International Journal of Psychophysiology 61 (2006) 5 – 18

www.elsevier.com/locate/ijpsycho

Basic emotions are associated with distinct patterns of

cardiorespiratory activity
Pierre Rainville a,*, Antoine Bechara b, Nasir Naqvi b, Antonio R. Damasio c
a
Département de stomatologie et Centre de recherche en science neurologique, Université de Montréal, CP 6128,
Succ. Centre-ville, Montréal Qc, H3C 3J7, Canada
b
Division of Behavioral Neurology and Cognitive Neuroscience, Department of Neurology, University of Iowa College of Medicine, Iowa City IA 52242, USA
c
Department of Psychology, University of Southern California, Los Angeles, CA 90089-2520, USA

Received 18 October 2005; received in revised form 20 October 2005; accepted 27 October 2005
Available online 24 January 2006

Abstract

The existence of specific somatic states associated with different emotions remains controversial. In this study, we investigated the profile of
cardiorespiratory activity during the experience of fear, anger, sadness and happiness. ECG and respiratory activity was recorded in 43 healthy
volunteers during the recall and experiential reliving of one or two potent emotional autobiographical episodes and a neutral episode. Univariate
statistics indicated that the four emotions differed from each other and from the neutral control condition on several linear and spectral indices of
cardiorespiratory activity. Dependent variables were further reduced to five physiologically meaningful factors using an exploratory principal
component analysis (PCA). Multivariate analyses of variance and effect size estimates calculated on those factors confirmed the differences
between the four emotion conditions. A stepwise discriminant analyses predicting emotions using the PCA factors led to a classification rate of
65.3% for the four emotions (chance = 25%; p = 0.001) and of 72.0 – 83.3% for pair-wise discrimination (chance = 50%; p’s < 0.05). These findings
may be considered preliminary in view of the small sample on which the multivariate approach has been applied. However, this study emphasizes
the need to better characterize the multidimensional factors involved in cardio-respiratory regulation during emotion. These results are consistent
with the notion that distinct patterns of peripheral physiological activity are associated with different emotions.
D 2005 Elsevier B.V. All rights reserved.

Keywords: Basic emotion; Autonomic nervous system; Heart rate variability; Respiratory sinus arrhythmia; Respiration

