AVERSIVE LEARNING, INDIVIDUAL DIFFERENCES, AND
MARK KENNETH GREENWALD
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
Although my graduate training has taken me through several distinct
phases, several people have been my constant supporters, mentors and
friends. Foremost among these is Peter Lang--whose role as my supervisor
and model researcher have made an indelible mark on the way I perceive
the broad field of psychology, and how I will conduct my future research.
I thank Peter profusely for helping me to do my best.
Two equally important and outrageously talented people, Margaret
Bradley and Bruce Cuthbert, provided daily inspiration and guidance, and
served as constant reminders of my own humility. For their patience and
willingness to listen to my problems, and for the myriad times they came to
my rescue over the years, I am deeply indebted to them.
I have also been very fortunate to have two friends whose collaboration
on various projects made me appreciate their growing talents. To Alfons
Hamm, whose early teamwork eventually resulted in this dissertation, and
to Christopher Patrick, with whom I have shared countless stimulating
discussions--I am very thankful. Other scientists whose ideas have also
excited my imagination, and with whom the Lang laboratory has been a great
place to work, are Ed Cook, Mark McManis, Margaret Petry, Ellen Spence,
Scott Vrana, and David York.
During my training, my committee members have been most helpful
and supportive of my work. Keith Berg is a marvelous role model for his
keen methodological eye and technical abilities; he also taught me many
fundamentals of psychophysiology. Ira Fischler and Michael Levy were
instrumental in helping me to negotiate the ever-complicated terrain of
contemporary cognitive psychology. Jim Cauraugh, always pleasant and
enthusiastic, coached me in my emerging interest in the cognition of motor
Finally, the hectic pace of doctoral work was actually enjoyable because
three incredibly dear people kept me sane and happy. My loving wife and
companion, Margaret, and my devoted parents, David and Rosemarie, have
been my cheerleaders and saviors throughout my education. The golden
opportunities they have given me help put all this work in the proper
TABLE OF CONTENTS
ACKNOWLEDGEMENTS ........................................ ii
ABSTRACT ......... ...... ..................................... vi
INTRODUCTION. ............................................. 1
The Startle Probe in the Assessment of Emotion ................. 1
Aversive Learning........................................... 3
Sensitization .......... ............... .................. 4
Associative Learning ..................................... 5
Individual Differences .................................. 10
Statement of the Experimental Problem ........................ 13
METHODS ..................................................... 20
Subjects ............... ........ ......... ................. 20
Stimulus Materials ................... ..................... ... 20
Apparatus and Psychophysiological Recording ................... 22
Scoring and Definitions of Physiological Response Measures ...... 23
Procedure ................................................. 24
Design and Data Analysis ...................................... 31
RESULTS ..................... .............................. 36
Sensitization .................................................. 36
Startle Probe Response.................................. 36
Visceral Reactions ........................................ 40
Summary .......................................... ..... 42
Associative Learning................................... ... 42
Affective Ratings ..................... ................. 43
Startle Probe Response ................................... 47
Visceral Reactions ...................................... 54
Summary ........................... ................... 61
Individual Differences ......................................... 61
Temperament Characteristics .............................. 61
Rated Shock Aversiveness ............................. 68
Sum m ary ............................ ................... 75
D ISCU SSIO N .................................. .................. 76
Sensitization .................................................. 76
Startle M agnitude ......................................... 76
Methodological Considerations ............................ 78
Associative Learning .......................................... 80
Habituation and Sensitization in Associative Learning ....... 81
Affective Associative Change .............................. 85
Temporal Specificity and Preparatory Processes .............. 87
Brain-Behavior Relationships in Associative Learning ....... 90
Individual Differences ....................................... 91
Aversive Reactivity and Sensitization ...................... 91
Associative Learning ...................................... 93
Summ ary ................................................... 96
APPENDIX A STIMULUS SELECTION ............................... 98
APPENDIX B PROCEDURAL INSTRUCTIONS. .................. 103
APPENDIX C QUESTIONNAIRES ............................... 109
APPENDIX D AUXILIARY DATA TABLES....................... 113
REFERENCES .................................................... 138
BIOGRAPHICAL SKETCH ...................................... 150
Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy
AVERSIVE LEARNING, INDIVIDUAL DIFFERENCES, AND
Mark Kenneth Greenwald
Chairman: Peter J. Lang, Ph.D.
Major Department: Psychology
In this experiment, a psychophysiological analysis of human aversive
learning was conducted. There were three primary goals in this study. First,
sensitization of the eyeblink reflex to an acoustic startle probe (95 dBA noise
burst) and visceral activation (skin conductance, heart rate) following electric
shock exposure were examined. Subjects were randomly assigned to two
groups, which were exposed to shock either before an adaptation series of
two pictures or midway through this series. To control for unrelated effects,
each group completed questionnaires while the other was exposed to shock.
Shock exposure significantly increased startle magnitude relative to control
conditions, whereas visceral activity was not influenced.
Second, the degree to which the startle reflex, visceral activity, and
affective ratings index associative learning was tested, using differential
aversive conditioning. The series of two pictures continued and one slide
was paired with shock. Associative learning potentiated startle reflex and
electrodermal responses, and generated more cardiac acceleration and rated
aversiveness of the shocked slide (relative to the nonshocked slide). This
experiment also examined whether pairing shock with different affective
slides would yield variable degrees of associative change. Stimulus selection
was based on each subject's affective ratings, not specific content. Slides
varied in pleasantness and arousal across four groups. The results, however,
did not support the hypothesis that the affective foreground constrains the
degree of associative modulation.
Third, this study explored whether individual differences modulate
measures of aversive learning or reactivity to the prepotent stimuli used
here (i.e., shock, startle probe). Subject temperament dimensions, as well as
tolerance and rated aversiveness of the shock, were examined. Although
temperament characteristics were not generally related to learning, subjects
who were more fearful and inhibited exhibited greater reactivity to aversive
stimulation (i.e., larger startle responses, less shock tolerance, and more
reported fearfulness). In addition, subjects who rated shock as more aversive
demonstrated greater learning in startle reflex, electrodermal, and verbal
report measures. A cognitive learning model, in which representations of
the shock and conditioned stimulus are connected to the brain's aversive
motivational system, is used to explain the present results.
The startle probe reflex is reliably augmented when ongoing processing
involves aversive information. This affect-startle effect, repeatedly found in
aversive learning experiments with animal subjects (Davis, 1989a), now
forms the centerpiece for a broader theory of human emotion (Lang, Bradley,
& Cuthbert, 1990, 1992). In this dissertation experiment, a psychophysio-
logical analysis of human aversive learning is conducted. First, sensitization
of the startle reflex and measures of visceral activation (skin conductance,
heart rate) during electric shock exposure (i.e., nonassociative learning) are
examined. Second, the degree to which the startle probe reflex, visceral
activity, and affective judgments index associative learning, through pairing
photographic slides with shock, is tested. Third, the present experiment will
explore the extent to which individual differences modulate psychophysio-
logical indicators of these two forms of aversive learning.
The Startle Probe in the Assessment of Emotion
The startle reflex is an obligatory motoric response to a rapid-onset,
relatively intense, exteroceptive (e.g., acoustic) probe. Although the startle
response can be a generalized bodily reaction (Hunt & Landis, 1936), the
magnitude and latency of the eyeblink to the probe are the characteristics
most typically measured in experiments using human subjects (Anthony,
1985). This reflex occurs in many species and has been most widely
investigated in animals (e.g., the whole body response in rats), for whom the
neurobiological foundations of the primary startle circuit are now well
understood (Davis, Gendelman, Tischler, & Gendelman, 1982).
There is an additional and highly valuable use for this reflex behavior:
The blink-eliciting probe may be introduced unpredictably during
information processing (e.g., perception, mentation, learning), and changes
in the magnitude or latency of the probe reflex in these contexts (relative to
control conditions) permit one to infer the disposition of the organism. In
fact, the affective value of environmental stimuli has been repeatedly found
to modulate the magnitude of the startle probe reflex. Specifically, the
magnitude of the startle probe reflex is largest when perceiving aversive
pictures, relatively smaller during neutral pictures, and smallest during
pleasant pictures (Bradley, Cuthbert, & Lang, 1990a, 1990b, 1991; Greenwald,
Bradley, Cuthbert, & Lang, 1990; Vrana, Spence, & Lang, 1988). Similarly,
startle probe response is greater when imagining aversive scenes than
neutral or pleasant events (Cook, Hawk, Davis, & Stevenson, 1991; Cuthbert,
Bradley, York, & Lang, 1990; Vrana & Lang, 1990). Since this valence-reflex
relationship does not depend on sensory processing, it suggests that startle
modulation indexes a central motivational state.
A recent theoretical model offers a framework for interpreting this
affect-startle effect (Lang et al., 1990). In this view, emotions are conceived
as tendencies to act, represented in memory as an associative network of
context-specific perceptual and response information. A postulate of this
model is that all affective representations are fundamentally appetitive or
aversive, reflecting action dispositions of approach or avoidance (see also
Dickinson & Dearing, 1979; Konorski, 1967). Unpleasantness is associated
with the bias to avoid/withdraw, whereas pleasantness is associated with
approach/seeking behavior. Both extrinsic (environmental) and intrinsic
(subject) variables contribute to the formation of this centrally organized
response set. The efferent system, which includes verbal, overt motor (e.g.,
reflex), and visceral outputs, is motivationally tuned by this central set to
respond in accordance with ongoing processing. Physiological measures
can, therefore, be used to index this response organization in real time.
Using an affective matching principle, Lang et al. (1990) argue that
defensive behaviors (e.g., the startle response to an aversive probe) are
facilitated when the processing task also occasions an aversive state (in
which case a "match" occurs) and inhibited when the situation is pleasant
(a "mismatch"). The affect-startle effect thus reflects the biphasic conceptual
organization of emotional valence. Startle augmentation during aversive
processing thus provides a compelling link to previous dimensional
accounts of emotion, all of which emphasize the direction (valence) and
intensity (arousal) parameters of behavior (Duffy, 1972; Hebb, 1949;
Mehrabian & Russell, 1974; Osgood, Suci, & Tannenbaum, 1957; Schlosberg,
1952; Schneirla, 1959; Tellegen, 1985).
Since this study focuses on learning, a clarification of terms is first
presented. Here, learning is defined as a hypothetical process inferred from
measurable behavioral changes which result from specific experiences
(Mackintosh, 1983). Learning processes have been subdivided as either
nonassociative, e.g., within-session habituation and sensitization, or
associative, e.g., classical and operant conditioning. In this experiment,
sensitization and associative learning (via classical conditioning) are the
main subjects of interest, although habituation is also addressed as a
secondary issue. Both types of learning studied here are further
characterized as aversive by the type of stimulus which will be used to
modify behavior: electric shock. Sensitization will be inferred when a
response increment occurs to shock exposure in the absence of an
association, whereas associative learning will be inferred from response
increments which result from explicit pairings of a stimulus with shock.
It is well established that relatively intense stimulation can produce
response increments which reflect sensitization. Studies of the hindlimb
flexion reflex in the spinal cat (Thompson & Spencer, 1966) and siphon-
withdrawal reflex in Aplysia (Hawkins & Kandel, 1984) have shown that
repeated tactile stimulation can result in an initial increase in reflex
magnitude (sensitization), followed by decreased response (habituation)--
especially if stimulation is more intense. This finding led to the develop-
ment of "dual-process theory" (Groves & Thompson, 1970; Petrinovich,
1984), which proposes that sensitization and habituation are independent
neural processes, the net activation of which determines response output
at any moment. Dual process theory--which relates neurophysiological
mechanisms of habituation and sensitization to molar theories of behavior--
presumes that habituation occurs in the stimulus-response pathway,
whereas sensitization occurs because of tonic changes in the state of the
organism, which then modulate the stimulus-response pathway.
Startle modulation. Sensitization of the acoustic startle response has
been observed in several contexts. First, early habituation experiments
showed that an intense startle probe can, by itself, elicit transient reflex
sensitization prior to habituation (Davis, 1972; Groves & Thompson, 1970;
Szabo & Kolta, 1967). A second type of sensitization has been identified, in
which increasing intensity of constant background noise enhances the probe
reflex response (Davis, 1974a, 1974b; Hoffman & Searle, 1965; Ison &
Hammond, 1971; but see Putnam, 1975, for a qualification of this effect).
Third, and most recently, Davis and his colleagues have convincingly
shown that the startle reflex is sensitized following exposure to electric shock
(Boulis & Davis, 1989; Davis, 1989b; Hitchcock, Sananes, & Davis, 1989).
Davis (1989b) showed that sensitization (1) is attainable with a single shock,
but the effect increases with 5 or 10 shocks, (2) develops quickly, and (3)
cannot be attributed to rapid association with the experimental context. It is
proposed here, as Davis suggested, that this effect represents the action of
unconditioned properties of shock which can support aversive conditioning.
Therefore, sensitization appears to be a building block of conditional
learning (Hawkins & Kandel, 1984). Sensitization could also be interpreted
as a type of state transition in which response codes activated by stimulus
exposure become eligible for associative learning (cf. Sutton & Barto, 1981).
Konorski (1967) developed a similar cognitive model of learning,
proposing that exposure to an unconditioned stimulus (US) leads to an
internal sensory representation of the US, which then activates a central
motivational state related to the affective quality of the US (Figure 1). This
motivational state produces preparatory behavior which reflects the affective
character of the stimulus (e.g., withdrawal from aversive stimulation). This
account suggests why the startle reflex--a protective reaction--should be
sensitized after shock exposure. According to the affective matching
principle (Lang et al., 1990), shock exposure activates the aversive system (via
the US representation) and should prime related defensive behaviors.
The mechanism of classical conditioning, Konorski (1967) argued,
involves a simple extension of the cognitive events described for
sensitization. Following repeated presentation and pairing of the
conditioned stimulus (CS) with the US, the internal representation of the CS
aversive preparatory responses,
e.g., freezing, sweating, cardiac
facial expression, reports of fear
Figure 1. Cognitive model of aversive learning (adapted from Dickinson &
Dearing, 1979). Presentations of the CS and aversive US lead to internal
representations of each stimulus (shaded circles). The US activates (via
pre-existing connections) the aversive motivational system which, in turn,
generates associated behaviors. Reinforcing the CS with the US generates
learned connections with the US representation and aversive motivational
system. Thus, direct US exposure (as in sensitization), and CS-US pairings
(as in associative learning) can both potentiate aversive behavior.
develops independent excitatory connections with both the US
representation and the relevant motivational system (Figure 1). Due to its
association with the motivational system, the CS representation comes to
activate preparatory behavior related to the affective quality of the US.
Konorski's theory not only deals with connections and representation
but, unlike many extant cognitive models of associative learning (e.g., Estes,
1973; Frey & Sears, 1978; Kamin, 1969; Kehoe, 1988; Pearce & Hall, 1980;
Rescorla & Wagner, 1972; Wagner, 1978), also explicitly models excitatory
and inhibitory interactions between hedonic states. In this framework (see
also Dickinson & Dearing, 1979; Wagner & Brandon, 1989), aversive and
appetitive motivational systems--which comprise the directive forces of
adaptive behavior--are assumed to interact via reciprocal inhibitory
connections. Since conditioning procedures modulate affective stimulus
associations, they can be used to transform the affective meaning of a CS
(e.g., by pairing an appetitive stimulus with shock) and, thus, demonstrate
the operation of these opponent systems. Traditionally, these aversive-
appetitive interactions have been measured in animals using conditioned
suppression (e.g., Dickinson, 1977; Rescorla, 1971). However, use of startle
reflex modulation affords a number of advantages in evaluating affective
associative learning, including its independence from operant behavior.
