

40-

}?--—X’ Simple Ichikawa Stimuli

30-

20. r--o Complex Ichikawa Stimuli

I’EBIZCZIBI'1P (ZCJIQIIIBCZUF

10-

 

 

0

fi 1 fl

0 1'0 40 70 1 00 130 ' 160

I NTERFRAME I NTERVAL ( MSEc )

Figure . Percent correct for simple and complex
Ichikawa stimuli as a function of extended
interframe intervals in experiment 5.

54

were found to be significant either, as can be seen by
the following F values and probabilities: block x
stimulus type, F(1,3) - 2.70, p - .1987; block x delay,
F(5,15) - .38, p - .8559; stimulus type x delay, F(5,15)
= .16, p - .9725; and block x stimulus type x delay,
F(5,15) - .69, p - .642.

Planned comparisons were then performed on the delay
data (Bonferroni halfwidth - 21.32). 10 ms was found to
differ significantly from 70, 100, 130, and 160 ms; 40 ms
differed significantly from 100, 130, and 160 ms; 70 ms
differed significantly from 130 and 160 ms; and finally
100, 130, and 160 ms did not differ significantly from
one another.

In summary, performance deteriorated as the
interframe interval increased, but once again, there was
no effect of stimulus complexity on accuracy or
persistence duration. The mean and individual results
for the typestim x delay interaction are shown in Table
5. In fact, there was some suggestion that performance
on the simple patterns was better than performance with
the complex patterns. It is conceivable that the simple
patterns aided performance due to their gestalt-like
quality; with simple patterns, subjects may have been
able to use short-term memory to remember some dot
locations, rather than having to rely just on visible

persistence. This is not without precedent, since

55

 

 

 

 

 

 

 

Table 5. Individual and mean results (% accuracy) from
experiment 5, as a function of interframe
interval

Stimuli: Simple Ichikawa Patterns
Interframe Interval
Subject # 10 40 70 100 130 160
1 98.75 98.75 88.75 76.25 67.50 57.50
2 98.75 100 96.25 85 63.75 57.50
3 97.50 90 52.50 36.25 32.50 38.75
4 91.25 87.50 62.50 46.25 32.50 27.50
MEAN 96.56 94.06 75.00 60.94 49.06 45.31
Stimuli: Complex Ichikawa Patterns
1 97.50 97.50 85 67.50 55 50
2 98.75 98.75 98.75 85 60 63.75
3 95 91.25 55 38.75 43.75 32.50
4 93.75 85 51.25 41.25 28.75 23.75
MEAN 96.25 93.13 72.50 58.13 46.88

42.50

 

56

Phillips (1974) showed evidence for persistence in short—
term visual memory for durations up to 9 seconds, using a
same-different judgment and square matrices with randomly
filled cells. Furthermore, performance was better for
less complex (as defined by the size of the square arrays
[i.e., number of cells]) arrays than for more complex
arrays when the interval between the two patterns
exceeded about 100 ms. Thus, it is important that one'
recognizes the possibility of short-term memory
influencing performance on these types of visual tasks.
Regardless of these cautions, there was no support found
for the "process” hypothesis of visible persistence.
Also, there was no indication of a practice effect in
this experiment, which is not really surprising since
experienced subjects were run.

Thus far, most of the evidence we have found has
suggested that complexity does not affect visible
persistence duration. Some researchers have shown
effective complexity manipulations however (e.g., Erwin,
1976), so our next step was to try to discover the reason
for the discrepant findings. The following experiments
were performed with two goals in mind: (1) to determine
the effectiveness of our complexity manipulations; and
(2) to investigate the effect of forced identification on
performance. It is possible that the preceding

57

experiments have not manipulated complexity effectively.
Our failures to find a complexity effect may simply be
due to the fact that we have not changed or measured the
correct attributes of the stimuli. Thus, it is important
that we demonstrate a complexity effect with our stimuli,
using' a different (perhaps more subjective) paradigm.
The paradigm that we employed, namely the onset-offset
subtractive reaction time technique, has been used by
Erwin (1976) to provide some of the most compelling
positive complexity effects that have been found. Thus,
the overall aim of the next set of experiments was to
demonstrate a complexity effect with a subjective
technique, and then try to find a complexity effect using
the same stimuli with our objective temporal integration
paradigm. Our second goal, investigating the effect of
forced identification, will be elaborated upon later.

Experiment 6

Experiments 4 and 5 found no complexity effect
(using a temporal integration technique), with stimuli
that had previously been rated as varying in complexity
(Ichikawa, 1985). The technique used was objective since
it required visible traces from both frames to be used.
Erwin's (1976) task is more subjective, so if we can find
a complexity effect using the Ichikawa stimuli, then it
is quite likely that our objective technique will point

58

out the failings of the onset—offset reaction time
paradigm and show why a complexity effect is obtained
when Erwin's task is employed. Thus, our prediction is
that the onset-offset technique will show a complexity
effect with the Ichikawa (1985) stimuli.

Materials and Methods

 

Subjects. Seven subjects were used in this
experiment . All subjects had normal or corrected—to-
normal vision . Subjects were paid $5 for one session

which lasted about 1 1/2 hours.

Stimuli. The stimuli employed were the same dot
patterns that were used in experiments 4 and 5, although
only the first frame was used. In other words, on each
trial one of the Ichikawa patterns was presented, but the
other 7 of 8 locations in the 4 x 4 matrix were not
filled in by a second frame. The dots and the arrays
subtended the same visual angle as in the previous
experiments (i.e., 4 and 5), since the same viewing

distance (13") was used.

Equipment. The same equipment was employed in this

experiment.

Procedures. Subjects initiated each trial by
pressing the return key on a computer keyboard. A

fixation point was then exposed for a period of 500 as.

59

There was then. a variable length delay (500-1300 m5)
between the offset of the fixation point and the onset of
the stimulus. Stimuli were presented for 10 ms (to allow
more accurate comparisons with experiments 4 and 5) for
regular trials, and 500 ms on catch trials. Subjects were
given the task of indicating the onset of stimuli on some
blocks, and the offset of stimuli on other blocks. In
each offset block, one third of the trials were designated
as catch trials to insure that subjects really were
responding to the offset of the stimulus. On half of all
trials, the subjects were asked to write out, on a
response sheet, a full report of the stimuli observed, and
this factor was also blocked. This report served the same
function that it did in the Erwin (1976) experiment,
namely that it forced the subjects to process the stimuli
to a greater degree. Thus, there were four types of
trials: (a) respond to onset with no report; (b) respond
to onset with a report; (c) respond to offset with no
report; and (d) respond to offset with a report. Figure 9

shows the general procedure that was employed.

Conditions. There were two levels of complexity in
this experiment (i.e., the low complexity Ichikawa stimuli
and the high complexity Ichikawa stimuli). There were two
levels of report (i.e., no report and full report). In

addition , we incorporated 5 levels of delay

 

Fixation

Point _+_

60

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Delay l J
cl .0
I
Stimulus
O O
O { C
Figure 9. General procedure used for the onset-offset

eXperiments (i.e., 6, 7, and 8). Subjects
responded to the onset of the stimulus on
half of the blocks, and to the offset of the
stimulus for the other half of the blocks.

 

sauuier

61

between the fixation spot and the stimulus. The purpose
of this manipulation was to prevent subjects from
anticipating the onset of the stimulus. The five levels
used were: 500, 700, 900, 1100, and 1300 ms, however,
the catch trials only incorporated the 500 and 700 ms
delays due to a programming error. Finally, two types of
reaction times (RT's) were collected: RT's to onset and
RT's to offset. For each subject, an average RT was
calculated for each of these two measures, and the RT
(avg) to onset was subtracted from the RT (avg) to
offset. It was not necessary to subtract out stimulus
duration, since the program running the experiment did
this while collecting the RT's. This resulted in an
estimate of persistence duration (see Briggs &
Kinsbourne, 1972; Erwin, 1976). These persistence
duration estimates were determined for each of the two
complexity levels at each of the two report levels.
Subjects completed four blocks of experimental
trials. A counterbalanced order of blocks was randomly
assigned to each subject. Onset blocks consisted of 80
trials, while offset blocks consisted of 120 trials
(i.e., there were 40 catch trials). Before each block,
subjects were run in a practice session that consisted of
20 trials for onset and 30 trials for offset (i.e., 10

catch trials included). Before beginning each practice

62

session and accompanying block, subjects were advised as
to whether they should respond to the onset or the offset
of the stimulus, and also whether or not they would have
to give a report on each trial. Each of the stimuli
(i.e., 40 simple and 40 complex) were randomly intermixed
to construct the experimental trials for both onset and
offset blocks. On the catch trials (for the offset
blocks), the stimuli were randomly created on each trial,
so that complexity would have no systematic effect.
Subjects were not advised of the complexity manipulation

that had been implemented on the experimental trials.

Results and Discussion

The results from experiment 6 are shown in Figure 10.
There were no significant main effects or interactions in
this experiment: condition, F(1,6) . .17, p - .6906;
report, F(1,6) - .07, ‘p - .8053; condition. x delay,
F(4,24) - .34, p - .8511; however, some of the analyses
showed marginally significant results: delay F(4,24) =
2.14, p - .1072; report x delay, F(4,24) - 2.26; p =
.0925; condition x report, F(1,6) - 2.31, p - .1797; and
condition x report x delay, F(4,24) - 2.21, p - .0979.