1. Introduction by this debate is the role of somatic states, constituted by

There is a long history of debates surrounding the role of
somatic states in emotions. William James proposed in 1894
(James, 1994) that the felt afferent signals from the viscera are
essential for the unique experience associated with distinct
emotions. In contrast, Cannon (1987) proposed that the slow,
diffuse, and unspecific, visceral activity could not be the source
for the qualities of felt emotions (also see Cannon, 1931). The
contribution of cognitive factors was later emphasized in
theories in which peripheral activity only contributed to the felt
level of arousal and not to the quality of the emotion experienced
(Schacter and Singer, 1962). One of the central questions raised
* Corresponding author.
E-mail address: pierre.rainville@umontreal.ca (P. Rainville).
0167-8760/$ - see front matter D 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.ijpsycho.2005.10.024
endocrine, visceral, musculoskeletal, and behavioral response,
in the experience of emotions. Indeed, while James’ original
theoretical proposal requires that emotions be associated with
distinct patterns of somato-visceral activity, the cognitive
perspective implies that the peripheral activation associated
with emotions can be reduced to a single dimension. Here, we
revisited James’ hypothesis using cardiorespiratory measures.
Previous studies have provided some evidence for James’
view. Specific patterns of autonomic activity have been reported
across individuals during the voluntary production of emotional
facial expressions (Ekman et al., 1983; see also Levenson et al.,
1990), and in response to visual (Collet et al., 1997), olfactory
stimuli (Vernet-Maury et al., 1999), and film clips (Christie and
Friedman, 2004). However, based on a meta-analysis of the
studies investigating the physiological responses observed
during basic emotions evoked using a variety of methods,
6 P. Rainville et al. / International Journal of Psychophysiology 61 (2006) 5 – 18
Cacioppo and his colleagues concluded that the literature
provided only equivocal evidence for the existence of distinct
patterns of peripheral activity associated with basic emotions
(Cacioppo et al., 2000). Although this conclusion may be
regarded as particularly conservative in view of the evidence
already available, it nonetheless suggested the need for
additional investigations of somatic states associated with
emotions.
There are at least two reasons for the limited success in
uncovering distinctive physiological patterns associated with
basic emotions: the inadequate experimental elicitation of the
target emotions and the incomplete physiological characteriza-
tion of the ensuing somatic states. The first limitation may be
surmounted by acquiring and analyzing physiological data
when subjects actually report feeling the emotions, rather than
simply relying on stimulus or task-related criterion. Here, we
use data obtained when subjects reported feeling a target
emotion in a paradigm involving the autobiographical recall
and on-line ‘‘reliving’’ of target emotions. This method has
been effective in demonstrating some level of discrimination
between basic emotions based on autonomic measurements in
previous studies (e.g. Ekman et al., 1983). Improvements in the
characterization of emotion-related physiological activity may
be achieved by measuring additional physiological signals or
by refining the analysis of commonly used physiological
measures. In the present study, we took this latter approach and
applied modern chronometric measurement techniques to
assess cardiorespiratory activity during emotional states.
1.1. Multi-dimensionality in cardiorespiratory regulation
One of the conclusions of the meta-analysis of Cacioppo
et al. (2000) was that a better characterization of sympathetic
and parasympathetic responses may provide some discrimina-
tive power to distinguish patterns of visceral activity associated
with basic emotions. A number of non-invasive computational
techniques have been developed over the recent decades to
characterize the autonomic processes involved in the chrono-
metric regulation of cardiac function (e.g. Task force of the
European Society of Cardiology and the North American
Society of Pacing and Electrophysiology, 1996; Berntson et al.,
1993b; Malliani et al., 1991; Pagani et al., 1997). These
techniques provide estimates of the relative contribution of the
parasympathetic and sympathetic branches of the autonomic
nervous system by assessing changes in heart rate variability
(HRV). A continuous tachogram can be derived from the ECG
by calculating the successive intervals between R-waves. From
the resulting RR-tachogram, one can extract several indices of
HRV based on linear calculations (time-domain measures such
as the standard-deviation of RR intervals for a given epoch and
the mean absolute difference between successive RR intervals).
The RR-tachogram can also be analysed using Fast-Fourier
Transform (FFT) methods to derive indices of HRV in specific
frequency bands (frequency-domain measures such as the
spectral power in the high frequency range).
Parasympathetic regulation is mediated by the vagus nerve
and by the release of acetylcholine (Ach) at the neuro-muscular
junction. This system is relatively rapid because the Ach is
quickly degraded by acetylcholinesterase in the extracellular
compartment (Talman and Kelkar, 1993). The dynamic
regulation of vagal tone can therefore be indexed by a number
of chronometric variables sensitive to high frequency changes
in the RR interval. By contrast, sympathetic regulation relies
upon the release and relatively slow re-uptake of adrenaline.
The difference in the dynamics of parasympathetic and
sympathetic systems implies that the rapid changes in heart
rate are mediated by parasympathetic activity while slow
changes can result from either sympathetic or parasympathetic
activity. Rapid changes in heart rate are further coupled, in part,
with changes in respiration because the increase in intra-
abdominal pressure during inhaling activates the baroreceptor
reflex and produces rapid increases in heart rate mediated by
vagal mechanisms. The resulting respiratory sinus arrhythmia
(RSA), typically observed in normal cardiac activity, can be
characterized by the amplitude of changes in heart rate within
each respiratory cycle, and generally contributes to the high
frequency components of HRV. Based on these observations,
several dependent measures, more or less sensitive and specific
to various aspects of cardio-respiratory control, can be
extracted from continuous recordings of cardiac and respiratory
signals. These measurements reflect three component of auto-
nomic regulation: (1) a sympathetic component; (2) a
parasympathetic component coupled with respiration (RSA)
and reactive to baroreceptor reflex activity; and (3) a para-
sympathetic component independent from respiration and
possibly reflecting top-down neural influences on vagal output.
1.2. Hypotheses
The experience of several basic emotions has been
consistently associated with changes in heart rate (Cacioppo
et al., 2000). By contrast, the contribution of heart rate
variability to emotions has been documented mostly in fear.
For example, panic attacks have been associated with higher
heart rate levels coupled with robust decreases in HRV
(Yeragani et al., 1991; Friedman and Thayer, 1998; Rao and
Yeragani, 2001; George et al., 1989; Yeragani et al., 1994b).
However, it is not clear whether those reductions in HRV are
simply secondary to changes in respiration (e.g. hyperventila-
tion) and thus reflect changes mediated by the baroreceptor
reflex as the breathing pattern changes. Few studies have
examined specifically the effect of other emotions on HRV.
Exceptions include a study suggesting that the increase in heart
rate during anger reflects mainly a sympathetic activation
characterized by a relative increase in the low frequency range
of HRV, while positive emotions may be associated with a shift
towards the high frequency range (McCraty et al., 1995).
In the present study, we tested the contribution of
cardiorespiratory activity to the production of distinct somatic
states associated with the feeling of different emotions in
normal human volunteers. We hypothesized that (1) cardiore-
spiratory activity associated with emotions can be characterized
along the dimensions of sympathetic and parasympathetic
influences, and that the latter factor can be further sub-divided
 P. Rainville et al. / International Journal of Psychophysiology 61 (2006) 5 – 18
into respiration-coupled and respiration-uncoupled compo-
nents; and that (2) specific basic emotions can be predicted
from the observed pattern of cardiorespiratory activity. In order
to facilitate the observation of robust and sustained emotions,
we relied on a self-induction paradigm in which subjects
vividly recalled an autobiographical episode that elicited a
strong emotion and we analyzed cardiac and respiratory
activity recorded immediately after the subjects indicated
starting to feel the emotion. This paradigm has demonstrated
its effectiveness in producing robust and partly segregate
patterns of brain activity associated with fear, anger, sadness
and happiness (Damasio et al., 2000). We extracted several
measures of heart rate, respiratory period and relative
respiratory amplitude, as well as indices of variability in both
heart rate and respiration. Since many of those measures were
expected to be partly redundant, we applied an exploratory
principal component analysis (PCA) to reduce the number of
dependent variables and extract possible orthogonal factors.
This analysis should be considered exploratory in view of the
small sample size on which it is applied. The independent
factors extracted form the PCA were used in discriminant
analyses to test the possibility that the emotion experienced can
be predicted by changes along those distinct dimensions of
cardiorespiratory activity.
2. Methods
2.1. Subjects
Forty-three healthy volunteers were recruited from the
University of Iowa Hospitals and Clinics and the University
of Iowa students and staff. Subjects were first contacted by
phone and were scheduled to participate in the experiment if
they could vividly remember one or two autobiographical
episodes where they experienced a strong emotion of fear,
anger, sadness, or happiness. All procedures were approved by
the Internal Review Board of the Human Subjects Office of the
University of Iowa Hospitals and Clinics.
2.2. Experimental procedure
Upon arrival to the psychophysiological laboratory, subjects
gave informed consent, and the physiological monitoring
equipment was installed. Subjects described the personal
emotional episode they indicated upon recruitment in as much
detail as possible. Then, the physiological recording started and
the subject was instructed to close his/her eyes and relive the
target emotion as vividly as possible while remembering the
autobiographical episode, and to try to maximize the target
emotion until the experimenter asked them to stop. The
experimenter provided verbal cues to guide mental evocation
based on the subject’s prior description of the events and
emphasized the events that precipitated the target emotion.
Subjects indicated by a slight movement of the hand when they
started to feel the emotion or when they were in the imagined
neutral situation (e.g. getting ready that morning, coming to the
laboratory; the last time they did the laundry. . .). The
7
experimenter then stopped further cueing and the physiological
recording was time-stamped for off-line analysis. Subjects were
told that they could stop experiencing the emotion after 90 s
and the physiological recording was time-stamped again and
the data saved. Subjects provided subjective ratings and
immediately performed a second trial in the same emotion
condition. Different experimental conditions were separated by
at least five minutes during which the subjects was invited to
relax. Each subjects completed one (n = 24) or two (n = 19)
emotion conditions and the control neutral condition. The study
was designed to include more than one emotion per subject, if
subjects could vividly recall other recent episodes in which
they had experienced another strong emotion. Ideally, we had
planned to gather sufficient data to perform within-subject pair-
wise contrasts for all pairs of emotions (possibly with different
subset of subjects contributing to the different contrasts).
However, since only some subjects could vividly recall two
strong emotional episodes, this was not possible and all
emotion data was treated as between-subject observation.
Physiological data were irrecoverable due to equipment failure
in two trials and cardiac physiological data were discarded due
to abnormal ECG activity in one subject (sustained bradycar-
dia). The final data set consisted in 15 subjects in the anger
condition, 15 in fear, 15 in happiness, and 17 in sadness.
2.3. Self-report
After each trial, subjects provided Likert-scale ratings (0 –4)
of the emotion(s) felt during the trial. The strongest emotion
reported was always the target emotion. Non-target emotional
feelings were occasionally reported mainly in the neutral
condition (e.g. happy or anxious) but these secondary emotions
were always of a lesser magnitude than the target emotion.
2.4. Physiological measures
Physiological activity was monitored continuously using
Grass amplifiers and data was digitized (200 Hz), recorded and
processed using the MP100 system and AcqKnowledge
software (Biopac Systems Inc.). Respiration was indexed by
relative changes in thoracic-abdominal expansion monitored
using a tension transducer attached to a strain-gage belt placed
over the lower floating rib and adjusted individually to produce
the maximal deflection during normal breathing in the pre-
experimental set up phase. ECG was recorded using a standard
3 leads montage (Einthoven lead 2 configuration). In addition
to those target physiological channels, we recorded palmar skin
conductance (SC) of both hands. Facial EMG activity was
recorded over the left zygomatic, masseter, and currogator
muscles in a subset of subjects. These measures are not des-
cribed further in this report, which focuses on cardiorespiratory
variables.
2.5. Physiological signal processing
All physiological recordings were continuously monitored
during the experiment and visually inspected off-line. Record-
8 P. Rainville et al. / International Journal of Psychophysiology 61 (2006) 5 – 18