Startle modulation. Experiments using animal subjects have shown
that the startle reflex can be augmented by aversive classical conditioning,
i.e., the "fear-potentiated startle" effect. During training, a previously
neutral light CS is paired with an aversive electric shock. In subsequent test
trials (in which shock ceases), an acoustic probe is presented during
presentation of the light cue, and in the absence of the light. In these studies,
fear-potentiated startle is defined as the within-subject difference in startle
amplitude during the cue versus its absence (Davis, 1986). Numerous
investigators have demonstrated this effect (Berg & Davis, 1984, 1985; Brown,
Kalish, & Farber, 1951; Cassella & Davis, 1986; Davis & Astrachan, 1978; Kurtz
& Siegel, 1966; Leaton & Borszcz, 1985).
However, few experiments have been conducted using human subjects.
The first studies which examined potentiated startle in humans were
originally completed about 30 years ago, and were primarily concerned with
the temporal delay of CS/probe presentation on reflex potentiation. One
conclusion from this early research (Ross, 1961; Spence & Runquist, 1958),
as well as some recent related work (Vrana et al., 1988), is that probe blink
response to an aversive cue is less reliably potentiated when the probe
follows stimulus onset by relatively short intervals (e.g., 500 msec). The
selective absence of reflex potentiation at short time lags suggests that
prepulse inhibition--an attentional mechanism--may obscure measurement
of fear induction (Davis, Schlesinger, & Sorenson, 1989).
Recently, Hamm, Greenwald, Bradley, and Lang (1991) reintroduced the
fear-potentiated startle paradigm with human subjects, suggesting that the
hiatus in experimentation (despite Konorski's enlightened views) may have
been due to a disinterest in conditioning during the "cognitive revolution."
However, the basis for understanding the relationship between learned fear
and startle potentiation was also complicated by Hull's (1943) explanatory use
of the motivational construct drive, which was presumed to have a broad
energizing effect on behavior (and is therefore analogous to arousal rather
than valence). According to this position, pairing a light with aversive
shock increased organismic drive, which then led to the increase in startle
magnitude. However, early animal studies failed to show augmented startle
effects when appetitive reinforcement was used: Probe startle responses
during a stimulus paired with food were unchanged (Trapold, 1962) or even
diminished (Armus, Carlson, Guinan, & Crowell, 1964; Armus &
Sniadowski-Dolinsky, 1966), relative to control conditions. Therefore, these
results argue against a drive theory interpretation of startle potentiation.
Hamm et al. (1991) thus adopted the Konorskian model to study
emotional learning. Unlike the single cue paradigm used with animals,
Hamm et al. used a differential learning paradigm, in which each subject
viewed two alternating slides during the experiment. For training, one slide
was always followed by shock, whereas the second slide was never followed
by shock. This enables within-subject control over the associative process.
Before and after training, the startle probe was presented during the slide
associated with shock (CS+), the slide never followed by shock (CS-), and
during interslide intervals. Visceral (heart rate, skin conductance) responses
and affective (valence, arousal) ratings of the stimuli were also measured.
Results indicated that blink reflexes were augmented when the startle probe
was administered during the slide that was paired with shock. Reflexes
evoked in the presence of this cue were significantly larger than those
elicited during the slide never paired with shock, or during the interval
between slide presentations. These data replicated and extended findings
from the animal literature.
A second objective of the Hamm et al. (1991) experiment was to explore
whether a priori stimulus affective properties influenced the degree of
emotional learning. Thus, different groups of subjects were assigned to view
slide contents which varied in affective valence and arousal throughout
learning. Two alternative hypotheses were evaluated. One hypothesis,
based on belongingness theory (Garcia & Koelling, 1966), predicted that better
conditioning should be obtained for slides with affective values which were
initially similar to shock. The second hypothesis, based on the "delta rule"
(Rumelhart, Hinton, & McClelland, 1986), predicted that better conditioning
should be obtained for slides with affective values which were initially
dissimilar to shock. Results indicated that the degree of startle potentiation
varied linearly with the a priori affective valence of the slide contents: Slides
of unpleasant content (i.e., mutilated bodies, threatening animals) yielded
significantly less startle potentiation than slides depicting pleasant content
(i.e., nature scenes, attractive nudes). Conditioned startle potentiation thus
reflected changes in hedonic value (i.e., pleasure). In contrast, the degree of
skin conductance discrimination in extinction varied with the a priori level
of arousal of the conditioned stimulus: Slides depicting arousing content
(i.e., attractive nudes, mutilated bodies) yielded significantly less conductance
discrimination than slides of calm content (i.e., household objects). These
data suggest that electrodermal differentiation reflects arousal (rather than
hedonic) change. Therefore, the results supported the delta rule, rather than
the belongingness hypothesis.
In the myriad contemporary conditioning studies which have exposed
human subjects to electric shock USs (e.g., Cook, Hodes, & Lang, 1986;
Hammond, Baer, & Fuhrer, 1980; Hugdahl & Ohman, 1977; Lanzetta & Orr,
1986; McNally & Reiss, 1984), the subject is asked--for obvious ethical
reasons--to set the shock level to be moderately to highly annoying
(uncomfortable), but not painful. In other words, the final shock level is
correlated with the subject's pain tolerance. However, none of these studies
has addressed whether a subject's tolerance for shock relates to personality
characteristics, nor whether temperament or shock tolerance differences can
predict the degree of learning. Even so, there is an older set of studies and
theory which bears on this issue.
Extensive, but inconclusive, research has been conducted on the
relationship between personality and conditioning (see Levey & Martin,
1981, for a review). As noted above, much of early conditioning research was
influenced by Hull's concept of drive; this approach was wedded with the
study of individual differences in anxiety by Spence's group at Iowa in the
1950s and early 1960s. This work was not related to a theory of personality
but, rather, tested the influence of a single trait on learning. Spence (1958,
1964) demonstrated that anxiety level (which was equated with drive)
positively related to the rapidity of conditioning, especially when the subject
reported acute anxiety. Although many learning (particularly eyelid
conditioning) experiments were conducted within this framework, aversive
shock reinforcement was not employed.
In contrast to the work above, Eysenck (1967) developed a two-factor
theory of personality emphasizing extraversion-introversion and
neuroticism as central dimensions. This scheme evolved from Pavlov's
(1957) observational typology of nervous system "strength" into modern
psychophysiological variants. The basic idea is that introverts possess
weaker (i.e., less strong) nervous systems than extraverts, rendering them
more susceptible to intense, especially aversive, stimulation. Gray (1967) and
Nebylitsyn (1972) generated research testing the hypothesis that introverts
should form conditioned responses more readily than extraverts. Gray
noted that the interaction of neuroticism--which involves sensitivity to both
rewarding and punishing stimulation--and extraversion personality factors
implies that neurotic introverts should respond more than all others to
aversive stimulation. This quadrant of the two-factor space has been
associated with high trait anxiety (Gray, 1973, 1982).
Unfortunately, there is some lack of clarity in two-factor theory and
research. First, it is not clear whether anxious (neurotic introverted) subjects
should be more reactive to aversive stimulation (e.g., shocks, loud noises),
whether they should exhibit better associative learning with aversive
reinforcers (e.g., better discrimination between a stimulus paired with shock
and one not paired with shock), or both. Second, Eysenck's early research
showed that there was a positive correlation between introversion and
neuroticism, which suggests (cf. Gray, 1982) that the Eysenckian structure is
impure. Third, impulsivity (or, conversely, inhibition)--considered by these
researchers to be a secondary factor, but given primary importance by others
(e.g., Kagan, Reznick, & Snidman, 1987)--loads positively on the extraversion
dimension and, thus, may account for variance in conditioned responding
(Barratt, 1971; Eysenck & Levey, 1972).
To minimize these problems, the present experiment uses a related
alternative framework to explore individual differences in aversive
reactivity and learning. The EASI (Emotionality [fear, anger], Activity,
Sociability and Impulsivity) temperament theory of Buss and Plomin (1975)
was chosen because of its basis in heritability studies, and because the factor
structure of the EASI assessment instrument relates to two-factor theory. For
example, Eysenck's extraversion could be roughly translated into relatively
high sociability and activity in the Buss and Plomin scheme. Thus, if the
sociability and activity dimensions were to control variance in aversive
reactivity or learning, this would argue against Eysenckian theory. In fact,
the most logical expectation is that fearful and/or inhibited temperament--
analogous to neurotic introversion--would facilitate aversive responding.
Conditioning researchers have also not addressed whether perceived
aversiveness of the shock US, regardless of personality factors or shock
tolerance, relates to the amount of affective associative learning. Within the
cognitive model of learning proposed above, this perceptual variable would
appear to be rather important. To reiterate, the US representation is
presumed to develop an excitatory connection with the CS representation
after repeated pairings, which activates a central motivational state and
produces preparatory behavior which reflects the affective character of the
stimulus. Since each representation is assumed to be influenced by
perceptual factors, then a more aversive US should enhance learning.
Statement of the Experimental Problem
The flow of experimental events generally follows the procedure of
Hamm et al. (1991) described earlier. First, prior to learning, all subjects
viewed and rated a series of colored photographic slides which varied widely
in affective valence and arousal. Based on these ratings, picture stimuli for
conditioning were selected.
Second, subjects were exposed to an aversive electric shock to evaluate
its effect on response sensitization (Figure 2). To control for passage of time,
dishabituation (from the experimenter leaving/re-entering the room), and
other factors not related to sensitization, there were two groups that served
as each other's control. For one group, shock exposure occurred prior to an 8
trial adaptation series of two slides, whereas the second group was exposed to
shock midway through the adaptation series. During each group's exposure
to shock, the other group completed questionnaires to serve as a control for
shock sensitization. For both groups, startle probes were presented before
shock exposure, and during and between slide presentations throughout the
+e r 4 -
c Wo 01
c In '
fn CU *
*0 o n
o 0 nC -
-o w vu -5
3 0 C r:
2S c- Q-
10 i c' 1-
. Qj C >
(Ug m. "^
[!-- [- cr-
u "Wo 7 -N 1
- ?i g^go, t
+ + ~d
(1 (A dJ
S^ -m .A
____ -i ?-
,u u S
za r. 6
+ U Cl .L; n+
cu n u cU( -
Third, the subject continued to view this unpredictable series of two
slides, which now served as the CS+ and CS- in a discriminative learning
procedure (Figure 3). In acquisition, one of the two habituated slides (CS+)
was always reinforced with shock at slide offset, whereas the other (CS-) was
not. Each slide was presented 8 times, but startle probes were not presented
during associative learning. This acquisition sequence was followed by a 16
trial extinction series, during which probes were again presented during and
between slides (8 presentations each slide). Finally, after the learning
procedure, a sequence of 12 slides varying in affective valence and arousal
(including each subject's CS+ and CS- pictures) was presented. Subjects rated
the slides as well as the perceived aversiveness of the shock US, CS+ slide,
CS- slide, and the startle probe.
There were three primary goals of this experiment. The first goal, based
upon demonstrations in the animal literature (Boulis & Davis, 1989; Davis,
1989; Hitchcock et al., 1989), was to assess the sensitizing effects of simple,
noncontingent, shock exposure on the startle probe reflex, as well as
activation in the cardiac and electrodermal systems. Using human subjects,
Greenwald et al. (1990) recently demonstrated that exposure to electric shock
significantly increased startle reflex magnitude, but that skin conductance
was not similarly augmented. However, their experiment was not explicitly
designed to test this hypothesis, e.g., the design did not control for response
increases which might have resulted from the passage of time.
In the present study, subjects were randomly assigned to two groups,
varying only in the timing of shock exposure, so that each group served as
the other's control in testing sensitization learning. It was expected that
shock exposure would increase the startle probe response in humans, much
as it does in animals. In contrast, studies with animal subjects have not
recorded autonomic activity before and after shock exposure; thus, although
electrodermal sensitization was not obtained by Greenwald et al. (1990), only
a weak hypothesis of "no effect" can be advanced. From the standpoint of
dimensional emotion theory (Lang et al., 1990), startle reflex sensitization
should reflect changes primarily in the hedonic (valence) state of the
individual, whereas sensitization of visceral response should indicate a
change in arousal level.
The second goal of this study, based upon early human aversive
conditioning experiments (Ross, 1961; Spence & Runquist, 1959) and recent
follow-up studies (Greenwald et al., 1990; Hamm et al., 1991), was to conduct
a psychophysiological analysis of shock associative learning using startle
reflex, visceral response, and affective judgment measures. The objective
here was to examine the sensitivity of each of these measures to simple
association (i.e., measured by potentiation of response to the slide paired
with shock [CS+]), and to associative discrimination (i.e., measured by
differences in response between CS+, the slide not paired with shock [CS-],
and in the absence of slides). The data generated by Hamm et al. (1991)
suggest that both startle potentiation and discrimination should occur. The
results of previous autonomic learning experiments (e.g., Hodes, Cook, &
Lang, 1985) suggest that cardiac response discrimination should primarily
occur in acquisition, and that skin conductance differentiation should be
present during both acquisition and extinction of the association. It is not
clear, however, whether affective judgments should better reflect the simple
association (i.e., CS+ increases in displeasure and arousal), or the
discriminative aspects (i.e., CS+/CS- difference) of the learning situation.
In addition, this experiment examined whether pairing aversive shock
with different emotional slides would yield variable degrees of affective
associative change. Whereas Hamm et al. (1991) assigned subjects to view
slide pairs which differed in affective valence and arousal, slide content was
confounded with this manipulation. In this study, there were four groups
who viewed either unpleasant-calm, unpleasant-arousing, pleasant-calm, or
pleasant-arousing slide foregrounds during conditioning. (Unlike Hamm et
al., 1991, neutral contents were not used as foregrounds in this experiment.)
Each subject viewed two pictures which he or she rated as most extreme
(regardless of content) in terms of his or her stimulus assignment to one
quadrant of the two-dimensional (valence X arousal) affective space.
The present approach is a deliberate, extreme-case test of the hypothesis
that slide affective evaluation should influence classical conditioning.
Although pairing any stimulus with shock should generally increase its
unpleasantness (and arousal), when the CS itself is aversive initially, the
amount of affective change due to association with shock may be minimal.
Conversely, a stimulus which is initially pleasant (or calm) might accrue
more displeasure (or arousal). If startle blink potentiation indexes the
amount of learned aversion, then greater changes in rated displeasure and
the startle reflex should occur for pleasant foregrounds, whereas smaller
changes are predicted for slides initially high in unpleasantness. Similarly, if
skin conductance response potentiation is an index of learned arousal, then
larger changes in rated arousal and electrodermal reactions should occur for
calm slide foregrounds, while smaller changes are predicted for slides
initially high in arousal. Such results, if obtained, would confirm the
findings of Hamm et al. (1991).
The third goal, clearly exploratory at present, was to identify individual
differences which mediate emotional learning. Since it is possible that a
subject's shock tolerance level is determined by personality characteristics,
and that temperament or shock sensitivity may relate to aversive learning,
these variables were examined. The EASI Temperament Survey (Buss &
Plomin, 1975) was used here to assess differences in aversive reactivity and
learning. The EASI subscales measure temperament dimensions of
fearfulness and impulsivity, as well as anger, activity, and sociability. If
research relating to Eysenck's personality theory is correct in suggesting that
anxious and/or inhibited subjects should either respond more to aversive
stimulation or show better conditioning, then subjects who score relatively
higher on the EASI fear subscale and lower on the impulsivity subscale
should demonstrate these effects maximally.