Although none of the above analyses showed
significance, further analyses were done on several of the

effects and interactions that showed marginal

63

 

 

 

 

 

 

 

 

 

 

A

U

5...:

Z

V

z 1401

O

H

b 120.. ‘ . +I‘"'_—I’

g l r— ~—

8 l F—I-I-w
100«

Id

U

5

P

m

0: 51:91.8 COMPLEX 51:41:11.5: COMPLEX

2 no nsporrr no REPORT REPORT REPORT

CZCDIUIDHI'I'JUCDIU

Figure 10. Onset-offset persistence duration as a
function of complexity (simple versus
complex Ichikawa patterns) and report
conditions for experiment 6.

 

64

significance. Post—hoc comparison on delay (Scheffe
halfwidth = 31.46) showed no significant differences.
Furthermore, planned comparisons on the condition x
report interaction (Bonferroni halfwidth . 14.71) and
post-hoc comparisons on the condition x report x delay
interaction (Scheffe halfwidth = 38.02) showed no
differences to be significant. Finally, a t-test for
correlated means was performed on the two types of
stimuli (i.e., condition) and it was found that the
difference was not significant (t(6) - .418, n.s.).

Figure 10 illustrates that performance was virtually
identical in each of the condition x report levels when
collapsed across subjects, however, the results for
individual subjects show that at least two different
styles of responding were present. These results are
shown in Table 6. Subjects 3 and 5 showed much shorter
persistence durations in both of the report conditions,
suggesting that they may have sped up their responses so
as to improve their reports. The other five subjects
showed either flat results or a tendency for persistence
duration to increase when a report was required. These
results also make it clear that there was little effect
of complexity upon persistence duration.

It was thought that some people might not respond
appropriately' on the offset trials, and this was the

reason for the inclusion of catch trials. The results of

65

 

 

 

Table 6. Individual and mean persistence durations (in
ms) from experiment 6, as a function of
complexity and report condition

Simple Complex Simple Complex

Subject # Ichikawa Ichikawa Ichikawa Ichikawa

No Report No Report Report Report

1 129.98 118.72 133.32 107.16

2 133.26 131.50 130.43 126.01

3 115.88 122.90 56.74 46.14

4 105.57 126.81 157.58 147.74

5 157.07 153.96 77.84 65.87

6 118.38 123.95 170.71 186.36

7 69.67 70.28 95.95 103.51
MEAN 118.54 121.16 117.51 111.83

 

66

the performance on catch trials for each of the subjects
in experiment 6 are listed in Table 7, as well as the
results of incorrect responding on the experimental
trials. A criterion of fewer than 10% (or less than 4)
errors on catch trials was set as acceptable performance.
Table 7 shows that subject 2 failed to meet this
criterion. The data were reanalyzed with this subject's
data excluded, however, there was no appreciable change
in the results, so these data shall not be described in
any more detail. Thus, incorrect performance on the task
was probably not the explanation for failure to achieve a
significant complexity effect.

Finally, the mean persistence duration in this
experiment was 117.26 ms, which is in line with various
estimates of persistence that have been obtained
previously (e.g., DiLollo, 1980; Erwin, 1976).

To summarize the results of experiment 6: (1) no
main effects or interactions were significant; (2)
results showed at least two different subjective styles
of responding; (3) persistence durations were roughly
equivalent to values that have been obtained previously;
and (4) an analysis of the catch trials ruled out
incorrect responding as the reason for the lack of a
complexity effect.

Thus, these findings failed to support Erwin (1976),

since no complexity effect was obtained. This is

7
6

 

 

 

 

 

 

we e we e wm v wm v .Hmumxm

we e we e II II in II noumu n

we e we e we e we e .Hmuaxm

wm.m H wm m It In II it noumo e

we e we e we e wmm.w H .Hmuaxm

we e wm N II II II It noumu m

we e we e we e we e .Hmudxm

we e wm N II I: it It noumu v

we e we .e we e we e .Hmuaxm

we e wm.m H In In In in novmo m

wm~.H A we e wm~.ww e wme.m m .wmumxm

we e wm.m~ m In it II II noumu N

we e we e we e wme.m m .Hmuaxm

we e we e in It In In noumu H
.uomoum ucsou .uomoum unsou .uomoum ucsoo .Homoum ucsou

mama Hmafisz

uuommmiummmmo Duommm Czaummmmo uuommmlummco uuommm ozlummco awake Homnndm

 

cowwweeoo

 

e ucoEeummxw How Anoumo Ho Hmucmfieuwmxm
..m.ev amenu mo mean can :oeuwccoo mo coauocsm 0 mm .waxowsv oou
popcommou muomnnsm noesz cw mameuu mo cowuuomowm onMHOOmmm can “03552

.h manna

68

somewhat unsatisfying, since we had hoped to replicate
Erwin with his paradigm, and then test the complexity
effect using our more objective temporal integration task
(although in fact, we already did this in experiments 4

and 5).

Experiment 7

The preceding experiment failed to demonstrate a
complexity effect, even when full report was required,
but perhaps the jpreceding' Ichikawa. complexity
manipulation was too subtle, especially with 10 ms
exposures (e.g., Erwin, 1976; used 50 ms presentations).
It might be possible to demonstrate a complexity effect
with a different type of stimulus. Experiment 3 showed
that performance with X's and random letters in a
temporal integration task was virtually identical. Will
these stimuli show a difference in the Erwin task?
Again, it was thought that these stimuli. might show
differential persistence durations due to the subjective
nature of the task. Thus, our hypothesis was that random

letters would persist longer than X's.

Materials and Methods

Subjects. Six subjects were ‘used in this
experiment. All subjects had normal or corrected—to-
normal vision, and each was paid $5 for a single session

lasting about 1 1/2 hours.

69

Stimuli. The same stimuli that were employed in
experiment 3 were used in this experiment (i.e., random
letters arrays and arrays of X's). These arrays were
formed by randomly filling in 8 locations in a 4 x 4
matrix with one of these two stimulus types. Thus, the
stimuli consisted of random frames of 8 elements that were

presented for a constant duration (for all trials).

Equipment. The same equipment was used for this

 

experiment.

Procedures. The same procedures that were used in

 

experiment 6 were employed again, except that exposure
durations were 20 ms in length. The general procedure
shown in Figure 9 once again applied, except for the

different exposure durations that were used.

Conditions. The same conditions were used in this

 

experiment as in experiment 6, except that complexity was
manipulated by using letters and X's instead of high and
lOW' complexity Ichikawa. stimuli. Also, the jprogramming
error that resulted in only two delays (between fixation
and stimulus) being used for catch trials in experiment 6,

was corrected for in this experiment.

Results and Discussion
The results for experiment 7 are shown in Figure 11.

The only main effect that. was significant. was delay,

70

d
m
2..

a.
3

.3
N .
S

100‘

a
O
l

 

a
O

 

 

 

 

 

 

 

 

'8.

 

SIMPLE CI»flHJD( SIMPLE CI»flHJ§(
NO REPORT NO REPORT REPORT REPORT

I3IEIQESJESE1FIBIUCZIE IDIJIIIXUFJECDI' (:thSIZCZI)

(:CDIUIJHIVI'JUCDIU

Figure 11. Onset-offset persistence duration as a
function of complexity (X-array versus
random letters) and report conditons
for experiment 7.

 

71

F(4,20) = 4.39, p = .0104. The report x delay
interaction was also found to be significant, F(4,20) =
3.11, p s .0384; and the condition x delay interaction
was marginally significant, F(4,20) = 1.86, p = .1571.

All other main effects and interactions failed to show

significance: condition, F(1,5) - .65, p = .4568;
report, F(1,5) = .33, p -' .5898; condition x report,
F(1,5) = 1.24, p = .3153; and condition x report x delay,

F(4,20) = 1.08, p = .3923.

Post-hoc comparisons on the report x delay
interaction (Scheffe halfwidth - 32.66) showed that the
1100 ms delay resulted in significantly longer
persistence durations than the 500 to 700 ms delays for
the report condition. When post-hoc comparisons were
done on the delay term (Scheffe halfwidth = 26.85), it
was found that only the 500 ms and 1100 ms delays
differed, with greater persistence being exhibited at the
1100 ms delay. Thus, longer persistence for the 1100 ms
delay term is the likely cause of the one significant
post-hoc comparison in the report x delay interaction.
Post-hoc comparisons on the condition x delay interaction
(Scheffe halfwidth =- 35.41) showed nothing to be
significant. Planned comparisons on the condition x
report term (Bonferroni halfwidth . 30.63) showed no
significant differences. A t test for correlated means

was also performed on the condition term, and the

72

difference was found to be not significant (t(5) = .806,
n.s.).

