The effects of display coding factors on observer visual signal detection

MISSING IMAGE

Material Information

Title:
The effects of display coding factors on observer visual signal detection
Physical Description:
x, 105 leaves : ill. ; 29 cm.
Language:
English
Creator:
Montgomery, Demaris A., 1965-
Publication Date:

Subjects

Subjects / Keywords:
Decision making   ( lcsh )
Visual communication   ( lcsh )
Genre:
bibliography   ( marcgt )
theses   ( marcgt )
non-fiction   ( marcgt )

Notes

Thesis:
Thesis (Ph. D.)--University of Florida, 1993.
Bibliography:
Includes bibliographical references (leaves 94-97).
General Note:
Typescript.
General Note:
Vita.
Statement of Responsibility:
by Demaris A. Montgomery.

Record Information

Source Institution:
University of Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 001962551
notis - AKD9225
oclc - 31473024
System ID:
AA00003249:00001

Full Text










THE


EFFECTS


OF DISPLAY CODING FACTORS
VISUAL SIGNAL DETECTION


ON OBSERVER


DEMARIS


MONTGOMERY


A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY


UNIVERSITY


OF FLORIDA


1993














ACKNOWLEDGEMENTS


especially


appreciate


the


encouragement


and


support


offered


husband,


Derek.


When


the


workload


was


heavy,


Derek


remained


patient


supportive


despite


his


own


con-


cerns.


Thanks,


also,


good


friend,


Toki


Sadraloda-


bai,


her willingness


to listen and


offer assistance.


Special


guidance


thanks


Robert


assisting me


with


Sorkin


patience


dissertation,


well


his


encouragement


and


direction


preparing


from


career.


would


also


like


thank


Bruce


Berg


for


his


guidance


learning


the


weight


analysis


technique.


Finally,


would


like


to gratefully


acknowledge


committee


members


, Dr.


Joe


Alba,


. Keith


Berg,


David


Green,


. Keith


White,


especially


David


Green


and


Dr. Keith


Berg


for their valuable


time and assistance with my


study.


This


research


was


supported by


grants


from


the Air


Force


Office of


Scientific


Research.
















TABLE OF CONTENTS


page

ACKNOWLEDGEMENTS........... ................ ...............11

LIST OF TABLES......... ... .. ... ...... . .. .... iv

LIST OF FIGURES ................ ....................... .. .v

ABSTRACT. .. .... ... ......... .. .. .. .. .. .. .. .ix

MATHEMATICAL MODELS. ... ...... ........................... 1

Introduction. ... .................... ....... .......1
Defining the Optimal Observer in Yes/No Detection......4
Defining the Optimal Observer in 4AFC Detection.......15

EXPERIMENT I.. ........................ .......... ...... 21


Introduction.
Method.......
Results......
Discussion...


* a..a. te.....
()(i ii





* ee.......a..


~)I((((i
* .eta
* a a .....
.C.............


EXPERIMENT


2...... ........ . . ........ ........ .65


Introduction .......
Method.............
Results... .........
Discussion.........


...0.0....000..
.... 0...0 a .a a.
...e...... a..a
........O....


.a.*.t.a
.0 ... a.a
......0.
0.....0.


.......... ..65
..... . .72
......... .78
. .. .... .85


GENERAL CONCLUSIONS.. ................ 89

REFERENCES.. ..................................... .. .....94

APPENDIX A YES/NO DECISION STATISTIC..................... 98

APPENDIX B BERG'S THEORETICAL SOLUTION OF THE WEIGHTS....99

APPENDIX C 4AFC DECISION STATISTIC......................101

APPENDIX D RELATIONSHIP BETWEEN YES/NO d' AND mAFC
PERCENT CORRECT ..............................103


21
...29
...37
...59














LIST OF TABLES


Table

1.


page


The means and standard deviations of


five


informational


sources


in which


sources


alternated


in reliability;


the even elements


have


the highest


The stimulus-response matrix


for the 4AFC task.


The sequence





represents


a signal


first spatial


position and noise


second,


third,


and


forth


positions


. . 17


The mnemonics


for the experimental


conditions


found


in experiment


Average observer performance


experimental


for
. .


condition......


Observer weighting


efficiency


.....33


each
........ .38


estimates


stimulus duration and condition....


... .. ..41


Weighting


Efficiency


Estimates


Condition.......


Arrangement
..... .. ... 56


Average weighting


efficiency


estimates


for the


experimental
arrangement,
efficiency


variables


(block-type condition,


and stimulus duration).


measure


is averaged


Each


over the


four observers......................... . .. 57


reliability................... .7













LIST OF FIGURES


Figure


The COSS functions derived from a
observer using an equal, left pan
unequal, right panels, weighting
The top functions with the square
bottom functions with the circles
the signal and noise trials, resp


page


simulated
els, or
strategy.
and the
represent
ectively........10


The weights
depicted in
represent
and noise
and the da
signal and


unequal


derive
Figure
the wei
trials
shed li
noise w


weighting


d from the COSS f
1. The squares
ght estimates for
, respectively.
nes represent the
eights for the eq


strategies,


unctions
and circles
the signal
The solid
average of
ual and


respectively....... 11


Demonstration of the nine graphical
found in experiment 1..............


elements


.....


. ... .25


Trial


sequence


for the


first experiment..........36


Subject Sl
source rel
block-type
the left,
respective
separate s
the solid


Subject Sl
source rel
block-type
the left,
respective
separate s
the solid


's average weights for the f
ability arrangements in the
condition. Panels a, b, c,
right, even and odd arrangem
ly. The smaller symbols are
ignal and noise weight estim
lines represent the optimal


's average weights for
ability arrangements i
condition. Panels a,
right, even and odd arr
ly. The smaller symbol
ignal and noise weight


lines


represent


our
UNcM
and d ar
ents,
the
ates, and
weights..


the four
n the UNcP
b, c, and
angements,
s are the
estimates,


the optimal


e



..44


d are


and


weights....45


Subject S1
source rel
block-type
the left,
respective


's average weights for the four
ability arrangements in the UCM
condition. Panels a, b, c, and d are
right, even and odd arrangements,
ly. The smaller symbols are the







Subject S2
source rel
block-type
the left,
respective
separate s
the solid


's average weights for the f
ability arrangements in the
condition. Panels a, b, c,
right, even and odd arrangem
ly. The smaller symbols are
ignal and noise weight estim
lines represent the optimal


our
UNcM
and d ar
ents,
the
ates, and
weights..


Subject S
source re
block-typ
the left,
respective
separate
the solid


's average w
ability arr
condition.
right, even
ly. The sma
ignal and no


lines


eights for
angements i
Panels a,
and odd arr
ller symbol
ise weight


represent


the f
n the
b, c,
angem
s are
estim


the optimal


our
UNcP
and
ents,
the
ates,


d are


and


weights.


Subject S
source re
block-typ
the left,
respective
separate
the solid


's average weights for the f
ability arrangements in the
condition. Panels a, b, c,
right, even and odd arrangem
ly. The smaller symbols are
signal and noise weight estim
lines represent the optimal


our
UCM
and d ar
ents,
the
ates, and
weights..


Subject S
source re
block-typ
the left,
respective
separate
the solid


's average weights for the four
ability arrangements in the UNcM
condition. Panels a, b, c, and d are
right, even and odd arrangements,
ly. The smaller symbols are the
ignal and noise weight estimates, and
lines represent the optimal weights....50


Subject S
source re
block-typ
the left,
respective
separate
the solid


's average weights for
ability arrangements i
condition. Panels a,
right, even and odd arr
ly. The smaller symbol
ignal and noise weight


lines


represent


the four
n the UNcP
b, c, and d are
angements,
s are the
estimates, and


the optimal


weights... .51


Subject S3
source rel
block-type
the left,
respective
separate s
the solid


's av
iabil
cond
right
ly.
signal


erage weights for the four
ity arrangements in the UCM
ition. Panels a, b, c, and
, even and odd arrangements,
The smaller symbols are the
and noise weight estimates,


lines represent the optimal


d are


and


weights....52


e



..47


e



..49


. ..48








Subject


source
block-
the le
respect
separa
the so


re
typ
ft,
tiv
te
lid


s average we
ability arra
condition.
ight, even a
y. The small
gnal and noi
ines represe


ig
ng
Pa
nd
le
se
nt


hts


ements i
nels a,
odd arr
r symbol
weight
the opt


the


n
b,
an
s
es
im


four


the
C,
gem
are
tim
al


UNcM
and d ar
ents,
the
ates, and
weights..


e



..53


Sub
sou
blo
the
res
sep
the


average we
ility arra
ndition.
ht, even a
The small
al and noi
es represe


ig
ng
Pa
nd
le
se
nt


hts for
events
nels a,
odd ar
r symbo
weight
the op


the
in th
b, c
range
Is ar
esti
timal


four
e UNcP
, and d are
ments,
e the
mates, and
weights....54


Sub
sou
blo
the
res
sep


ect S
ce re
k-typ
left,
ectiv
rate
solid


's ave
iabili
condi
right,
ly. T
signal
lines


rag
ty
tio
ev
he
and


ights for
ngements
Panels a,
nd odd ar
ler symbo
se weight


represent


the four
n the UCM
b, c, and
angements,
s are the
estimates,


optimal


are


weights.


The
and


average
reversed


weights for subject
cue conditions....


the
. .


UCM
. . .61


Demonstration of
four arrangements
used to identify


Demonstration of
demonstrating an
figure produces


the possible patterns for
which observers may have
the more reliable sources....... 63


three line graph arrangements
emergent feature. The middle
an emergent object-like feature..67


Line g
are th
Non-Li
Square
present
angle
figure


raph di
e Linea
kelihoo
(LSQ-N
ted in
subtend


des. F
hood (L
L), and
ays. E
a gray
sted at


igur
IN-L
Lin
ach
gri
the


s a, b
), Line
ar Non-
isplay
and th
bottom


and c
ar
Likelihood
was
e visual
of each


. . .74


Angu
are
Non-
Like
press
angel


lar display
the Object
Likelihood
lihood (ANG
ented in fr
e subtended


code
ikel
OBJ-
NL)
nt o
is 1


s.
iho
NL)
dis
f a
ist


pl
g
ed


figures
(OBJ-
and An
ays.
ray gr
at th


a
LR
gu
Ea
id
e


)
la
ch
a
bo


b and c
Object
r Non-
display
nd the v
ttom of


was
isual
each


--- -


'7e


.

-
.
.
.


...55


I








The observers'


measures


average


performance


for the six arrangements


(d')
in the


Yes/No


task.


Panels


a-c


subjects data


represent


panel


individual


is the average data


The error bars are one standard


error


mean..... ...


The observers'


... .79


average reaction time measures


(measured


from the offset of the mask to


for the six arrangements


in the


response


Yes/No task.


Panels


a-c


represent


individual


subjects data


and panel


are one standard


is the average data.


error of


The error bars


the mean...........


observers'


measures


task.


for the


Panels


subjects data
The error bars
mean......... .


average


performance


six arrangements


represent


panel


are one


d is the
standard


(d')
in the


4AFC


individual
average data.


error


. . . . ..82


observers'


average


reaction time measures


(measured


for the


from


six


the offset


arrangements


the mask to


in the


4AFC task.


response)


Panels


a-c


represent the


panel


individual


is the average data.


subjects data a
The error bars


are


one


standard


error


the mean. ........ ..... 83


...80














Abstract


of Dissertation


Presented


the Graduate


School


the University


of Florida


in Partial


Fulfillment


of the Requirements


for the


Degree of Doctor


of Philosophy


THE EFFECTS OF DISPLAY


CODING FACTORS ON OBSERVER


VISUAL SIGNAL DETECTION

By


DeMaris A.

August,


Montgomery

1993


Chairman:


Robert D.


Sorkin


Major Department


: Psychology


Two


observers'


detection.


studies


examined


ability


The


use


information


the


effects


multiple


display


sources


provided by


factors


visual


given


signal


source


was


represented


as a value


on a graphical


element.


Each


dis-


played


value


was


independent


sample


from


one


two


normal


distributions,


depending


on the


type


trial


(Signal


Noise)


and


the


task


being


performed


(Yes/No


Four-


Alternative-Forced-Choice,


4AFC).


The


first


study


examined


observers


ability


use


differences


source


reliability


performing


Yes/No


deci


sion


task.


The


reliability


the


different


display


elements


was


controlled


manipulating


variance


distributions


from


which


the


element


values


were


sampled








estimated.


Observers were relatively


inefficient at using


reliability


information


forming


two-alternative


deci-


sion


(signal


noise)


Only


when


luminance


cue


source


reliability


was


introduced


stimulus


durations


equal


greater


than


400


was


observer


performance


equivalent


equal


reliability


condition.


The


evidence


suggests


that


luminance


cues


aid


observers


prioritizing


visual


information


sources


according


their


importance


task.


second


arrangement


study


examined


on observers'


effects


performance


display


in both


element


Yes/


and


4AFC


visual


signal


detection


tasks.