ing artifacts were identified and corrected by interpolation or
else discarded. Instantaneous R– R intervals were calculated
from the ECG using a template-matching autocorrelation
function (template centered on the R-wave) and a peak
detection algorithm to obtain a continuous RR tachogram.
Careful examination of the ECG and the tachogram insured
that the automatic R-wave detection procedure had been
performed correctly. The respiratory signal was smoothed
using the mean of a 1-s moving window and transformed to
obtain instantaneous (cycle-by-cycle) measurements of period
and amplitude.
Dependent variables reported in Table 1 were extracted
from 90-s epochs of recording starting at the point when
subjects indicated feeling the target emotion. Multiple linear
(time-domain) and spectral (frequency-domain) indices of
HRV were computed following the recommendation of the
Task force of the European Society of Cardiology and the
North American Society of Pacing and Electrophysiology
(1996) and Malliani et al. (1991). Various RR interval
measurements were further extracted for each respiratory
cycle to document the variability between cycles (i.e. slow
changes) and within cycle (i.e. fast changes). RR variability
within respiratory cycle reflects the RSA and was calculated
using the difference between the maximum and minimum RR
interval within each respiratory cycle based on the peak-to-
valley method of Grossman et al. (1990). In addition to the
variables listed, minimum and maximum RR intervals were
systematically extracted from each recording as a quality
control measure to insure that no extra beat had been included
and that no beat had been missed (Berntson and Stowell,
1998). Exact Fast Fourier Transform (FFT) analysis was
performed on 214 samples (81.92 s) of the continuous
tachogram and the smoothed respiratory data. Because of
the relatively short duration of the recording, valid estimates
of HRV could be obtained only in the high frequency range
(0.15 –0.40 Hz), sensitive to changes in parasympathetic
activity.
2.6. Data analysis
Physiological measures were averaged within subject and
condition and univariate statistics were computed on each
dependent variable as an exploratory measure to examine the
effect of the emotions compared to the neutral condition
(within-subject T-tests). A uniform log transformation of the
ratio of each emotion over the neutral condition was then
applied to account for individual differences in baseline
physiological activity in the control condition. This transfor-
mation was chosen because it effectively reduced the skewness
in the distribution of most dependent variables. The distribu-
tion of each variable was further examined after the transfor-
mation to identify possible remaining outliers (mean T 3SD).
Based on this criterion, 6% of all measurements were excluded
from the group analysis. The transformed scores were then
compared across emotions using between-subjects ANOVAs
and effect sizes were calculated between pairs of emotions to
control for type 2 error (Cohen, 1988).
Because this study included several measurements sensitive
at least in part to common physiological mechanisms, an
exploratory principal component analysis (PCA) was performed
to reduce the number of dependent variables and account for the
inherent structure of cardiorespiratory activity (Rotation Meth-