Once again, performance in each of the condition x
report levels was similar, as seen in Figure 11. Report
conditions showed slightly longer persistence durations,
but this was not significant. Part of the reason for
this result was that at least two different types of
responding styles were adopted by individual subjects.
These results can be seen in Table 8. Subjects 1 and 2
showed much shorter persistence durations in both of the
report conditions. This was analogous to the finding in
experiment 6, and in fact subject 2 was one of the two
subjects from experiment 6 who adopted this responding
style. It was suggested previously that this quick
responding may have been used as a means" to improve
report accuracy (i.e., perhaps subjects did not wait for
complete fading of the visible trace). The other four
subjects showed either flat results, or a tendency for
persistence duration to increase when a report was
required.

The type of stimulus (i.e., complexity condition)
seemed to have little effect, although simple stimuli
appeared to have slightly longer persistence durations
(albeit a nonsignificant difference). It is unclear why
this difference might occur in this direction, except
perhaps that it is easier to group figures that are

73

Table 8. Individual and mean persistence durations (in ms)
from experiment 7, as a function of complexity
and report condition

 

 

Simple Complex Simple

Complex
Subject # (X-Array)

(Rand Lets) (X-Array) (Rand LetS)

 

No Report No Report Report Report
1 133.50 124.28 98.99 81.46
2 153.63 160.24 132.58 125.74
3 97.09 103.91 95.73 105.51
4 142.95 124.95 211.48 129.31
5 108.13 91.18 111.10 112.51
6 110.15 132.72 194.40 205.43
MEAN 124.24 122.88 140.71 126.66

 

74

identical than those that differ. Thus, the X's might
serve as placemarkers in the same way that the Ichikawa
dot patterns may have, and short-term memory might be
more readily applied to these stimuli. Random letters,
on the other hand, would make the task more difficult
because each stimulus element is different from its
neighboring elements, and thus the individual identities
might interfere with its functioning purely as a
placeholder.

Again, the data from catch trials were considered.
Table 9 shows the number (and percentage) of trials on
which subjects performed incorrectly. The same criterion
was applied (i.e., fewer than 10% errors on catch
trials), and in this experiment subjects 3 and 4 failed
to perform at an acceptable level. A reanalysis
excluding these subjects failed to reveal any different
findings, so these results shall not be described in any
more detail. Thus, once again, incorrect performance of
the task is unlikely to be the cause of nonsignificant
complexity effects.

The mean persistence duration in this experiment was
128.62 ms, which is once again similar to several results
reported in the literature (e.g., Erwin, 1976).

To summarize the results of experiment 7: (1) no
main effects were significant except for the delay term;

however, this was simply a control factor inserted to

75

 

 

 

 

 

 

 

 

we e we e wm~.H H we e .Hmuaxm
we e we e II II I: II noumu e
we e we e wmm.H H wm~.e m .Hmumxm
we e wm.> m II II In In noumo m
wm~.H H we e we.mH NH wme.wH HH .Hmuaxm
we e weH v II II II II noumu v
we e we e we e we e .Hmuaxm
weH e wm.~ H it I: it It soumu m
we e we e we e wm.m m .Hmumxm
we e wm.~ H it it II II coumu m
we e we e we e we e .Hmuaxm
we e we e in in II In noumu H
.uomoum ucsou .Homoum ucsou .Homoum ucsou .Homowm ucsou
mama Hmnadz
uuommmnummmmo uuommm CZIummmmo unommmiummco uuoamm oziummco HwHHB uomflnsm
GOHuHccou
h ucmEHummxo How Escudo no HmucmEHuomxo
..m.Hv HMHuu mo beau can coHuHocoo mo :oHuocsm m mm .meOHSU oou
popcommmu muomnnsm :OHS3 :H mHmHuu mo coHuHomoum emHMHoommm can gonadz .m mHnt

76

prevent subjects from anticipating the onset of stimuli;
(2) the report x delay interaction was significant;
however, this result seemed mainly to be due to the
effect of the delay term; (3) subjects once again showed
at least two different styles of performing the task,
thus demonstrating the subjectivity of the task; (4)
persistence durations were comparable to those of
experiment 6, and previous reports in the literature; and
(5) an analysis of catch trials ruled out incorrect
responding as the reason for the lack of a complexity
effect.

In conclusion, the findings from this experiment
also represent a failed replication of the Erwin (1976)
results. Our complexity manipulation was less subtle in
this experiment (i.e., X's versus random letters), than
in experiment 6, where complexity was formed by the
overall configuration of dots rather than the individual
stimulus elements. Once again, our results would have
been more satisfying if we could have found support for
Erwin (1976), and then used the same stimuli in a
temporal integration task, and fail to find a complexity

effect (as we did in experiment 3).

Experiment 8
Thus far, we have failed to obtain the same results

as Erwin (1976), since we have been unable to obtain a

77

complexity effect, even when a report of the information
has been required. Perhaps it will be possible to get a
complexity effect with stimuli that are particularly
complex (i.e., they do not have name codes attached to
them, and should be unfamiliar to the subjects). To meet
this demand, a set of pseudo-letters was created and
pitted against the random letters. It was hypothesized
that pseudo-letters would persist longer than random

letters, particularly when a report was required.

Materials and Methods

Subjects. Six subjects were run in this experiment.
All subjects had normal or corrected-to-normal vision,
and were paid $5 for a single session lasting for about 1

1/2 hours.

Stimuli. Arrays of random letters (all of the
consonants except y) and arrays of random pseudo-letters
(created by the author by interchanging the line segments
of each of the 20 letters used), were presented. Figure
12 illustrates the pseudo-letters that were used, as well
as the letter that each was derived from. These arrays
consisted of 8 letters (or pseudo-letters), placed
randomly in a 4 x 4 matrix like that used in several of
the preceding experiments. Again, the viewing distance
was 13", so the letters subtended .3 degrees vertically

and .4 degrees horizontally, while each space between the

2* If we

47

Figure 12.

78

or
en
7+
639:1

be
7m:
r-L.
3::

“UV-c
”7'4
(not

930!

Pseudo-letters employed in experiments
8, 9, and 10. These pseudo-letters were

created by interchanging the line segments
of all the consonants with the exception

of y.

79

letters occupied .4 degrees of visual angle in both the
vertical and horizontal directions. Letters and pseudo—
letters were equated for the .number of' light points
making them up, and both sets averaged 16 points per
letter. The entire array subtended 2.8 degrees of visual
angle vertically, and 2.4 degrees of visual angle
horizontally. Pseudo—letters were assumed to be
cognitively more complex than letters, but equivalent
featurally (i.e., they were constructed with. this in
mind).

Equipment. The same equipment was used in this

experiment.

Procedures. 20 ms exposures were used for both
types of stimuli in this experiment, since the letter
arrays used in the previous experiments had this exposure
duration. All other procedures were identical to those
employed in experiment 6. Again, answer grids were
provided to subjects for their full report blocks. Once
again, the general procedure sketched in Figure 9
applied, except that exposure durations were 20 ms rather
than 10 ms. Also, due to the difficult nature of the
report with these stimuli, subjects were asked to report
as many of the array elements as they could without

relying completely on guesses.

80

Conditions. The same conditions were used in this

 

experiment, except that complexity was manipulated by
using letters and pseudo-letters rather than the rated
low and high complexity Ichikawa stimuli (experiment 6),

or X's versus random letters (experiment 7).

Results and Discussion

 

The results for experiment 8 are shown in Figure 13.
The only main. effect. that. was significant. was delay,
F(4,20) -- 11.7, p < .0001. No other main effects or
interactions were significant as can be seen by the
following results: condition, F(1,5) - 1.28, p - .3089;
report, F(1,5) - .33, p - .59; condition x report, F(1,5)
= 1.27, p - .3106; condition x delay, F(4,20) - .13, p -
.97; report x delay, F(4,20) - 1.69, p =- .1911; and
condition x report x delay, F(4,20) - .41, p - .7959.

Post-hoc comparisons on delay (Scheffe halfwidth =
27.46) showed that 500 ms differed significantly from
both 1100 ms and 1300 ms, and that 700 ms differed
significantly from 1300 ms. The shorter delays showed
shorter persistence durations in these cases. Planned
comparisons were performed on the condition x report
interaction (Bonferroni halfwidth = 13.22), and it was
found that no report trials led to significantly longer
persistence durations than report trials for pseudo-

letters, and with random .letters, the same 'trend. was

81

 

 

 

 

 

 

 

 

 

 

8
I81 1601
l0
2
V
2 I40-
0
Pl
5 120‘
t . ....
a I ‘ .
10°” '—"'l 'F—TI
‘3
a 80-
f:
la
U] 60;: ‘ 1 v 1'
O: SIMPLE COMPLEX SIMPLE COMPLEX
33 no REPORT NO REPORT REPORT REPORT

CZCDTWIDCI'I‘JDCDIU

Figure 13. Onset-offset persistence durations as a
function of complexity (random letters
versus pseudo-letters) and report conditions
for experiment 8.

 

82

Observed, although the difference was not quite
significant. Within each Of these report conditions,
however, complexity did not vary significantly. A t test
for correlated means showed that the two types of stimuli
did not lead to significantly different persistence
durations (t(5) = 1.132, n.s.).