Finally, as the cognitive model of conditioning adopted here implies,
perceptual factors may relate to the degree of emotional learning. Therefore,
it was predicted that psychophysiological measures of learning would be
enhanced in subjects who report greater aversiveness of the shock US.
Sixty-four introductory psychology subjects (32 males, 32 females)
volunteered to participate in this experiment and received experimental
credits toward fulfilling their course requirements. Subjects were informed
before arriving at the laboratory that the study involved "cognitive and
physiological responses to stimuli in different sensory modalities." Total
participation time for each subject was two hours.
Affective pictures. Thirty-nine 35 mm colored photographic slides,
selected from catalogue of the International Affective Picture System (lAPS;
Lang, Ohman, & Vaitl, 1988), were used in this experiment. These slides
depict a wide variety of content, including wild and domestic animals,
natural scenes, objects, human faces and bodies, social interactions, and
conditions of life. The IAPS slides have relatively simple figure-ground
relationships which facilitate perceptual resolution and evoke a broad range
of affective responses within the limits of this stimulus format.
The composition of the initial slide set was based on an inspection of
normative ratings for all 240 IAPS slides (Lang, Greenwald, & Bradley, 1988).
Appendix A lists all slide materials used in this study and their normative
affective values on each dimension. The slides were distributed across the
four quadrants in the coordinate space defined by the intersection of affective
valence and arousal axes. Eight slides from each of these four quadrants
(unpleasant-arousing, unpleasant-calm, pleasant-arousing, pleasant-calm)
and four neutral-calm slides comprised the slide set. Additionally, three
practice slides (unpleasant, neutral, and pleasant) preceded presentation of
the above 36 slides during the pre-experimental viewing/judgment series, to
anchor subsequent ratings of these materials.
Acoustic startle probe. The startle-inducing stimulus was a 50 msec
burst of 95 dBA white noise (20 20,000 Hz), with instantaneous rise time
(< .05 msec). The white noise was generated by a Coulbourn S81-02 Noise
Generator, gated through a Coulbourn S82-24 amplifier, and delivered
binaurally via matched Telephonics TDH-49P earphones. To calibrate noise
intensity, the earphone was clamped to a Bruel and Kjaer Type 4153 artificial
ear (1/2" Type 4134 condenser), thus directing the sound wave perpendicular
to the diaphragm (Putnam, Graham, & Sigafus, 1975). Sound pressure level
of the continuous noise was calibrated using the fast-filtered A scale of a
Bruel and Kjaer Type 2203 sound level meter.
Aversive unconditioned stimulus (US). The electric shock US was a
continuous 500 msec train of rectangular pulses generated by a Farrall
Instruments Mark 300C Aversive Conditioner. The tactile stimulus was
delivered through a concentric Tursky electrode. The electrode was attached
to the volar surface of the left upper arm, which was first rubbed lightly with
Ultraphonic conductive gel. Intensity of the US was set at a level which the
subject described as "highly annoying, but not painful" (see Procedure).
Shock intensity was calibrated using a Micronta DC analog meter (85 ohm
internal resistance), a bridge circuit (to rectify the shock pulse), a resistance
load of 27 Kohms, and a 100 iF capacitor across the output terminals of the
Farrall shocker. For each step on the shock current knob, the value on the
Micronta meter was multiplied by the duty cycle (32.05) to obtain peak
Apparatus and Psychophysiological Recording
Stimulus presentation and data acquisition were coordinated by
distributing the processing tasks (via digital input/output connections)
between PDP 11/23 (Data Translation 12 bit analog-to-digital [A-D] converter
board) and Apple IIe microcomputers. The Apple IIe triggered a Gerbrands
electronic shutter (for slide onset/offset; rise/fall time < 5 msec) and the slide
carousel advance/reverse switches of a Kodak Ektagraphic II AM projector.
The projector lens was located behind a clear plexiglass sound baffle in the
experimenter's room; thus, the subject had no audible cues as to which slide
(CS+, CS-) would be presented next during conditioning. The PDP computer
recorded all physiological data and controlled the timing of slide, startle
probe, and shock presentations. Physiological signals (except heart rate) were
sampled at 10 Hz, except during startle probes, during which the sampling
rate was temporarily accelerated to 1000 Hz from -50 to +250 msec relative to
The eyeblink component of the startle response was measured by
recording the electromyographic activity over the left orbicularis oculi site,
using Sensor Medics miniature silver/silver chloride electrodes filled with
Teca electrolyte gel. The raw signal was bandpass-filtered (from 90 to 250 Hz)
and amplified (20,000 gain factor for all subjects) by a Coulbourn S75-01
Bioamplifier, and integrated on-line using a Coulbourn S76-01 Contour
Following Integrator (at a time constant, calibrated off-line, of 120 msec).
The analog heart rate signal was measured using collarbone and arm
leads with Sensor Medics standard size silver/silver chloride electrodes,
using Teca electrolyte gel. Following input to a Coulbourn S75-05 Isolated
Bioamplifier, the signal was filtered from 8 to 40 Hz. A Schmitt trigger
detected R-wave peaks and interrupted the computer to record interbeat
intervals to the nearest msec from 3 sec prior to slide onset to 2 sec after slide
Skin conductance activity was recorded with Sensor Medics standard
size silver/silver chloride electrodes attached adjacently to the hypothenar
eminence of the left palmar surface. The skin was first cleaned with distilled
water and sensors were filled with K-Y jelly electrolyte. A Coulbourn S71-22
skin conductance coupler provided a constant 0.5 volt across electrodes. The
signal was calibrated prior to each session to detect activity in the range from
0 to 40 gSiemens (pS). The calibration value was used to convert the
digitized raw signal off-line to analog skin conductance values.
Scoring and Definitions of Physiological Response Measures
Startle reflexes were scored using a computer program that scored the
blink waveform for onset latency and amplitude. Eyeblinks were scored by
the algorithm if (1) minimum response amplitude was 20 A-D units and if
the response had a (2) minimum onset latency of 20 msec and (3) maximum
peak latency of 150 msec relative to probe onset. Null response trials (due to
violations of criteria 1-3) were assigned an amplitude of zero A-D units.
Startle reflex magnitude (including zero amplitude responses) was the
principal measure of interest. For startle onset latency, null response trials
were excluded from analyses.
Heart rate was converted from digitized interbeat intervals to analog
rate data in half-second bins, with intervals weighted in proportion to the
time occupied (Graham, 1980). Initial and late deceleratory (Dl and D2) and
midinterval acceleratory (Al) components of the cardiac waveform were
scored (Gatchel & Lang, 1973), since these peaks and troughs in the
waveform reflect psychologically significant events (Bohlin & Kjellberg,
1979). These component scores were expressed in beat/minute change
scores, deviated from a 1 sec prestimulus baseline.
Skin conductance responses were scored using an adaptation of the
Strayer and Williams (1981) SCORIT algorithm. The First Interval Response
(FIR)--which presumably reflects orienting to stimulus onset--was defined as
the largest response during the window from 0.9 to 4.0 sec after stimulus
onset (Prokasy & Raskin, 1973). The Second Interval Response (SIR)--which
reflects anticipation of the shock US during acquisition and extinction--was
scored as the largest response on each trial between 4.1 and 6.8 sec after slide
onset (i.e., ending 0.8 sec after slide offset). Third Interval Responses (TIRs)--
which index the UCR during acquisition and orienting to shock omission
during extinction--were scored as the largest responses between 0.9 and 4.0
sec after slide offset. Skin conductance response data were later transformed
(log [SCR+1]) to normalize the distribution of responses.
Equipment preparation and instructions. After arriving at the
laboratory, each subject completed a consent form to participate in a brief (ca.
30 min) slide viewing/rating experiment. The subject was informed about
the different emotional pictures, a "brief noise click" and recording of
physiological responses. The subject was also told that that there would be a
second optional experiment, which the experimenter would describe later.
No mention of shock was made at this time. The subject was seated in a
comfortable recliner in a sound-attenuated, dimly lit room adjacent to the
equipment room. Physiological sensors were attached at this time, but
(unknown to subjects) physiological responses to slides were not actually
recorded during the first experiment.
Subjects were then instructed in using the Self-Assessment Manikin
(SAM; Lang, 1980). Subjects used the pencil-and-paper version of SAM (see
Appendix B, Figure 19) to make affective valence and arousal judgments of
experimental stimuli. To represent changes on the valence dimension, the
SAM figure ranges from smiling and happy to frowning and unhappy. To
represent changes in arousal, SAM ranges from active and wide-eyed to a
static, eyes-closed figure. There are 5 cartoon-like figures for each dimension.
Appendix B also provides the experimental instructions for using SAM.
Briefly, subjects were instructed to place an "X" on the figure, or between
figures, which best represented their affective reaction to the slide on each
dimension. This scheme formed a 1 9 rating scale for each affective
continuum. "Low" (1) on the valence dimension was unpleasant, whereas
"high" (9) was pleasant. "Low" (1) on the arousal scale was calm, and "high"
(9) was excited. During initial practice with SAM, the experimenter
instructed subjects with verbal labels for the endpoints and middle of each
scale. These labels correspond to the anchor adjectives of the semantic
differential (Mehrabian & Russell, 1974).
Prior to leaving the room, the experimenter instructed the subject that
the noise would occur "at various times during the session" and then placed
on the headphones.
Pre-experimental slide series and selection of conditioning slides.
Subjects paced themselves through a series of 39 slides, the first three of
which (unknown to subjects) were practice slides and did not contribute to
their ratings distribution. The subject was instructed to press a button which
initiated slide onset, watch the slide closely while it was on the screen, and
rate each slide for valence and arousal immediately after slide offset. After
the subject initiated slide exposure, it remained on for 2 sec. The subject was
only told that the slide exposure would be "relatively brief," so that after
pressing the button (1 sec delay), he or she should be ready to view the slide.
After each slide trial, the subject repeated this sequence. Noise probes were
not delivered during this slide sequence.
The subject was instructed to relax after the final slide trial (which was
evident from the numbered rating form). At that time, the experimenter
determined from a video monitor that the subject was finished writing and
resting comfortably. When this was confirmed, two acoustic startle probes
were presented (16 sec apart) to assess prelearning probe response.
The subject's rating form was then collected, and the experimenter
examined the two-dimensional (valence, arousal) distribution of affective
judgments. Before arriving at the laboratory, the subject had been randomly
assigned to view one of four slide types (unpleasant/arousing, unpleasant/
calm, pleasant/arousing, pleasant/calm) during learning. Accordingly, the
experimenter first selected the two conditioning slides which the subject had
rated most extremely in his or her assigned category. The experimenter then
selected ten adjunct slides--two in each of five affective categories (same four
types as above, plus neutral/calm)--which were not used in learning but
were rated again after the experiment (see below). Appendix A explains the
decision rules for choosing these conditioning and adjunct slides.
Shock exposure. The subject was instructed that a new slide sequence
would begin, and to "watch each slide for the duration of exposure and
ignore the noise." While subjects were not informed of the actual slide
exposure duration (6 sec) or when the noise would occur, probes were
presented equally often at each of 4 times during slide exposure--either 3.00,
3.75, 4.50 or 5.25 sec--as well as between slides (i.e., interslide intervals).
Subjects were instructed to relax quietly during interslide intervals, which
ranged from 10 to 20 sec in 5 sec increments.
Next, the procedure diverged for two groups (see Figure 2). Forty-eight
subjects were told, "Experiment 1 is now over." These subjects--designated
as the Immediate shock exposure group--were thanked and invited to
continue in the next experiment, which provided additional credits toward
their class research requirements. This group was told that there would be a
new consent procedure for this second experiment and that there was no
obligation to participate. If the subject agreed to continue, he or she was
given the next informed consent to read. Three subjects (in addition to the
sample of 48) declined to continue; they were thanked for their participation,
given credit, and debriefed.
The second consent procedure informed the subject about electric shock.
Otherwise, the description of slide viewing and noise probes was the same as
before. After written consent was obtained, the experimenter attached the
shock electrode and initiated shock exposure (Appendix B). In this standard
procedure, the intensity of shock was adjusted for each subject to a level
which was reported as "highly annoying but not painful." The shock level
began just above perceptual threshold and was increased gradually in fixed
steps to the subjective criterion. Increments in shock level were calibrated to
approximate a power function (Appendix B, Figure 20). All subjects reached
at least the second shock intensity.
Once the shock level was determined, the subject was told that, at
various times during the remainder of the experiment, the shock would
occur at the intensity and duration agreed upon. No mention was made of
"conditioning," association, nor timing of stimulus presentation. The
experimenter answered any questions from the subject, informed the subject
that the next phase would begin, and exited the room.
The adaptation sequence of to-be-conditioned slides began. Each slide
(CS+, CS-) was presented twice and probed twice; two startle probes were
presented randomly between slides. This was the first block of adaptation.
Then the experimenter re-entered the room to break the procedure for an
unanticipated questionnaire session. After questionnaires, the experimenter
left the room for the last time. The second block of adaptation (identical to
the first block) began, and acquisition and extinction slide series continued
uninterrupted from this point.
The remaining 16 subjects--designated as the Delayed exposure group--
completed questionnaires first instead of being exposed to shock. Following
questionnaires, they were told that a new slide sequence would be presented
and were given the instructions for slide viewing above. Adaptation block 1
slide and probe presentations were then administered in the same manner
as for the Immediate exposure group. The experimenter then re-entered the
room to inform subjects in this group that experiment 1 was over. The
experimenter then obtained the second informed consent, conducted the
shock exposure protocol, and left the room a final time. After this point, the
second adaptation block, acquisition, and extinction slide series proceeded
without interruption--identical to that for the Immediate exposure group.
To summarize, when the Immediate group was first exposed to shock,
the Delayed group completed questionnaires. This process was reversed, so
that when the Delayed group was exposed to shock, the Immediate group
completed questionnaires. (Both shock exposure and questionnaires lasted
about 10 mins.) This interlude enabled each group to serve as the other's
control group at their time of shock exposure, as well as an opportunity to
collect individual difference data. All subjects completed an abridged
version of the Fear Survey Schedule (FSS; Wolpe & Lang, 1964), consisting of
40 items describing different types of fears depicted by IAPS slide contents. In
addition, the 25 item EASI Temperament Survey (Buss & Plomin, 1975) was
administered. The EASI subscales tap constructs of emotionality (fear and
anger), activity, sociability, and impulsivity. Items on both questionnaires
were rated on a 1 5 scale, yielding a 40 200 range for this FSS version and a
5 25 range for each EASI subscale. These questionnaires (see Appendix C)
were used to address the relationship of temperament dimensions and
situational fears to aversive reactivity and learning.
