ArticlePDF Available

Negative Information Weighs More Heavily on the Brain

Authors:

Abstract and Figures

Negative information tends to influence evaluations more strongly than comparably extreme positive information. To test whether this negativity bias operates at the evaluative categorization stage, the authors recorded event-related brain potentials (ERPs), which are more sensitive to the evaluative categorization than the response output stage, as participants viewed positive, negative, and neutral pictures. Results revealed larger amplitude late positive brain potentials during the evaluative categorization of (a) positive and negative stimuli as compared with neutral stimuli and (b) negative as compared with positive stimuli, even though both were equally probable, evaluatively extreme, and arousing. These results provide support for the hypothesis that the negativity bias in affective processing occurs as early as the initial categorization into valence classes.
Content may be subject to copyright.
Journal of Personality and Social Psychology
1998,
Vol. 75, No- 4, 887-900Copyright 1998 by the American Psychological Association, Inc.
0022-3514/98/S3.00
Negative Information Weighs More Heavily on the Brain:
The Negativity Bias in Evaluative Categorizations
Tiffany A. Ito, Jeff T. Larsen, N. Kyle Smith, and John T. Cacioppo
Ohio State University
Negative information tends to influence evaluations more strongly than comparably extreme positive
information. To test whether this negativity bias operates at the evaluative categorization stage, the
authors recorded event-related brain potentials (ERPs), which are more sensitive to the evaluative
categorization than the response output stage, as participants viewed positive, negative, and neutral
pictures. Results revealed larger amplitude late positive brain potentials during the evaluative categori-
zation of (a) positive and negative stimuli as compared with neutral stimuli and (b) negative as
compared with positive stimuli, even though both were equally probable, evaluatively extreme, and
arousing. These results provide support for the hypothesis that the negativity bias in affective pro-
cessing occurs as early as the initial categorization into valence classes.
A growing catalog of errors, biases, and asymmetries points
to the conclusion that negative information more strongly influ-
ences people's evaluations than comparably extreme positive
information (Kanouse & Hansen, 1971; Peeters & Czapinski,
1990;
Skowronski & Carlston, 1989). Impression formation is
one area in which this is especially evident. In an illustrative
study, Anderson (1965) found that evaluations of people de-
scribed by multiple positive traits of differing extremity fol-
lowed an averaging rule. The evaluation of such a person was
similar to the average of the evaluations that had been given to
people possessing each of the positive traits in isolation. By
contrast, the evaluation of a person described by multiple nega-
tive traits of differing extremity was less favorable than expected
from an averaging model. This suggests that negative traits are
given greater weight in overall evaluations than are positive
traits (see also Bimbaum, 1972; Feldman, 1966; Fiske, 1980;
Hodges, 1974). A greater weighting for negative information
than positive information can also be seen in risk-taking re-
search, where the axiom that losses loom larger than gains often
holds.
The distress that people report in association with the
loss of a given quantity of money typically exceeds the amount
of pleasure associated with gain of that same amount (e.g.,
Kahneman & Tversky, 1984). More generally, Taylor (1991)
has noted a tendency for negative events to result in a greater
Tiffany A. Ito, Jeff T. Larsen, N. Kyle Smith, and John T. Cacioppo,
Department of Psychology, Ohio State University.
Reparation of this article was supported by National Science Founda-
tion Grant SBR-9512459 and National Institute of Mental Health Grant
P50MH52384-01A1.
We thank Allyson Holbrook for assistance with
statistical analyses.
Correspondence concerning this article should be addressed to Tiffany
A. Ito, who is now at the Department of Psychology, CB 345, University
of Colorado, Boulder, Colorado 80309, or to John T. Cacioppo, Depart-
ment of Psychology, Ohio State University, 1885 Neil Avenue Mall,
Columbus, Ohio 43210-1222. Electronic mail may be sent to tito
@psych.colorado.edu or cacioppo.l@osu.edu.
mobilization of an organism's physiological, cognitive, emo-
tional, and social responses.
These disparate instances of greater sensitivity to negative
information represent the operation of what has been termed a
negativity bias. Cacioppo and colleagues have incorporated the
negativity bias into a more general model of evaluative space
in which positive and negative evaluative processes are assumed
to result from the operation of separable positive and negative
motivational substrates, respectively (Cacioppo & Berntson,
1994;
Cacioppo, Gardner, & Berntson. 1997; Ito & Cacioppo,
in press). Positivity and negativity are further posited as having
partially separable neurophysiological substrates that have func-
tional outputs best viewed within a multidimensional bivariate
space as opposed to a single bipolar continuum.
An important advantage of the model of evaluative space is
that it incorporates instances in which positivity and negativity
are activated in a reciprocal (i.e., bipolar) manner and instances
in which positivity and negativity vary in other combinations.
Research by Goldstein and Strube (1994) demonstrated that
positive and negative evaluations do not always operate recipro-
cally by revealing independence in the positive and negative
reactions students reported at the beginning and end of three
consecutive class periods. Specifically, the intensity of positivity
and negativity within each class period was uncorrelated, and
the two valent reactions were differentially affected by exam
feedback. Cacioppo and colleagues referred to instances in
which the positive and negative motivational systems operate
independently as uncoupled activation (Cacioppo & Berntson,
1994;
Cacioppo et al, 1997).
Multiple modes of evaluative activation were also observed
by Ito, Cacioppo, and Lang (1998), who assessed the relation
between positive and negative evaluations of nearly 500 color
pictures from the International Affective Picture System (IAPS;
Center for the Study of Emotion and Attention, 1995). For many
of the pictures, positivity and negativity were negatively corre-
lated, suggesting a reciprocal relation between the two motiva-
tional systems. However, positivity and negativity were uncorre-
lated for other pictures, revealing uncoupled activation. Uncou-
887
ITO,
LARSEN, SMITH, AND CACIOPPO
pled activation can occur as singular activation of either the
positive or the negative motivational system, both of which were
observed by Ito et al. Finally, the model of evaluative space
proposes a nonreciprocal mode of activation in which both va-
lent systems are coactivated or coinhibited, resulting in a posi-
tive correlation between positivity and negativity ratings. Racial
prejudice is one area in which coactivation has been observed
such that White participants sometimes report both strong posi-
tive attitudes and strong negative attitudes toward African
Americans (Hass, Katz, Rizzo, Bailey, & Eisenstadt, 1991).