Table 1
Dependent variablesa
Domain
Respiration (linear)
RR changes per respiratory
cycle
RR interval (linear)
RR interval (spectral)
Respiration (spectral)
a
Linear and spectral indices of RR variability are taken from Malliani et al. (1991) and from the Task force of the European Society of Cardiology and the North
American Society of Pacing and Electrophysiology (1996).
Variable Description
RespPeriod-mean Mean respiratory period
RespPeriod-median Median respiratory period
RespPeriod-sd Standard deviation of the respiratory period
RespAmp-mean Mean relative respiratory amplitude
RespAmp-median Median of the relative respiratory amplitude
RespAmp-sd Standard deviation of the relative respiratory amplitude
RRmean/RespCycle-mean Mean of the averaged RR within each respiratory cycle
RRmean/RespCycle-sd Standard deviation of the averaged RR across respiratory cycles (slow RR changes between
respiratory cycles)
RRsd/RespCycle-mean Mean of the standard deviation of RR calculated within each respiratory cycle (fast RR changes
within respiratory cycles; respiratory sinus arrhythmia)
RRsd/RespCycle-sd Standard deviation of the standard deviation of RR calculated within each respiratory cycle (fast RR
changes within respiratory cycles; respiratory sinus arrhythmia)
RRmax-min/RespCycle Mean of the maximum excursion of the RR interval (RR max RR min) within each respiratory
cycle (fast RR changes within respiratory cycles; respiratory sinus arrhythmia)
RR-mean Mean RR interval
RR-sd Standard deviation of the RR interval (overall variation in RR)
DeltaRR-mean Mean difference between successive RR intervals (RRi+1 RRi ; fast beat-by-beat RR changes)
DeltaRR-sd Standard deviation of the difference between successive RR intervals (fast RR changes)
DeltaRR-rms Square root of the mean of the sum of square of differences between successive R – R intervals
(fast RR changes)
RR-HighFreq High frequency spectral power (FFT) of the continuous tachogram (0.15 – 0.40 Hz; fast RR changes)
Resp-HighFreq High frequency power (FFT) of the respiratory signal (0.15 – 0.40 Hz)
 P. Rainville et al. / International Journal of Psychophysiology 61 (2006) 5 – 18 9