Shock conditioning. The remainder of the experiment involved the
discriminative learning procedure. Subjects were randomly assigned to
view an unpredictable sequence of either two unpleasant/calm, unpleasant/
arousing, pleasant/calm, or pleasant/arousing slides. These two slides
served as CS+ and CS- (see Figure 3). During acquisition, the CS+ slide was
always reinforced with a 500 msec shock US which coincided with slide
offset. The CS- slide was never reinforced. Each slide was presented 8 times
(16 total trials; 8 different CS+/CS- orders, counterbalanced across subjects).1
1 For all subjects, the last trial of adaptation was a CS+ trial, followed by the
shock US. Since subjects could not anticipate this contingency, preceding
events were functionally part of the adaptation phase. Thus, the first CS+
presentation in acquisition accompanied a shift in the subject's experience of
event-relations and is a pure acquisition trial. By similar logic, the last CS+
trial of acquisition was not reinforced with shock. As before, subjects could
not anticipate this new contingency and, therefore, all preceding events were
functionally part of acquisition. In addition, to minimize the tendency for
subjects to expect a reversal of the contingency, an unreinforced CS- buffer
trial--which was not analyzed--was introduced between acquisition and
extinction for all subjects. The intent of these latter two manipulations was
to purify interpretation of the extinction phase.
Since the startle probe is prepotent aversive US, probes were not presented
during acquisition so as not to be associated with learning.
During prior adaptation and subsequent extinction phases only, probes
were presented during and between slides. As noted earlier, each of the 2 CS
slides was presented and probed 4 times during the adaptation series. These
8 trials were intermixed with 4 no slide/probe trials (12 total trials). Because
of the between-group shock exposure manipulation, adaptation was divided
into two equal parts. In each of four extinction blocks, each slide was seen
and probed twice, and 1 no slide/probe was presented. Thus, there were a
total of 8 CS+, 8 CS-, and 4 no slide/probe trials.
Throughout conditioning, slide exposure was 6 sec. Variable interslide
intervals ranged from 10 to 20 sec in 5 sec increments.
Postlearning: Affective ratings and debriefing. After the last extinction
trial, the experimenter removed the recording sensors, shock electrode, and
headphones. The experimenter informed the subject that the noise and
shock would no longer be presented. The experimenter then instructed the
subject that he or she would view a final series of slides and rate each slide
again using SAM. The procedure was identical to prelearning ratings, except
that there were only 12 slides. Two of these slides were the subject's CS+ and
CS-. The remainder were the 10 adjunct slides falling into the five affective
categories mentioned above. After completing the slide ratings, the subject
also rated (from memory) the noise and shock stimuli with SAM.
After affective judgments were completed, subjects were asked in a
debriefing session to state (1) their estimate the number of shocks received;
(2) whether they recognized any rule during the procedure; (3) whether some
stimuli were followed by shock more often than others; (4) whether they
heard the noise over the headphones; (5) if they knew which physiological
measures were being recorded; and (6) whether they knew or could guess the
purpose of the experiment. In addition, they were also asked to rate each of
the following items for their aversiveness (i.e., "the extent to which you
would avoid the stimulus") on a 1 (not at all aversive) to 7 (extremely
aversive) scale: "noise over the headphones," "shock," "shocked slide," and
"nonshocked slide." After these debriefing questions, the subject was
informed about the purpose of the experiment, thanked, given class credit,
Design and Data Analysis
In previous human aversive learning studies, subjects have received
shock exposure prior to learning. In this experiment, however, designed
variations in the timing of shock exposure across groups complicate
interpretation of the adaptation phase data. Thus, to facilitate interpretation
of the conditioning data within this context, the effects of shock exposure
timing on startle reflex, tonic visceral levels, slide visceral responses, and
habituation rates for these measures are reported first.
Otherwise, this study largely repeats (with minor improvements) the
discriminative conditioning procedure used by Hamm et al. (1991). To
facilitate comparisons with those data, univariate analyses of physiological
measures (startle magnitude and onset latency, electrodermal response, and
heart rate response) were conducted separately for each phase of
conditioning (adaptation, acquisition, and extinction). Affective ratings
(valence and arousal) were assessed before and after associative learning.
The experiment was divided, both from the standpoint of design and
analysis, into several parts, hereafter referred to as (1) Prelearning, or initial
slide viewing/rating; (2) Sensitization; (3) Associative learning, comprised of
adaptation, acquisition, and extinction phases; and (4) Postlearning. It is
important to note that the adaptation slide sequence is common to both the
sensitization and associative learning segments.
Prelearning. The first objective was to assess the ratings distribution for
each subject and--based on the subject's a priori group assignment--use these
ratings to select two matched slides as foregrounds for conditioning. Thus,
analyses of these ratings served as a manipulation check that the four
conditioning groups significantly differed in valence and arousal, and that
conditioning slides within each group did not differ prior to learning. Slide
Valence (2: Unpleasant, Pleasant; between-subject) X Slide Arousal (2: Calm,
Aroused; between-subject) X Stimulus (2: CS+, CS-; within-subject)
univariate analyses of variance (ANOVAs) were conducted on both valence
and arousal ratings. Similar analyses were conducted on adjunct (i.e.,
nonconditioned) slides to determine their distribution in the affective space.
Sensitization. To assess shock sensitization of the startle reflex without
the confound of different affective slide foregrounds, analyses focused on
startle probe responses during interslide intervals. Missing data (2% of all
prelearning and adaptation trials) were estimated by regression, separately
for each exposure group. One subject in the Immediate exposure group was
excluded because more than 50% of his responses had been rejected, leading
to insufficient data for estimation. Changes in the probe startle response as a
consequence of shock exposure were analyzed in an ANOVA for Exposure
Group (2: Immediate, Delayed; between-subject) X Interslide Probes (2) X
Probe Blocks (3: prelearning, first half adaptation, second half adaptation;
To examine sensitization of tonic visceral levels during shock exposure,
average heart rate and skin conductance for the 1 sec prior to each adaptation
trial (in the absence of slide stimulation) were computed for each subject.
Since there were 12 adaptation "trials"--interslide intervals were treated as
trials here--the change in pretrial baseline levels during shock exposure
could be examined. Although tonic visceral level data were not available
during the prelearning rest period, the effect of shock exposure on visceral
levels could be tested for the Delayed exposure group since, for these subjects,
adaptation block 1 preceded exposure, and adaptation block 2 followed
exposure. These data were explored using Exposure Group (2: Immediate,
Delayed; between-subject) X Trials (6: consecutive, regardless of CS+, CS- or
interslide) within Trial Blocks (2: first half adaptation, second half
adaptation; within-subject) ANOVAs.
To examine effects of shock exposure on slide visceral reactivity, slide-
by-slide heart rate leg scores and skin conductance responses (consecutive
trials, without regard to CS+ or CS-, within each half of adaptation, as well
as acquisition and extinction) were analyzed, separately for each exposure
group. Slide visceral responses were tested using Exposure Group (2:
Immediate, Delayed; between-subject) X Slide Trials (4: consecutive,
regardless of stimulus) X Blocks (2: first half adaptation, second half
adaptation; within-subject) ANOVAs.
Associative learning. All physiological data were analyzed in blocks of
two trials separately for each phase of conditioning (adaptation series [2 trial
blocks], acquisition [4 trial blocks], and extinction [4 trial blocks]) with
univariate ANOVAs. Subjects were randomly assigned to receive one of
four slide affective foregrounds during conditioning. Slide Valence and
Arousal (2 levels each) constituted the between-group factors in the design.
To evaluate conditioning within groups over trials, Stimulus (2: CS+, CS-)
and Trial Block (2 or 4 within phase) factors formed the remaining factors of
the design. For ANOVAS involving the trial block repeated measure term,
unadjusted degrees of freedom and Greenhouse-Geisser epsilon (e)
corrections are reported.
In analyses of associative learning, "discrimination" was defined as the
extent to which CS+ response exceeded CS- response during acquisition or
extinction. Startle potentiationn" was defined as the degree to which CS+
response increased from adaptation block 2 to extinction block 1. Since
visceral responses rapidly habituate, potentiation of visceral activity was
tested as the change in CS+ response from adaptation block 2 to acquisition
Postlearning. For affective judgments, changes in affective ratings of
each subject's CS+ and CS- before and after conditioning were tested in
univariate Phase (2: Pre, Post) X Stimulus (2: CS+, CS-) X Slide Valence (2:
Unpleasant, Pleasant) X Slide Arousal (2: Calm, Aroused) analyses of
variance for valence and arousal ratings. Statistical tests evaluated the
hypothesis that CS+ ratings should become (or remain) more unpleasant
and arousing than CS- ratings. Similar analyses were conducted on adjunct
slides to assess the degree of affective change.
Individual differences. Relationships of the temperament dimensions
to both aversive reactivity and learning were initially examined with zero-
order correlations and stepwise multiple regression. The EASI subscale
scores were entered as independent variables in the regression models to
predict measures of aversive reactivity (e.g., shock tolerance, prelearning
startle response, shock aversiveness, situational fearfulness) and the extent
of aversive learning (e.g., startle sensitization and potentiation, cardiac and
electrodermal conditioning, affective ratings change). Follow-up ANOVAs
were conducted to describe group mean differences.
Relationships of shock tolerance and rated shock aversion to learning
measures were also examined. Stepwise multiple regression was used to
enter both variables to predict the extent of learning in the psychophysio-
logical response measures. Again, ANOVAs were conducted to describe
mean differences among groups.
For all statistical tests, the size of the rejection region was set at .05.
The strategy for data analysis and presentation proceeds as follows.
First, the effect of shock exposure on response sensitization during the
initial slide adaptation series is examined. Second, the influence of aversive
classical conditioning on psychophysiological responses is evaluated,
separately for initial adaptation, acquisition, and extinction slide series.
Third, effects of individual differences (temperament characteristics,
tolerance of shock, and judgments of the aversiveness of stimuli) on
psychophysiological reactivity and associative learning are assessed.
Startle Probe Response
Reflex magnitude. It was predicted that after exposure to shock, each
group would show a marked rise in startle magnitude, and that this would
not occur under control conditions. As mentioned earlier, this was tested
using interslide interval data to avoid the confounding influence of slide
valence or arousal on sensitization. As predicted, startle magnitude was
clearly sensitized by shock exposure (see Table 1). The omnibus test of
sensitization learning was significant, Exposure Group X Trial Block
F(2,124)= 7.11, e= .91, p< .002. Planned comparisons testing the sources of this
interaction used blocked averages for each subject, since interslide responses
within prelearning and adaptation series probe blocks did not differ.
There was no group difference in startle magnitude during prelearning
(i.e., before shock exposure), F(1,62)= 1.77, p= .19. For the Immediate
exposure group, the increment in startle magnitude following shock
Startle magnitude means (ADUs + standard errors)
during prelearning and interslide intervals of adaptation.
Exposure Group Prelearning Adapt. block 1 Adapt. block 2
Immediate 364 (shock 502 424
(51) exp.) (59) (56)
Delayed 236 290 (shock 415
(65) (72) exp.) (81)
Startle onset latency means (msec + standard errors)
during prelearning and interslide intervals of adaptation.
Exposure Group Prelearning Adapt. block 1 Adapt. block 2
Immediate 38.3 (shock 35.1 36.1
(1.0) exp.) (1.0) (1.1)
Delayed 45.9 39.2 (shock 36.7
(2.1) (1.2) exp.) (1.6)
/ I- ia
(siiun a-V) m!j4 4luuealajd Isuj
wuoi apnlrUeuw -uilq ui afueqD
'0 r-; 0
baa) E j
~c fi a
rn ~ <"
exposure (A= +138 ADUs) was significant, F(1,47)= 18.57, p< .005. After the
sensitization peak, startle response habituated, F(1,47)= 12.30, p< .005.
For the Delayed exposure group, the increase from prelearning to
adaptation series block 1 (control period; A= +54 ADUs) was not significant,
F(1,15)= 3.22, p> .09. As predicted, the subsequent increase in startle reflex
magnitude directly attributable to shock exposure (A= +125 ADUs) was
statistically supported, F(1,15)= 9.42, p< .01.
As seen in Figure 4, startle magnitude increased markedly at the first
probe after shock exposure for each group. The size of the sensitization effect
was equal across groups (about 40%), and absolute startle magnitude in block
2 of the adaptation series--after both groups had been exposed to shock--was
nearly identical (see Table 1). This is noteworthy, since the end of the initial
adaptation series was intended to serve as the general baseline for the test of
startle potentiation. Also, startle reflex sensitization was independent of
orbicularis oculi tension (Appendix D, Table 7) and responses during slides
were comparable with responses during interslide intervals.
Startle onset latency. Although latency differences were obtained, the
findings were not consistent with a shock sensitization effect (Table 2).
Shock exposure was expected to speed onset latency (first for the Immediate
group, later for the Delayed group), but latencies were generally speeded
from prelearning to the first block of the adaptation series for both groups,
F(2,122)= 15.84, e= .88, p< .0000. A Group X Block interaction did emerge
(which would be anticipated for sensitization), F(2,122)= 5.06, e= .88, p< .02,
but this was due to overall slower responses in the Delayed group, F(1,61)=
6.66, p< .02, which became significantly quicker in onset by the second block
of the adaptation series. Thus, the present data do not support the
conclusion that startle onset latency is reliably related to shock exposure.
Skin conductance. Electrodermal tonic levels habituated within blocks
of adaptation trials for both groups. Although Table 3 suggests a group
difference in levels in block 1 of the adaptation series, this effect was weak
(p= .15). There was a large increase in conductance level for the Immediate
group from adaptation block 1 to 2 (i.e., after questionnaires), but not for the
Delayed group (i.e., after shock exposure), Group X Trial Block F(1,60)= 6.42,
p< .02. The source of this interaction was the significant increase in the
Immediate group. Importantly, conductance levels of the Delayed group
were not sensitized by shock exposure (p= .64). Tonic levels of the two
groups did not significantly differ in acquisition or extinction.
Slide conductance responses of both groups increased from block 1 to 2
of the adaptation series; the slightly greater increase for the Delayed shock
exposure group was not significant, Group X Trial Block F(1,62)< 1 (Table 3).
Tested alone, the increase in the Delayed group following their exposure to
shock was not reliable (p= .14). Since slide electrodermal response increased
in both groups, dishabituation may have contributed here. Incidentally, the
Immediate group had larger mean skin conductance responses to slides
throughout conditioning, but this overall group effect was only significant in
acquisition, F(1,62)= 3.85, p= .05. The Immediate group also had larger
Second Interval anticipatory responses than the Delayed group in acquisition
(Ms= .14 and .09 pS), F(1,62)= 5.33, p< .03, but the groups did not differ in
Third Interval Response magnitude. Shock exposure groups did not differ
on Second Interval or Third Interval response measures in extinction.
Heart rate. Heart rate levels of the Immediate group generally decreased
from block 1 to block 2 of the adaptation series, whereas levels of the Delayed
group increased slightly (see Table 3). There was a Group X Trial Block
Mean visceral levels and slide responses for
Immediate and Delayed shock exposure groups.
Visceral Adaptation Acquisition Extinction
Measure block 1 block 2
Skin conductance tonic level (pS)
Immediate 6.42 7.76 7.05 6.79
Delayed 7.75 7.92 8.03 7.69
Slide skin conductance response (log [FIR+11 pS)
Immediate 0.23 0.24 0.22 0.13
Delayed 0.15 0.19 0.13 0.08
Heart rate tonic level (bpm)
Early-interval slide deceleration
Mid-interval slide acceleration
Late-interval slide deceleration
interaction, F(1,61)= 4.33, p< .05, which was due solely to habituation of
levels across blocks in the Immediate group, F(1,46)=9.23, p< .005; that is,
tonic heart rate levels did not increase (i.e., subjects were not sensitized) in
the Delayed exposure group (p= .53). Heart rate levels did not differ across
groups during conditioning.