Certainly, the model of evaluative space is not the first model
to note the separability of positive and negative evaluative pro-
cesses (for a review, see Cacioppo & Gardner, in press). This
notion has a long tradition, for example, within the attitude and
judgment literature. Although bipolar measures of attitudes are
used widely and attitudes are often conceptualized as the net
difference between the positive and negative valent processes
aroused by a stimulus, many attitude theorists have nevertheless
grappled with the separability of positive and negative evalua-
tions (e.g., Edwards, 1946; Kaplan, 1972; Priester & Petty,
1996;
Scott, 1968; Thompson, Zanna, & Griffin, 1995). The
bivariate structure of evaluations was noted by Scott (1968) in
his review of attitude measurement:
The conception of favorable and unfavorable as "opposites" im-
plies that persons will not be found with attitudes simultaneously
at both ends of
the
dimensions. Yet an alternative formulation might
treat degree of favorableness and degree of unfavorableness as
conceptually distinct (although no doubt empirically correlated)
components, on which persons may make, simultaneously, a variety
of position combinations. In other words, it is only by convention
that direction of an attitude is conceptualized as a single bipolar
continuum, (p. 206)
For our present purposes, the most important implication of
conceptualizing positivity and negativity as separable is the abil-
ity to stipulate different currency or activation functions for the
two systems. Activation functions can be thought of as a means
of expressing the value of separate and multifarious appetitive
and aversive inputs on a common scale of positivity and negativ-
ity, respectively. In the model of evaluative space (Cacioppo &
Berntson, 1994; Cacioppo et al., 1997), the negative motivation
system is characterized by a negativity bias. This refers to a
tendency for the negative motivational system to respond more
intensely than the positive motivational system to comparable
amounts of activation. That is, the gradient for the currency
function for negativity is steeper than the one for positivity (see
also Lewin, 1935; Peeters & Czapinski, 1990).
The steeper gradient for the negative motivational system was
evident in the evaluations of IAPS stimuli reported by Ito et al.
(1998).
To assess the negativity bias, Ito et al. performed a
regression analysis, in which mean negativity scores were re-
gressed onto mean arousal ratings used as a proxy for motiva-
tional activation, for the 212 slides in the set that participants
found more negative than positive. They similarly regressed
mean positivity scores onto arousal ratings for the subset of 258
slides that participants found more positive than negative. As
predicted, the slope of the regression line for negativity was
steeper than the regression line for positivity. Similar results
were obtained using bipolar valence scores as the measures of
evaluative activation, in which the dataset was dichotomized at
the scale median.
To assess the generalizability of this effect across stimulus
items,
we have replicated this analysis on data for the English
Affective Lexicon (Bradley, Lang, & Cuthbert, 1997). This
stimulus set currently contains 620 verbs (e.g., activate), nouns
(e.g.,
lion),
adverbs (e.g., leisurely), and adjectives (e.g., quiet)
for which normative ratings of bipolar valence, arousal, and
dominance are available. Prior to the analyses, the data were
dichotomized at the median of the mean valence ratings. This
median split yielded two subsets of words that elicited either
predominantly positive or predominantly negative evaluations.
Separate regression analyses in which valence ratings were pre-
dicted from arousal ratings for the two subsets of words were
then performed. As in Ito et al. (1998), arousal ratings were
used as a proxy for motivational activation. Consistent with the
regression analyses in Ito et al., a negativity bias was found
such that the regression line relating arousal to valence among
evaluatively negative words was steeper than the line relating
arousal to valence among evaluatively positive words.1
Although the extant research clearly reveals evidence of the
negativity bias (e.g., see Skowronski & Carlston, 1989), much
less is known about the stage at which this bias operates. Ob-
servable expressions of an evaluation represent the output of at
least two stages—evaluative categorization and response out-
put—and the negativity bias could operate at either stage. A
negativity bias may be produced through processes occurring
at the response-output stage by response priming, for example.
This could occur if negative stimuli are more likely to prime
or activate a fight-or-flight response, thereby producing more
extreme reactions to (including more extreme ratings of) nega-
tive stimuli than positive stimuli (Cannon, 1929). Although not
denying the possibility of response priming, the model of evalua-
tive space views the negativity bias as an inherent characteristic
of the underlying motivational substrate. This led to the predic-
tion that the negativity bias will manifest at the initial evaluative
categorization stage of information processing (Cacioppo &
1 In fact, Cacioppo and Berntson (1994) proposed that the activation
functions for positivity and negativity were nonlinear with exponents less
than 1. Therefore, nonlinear regression analyses were also performed on
the IAPS ratings in Ito et al. (1998). These analyses used the equation
E = Ax + h, where E is either unipolar negativity ratings or positivity
ratings, A is arousal ratings, X is the exponent that represents both the
slope of the line and the rate of deceleration in the impact of increasing
activation of the valent system of interest, and b represents the intercept
value. To model activation of the negative motivational system, we com-
puted the above equation using data from the 212 slides that participants
found to be more negative than positive, with unipolar negativity ratings
serving as E. To model activation of the positive motivational system,
we computed the above equation using data from the 258 slides that
participants found more positive than negative, with unipolar positivity
ratings serving as E. As predicted by the negativity bias, the exponent
was larger in the model estimating activation of the negative motivational
system as compared with the model estimating activation of the positive
motivational system. Similar results were obtained when bipolar valence
ratings were used as E, dichotomizing the dataset at the scale median.
Comparable analyses were also performed for stimuli in the English
Affective Lexicon. Results again revealed a larger exponent in the model
estimating activation of the negative motivational system.
NEGATIVITY BIAS IN EVALUATIVE CATEGORIZATIONS889
Berntson, 1994; Cacioppo et al., 1997). In the present research,
we report two experiments designed to test this hypothesis.
This research makes use of event-related brain potentials
(ERPs) as measures of
the
evaluative categorization stage (Caci-
oppo,
Crites, Berntson, & Coles, 1993). ERPs are changes in
electrocortical activity that occur in response to discrete stimuli.
Time-locked topographical features of the ERP are referred to
as components and are typically identified by the peaks in and
the spatial distributions of the waveforms (for reviews, see
Coles,
Gratton, & Fabiani, 1990; Coles, Gratton, Kramer, &
Miller, 1986). An ERP component is assumed to reflect one or
more information-processing operations, and the amplitude of
the component is thought to reflect the extent to which an infor-
mation-processing operation is engaged (Donchin & Coles,
1988;
Gehring, Gratton, Coles, & Donchin, 1992).
The paradigm we use is a modification of the oddball para-
digm frequently used to study the P300 component of the ERP.
In the standard oddball paradigm, simple stimuli representing
two distinct categories (e.g., low- and high-pitched tones) are
presented with differing probabilities to participants. On aver-
age,
the low-probability stimulus (also called the oddball or
target stimulus) evokes a larger positive-going potential, called
the P300, as compared with the high-probability stimulus. The
P300 has a maximal amplitude over central and parietal scalp
areas,
and manifests from approximately 300 to 900 ms follow-
ing stimulus onset (Donchin, 1981).