The


information


was


displayed


graphically


one


six


formats


constructed


from


display


combination


elements


two


were


factors:


arranged


whether


to produce


not


global


the


feature


that


resulted


from


the


interaction


the


separate


display


elements,


"emergent


feature,


" and


whether


or not


magnitude


this


global


feature


was


monotonically


related


the


optimal


decision


statistic


(for


the


Yes/


task).


results


emergent


indicate


that


feature


performance


Yes/


task


was


was


facilitated


hindered


presence


an emergent


feature


in the


4AFC


task.


Due


the


relatively


high


performance


produced


angular


element

visual


code,

signal


was


not


detection


possible


was


affected


to

by


determine


the


whether


presence













MATHEMATICAL MODELS

Introduction


Every


day


humans are


faced with


uncertain circumstances


which


they


have to


form decisions


based


multiple


sources of


information.


In some situations


these


decisions


have


to be made


rapidly


, possibly to avoid


unfortunate


outcome.


example,


air traffic


controllers


have


detect


and respond


to selected


events


under time stress


order


situations,


avoid

the


potential a

information


aircraft


collisions.


is conveyed


many


decision


maker


via


visual


displays.


As a


result


, researchers


are


interested


determining how


efficiently


observers


can


combine


spatially


and


temporally presented visual


informa-


tion sources,


identifying the


factors


which


influence


overall


processing


efficiency.


current


investigation


examines observers'


use


multiple,


tion


spatially presented,


sources


independent,


in forming detection


visual


decisions.


informa-


Using


Theory


of Signal


Detectability


(TSD,


Green


Swets,


1966)


paradigm


, we


can


specify the


performance


optimal


observer


in different detection tasks.


The


central


theme of


this


investigation was


to identify whether selected


display








The


first study


examines observers'


ability to


combine


nine


independent,


informational


sources


form


Yes/No


detection


decision


(Signal


or Noise).


information


coded as graphical


elements


in the


visual


field,


in some


conditions

Frequently,


sources


decisions


differed


are


based on


their

multiple


reliability.


sources


information


that differ


their


informativeness


reli-


ability).


An optimal


observer


includes


this


information


her


detection decision.


That


she weights


informa-


tion according to


informativeness.


However


, when


decisions


need


to be made


rapidly,


observers do not


always


consider


relevant


information.


The observer may not


consider all


sources


or she may not apply


an optimal


weighing


strategy.


Thus,


main


concern


of this


study


was


determine


whether selected


factors assist


observers


in directing their


attention


to more


reliable


informational


sources.


The


second


study


further


examines


the effects of


selected

Observers


display


formats on


were given


four


observers' detection

informational sources


decisions.


perform


either


Yes/No


task,


first study,


Four-


Alternative-Forced-Choice


(4AFC)


detection


task


Bennett


Flach


studies


(1992)


which suggest


summari


that


results


factors


from a


related


number


display


element arrangement can differentially


affect


performance


these


detection


tasks.


That


, selected


display








Yes/No


task,


attention


, as


than


performance


in a 4AFC


task.


in tasks


This


which


study


require


attempts


focused


iden-


tify


importance


factors


related


to display


ele-


ment


arrang


ement


which


contributing


possible


diff


erences


in performance


between


tasks.


/No


deci


sion


task,


an observer


is given


a sample


independent


elements


. Xn)


1' X2


to decide


which


two


evidence


alternative


observe


events


a given


signal

trial


or noi


, one


two


stimu-


alternatives


is true


each


element


conveys


inde-


pendent


information


about


current


state.


signal


trials


each


drawn


from


a normal


stribution


with


mean


a standard


deviation


On noi
S


trial


each


i is


drawn


from


a normal


distribution


with


a mean


and


a standard


deviation


Alternatively,


in a 4AFC


task,


on each


trial


four


independent


elements


are


present


values


three


sources


are


drawn


from


the


noise


distribution


one


source


value


drawn


from


signal


distribution


The


server


decide


which


source


represents


"signal


" event


Employing


TSD


paradigm


we can


use


information


about


the


underlying


stribution


parameters


to identify


how


optimal


observer


should


perform


in each


these


tasks.


That


, we can


observer


who


identify


only


optimal


limited


performance


the


level


uncertainty


V~ ^.*^-


--


W ,


.L


a








performs


relative


to the


ideal,


and 2)


attempt


to facilitate


observer


ideal,


performance,


by presenting the


them to act


which is generally


information


like a mathematically


Defining the Optimal


Observer


inferior


in a manner which helps


ideal


observer.


in Yes/No


Detection


The Theory


of Signal


Detectability


(TSD,


Green


Swets,


1966;


Green,


1992)


provides


quantitative


model


describing


decisions


based


on uncertain evidence.


Since


normative


theory,


it prescribes


an optimum


means


combining the


information to


observer can base her


form a


decision.


statistic upon which


According to TSD,


opti-


decision


that


statistic


is monotonically


likelihood


ties


ratio


for the current


a likelihood ratio,


related


ratio of


trial


to the likelihood


conditional


evidence,


or some


value


ratio.


probabili-


That


L(x)


= f(xls)


is assumed


that


underlying distributions


are


normal


such


that


the conditional


probabilities can be expressed


f(xln)


[1/(


raA)


]EXP[


Sn)/o )


[1/(


27r2)


]EXP[


s)/as)


, (2)


where a


= 0n


, and


to simplify the derivations


For


n independent sources of


information,


definition the probability


of their joint occurrence


product


P(x,)p(x2)


of their separate


Similarly


probabilities,


likelihood ratio


P(x,


for multiple


(1)








Since


natural


logarithm


likelihood


ratio


monotonic


with


the


likelihood


ratio


the


InL(X)


also


optimal


deci


sion


stat:


istic


Thus


we have


following


equation:


= InL(x,


. xn)


InL(x1)


+ InL(


...+lnL(


. (3)


When


the


definitions


the conditional


probabilities


likelihood


ratios


are


included


equation


and


this


equation


reduced


, it


turns


that


optimal


dec


sion


stati


stic


is a weighted


sum


evidence


see


appendix


derivation),


= xi((Li


((PSI


-2 )/o


where


source


drawn


from


either


signal


distribution


Normal [ ni

distribute


l"ei


Normal[(p


Given


since


si ei ] '


large


the


or the

sample,


sum


noise


distribution,


is also


mutually


normally


independent


random


variable


, and


mean


variance


given


the


alternatives


are


=S l


si- ni)


, E(Z


= -Z


VAR(


= Z((-


si ^ni)


The


assumption


that


on a given


trial


ideal


observer


will


compare


some


deci


sion


criterion,


si- ni)/a


I


n)/0


1


__







When

reliable,


separate pieces of


an ideal


observer


evidence are


is sensitive


not

these


equally

differ-


ences and weighs


informational


sources accordingly.


informativeness,


thus


the appropriate


weight,


given


source


can


be represented by the


statistic


follows:


- mi )


1= ( lsi


Expected


optimal


performance


in a


detection


task is


limited


by the


informativeness


underlying


evidence.


That


observer


performance given a


single source


will


exceed


Based on


equation


informativeness


particular


source can be manipulated by


either


changing


distance


or the


between


size


two distribution means,


the standard deviation,


first study,


6. sis held constant


informativeness of


(6 1


the separate


sources


controlled


changing


Smaller values of


produce


larger


sources.

spending


values


Table

with


thus


lists


a condition


represent


the distribution


in which


more


informative


parameters


the even


corre-


sources,


have


lower variability,


making them relatively more


informa-


tive.


The


the d'


optimal


value.


weight


When ji


for a


is held


given source


constant,


related


iis proportional


reciprocal


the variance


for that source,


6Si ,


. ) ,


, at,














Tabi


The


mean


and


standard


deviation


five


informational


sources


even


in which


elements


have


the
the


sources
highest


ernate


in reliability;


reliability


element 1 2 3 4 5

signal
"s 1 1 1 1 1

as 1.5 0.75 1.5 0.75 1.5

noise
0n 0 0 0 0

0n 1.5 0.75 1.5 0.75 1.5


d' 0.67 1.33 0.67 1.33 0.67







When


an observer's decision


optimal


performance


is based


is expressed


multiple


terms of the


sources,

following


' statistic:


n
r1]
[Cdt] 1


Equation


can,


then,


be rewritten


include


the


optimal


weights


follows:


ideal


aS]


[ (Sa a


~~1].


(10)


weights


pattern requires


10 equals


are normalized and


equal


product of


preceding


optimal


weight across elements


the square-root of


equations allow us


weighting


, then equation


n and


to define


perform-


ance of


an ideal


observer who


is only


limited by the


uncer-


tainty


the evidence.


This provides a standard by


which


can


compare an


observer' s


performance


which


fre-


quently


inferior to


ideal.


For a multiple


observation


task,


some of


loss


in performance may be a


product


observers


using


a nonoptimal


weighting


strategy.


Other


loss


may be more generalized


(e.g.,


some


form of


internal


noise),


showing up


as an overall


performance


loss.


discriminate


the effects of


these


two sources of


error


, we need


iden-


tify


how


observer weights


separate


sources.


technique designed by


Berg


(1989,


1990)


provides a means


estimating the observer'


relative weights.


- ( ^2


(S(6 11


[6 i







which


Berg


refers


as Conditional-On


-A-Single


-Stimulus


COSS


functions


. A COSS


function


is a plot


the


proportion


times


an observer


magnitude


a given


responded


element


"signal"


across


as a


function


experimental


trial


Two


COSTS


functions


are


calculated


each


element,


one


signal


trial


one


noise


trial


Figure


shows


the COSS


functions


derived


from


simulated


data


an observer


using


three


informational


sources


perform


left


a Yes/No


represent


section


an ob


sion.


server


The


using


COSS


functions


equal


weighting


strategy.


observer


COSTS


functions


weights


first


on the


source


right


represent


most


third


source


eas


upper


curves


with


the


square


symbol


each


graph


signal


of figure


trials


represent


lower


curves


COSTS


with


functions


circles


each


graph


represent


COSTS


functions


the


trial


Figure


depicts


weights


three


sources


derived


from


COSS


functions


shown


Figure


small

for


squares


the


circle


signal


represent


trials


the


weight


, respectively


estimates


The


lines


connecting


points


represent


average


weight


estimates.


solid


and


dashed


lines


represent


equal


and


unequal


weighting


strategies


, respectively


comparing


figures


and


, it


can


seen


that


when


the


COSTS


functions


have.


similar


slopes


weights


are


relatively


By





























0 1 2 3 4 5 6 7


Itude of X l


I ude


of X1


2 3 4 5


6 7 8 9 10


Itude


of X2


de of X2


0.20-


-1 a.00
10 .


itude


of X3


Magn itude


of X3


Figure


The COSS


observer using


functions derived


an equal,


left


from a


panels,


simulated


r unequal, r


ight


o/
/



/
/o



a /


0/


07


0.00


Q 10


0. 40


/


I
























0. 8







0.4





0.2




0.0


O --O--
''^
U 0^
D S


0 1 2


emen


Figure
dep
rep


2. The
iected
resent


rights deriv
Figure 1.
weight est


ed from
The squ
mates


the COSS
res and
or the s


functions
ircles
gnal and noise







correspond


with


shallower


slopes


in the


COSTS


functions.


actual


weights


depicted


Figure


are


based


on the


variance


of a cumulative


normal


distribution


which


best


-square


, VAR[Yi]


observer


s COSS


func-


tions


repre


detail


sented


description


solid


of Berg


lines


1989)


figure


theoretical


solution


relative


weights


found


appendix


estimated


variance


is added


variance


the


distribu-


tion


from


which


items


were


sample


Then


derive


the relative


weights


, the


sum


the


variances


each


source


is divided


sum


variances


corre-


spending


with


one


source


to unity,


VAR[Y


+ a 2
ej


Za a
1


(11)


VAR[Y ]k


Finally,


weights


are


normalized


such


that


- 1
1


Note


that


choi


of whi


source


is to be


unity


arbitrary


. That


, the


investigator


should


decide


which


item


t choi


given


hypothesis


that


being


addre


sse


For


instance


, Berg


Green


(1990)


used


COSTS


technique


in an auditory


profile


anal


task.


profile


task


involves


dete


cting


an increment


in the


eve


single


component


(tone)


among


multi


-component


back-


ground.


Given,


an optimal


deci


sion


strategy


that


comp


areas


, to


+ G e
ek







zero


the more


likely


increment was added.


the


weight


assigned


to the signal


component


is set


to unity


then


optimal


weighting


for the


nonsignal


components


should


equal


(n-1)


ment


(where


the element


there are


found


n components)


the center of


For this


visual


experl-


field


fixation point,


was set


to unity.


Given


observers'


estimated


weights


Berg


(1990)


shows


how these weights can be


incorporated


into a


measure


the observers'


weighting performance.