Shock exposure also did not sensitize cardiac response to the slides
during the adaptation series, although group differences emerged later in
conditioning (Table 3). For initial deceleration (D1 leg), the Delayed group
decelerated more than the Immediate group during acquisition, F(1,61)= 8.02,
p<.01, and extinction, F(1,61)= 4.41, p< .05. The Delayed group showed
greater mean deceleration late in the slide interval (D2 leg) in acquisition,
but this was not significant (p= .13). In extinction, the group difference in
late deceleration was significant, F(1,61)= 4.76, p< .04. There were no group
differences in Al leg response during any part of conditioning.
Only startle magnitude was significantly influenced by shock exposure:
Startle responses were 40% larger immediately following shock, relative to
control conditions, suggesting that this increase cannot be attributed to
dishabituation or simple passage of time. In contrast, neither startle onset
latency nor visceral reactions were sensitized by exposure to shock.
Prior to learning, the CS+ slide, CS- slide, and 10 adjunct slides were
selected from each subject's ratings distribution (see Appendix A).1 Mean
1 The terms "Negative," "Positive," "Calm" and "Arousing" and their
combination (e.g., "Negative/Calm") refer to the different conditioning
groups viewing slides of different affective valence or arousal. To minimize
confusion, these group references will be distinguished from the lower-case
valence and arousal ratings for conditioning and two control slides--which
were drawn from the same-affect category as each subject's CS slides--are
presented in Appendix D (valence, Tables 8-9; arousal, Tables 10-11). Adjunct
slide ratings (Appendix D, Tables 12-13) were not significantly affected by
Valence judgments. Figure 5 illustrates each group's affective valence
and arousal ratings of their foreground slides both before and after learning.
Valence ratings of CS+ and CS- slides did not differ prior to learning. Not
surprisingly, affective valence ratings by subjects assigned to view pleasant
CSs were higher than ratings by subjects assigned to view unpleasant CSs
(regardless of arousal), F(1,60)= 1910.38, p< .0000.
Regardless of content, CS+ was rated more unpleasant than CS- after
learning (Ms= 3.95 vs. 4.55), F(1,60)= 9.78, p< .003; i.e., valence discrimination
did not significantly vary by affective slide condition.
Analysis of valence change scores also indicated that, overall, CS+
became significantly more unpleasant than CS- (As= -.81 vs. -.25 units),
F(1,60)= 7.87, p< .01. Figure 5 shows that the direction of ratings change was
opposite for subjects viewing unpleasant vs. pleasant foregrounds, F(1,62)=
54.86, p< .0000. Follow up tests found that for Positive slide groups, CS+
became significantly more unpleasant (A= -2.19) whereas for Negative
groups, CS+ did not significantly change (A= +.56).
Depending on the slide foreground, conditioning and control slides
(from the same affective category as the subject's CSs, but not conditioned)
terms "unpleasant," "pleasant," "calm," and "arousing," which denote the
affective qualities of slide pictures or subjects' ratings of the pictures.
S 2 3 4 5
Figure 5. Affective valence and arousal ratings of conditioned
stimuli before learning (origin of lines) and after learning (terminal
point of arrows), separately for each slide foreground group.
J I I
.n b 4
bo < ne
0 0 b I
s s s
u54 -V .
n > >
(s4unm Supei faid-4sod) aueq)ip aualeA
a oa 9
showed different relative change, Slide Valence X Stimulus F(2,124)= 9.93,
p< .0001 (see Figure 6). Follow up tests found that, for Negative groups, CS-
became significantly more pleasant than both CS+ and control slides. In
contrast, a gradient resulted for Positive slide groups, in that CS+ became
marginally more unpleasant than CS- (p< .07), whereas both CSs became
significantly more unpleasant than control stimuli.
The shock US and startle probe were rated as equally unpleasant after
learning (Ms= 1.97 and 2.16, respectively).
Arousal judgments. Arousal ratings of the CS+ and CS- slides did not
differ before learning. Due to the experimental selection criteria, arousal
judgments by subjects assigned to view more exciting CSs were significantly
higher than ratings by subjects assigned to view calm CSs (regardless of
valence), F(1,60)= 62.31, p< .0000.
Unlike valence ratings, classical conditioning did not produce arousal
judgment discrimination. The overall difference between CS+ and CS- was
not significant, F(1,60)< 1, and did not vary by slide foreground.
Arousal change scores also failed to provide evidence of learning.
Although CS+ maintained slightly more arousal than CS-, the overall
difference between CS+ and CS- change was not significant, F(1,60)< 1.
However, the direction of ratings change was opposite for subjects viewing
calm vs. arousing foregrounds (see Figure 5), F(1,62)= 15.96, p< .0005. For
Arousing groups, CS+ became significantly more calm (A= -1.59), whereas
for Calm groups, CS+ did not significantly change (A= +.60). There were no
other significant findings.
Depending on the slide foreground group, conditioning and control
slides showed different relative change, Slide Arousal X Stimulus F(2,124)=
5.16, p< .01 (see Appendix D, Table 11). For Arousing slide groups, both CS+
and CS- became significantly more calm than control slides (ps< .02). Calm
slide groups failed to show this effect; differences in arousal change between
control slides and both CS+ and CS- slides were not significant.
Arousal ratings of the shock and startle probe were similarly high (Ms=
7.19 vs. 6.77); even so, this small difference was reliable, F(1,60)= 4.08, p< .05.
Startle Probe Response
Reflex magnitude. As expected, CS+ and CS- startle magnitude did not
differ before learning (see Appendix D, Tables 14-15 for reflex magnitude
means for all groups during conditioning). There were no other significant
effects on startle magnitude during the adaptation slide series.2
It was predicted that startle magnitude evoked during the CS+ would
increase immediately after learning. Clearly, however, the ability to detect
change at extinction relies on the initial level from which change is assessed.
Recall that delaying shock exposure for some subjects in this sample was a
departure from other conditioning studies (in which subjects have routinely
been exposed to shock prior to the adaptation series). Although mean startle
responses of the two groups did not differ in the second block of adaptation,
responses of the Delayed group were relatively high within that group due to
their recent shock exposure; conversely, responses in the Immediate group
had decreased from their earlier postshock exposure (see Table 1). Thus, the
larger responses for Delayed subjects may not have fully habituated, making
it more difficult to detect potentiation.
2 Subjects viewing unpleasant slides had larger mean startle responses
(across CS+ and CS- pictures) than subjects viewing pleasant slides during the
adaptation series (Ms= 518 vs. 360 ADUs), F(1,60)= 3.18, p< .08. Covarying for
prelearning startle magnitude levels, however, eliminated this marginal
effect. These apparent slide foreground response differences could therefore
be attributed to initial subject response differences.
Overall analysis indicated that the CS+/shock pairing produced a mean
increase in startle magnitude from block 2 of the adaptation series to block 1
of extinction (A= +41 ADUs, from 443 to 484 ADUs), but this change was not
significant, F(1,62)= 1.85, p< .18. A closer look at the trials immediately
surrounding the acquisition phase (i.e., the last trial of adaptation and first
trial of extinction) showed that startle was significantly potentiated across all
subjects (A= +139 ADUs, from 414 to 553 ADUs), F(1,61)= 5.20, p< .03.
There was, however, a dramatic difference in startle reflex potentiation
between the two groups differing in the timing of shock exposure (Figure 7).
The Immediate exposure group showed startle potentiation (A= +83 ADUs),
whereas the Delayed group showed a commensurate decrease (A= -86 ADUs),
F(1,62)= 6.62, p< .02. Tested alone, the increase in the Immediate group was
significant, F(1,47)= 6.77, p< .02. Thus, in a statistical test directly comparable
with previous studies, fear-potentiated startle was replicated.
Overall mean CS+ response was significantly larger than CS- response
throughout extinction, F(1,62)= 5.84, p< .02, and this discrimination did not
vary over slide trial blocks (see Figure 8). Clearly, startle magnitude declined
overall during extinction, F(3,186)= 35.94, p< .0000. There were no Slide
Arousal or Shock Exposure group effects on extinction startle discrimination.
In two recent experiments, startle discrimination tended to be greater
among subjects viewing pleasant slides than unpleasant slides. In this study,
this effect failed to replicate, Slide Valence X Stimulus F(1,62)< 1 (Figure 9).
An effect of Slide Valence on CS+ startle potentiation (cf. Hamm et al., 1991)
was also not replicated, F(1,62)< 1. Consistent with these recent studies, the
Negative/Arousing group showed sma.llir mean startle potentiation than
other groups (-7 vs. 122, 15, and 32 ADUs for Negative/Calm, Positive/Calm,
and Positive/Arousing groups, rcspc'il\'l.v'). I however, statistical analyses of
Blocks of 2 CS+ Trials
Figure 7. Extent of CS+ startle potentiation in groups exposed to
shock immediately before the adaptation series (open circles) or
delayed exposure to shock midway through the adaptation series
Figure 8. Startle magnitude differentiation in the overall
sample (n=64). Mean responses are presented in blocks of
2 trials during the adaptation and extinction slide series.
Note that probe reflexes decrease in extinction to a point
which is below prelearning levels (dashed line), and that
CS- response reaches this level before CS+ response.
I I I
Ext2 Ext3 Ext4
Adapl Adap2 Extl
Figure 9. Startle magnitudes during adaptation and extinction
slide series of classical conditioning, separately for groups (n=32
each panel) viewing either pleasant slides (above) or unpleasant
pictures (below). Note the pattern of reflex habituation in each
group relative to prelearning levels (dashed lines).
(,vsuI),(:)ua~e-l 4asuo alijuls
potentiation scores did not support these impressions (p> .20). Finally,
retesting Slide Valence effects only for the Immediate exposure group (which
is comparable with previous studies) did not change these null findings.3
The time at which probes were delivered influenced startle magnitude
in extinction. Startle response was generally larger to probes presented late
during the slide interval (regardless of CS+ or CS-) relative to earlier probes
(Ms= 358, 340, 374, and 387 ADUs for times 1-4). Variation across probe times
was significant, F(3,186)= 3.83, e= .90, p< .02; linear F(1,62)= 5.11, p< .03;
quadratic, p< .15. Further confirming this effect, mean response at the last
two probe times (381 ADUs) was greater than overall startle response at the
first two probe times (349 ADUs), F(1,63)= 8.29, p< .01. Startle discrimination
was consistent across probe times, Stimulus X Probe Time F(3,186)< 1.
Onset latency. There were no significant slide group, stimulus or trial
block effects for startle latency before aversive learning. (See Appendix D,
Table 16, for onset latency means in all groups and the overall sample.)
Startle onset latencies were evaluated for both learned facilitation and
discrimination, analogous to startle magnitude. Figure 10 illustrates startle
latencies from the end of adaptation (i.e., the base for facilitation) through
extinction. Due to missing data, nine subjects were excluded from these
analyses (n= 55).
3 Unexpectedly, Slide Valence influenced change in CS- startle response
from adaptation to extinction, F(1,62)= 4.63, p< .04. Probe response to CS- was
significantly potentiated (A= +82 ADUs) for Negative slide groups, F(1,31)=
4.16, p= .05, but CS- response did not significantly change (A= -25 ADUs) for
Positive groups. Although this finding suggests greater generalization of CS+
startle potentiation to CS- when subjects viewed unpleasant slides, it should
be viewed cautiously: Different patterns of CS- startle change were obtained in
previous experiments assessing affective modulation of the startle response
(Greenwald et al., 1990; Hamm et al., 1991).
From adaptation to the first block of extinction, CS+ latency was
speeded, whereas CS- latency was slowed. The interaction of these changes
was significant, F(1,51)= 4.41, p< .05. Whereas CS- slowing (A= 1.4 msec) was
significant, F(1,51)= 4.15, p< .05, CS+ speeding was not (A= 1.1 msec, p= .24).
Probe response to CS+ was faster overall than CS- response during extinction
(Ms= 32.4 vs. 34.1 msec), F(1,51)= 8.31, p< .01. Although CS+ onset latencies
returned to prelearning levels, discrimination did not significantly vary over
trial blocks. Onset latency of the reflex between slides (M= 39.0 msec) was
significantly slower than during slides.
Similar to startle reflex magnitude, mean onset latency discrimination
in extinction was smaller for the Negative/Arousing group than the other
slide groups (0.1 vs. 2.2, 1.5, and 2.5 msec for Negative/Calm, Positive/Calm,
and Positive/Arousing groups, respectively). However, the statistical test of
this impression was not significant (p= .19). There were no other remarkable
effects of affective slide content on modulation of startle onset latency.
Skin conductance. Appendix D (Tables 17-18) presents First Interval
skin conductance responses for all slide foreground groups throughout
conditioning. Electrodermal response magnitude decreased from adaptation
trial 1 to 2, increased at trial 3, and decreased again from trial 3 to 4. This
within-block response habituation was significant, F(1,60)= 24.68, p< .0000.
The increase between adaptation blocks probably represents dishabituation
due to either shock exposure or completion of questionnaires (see above).
Analogous to startle potentiation, the Immediate exposure group
showed increased conductance response from the end of the adaptation
series to the first block of acquisition, whereas the Delayed exposure group's
response decreased (As= .051 and -.054 pS; see Figure 11). This difference in
Blocks of 2 CS+ Trials
Figure 11. Extent of CS+ skin conductance response potentiation
in groups exposed to shock before adaptation (open circles) or
delayed exposure to shock midway through adaptation (dark
M 0 0
Z &=4 8
oo d .
JTb4 0 ;C4
0 0 0 ft8 0
r( 9 0
conductance potentiation was significant, Group X Trial Block F(1,62)= 3.96,
p= .05. The two groups did not differ in their skin conductance responses at
the end of adaptation.
In acquisition, First Interval Response magnitude to CS+ was larger
than to CS- overall (Ms= .24 and .15 pS), F(1,60)= 38.05, p< .0000 (Figure 12).
Conductance response magnitude habituated during learning (Ms= .22, .21,
.19, and .17 pS, for blocks 1-4), F(3,180)= 5.19, e= .84, p< .005. Second Interval
Response magnitude, reflecting shock anticipation, was greater during CS+
than CS- overall (Ms= .18 and .07 pS), F(1,60)= 47.96, p< .0000. Relative to
CS-, the Second Interval Response to CS+ increased monotonically across
trial blocks, F(3,189)= 5.47, e= .97, p< .002. Not surprisingly, unconditioned
responding to shock (i.e., Third Interval Response magnitude after CS+; M=
.50 pS) was dramatically greater than activity after CS- offset (M= .06 pS),
F(1,60)= 320.53, p< .0000. These unconditioned responses to shock declined
over trial blocks (Ms= .57, .51, .45, and .45 pS), F(3,189)= 4.27, e= .80, p< .01.
In extinction, the First Interval Response was greater for CS+ than CS-
(Ms= .14 and .09 pS; see Figure 12). This overall discrimination effect was
significant, F(1,60)= 11.65, p< .002. First Interval Response magnitude, which
declined from acquisition, continued to decrease in extinction. This general
decline across extinction trial blocks (Ms= .14, .12, .11, and .10 pS) was
significant, F(3,180)= 4.09, e= .99, p< .01.
Whereas both Hamm et al. (1991) and Greenwald et al. (1990) found that
electrodermal discrimination in extinction was larger in the context of calm
slides, neither Slide Arousal nor Valence influenced conductance response
differentiation in this experiment (both main effect Fs< 1). Consistent with
these past studies, conductance discrimination was virtually absent in the
Negative/Arousing group, relative to other rmrupps, in extinction (.01 vs. .06,
.04, and .08 pS in Negative/Calm, Positive/Calm, and Positive/Arousing
groups, respectively). This impression of an isolated group difference (as
suggested by the Slide Valence X Slide Arousal X Stimulus interaction),
however, was not supported, F(1,60)= 2.50, p< .12.