In this experiment, the various condition x report
levels showed slightly different results, since the
report conditions actually resulted in shorter
persistence durations than the no report conditions, as
can be seen in Figure 13. This is probably due to the
fact that 3 subjects (numbers 1, 5, and 6) showed a
pattern of results in which report conditions led to
shorter persistence durations. These three subjects were
the same ones that had responded this way in at least one
of the previous onset-offset experiments (6 and 7).
Thus, there seem to be subjective styles of responding
that fall out when using this technique. The other three
subjects in this experiment showed either flat results or
the opposite pattern (i.e., report conditions lead to
longer persistence durations). These trends can be seen
in Table 10.

Once again, the catch trial data were considered in
more detail. Table 11 shows the number (and percentage
of trials on which subjects performed incorrectly. The

same criterion for adequate performance was used (i.e.

I

83

Table 10. Individual and mean persistence durations (in ms)
from experiment 8, as a function of complexity
and report condition

 

 

 

Simple Complex Simple Complex
Subject # (Rand (Pseudo- (Rand (Pseudo-
Lets) Lets) Lets) Lets)
No Report NO Report Report Report
1 102.83 101.16 50.95 55.21
2 103.18 122.19 107.68 106.17
3 76.80 78.50 90.49 82.47
4 121.25 123.10 216.72 201.65
5 135.22 139.26 84.18 96.37
6 119.95 132.19 43.64 55.83
MEAN 109.87 116.07 98.94 99.62

 

84

 

 

 

 

 

 

 

 

me.H H we e we e wm.N N .Hmumxm
wm.N H wm.N H II In II II noumo e
we e wm~.H H we e we e .Hmwdxm
we e wm.N H II In It It noumu m
we e we e we e wm~.H H .Hmuaxm
we e wm.h m it II II II soumu v
we e we e we e wm~.H H .Hmumxm
wm.N H me m II II II II nouwo m
we e we e wme.w e www.mH HH .Hmuaxm
wm.N H wm.h m it it It In soumu N
wm~.H H we e we e wmm.H H .Hmuaxm
wm.N H wm.N H II II It In noumu H
.uoeoum unsou .Homoum ucsou .Homoum ucsoo .Homoum ucsoo
meme HOnEsz
HHOQOminmmHO whommm oqummmwO phonemiuowco . uuomwm ozlummco HmHHB pumnnsm
EOHHHOEOO
m unmEHHmmxm How Anouwo Ho Hmucmememxm 4.0.H. HMHHH
mo mew» cam :oHqucoo mo EOHPOCOH m mm .HHxOHOW oou pmecommmu
muomnnsm SOan EH mHmHuu mo :oHuHomoum OOPOHOOmmm can HOQEOZ .HH OHQOB

85

fewer than 10% errors on catch trials), and only subject
3 failed to perform at this level. A reanalysis with the
data from this subject excluded did not change the
results in any way, so they shall not be discussed in any
more detail. Thus, quick responding is unlikely to be
the cause of failure to find significant complexity
effects with this paradigm.

The mean persistence duration in this experiment was
106.13 ms; which is shorter than it was in the previous
two experiments, although still comparable to results
found in the persistence literature.

To summarize the results of experiment 8: fit) the
only significant main effect was delay, which was not a
variable of interest, and there were no significant
interactions either; (2) half of the subjects showed one
style of responding in which the report conditions
resulted in longer persistence durations, while the other
half of the subjects responded in the opposite manner;
(3) the persistence durations were similar to those of
experiments 6 and 7 and previous reports in the
literature; and (4) the catch trial analysis once again
showed that the failure to achieve a complexity effect
was not due to incorrect responding.

In conclusion, this experiment failed to obtain
results comparable to those of Erwin (1976), even when a

seemingly strong complexity manipulation was employed

86

(i.e., random letters versus pseudo-letters). It was
hoped that this experiment would show a complexity
effect, and that we could eliminate the effect, using the
same stimuli in a temporal integration task. Overall,
the Erwin onset-offset paradigm seemed less satisfactory
than the temporal integration task, since several
responding styles could be adopted by subjects. Also,
these three onset-offset experiments suggest that the
"neural" hypothesis is the more tenable of the two, since
there was no evidence of a complexity effect throughout
any of our stimulus manipulations.

It could be argued, however, that the complexity
manipulations used in the present set of experiments
simply did not work. In other words, perhaps a different
complexity manipulation would be more successful with the
Erwin paradigm. The experiments had originally been
designed in such a way that it was taken for granted that
the Erwin results were reliable. We planned to obtain
complexity effects with this technique, and then show
them to be artifacts of this subjective method, by
removing them with the very Objective temporal
integration technique. It seems likely that the Erwin
paradigm is untrustworthy, since the stimuli used for
experiment 6 were independently rated as varying in
complexity. It could very well be the case that a more

"forceful" experimenter could obtain complexity effects

87

using the onset-offset technique with the very same
stimuli. Also, it seems likely that the pseudo—letters
provided a true complexity manipulation due to their

unfamiliarity and lack of name codes.

Experiment 9

The purpose of this experiment was originally to try
to eliminate the complexity effect found in experiment 8
using the more objective temporal integration paradigm.
The problem encountered was that we could not obtain a
complexity effect with the Erwin paradigm (although some
subjects showed a tendency toward this pattern of
results). :rt is still of interest, however, to see what
results will be Obtained using the same stimuli (i.e.,
random letters and pseudo-letters) in the temporal
integration task. The "neural" hypothesis predicts that
there will be no difference in performance between the
letters and pseudo-letters, especially since ‘they' have
been equated for the number of light points used, and the
features (line segments) used in constructing them. The
"process" hypothesis predicts that subjects will show a
complexity effect, especially under the report conditions
(see experiment 10). The preceding experiments lead us to
the prediction that. no persistence: difference will be
found for the two types of stimuli.

88

Materials and Methods

 

Subjects. Six subjects were run in this experiment.
All had normal or corrected-to-normal vision, and each
was paid $5 for a single session lasting approximately 1

1/2 hours.

Stimuli. The same stimuli were used in this
experiment as were employed in experiment 8. The only
difference was that the stimuli once again consisted of
two frames which filled in 15 of 16 possible locations in
a 4 x 4 matrix when perceptually integrated. The first
frame consisted of either letters or pseudo-letters in 8
positions within the 4 x 4 matrix. The second frame was
formed by filling in 7 of the 8 remaining locations with
letters (or pseudo-letters, if the first frame consisted

of pseudo-letters).

Equipment. The same equipment was employed in this
experiment, as had been used in each of the preceding

experiments.

Procedures. This experiment was analogous to
experiments 4 and 5, except that letters and pseudo-
letters were used rather than Ichikawa stimuli. The
subject's task was to report the location in the 4 x 4
matrix where no letter (or pseudo-letter) was presented.

Subjects were once again required to type this location

89

into the computer. Each subject was run in 2 blocks of
300 trials, thus receiving 50 trials in each of the 2

complexity by 6 interframe interval conditions.

Conditions. Frame 1 and frame 2 were both presented

 

for 20 ms. There was a variable length interframe
interval (IFI) separating these two frames, and the IFI
values that were used in this experiment were the
following: 10, 40, 70, 100, 130, and 160 ms. Random
letter trials and random pseudo-letter trials were
randomly intermixed. Subjects were given 30 practice
trials before their first block of 300 experimental
trials, and 20 more practice trials before their second

block of 300 experimental trials.

Results and Discussion

 

The mean results from experiment 9 are shown in
Figure 14. The only main effect that was significant was
delay (interframe interval), F(5,30) - 74.03; p < .0001.
No other main effects or interactions were found to be
significant as can be seen by the following: block,
F(1,6) = .24, p - .6429; typestim, F(1,6) .. 1.57, p
.2571; block x typestim, F(1,6) - 1.29, p - .2990; block
x delay, F(5,30) - .68, p - .6405; typestim x delay,
F(5,30) = .76, p = .5750; and block x typestim x delay,
F(5,30) = .94, p = 4702.

90

ml

XF-—%K Random Letters
90- \

\ ...... .. Pseudo-Letters

80'

70'

604

50‘

40-

30-

 

20-

E’EBI§CZI§IU1F CZCJIQIQIBCZUP

10-

 

 

O

‘ 1 j j

0 1'0 40 70 100' 130 ' 1 60

I NTERFRAME I NTERVAL ( MSEC )

Figure 14. Percent correct for random letters and pseudo-
letters as a function of interstimulus
interval in experiment 9.

91

Planned comparisons were performed (Bonferroni
halfwidth a 15.31) on the delay term. 10 ms was found to
differ significantly from 40, 70, 100, 130, and 160 ms.
40 ms differed from all other delays. Finally, 70, 100,
130, and 160 ms were found not to differ significantly
from one another. Performance was found to decrease as
the interframe interval (delay) was increased. To
determine whether there were any persistence differences,
planned comparisons were performed on the typestim x
delay interaction (Bonferroni halfwidth - 6.71), however
no significant differences were found.

Table 12 shows that individual subjects displayed
basically the same pattern of results, as were found by
the overall mean. There were no radically different
styles Of performing, as there had been when using the
Erwin (1976) paradigm (especially considering the fact
that the same subjects were used).