This


measure


the same


as equation


except


the observer


s weights,


are


used


instead


ideal


weights


Ea1]


[(a ao 2)
Ser


(12)


observer applies a nonoptimal


weighting pattern


observer's weighting


performance,


will


lower than


that of


an ideal


observer,


ideaL


Furthermore,


we can obtain a measure of


the


observer's


overall


performance,


on the


task by


calculating


absolute value of the difference between


the


Z-scores corre-


spending


with her hit and


false alarm probabilities on


experimental


trials.


then


thought


If this measure,


additional


of as the effects of


loss


lower


performance


the observer's


internal


than


can


noise,


That


unlike an


ideal


decision maker,


observer


wint
will


often be


less


reliable at


transferring


information


from


, ai,







Finally,


once


three


performance measures,


ideal'


have been derived,


performance


can


summarized


in terms of


an efficiency measure


(Tanner


Birdsall,


1958).


Berg


(1990)


describes


observers'


per-


formance

senting


in terms


three efficiency measures:


the observers overall


performance,


one


another


repre-

repre-


senting


observers


' weighting performance,


third


representing

general meas


residual


ure of


factors such as


the observer's overall


internal


noise.


efficiency,


T obs'


squared ratio of her performance


relative


performance of


ideal


observer,


ideal


That


d' ideal) 2


(13)


the observer


is optimal,


1.0.


Any


decrement


observer' s


performance will


correspond with a


decrease


efficiency


, where


< Tobs


< 1.


The other two


effi


ciency measures


, rwgt and


, allow


us to separate


loss


in observer


efficiency


non-


optimal


weighting


from


loss due


to observer


internal


respectively.


The weighting


efficiency,


like


the overall


efficiency,


Sobs'


is the measure of


the observer's weighting


performance,


relative


to the


ideal


observer,


ideal


who


uses an


optimal


weighting


strategy:


wgt /


d ideaL ) 2


(14)


/noise


accounts


additional


loss


in d'


obs not explained


by the weights,







relationship


among


these


measures


= 7wgt


(16)


*noise


Defining


the Optimal


server


in 4AFC


Detection


Four-


Alternative


-Forced


-Choice


(4AFC)


task,


observer


is given


four


independent


sources


information,


where


each


source


represents


one


two


alternatives,


"signal"


elements


"noise


is randomly


a given


select


trial


to represent


one


four


signal


event.


This


source


value


drawn


from


a normal


distribution


with


mean


a standard


deviation


The


remaining


three


source


values


are


drawn


from


normal


distribution


with


a mean


a standard


deviation


where


observer


s task


identify


which

than


combining


four

the


sources

informati


represents

on to make


signal


a single


res


Rather


ponse


with


the


task


a 4AFC


task


requires


server


independ


ently


assess


each


value


identify


which


source


represents


signal


second


study


, there


are


four


informational


sources


represent


ed by


graphical


elements


located


four


separate


spatial


position


the visual


field.


4AFC


task


there


are


four


oss


stimulus


sequences


,


,





or

n.s>


where





repre-


sents


stimulus


value


first


spatial


ition


three


noi


values


second,


third,


and


fourth


spatial







four possible


responses,





,


,


,


corresponding with


four equally


likely


loca-


tions where


Table


the signal


depicts


can occur.


stimulus-response matrix


decision


task.


The matrix cells


falling


along


minor


diagonal


represent


correct


responses.


represents


total


correct


responses


for the


ith ordering


stimuli,


Sgi.


percentage correct


in a


4AFC task,


P (C),


P,(C)


= ST.,
1


NTota
Total


where NT

lus orders.


total


number


of trials across all


stimu-


Green (

to maximize


1992)


shows


percent


that an


correct


ideal


will


observer


choose


who


attempts


source with


largest


value


since


this


value also has the


largest


likeli-


hood ratio.


To expedite


the derivation,


Green


characterizes


task as detection of


1-of-m possible signals


relative


noise


alone.


Using this approach,


each sequence


would


represented


a separate signal,


e.g.,





Sg .


Thus,


likelihood


that


the evidence


presented


on a


given


trial


represents


signal


com-


pared


to noise


alone may


be expressed


as follows:


l(xlSgi)


= EXP[xi( (Ms-n


sC 2n


/a) ,


where


, and a


e xi


monotonically


related


the optimal


decision statistic


Sg ).


Thus,


, 1(


_


I
















The stimulus-Response
sequence repr
spatial position and
third, and fourth spati


Table
matrix f
esents a
three no
al postit


2.
or the 4AFC ta
stimulus value
ise values in
ions.


sk. The
in the first
the second,


Sgl Sg2 Sg3 Sg4


T1

T2

T3

T4







derivations of the optimal


decision statistic


for an Four-


Alternative-Forced-Choice


task are


found


in appendix C.


alternative calculation of


the optimal


decision


statistic


that considers


yields a slightly


the sequences as


four separate hypotheses


different decision statistic,


the same


decision strategy.


choose the


That


source with


is, an optimal


observer should


largest value since


it also


largest


likelihood ratio.


The accuracy


an observer using this


predicted


optimal


decision


strategy


depends


upon the


probability


that


largest value was


actually


sampled


from


signal


distribu-


tion.


That


P, (C)


depends on the


probability


that


sample


from


the signal


distribution,


, is greater than


the samples


from


the noise distribution,


f(xln)


Consider-


ing two alternatives,


probability that


one


random varia-


larger than another


can be expressed as


follows:


P2(C)


I f(ul s)
J -o


f(vln)
J -0


dvdu.


(19)


Equation


represents


the probability that


the value of the


noise


sample,


less


than the


value of


signal


sam-


ple,


1992).


summed


Since


across all


possible


same probability


values


density


(Green,


functions,


f(x


f(xln),


are


used


to define


the hit


and


false


alarm


probabilities


found


in a


Yes/No


ROC curve


, it


possible


relate performance


in an m-Alternative-Forced-Choice


task to








P, (C)


[1-P


n)]dP


(20)


where


-Pu(S


the complement


the


alarm


proba-


ability,


and


-dP (


the derivative


complement


probability


Appendix


D for


the


derivations


Equation


related


shows


percent


area


correct


under


in a 2AFC


a Yes/No


task


ROC


curve


(Green,


1992


Green


Swet


, 1966)


Finally,


equation


can


be rewritten


to account


multiple


alternati


ves


follows


P (C)


dP (S


where


Thus


poss


to convert


a percent


correct


value


an m-Alternative-Force


d-Choice


task


to a Yes/No


meas-


ure.


Hacker


Radcliff


1979)


public


shed


tables


which


allow


us to make


conversions


from


percent


correct


Alternative


-Forced-Choice


task


/No


This


tabi


takes


into


account


uncertainty


associated


with


larger


numbers


ernatives.


instance


, when


P,(C)


= 0


2AFC


task


.19;


however,


in a 4AFC


task


same


percent


correct,


P4(C


, yie


Finally,


given


the


relationship


between


performance


/No


mAFC


tasks


, maximum


perc


correct


in a 4AFC


task


will


also


limit


ed by


underlying


stribution


parameters


(Mac-


Millian


& Cree


Iman


, 1991)


That


, when


percent


correct







The


preceding


definitions


the optimal


observer


both


a Yes/No and


4AFC detection decision


, provide a


base-


line


comparing


observer


performance


under


different


experimental


conditions.


situations


where


observer


performance


falls


short


of the


ideal,


performance


may


facilitated


presenting the


information


some


manner


which


helps


them to act more


like an


ideal


observer.


studies


optimizing


, to


be described,


human performance.


address


That


this


approach


these studies


look


the effects of


selected display


coding


factors which were


designed


to help


observers


function


as optimal


observers


their detection decisions.












EXPERIMENT

Introduction


Visual


displays


are


commonly used


convey


system


information,


production


such


line,


as air traffic


flow or the


a human decision


maker.


status


complex


visual


display may


include


several


subordinate displays


display


"elements.


Each


display


element provides a


poten-


tial


source


information


for the human operator.


However,


may


impossible


for the operator


obtain


useful


information


from more


than a


few of


the display


elements


one


time.


This


problem may


be minimized


the operator can


priority


display


elements


terms


their


criticality


informativeness,


operator


can


allocate


attention


accordingly.


This


study


examined


several


factors


that affect an operator's


ability


to allocate attention


to display


elements


that are differen-


tially


informative.


previous


experiment


(Sorkin


Mabry,


Weldon,


Elvers


, 1991) ,


observers


examined a


multi-element


display


and


then


reported whether the display represented


the occur-


rence


a signal


or nonsignal


event.


Using


technique


derived


from the Theory


of Signal


Detectability


(TSD,


Green


Swets,


1964),


Sorkin


et al.


estimated


importance








making


a detection decision


(Berg,


1989,


1990).


optimal


decision-theoretic


observer weights


input


from


each


element


according to


the element


informativeness


reliability;


highly


reliable display


elements are


weighted


more


highly


in the detection decision


than


less


reliable


elements


(Durlach,


Braida,


Ito,


1986;


Berg,


1990;


Berg


Green,


1990).


the Sorkin et al.


(1991)


study


display


elements


were


equally


informative;


hence


, each


element


should


have


been weighed


equally


the observers'


decisions.


When


observation


durations


were


long,


weights


were


equal


across


spatial


array


of display


elements.


However


, when


the observation durations were brief


the display


coding


was


complex,


the highest decision weights


were


associated


with


display


elements


the center of


visual


field,


around


the observer'


fixation point.


The extent


which


weighting


functions were peaked


corresponded


with


performance


level


(low


performance


was


associated


with


peaked


functions)


Sorkin et al.


(1991)


concluded


from


these results,


that


under difficult conditions,


observ-


er's


allocation


of attention was


restricted


central


portion of


the display.


This


interaction between


availability


of information


the difficulty


from different


task


regions


display


is not


c ri


c ina


A number


a-C IH ~ ~ J


variables


are


a& 4


%








include


the number


Wickens,


1988)


(Perrott et al.;


irrelevant,


1991)


and spacing


or distracted,


items


(Andre

found


Wickens


visual

1987;


field,


Legge,


type of

Luebker,


display

1989;


code


Sandersol


(Boles &

n, Flach,


Buttigieg,


Casey


, 1989;


Sorkin et.


, 1991) ,


display


item


intensity


(Eriksen


Rohrbaugh,


1970)


, and


task


com-


plexity


(Williams,


1982).


When


the stimulus durations


the Sorkin et al.


(1991)


experiment


were


long


(more


than


500 ms),


display


element


weights


were


equal


, indicating that


observers


could


process


information


from all


regions of


the display.


Since


reliability


of all


elements was also equal,


an equal


weight


strategy


was


optimal


for that


task.


important


question

when the


is whether


reliabilities of


observer can employ

the elements are


optimum

not equal


weights

across


the


visual


array.


Obviously,


the ability to match


decision


weights


observer


element


is to prioritize


reliability


the display


necessary


elements according


their


importance


task.


When an


informational


source does


provide a


consist-


ent


report


of an unchanging


event,


the source


is not


reli-


able.


instance


, if


a sensor measures a


specific


lumi-


nance


value


to be


x at one


time and


x+ n


subsequent


reading,


sensor


showing variability


measure-


ment.


Thus,


th i s


sensor would be


less


reliable


than


one


__


II-








greater


weight on the more


reliable


source.


However,


evi-


dence


suggests


that people


tend


to overrate


importance


of unreliable sources


(Schum,


1975).


Wickens


(1984)


states


that


when


equally


people are confronted


informative,


they


with sources


perform the


which are


task


sources were equally reliable.


The


present study addressed whether observers


can


use


differences


display


element


variability


identify


source


reliability


use


this


information


forming


simple


Yes/No detection decision.


In addition,


this


study


was


designed


imposes


"overhead"


factors could


to determine whether using


significant


amount


the observer,


reduce


this


additional


and whether


related performance


selected


loss.


information


processing


display

in Sorkin


et al.


(1991),


the observers


in the current


study performed


a multi-channel


visual


detection


task.


On each


trial


of the


experiment,


observers were


nine display


presented with a


elements.


The display


display


consist-


elements


were


nine


array


graph


pendent,


signal


vertical


(see


gauges,


line-graph


figure




normally


trial,


gauges arranged


horizontal


The values displayed


.., x9


distributed,


the values


line-


were determined


random


inde-


variables.


the nine elements were


selected


from a


distribution with a mean of


and a


standard


devia-


tion of as.


On a


noise


trial


, the values were drawn


from











































Figure


-


Demonstration


nine


graphical


elements


found


experiment








whether


data


displayed had been


generated


from


signal


or no


distribution.


The


reliability


of different display


elements


was


con-


trolled


by manipulating the variance of


the


distributions


from which


the element values


were


sampled:


high


reliability


elements were


sampled


from distributions with


lower variance


than


low reliability


elements.


That


, a source


high


reliability would be analogous


to an


instrument which


shows


consistent measurements across

reliability would not provide


time whereas a source


consistent measurements.


variance of


a display


element


at a particular position


was


same


elements


signal


depending


and noise


on the


trials,


experimental


differed


condition.


across


Table


illustrates


the mean and standard deviations


that could


employed


five element


display


in which


even


elements alternated


their


level


of reliability.