To study evaluative processes, we have presented stimuli that
are either positive, negative, or neutral in valence, with stimuli
from one evaluative category occurring more frequently than the
others (e.g., Cacioppo et al., 1993; Crites, Cacioppo, Gardner, &
Berntson, 1995) ? We refer to the frequently presented stimuli
in each sequence as the context and those from the less probable
categories as targets. Evaluative inconsistency between the tar-
get and context (e.g., a negative target stimulus embedded within
a sequence of positive-context stimuli) results in an enhance-
ment of a late positive potential (LPP) of the ERP, which shares
many of the signature characteristics of the P300: (a) The LPP
is typically largest over the parietal scalp area, intermediate over
the central scalp area, and smallest over the frontal scalp area;
(b) larger amplitude LPPs are elicited by the (evaluatively)
inconsistent stimuli than by (evaluatively) consistent stimuli,
particularly over central-parietal regions; (c) the average latency
of the LPP falls within the 300 to 900 ms latency window
typical of the P300; and (d) the amplitude of the LPP elicited
over the central-parietal region varies as a function of the evalua-
tive distance of the target from the context even when targets
are equally probable (Cacioppo, Crites, Gardner, & Bemtson,
1994;
Crites & Cacioppo, 1996; Crites et al., 1995; Gardner,
Cacioppo, Berntson, & Crites, in press). Furthermore, these
LPP variations are found when individuals perform evaluative
categorizations of the stimuli but not when they perform various
nonevaluative categorizations (Cacioppo, Crites, & Gardner,
1996;
Crites & Cacioppo, 1996).
As we have noted, we conceptualize evaluative categorization
as separate from response selection and execution (or output;
Cacioppo & Berntson, 1994; Cacioppo et al., 1993), which
raises the issue of whether greater responsivity to negative cues
is a function of processes operating at either the evaluative-
categorization stage or response selection-execution stage.
ERPs provide a means of assessing the evaluative-categorization
stage independent of response selection and execution pro-
cesses. This was demonstrated by Crites et al. (1995), who
recorded LPPs to positive, negative, and neutral stimuli embed-
ded within sequences of positive-context stimuli. On some trials
participants accurately reported their evaluations, whereas on
others they were instructed to misreport their evaluations of
either negative or neutral items as being positive. The misreport
instructions had the intended effect on overt responses. However,
the LPPs to evaluatively inconsistent stimuli, as compared with
consistent stimuli, were enhanced, regardless of the accuracy of
the overt evaluative report. These results were also replicated
in a negative evaluative context in which participants either
accurately reported their evaluations or misreported their neutral
or positive evaluations (Crites et al., 1995; see also Gardner et
al.,
in press). Therefore, ERPs provide an especially sensitive
probe of the evaluative-categorization stage, allowing us to as-
sess whether the underlying negative motivational system re-
sponds more intensely than does the positive system to compara-
ble amounts of activation.
To test the hypothesis that the negativity bias operates at the
evaluative-categorization stage, we performed two experiments
in which ERPs were recorded while participants evaluated posi-
tive,
negative, and neutral pictorial stimuli. Neutral pictures
served as the most frequently presented contextual stimuli. The
positive and negative pictures were equated for (a) probability
of occurrence, (b) evaluative extremity relative to the neutral
pictures, and (c) level of arousal, resulting in a design in which
the positive and negative pictures differed primarily in terms
of whether they activated the positive or negative motivational
system. If the negativity bias operates at the evaluative-categori-
zation stage, it should manifest itself as larger LPPs to evalua-
tively negative pictures as compared with positive pictures.
Experiment 1
Method
Participants. Thirty-three Ohio State University (OSU) undergrad-
uates (24 men) participated in the experiment for partial class credit.
All were right-handed and had right-handed parents. Data from
S
partici-
pants were removed because of equipment malfunction (n - 5), volun-
tary withdrawal from the study (n - 2), or excessive artifact in the
electroencephalograph (EEG) from vertical eye movement (n = 1).
Analyses were conducted on the data obtained from the remaining 25
participants.
Materials. Thirty-six affectively neutral, two positive, and two nega-
tive pictures were selected from Sets 1-8 of the LAPS (Center for the
Study of Emotion and Attention—National Institute of Mental Health,
1995).
Because neutral pictures were shown much more frequently than
positive or negative pictures, the inclusion of a greater number of neutral
pictures ensured that exemplars from all three categories were presented
2 In the traditional oddball paradigm, stimuli are presented in long
sequences (e.g., 200 stimuli). Evaluative categorization of long se-
quences of affectively valenced stimuli proved difficult for participants
to perform (see Cacioppo et al., 1993, Note 1). As a result, we present
stimuli in short sequences of 5 or 6 stimuli in our modified paradigm.
The shorter sequences reduce variability in the ERP by presumably
increasing participants' attention to and discrimination of the stimuli.
890
ITO,
LARSEN, SMITH, AND CACIOPPO
an equal number of times.3 Using normative data collected from OSU
undergraduates in a previous term (Ito et al., 1998), we selected neutral
pictures that had (a) bipolar valence ratings near the midpoint (5.0)
and median (5.19) of the scale (M = 5.10 on a 1-9 scale; range =
4.21-6.15); (b) low levels of positive activation as measured by a
unipolar positivity scale (M = 2.10 on a 5-point scale, where lower
values indicate less positivity); (c) low levels of negative activation as
measured by a unipolar negativity scale (M = 1.48 on a 5-point scale,
where lower values indicate less negativity); and (d) low levels of
arousal (M = 2.75, as measured on a 9-point bipolar scale where lower
values indicate greater calmness). The 36 neutral pictures were divided
into two equal-sized groups with comparable normative ratings: (a)
bipolar valence M = 5.10, 5.10; (b) unipolar positivity M = 2.12, 2.08;
(c) unipolar negativity M = 1.49, 1.47; and (d) arousal M = 2.74, 2.75.
Examples of neutral pictures include a plate, hair dryei; and an electrical
outlet.
The two positive and two negative pictures were selected to have high
valence and arousal ratings that were equally extreme from the mean
values for the neutral pictures. The positive pictures, which depicted a
red Ferrari and people enjoying a roller coaster, had the following norma-
tive ratings: (a) bipolar valence M = 8.31, (b) unipolar positivity M =
4.19,
(c) unipolar negativity M - 1.23, and (d) arousal M = 7.43. The
negative pictures, which depicted a mutilated face and a handgun aimed
at the camera, had the following normative ratings: (a) bipolar valence
M = 1.89, (b) unipolar positivity M = 1.16, (c) unipolar negativity M
= 4.07, and (d) arousal M =
1.34.4
Procedure. Potential participants were informed that the purpose of
the study was to measure electrical activity occurring in the brain when
people view pictures. Once they arrived for the experimental session,
they received a brief overview of the procedures, then read and signed
an informed consent form. Participants then had the electrodes attached
and received more detailed task instructions.
Participants were seated in a comfortable reclining chair in a sound-
attenuated, electrically shielded room. Following procedures used in
prior research on evaluative categorization (Gardner et al., in press),
pictures were shown to participants in sequences of five on a color
computer monitor located approximately 76 cm in front of the chair.