Electrodermal activity was larger in anticipation of shock onset during
extinction. The Second Interval Response to CS+ was significantly greater
than to CS- (Ms= .24 and .19 pS), F(1,60)= 11.14, p< .002, and habituated
overall (Ms= .33, .22, .17, and .16 pS), F(3,180)= 36.39, e= .93, p< .0000. Third
Interval Response magnitude (i.e., due to shock omission after CS+) was
small and undifferentiated in extinction.
Heart rate. Tables 19-22 in Appendix D present heart rate level and leg
scores, by slide foreground group, throughout the experiment. Adaptation
prestimulus heart rate levels tended to be higher for Positive slide groups
than Negative groups, F(1,59)= 3.68, p< .06. Levels were also higher prior to
the CS+ slide than the CS- slide for the Negative groups, Slide Valence X
Stimulus F(1,59)= 5.01, p< .03. Because heart rate levels varied within and
between groups, analyses of leg scores during the adaptation series were
conducted using baseline heart rate for each CS at each trial block within
each affective valence group as covariates.
In adaptation, there was significantly less cardiac acceleration (Al leg)
for CS+ than CS- (Ms= 2.55 vs. 4.28 bpm), F(1,59)= 9.62, p< .005, which was
not altered by covariance analysis, F(1,58)= 5.75, p< .02. Initial deceleration
(Dl leg) and late-slide deceleration (D2 leg) tended to be larger for CS+ than
CS-, F(1,59)= 3.72, p< .06, and F(1,59)= 4.06, p< .05, but adjusting for tonic
heart rate attenuated both findings in both instances (ps> .35).
In acquisition, cardiac acceleration was the best index of learning. As
Figure 13 suggests, the initially smaller acceleration to CS+ (originating in
C C? <
-4 ('. et 4 9 .
B as JU
(udq) agueqD alet Je-aH
adaptation, but not significantly different from CS- response in acquisition
block 1) gradually increased; there was a concomitant decrease in acceleration
to the CS- slide over trials. Although the overall learning effect (i.e., CS+
more acceleratory than CS-) was not significant, F(1,59)= 1.84, p= .18, there
was a linear increase in discrimination with repeated slide/shock pairings.
The Stimulus X Trial Block interaction was marginal, F(3,177)= 2.60, e= .93,
p< .06, but the difference in CS+ vs. CS- linear trends across blocks was
significant, F(1,62)= 7.99, p< .01. Thus, associative learning reversed the
initial disparity in slide cardiac response from the adaptation series and did
so monotonically. Furthermore, the learning effect was a function of both
incremental acceleration to CS+ and declining acceleration to CS-.4
Unconditioned cardiac acceleration (A2 leg) to shock after CS+ offset
was greater than CS- postslide response (Ms= 10.7 vs. 5.8 bpm), F(1,59)= 45.62,
p< .0000. The unconditioned response diminished over trial blocks (Ms=
11.6, 11.0, 9.9, and 7.0 bpm), whereas cardiac response after CS- stayed smaller
and constant. This differential habituation effect was also significant,
Stimulus X Trial Block F(3,180)= 3.28, e= .94, p< .03.
In extinction, late cardiac deceleration (D2 leg) was greater during CS+
than CS- (Ms= -5.2 vs. -4.2 bpm), reflecting anticipation of shock, F(1,59)= 4.78,
p< .04. There were no other significant leg score findings during extinction.
4 Groups viewing arousing slide foregrounds showed the increasing
cardiac accelerative differentiation more than Calm slide groups, Slide
Arousal X Stimulus X Trial Block F(3,183)= 2.76, e= .93, p< .05. This was due
to incremental learning across blocks in the Arousing slide groups but also
reversed conditioning for Calm groups in trial block 2 (see Appendix D, Table
21). However, modulation of cardiac acceleration by slide foreground has not
been observed in previous conditioning experiments; its appearance here
may be fortuitous.
Positive slide groups, however, had higher heart rate levels than Negative
groups, F(1,59)= 4.03, p< .05.
Clear, significant conditioning effects were obtained across a variety of
measures. For the startle probe reflex, magnitude was potentiated during
CS+, especially for the Immediate exposure group, and was larger than CS-
throughout extinction. Similarly, CS+ startle latencies were faster than CS-
in extinction. For visceral reactions, skin conductance response during CS+
was rapidly potentiated, especially for the Immediate exposure group, and
remained larger than that of CS- during both acquisition and extinction. In
contrast, cardiac accelerative conditioning proceeded more slowly, with CS+
response gradually exceeding CS- response late in acquisition; late-slide
deceleration was greater for CS+ than CS- in extinction. For affective valence
judgments, CS+ became more unpleasant (relative to CS-) after learning,
especially for subjects who viewed pleasant slides; conditioning produced no
systematic changes in arousal ratings overall. In general, startle and visceral
responses were not significantly modulated by the affective slide foreground,
but valence and arousal judgments were influenced by this manipulation.
The relationships of temperament characteristics, shock tolerance, and
rated shock aversiveness to stimulus reactivity and associative learning were
assessed. Temperamental influences are treated as temporal precursors to
experimental variables in the subsequent analyses.
Consistent with Gray's (1982) broad hypothesis that neurotic introverts
are more sensitive to aversive stimulation, the EASI fear and impulsivity
temperament dimensions accounted for significant variability in aversive
reactions; in contrast, dimensions of sociability, activity, and anger did not
predict measures of aversive reactivity or learning. Figure 14 illustrates the
distribution of fear X impulsivity scores for all subjects and cutpoints used in
subsequent group analyses. Fearfulness and impulsivity subscale scores were
independent in this sample of subjects (r= .09, ns) and the cutpoint-defined
groups did not differ on the other temperament dimensions.
Reactivity to aversive stimulation. Each person set his/her level of
shock to be "highly annoying but not painful" during initial exposure.
However, the physical intensity of shock at this subjective level--hereafter,
"shock tolerance"--varied widely and was significantly lower for subjects
whose temperamental disposition was fearful and inhibited.
Stepwise regression (Table 4) indicated that fearfulness and inhibition
predicted lower tolerance for shock, multiple r= -.45. Group analysis (see
Table 5) revealed a significant main effect for inhibition, F(1,60)= 5.45, p< .03,
a marginal effect for fearfulness, F(1,60)= 2.98, p< .09, and no interaction (p=
.84). In follow up tests, only highly fearful/inhibited and low fear/impulsive
subjects differed from one another in shock tolerance.
Temperamental fear and inhibition also predicted prelearning startle
magnitudes which were significantly larger (multiple r= .32). Group analysis
indicated a significant fearfulness effect, F(1,60)= 4.01, p< .05, a marginal effect
for impulsiveness, F(1,60)= 3.74, p< .06, and no interaction (p= .87). Again,
follow up tests showed that only fearful/inhibited subjects and nonfearful/
impulsive subgroups significantly differed from one another. (Mean startle
magnitude of fearful subjects continued to be higher overall throughout
conditioning, but this difference was not significant). Consistent with the
startle magnitude data, mean prelearning startle onset latencies of fearful
0 0 000 -0
0 1o (C 0
-- r -T
Figure 14. Distribution of fearfulness and impulsivity scores (EASI Temperament
Survey) for all subjects. Placement of dashed axes indicates that cutpoints of 14 on
each dimension which were used in determining subgroups for statistical analyses.
Stepwise multiple regressions of temperamental fear and inhibition
(low impulsivity) on measures of aversive reactivity.
Dependent Multiple R
Variable Fearful Inhibited Statistical Summary
FSS total score .48 F(1,62)=18.31, p< .001
Shock tolerance -.33 -.45 F(2,61)= 7.86, p< .001
Prelearning startle reflex:
magnitude .22 .32 F(2,61)= 3.46, p< .05
onset latency .25 F(1,60)= 3.88, p< .06
Means and standard deviations for measures of aversive reactivity
as a function of temperamental fear and impulsivity subgroups
(cutpoints = 14 for each subscale).
Low Fear High Fear
Dependent Inhibited Impulsive Inhibited Impulsive Mean
Variable n= 20 n= 10 n= 20 n= 14 n=64
FSS total score 87.2 b 105.3 ab 112.0 a 107.6 a 102.2
(40-200 scale) (16.4) (28.5) (19.0) (19.6) (22.3)
Shock tolerance 6.1 ab 8.3 a 4.2 b 6.8 ab 6.0
milliamperess) (4.1) (4.5) (3.1) (4.5) (4.1)
magnitude 309.9 ab 131.6 b 465.9 a 315.8 ab 332.1
(A-D units) (243.4) (83.3) (364.3) (453.2) (336.9)
onset latency 41.6 a 42.8 a 40.2 a 37.6 a 40.4
(msec) (8.8) (7.2) (7.5) (8.1) (8.0)
Means without common letters significantly differ (p< .05; using Tukey HSD,
MSerror [df= 4,60] and harmonic mean= 14.74).
ct a 44
6 (D i a .
-- -~ Ue e'
a P > E 0
1 o E 4
at .5 us
cu t -a
* w .1
(2) L" x
1 t z c0
li~ ~. .f
subjects were marginally faster than for nonfearful subjects (r= .25). Group
analysis yielded a somewhat weaker fearfulness effect, F(1,58)= 2.33, p= .13.
As was expected, temperamentally fearful subjects (independent of
impulsivity) reported more fear across a variety of situations (Fear Survey
Schedule; r= .48). This was confirmed by a significant difference in the group
analysis, F(1,60)= 6.69, p< .02. Follow up tests indicated that the nonfearful/
inhibited group reported less fear than other subgroups. Finally, fearful
subjects also tended to rate their shock as more aversive (r= .20, ns) which, to
some extent, explains the counterintuitive finding that less shock tolerance
was related to significantly higher shock aversion in the overall sample (r=
Figure 15 summarizes the relationships between temperament, low
shock tolerance, high shock aversion, and high situational fear. This pattern
of data supports the conclusion that fearful/inhibited subjects are more
generally reactive to aversive stimulation--in terms of reported situational
fears, behavioral tolerance of shock, and reflex excitability.
Sensitization. Temperament characteristics were not significantly
related to sensitization learning measures.
Associative learning. Temperamentally impulsive subjects exhibited
more skin conductance response potentiation and discrimination during
acquisition. Both effects were supported statistically by regression analysis,
Fs(1,62)= 4.36 and 6.67, ps< .05. Correlations between impulsivity and both
conductance potentiation and discrimination were modest but significant
(rs= .26 and .31). Fearfulness, entered stepwise in these same regressions,
did not significantly predict electrodermal learning. Temperament did not
interact with slide affective foreground in determining aversive learning.
Rated Shock Aversiveness
The role of the shock in mediating aversive learning was explored next,
using stepwise multiple regression. Rated shock aversiveness and
behavioral shock tolerance served as predictors and psychophysiological
measures of learning (sensitization, potentiation, and differentiation scores)
were the dependent measures. Interestingly, shock tolerance did not
significantly relate to any of these learning measures.
Sensitization. Rated shock aversiveness was not significantly related to
Associative learning. In contrast, shock aversion systematically affected
associative learning (see Figure 15). One group of subjects judged the shock
as highly aversive (rating of 7 on the I 7 scale; n= 27). Others, forming a
Moderate Aversion group (n= 20), rated shock aversiveness equal to 6. The
remaining subjects (n= 17), with aversion ratings from 3 to 5, constituted the
Low Aversion group. SAM ratings of the shock were each highly correlated
with shock aversion (pleasure, r= -.57; arousal, r= +.51). Therefore, reported
"aversion" can be decomposed into elements of displeasure and arousal.
Stepwise regression indicated that greater shock aversiveness predicted
significantly greater CS+ aversion, F(1,62)= 17.59, r= .47; startle potentiation,
F(1,62)= 13.98, r= .43; and skin conductance response potentiation, F(1,62)=
6.87, r= .32. Table 6 presents mean data from the group analyses. The group
mean difference in rated CS+ aversion was significant, F(2,61)= 7.84, p< .001.
Follow up tests found that High and Low Aversion groups significantly
differed; these extreme groups both differed marginally from the Moderate
Aversion group (ps< .07). However, the three groups did not significantly
differ in CS- aversion ratings.
Shock Aversion groups differed overall in startle potentiation, F(2,61)=
7.83, p< .001 (see Figure 16). In t-tests, High and Moderate Aversion groups
did not differ (As= +120 vs. +82 ADUs) and both showed significantly more
potentiation than Low Aversion subjects (A= -134 ADUs).5 The lack of
startle potentiation in the Low Aversion group was not due to more of these
subjects receiving delayed shock exposure (see Appendix D, Table 23).
Regression analysis indicated that shock aversiveness was not significantly
related to startle discrimination.
Subjects who rated shock as more aversive showed larger conductance
response increases to CS+ from adaptation to acquisition (r= .32; see Figure
17). The Aversion group difference in conductance response potentiation
was significant, F(2,61)= 3.59, p< .04. As with startle potentiation, electro-
dermal response potentiation for Moderate and High Aversion groups (As=
.04 and .08 pS) did not differ and changes in both groups were significantly
greater than the decreasing response in the Low Aversion group (A= -.07 pS).
Higher shock aversion was marginally related to better skin conductance
response discrimination in acquisition, Aversion X Stimulus F(2,61)= 2.54,
p< .09, but not during extinction.
High Aversion subjects showed greater heart rate acceleration to CS+ in
the final trial block of acquisition. However, the importance of this effect
was diminished by a nearly complete reversal of this effect in the preceding
trial block (see Appendix D, Table 24, for data and a brief discussion).
5 Although Figure 16 might suggest that the Low Aversion group had
higher CS+ startle responses than the other two groups before learning, the
group difference was not significant (p= .12). Thus, a direct comparison of the
amount of potentiation across groups is permissible.
Means and standard deviations for measures of associative learning
as a function of shock aversiveness (see Figure 15 for correlations).1
Shock Aversion Group
Dependent Low Moderate High Mean Refer to
Variable n=17 n=20 n=27 n=64 Figure:
CS+ aversiveness 4.5 b 5.5 ab 6.2 a 5.5
(1-7 rating scale; (1.5) (1.6) (1.0) (1.7)
CS- aversiveness 1.8 a 2.2 a 2.5 a 2.2
(1-7 rating scale; (1.1) (1.5) (1.7) (1.8)
Startle potentiation -134 b 82 a 120 a 41 16
(A-D units)2 (273) (98) (235) (274)
Skin conductance -.07 b .04 a .08 a .02 17
potentiation (pS)3 (.11) (.17) (.21) (.20)
Change in CS+ -1.7 a -1.2 a -3.2 b -2.2 18
pleasure4 (0.8) (1.5) (1.8) (1.7)
1. Means without common letters are significant different (t-tests; p< .05).
2. Change in CS+ response from adaptation block 2 to extinction block 1.
3. Change in CS+ response from adaptation block 2 to acquisition block 1.
4. Change in CS+ valence from pre- to post-learning, for subjects viewing
pleasant slides only.