Berg


(1990)


found


that


reliability


of elements


in an


auditory task


similar to the one


used


this


study


could be


used


observers when


the most


reliable


tones


were


much


louder


than the


less


reliable


tones.


The


loudness cue


was


much


less


effective when


reversed,


in which


case


louder


cue


indicated a


lower reliability.


Berg's


results


suggest


that


under some conditions cuing


element


reliability


(e.g.,


with


intensity


or color)


may


observers


accurately


weighting


display


sources


by their


importance.








specific


items


in a


display.


For


instance,


researchers


have


found


that


correct


utilization of


color


coding


(Christ,


1990;


item


showed


Fisher


in a

that


Tan,


1989)


can


reduce search


display. Furthermore, Wic

color coding a particular


time


in locating


kens and Andre


item


(1990)

object


display


leads


improved accuracy


in recalling the specific


value associated with


that


item relative


to a


monochromatic


display.


Thus,


given Berg's


results and


the evidence


cited


above,


we predicted


that


observer weighting


efficiency


present experiment should be higher


for a


condition


which


luminance cue


signals


the element


reliability.


order


test


the efficacy


cue


element


reliability


present


experiment,


the spatial


position


the high reliability


display


elements was


randomly varied


over


trials.


The overall


luminance of


the display


element


varied


in accordance with


reliability


(high


or low)


element.


We expected


that luminance would


provide


natural


code


allocating


observer attention


hence


weight,


case,


to the high reliability


the efficiency


elements.


observers'


If that


weighting


were


strategy


would be much higher


in a


cued


than


in an uncued


condition.


The duration


the stimulus and


the spatial


arrangement


the


element


reliabilities also


should


influence


how


efficiently the observers match


The rPesn1 ts


their weights


from the Sorkin at


the element


a .


(1991n


reliabilitiF. -








reliable


, graphically


coded display


elements.


However


it is


possible that sensing the element

entially weighting the elements,


reliabilities and


may require


differ-


some additional


processing


steps


or "overhead"


an observer.


duration


of 233-ms may be at


the margin of


an observer's ability


extract


differences


information needed


in element reliability.


to discriminate


example,


employ


slower,


serial


search


may be


required


to extract


reliability


information


and weight


the elements accordingly.


that


case,


it might be advantageous,


when


processing


short


dura-


tion stimuli,


ignore


reliability and differential


weight-


information.


Our


experiments


tested


three


levels


stimulus


duration


(150


400


800-ms).


We expected


that


weighting


durations


efficiency


(400-ms


would be greatest


800-ms


long


and very poor


stimulus


shortest


duration


(150-ms)


Observer sensitivity to element


reliability


also may


affected by the spatial


If attention


arrangement


of element


is distributed more effectively


reliability.


among


spatially


contiguous


than separated


items,


grouping


sources


similar


reliability


should


aid performance.


Posner,


Snyder,


Davidson


(1980)


found


that simple


reaction


time


to detect


light at a


second most


likely position was


facilitated


only


when


this


item was adjacent


to a


cued


location


(the


most


likely target


location).


When


the second most


likely


posi-








should


better


for displays with


elements


grouped


similar


reliabilities


than


displays


that


distribute


element reliabilities


across


the array.


Finally,


we were


interested


in whether


observers


would


sensitive


reliability


individual


elements


without


cues


to element


reliability.


That


can


observers


estimate


(and


employ)


information


about


element


reliability


based


only


trial-by-trial


variability


readings


from


individual


display


elements and


feedback


about


events?


To answer that question,


added


conditions


in which


relationship


between


element spatial


position


reliability was


fixed,


rather


than


random,


over


a block of


200


trials.


If the observer


can


estimate


variance of


the element readings


from the


first


k trials


a block,


the observer may


able


to partition


ele-


ments


into those with high and low reliabilities.


that


process


assi


gnment of higher weights


more


reliable elements,


the observer's weighting


efficiency would


be enhanced


that


condition.


Method


Subjects


Four

corrected


University


normal,


of Florida


visual


students


acuity


with


participate


normal,

d in


this


study.


One


subject,


, was


later discovered


color


deficient,


and another


, S4, was highly trained


on the


task.







Apparatus


and Stimuli


Observers were seated


in a sound


isolated booth approxi-


mately


inches away


from a


10.5


inch


color monitor


(EGA)


driven by


mum


an 80386


contrast,


computer.


intensity was


The monitor was set


set at


maxi-


approximately


cd/m2


, measured


from a


inch by


inch


uniform


white


field.


On a


given


trial,


nine gauges were presented


monitor;


subtending a horizontal


by vertical


visual


angle of


approximately


Each gauge was composed


parallel


white


lines,


with


tick-marks


falling


equal


intervals


left


line


conditions


except


luminance


cue


condition.


For this


latter


condition


high


reliability


gauges were white and


the remaining gauges


were


gray.


The


intensity


the white gauges was


approximately


cd/m2


intensity


the gray gauges was


approxi-


mately


cd/m2


measured


from


7.5


inch by


inch


uniform


white


and gray


fields,


respectively.


Each


tick-mark represented a


display


increment


1.0,


ranged


from 0.0


to 10.0.


longer


blue


lines,


located


near


tick-marks,


indicated


positions


of the


signal


noise distribution means.


The value displayed by


each


gauge


was


"signal"


determined by


or "noise"


sampling a number


distribution,


depending


from

on the


either

type


trial.


This


number was


converted


to the


vertical


displace-


ment


of a horizontal


white


line


from the bottom


(e.g.


zero







trial


The


mean


gauge


values


on signal


trial


was


equal


to 5


mean


on noise


trial


' Mn


was


equal


The


standard


deviation


gauge


values


on signal


trials


depended


on the particular


experimental


condi-


tion.


There


were


five


diff


erent


element


reliability


condi-


tions


: (1)


Equal,


Group


ed-Left-High,


Grouped


-Right-


High


, (4)


stributed


-Even


-High,


and


Distributed-Odd-


High


. In


Equal


condition


, the


standard


deviation


of all


gauge


elements


was


equal


to 1.


the


Grouped


-Left


-High


condition


, the


standard


deviation


four


left


elements


was


equal


to 0


, and


five


right


elements


was


equal


That


pattern


was


reversed


the


Grouped-Right


-High


condition.


stributed


-Even


-High


condition,


standard


deviation


four


even


elements


element


, and


was


equal


to 0


standard


deviation


the


remaining


elements


was


equal


1.5.


the


Distribut-


ed-Odd


-High


condition


the standard


deviation


the


five


elements


element


, and


was


equal


.85,


the


standard


deviation


remaining


elements


was


equal


1.3.


The

different


unequal

trial


reliability


block


conditions


conditions


were


: Pure


run


under


Block


two

xed


Block.


Pure


Block


condition


display


stri-


bution


parameters


were


xed within


a block


of 200


trials.







relationship


between


element


reliability


and spatial


posi-


tion was


fixed


throughout


the block


trials.


the Mixed


Block


conditions,


trials within a block


trials


alternated


randomly


among the


Grouped-Left-High,


Grouped-


Right-High,


conditions.


Distributed-Even-High


Distributed-Odd-High


the Mixed Block conditions,


would


impossible


an observer to


identify the


reliability


any given spatial


element,


unless


the observers were


provid-


with an additional


trial-by-trial


cue to


element


reli-


ability.


three


Finally,


levels of


trial


stimulus duration


block conditions


(150,


were


400,


tested at


800 ms).


duration of


stimulus


presentation was


synchro-


nized


between


with


refresh


traces


was


traces of


approximately


the monitor.


ms.


The


The


period


onset


offset of


the display was delayed until


a retrace was


ready


to occur.


Once


controlled


stimulus was presented


counting the number


corresponded with


of refresh


the duration


t


selected stimulus duration


races

(150


was


which

400,


or 800 ms).


experimental


conditions are shown


table


mnemonics


in each


table cell


describe the


trial-block condi-


tions.


three


trial-block


conditions


which


contained


elements which differed


in reliability across


spatial


posi-


tions


are denoted by the


letter U


the mnemonic,


meaning


sources were


unequal


in reliability.


The


equal


reli-










The mnemonics
experiment 1.


Table 3.
experimental


conditions


found


UNCUED CUED

150 400 800 150 400 800

LEFT

RIGHT
MIXED BLOCK ----UNcM UCM
ODD

EVEN


PURE


BLOCK


LEFT


RIGHT


ODD


EVEN


UNcP


EQUAL RELIABILITY ENcP







were


run


three


leve


stimulus


duration.


addition


within


each


unequal


reliability


trial


-block


conditions


the four


element


reliability


arrangements


were


presented.


The next


letters


mnemonics


represent


whether


a luminance


cue


was


present


= cue


cue)


Finally


, the


ast


better


or M


denotes


manner


which


element


reliability


arrangements


were


sent-


The


repr


esents


a pure


block


design


which


arrangements


remained


constant


across


experimental


trials


a given


block


, and


repre


sents


a mixed


block


design


which


arrangements


varle


across


trial


Thus


, the


mnemonic


UNcM


stands


an Unequal


reliability


cue


Mixed


block


design.


Procedure


Observ


ers


were


told


to make


their


decis


ions


based


on the


leve


the


markers


The


gauges

y were


relative


told


signal


to rank


and


likelihood


noise

that


mean

the


evidence


represent


a signal


using


II 3 II"


I i 2 11
/ *


keys


, where


repr


esent


very


sure


was


signal


represent


noise.


fact,


observers


tended


only


use


two


middle


keys


Thus


, responses


on keys


were


combined


to represent


noise


, and


responses


keys


responses


were


combined


to represent


signal


data


analy


ses.


On conditions


where


reliabilities







The


trial


sequence,


shown


in figure


proceeded


follows.


First,


observers were given a


0.5"


fixa-


tion


cross


the center


the display


ms.


This


was


replaced by the nine


line-graph


gauges


stimulus


duration


either


150,


400,


or 800


ms.


Following


stimulus


white blanking mask was presented


ms.


Then,


the


display


was


completely black


second,


which


time


observers were allowed


respond.


Any


responses


made


prior to or


following this


period were


dis-


carded


given


Within a


Response"


feedback


trials.


the center of


given session,


Finally,


the display


an observer ran


the observers were


ms.


10 blocks


through


trials.


Across


sessions


there were


1500


trials


(750


signal


noise)


collected


for each


condition.


Due


time constraints


imposed by the need


collect


multiple


trials,


some of


the observers


received


less


prac-


tice


than


others.


Subject S4


was highly practiced.


ran


through at


prior


least eight


to collection


practice sessions

the experimental


for each

trials.


condition

Subjects


and S3


were highly practiced on


the


Yes/No detection


task,


but


they


only ran


through


one


practice


session


each


individual


conditions.


control


any possible


practice


effects


experimental


sessions,


each


observer received


different


nrmT-i


ri n' +i r n


/*^*rr r n 7 r


fniir "1- 1


hln r


cnif rh


fh a-


IIll
















Fixation:


200


Stimulus
Duration


or 800 ms


Mask:

200 ms

Response
Duration:
1000 ms

Feedback:

250 ms


White Field



Black Field



Correct


Figure


Trial


sequence


first


experiment.







Results


Average


observer


performance measures


experimental


conditions are shown


table


In order


consider differences between


the equal


and


unequal


reliabil-


block-type


conditions


the data


were


collapsed


across


source


reliability


arrangements


for the


unequal


reliability


conditions.


An analysis of variance


performed


on the


aver-


showed


a significant main


effect


stimulus


duration


(F(2


,6)=12.49,


0.01).


Performance


improved


stimulus duration


increased.


of block-type condition


(F(3


There was


,9)=5.285,


also a main


.05) .


effect


four


observers


showed greater performance


in the equal


reliabili-


condition


relative


the


two unequal


reliability


condi-


tions


which


did not


include a


luminance cue


more


reliable


sources


(UNcM and UNcP).


An analysis


performed


on the observers d'


oab measures


unequal


reliability


effects


conditions


the experimental


indicated

variables


significant

(block-type


condition,


stimulus duration,


and arrangement)


, and


their


interactions,


except


for the


three-way


interaction.


Howev-


only a


of these differences were evident


the data


individual


ance


subjects.


improvement


observers


stimulus


showed a


duration


perform-


increased


(F(3


,6)=13. 66,


p0.01).


There was also a


performance advan-


tage


for the cued block-type condition,


UCM,


relative


to the










Table


Average
condition.


observer


performance


each


experimental


UNCUED CUED

150 400 800 150 400 800

LEFT 1.92 2.11 2.12 2.07 2.37 2.40

RIGHT 1.77 1.89 2.08 1.90 2.21 2.34
MIXED BLOCK
ODD 2.06 2.22 2.28 1.92 2.06 2.14

EVEN 1.93 2.02 2.18 2.01 2.20 2.23


PURE


BLOCK


LEFT


.05


RIGHT 1.68 1.92 2.04

ODD 1.94 2.07 2.06


EVEN


.98


EQUAL RELIABILITY 2.24 2.44 2.57


OBSERVERS








Finally,


there was consistency among the observers


two of


the


interactions,


as well.