LPPs are affected by surrounding contextual stimuli as well as by the
stimuli currently being processed. As a result, LPPs observed in prior
research (in which targets were embedded in positive or negative stimu-
lus sequences) may have varied with the valence of the context stimuli
as well as the target stimulus. To examine the negativity bias in the
present experiment, we therefore established a neutral evaluative context
and recorded ERP responses to pictures that were either evaluatively
consistent (i.e., neutral) or inconsistent (i.e., positive or negative) with
that context. To accomplish this, all participants were exposed to 120
sequences of five pictures. These sequences were divided into two 60-
sequence blocks. Both blocks contained primarily neutral pictures but
differed in whether positive or negative pictures were also embedded in
some of the sequences (see Table 1). Specifically, in half of the se-
quences in each block, a single positive or negative target (depending
on the block) was embedded in the neutral context. In the remaining
sequences in each block, all pictures within the sequences were neutral,
and one of these neutral pictures was designated as the target picture.
In all sequence types, targets randomly appeared in either the third,
fourth, or fifth position in a sequence, thereby ensuring that targets were
always preceded by at least two neutral pictures and that participants
could not easily predict when a positive or negative picture might appear.
This resulted in 12 trial types (see Table 1). Types 1-6 were shown to
each participant an equal number of times within one of the blocks, and
Types 7-12 were shown an equal number of times within the other
block. Order of the trial types was randomized within each block for
each participant. Participants were not informed of the distinction be-
tween context and target pictures and evaluated all pictures in a similar
Table 1
Types of Five-Picture Sequences Used in Experiment 1
Stimulus position
Sequence type
Block with positive targets
1.
Neutral in neutral
2.
Neutral in neutral
3.
Neutral in neutral
4.
Positive in neutral
5.
Positive in neutral
6. Positive in neutral
Block with negative targets
7.
Neutral in neutral
8. Neutral in neutral
9. Neutral in neutral
10.
Negative in neutral
11.
Negative in neutral
12.
Negative in neutral
N
N
Note. Within each sequence type, psychophysiological data were re-
corded to an evaluatively consistent or inconsistent target located in
either the third, fourth, or fifth stimulus position; these stimuli are desig-
nated by boldface characters.
<$>
= a stimulus position in which neutral
targets were presented; P = a stimulus position in which positive targets
were presented; N = a stimulus position in which negative targets were
presented.
fashion. Psychophysiological data were recorded during the presentation
of the single target picture in each sequence.
Half of the participants saw the block with embedded positive targets
first, and the remaining participants saw the block with embedded nega-
tive targets first. Moreover, a different set of 18 neutral pictures was
presented in each block. This was intended to prevent attenuation of
LPP amplitude to the neutral pictures in the second block because of
repeated presentations of specific neutral pictures. Prior to viewing the
second block, participants took a short (<5 min) break and read a brief
history of OSU. The passage was intended to induce a neutral mood
and mitigate potential carryover effects.
Each picture in a sequence was presented for 1,000 ms. Participants
were instructed to look at the picture for its entire presentation and to
think about whether it showed something they found to be positive
3 Although we intended to present each picture 15 times, an error
in the stimulus randomization program, which was detected after data
collection, resulted in one exemplar from each valent category being
presented 20 times, whereas the other exemplar from the same valent
category was presented 10 times. Thus, for each participant the probabil-
ity of seeing a given valent picture was eitfier .033 or .067 instead of
being equiprobable. This difference in probability of presentation be-
tween exemplars of the same valenl category was crossed, not con-
founded, with experimental variables, however, and had no effect on LPP
amplitude. Therefore, we collapsed across this factor in all subsequent
analyses.
4 The IAPS pictures in Neutral Subset A were 1910, 2190, 4100,
5500,
5800, 6150, 7002, 7025, 7035, 7040, 7080, 7090, 7140, 7217,
7224,
7285, 7550, and 7820. Those in Neutral Subset B were 2230,
2840,
5900, 7000, 7006, 7009, 7010, 7030, 7050, 7100, 7130, 7150,
7170,
7190,
7233,
7235, 7284, and 9210. The TAPS pictures in the
positive category were 8490 and 8510; those in the negative category
were 3030 and 6230.
NEGATIVITY BIAS
IN
EVALUATIVE CATEGORIZATIONS891
(or negative, depending
on the
block)
or
neutral. After stimulus offset,
participants registered their evaluation
by
pressing
one of
two labeled
keys
on a
computer keypad. Either
the
left
(or
right, counterbalanced)
key indicated neutral.
The
other
key
indicated positive
or
negative,
depending
on
which block
was
being presented. Participants used
the
left
and
right thumbs
to
press
the
left
and
right keys, respectively. After
a
1,000 ms
interstimulus interval,
the
next picture
was
shown.
We
stressed
to
participants that there were
no
right
or
wrong answers
in
evaluating
the
pictures
and
that
we
were interested
in
their first impres-
sions.
After
the
fifth picture
in a
sequence,
the
wprd pause
was
shown
on
the
screen. Participants were instructed
to use
either thumb
to
press
a third button
on the
keypad when they were ready
to
initiate
the
next
sequence
of
five pictures.
To summarize, data were recorded from both target (positive
and
negative)
and
context (neutral) pictures. The design, therefore, featured
two within-subject variables: picture category (valenced
or
context)
and
target valence (positive
or
negative). There were also three between-
subject variables, representing the hand used
to
register
a
neutral evalua-
tion (left
or
right), block order (positive
or
negative block first),
and
set
of
neutral pictures shown
in the
first block (Subset
A or B). An
additional within-subject factor, sagittal scalp site,
is
described below.
Psychophysiological data collection
and
reduction. EEG data were
recorded
at
sites over midline frontal
(Fz),
central
(Cz), and
parietal
(Pz) scalp areas using
tin
electrodes sewn into
an
elastic
cap
(Electro-
Cap International, Eaton, OH).5
An
additional site
at the top of the
forehead served
as an
electrical ground. Miniature
tin
electrodes were
also placed over
the
left
and
right mastoids. Active scalp sites were
referenced on-line
to
the left mastoid. Additional miniature tin electrodes
were placed above
and
below
the
left
eye and on the
outer canthus
of
each
eye to
monitor vertical and horizontal eye movements, respectively.
Electrode impedances were below 5 KO at all sites. EEG and electroocu-
logram (EOG) recordings were amplified
by
NeuroScan Synamps
am-
plifiers with
a
bandpass
of
0.1-30
Hz
(12-dB roll-off)
and
digitized
at
1,000
Hz. For all
targets,
EEG and
EOG data recording began 128
ms
before picture onset
and
continued throughout
the 1,000 ms
picture
presentation.