To summarize the results of’ experiment 9: (1)
performance decreased as the interframe interval was
increased; (2) there was no effect of complexity, since
pseudo-letters did not persist significantly longer than
random letters; and (3) there was no evidence of a

practice effect, as shown by the lack of significance for

the block factor.

92

Table 12. Individual and mean results (% accuracy) from
experiment 9, as a function of interframe
interval

 

 

Stimuli: Random Letters

 

Interframe Interval

 

 

Subject # 10 40 70 100 130 160
1 100 96 82 56 50 42
2 98 54 14 10 14 16
3 96 58 30 28 22 12
4 92 54 16 18 14 16
5 92 76 44 24 32 28
6 100 94 58 44 36 32
7 86 26 10 08 12 08
MEAN 94.8 65.4 36.3 26.9 25.7 22.0

 

Stimuli: Pseudo-Letters

 

1 100 98 80 66 42 54
2 100 52 12 22 16 16
3 94 68 24 16 20 12
4 88 56 22 28 20 20
5 100 80 48 36 40 30
6 100 96 60 32 32 24
7 86 42 10 10 04 08

MEAN 95.4 70.3 36.6 30.0 24.9 23.4

 

93

Experiment 10

The temporal integration experiments that have been
run. thus far suggest that the "neural" hypothesis is
correct. The slight complexity effect that was obtained
in experiment 1 was ruled out as being due to suppressive
forces acting among array elements near each other in
space and/or time. Furthermore, although several
replications Of Erwin (1976) were attempted (using three
different complexity manipulationS), no complexity effect
was obtained. A "process" advocate could argue that the
temporal integration experiments fail to show a
complexity effect because they do not require
identification of the stimuli. In fact, Erwin (1976)
stresses that these complexity effects only occur when
complex processing of the stimuli is required (e.g., if a
report must be made). Thus, the purpose of this
experiment was to force identification of the stimuli
while still maintaining the task of detecting the missing
location in the array (i.e., the temporal integration
paradigm). The "neural" hypothesis suggests that this
deeper processing requirement will not result in a

complexity effect.

94

Materials and Methods

 

Subjects. Six subjects were run in this experiment.
All had normal or corrected-to-normal vision. Subjects

were paid $5 for each of two 1 1/2 hour sessions.

Stimuli. Stimuli were identical to those used in

experiment 9.

Equipment. The same equipment was employed in this

 

experiment, as had been used in each of the preceding

experiments.

Procedures. In addition to typing their row-column
coordinate for every trial, subjects also had to record
the letters (or pseudo-letters) that were in the same row
as the unoccupied location. These are designated as
partial report trials, since this method is analogous to
the partial report technique seen throughout the
persistence literature (e.g., Sperling, 1960). To aid
the subjects on these partial report trials, a diagram
was provided of all of' the pseudo-letters, so as to
improve drawing accuracy. Subjects were run in the same
number of practice and experimental trials as they had

been in experiment 9.

Conditions. The conditions for this experiment were
the same as those used in experiment 9, except that a

partial report was required on every trial.

95

Results and Discussion

 

The mean results for experiment 10 are shown in
Figure 15. The only main effect that was significant was
delay, F(5,30) - 52.48, p < .0001. Also, the typestim
factor was marginally significant, F(1,6) - 5.90, p a
.0512. Pseudo-letters were found to result in better
performance than random letters. The other main effect
(block) was not significant, F(1,6) - .10, p - .7616.
One interaction showed a trend toward significance:
namely the block x typestim x delay, F(5,30) - 1.66, p -=
.1751. The other interactions were not significant as
shown by: block x typestim, F(1,6) - .00, p - .9536;
block x delay, F(5,30) - .80, p - .5573; and typestim x
delay, F(5,30) - 1.38, p - .2591.

Planned comparisons on delay (Bonferroni halfwidth -
17.71) showed 10 ms to be significantly different from
40, 70, 100, 130, and 160 ms. 40 ms was significantly
different from 70, 100, 130, and 160 ms. 70 ms was
significantly different from 160 ms. Finally, 100, 130,
and 160 ms did not differ significantly from one another.
Thus, as the interframe interval increased, performance
was found to decrease.

The typestim factor showed results very near
significance, with pseudo-letters resulting in higher
accuracy than random letters (when a report was

required), and it also appears that the pseudo-letters

100'

90-

80'

70-

60-

50-

1?1312(:IBI'1P CZCJIQIQIEC21F

 

96

XF-fix Random Letters

\ o- - -. Pseudo-Letters

 

 

0 1O

1 V T

40 70 100 730 ' 160

I NTERFRAME I NTERVAL ( MSEc: )

Figure 15.

Percent correct for random letters and
pseudo-letters when report was required
as a function of interframe interval in
experiment 10.

97

asymptote later than the random letters. This supports
the "process" account. Why might this state of affairs
have occurred? One possibility is that a complexity
effect was Obtained. Another is that subjects experienced
some sort of "frustration effect" when the pseudo-letters
were presented. In other words, when the first frame
consisted of pseudo-letters, subjects decided that they
would at least get the location correct, since the pseudo-
letters were so difficult to report. Subjects continually
complained of the difficulty of reporting the pseudo-
letters, and they generally felt as if they were guessing
on these reports. Since the random letter trials were
somewhat easier, subjects may have more evenly divided
their efforts on these trials. In any event, the
differences between stimulus types were quite small.

The typestim x delay interaction was also analyzed
with planned comparisons (Bonferroni halfwidth - 10.47).
No significant differences were obtained, however, as the
10-130 ms delays, pseudo-letters resulted in better
performance than the random letters at near significant
levels. Individual subjects showed slightly different
patterns of results, as can be seen in Table 13. This
result can be contrasted with the individual results from
experiment 9 (i.e., listed in Table 12), where responding

styles were remarkably consistent across subjects.

98

 

 

 

 

 

 

 

Table 13. Individual and mean results (% accuracy) from
experiment 10, as a function of interframe
interval

Stimuli: Random Letters with Report
Interframe Interval
Subject # 10 40 70 100 130 160
l 100 98 84 60 42 48
2 92 50 20 12 14 18
3 92 52 24 20 16 24
4 90 44 24 18 26 26
5 94 70 52 38 38 20
6 96 94 70 50 42 44
7 98 '78 36 24 10 14
MEAN 94.6 69.4 44.3 31.7 26.9 27.7
Stimuli: Pseudo-Letters with Report
1 100 100 88 64 54 40
2 94 50 30 20 14 08
3 98 76 24 22 12 18
4 96 42 20 28 16 18
5 100 72 44 42 44 34
6 98 98 86 58 42 16
7 96 88 3O 14 18 18
MEAN 97.4 75.1 46.0 35.4 28.9 21.7

 

99

There was another trend that is troublesome for the
"neural" hypothesis to accOunt for however, as subjects
appeared to perform better in this experiment than in
experiment 9, even though the task was substantially more
difficult (i.e., a report of the stimuli was required).
If performance was superior when a report was required,
then this is consistent with the "process" hypothesis.
In fact, this result is very much like that of Erwin
(1976). It should also be pointed out that subjects
always performed experiment 10 after experiment 9, so it
is of interest to see whether the observed effect was due
to practice or to a true positive complexity effect.

In order to ascertain this information, the data
from experiments 9 and 10 were pooled for those subjects
who had taken part in both studies (i.e., six of the
seven subjects in each experiment had participated in
both). When the data were pooled, most of the main
effects and interactions were not significant, as can be
seen by the following: experiment, F(1,5) - .98, p a-
.3680; block, F(1,5) - .11, p - .7561; experiment x
block, F(1,5) - .02, p - .8911; experiment x typestim,
F(1,5) - .04, p - .8516; block x typestim, F(1,5) - .45,
p - .5334; block x delay, F(5,25) - 1.00, p - .4369;
experiment x block x typestim, F(1,5) - .00, p - .9534;

experiment x block x delay, F(5,25) .. .43, p - .8227;

100

experiment x typestim x delay, F(5,25) = .98, p = .4509;
and block x typestim x delay, F(5,25) - .67, p = .6465.

Only one main effect was significant, namely delay,
F(5,25) = 56.09, p < .0001. There was a trend toward
significance for typestim however, F(1,5) = 3.41, p =
.1368. The typestim x delay interaction also showed a
trend toward significance, F(5,25) - 1.89, p - .1316.
Finally, two other interactions were marginally
significant, experiment x delay, F(5,25) -= 2.25, p =
.0810; and experiment x block x typestim x delay, F(5,25)
= 2.13, p - .0954.

Further analyses were performed on the theoretically
interesting data as well as those factors and
interactions that approached significance. Planned
comparisons were performed on the delay term (Bonferroni
halfwidth = 15.29), and it was found that 10 ms differed
from all other delays, as did 40. ms; 70 ms differed from
160 ms, and all other delays were not significantly
different. As has been found for all of the temporal
integration experiments, accuracy decreased as the delay
was increased.

A t test for correlated means was performed on the
typestim main effect (t(5) - 1.771, n.s.), and it was
found that the two types of stimuli were not
significantly different, although 4 of the 6 subjects

101

showed a tendency for pseudo-letters to lead to higher
accuracy than random letters.