For the


four arrangements


there


was a


tendency


for observer performance


to be


highest


when


the most reliable sources were grouped


four


left


spatial


positions


stimulus


durations


greater.


Alternatively


, at


the shortest stimulus


duration


performance


was


highest when


the most


reliable


sources


were


distributed among the odd spatial


positions.


effects


arrangement


also depended


on the


particular


condition.


general,


observers


showed


a performance advantage


arrangement


over the other three


arrangements


two,


luminance cue conditions


(UNcM and UNcP).


Alterna-


tively,

observer


the grouped

performance


eft arrangement

in the mixed


yielded

block,


the

cued


greatest

condition


(UCM).


Moreover,


the right arrangement


tended


to show


poorest


performance


in the


UNcM and


UNcP


conditions,


but


relatively


high


performance


UCM condition.


these


differences


were


found


significant


through


subsequent


paired


comparisons


using


a Tukey test.


Thus,


the evidence


from these analyses


indicated


that


stimulus duration and block-type condition had a


consistent


effect on


observer's performance.


In addition,


the arrange-


ment of

ance.


source reliabilities


However,


influenced


the direction of


observers'


perform-


effects depended


htd nn4-o1-Vt nnir-*'n nnl+rnri i -in ron


E h i Tnii 1 n c /^n =i+ ; rnn ar^


Def o TmFnrn Pnr








unequal


reliability


patterns alone,


performance


was


best


when


sources


high


reliability were


cued


grouped.


However,


performance was


also


relatively


high


arrangement


UNcM condition.


Since


location


reliable


elements


in the


visual


field affected


performance


differently


under


specific


conditions,


logical


suspect


that


the differences


in performance


are


related


observers'


weighting


strategies.


For


example,


when


observ-


ers are under time constraints


or there


is uncertainty


about


location


efficient


of reliable sources,


at applying weights


observers


appropriate


may


to the


less


weighting


strategy


selected.


Weight Analysis


Observers'


weights


were estimated


using


Berg's


(1989,


1990)


Conditional-On-A-Single-Stimulus


COSTS )


analysis


technique,


based


best


described


the slopes


Chi-Square


earlier.


The estimated


of cumulative normal


with


weights


functions


the corresponding


that


COSS


were

had


func-


tions.


Two weight


estimates were calculated


each


ele-


ment


each


condition,


one


for signal


one


noise


trials.


this


analysis,


out of


2808


COSS


functions


Chi-


square


fitted


to cumulative normal,


6.7%


significantly


differed


The weights


<= 0.05)


reported


from


are


the observers'


average of


COSS


signal


functions.


and noise


weight estimates.








Table 5.
Observer Weighting Efficiency Estimates
Duration and Condition.


for Stimulus


Subject


Stimulus
400


Condition


Duration
800 Average


ENcP
UNcM
UNcP
UCM


ENcP
UNcM
UNcP
UCM


.650
.540
.675
.735

.720
.485
.510
.607

.710
.657
.495
.715


ENcP
UNcM
UNcP
UCM


ENcP
UNcM
UNcP
UCM


Avg:


.940
.792
.727
.900


ENcP
UNcM
UNcP
UCM


.750
.619
.602
.739


0.780
0.645
0.755
0.835

0.820
0.563
0.617
0.740


0.830
0.688
0.650
0.758


.960
.745
.770
.958


.848
.660
.698
.822


0.850
0.683
0.797
0.850


.790
.633
.627
.795


.860
.705
.680
.802


.970
.780
.740
.960


0.868
0.700
0.711
0.852


.760
.622
.743
.807

.777
.560
.585
.714

.800
.683
.608
.758

.957
.772
.746
.939

.823
.660
.670
.805


Equal reliability No cue Pure block design (ENcP)
Unequal reliability No cue Mixed block design (UNcM)
Unequal reliability No cue Pure block design (UNcP)
Unequal reliability Cue Mixed block design (UCM)







first ANOVA described


earlier.


Again,


for the


three unequal


reliability


conditions


the efficiencies


represent


aver-


four arrangements.


Across


observers


conditions

duration


weighting


increased.


efficiency


increased as the


To identify whether


these


stimulus


differences


were significant,


a Monte Carlo simulation was


run to


esti-


mate


the expected variance


Swgt


The sampling distribu-


tion


of wgt which best


reflected


observers


weighting


efficiencies was used


to identify


significant differences.


The


criterion selected


significant differences


was


standard


distribution.


deviations


Differences


= 0.04)


weighting


this


efficiency


sampling


which


exceeded


0.08


were


identified as


significant


differences.


This


was a


fairly


conservative estimate,


considering


that


these


distribution


parameters corresponded best


with


data


from


poorest observer.


Given


this


criterion,


weighting


efficiency was


significantly


greater


stimulus


durations of


400 ms or higher,


efficiency was


highest at


ms.


Thus,


significant


improvements


in weighting


effi-


clency

overall


least partially


accuracy


account


found when stimulus


for the

duration


improvement

increased.


Moreover,

efficiencies


with


differences


for the


the differences


among the


observers


block-type conditions


observed


are


in the overall


weighting

consistent

measures.


Performance


was


greatest


ENcP and


UCM


conditions







maintained


at all


stimulus durations


fairly


consistent


across


four


observers.


Since


performance


was highest at


longest


stimulus


duration,


differences among the conditions were consist-


across


levels


of stimulus


duration,


figures


depicting


reliability


separate observers'


conditions are based


weights


on the data


unequal


obtained at


stimulus


duration,


only.


Figures


5-16


show


observers'


average


weights


for the


four


arrangements


source


reliability


three


unequal


reliability


block-


type


conditions.


There are


three


figures


representing


three


block-type conditions


(UNcM


, UNcP,


and UCM)


each


observer.


In each


figure


there are


four graphs


representing


four source


reliability arrangements


, where a,


represent


left,


right,


even


and


arrangements,


respectively


The


larger symbols


are


the


average


weight


estimates


the smaller symbols


are


the signal


noise


weight


estimates.


The solid


line represents


optimal


weights


for the separate arrangements.


All


four


observers


show


similar


changes


their weights


across


three


conditions,


where


their weights


best match


the


ideal


weights


for the UCM condition.


Table


lists


the observers'


weighting


efficiencies


conditions


ms duration


found


is shown.


figures


Table


5-16.


shows


Again,


the average


only the


weighting














0.35


0.30


0.25
J
-c
C)
a 0.20
2


0,
7 0.15
C




0,00
8' .10

4,8


0 1 2 3 4 5 6 7 8 9


at a l Pos


of ua


uge


Spoa t a


IL tIon


o f Gau


., 35


0.30-


0.25-


0.20-


0. 15-


0.10-





0.00-


O. 35


0.S30


O.25-


0.20-


oT 0, 15
0 10
(.






e.00


+

/


t ia Po


tL Ion


Gouge


Ltal Po


i tL on


of Go


Figure


Subject Sl


reliability


tion


Panel


s average weights


arrangements
s a. b. c. a


for the


four source


in the UNcM block-type


nd d are


L'-A A


left.


riaht.


condi-
even


0.25


0.20


C







+




0-
0 0
+


6j


uge


I Pos






































0 1


9 10


cO 0. 15
0
L
C
> 0. 10e
4-


0.05


0.00


0 1 2 3 4 5 6


Opot


t Pos


of Go


uge


So at to


i Pos


i t Ion


of Ga


0. 30-


0. 25-


0.15-


0.10-


0.05


0.00


I I I I IT -
1 2 3 4 5 6 7 8


0.30


0.25-
01
(-

a 0.20-


0o 0.15-

0
e.


0.00


otio


os it Lon


of Gaug


Spoa


ta I Pos


of Go


Figure


Subject Sl


s average weights


for the


four source


reliability


arrangements


the UNcP block-type condi-


tion.


Panels a,


and d are the


1 left.


ri cht


even


10


uge


uge


.


, V






















3.20-


0. 15-


0. l10-


S0.05-


0.00-


0.35-


0.30-


0.25


0.20-


0.15


0.10-


0.05-j


0.00


I Pos


It on


of Gaug


Spa t a i Pos


of Go


0.35


0 1 2 3 4 5 6 7 8


0.35


9 10


0 1 2 3 4 5 6 7 8


9 10


tL o Po


Figure


it ion


Subject Sl


reliability


tion.


of Goug


L ta Pos


s average weights


arrangements


Panels a, b, c


in the UCM


and d are the


r


for the


it ion


of Go


four


block-type


left,


right,


ugo


source
condi-


even


a



A A











+
A


uge




































1 2 4 5 6 8


9 10


0.25


a 0.20





S0, .10-


0.05-


0.00-


a t t I Pos


i t ",on


of J uge


a t ao i Pos


It ton


of Gou


0. 30


9 10


0 1 2 3 4 5 6 7 8 9 10


Spot


lal Po


of Ga


uge


latL


I Po


i tIon


of G


auge


Figure


Subject S2


reliability


tion.


s average weights


arrangements


Panels a.


S. -. a


for the


in the UNcM block-


nd d are


the left.


four source
type condi-
riaht. even


0.20


0 1


L


__













o. 35 -





3.25-


0. 20-


8.8-*




0. 05-
0. 10


2 I I 7 8 9
2 2 4 5 6 7 8 9


0.35-


0. 30


0.25-


0.20-


0.15-


0.105


0.05-


0.00


O/
C O
+- j -
J

i"


atao


I Pos


of Gau


Ltl I Pos


I t (on


of Gau


3. 35-


0. 30-


0.15-


0.10-


0.05-


0.00


0.35-


0.30-


0.25-


0.20-


0.15-


0.10-


0.05


0.00


of Go


uge


SLta Pos


i t Ion


ougq


Figure


tion.


Subject S2's average weights


arrangements


Panels a, b, c,


for the


four source


the UNcP block-type condi


and d are


left,


right,


even


t j Po


reliability



















0, 25


S0.20-


n0


S0.10


0.05-


0.00


A A


SAt


A A


0.00


o f bo


uge


SO o I aI F


I t


of Ga


9 10


it ion


of Ga


uge


t ilI Po


ItL on


of Ga


tion.


Subject S2's average weights


arrangements


Panels a,


for the
block-t


and d are the left,


ri


four source
ype condi-
aht. even


-S


D a t


uge


t L I Po0


9 10


Figure


uge


reliability


in the UCM


, v














0.353





0.25

0-20
q, 0.20-


**
4-*


0. 35


0.3025-
-1


-c A
-S

^ 0.20-



oAEJ
@I






0.05-


0.00-4


aL it P


os it o n


of Ga ug


atla


I Pos


It. on


of i' u


353

0. 30-


0. 25-
C
Cr

- 0.20

L-7
0.15-





0.05


0.004


WV


0.30


0.251


S0.20


oi 0.15
CD




0.05-


0.00


of Go


reliability


average weights


arrangements


Panels a, b,


for the


fou


in the UNcM block-type


and d are


left,


riqht


r source
condi-


even


t ; 3 ^


So t I l Po0


It-


o f 1o


K,


Figure


uqO


Subject S3's


tion.


*) o 0


r


I,





















. 2)5


0.20


0 1 2 3 4 5 6 7 8


a 0.20-
1


9 10


+

O O



otti


i Po s


of anuge


Llai P


os ttt


of Go


0.35-





0.25


0.20-


0.15-


0. 10-


0,05-


0.00-


1 I I I T
2 3 4 5 6 7 8 9


0.25-


0.20-


0.15-


0. 10-


0.05-


0.00


Sp otta


I Po


itLo


of Go


uge


tial Po


It L on


of Ga


Figu


re 12.
reliabil


tion.


Subject S3's average weights


ity


arrangements


Panels


for the


four source


in the UNcP block-type condi


and d are


*tha


1 a -,


Ltat *-t t- I i-a -


ric oht -


even


uge


uge


0.25^


U U. .















0. 35
I
-A
O. 30i

0.25-1
- A


: 3.20-i

n 4,
I'
0 .1
m 0. 10


9 10


0.35j


0.301


) 0.25-i
r 1



-c Isi
o -4
CT) 01


0. 10J


0.05o
-I
i
-J
ooa io


a i0


Lio tI


i t


.DNQ


t II Po


s it Lon


o0f :


i-
3.30-


3.25

rQ
0 0.20
-7

C 0.15
0
L .
> 0. 10-
<
I


p
/ ,


0.35


0.30


0.25


0.20


0.05-
4 -

0. 00
0 1 2 3 4


t 1 I Po


i t ion


Subject S3


reliability


tion.


uge


L t aI Po


s average weights


arrangements


Panels a, b, c,


the UCM


and d are


It ton


for the


of o~


uge


four source


block-type condi-
left, right, even


uge


Figure














0.35 !
-4
0.301


S0.25-
r 4

" 0.201
- 1
0


0
0 0
4-


0 1 2


9 10


0.00


0 1 2 3 4 5


L i oI R s


it ion


cof Ua


uge


Lt l i Pos


tLII n


of Go


9 10


0.25


0.20


0.15


0. 10-a


0.05-


0.004


A
/ \ /

/ I / A


o t a


i Po


of bo


Subject S4's


reliability


tion


uge


Spo t


average weights


arrangements


PanPels a.


kI -. -


ila Poa


it ion


for the


in the UNcM block-t


and are the left.


of Go


uge


four source
ype condi-


0.10


*0


ugs


Figure


riaht


Seven


.