(4 C- (
10v0 10 *'1' -'I
.o / o
S\ / oxs
I) \ I
0 C il
o 0P a
(Sd I -I+ jI Bol) apnimj) asuodsaN aluel:npuoD uijS
I I --
(sIui-i sSuWeIP) agueiq3 azualeA
The decrease in CS+ pleasure from pre- to postlearning (i.e., negative
change scores indicating valence potentiation) tended to be greater for High
Shock Aversion (A= -1.4 units) than Moderate or Low Aversion subjects
(both As= -0.4 units) in the overall sample, Aversion Group X Stimulus,
F(2,61)= 2.88, p< .07. However, as seen in Figure 18, high shock aversiveness
potentiated CS+ valence change only for subjects viewing pleasant slides,
Slide Valence X Shock Aversion F(2,58)= 3.18, p< .05. Post hoc tests showed
that, for unpleasant slide foregrounds, shock aversion did not systematically
modulate CS+ valence change. But among subjects viewing pleasant slides,
subjects highly averse to shock judged their CS+ as more unpleasant after
learning, F(2,29)= 5.73, p< .01. Figure 18 also illustrates, in comparison, that
CS- change was not modulated by rated shock aversion.
Shock aversiveness did not interact with affective slide content in
determining psychophysiological response during associative learning.
Frequency judgments of shock occurrence. Subjects underestimated the
number of shocks they received during conditioning (i.e., excluding the
shock exposure procedure). Although each subject experienced 8 shocks, the
mean estimate was 6.5 times. This discrepancy from actual exposure was
significant, F(1,61)= 29.22, p< .0000. Although no individual difference
factors were reliably associated with shock frequency estimates, subjects
reporting less aversion of shock tended to under-report frequency of shock
occurrence (r= .23, p< .10). Frequency judgments of shock occurrence were
not influenced by the affective slide foreground during conditioning.
Contingency recall. Eighty-six percent of subjects clearly recalled their
CS+ when probed by the following questions: (1) Did you recognize a rule in
this experiment?; (2) Were some stimuli followed by shock more often than
others?; and (3) If yes, which? An additional four subjects did not clearly
understand the probe questions but, when clarified, correctly identified their
CS+. (It is not clear whether the experimenter inadvertently cued the subject
in these cases; thus, it is not certain whether correct identification was a
recall or a recognition decision. If these four subjects are included in the
correct recall category, the recall rate becomes 92%.)
Four subjects could not report which slide had been shocked and a fifth
subject incorrectly believed that CS- had been shocked. Interestingly, four of
these five subjects were in the Low Shock Aversion group (the other was in
the High Aversion group). However, due to the rather small subsample,
these data were not subjected to statistical analysis. Contingency recall was
not related to affective slide content. Re-analysis of the data set, excluding
subjects who could not recall the contingency, did not change the findings.
The correlational pattern shown in Figure 15 supports several major
conclusions. First, temperamental fear and inhibition modulated sensitivity
to aversive stimulation, but did not generally determine aversive learning.
Second, the perception of shock aversiveness--and not behavioral tolerance
of shock intensity--predicted the degree of associative learning. Third, these
relationships were present only for measures of simple (but not differential)
association. Finally, whereas rated shock aversiveness directly influenced
physiological measures of learning, CS+ aversiveness showed similar, albeit
The present results indicate that startle reflex magnitude was sensitized
after exposure to shock and was also potentiated during a stimulus which
was paired with shock (relative to a stimulus not paired). In short, startle
magnitude is a reliable index of both nonassociative and associative learning
when an aversive reinforcer is used. Whereas startle onset latency and
visceral (heart rate and electrodermal) reactions also reflected associative
learning, these measures failed to index the sensitization learning process
which is presumed to occur when subjects are exposed to shock. Consistent
with the startle reflex data, ratings of affective valence (but not arousal) for
the conditioned stimuli reflected overall associative learning.
In addition, individual differences in aversive reactivity and learning
were identified. First, subjects who were both temperamentally fearful and
inhibited were more sensitive to aversive stimulation (i.e., they showed less
tolerance of shock, larger initial startle responses, and reported more anxiety
across a variety of situations) than other subjects. However, temperament
characteristics were not generally related to learning measures. Second,
perceived aversiveness of the shock US--and not physical intensity of shock--
modulated the degree of learning in this experiment.
In this experiment, a brief series of shocks was administered to each
subject to calibrate US intensity just below the subject's reported pain
threshold level. Shock exposure prompted an immediate, vigorous (40%)
increase in startle magnitude on the first postexposure trial, whereas this
effect was absent in control conditions. Thus, significant startle sensitization
developed very rapidly and could not be attributed to either dishabituation
or the simple passage of time. In addition, the increase in startle magnitude
from pre- to postexposure in the Immediate group--for whom duration of
this sensitization effect could be more completely measured--persisted
throughout adaptation (ca. 10 mins). The present findings are consistent
with the animal literature on shock sensitization of startle (Boulis & Davis,
1989; Davis, 1989; Hitchcock et al., 1989). To the extent that similar results
were obtained in recent experiments using human subjects (Greenwald et
al., 1990; Hamm, Stark, & Vaitl, 1990), it may be concluded here that shock
sensitization of the startle response is a robust phenomenon (i.e., large
magnitude, rapid development, relatively persistent, and independent of
species, dishabituation, and time).
In contrast, neither startle latency nor visceral activation (i.e., tonic
levels or slide responses) was reliably facilitated after shock exposure. Taken
together, the data suggest that shock exposure prompts an aversive
preparatory set--uniquely indexed by the startle reflex (Lang et al., 1990)-
which, lacking visceral arousal, may not require abrupt mobilization for
action. Consistent with this idea, freezing has been shown to emerge as the
dominant tactical response following shock exposure in animals (Fanselow,
1980; Fanselow & Bolles, 1979); furthermore, the extent of motoric inhibition
correlates highly positively with increased startle response in individual
subjects (Leaton & Borszcz, 1985). Shock exposure can also influence
behavior in future aversive situations. For instance, pre-exposing animals
to a brief series of relatively intense shocks facilitates "fear" of subsequent
aversive stimulation, as measured by suppression of operant rates (Pearl,
1963; Pearl, Walters, & Anderson, 1964) and avoidance in a conflict test
(Kurtz & Walters, 1962; Walters, 1963).
Current evidence (Lang et al., 1992) suggests that startle augmentation
reflects the motivational state of the subject, specifically the presence of an
aversive response disposition. The selective effect of shock exposure on
sensitization of startle magnitude (i.e., in the absence of visceral activation)
obtained here, as well as the varied effects of shock exposure on behavior
just described, encourages the conclusion that the startle response is not
tactically specific (i.e., associated with a particular behavioral act) but, rather,
a more general measure of aversive processing.
One question which cannot be answered from this experiment is
whether actual exposure to shock is necessary to obtain performance
changes, since exposure is confounded here with the knowledge that shock
will be presented. Thus, it remains possible that instructed exposure can
produce sensitization. Several researchers have, in fact, examined reactions
to threat of shock. Chattopadhyay, Cooke, Toone, and Lader (1980; Exp. 1)
conducted a study in which subjects were threatened or not threatened with
shock (within-subject, sessions separated by 1 week; order counterbalanced
across subjects). They found that electrodermal levels and anxiety ratings
increased during threat sessions, relative to no threat. Unfortunately, the
authors did not report whether responses increased relative to resting
baselines within the threat session, nor whether they conducted a between-
group analysis of the first session. Thus, it is not clear whether within-
session changes were significant, or whether between-session (i.e., threat
order) effects influenced the results. Farha and Sher (1989) also showed that
autonomic measures, somatic activity, and reports of anxiety all increase
after shock threat (relative to a "naive" baseline). Unlike the present study,
their experiment did not include a control group. Therefore, neither
dishabituation (i.e., after the experimenter re-entered the testing room) nor
the sheer passage of time can be ruled out as potential explanations of their
Unlike the studies above, shock exposure did not sensitize either cardiac
or electrodermal activity here. There are several possible explanations for
this discrepancy. First, it could be argued that the test of shock sensitization
of visceral activation was not powerful: Pre-exposure tonic levels and slide
responses were not available for the majority of subjects (i.e., Immediate
exposure group), yielding a weaker test in a smaller subset of subjects (i.e.,
Delayed exposure group). On the other hand, although conductance levels
and slide responses slightly increased (nonsignificantly) after shock exposure
in the Delayed group, levels and responses in the control condition (i.e.,
Immediate group) also increased. These data suggest that the increases in the
Delayed group were not specific to shock exposure and could be accounted
for by another factor (e.g., dishabituation). A second explanation is that
exposure to shock and threat of shock may differentially affect visceral
activity. During shock threat, the subject might anticipate aversive
stimulation and thus maintain visceral activity at higher levels. During
shock exposure, however, visceral activity may temporarily increase (before
it could be measured here?)1 and then rapidly habituate. The present heart
1 There was clearly an unconditioned reaction to shock presentation.
However, the issue here is whether shock exposure generated a fleeting
increase in visceral tonic level before the experimenter could continue the
procedure to measure such a change. In fact, there was a period of about 3 to
5 minutes after setting the final shock intensity when the experimenter had
to instruct the subject on the subsequent conditioning procedure, and initiate
rate tonic level data are partially consistent with this interpretation, since
levels of the Delayed group tended to increase after exposure, whereas
control levels significantly habituated. A third, equally plausible,
explanation is that the results of other studies may arise from uncontrolled
influences discussed above. In future work, one would ideally separate the
incremental effects of threat and exposure, using a paradigm similar to that
of Farha and Sher (1989), while including a control group and continuous
monitoring of visceral activation.
Finally, it is interesting that during the initial control period for shock
exposure (Delayed group) there was a mild, but nonsignificant, increase in
startle magnitude. This increase is not inconsistent with previous findings:
Some habituation experiments (Groves & Thompson, 1970; Szabo & Kolta,
1967) have shown that the startle reflex elicited by an intense acoustic probe
(in the absence of shock) may briefly increase before habituation occurs.
Thus, it is possible that reflex sensitization in the Immediate exposure group
reflects summation from two events--a strong, persistent increase from
shock exposure and a weaker, probe-related increase. One suggestion for
future research would be to repeat the present study design, adding a longer
preshock series of startle probes to induce certain response habituation. This
would insure that the amount of sensitization following shock exposure
could not be attributed to characteristics of the probe itself.
Magnitude of the eyeblink reflex to acoustic probes presented during the
CS+ slide was larger at the start of extinction, relative to adaptation. Startle
the computer program. Thus, if visceral changes did occur, they could have
subsided by the time the procedure continued.
reflexes evoked during CS+ were also significantly larger and faster in onset
than those during CS- or interslide intervals in extinction. Additional
performance changes--including skin conductance response potentiationn
and discrimination), differential cardiac acceleration, and affective ratings
change--provide converging support for the conclusion that aversive
learning occurred. The present findings thus replicate and extend results
from single cue (i.e., nondiscriminative) conditioning experiments with
animals (Berg & Davis, 1985; Brown, Kalish, & Farber, 1951; Cassella & Davis,
1986; Davis & Astrachan, 1978; Kurtz & Siegel, 1966; Leaton & Borszcz, 1985)
and recent differential learning experiments with humans (Greenwald et al.,
1990; Hamm et al., 1991; Hamm, Stark, & Vaitl, 1990).
Sensitization and Habituation in Associative Learning
The differential learning paradigm was used in this experiment because
it controls for nonassociative influences such as (within-session) habituation
and sensitization. Since the assessment of associative learning involves
comparing CS+ and CS- responses, it is important to rule out nonassociative
factors as alternative explanations.
Sensitization, habituation, and potentiation. CS+ startle magnitude was
potentiated early in extinction (relative to adaptation), whereas interslide
responses decreased over this same period. Recent experiments (Greenwald
et al., 1990; Hamm et al., 1991) also demonstrated this same pattern. If both
CS+ and interslide responses had increased, this would have suggested that a
nonassociative factor influenced potentiation. Thus, the data indicate that
the CS+ increase was an associative effect (since shock was never paired with
the absence of slides). By similar logic, sensitization cannot account for CS+
skin conductance response potentiation, since CS- response decreased from
adaptation to acquisition. Also, sensitization cannot account for heart rate
accelerative changes during training, since CS+ and CS- responses developed
slowly and in opposing directions.
For subjects whose exposure to shock preceded adaptation in this study
(as in past experiments), startle probe responses habituated before learning
and there was a clear replication of the startle potentiation effect. In contrast,
for the subgroup whose exposure to shock was delayed, startle potentiation
was not only absent but significantly decreased from adaptation to extinction.
These data suggest that habituation of startle responses in the Delayed group
prior to acquisition was not as advanced which, in turn, may have masked
startle potentiation. It is therefore possible that the absence of startle
potentiation in the Delayed exposure group may be a methodological artifact
resulting from insufficient habituation. A similar pattern was observed with
skin conductance response potentiation, i.e., initial pairing of the CS+ and
shock produced increased conductance response in the Immediate group, but
CS+ response in the Delayed group decreased--even though the two groups'
slide responses prior to shock pairing did not differ. An obvious remedy for
this artifact (if one were to repeat this procedure) would be to extend the
adaptation series, thus insuring response habituation in the Delayed group.
Habituation and discrimination. The psychophysiological measures of
learning employed here--especially startle magnitude and skin conductance
response--clearly habituated during the procedure, with the exception of
transient potentiation due to the shock pairing. Yet there was significant
discrimination between CS+ and CS- in extinction for startle (magnitude and
onset latency), skin conductance, and late cardiac decelerative responses and,
during acquisition, for heart rate acceleration and electrodermal responses.
These data suggest that the associative effect during and after learning was
specific to CS+. Furthermore, CS+ and CS- responses did not extinguish at
different rates, i.e., response decline was parallel for the CS+ and CS- slides,
to the extent that Stimulus X Trial Block interactions were not obtained for
these measures. Thus, discrimination learning is assessed here with
habituation as a constant, background factor.
Bradley, Cuthbert, and Lang (1990a) demonstrated that the startle reflex
habituates with repeated probe presentations within a session, a result which
presumably reflects decremental activation in the obligatory startle circuit.
They also found that the affect-startle effect was present throughout this
general habituation, attesting to persistent activity in the modulatory circuit.
In the present experiment and other recent human conditioning studies,
CS+/CS- response discrimination throughout extinction similarly reflects
the continued influence of the modulatory system--as induced by associative
The pattern of cardiac acceleration was different from startle reflex and
electrodermal measures during associative learning. In acquisition, heart
rate acceleration during the CS+ slide increased with repeated pairings
whereas acceleration to CS- decreased with repeated nonreinforcement. This
cardiac pattern suggests that aversive learning may influence the heart
rhythm in such a way as to reflect both excitatory and inhibitory influences.
However, slightly different results were reported by Hamm et al. (1991), who
found that cardiac response was more acceleratory overall, that CS- response
became more accelerative over trials with CS+, and that the size of
differential acceleration was slightly larger than in the present study. At first
glance, these differences suggest that Hamm et al. produced a more powerful
manipulation (e.g., a more intense shock); however, the size of the startle
reflex and skin conductance learning effects are virtually identical in the two
studies. Therefore, it does not seem warranted to conclude that the present
data reflect a less effective manipulation. A more likely explanation is that
cardiac acceleration is simply less reliable than other measures of slide
response (Lang, Greenwald, Bradley, Hamm, in press). Thus, specific aspects
of the cardiac pattern may be less likely to replicate.
Finally, there was evidence that (1) affective ratings habituatee" (regress
towards neutrality) with stimulus repetition and that (2) CS+ valence and
arousal judgments generalized to CS-. On the first point, recall that adjunct
slides varying in affective content were rated pre- and postlearning, but were
not subjected to conditioning. Mean valence judgments of the pleasant and
unpleasant control slides became more neutral, whereas initially neutral
slides showed no mean change. Similarly, arousal ratings of adjunct slides
became moderate (i.e., calm slides became more exciting and vice versa).