Off-line,
the
data were rereferenced
to a
computed average
of
the left
and right mastoids.6 EEG data were next corrected
to the
mean voltage
of the 128-ms prestimulus recording period before applying
a
regression
procedure to remove the effects
of
vertical eye movements from the EEG,
which
can
distort measurements from scalp sites (Semlitsch, Anderer,
Schuster,
&
Presslich, 1986). The regression correction
was
applied
to
14
of
the participants' data. The remaining 11 participants
did not
blink
enough during
the EEG
recording period
for the
regression procedure
to reliably estimate
eye
activity from
the
vertical
EOG
channel.
For
these participants,
we
visually inspected
the EEG
data
and
deleted
any
trials
on
which ocular
or
other artifact occurred (e.g., because
of
move-
ment).
We similarly inspected the EEG data from those participants
for
whom
the
regression procedure
was
applied
for
remaining ocular
or
other artifact.
For all
participants,
if
artifact was detected
at any of the
three scalp sites, data from
all
sites
for
that trial were eliminated from
further analysis.
A 9-Hz
low-pass digital filter
was
then applied
to the
remaining data.
We next constructed ensemble averages
to
extract the LPP component
from
the EEG
signal (Coles
et al.,
1990).
For
each participant's data,
we computed four averaged waveforms. These waveforms aggregated
the electrical activity associated with
the
evaluation
of
positive targets,
neutral targets
in
the positive block, negative targets, and neutral targets
in
the
negative block.
We
calculated separate ensemble-averaged
ERP
waveforms
for
each scalp site
for
each participant.
The
amplitude
of
the LPP
of
the ERP
was
quantified
by
locating within each
ERP
wave-
form
the
largest positive-going potential
at Pz
between 400-900
ms
after stimulus onset.
The
amplitude
of the LPP in the
other
two
sites
was defined
as the
largest positive-going potential occurring within
±
100
ms of
the LPP
at
Pz.
Results
We first examined whether we replicated prior research on
evaluative categorizations showing that the LPP amplitude has
a central-parietal scalp distribution and varies as a function
of evaluative inconsistency. To do this, we subjected the LPP
amplitudes to a 2 (picture category: context, valenced) x 2
(target valence: positive, negative) x 3 (sagittal scalp site: fron-
tal,
central, parietal) X 2 (hand for neutral response) X 2 (block
order: positive first, negative first) X 2 (set of neutral pictures
paired with positive targets: Subset A, Subset B) multivariate
analysis of variance (MANOVA). All F tests reported represent
the Wilks's lambda approximation. Our results confirmed both
prior effects. First, we obtained a main effect of sagittal scalp
site, F(2, 16) = 22.15,/? <
.0001.
Two sets of planned contrasts
were conducted to test specifically for the central-parietal scalp
distribution. The first contrast compared the mean of the LPP
amplitudes at Cz and Pz with the mean at Fz, revealing signifi-
cantly larger LPPs at the combined Cz-Pz areas (combined
Cz-Pz M = 7.42 fN) than at Fz (M = 3.28 fN), F(l, 17) -
34.46,
p <
.0001.
The second planned contrast compared LPP
amplitudes at Pz and Cz, revealing larger LPPs at Pz (M - 8.49
fN) than at Cz (M - 6.35 //V), F(l, 17) = 9.84, p < .01.
Second, the MANOVA revealed main effects of picture category,
F(l, 17) = 59.98, p <
.0001,
and target valence, F(l, 17) =
8.79, p < .01, which were qualified by a significant Picture
Category X Target valence interaction, F(l, 17) = 19.45, p <
.01.
Both positive and negative targets (M^a^ 7.43 fN,
^^3^
= 10.90 fN) were associated with larger LPPs than their
corresponding context pictures (M = 2.95 fN and M ~ 2.89
fN, respectively), F(l, 17) - 27.08, p < .001, and F(l. 17)
52.66, p < .001, respectively. Across both blocks, then, LPPs
were larger for evaluatively inconsistent pictures than consistent
pictures.
It is importanl to note that this interaction also revealed the
predicted negativity bias at the evaluative categorization stage.
Negative targets resulted in larger LPPs than positive targets,
F( 1, 24) =
15.51,p
< .001, whereas LPP amplitudes to neutral
pictures in the two blocks did not differ (see Figure 1). Thus,
the evaluative categorization of negative stimuli was associated
with a larger amplitude LPP than was the evaluative categoriza-
tion of equally probable, equally evaluatively extreme, and
equally arousing positive stimuli.
5 Data were also recorded from
25
additional scalp sites (F3, F4,
F7,
F8,
FT7, FT8, FC3, FC4, T3, T4, TP7, TP8, C3, C4, CP3, CP4,
T5,
T6,
P3, P4,
01, 02, and OZ) for
initial exploratory analyses
of
dipole
sources. Thus, they
are not
relevant
to the
present psychological
hypotheses.
6 The off-line referencing
nf
EEG data
to a
combined left
and
right
mastoid reference eliminates bias
in
asymmetry measures that
can be
introduced from even slight differences
in
impedances between
the two
mastoids
(R. J.
Davidson, personal communication, September
21,
1995).
Other referencing schemes have been used
to
eliminate
the
bias-
ing effects
of
differential impedances, such
as
common referencing
(ref-
erencing each site
to an
average
of all the
sites),
but the
method used
here results
in
more precise and less biased measures
of
ERP amplitudes.
892
[TO,
LARSEN, SMITH, AND CACIOPPO
-6-1
-2-
I
•o
"5.
neutral (positive block)
neutral (negative block)
positive
negative
Latency (ms)
Figure
1.
Averaged event-related brain potential waveforms at the midline parietal electrode (Pz) to neutral
and positive targets in the block of trials containing neutral (frequent) and positive (rare) targets and to
neutral and negative targets in the block of trials containing neutral (frequent) and negative (rare) targets.
The amplitude of the late positive potential is not only larger to the rare (positive and negative) targets than
the frequent (neutral) targets, but it is larger to the negative targets than the positive targets. These results
were obtained even though the positive and negative targets were equally extreme, arousing, rare, and task
relevant. These results, therefore, are consistent with the operation of a negativity bias at the evaluative
categorization stage of information processing.
A Picture Category X Sagittal Scalp Site interaction was also
obtained, F(2, 16) = 16.38,/? <
.0001.
For the target pictures,
follow-up comparisons revealed larger amplitude LPPs at the
central-parietal area (combined Cz-Pz M = 10.78 (N) as com-
pared to Fz (M = 4.55 iN), F(U 24) = 74.30, p <
.0001.
LPPs at Pz (M - 12.54 pV) were also significantly larger than
those at Cz (M = 9.01 (N) for the target pictures, F(\, 24) =
23.92,
p <
.0001.
For the neutral pictures, follow-up compari-
sons revealed larger amplitude LPPs at the central-parietal area
(combined Cz-Pz M = 3.55 fN) as compared to Fz(M = 1.48
//V),
F( I, 24) - 9.89, p < .005; LPP amplitudes at Pz and Cz
did not differ for the context pictures.