3a,35-


0.30-


0.25-


0.20-


0. 15-


0. 10-


0.05-


0.00-


0.25


aC 0.20-
3

oe 0.15-
o
.-


0.05


+
[] [


L1 a i ros


of Gouge


atlo! PoHs


SL It on o "


ouge


0.35-
a, 3-


0.30


0.25-


0.20-


0.15-


0.10 -


0.05-


0.00


0.35-
-


0.30-


0.252
Cr

S0.20-


na 0.15-

.DJ


I
0.05 -i


0.00


SLl I Po


ItL on


of Ga


uge


Lt I Po


of Oa


Figure


Subject S4's average weights


for the


four source


reliability arrangements


in the UNcP block-type condi


tion.


Panels a,


and d are


left,


riaht.


F V -


even


c













J \\

//A
/ *3\ C+ U

11; 4


uge


+/o


v w














0.35-r


-4
0.30


S0.25

-J0 _2
, 0.20


i 0, 15

Q .1
i^* 1v'i !-


4
Ai


0 1 2 3 4 5 6 7 8 9 10


attla Pos


itia on of


0 1 2 3


o f Ua


9 10


uge


0 1 2 3


9 10


1
0 1 2


Lt i l Po


of bo


uge


LlaI Po


L iLon


tion.


Subject S4's average weights


arrangements


Panels a,


in the UCM


and d are


the


for the


block-type


left,


four source


condi-


even


0. 35


3.25


9 10


of Uo


uge


Figure


reliability


right,


I











Table 6.
Weighting Efficiency Estimates for Arrangement and
the Unequal Reliability Trial-Block Conditions.

Subject
Arrangement UNcM UNcP UCM


Left
Right
Even
Odd


0.880
0.820
0.690
0.800


.940
.940
.790
.730


Left
Right
Even
Odd


.650
.560
.630
.690


0.620
0.490
0.640
0.760


0.860
0.850
0.780
0.690


Left
Right
Even
Odd


Left
Right
Even
Odd


Avg:


Left
Right
Even
Odd


.700
.570
.750
.700


0.710
0.780
0.730
0.900


.700
.623
.683
.795


0.770
0.780
0.810
0.850


.840
.730
.600
.790


.760
.653
.670
.763


Equal reliability No cue Pure block design (ENcP)
Unequal reliability No cue Mixed block design (UNcM)
Unequal reliability No cue Pure block design (UNcP)
Unequal reliability Cue Mixed block design (UCM)











Table 7.
Average weighting efficiency estimates for the
experimental variables (block-type condition,
arrangement and stimulus duration).

Stimulus
Duration Arrangement UNcM UNcP UCM


Left
Right
Even
Odd


Left
Right
Even
Odd


.640
.520
.643
.673


.668
.585
.660
.728


0.633
0.545
0.620
0.610


.823
.757
.743
.635


0.730
0.675
0.658
0.730


.845
.890
.815
.740


Left
Right
Even
Odd


.700
.623
.683
.795


.760
.653
.670
.763


.885
.885
.833
.805


AVG


Left
Right
Even
Odd


.669
.576
.662
.732


0.708
0.624
0.649
0.701


.851
.844
.797
.727


Equal reliability No cue Pure block design (ENcP)
Unequal reliability No cue Mixed block design (UNcM)
Unequal reliability No cue Pure block design (UNcP)
Unequal reliability Cue Mixed block design (UCM)







shows higher weighting


efficiency than the other two


condi-


tions,


UNcM


and UNcP,


each


arrangement


stimulus


duration,


except when


the most reliable elements were


found


positions.


Table


indicates


that


four


observer


show


the same


pattern of


effects at


stimulus duration.


They tended


to show a


little more


varla-


ability,


similar patterns across


subjects


the


shorter stimulus durations.


As with


the d'


data


we see


a slight,


but not


signifi-


cant,


advantage


for grouped arrangements


in the UCM


condi-


tion


for the odd arrangement


UNcM


condition


relative


the other


arrangements.


The


interaction between


stimulus


duration and


arrangement


found


in the


data


was


supported by differences


observers


weighting


efficiencies.


From table


we also


see


that


less practiced


observ-


ers


tend


to show


the


lowest


weighting


efficiency


non-cued


conditions,


UNcM


UNcP,


for the


right arrangement.


The


same


is true


for the shorter


durations.


However,


once


the cue was


introduced,


efficiency


was


indistinguishable


for the weighting


efficiency


left


arrangement


Finally, as far

were concerned,


which


was


as the observers'


the only


consistently


residual


fairly


high.


efficiency,


consistent difference


among


observers


was


tendency


for residual


efficiency







lus duration by


arrangement


interaction


found


the observ-


ers'


measures was


not driven exclusively


changes


rather by


a combination


these


fects.


Discussion


The primary

the effects of


goal


selected


this in

display


vestigation was


factors


determine


in directing observ-


ers'


attention


informative sources.


There was an


overall


improvement


in observer performance


stimulus


duration


increased;


this


was mainly


function


improved


weighting


efficiency.


In general,


when no additional


cues


source


reliability were available,

est when sources were equal


weighting e

rather than


efficiency was

unequal in r


great-


eliabil-


ity.


There was a


tendency


for better performance when


more


reliable


sources were grouped,


rather than


distributed


the cued block-type condition,


especially when


stimulus


duration was at


least


400 ms.


Alternatively,


the no


cue


block-type conditions


performance was


highest


when


observers


were presented a


ences


distributed


odd arrangement.


in performance were mostly


These differ-


differences


found


in the observers'


weighting


efficiency measures.


When


sources


have


to be prioritized


terms


underlying


statistical


properties


of the


information,


this


study,


estimate


these


people are


properties


limited by both


(e.g.


their


, variability


ability


source


rlnoi se '







conditions may have been due


their


inability to


estimate


variability


sources when


this


information


was


relevant


their decisions


.g.,


UNcP


condition).


That


, the observers may not have been


sensitive


trial-to-trial


in weighting


condition,


toward


variability


efficiency,


of the sources.


given the


indicates greater


these elements.


The


luminance cue


attention


weight


It does not necessarily


improvement


UCM


directed


suggest


that


observers


have


improved sensitivity to the


differences


element


reliability.


test


this


possibility


, observer S4,


contributed


data


to an additional


condition


in which


gauge


luminance


was


inversely


related


to gauge


reliability.


Figure


shows


subject S4


s average


weight


estimates


luminance cue conditions,


four reliability patterns,


UCM and reverse


Grouped-Left and


cue,


-Right


Distributed-Even


direct


-Odd.


relationship between gauge


circles


luminance and


represent


reliability,


UCM condition.


condition.


triangles


The weights estimated


represent


for these


two


reverse


cue


conditions


are


nearly


identical.


This was


reflected


weighting


efficiencies.


largest difference between the


lumi-


nance cue conditions


in weighting


efficiency was


0.02;


the Grouped-Left pattern


= 0.97


for the UCM condition,


= 0.95


for the


reverse cue condition.


Thus,


least


on the


initial


trials


in a


given block the observer























0 A 2
A


S A
A

0 1 2 3 4 5


9 10


9 10


Spo t


lal Po


si ,


otf ba


bpoto La


0.30


0. 25


0.20


Pos t


of G


auge


0.15


1 2 3 4


9 10


0 1 2 3 4 5 6 7


9 10


Spot


I al Pos


it Ion


of Go


uge


Spo t


toi Po


i t ion


of G


Figure
and


The average weights


reversed


for subject S4


the UCM


cue conditions








However,


there


remains a


question


as to whether or


observers


were actually using trial-to-trial


variability


sources to make


their decisions.


Some of


observers


showed


fair


amount


accuracy


weighting


sources


according to


reliability


the UNcM condition where


trial-


to-trial

sources


variability

were more re


could not be


liable.


used


One possible


identify


which


explanation


this

terns


performance

produced


is that observers were s

by gauge markers when the


sensitive


variability


pat-

was


low.


a given


trial


, the markers of


high


reliability


gauges


tended


to fall


at a


common


vertical


position


display


causing


them


to line


as shown


figure


Thus,


this


pattern


may have


drawn


observers


attention,


helping

After


them to identify which sources were more

questioning the observers about strategies


reliable.


used


these


trials,


such


it was confirmed


patterns


in the displ


that observers

av. Based on


were

this


sensitive

evidence


alone


is not


conclusive


that


observers were


sensitive


underlying


variability


the sources.


Rather


more


probable


that cues


provided and


patterns


inherent


in displays


helped observers


to weight sources according


reliability.


Finally,


especially


good performance


arrangement


raises another


question.


Why


is performance


_ -










Grouped









Grouped


-Left


-Right


High


High


Distributed-Even


l L


Distributed-Odd


High






K


High


Figure 18
four


Demonstration


arrangements


identify


more


which ob
reliable


ser


possible
vers may


patterns


have


used


sources.







the arrangement.


The odd arrangement was the only


condition


with


five


sources,


rather than


four


, which


were


high


reliability.


a result,


it had


more


high


reliability


sources distributed


throughout


the visual


field,


was


only


arrangement


that had a source high


reliability


located at


fixation


point.


Secondly,


in order to


main-


tain


equal


levels of predicted


ideal


performance,


there


were smaller differences between


the variances


for high


reliability


sources.


Thus,


based


both


these


factors, t

reliability


his condition most


condition.


closely


If observers


approximated an


resorted


equal


weighting


sources equally when


they


are


under time


stress or unable


identify


which


sources


are


reliable,


this


strategy


would


prove most useful


condition.


conclusion,


from


this


study we can


state


that


when


observers


have


to utilize


information


from multiple


subordi-


nate displays


they


are


relatively


inefficient at


identify-


ing differences


in reliability among the displays.


However,


there


coded


improved


luminance.


efficiency when


The assumption


the display


that


elements


this


are


cue,


possibly


other


cues


, help observer's


prioritize


information by


indicating where


attention should be


direct-


Additional


assistance may be gained by


organizing


displays such


that


sources


similar


in reliability are


proxi-


mate


to one another.













EXPERIMENT


Introduction


This

ability


study


use


continues the


investigation


independent visual


information


observers'

sources ir


forming


detection


decisions.


Since


humans


are


often


required


to make decisions


under time stress


many


real


world settings


, researchers have been


interested


identi-


flying means


of coding visual


information


to reduce


potential


errors and


optimize human performance.


One approach


assist performance by


creating


display


codes which


capital-


ize


on our


knowledge of human


sensory


and perceptual


mecha-


nisms.


For


instance


, Woods,


Wise,


and Hanes


(1981)


reduced


complexity


values


of integrating multiple


form a detection decision by


independent


combining


sensor


display


elements


into a single geometric


form.


This allowed


human


monitors


use shape distortions


identify


important


system states.


The


primary


concern


of this study


was


determine


whether


two


factors


related


to display


element


arrangement


affect


observers'


detection deci


sions.


The


first


factor


concerns


the


influence of


emergent


features


observers'


detection


decisions.


Emergent


features


are


defined


.I


I


1


I







(Treisman,


1986).


instance,


if the elements are


repre-


sented


three


line


segments,


then


depending


arrangement chosen we could


figure


create one of


forms shown


Particular element arrangements produce


features


such


as angles and


intersections which are


observable


given


individual


lines.


Moreover


, some element arrange-


ments


produce


global


features


which


are


recognizable


objects.


instance,


first and


last element


arrange-


ments


in Figure


as the


There


triangle

is mixed


19 do not possess as strong


found


an object quali-


the middle.


evidence


the object perception litera-


ture


regarding whether


emergent,


object-like,


features


simple


element arrangements


facilitate or hinder


detection


the


dence


underlying

suggesting


elements.


Some studies have


an "object-superiority


effect.


found


That


evi-

is,


when a


target


feature


(e.g.


a line segment


of a given


orien-


station)


is embedded


in a


contextual


pattern,


observer detec-


tion


performance


context


form


facilitated when


a recognizable object


target


(Weisstein


feature


Harris,


1974).


Similarly,


Ankrum


and Palmer


(1991)


found


that


observers


were better at detecting


differences between


element


arrangements


which


formed


objects


than


between


element


arrangements


in which


one was an


object


and


other


was


part


of an object.


This


enhanced


detectability


may


be related


familiarity


organized


pattern




















































Figure


Example


demonstrating


three


an emergent


line-graph
feature.


arrangements


The


middle


figure


produces


an emergent


object-like


feature







effect"


where


is eas


ier to detect a


specific


letter


word


than


in a nonword


Reicher


, 1969).