Planned comparisons were also performed on the
typestim x delay interaction (Bonferroni halfwidth =
5.73), and it was found that there were no significant
differences between the stimulus types at any of the
delays, although the difference was close to significant
at the 100 ms delay. Figure 16 illustrates the results
of this interaction, and Table 14 lists the mean and
individual results. The pseudo-letters led to higher
accuracy at all delays except 160 ms, however the
differences were extremely small in all cases.
Furthermore, both types of stimuli appeared to asymptote
at the same point, so there is little evidence that one
stimulus type persists longer than the other.

Since the pseudo-letters appear to lead to
consistently better performance than the random letters,
one is left wondering if perhaps suppressive interactions
are once again affecting the results. The pseudo-letters
were constructed by maintaining the same line segments,
while interchanging their relations to one another.
Several naive subjects were also consulted as to whether
the pseudo-letters looked like any of the letters of the
alphabet, or familiar shapes. When the answer was
affirmative, then the pseudo-letters were further

modified, until the final set (listed in Figure 12) was

102

9°. \\ x——X’Random Letters
.——0 Pseudo-letters

50-
4o-

3Od

 

201

r=rsrz<215111r CZCJIQIQIECZWP

10-

 

 

O

I j '1

0 1‘0 40 70 1 00 130 ' 160

I NTERFRAME I NTERVAL ( MSEC )

F'ggre 16. Percent correct for random letters and pseudo—
letters as a function of interstimulus interval for the
Pooled data of experiments 9 and 10.

103

Table 14. Individual and mean results (% accuracy) from
the pooled data of experiments 9 and 10, as a
function of interframe interval

 

 

Stimuli: Random Letters

 

Interframe Interval

 

 

Subject # 10 40 70 100 130 160
1 100 97 83 58 46 45
2 95 53 19 15 15 20
3 94 54 25 20 18 15
4 91 49 20 18 20 21
5 93 73 48 31 35 24
6 98 94 64 47 39 38

MEAN 95.17 70.00 43.17 31.50 28.83 27.17

 

Stimuli: Pseudo-Letters

 

1 100 99 84 65 48 _ 47
2 99 64 18 22 14 17
3 94 59 ‘ 27 18 17 10
4 92 49 21 28 19 19
5 100 76 46 39 42 32
6 99 97 73 45 37 20

MEAN 97.33 74.00 44.83 36.17 29.50 24.17

 

104

arrived at. This construction method certainly did not
rule out the possibility of differential suppression
effects, however the main point to be made is that there
was little support for a complexity effect as the two
types of stimuli did not persist differentially (i.e.,
one type did not asymptote later than the other), even if
accuracy was slightly higher for pseudo-letters.

It seems likely that subjects did better in
experiment 10 because they had become familiar with the
task and stimuli during experiment 9. This is especially
evidenced by the fact that the experiment x typestim x
delay interaction did not approach significance. Thus,
experiment 10 appears to exhibit more of a practice
effect, than support for the "process" hypothesis.

Post—hoc analyses were also performed on two other
interactions, since they showed marginal significance.
Analysis of the experiment x delay interaction (Scheffe
halfwidth = 8.19) showed no differences to be
significant. In general, experiment 10 resulted in
higher accuracy than experiment 9, particularly at the
longer delays, but these differences were small and
nonsignificant at each delay.

Finally, a post-hoc analysis of the experiment x
block x typestim x delay interaction (Scheffe halfwidth =
14.80) showed no comprehensible pattern of results. In

fact, when comparisons were restricted to each of the

105

particular delays, only one difference was significant,
and it is reasonable to assume that this comparison was
the result of a Type I error. Several comparisons across
delays were significant, however, it has been well
established that delay (i.e., interframe interval) is a
very powerful determinant of performance in these
experiments.

To summarize the results of experiment 10: (1)
performance was found to deteriorate as the interframe
interval increased; (2) there was some suggestion that
pseudo-letters resulted in better performance than random
letters when a report. was required, however this ‘was
shown to be more likely to represent a practice effect,
since subjects always took part in experiment 10 after
experiment 9; (3) there was little evidence of a
complexity effect, since the two types of stimuli did not
appear to asymptote differentially; and (4) there was
some suggestion that the random letters and pseudo-

letters may have shown differential suppression effects.

GENERAL DISCUSSION

The preceding experiments were designed to
investigate some ways in which human perception operates
upon an environment filled with many diverse objects.
Some aspects of the environment are simple, while others
are complex (e.g., predictable groupings of objects
versus random arrays). Does this make any difference to
the visual system? There are two possible answers to
this question: yes and no. If the complexity of the
perceived stimulus affects the duration of visual
processing, then this can be described as support for the
"process" hypothesis, whereas the failure of complexity
to affect the duration of visual processing can be
described as support for the "neural" hypothesis (i.e., a
standard neural sequence of events occurs regardless of
what is viewed).

In order to address this question, there are two
important aspects of the procedure to be considered: (1)
the stage of visual perception at which to look for a
complexity effect; and (2) complexity itself. It is
obvious that complexity becomes important at higher
levels of cognition (e.g., differences in reading speed
for text inverted in various ways--Kolers, 1975), but it

106

107

is interesting to determine if complexity differences
might exist from the earliest stage of visual processing.
Thus, the goal of this project was to look for a
complexity effect at the level of visible persistence,
which can probably be equated with neural persistence
(see Coltheart, 1980). Coltheart's definition therefore
implies that this stage is neural, and hence unlikely to
be affected by complexity manipulations. DiLollo seems
to take a similar stance. This "neural" hypothesis is
not universally accepted, however, since other
researchers purporting to study the same stage have
demonstrated positive: complexity effects (e.g., Erwin,
1976; Avant 8 Lyman, 1975). Since these researchers
believe that the duration of visible persistence is
directly related to the amount of information to be
coded, their view (i.e., the "process" hypothesis)
suggests that complexity can affect visible persistence
duration.

The way in which to define complexity is a much more
difficult aspect of the research plan, since there are so
many possible manipulations. A quick scan of the
literature shows many ways of defining this variable.
For example, orders of approximation to English (Erwin s
Hershenson, 1974; Erwin, 1976); number of interior angles
of a line drawing (Attneave, 1957); the ratio of the

squared perimeter to the area of a figure (Stenson,

108

1966), as well as several other operational definitions
have been proposed. Initially we decided to use a global
measure (i.e., rectangles versus letters) since these
stimuli varied both cognitively and featurally, with the
stipulation that more fine-grained‘ complexity
manipulations were planned if a true complexity effect
was found. The rectangles and random letters were
equated for the number of dots comprising them (as were
the stimuli in later experiments), and this served to
remove a potential confound of differential retinal
excitement. .
It was critical to the success Of this project that
we demonstrate the viability of our complexity
manipulations. Thus, various complexity checks were
instituted throughout the design stage of these
experiments. These aspects can probably best be
characterized as a set of rules or criteria. Each of the
experiments employed at least one of these criteria,
although these complexity principles were generally not
mutually exclusive from one another. The major
Complexity Criteria considered in this research project

were the following.

Symmetry
In. general, symmetry can. be associated with

complexity (i.e., the less symmetrical a stimulus is, the

 

109

more complex it is). Symmetry can be measured by how
much stimulus halves would overlap, if they were folded
on an imaginary midline. Some evidence for this can be
seen by the fact that the subjects in the Ichikawa (1985)
experiment rated symmetrical dot patterns to be more
simple. Symmetry can also be at least partially equated
with the Gestalt term "goodness.”

Subject Ratings

As suggested previously, subject ratings are another
potential complexity criterion. This is different from
the other types of complexity, and in fact is probably
heavily dependent on some introspective awareness of
these other aspects (e.g., symmetry). An example of this
complexity manipulation can be found in the experiments
of Ichikawa (1985). The simplest rated dot patterns are
those that appear to conform to many of the Gestalt
grouping principles. In general, the simple patterns
might be refered to as "good" Gestalts. The differences
between these two types of patterns can be seen by

referring to Appendix A.

Groupability
As suggested in the subject ratings section,
subjects often appear to base their complexity judgments
on how "good" the arrangement of elements seem to be.

For example, Figure 6 has more of a sense of randomness

 

110

to it than Figure 5, and thus is rated by subjects as
being more complex. If one were to attempt to verbally
describe Figure 6, this description would be more complex
than that of Figure 5 as well, thus suggesting that how
elements group may well play a role in the ease or

difficulty of their coding.

Nameability
It seems plausible that an object which has a name
may be more easily coded than a similar Object with no
name. For example, reports Of pseudo-letters in
experiment 10 were found to be much more difficult than
the reports of random letters, since the attached name
codes (i.e., A, B, C, etc.) can more easily be placed in

durable memory storage.

Distinctiveness

Another complexity manipulation that was employed in
the present series of experiments was the distinctiveness
of array elements from one another. This section shall
argue that if an array is composed of identical elements
(e.g., dots, rectangles, X's, etc.), it will have a lower
complexity value than an array composed of a random
collection of letters or pseudo-letters (i.e., each
element in these arrays will differ from its neighbors).
With identical elements throughout the array, subjects

would only have to process the arrangement, as opposed to

111

the variable element arrays in which subjects would have
to remember identities as well as the spatial
arrangements. In other words, each of the elements in
the identical array (e.g., X'S) could serve as some sort
of placeholder for the spatial arrangement when a report
is required, whereas this would not be the case for an

array of random figures (e.g., letters).