More interesting from a theoretical perspective, we obtained
a Picture Category X Target Valence X Sagittal Scalp Site inter-
action, F(2, 16) = 6.73, p < .01. The mean amplitudes as a
function of picture category, target valence, and sagittal scalp
site are shown in Table 2, and the stimulus-aligned averaged
waveforms at Pz are shown in Figure 1. LPP amplitudes were
larger to negative than positive targets at Pz, Cz, and Fz, all ps
< .05. The interaction was attributable to a greater sensitivity
to evaluative inconsistency at central and parietal areas. When
participants were viewing negative target pictures, the target
pictures resulted in larger LPP amplitudes than the neutral pic-
tures did at all sites, all ps < .01. When participants were
viewing positive target pictures, the target pictures resulted in
larger LPP amplitudes than neutral pictures did at Pz and Cz,
ps < .005, a difference that was not significant at Fz, p < .06/
Discussion
The results of Experiment 1 reveal three important findings.
First, the present results replicate prior research results, reveal-
ing both the scalp distribution of the LPP and the LPP's sensitiv-
ity to evaluative inconsistency (e.g., Cacioppo et al.,
1993;
Caci-
oppoetal., 1994; Crites & Cacioppo, 1996; Crites et al., 1995).
Specifically, the LPP was largest over central-parietal regions,
7 The MANOV\ also revealed two higher order interactions that did
not bear on theoretical issues: (a) Picture Category x Block Order x
Set of Neutral Pictures Paired with Positive Targets, F(l, 17) -
11.23,
p < .01; (b) Picture Category X Hand for Neutral Response x Block
Order X Set of Neutral Pictures Paired with Positive Targets, F(l, 17)
= 731, p <
.01.
None of these interactions were theoretically interesting,
nor did they qualify the effects reported in the text.
NEGATIVITY BIAS IN EVALUATIVE CATEGORIZATIONS893
Table 2
Mean LPP Amplitude and Standard Error of the Mean As a Function of Picture Category,
Target Valence, and Sagittal Scalp Site in Experiment 1
Site
Pz
Cz
Fz
M
Negative
M
14.99
11.51
6.20
10.90
SEM
1.22
1.20
1.16
1.09
Target
Positive
M
10.87
7.38
4.04
7.43
SEM
1.04
1.03
1.07
0.87
Neutral
(negative
block)
M SEM
3.67 0.59
3.26 0.64
1.57 0.78
2.89 0.43
Neutral
(positive
block)
M SEM
4.44 0.82
3.26 0.87
1.14 0.93
2.95 0.73
M
8.49
6.35
3.28
SEM
0.72
0.68
0.77
Note. All values are in
/JV.
LPP = late positive potential; Pz = midline parietal electrode; Cz = midline
central electrode; Fz = midline frontal electrode.
and its amplitude was enhanced for the evaluatively inconsistent
positive and negative pictures as compared with the neutral-
context pictures. Moreover, the LPP was maximally sensitive to
evaluative inconsistency effects at Pz and Cz.
Second, the sensitivity of the LPP to evaluative inconsistency
in the present experiment builds upon prior research by using
an evaluatively neutral context. In prior research, LPP enhance-
ment to evaluative inconsistency was obtained across a range
of evaluative context-target combinations, including positive
contexts with negative targets (Cacioppo et al., 1993; Cacioppo
et al., 1994; Crites et al., 1995; Crites & Cacioppo, 1996) and
neutral targets (Crites et al., 1995), as well as negative contexts
with positive targets (Cacioppo et al., 1993; Crites et al., 1995)
and neutral targets (Crites et al., 1995). Note that the context
in these prior studies was always either positive or negative.
The present research demonstrates that LPP enhancement to
evaluative inconsistency also occurs with positive and negative
targets in a neutral context. As predicted, LPPs for the positive
and negative targets were larger than LPPs for the neutral con-
text. When added to the cumulative research, the present results
demonstrate that the effects of evaluative inconsistency are not
dependent on the valences of the contextual stimuli.
More important, larger LPPs were obtained in response to
negative target pictures as compared with positive target pic-
tures,
suggesting the operation of a negativity bias as early as
the evaluative-categorization stage. This occurred even though
the positive and negative pictures were equally improbable in
the stimulus sequences and equally discrepant from the neutral
pictures in terms of mean valence and arousal ratings. The stim-
uli in Experiment 1 were chosen based on the normative re-
sponses from a separate sample of OSU undergraduates. Exami-
nation of the behavioral responses in Experiment 1 indicated
that 10 of the 25 participants consistently categorized at least
one of the four normatively valenced targets as neutral. There-
fore,
we conducted two ancillary analyses to determine whether
the ERP evidence for evaluative inconsistency or negativity bias
effects could reflect the operation of a possible confounding
variable. Neither analysis provided any evidence for this possi-
bility. First, analyses of the 15 participants whose behavioral
responses matched the normative classifications revealed the
same evaluative inconsistency and negativity bias effects as were
obtained in the full sample. Second, among those participants
who tended to misclassify one of the positive or negative targets,
the LPP amplitude for the misclassined target did not differ
from the amplitude for the target of the same valence that was
correctly classified. For instance, the amplitude of the misclassi-
fied positive target did not differ from the correctly classified
positive target. Further, LPP amplitude for the misclassified posi-
tive or negative target was uncorrelated with LPP amplitude for
the neutral pictures from the relevant block. At the electrocorti-
cal level, then, evidence shows that positive and negative targets
were perceived as evaluatively inconsistent from the neutral con-
text, even when the participants pressed the neutral rather than
normatively consistent positive or negative button on the keypad.
Experiment 2
Although the ancillary analyses suggest that the results of
Experiment 1 were not qualified by normatively inconsistent
classifications, we were nevertheless concerned about the occur-
rence of these responses. The normative data on which stimulus
selection was based were collected in sessions in which roughly
equal numbers of positive, negative, and neutral pictures were
shown (Ito et al., 1998). The judgments of pictures in Experi-
ment 1, in contrast, were made in a context of primarily neutral
pictures. The numerous presentations of neutral pictures may
have affected participants' subjective evaluative criterion by ex-
panding the range of neutral classifications. In essence, this
would have produced an assimilation of valent stimuli to the
neutral category. To counter this possibility, we replicated Exper-
iment 1 but preceded the experimental trials with a picture
preview period in which positive, negative, and neutral pictures
were shown to participants. We reasoned that participants would
be less likely to expand their subjective range of neutral stimuli if
they had recently been exposed to anchors of relatively extreme
positivity and negativity.
In addition, we intermixed positive and negative targets in a
single block in Experiment 2 to increase external validity.