Others


have


found


contradicting


evidence


Pomerantz


1981)


points


to evidence suggesting


that


when


elements


perceptually


group,


the emergent


feature


created


configuration may be more


perceptually


salient and selective


attention to the


underlying


elements may


impeded


(Pomer-


antz


& Schwaitzberg,


1975)


Similarly,


Navon


(1977)


found


Stroop


interference of


global


configurations


subj ects'


processing


Flach


local


(1992)


elements


summarize


but not


results


the opposite.


from


Bennett


number


studies


which have applied


this concept


to real


world


set-


tings.


These studies


indicated


that an emergent object-like


property


had no affect


or adversely


affected,


perform-


ance


in tasks


requiring


focused attention


the


individual


elements.


However,


they pointed


that


performance


been


facilitated


by the same emergent


features


when


detection


task required


information


integration.


hypothesis,


tested


in earlier


studies,


stated


that


magnitude of


performance advantage


integration


tasks,


relative


to selective attention tasks,


depended


upon


degree


to which


the element configuration


produced


object-like


feature


(Carswell


Wickens,


1987;


Wickens


Andre,


1988).


The degree of


"obj ectness"


depended


on wheth-


or not


the element configuration possessed


enclosed







emergent


feature


carried


important


information


about


underlying


state,


rather than simply the object


quality


configuration.


instance,


Buttigieg


and


Sanderson


(1991)


found


that object displays did not always produce the


best


performance


in integration


tasks,


whereas


"well-mapped"


emergent


features did.


However,


recent


investigations,


designed specifically to


address


relationship between


the configuration and


task,


have


produced mixed results.


Some


researchers


have


found


support,


suggesting that performance was


better


when


there was

emergent


a strong relationship between some


feature and


property


the decision statistic than when


this


relationship was weak


(Bennett


, Toms,


Woods,


1993;


Mitch-


Biers,


1992;


Schmidt


Elvers,


1992).


Others


researchers


did not


find a


performance advantage


emer-


gent


features


that were


"well-mapped"


the deci


sion


task


(Sanderson,


Haskell


Flach,


1992).


The


second


factor addressed


this study


concerns


importance of


relationship between the emergent


feature


the optimal


decision statistic


for the


task.


One


nice


characteristic


of the Theory


of Signal


Detectability


(TSD,


Green and Swets


, 1966)


paradigm


that we


can


mathemati-


cally


specify the optimal


decision statistic


in a


detection


task.


The


optimal


decision statistic


likelihood


ratio


or some


value which


is monotonically related


I a -*


I







optimal


observers can


decision


use changes


statistic,


expected


in the magnitude of


this


that


emergent


property to make


their decisions.


this


study,


observers


were


presented


four


independent


informational


sources


coded


as graphical


elements


visual


display


Six display


codes were constructed


from


combination


of two


factors,


whether


or not


display


element

whether


arrangement


or not


produced an emergent


the emergent


feature,


feature had some property


2)

that


was directly related


to the optimal


deci


sion statistic


(for


Yes/


No task).


Observers used


this


information


per-


form either

The ma


normal


Yes/No


gnitude


random


task or


a 4AFC task.


a given source was


variable which depended


determined


underlying


state


(Signal


or Noise).


signal


events,


source


values


were


selected


from a


distribution with a mean


and a standard deviation


values of


noise sources


were


drawn


from


a distribution with


a mean


and
n


standard


deviation


where un
n


< 4 and as


= an


the


Yes/No


task,


observers had


decide


whether


source


values


presented


on a


given


trial


represented


either


a signal


state or noise.


4AFC task,


the observer


had


decide


which


four


elements


represented


sicrnal


event.


emergent


features


facilitate


processing







tasks


when


such


features


are


present.


Alternatively,


these


emergent


features


are


processed


faster


than


the


under-


lying


elements


then


deci


sion


tasks


which


require


sensi-


tivity


to the


underlying


elements


may


be hindered


display


codes


that


ssess


these


features.


For


instance,


equal


reliability


Yes/No


task,


sensitivity


the


separate


underlying


than


elements


in a 4AFC


is of 1


task


ess


importance


As a result,


performance


the


emergent


feature


hinders


observer


sensitivity


the


underlying


elements


, performance


on a 4AFC


task


may


be 1


ess


effi


cient


when


the


information


is arranged


form


an emergent


feature


than


when


element


arrangement


does


not


posses


this


feature.


With


respect


second


factor


the


relationship


between


the


magnitude


some


emergent


feature


property


optimal


deci


sion


statistic


important


then


should


see


performance


advantage


when


this


relationship


present


the


current


study


, thi


relationship


was


coded


only


the


Yes/No


decision


task.


Two


the


six


display


arrangements


produced


an emergent


feature


in which


the


width


area


this


feature


was


directly


related


optimal


was


deci


expected


sion


that


statistic


detection


/No


performance


task.


would


Thus


, it


facilitated


the


/No


task


given


ese


display


codes


relative


the


other


codes which


do not


poss


ess


relationship.


--


r- I


1


I







arrangements


produced


emergent


features which


possessed


enclosed contour,


whereas others did not.


the configural


property


of the display arrangement


is an


important


factor


(Carswell


Wickens,


1987) ,


then display arrangements


which


possess


this property may


emergent


be more


feature effects


likely to show the expect-


than arrangements


without


enclosed


contour.


Method


Subjects


Three of


four subjects who participated


in the


first


study


also contributed


data


this


study.


observers


were


paid an hourly wage


plus a bonus


based


on performance.


Apparatus and Stimuli


Observers were seated


in a sound isolated booth approxi-


mately 27


inches away


from a


10.5


inch


color monitor


(EGA).


The monitor was


for maximum contrast.


Intensity was


approximately


cd/m2


measured


from a


uniform


white


field


covering the monitor.


On a


given


trial,


one


six


arrangements of


four graphical


elements was presented


on the


monitor


against


a gray grid.


The


maximum


horizontal


vertical


visual


angles were


13.5


respectively.


(The


separate


measures


of visual


angle


each


displays are


found


on figures


The values


depicted by the graphical


elements were either drawn


from


signal


noise


distribution,


depending


on the


trial







were


= 40 and


respec-


tively


Element


magnitude


ranged


from


to 99.


Figures


depic


three


line-graph


display


codes


which


element


magnitude


was


coded


the


length


(number


of pixel


of a hori


zontal


line


segment.


Figure


repre-


sents


the


linear


likelihood


(LIN


-LR)


arrangement.


LIN-LR


arrangement


there


was


fixe


separation


between


one


line


segment


beginning


next.


Thus


, in


task


total


ength


play


produced


the separate


segments


varied


directly


with


likelihood


ratio.


Figures


represent


ver-


sions


linear


non


-likelihood


arrangement


both


cases


the


onset


eac


line


segment


began


spec


location


visual


field


; total


display


length


vary


one


case


elements


were


arranged


hori


zontally


other


case


they


were


arranged


square


(LSQ


-NL)


, to control


differences


vis


angle.


Figures


depict


three


angular


displays


angle


formed


line


segments


in a given


quadrant


was


directly


related


magnitude


the


underlying


element.


end


of each


line


segment


was


fixed


in position


the


display


S The


opposite


ends


two


segments


joined


form


an angel


The small


arrows


in figures


desig-


nate


angel


being


described.


The


angle


varied


with


element


magnitude


follows:


-NL) ,








Linear


Likelihood


(LIN-LR)


Arrangement


es/No


Signal


Yes/No


H-


Trial


Trial
Trial


K-


Average


Visual


Angle


: Hori


zontal


= 6


Vertical


Linear


Non-Likelihood


Horizontal


(LIN-NL)


Arrangement


Yes/No


Signal


Trial


Yes/No


Trial


F-


K-


H-


Average


Visual


Angle:


Hori


zontal


11.25


Vertical


Linear


Yes/No


-Likelihood


Signal


Square


(LSQ


Trial


-NL)


Yes


Arrangement


/No


Trial


H-


H-


Average


Visual


Angle


: Hori


zontal


= 6.2


Vertical


= 6


Figure


Line


graph


display


codes


Figures


b and


are


hood


ear


(LIN-NL),


displays.


Each


Likelihood


Linear


display


(LIN


Non


was


Linear


-Likelihood


sented


Non


Square


in front


-Likeli-
(LSQ-NL)


a gray


. .


I


-LR),


* 1


*


i


11 _







Object


Likelihood


(OBJ


-LR)


Arrangement


Yes


/No


Signal


trial


Yes/No


Noise


trial


Average


Visual


Angle:


Horiz


ontal


Vertical


= 6.3
= 6.3


Object


Non


-Likelihood


(OBJ-NL)


Arrangement


es/No


Signal


trial


Yes/No


Noi


trial


F-'


Average


Angular


Visual


Non


Angle:


-Likelihood


Horizontal
Vertical


(ANG-NL)


= 6.3
= 6.3


Arrangement


/No


Signal


trial


Yes/No


Noi


trial


C- ^-^


Average


Visual


Angle


: Hori


zontal


= 6.3


Vertical


Figure
the


Angular


display


Likelihood


(OBJ-NL),


Angular


codes.


(OBJ


Non


-LR)


Figures
Object


-Likelihood


b and


c are


-Likelihood


(ANG-NL)


i---n frnnt arr


displays.
ann r


~c~-~c~


r^ i cnl si


Ta^Hh


Y^VQCnt +-Qn


a rtrr~


uraC







21a,


the smaller


element values correspond with angles


which


point

21c,


toward

smaller


the center

element


the display.


values correspond


In figures


with


angles


which


point downward


two of the


figures


(21a and


21b)


the


elements


were


arranged


to form an enclosed contour,


producing


object-like


shapes.


In figure


element arrangement produced


object


which


the area


was


directly


related


likeli-


hood


ratio


for the


Yes/


task


(OBJ-LR).


That


Area


Figure


SSlOOx..


represents an object-like display that does


have


a property


(OBJ-NL) .


related


Finally,


figure


the optimal


21c depicts


decision


ANG-NL


statistic


display


which


identical


the OBJ-NL display;


however


, it


does


not posses a


continuous enclosed


contour.


Procedure


Yes/No


task,


observers were told


to make


their


decisions


based


on the average magnitude


the gauges


, and


rank


likelihood


that


evidence


represented


signal by using the


I, 1 3


, "2


" and


keys of


101-key


keyboard,


where


represented


"very


sure


it was a


signal"


represented


"noise"


Again,


and


responses


were


identified


identified


as noise,


as signals


responses


the analyses.


4AFC task,


were


observ-







Subj ects


indicated


the


signal


location


using


either


same


keys


found


in the


/No


task,


or the


"Ins"


, "Home"


"Del"


"End"


keys


, depending


on the


element


arrangement.


The


trial


sequence


was


same


as the


first


study


see


Figure


However


, instead


of nine


line-graph


gauges,


one


display


arrangements


described


above


were


pre-


sented


duration


a stimulus


was


duration


controlled


of 2


same


manner


S(The


stimulus


as described


experiment


Each


observer


received


eight


blocks


practice


each


display


arrangements


tasks


before


experimental


data


was


coll


ected.


both


practice


experimental


trial


, the


display


arrangements


were


randomly


prese


nted


each


observer


received


a differ-


random


order


a given


session,


a subject


ran


through


eight


blocks


task and


eight


blocks


4AFC


task,


they


contributed


data


to eight


blocks


one


task


before


beginning


the


next


task.


The


order


tasks


alternated


across


experimental


sessions.


Since


performance


in the


initial


experimental


blocks


was


nearly


eal,


a random


noise


pattern


was


added


plays


degrade


performance.


The


random


noise


pattern


consi


sted of


white


spots


, two


pixels


in width.


each


trial


locations


spots


were


randomly


determined.


Thus


, the


pattern


varied


across


trials


This


noise


pattern


overlaid


the


graphical


elements


background


grid,


such







Approximately


the grid region was


covered


random noise pattern.


Subjects performance


in the


Yes/No task


was


poorer with


random noise pattern


than without.


Overall


efficiency


dropped


size of


pattern


approximately


16% when


the differences


of effects did not


random noise was


increased.


change.


However


Thus,


added.


, the general


experimental


trials


included


random noise pattern.


Results


Yes-No Task


The


observers'


accuracy


(d')


mean


reaction


time


measures

are shown


for the


six display


figures


arrangements


in the


, respectively.


Yes/No task


Three


panels,


in each


figure


fourth,


represent the

is the three


individual

observers


subj ects

average


' data,

data.


Each


of the subjects d


' and


reaction time measures are based


on eight blocks of


trials,


and the error bars


represent


one


standard


error


the mean.


Separate


repeated measures ANOVAs were performed


observers'


eight


display


d' and reaction


trial blocks


arrangement


time measures,


There was


,10)=11.263


collapsed


an effect


, <0.001


across


type


observer


accuracy


(d's).