Predictability

The above criterion also seems to tie in with the
predictability of an array, as do some of the other
criteria (e.g., groupability and symmetry). In fact, it
seems likely that predictability plays a big role in
subject's ratings of complexity. With symmetrical
objects (or arrangements), one only has to know (or see)
part Of the Object (or arrangement) to become aware Of
what the whole object (or arrangement) looked like.
Similarly, groupability allows a subject to more easily
form a pattern during the first half of a matrix, so that
the second half of the array will be more predictable.
Erwin (1976; and Erwin & Hershenson, 1974) has
manipulated complexity in a similar fashion with his
orders of approximation to English” Thus, a random
string of consonants is much less predictable than a
string of letters that bears more similarity to the rules

of English word formation.

‘ dull-tum—
. - ' 7.‘ i‘_

112

The first three experiments varied several
parameters and found little evidence for a complexity
effect. There was a finding that suppression (similar to
metacontrast masking) Operated between stimuli
neighboring each other in space and/or time. Suppression
was found to increase as intercontour distance between
array elements was decreased, with rectangles having the
smallest amount of intercontour separation, and X's
having the largest. Actually, dot arrays have even
greater separations and pilot data have shown them to
persist even longer than random letters and X's. Thus,
the first three experiments must be viewed as
demonstrating the role that suppression plays in temporal
integration experiments.

Experiments 4 and 5 used stimuli that had
independently been rated for complexity (see Ichikawa,
1985), and found no complexity effect when using the
temporal integration paradigm. This can be taken as
fairly strong evidence against the ”process" hypothesis
of visible persistence. This claim would be made even
more convincing however, if a complexity effect could be
demonstrated with these same stimuli and a more
subjective technique. The task we chose to use was the
Erwin onset-offset technique, since it is free from many
of the problems found with other techniques (e.g., there

is a memory component involved when performing the task

 

5- VI- I'JI‘I'!‘ Q sham—7'

 

113

used by Avant and Lyman, 1975), although even this
technique relies on judgments that are not visible to the
experimenter. Thus, most of our efforts were directed at
showing that the complexity effects Obtained with
subjective techniques do not stand up when the same
stimuli (and complexity manipulations) are used in our
more objective temporal integration paradigm.

The above rationale was used for experiments 6-10.
In experiment 6, we attempted to find a complexity effect
using the Erwin onset-offset technique and stimuli that
showed no such effect with the temporal integration
paradigm (i.e., the Ichikawa (1985) rated complexity
patterns). Likewise, experiment 7 attempted to find a
complexity effect with the Erwin method, and random
letters versus X's. Remember that experiment 3 found no
complexity effect with these stimuli and the temporal
integration technique. Finally, we also 'used random
letters and pseudo—letters in the Erwin technique to try
to demonstrate a complexity effect (experiment 8). None
of these attempted replications of Erwin were successful
however, so it could be argued that we did not manipulate
complexity effectively. It is unclear whether this is
true or not, since some subjects did show an effect of
complexity, however this could have occurred simply due
to chance. This leads us to believe that Erwin's task is

unreliable, and that his results may have been due to any

114

number of reasons (e.g., particularly motivating
instructions, predictability of' the onset of stimuli,
etc.).

Why did Erwin (1976) get a complexity effect with
the onset-Offset technique while we could not? This
issue is worthy of further consideration. There were a
number of differences between the two experiments. First
of all, Erwin used 50 ms exposures, whereas our exposure
durations were 10 ms (experiment 6) or 20 ms (experiments
7 and 8). It may be that our exposures were too brief to
allow a cognitive complexity effect (i.e., the complexity
difference was not perceived).

Another difference between the two procedures was
the size and shape of the arrays (i.e., Erwin used a row
of 7 letters, while we used 8 elements arranged randomly
in a 4 x 4 array). Erwin's arrays subtended 2.3 degrees
horizontally and .5 degrees vertically, while our arrays
were 3.1 degrees horizontally and 3.6 degrees vertically.
Combined with our short presentation times, it is
possible that our complexity manipulations were
ineffective because of the difficulty subjects may have
had perceiving the whole array (especially in experiment
6, where the entire pattern of dots determined how
complex that particular stimulus waS) . Since subjects
had difficulty reporting more than a few items from the
array, this view is at least tenable, although output

' 3"" . .'-Y4_'JIAIIJ.

 

115

interference is at least as good an explanation for this
result.

A third and probably most important difference
between our results and Erwin (1976) were the subjective
aspects of the two experiments. For example, some of our
subjects showed a complexity effect while others showed
either no complexity effect or some strange reporting
phenomena (e.g., shorter persistence durations when a
report was required). It may be that Erwin's subjects
were provided with more highly motivating instructions
than our subjects, and thus adopted a strategy that best
corresponded to the demand characteristics Of his
experiment. Our subjects were provided with relatively
neutral instructions, informing them whether to respond
to the ”onset" of the stimulus, or when the stimulus
appeared to disappear Off the screen (offset). Our
subjects were also instructed to report correctly as many
of the elements as possible in their correct locations,
including educated guesses. Erwin (1976) instructed
subjects not to guess, so this may have also had an
effect on performance of the task, although it is not
clear what effect this would have.

A final difference between our experiment and that
of Erwin was the complexity manipulations that we used.
A possible complexity manipulation that could have been

studied was the effect of using orders of approximation

 

116

to English in the same two tasks (in fact, Erwin used
this manipulation for his study). This would be an
exceedingly difficult manipulation for the temporal
integration task, although perhaps an effect would emerge
with these stimuli. This would suggest that the way in
which complexity is manipulated has a direct bearing on
whether a complexity effect is found.

The last point to be considered is the implications
that the findings of this research program will have.
Additionally, how can these findings be incorporated into
current models Of perception? As mentioned earlier, some
researchers have claimed all along that visible
persistence is neural in nature (e.g., Coltheart, 1980;
DiLollo, 1980). These researchers have proposed models
that assume an early neural stage which is unaffected by
such cognitive aspects as complexity. Thus, it is quite
likely that many models of perception will not have to be
altered, or at the most, only slightly altered to
accommodate the results if the "neural” hypothesis proves
true. The big surprise would have occurred if strong
evidence for the "process" hypothesis had been found,
however, this seems highly unlikely, since very little
evidence was found supporting this view. Experiment 10,
however, showed a trace of a report effect, so perhaps
complexity does have some effect that we have not

manipulated effectively enough. A complexity finding

. I! mi 15.-m3,

 

117

would result in a major ‘upheaval of’ several current
visual perception issues, particularly the level at which
cognitive complexity can have effects (i.e., this point
would be pushed back to the earliest stage of visual
processing). This would have implications for the
interaction of top-down and bottom-up jprocessing, and
would almost assuredly tie in with the skills of reading.
In fact,- some research had already been performed which
suggests that the duration of visible persistence affects
the quality of reading, and may be a determinant of
certain types of dyslexia (e.g., DiLollo, Hanson, &
McIntyre, 1983).

It might be illustrative to place visible
persistence within a current model of visual processing.
Irwin and Yeomans (1984, 1986) have proposed a model of
the early stages of visual perception. This model has
explicated in considerable detail the processes that
occur during informational persistence (or knowledge
about the visual properties of the stimuluS), and is
diagrammed in Figure 17.

This model assumes that sensory information from a
display (sensory representation) is ultimately translated
into relatively durable .nonvisual identity codes that ~
have associated with them some abstract representation of
spatial position (e.g., spatial coordinates). This

information is then transferred to short-term memory. As

 

118

TRANSLATION

I

 

 

 

 

 

 

«ownsuau. 105mm cont ...
_, WITH SPATIAL communes
v.50“ _. smsonv VISUAL

‘ 01591 Avg REPRESENTATION ANALOG

 

 

 

Figure 17. Irwin and Yeomans' (1986) model of the
early stages of visual processing.

' 2 'A.'S'.“-'

 

119

long as the display is present, translation can and will
occur; however, when the display is terminated, the
translation process must rely on stimulus persistence.
The experiments performed by Irwin and Yeomans suggest
that this stimulus persistence is best represented by a
visual analog. The duration of the analog representation
is sensitive to stimulus exposure conditions, such as the
presence of masking stimuli, and it is assumed to decay
with the passage of time. In other words, the visual
analog maintains form and location information about the
display for some period of time after stimulus Offset in
order to allow the translation process to extract further
information about the presented items. In general, the
analog lasts for 150—300 ms regardless Of exposure
duration. Translation is nonselective unless directed by
a cue, such as used in a partial-report experiment.

It should be noticed that there is no mention of
visible persistence in this model. This process has been
dumped into the sensory representation box in the model,
since not enough details were known about it. The
findings from this research plan. may ‘ultimately give
visible persistence a role in the model. The findings of
this project best support the "neural" hypothesis, and
this suggests that the role of visible persistence is
that translation into identity codes occurs using

information from the actual array or visible persistence,

120

whichever of the two is longer. This makes sense since
visible persistence has been described as the phenomenal
impression that the stimulus is still visibly present
(Coltheart, 1980). This mechanism guarantees the visual
system a minimum of about 100 ms for using veridical
information about the stimulus.