Whereas the valence of the target differed by blocks in Experi-
ment 1, people more typically encounter positive and negative
events in close proximity in everyday settings. A situation in
which the activation and deactivation of the positive and negative
894
1T0,
LARSEN, SMITH,
AND
CACIOPPO
motivational systems occur in a random sequence may therefore
more closely mimic the evaluative situations typically faced in
everyday life. Moreover, whereas participants in Experiment 1
were presented with only two response options (i.e., positive
and neutral in the positive target block and negative and neutral
in the negative target block), participants in Experiment 2 were
presented with all three options for every picture. When people
naturally evaluate objects in their environment, it is more likely
that they choose from the full range of evaluative responses,
which includes positivity, negativity, and neutrality. Thus, the
presence of all three options may also more closely correspond
to people's everyday evaluative experiences than did the situa-
tion in Experiment 1. Finally, to increase the generalizability of
the effects obtained in Experiment 1, we used different positive
and negative pictures in Experiment 2.
Method
Participants. Twenty-one
OSU
undergraduates
(11 men)
partici-
pated
in the
experiment
tor
partial class credit. Data from
7
participants
were unusable because
of
equipment failure
(n = 1)
or excessive artifact
(n
= 6).
Analyses were conducted
on the
data obtained from
the re-
maining
14
participants.
Materials. Twenty-six affectively neutral, two positive, and two neg-
ative pictures were used
in the
experiment. Eighteen
of
the neutral
pic-
tures
and all of the
positive
and
negative pictures were selected from
Sets
1
-8 of
the IAPS using normative data from Itoetal. (1998). Some
of
the
neutral pictures
(but
none
of the
positive
or
negative pictures)
used
in
Experiment
1
were also used
in
Experiment
2.8
The
18
neutral IAPS pictures
had the
following characteristics:
(a)
bipolar valence ratings near
the
midpoint
of
the scale
(M -
5.00),
(b)
bipolar valence ratings
not
greater than
1
scale point away from
the
midpoint (range
=
4.60-5.40),
(c) low
levels
of
positive activation
as
measured
by
unipolar positivity scale
(M =
1.91),
(d) low
levels
of
negative activation as measured by unipolar negativity scale (M
=
1.47),
and
(e) low
levels
of
arousal
(M =
2.64).
The two
positive
and two
negative pictures were selected
to
have high and equally extreme affect
and arousal ratings.
The
positive pictures, which showed
a
pizza
and a
bowl
of
chocolate
ice
cream,
had the
following normative ratings:
(a)
bipolar valence
= 7.81, (b)
unipolar positivity
= 3.65, (c)
unipolar
negativity
= 1.42, and (d)
arousal
=
6.17. The negative pictures, which
showed
a
dead
cat and a
dead and decomposing cow,
had the
following
normative ratings:
(a)
bipolar valence
= 2.10, (b)
unipolar positivity
= 1.09, (c) unipolar negativity
=
4.08, and
(d)
arousal
=
6.22. Eighteen
pictures
in the
IAPS
met our
criteria
for
neutral pictures.
To
ensure
equal rates
of
presentation
of
each individual picture,
we
included eight
additional neutral pictures from
the PC
Paintbrush PhotoLibrary
CD
(1994)."
Procedure. Experiment
2
followed
the
same procedure
as
Experi-
ment
1,
with
the
following exceptions. First, participants were exposed
to
90
sequences
of
five stimuli,
and
each
of
the
9
possible sequences
in
Table
3
was shown 10 times
in a
different random order
for
each partici-
pant. Second, just prior to the presentation
of
experimental trials, partici-
pants were preexposed
to 30
positive,
30
negative,
and 30
neutral
pic-
tures.
To provide anchors
of
extreme positivity and negativity,
all
partici-
pants viewed
the
neutral preview pictures last. Half
of the
participants
were randomly assigned
to
view
the
positive pictures first,
and the re-
maining participants viewed the negative pictures first. Six pictures were
shown at
a
time on the screen, and participants paced themselves through
the preview screens.
The 26
neutral,
2
positive,
and 2
negative pictures
selected
for use in the
experiment were among those shown during
the
preview period. These were augmented with
4
additional neutral pictures
from
the PC
Paintbrush PhotoLibrary
CD
(1994),
28
positive pictures
Table 3
Types of Five-Picture Sequences Used in Experiment 2
Stimulus position
Sequence type
1.
Neutral
in
neutral
2.
Neutral
in
neutral
3.
Neutral
in
neutral
4.
Positive
in
neutral
5.
Positive
in
neutral
6. Positive
in
neutral
7.
Negative
in
neutral
8. Negative
in
neutral
9. Negative
in
neutral
N
<f>
A
N
<b
N
Note. Within each sequence type, psychophysiological data were
re-
corded
to an
evaluatively consistent
or
inconsistent target located
in
either the third, fourth,
or
fifth stimulus position; these stimuli are desig-
nated
by
boldface characters.
<b
- a
stimulus position
in
which neutral
targets were presented;
P = a
stimulus position
in
which positive targets
were presented;
N = a
stimulus position
in
which negative targets were
presented.
from
the
IAPS,
and 28
negative pictures from
the
IAPS. Participants
were told that
the
pictures were shown
so
they could preview some
of
the pictures they would
see in the
experiment.
As
in
Experiment
1,
participants were instructed
to
make evaluative
categorizations
of
what was depicted
in
the pictures. Unlike Experiment
1,
participants were exposed
to
positive, neutral,
and
negative stimuli
across the 90 sequences. Therefore, participants were instructed to deter-
mine whether each picture showed something they found positive, nega-
tive,
or
neutral. They indicated their evaluation
by
pressing
one of
three
appropriately labeled keys
on a
computer keypad once
the
picture
was
removed from
the
screen.
The
middle
key was
always labeled neutral.
For half
of the
participants, keys labeled positive
and
negative were
located
to the
right
and
left, respectively.
The
order
of
these keys
was
reversed
for the
remaining participants.
The
right thumb
was
used
to
respond
to the
rightmost
key and the
left thumb
for the
leftmost
key.
Either diumb could
be
used
for the
middle
key. The
screen remained
blank
for 1,200 ms
after participants responded, then
the
next picture
appeared. Participants pushed
a
fourth
key to
initiate
the
next picture
sequence following
the
word pause.
Psychophysiological data collection
and
reduction. Experiment
2
used
the
same data-collection
and
reduction procedures
as
Experiment
1.
We
computed separately
for
each participant
an
ensemble-averaged
ERP
for
positive targets, negative targets,
and
neutral targets. This
was
done separately
at all
sites, producing nine separate ERP waveforms
for
each participant.
LPP
amplitude
was
quantified
in the
same manner
as
in Experiment
1.