Subsequent analytic comparisons,


using


pooled


variance


as the error term,


indicated that


perform-


ance


was


greater


element


arrangements


that


produced


emergent


features


(F(1


,10)=23.2,


S5o0.001)


relative


to those













ec t S1


I i


LIN-NL LS-.L LIN-LR


ANG-N OBJ-L OCBJ-LR


LIN-NL LSO-L LIN-LR


ANG-NL OBJ-NL


OBJ-LR


oct S3


erage


or mn


2.00-




1 .50H








-l
0.50-



0.00-


LIN-NL LSOQ-NL LIN-LR


ANG-MN OBJ-N 08-JLR


LIN-L LSO-MNL


LIN-LR ANG-NL OBJ-NL


GBJ-LR


Figure


The


observers


' average


performance


measures


the


SIX


arrangements


in the


Yes/No


task


Panels


a-c


2.00


1 50


1 .00


I li
T -* -


L UL


!


.













ect S 1


SS2


2300.0


200.0-





100.0-



-.

0.0-


300.0





200.0





100.0





0.0


LIN-NL LSO-NL


LIN-LR ANG-NL OBJ-L 08J-LR


LIN-N LSQ- NL


LIN-LR ANG-NL CeJ-NL


OBJ-LR


oC c I b


ernge


orma n ce


or S


300.0


200.0


100.


300.0-





200.0-





100.0-





0.0


LIN-NL


LSQ-NL


LIN-LR ANG-NL 8OJ-N OBJ-LR


LIN-NL


LSO-d


L:N-LR


ANG- NL


OBJ-LR


Figure


The


observers


' average


reaction


times


measures


(measured


from


offset


mask


to response)


the


siX


conditions


in the


Yes/No


task.


Panels







performance.


plays,


However,


performance was


for the non-emergent


significantly better


feature


(EF(


dis-


,10)=30.8,


0o.001)


for the ANG-NL display relative


LSQ-NL displays.


This


the


LIN-NL


performance advantage among the


non-


emergent


feature displays may


angular element


be a


code, especial


function

ly since


the


the


underly-


difference


between


the ANG-NL and OBJ-NL display


arrangements was


significant.


The


effect


likelihood


ratio


manipulation


was


significant


only


line-graph


displays


(F(1, 10)=21.499, E 0. 001)


Performance


in the


LIN-LR


condi-


tion was better than performance given


the other two


linear


displays


, LIN-NL and


LSQ-NL.


Finally


, there was


signifi-


cant difference among the observers'


reaction


time


measures


(E(5,10)=13.4,


p 50.001)


for the separate element


arrange-


ments.


Reaction


time was


slower given a


LSQ-NL display


code


relative


to the other


element arrangements.


4AFC Task


Figures


25 depict


the observers'


accuracy


reaction


time measures


for the


six display


arrangements


4AFC task.


Again,


three


panels,


a-c,


in each


figure


represent


the


individual


subj ects


' data


and


the


fourth


panel,


the average


for the


three observers.


Each


subjects


reaction


time measures


are


based


eight


blocks


100 trials,


the error


bars


are


one













Sub ject


2.00


1 .50




1 .00-




0.50




0.00


A


LIN-NL


LSOQ-N


LIN-LR


ANG-t OBJ-NL-


OBJ-LR


LIN-L LSQ- I


LIN-LR ANG- N


OB3-NL


OBJ-LR


Ject


Average


Per forman


for 31-2


2.00:




1.50+




S1.00


LIN-NL LSCQ-I LIN-LR ANG-NL OBJ-NL OCJ-LR


LIN-NL LSQ-NL LIN-LR ANG-NL


OSJ-LR


Figure
for


observers'


av


the six arrangements


erage
in th


performance
e 4AFC task.


(d')


Pan


measures
els a-c


2.00


1.50@


1 .00


Sub ject


_ _












ub ]ec L


200.0





'00.0





0.0


300.0





200.0





100.0





0.0


LIN-NL LSQ- L


LIN-LR


ANG-NL OBJ-NL


LIN-L LSO- N


LIN-LR


ANG-M NCBJ-NL


OBJ-LR


verge


Per formnce


for S1-3


300.0


200.0





100.0





0.0


300.0





200.0





100,0




a0.0


LIN-NL LSO- NL


LIN-LR ANO-NL OBJ-NL BY-LR


LIN-t


LSO-QL


LIN-LR


ANG-NL


OBJ-tN


OBJ-LR


average


from the offset


reaction t
the mask to


4AFC task.


times measures


response)


Panels a-c


repre-


sent


aa-


individual


subjects and nanel


d is the


average


I







I
-O
0 *


Figure


(measured


The observers


the six conditions


ub Jec t







Again,


separate


repeated measures ANOVAs were


performed


the observers'


' and reaction


time measures,


collapsed


across


eight


trial


blocks.


There


was


significant


effect


display


arrangement


(F(5,10)=4


134, '


<0.05)


observers'


accuracy.


Subsequent analytic comparisons,


using


pooled variance


for the error term,


indicated


that


there


was


an overall


difference between


displays


with


without


emergent


features.


However


, there was


a performance


advantage


that


(F(1,10)=5.63,


produced an emergent


P <0.05)


for display


feature with an


arrangements


enclosed


contour


(e.g.


displays)


relative


to an emergent


feature without


this


property


LIN-LR)


There


was


also


significant


(F(1

the


,10)

two


=11.75,

line-grap


E <0.01)

h display


difference between


arrangements


the ANG-NL


that did not


possess


an emergent


feature


(LIN-NL and LSQ-NL)


, though.


Given


this


latter


difference,


these effects may


best


characterized


terms


differences between


line-graph


angular


element


coding.


Based


on the


accuracy


data


alone


observers


tended


show poorer performance


when emergent


features were


present.


observers


display


showed


arrangement


lowest performance


and highest performance


for the


for the


LIN-LR


ANG-NL


display arrangement.


In addition,


among the angular display


codes all


observers


showed


lowest performance


for the OBJ-LR


arrangement.


TV-- A-'-- -I-A---


_ 1-_ 1_-_ ^ ^_ -. 1- ^ -1 -


__


TTA----





i_ *. l ^


TT _


_ ...







analysis


of reaction


times


indicated that


reaction


time


was


faster


(F(l,10)=11.6


, <0.01)


for display


element


arrange-


ments


that produced


emergent


features


than


those


that


possess


these


features.


There was evidence


speed-


accuracy


tradeoffs


among the


separate


linear


angular


displays.


For


instance


, among the


linear displays


the


LIN-


arrangement


showed


less


accuracy,


faster


reaction


times


than


the other two


line


graph


displays.


However,


there


remained a


performance advantage


angular


element


coding


compared


LIN-LR display


arrangement.


Discussion


There


is a


great


deal


interest


from both


theoretical


practical


perspectives


in how human detection


perform-


ance


affected by


element configuration.


One of


main


issues


concerns whether


emergent


features


produced


se-


elected


element arrangements help or hinder processing


underlying


elements.


Evidence,


so far,


suggests


that


impact on


performance may


depend upon which


feature


most


salient


(Pomerantz,


1981)


and how well


this


feature


relates


to the


task being performed


(Buttigieg


Sanderson,


1991)


This experiment


compared performance


in Yes/No and


4AFC


detection


tasks,


for different arrangements


line


element


components


of simple


visual


displays.


From


Yes/No


task data,


it was


found


that


observer


accuracy


was


affected


the display


arrangement


when o


observers


were


4 S -


* I*







emergent


feature


(LIN-LR),


performance


in this condition


was


indistinguishable


from the angular display


code arrange-


ments which


consistently produced superior performance.


Alternatively,


same


feature


appeared


hinder


performance


in the


4AFC task,


which


required


focused


atten-


tion to the separate elements.


That


accuracy was


rela-


tively


low and


reaction


time was


higher


for the


line-graph


display


arrangement


which possessed


emergent


feature.


Similarly,


there was


a tendency


for poorer performance given


emergent


angular display


, OBJ-LR,


relative


non-


emergent


angular display


ANG-NL.


This


pattern of


effects


would


expected


attention


automatically


directed


toward


the


capacity


has


emergent


to be


feature,


invoked


and


to gather


additional


information


processing


from


underlying


elements


(Navon,


1977;


Pomerantz


Schwaitzberg,


1975).


Although


there


no strong


evidence


suggesting


effect


of the


relationship between


the emergent


feature


decision


statistic,


is not


possible


completely


rule out


this


factor.


instance,


in the


Yes/No


task


emergent


feature advantage observed


for the


LIN-LR


arrange-


ment relative


the other


line graph displays could also


explained


emergent


true because


as a difference due


feature and


relationship


the decision statistic.


LIN-LR arrangement


possessed both


between


This


factors,







that


could draw


conclusions about


which


factors


were


influencing


performance.


exercised when


Furthermore,


interpreting the


some caution


results


since


should


poor


performance observed


may be


related to masking


LIN-LR display


effects of


in the 4AFC


processing nearby


task

items


in the


visual


field


(Eriksen


Eriksen,


1974)


Wickens


others have argued


that


this


pattern


effects may


be explained


in terms of


the proximity


compati-


ability principle


(Andre


Wickens


1988;


Carswell


& Wickens,


1987;


Wickens,


1992) .


According to the


proximity


compati-


ability principle,


tasks


that


require


integration are


better


supported by


display


arrangements which have high perceptual


proximity;


whereas,


tasks which


require


focused


attention


are better


tual

ences


supported by arrangements


proximity.


which have


there were more distinguishable


among the angular element display


arrangements


percep-


differ-

which


produced


consistently


high performance,


it may be


possible


further


eliminate


alternative hypotheses.


This


raises


another question.


Why was


performance


good


for displays composed


of angular


elements?


One


expla-


nation


for this


angular display


code advantage


is that


is easier to extract magnitude


information when


coded


as changes


length


size


of an angle


of a line segment


That


than when


is coded


the angle may


empha-


size


scale of the


underlying


element magnitude


Alter-







angle


to make


their


deci


sons


instance,


observers


may


have


identified


angles


pointing


toward


center


display


(OBJ-LR)


or downward


(OBJ-NL


or ANG


-NL)


repre-


senting


"noise


oppos


indicating


"signal"


events


Thus


their


deci


sons


would


have


been


based


binary


information


rather


than


actual


magnitudes


underlying


elements


simple


conduct


a study

those


test


latter


in which


used


poss


tasks


prec


eding


ibility


displays

study.


would


are

Howeve


identi-

r, the


distribution


paramet


ers


would


be manipulated


so that


one


case


observers


could


use


direction


angle


other


cases


they


could


not.


instance,


under-


lying


signal


noise


distributions


had


relatively


small


large


means


then


most


angles


would


be either


small


large,


respectively


Thus


, the


direction


angles


would


be 1


ess


useful


forming


a detection


decision.


observers


are


using


magnitude


angles


, there


should


no different


ces


among


selected


pairs


means


long


as the


stance


between


distribution


means


standard


deviations


were


held


constant.














GENERAL CONCLUSIONS


We investigated


whether selected display


coding


factors


would


assist


observers


in visual


signal


detection.


The


factors


investigated were


partially


selected based


on knowl-


edge of the predicted


optimal


observer as defined


the TSD


paradigm

studies


(Green

suggests


Swet


potential


1966) .

means


The evidence

for coding d


from


displays


these

that


will


assist observers


in forming detection decisions.


first study


, observers performed


Yes/No


detec-


tion


task where


separate sources differed


reliabili-


The


main concern was


to determine


whether


observers


included


these differences


in source reliability


their


detection decisions.


Evidence


from research


on human


deci-


sion making


suggests


that


when


informational


sources


differ


informativeness,


er these differences


decision makers generally


in forming their deci


do not


sons.


consid-


Instead,


they


acts


as though


the sources


are


equally


informative


(Wickens,


1984),


and weight


them accordingly.


Berg


(1990)


also,


found


that observers were better at weighting


sources


equal


rather


than


unequal


in reliability


auditory


frequency


discrimination task.


The


results


from the


first study were


consistent


with







when


sources


high


in reliability were cued by


gauge


lumi-


nance,


weighting


efficiency was equivalent


equal


reliability


condition.


Furthermore,


the


best


performance


and highest weighting


efficiency


reliability were cued by


gauge


occurred when sources


luminance,


high


presented


long


stimulus durations,


contiguous


rather than distrib-


uted


throughout


the visual


field.


The


performance advantage


associated


with


grouped source


reliabilities


consistent


with


results of Posner


et al.


(1980).


Examination


the observers


weights


indicate


that


observers


tended


to use a


relatively


equal


weighting


strate-


gy when


there


was


uncertainty about


location of


the more


reliable

advantage


sources.

for th


This may partly


le odd arrangement


explain

in the


performance


non-cued


condi-


tions,


since


this arrangement was most similar to an


equal


reliability pattern.


it

why


is not

it i


study


clear what


s not mainta


which


However,


based


factor accounts

ined in the cued


examines other


factors


on the current evidence


for this advantage,


condition.


that may


future


contribute


this


advantage may help to


understand


these


effects.


instance,


the si


the difference between


sources


high


low


in reliability


or the distribution


sources


reliabilities which


produce


this advantage?


Despite


the non-cued


relatively


conditions,


equal


weighting pattern used


observers showed moderate


sensitiv-