However, changes in CS+ valence ratings (particularly for Positive groups)
and CS+ arousal ratings (especially for Calm slide groups) were greater than
changes in ratings for control slides, suggesting that learning played a greater
role in affective change than regression to the mean.
On the second point, Figure 5 clearly shows that affective ratings of CS+
generalized to CS-. This was more the case for arousal than valence ratings:
Rated arousal, similar for CS+ and CS-, was the single performance measure
which did not reflect aversive learning. In contrast, both discrimination and
potentiation of valence ratings were obtained. Although it is not clear why
arousal ratings did not differentiate the stimuli here--Hamm et al. (1991) and
Greenwald et al. (1990) found significant ratings changes along both affective
dimensions--one hypothesis is that both earlier studies used neutral/calm
CS slides, which yielded the greatest arousal change of any affective slide
type. These neutral CS slides, because of the selection of extreme affective
stimuli here, were largely excluded.
Affective Associative Change
Results of this experiment strongly support the main hypothesis that
aversive learning modifies the principal motivational (valence, arousal)
parameters of emotional response. Startle reflex and electrodermal changes
(both potentiation and discrimination), cardiac acceleration, and affective
ratings change were all significantly modulated by the classical conditioning
procedure. The results do not, however, support the secondary hypothesis
that the affective quality of the slide foreground prior to learning constrains
the degree of associative change. More specifically, the physiological
measures used here do not supply information on the extent of affective
Affective modulation by specific content. As noted before, Hamm et al.
(1991) found that the amount of startle potentiation (and discrimination)
varied linearly with a priori valence of slide content: Less modulation was
observed for subjects who viewed unpleasant slide contents (mutilations,
threatening animals) than for subjects who viewed pleasant contents (nature
scenes, erotica). In addition, electrodermal discrimination varied linearly
with a priori slide arousal: Greater discrimination was observed for subjects
who viewed calm slides (household objects) than for subjects who viewed
the most arousing slides (mutilations, erotica).
Greenwald et al. (1990) attempted to replicate these associative effects,
by assigning subjects to view one of six slide content pairs distributed in the
valence X arousal space. The results of that experiment showed two changes
in the original startle modulation effect: (1) Discrimination, more so than
potentiation, tracked the amount of valence change; and (2) Rather than a
difference in magnitude, a smaller proportion of subjects who viewed
unpleasant slides showed startle discrimination than those viewing pleasant
slides. Electrodermal discrimination was also not a linear function of slide
arousal. Such differences across studies do not strongly support the affective
modulation hypothesis or, perhaps, suggest that other factors (e.g., specific
content) could mediate these effects. In contrast, changes in affective ratings
here supported findings of both earlier studies, namely that (1) CS+ valence
potentiation was greater for subjects viewing pleasant (than unpleasant)
slides and (2) CS+ arousal potentiation was greater for subjects viewing calm
(than arousing) slides. One problem with these differential ratings changes,
however, is that they may arise from floor and ceiling effects (i.e., initially
unpleasant slides cannot be rated much more unpleasant, and initially
arousing slides cannot be rated much more arousing, after aversive
Modulation along affective dimensions. In the present experiment,
stimulus content was deliberately obscured by choosing slides based on each
subject's extreme ratings. Inspection of Appendix A indicates that many
slide contents defined each of the four affective categories. Thus, it appears
that the effects obtained by Hamm et al. (1991) and Greenwald et al. (1990)
may have partly depended on specific slide contents.
The present approach to stimulus selection is more consistent with the
dimensional view of emotion, rather than relying on specific slide content to
influence the degree of associative change. Even with this more sensitive
approach, the results did not clearly support the affective change hypothesis.
On the other hand, subjects who viewed unpleasant/arousing slides tended
to show less differential learning than the other groups--not only in startle
magnitude and onset latency, but for electrodermal response, valence and
arousal ratings (Appendix D). Thus, a weaker version of the hypothesis
proposed by Hamm et al. (1991) could be advanced, in which a CS+ whose
affective character prior to learning most closely resembles that of shock (i.e.,
highly unpleasant and arousing) will undergo marginal associative change,
relative to any other effectively dissimilar stimulus. This notion-which is a
variant of the "delta rule" (Rumelhart et al., 1986)--has been applied to the
problem of affective change in conditioning elsewhere (e.g., Rescorla &
Wagner, 1972). In the present case, this hypothesis predicts a Stimulus X
Valence X Arousal effect (for discrimination scores) or a Valence X Arousal X
Trial Block effect (for potentiation scores). Despite the consistency of mean
differences across measures, however, no statistical test yielded a significant
Temporal Specificity and Preparatory Processes
Startle response. Several studies have demonstrated that associative
learning is temporally specific, since startle potentiation is largest nearest the
CS/US training interval (Davis, Schlesinger, & Sorenson, 1989; Siegel, 1967)
or when shock is expected to occur (Grillon, Ameli, Merikangas, Woods, &
Davis, in press). Grillon et al. signaled subjects to expect shock in one
("threat") period but not in another ("safe") period. In addition, subjects
were told that shock would only occur during the last 10 sec of the threat
period. Grillon et al. found that startle was significantly potentiated during
threat (relative to safe) periods, and was maximal during the last 10 sec of the
threat period. Interestingly, this effect was obtained before shock was ever
administered and improved over trials.
In the present study, startle magnitudes generally (across CS+ and CS-)
increased linearly over the CS/US interval in extinction. Thus, temporal
sensitivity of probe response was demonstrated here. However, the lack of
differentiation between CS+ and CS- along this temporal gradient indicates
that stimulus generalization also occurred. A similar, but non-significant,
increase in probe response was found by Hamm et al. (1991). One reason that
a significant finding may have been obtained here but not in the previous
experiment is that every presentation of a slide was probed here, whereas
Hamm et al. probed only 3 of every 4 CS occurrences. Thus, the difference
across studies may be partly due to statistical power.
Visceral reactions. Electrodermal and heart rate responses obtained in
this experiment offer a clearer example of learned temporal and stimulus
specificity. In acquisition, the SIR conductance response (preceding slide
offset) was greater to CS+ than CS-, presumably indexing greater orienting to
shock onset. This differentiation developed linearly over trials and persisted
into extinction. Although overall SIR magnitude decreased in the absence of
reinforcement, this anticipatory discrimination remained constant. Also,
late heart rate deceleration in extinction (D2 leg; occurring at the same time
during slide presentation as the SIR response) was significantly greater
during CS+ than CS-. Late cardiac deceleration presumably reflects
anticipation of events at stimulus offset in an S1-S2 paradigm (Bohlin &
Kjellberg, 1979). Thus, skin conductance and heart rate response provide
convergent evidence for temporally-specific stimulus differentiation.
Exactly what subjects are preparing to do during the training phase of
aversive classical conditioning has been widely debated. The midinterval
acceleratory and late-interval deceleratory components of the cardiac
waveform have been examined for their relationship to psychological
processes. Enhanced cardiac acceleration to an aversive cue has been
proposed to measure recruitment of a defense response (Fowles, 1980; Gray,
1982; Hodes et al., 1985; Obrist, 1981), whereas greater heart rate deceleration
just prior to US delivery has been thought to index an attentional set (Deane,
1969; Lacey & Lacey, 1970). To the extent that cardiac discrimination in
acquisition develops in the acceleratory component, this suggests that
temporary activation of a defense response may occur during learning. In
contrast, cardiac discrimination in the late decelerative leg during extinction
suggests that the information processing set shifts from a defensive
(stimulus rejection) posture to a more attentive (stimulus intake) mode with
cessation of aversive reinforcement.
Patrick and Berthot (1991) recently studied the effect of active vs. passive
preparation on subjects' psychophysiological responses in anticipation of
aversive noise. They found that only late cardiac deceleration varied by
preparatory set, with greater deceleration in the "active" group (who pressed
a microswitch to escape the noise) than either a yoked "passive" group or a
control group anticipating a less intense tone. In the active group, the extent
of cardiac deceleration was significantly correlated with the speed of escape,
which supports the idea that deceleration reflects sensory enhancement (cf.
Lacey, Kagan, Lacey, & Moss, 1963). These data are consistent with the results
of an experiment by Chase, Graham, and Graham (1968), who found that
cardiac acceleration was dominant only when subjects anticipated a vigorous
motor response (i.e., a leg lift), whereas cardiac deceleration was the clear
pattern when subjects anticipated a simple reaction time response.
In contrast to the active-passive group difference they found for cardiac
deceleration, Patrick and Berthot reported that startle magnitude during the
preparatory period was equal for the two groups, increasing closer to the
aversive US (just as in the present study). In contrast, startle responses of the
control group were smaller and decreased progressively closer to the weaker
US which that group received. Thus, the startle probe reflex did not index
the specific tactical response, which supports the hypothesis that it is a more
general strategic indicator of aversive motivation (Lang et al., 1990, 1992).
Brain-Behavior Relationships in Aversive Learning
The present analysis has focused on psychophysiological manifestations
of emotional learning and used these performance measures to infer
cognitive processing in aversive behavior. Yet this study also relates to an
expansive literature on the neural substrates of aversive learning. Thus,
although the present discussion is framed at a representational level of
analysis (Figure 1), the cognitive events described here are presumed to
parallel events at an implementational level (cf. Marr, 1982). Ultimately,
one goal of this work is to develop a psychological model of emotional
learning which is (at the very least) plausible or (at its very best) isomorphic
with neural events.
Davis (1989a) has outlined a neurobehavioral model of the aversive
system using both fear-potentiated startle and sensitization paradigms. The
amygdaloid complex, especially the central nucleus, appears to be the critical
anatomical node involved in aversive learning and memory (Kapp, Pascoe,
& Bixler, 1984; LeDoux, 1987; Sarter & Markowitsch, 1985). Lesions of the
central nucleus abolish both startle sensitization and potentiation (Hitchcock
& Davis, 1986; Hitchcock et al., 1989). Furthermore, drugs which have
anxiety-reducing properties in humans block learned startle potentiation
(e.g., Berg & Davis, 1984), whereas anxiogenic agents increase the learning
effect (e.g., Davis, Redmond, & Baraban, 1979).
Behavioral signs of aversion are presumed to occur when the amygdala
central nucleus is activated and projects to a wide array of brain areas. When
the shock US is presented--as in sensitization--it activates the amygdaloid
complex, which projects to several efferent nodes including the central grey
(leading to inhibition of behavior), the dorsal hypothalamus (cardiovascular
changes), the parabrachial nucleus (respiratory adjustments), and--in the case
of startle reflex excitability--the nucleus reticularis pontis caudalis (Davis,
1989a). Electrical stimulation of the amygdaloid central nucleus elicits these
unconditioned behaviors (Applegate, Kapp, Underwood, & McNall, 1983;
Davis et al., 1982; Gentile, Jarrell, Teich, McCabe, & Schneiderman, 1986;
Harper, Frysinger, Trelease, & Marks, 1984; Rosen & Davis, 1988, 1990).
The amygdala can also act as an associative node for the conjunction of
the CS+ and US. In classical conditioning, the result of pairing a visual CS+
with the shock US is that the amygdala can then be independently activated
by the CS+, via relatively direct subcortical sensory pathways, leading to
potentiated startle (Tischler & Davis, 1983). In short, the flow of events
within this neuroanatomical circuitry is remarkably consistent with the
Konorskian cognitive mechanisms of learning adopted here.
Aversive Reactivity and Sensitization
Each subject was pre-exposed to a series of shocks, the criterion level
being judged as very intense but below his or her nocioceptive threshold.
Subject temperament characteristics modulated tolerance of shock during
this procedure. Subjects who were relatively more fearful and inhibited
were less able to tolerate increments in shock intensity than other subgroups
(especially nonfearful, impulsive) defined by the fear X impulsivity
temperamental space. These trait fearful, inhibited subjects also reacted
more strongly to the startle probe before learning and reported more specific
fears across a variety of situations than trait nonfearful, impulsive subjects.
Thus, the present study identified subjects who are generally more sensitive
to aversive stimulation.
In the Chattopadhyay et al. (1980; Exp. 1) study noted earlier, auditory
evoked responses to quasi-startle probes (10 msec rise time) were measured
during threat of shock and its absence. Not unlike the larger probe startle
responses observed here for fearful/inhibited subjects, they found that the
P300 component of the auditory evoked cortical response was greater in high
trait anxious than low trait anxious subjects, independent of shock threat.
There may be some communality among those findings and the present
data, since other investigators (Roth, Blowers, Doyle, & Kopell, 1982; Roth,
Dorato, & Kopell, 1984) have found that startle blink amplitude and P300
amplitude covary reasonably well across a range of probe intensities. More
work is needed to determine whether temperamental fear potentiates both
the reflex eyeblink and evoked cortical potential to acoustic startle probes.
This would constitute a test of the hypothesis (Lang et al., 1992) that
emotional response sets are integrated efferent patterns extending from
"cortex to reflex."
It is possible that nonfearful/impulsive subjects, who were less affected
by the aversive stimuli (i.e., shock, startle probe) presented here, may be part
of a "disinhibited" subpopulation that is generally more tolerant of novel or
intense stimulation (Barratt, 1983; Zuckerman, Buchsbaum, & Murphy,
1980). In fact, Bradley (1992) found that the EASI Impulsivity dimension and
Disinhibition subscale of "sensation seeking" (Zuckerman, 1979) both loaded
on a global "behavioral constraint" factor (cf. Tellegen, 1985) in a factor
analysis of personality questionnaire data. Thus, it is worth noting that the
present data agree with a recent report (Pivik, Stelmack, & Bylsma, 1988) that
subjects scoring high on the Zuckerman Disinhibition subscale exhibited
reduced motoneuron excitability (i.e., electrically elicited Hoffman reflexes).
Naturally, one should be cautious in drawing conclusions from such diverse
reflex measures, but these findings suggest that the inhibition/ impulsivity
construct may relate to basic differences in the organization of efferent
output. Future research can determine whether these temperament group
differences are reliable, whether they apply to both aversive and appetitive
USs, and which mechanisms control these putative behavioral differences.
Overall, 77% percent of subjects in this study showed at least some
increase in startle magnitude following shock exposure. However, neither
individual differences in subjects' temperament nor shock tolerance were
significantly associated with the degree of startle sensitization. Since
Greenwald et al. (1990) also found that most subjects (i.e., 95%) showed shock
sensitization of startle magnitude, startle sensitization in humans may be a
fairly robust effect. One reason that slightly more subjects in the earlier study
showed startle sensitization may be that many more preshock startle probes
were administered (as part of a preceding experiment), presumably leading
to greater habituation and a lower prelearning baseline for assessing startle
sensitization. As argued above, this factor would enhance the effect across
Temperament and tolerance of electric shock. Individual differences in
temperament and shock behavioral tolerance did not generally influence
associative learning. Because the fearful/inhibited group attenuated their
exposure to shock initially, it is difficult to determine whether these same
subjects should learn better with an aversive reinforcer (as Eysenckian
theory might predict). More extensive pre-exposure to the US is known to
retard classical conditioning (Randich & LoLordo, 1979). Therefore, it would
be expected that nonfearful/impulsive subjects--whose greater tolerance for
shock resulted in more pre-exposure in this study--might actually show
poorer associative learning. Yet impulsive subjects performed better in skin
conductance (i.e., greater potentiation and discrimination). Thus, neither