In conclusion, the question of how complexity
affects visible persistence is an interesting and
challenging one. It may not have been answered fully by
this research plan, but certainly we took a step in the
right direction. Perhaps the biggest contribution this
research made was the introduction Of an Objective
technique into a line of study that has traditionally
been plagued by subjective methods. As so often happens
to topics in psychology, complexity turns out to be more

complex than meets the eye.

 

APPENDIX

 

 

Figure A.l.

Simple

121

Ichikawa Stimuli:

26-49.

in"

 

 

122

 

Figure A.2. Simple Ichikawa Stimuli: 50-65.

 

 

123

 

Figure A.3. Complex Ichikawa Stimuli: 101-124.

 

124

 

Figure A.4. Complex Ichikawa Stimuli: 125-140.

 

REFERENCES

 

REFERENCES

Adelson, E. H., & Jonides, J. (1980). The ;
psychophysics of iconic storage. Journal of
Experimental Psychology: Human Perception and
Performance, p, 486-493.

 

Allen, F. (1926). The persistence of vision. American
Journal of Physiological Optics, 1, 439-457.

 

Twp .

Attneave, F. (1957). Physical determinants of the
judged complexity of shapes. Journal of
Experimental Psychology, 53, 221-227.

Avant, L. L., & Lyman, P. J. (1975). Stimulus
familiarity modifies perceived duration in
prerecognition visual processing. Journal of
Experimental Psychology: Human Perception and
Performance, 1, 205-213.

 

Avant, L. L., Lyman, P. J., & Antes, J. R. (1975).
Effects of stimulus familiarity upon judged visual
duration. Perception and Psychophysics, 11, 253-
262.

Averbach, E., & Coriell, E. (1961). Short-term memory
in vision. Bell System Technical Journal, 39, 309-
328.

Baxt, N. (1871). Uber die Zeit welche notig ist damit
ein Gesichseindruck zum Bewusstein kommit und uber
die Grosse (Extension) der bewussten Wahrnehmung bei
einem Gesichtseindrucke von gegebener Dauer.
Pfluger's Arch. ges. Physiol.,'g, 325-336.

Bowen, R. W., Pola, J., & Matin, L. (1974). Visual
persistence: Effects of flash luminance, duration,
and energy. Vision Research, 11, 295-303.

Briggs, G. G., & Kinsbourne, M. (1972). Visual
persistence as measured by reaction time. gparterly
Journal of Experimental Psychology, 23, 318-325.

Cattell, J. McK. (1886). The time it takes to see and
name objects. Mind, 11, 277-292, 524-538.

125

126

Clark, S. E. (1969). Retrieval of color information
from preperception memory. Journal of Experimental
Psychology, p2, 263-266.

 

Coltheart, M. (1980). Iconic memory and visible

persistence. Perception and Psychophysics, 21, 183-
228.

Coltheart, M., Lea, C. D., 8 Thompson, K. (1974). In
defence of iconic memory. Quarterly Journal of
Experimental Psychology, 29, 633-641.

2AIMFZ- «.1 saw

Dick, A. O. (1969). Relations between the sensory
register and short-term storage in tachistoscopic
recognition. Journal of Experimental Psychology,
82, 279-284.

 

DiLollo, V. (1977). Temporal characteristics of iconic
memory. Nature, 267, 241-243.

DiLollo, V. (1978). On the spatio-temporal interactions
Of brief visual displays. In R. H. Day and G. V.
Stanley (Eds.), Studies inpperception (pp. 39-55).
Perth: UniversiEy Of Western Australia Press.

DiLollo, V. (1980). Temporal integration in visual
memory. Journal of Experimental Psychology;
General, 109, 75-97.

DiLollo, V., Hanson, D., & McIntryre, J. S. (1983).
Initial stages of visual information processing in
dyslexia. Journal of Ex erimental Psychology;
Human Perception and Per ormance, 2, 923-935.

DiLollo, V., & Hogben, J. H. (1985). Suppression of
visible persistence. Journal of Experimental

Psychology: Human PerceptiOn and Performance, 1;,
304-316.

Efron, R. (1970a). Effect of stimulus duration on
perceptual onset and offset latencies. Perception
and Psychophysics, p, 231-234.

 

Efron, R. (1970b). The minimum duration of a
perception. Neuropsychologia, p, 57-63.

Efron, R. (1970c). The relationship between the
duration Of a stimulus and the duration of a
perception. Neuropsychologia, p, 37-55.

127

Eriksen, C. W., & Collins, J. F. (1967). Some temporal
characteristics Of visual pattern perception.
Journal of Experimental Psychology, 13, 476-484.

Eriksen, C. W., & Collins, J. F. (1968). Sensory traces
versus the psychological moment in the temporal
organization of form. Journal of Experimental
Psychology, 11, 376-382.

 

Erwin, D. E. (1976). Further evidence for two
components in visual persistence. Journal of
Experimental Psychology: Human Perception and
Performance, E, 191-209.

 

 

Erwin, D. E., & Hershenson, M. (1974). Functional
characteristics of visual persistence predicted by a
two-factor theory of backward masking. Journal of
Experimental Psychology, Egg, 249-254.

 

Haber, R. N., & Standing, L. (1969). Direct measures of
short-term visual storage. anrterly Journal of
Experimental Psychology, EE, 43-54.

Haber, R. N., & Standing, L. (1970). Direct estimates
of the apparent duration of a flash. Canadian
Journal of Psychology, E5, 216-229.

 

Hogben, J. H., & DiLollo, V. (1974). Perceptual
integration and perceptual segregation of brief
visual stimuli. Vision Research, lg, 1059-1069.

Hogben, J. H., & DiLollo, V. (1985). Suppression of
visible persistence in apparent motion. Perception
and Psychophysics, pp, 450-460.

 

Ichikawa, S. (1985). Quantitative and structural
factors in the judgment of pattern complexity.
Perception and Psychophysics, §§, 101-109.

Irwin, D. E., & Yeomans, J. M. (1984). Nonvisible
visual persistence in iconic memory. Paper
presented at the 25th annual meeting of the
Psychonomic Society San Antonio, Texas, November,
1984.

Irwin, D. E., & Yeomans, J. M. (1986). Sensory
registration and informational persistence. Journal

9f Experimental Psychology: Human Perception and
PerfOrmance, l2, 343-360.

 

128

Kolers, P. A. (1975). Specificity of operations in
sentence recognition. Cognitive Psychology, 1, 289-
306.

 

Long, G. M. (1980). Iconic memory: A review and
critique of the study of short-term visual storage.
Psychological Bulletin, EE, 785—820.

 

Long, G. M., & Beaton, R. J. (1982). The case for
peripheral persistence: Effects of target and
background luminance on a partial-report task.
gournal of Expegimental Psychology: Human
Perception and Performance, E, 383-391.

Long, G. M., & McCarthy, P. R. (1982b). Target energy
effects on Type 1 and Type 2 visual persistence.
Bulletin of the Psychonomic Society, lg, 219-221.

Long, G. M., & Wurst, S. A. (1984). Complexity effects
on reaction-time measures of visual persistence:
Evidence for peripheral and central contributions.
American Journal of Psychology, 21, 537-561.

Mewhort, D. J. K., Campbell, A. J., Marchetti, F. M., &
Campbell, J. I. D. (1981). Identification,
localization, and ”iconic" memory: An evaluation of
the bar-probe task. Memory and Cognition, 2, 50-67.

Miller, G. A. (1956). The magical number seven, plus or
minus two: Some limits on our capacity for
processing information. Psychological Review, g3,
81-97.

Neisser, U. (1967). Co nitivepsychology. New York:
Appleton-Century-Crofts.

Phillips, W. A. (1974). On the distinction between
sensory storage and short-term visual memory.
Perception and Psychophysics, Eg, 283-290.

Sperling, G. (1960). The information available in brief
visual presentations. Psychological Monographs, lg,
(11, Whole NO. 498).

Sperling, G. (1967). Successive approximations to a
model for short-term memory. Acta Psychologica, E1,
285-292.

Stenson, H. H. (1966). The physical factor structure of
random forms and their judged complexity.
Perception and Psychophysics, l, 303-310.

 

 

129

Townsend, V. M. (1973). Loss of spatial and identity
information following a tachistoscopic exposure.
Journal of Experimental Psychology, 2E, 113-118.

Turvey, M. T., & Kravetz, S. (1970). Retrieval from
iconic memory with shape as the selection criterion.
Perception and Psychophysics, E, 171-172.

von Wright, J. M. (1968). Selection in immediate visual
memory. Quarterly Journal of Experimental
Psychology, 39, 62-68.

Yeomans, J. M., s Irwin, D. E. (1985). Stimulus '
duration and partial report performance. Perception
and Psychophysics, 31, 163-169.

 

 

I‘IICHIGAN STRTE UNIV. LIBRARIES
IIIIIIIIIIII IIIIIIIIII III II IIIIHIIIIHII
31293005636414