Results
As in Experiment 1, we tested for the expected scalp distribu-
tion and the effects of evaluative inconsistency before testing
8 The IAPS pictures
in the
neutral category were 6150, 7006, 7009,
7010,
7025, 7030, 7035, 7040, 7080, 7090,
7100, 7150, 7170, 7190,
7233,
7235, 7820,
and
7830. Picture 7830
was the
only neutral IAPS
picture used
in
Experiment 2 but not in Experiment 1. The IAPS pictures
in
the
positive category were 7340
and
7350,
and
those
in the
negative
category were 9140
and
9571.
9 The PhotoLibrary
CD
images used
had the
file names 5740081,
9320065,
9430083, 14060015, 14070024, 14070035, 20110049,
and
2040092.
NEGATIVITY BIAS IN EVALUATIVE CATEGORIZATIONS895
Table 4
Mean LPP Amplitude (and Standard Error of the Mean) as a Function of Target Variance
and Sagittal Scalp Site in Experiment 2
Site
Pz
Cz
Fz
M
Negative
M
17.01
9.03
3.14
9.73
SEM
0.85
0.91
0.90
0.65
Target
Positive
M
13.01
8.73
4.44
8.94
SEM
1.34
1.13
1.08
0.98
M
5.20
2.29
1.01
2.83
Neutral
SEM
0.62
0.91
1.45
0.88
M
11.95
6.68
2.87
SEM
0.65
0.68
0.93
Note. All values are in fN. LPP = late positive potential; Pz = midline parietal electrode; Cz = midline
central electrode; Fz = midline frontal electrode.
for the posited negativity bias. All F tests reported represent the
Wilks's lambda approximation. A 3 (target valence: neutral,
positive, negative) X 3 (sagittal scalp location: frontal, central,
parietal) X 2 (hand for positive response: left, right) X 2 (va-
lence of first preview pictures: positive, negative) MANOVA
revealed the expected main effect of sagittal scalp site, F(2, 9)
= 37.47, p <
.0001.
As in Experiment 1, we performed two
planned contrasts, the first of which compared the mean of the
LPP amplitudes at Cz and Pz to those at Fz, and the second of
which compared LPP amplitudes at Pz to those at Cz. As in
prior research and in Experiment 1, LPPs were larger at the
combined central-parietal area (combined Cz-Pz M = 9.53
}N) than at Fz (M = 2.87 //V), F(1, 10) = 40.64, p <
.0001,
and LPP amplitudes at Pz (M = 11.95 fN) were larger than
those at Cz (M = 6.68 //V), F(\y 10) =
83.13,
p <
.0001.
Thus,
the scalp distribution of the LPP conformed to the ex-
pected central-parietal maximum.
In addition, the main effect of target valence confirmed that
the LPP was sensitive to evaluative inconsistency, F(2, 9) =
20.94,
p <
.0001.
A planned contrast revealed that across all
sites,
LPPs were larger for the evaluatively inconsistent positive
and negative pictures (combined positive-negative M = 9.45
fJV) than for the neutral-context pictures (M ~ 2.83 fN), F(\,
10) = 45.95, p <
.0001.
We also obtained a significant Target
Valence X Sagittal Scalp Site interaction, F(4, 7) = 15.86, p
< .001. Follow-up contrasts revealed evaluative inconsistency
effects at each scalp site: LPPs for the evaluatively inconsistent
positive and negative pictures exceeded those for neutral pictures
atPz, F(l, 13) - 74.40, p <
.0001,
Cz, F(l, 13) = 26.60,77
<
.0001,
andFz, F(l, 13) = 5.10, p < .05.
The Target Valence X Sagittal Scalp Site interaction also
revealed evidence of the operation of the posited negativity bias
at the evaluative-categorization stage of information processing.
The mean LPP amplitudes as a function of target valence and
sagittal scalp site are shown in Table 4, and the stimulus-aligned
averaged waveforms for each valence at Pz are shown in Figure
2.
Follow-up contrasts comparing LPP amplitude for negative
pictures to LPP amplitude for positive pictures at all sites re-
vealed the negativity bias at Pz, where LPP amplitude was larger
for negative targets (M 17.01 //V) than for positive targets
(M - 13.01 fN),F{l, 13) = 5.29, p <
.05.10
Discussion
In Experiment 1, we noted unexpectedly high rates of norma-
tively inconsistent categorizations of the valenced targets. Al-
though ancillary analyses discounted the possibility that nonnor-
mative responses produced the evaluative inconsistency and neg-
ativity bias effects in Experiment 1, we nevertheless hoped to
decrease the rate of normatively inconsistent responses in Exper-
iment 2. Examination of the behavioral responses from partici-
pants in Experiment 2 suggests we were successful in doing so
in that only three of the participants in Experiment 2 displayed
normatively inconsistent classifications of valent stimuli.11
Several design changes in Experiment 2 may have increased
participants' agreement with the normative classifications. First,
we introduced the picture preview period in which positive,
negative, and neutral pictures were shown to participants before
the experimental trials. Presenting such a large number of neutral
pictures may have assimilated valent stimuli into the neutral
category in Experiment 1. To counteract this in Experiment
2,
we provided participants with relatively extreme anchors of
positivity and negativity before the experimental trials by pre-
senting the positive and negative preview pictures before the
neutral ones. Experiment 2 also used different exemplars of
positive and negative stimuli and had a partially nonoverlapping
set of neutral stimuli, raising the possibility that the particular
exemplars chosen in Experiment 1 resulted in fewer consensual
classifications than the exemplars in Experiment 2 did. This
seems unlikely, however, because the variability in the normative
ratings of stimuli in Experiments 1 and 2 was comparable. An-
other difference between the two experiments was the presence
of two (Experiment 1) as opposed to three (Experiment 2)
response options. If anything, the presence of an additional
response option would increase response variability, leading
10 Of less theoretical interest was a main effect of hand for positive
response, F(l, 10) = 7.84, p < .05. LPPs were larger for participants
who used their left thumb to indicate a positive evaluation (M = 8.70
/iV) as opposed to those who used their right thumb (M = 5.64
jJS/).
11 It is important to note that the evaluative inconsistency and negativ-
ity bias effects obtained in the full sample in Experiment 2 were repli-
cated in the subsample of participants whose categorizations of the
valenced targets matched the normative classifications.
896
ITO,
LARSEN, SMITH, AND CACIOPPO
"5.
E
-6-
-2-
2-
6-
10-
14-
neutral (positive block)
positive
negative
!"-) O> ^C <*1 f>
8
(S* >o oo
fl « CT* (S
0O
OO OO DO CTi
Latency (ms)
Figure 2. Averaged event-related brain potential waveforms at the midline parietal electrode (Pz) to neutral
(frequent), positive (rare), and negative (rare) targets. The amplitude of the late positive potential is not
only larger to the rare (positive and negative) targets than the frequent (neutral) targets, but it is larger to
the negative targets than the positive targets. These results were obtained even though the positive and
negative targets were equally extreme, arousing, rare, and task relevant. These results, therefore, are consistent
with the operation of a negativity bias at the evaluative-categorization stage of information pro