Table 2 median, RespAmp-median, RRmax-min/RespCycle, and Del-

P-values for the contrasts between each Emotion conditions and the Neutral taRR-mean, respectively). From the remaining variables, indices
control conditiona
of RR variability entered the PCA if they showed some effect of
Dependent Emotion vs. Neutralb
the emotion compared to the neutral condition, based on paired
Variable Anger Fear Happiness Sadness t-tests ( p = 0.05, uncorrected for multiple comparisons; see
RespPeriod-mean Ns ,0.0009 ,0.0007 Ns Table 2). Other variables may be considered emotion-insensitive
RespPeriod-median ,0.02 ,0.0006 ,0.001 Ns in the context of the present experiment. It should be noted that
RespPeriod-sd Ns Ns Ns j0.01
this permissive selection of putative emotion-sensitive variables
RespAmp-sd j0.003 j0.04 Ns Ns
RRmean/RespCycle-mean ,0.0006 ,0.00001 ,0.0001 ,0.0005 for the PCA has no incidence a priori on the hypothesis that
RRsd/RespCycle-mean Ns ,0.0006 ,0.002 Ns cardiorespiratory activity is organized along several dimensions
RRmax-min/RespCycle Ns ,0.0005 ,0.004 Ns expected to be captured by the PCA and on the hypothesis that
RR-mean ,0.0001 ,0.00004 ,0.0004 ,0.03 basic emotions can be discriminated based on the extracted
DeltaRR-mean Ns ,0.01 ,0.01 ,0.01
factors. Two additional variables describing respiratory activity
DeltaRR-sd Ns ,0.04 ,0.04 ,0.04
DeltaRR-rms Ns ,0.02 ,0.02 ,0.015 (RespAmp-median and Resp-HighF) did not meet this criterion
RR-HighFreq Ns ,0.006 ,0.02 Ns but were included in the PCA to improve the characterization of
a
Dependent variables are described in Table 1. Only variables from Table 1 cardiorespiratory space. The minimum Eigenvalue (0.5) was
for which at least one contrast is significant (paired T-test, p-uncorrected < .05) selected based on the inspection of the factor loading to insure
are included in Table 2. that we included at least one factor that captured respiratory
b
j: Emotion > Neutral; ,: Emotion < Neutral; Ns: not significant ( p > 0.05). variability.
A multivariate analysis of variance performed on the Factors
od, Varimax with Kaiser Normalization; SPSS v. 11.0). The extracted from the exploratory PCA tested for the main effect of
correlation matrix was first examined and confirmed the the emotion (MANOVA, using Pillai’s Trace). Post hoc
expected redundancy in measurements. Some of the variables univariate tests and contrasts were further computed on the
listed in Table 1 were excluded from the PCA (i.e. RespPeriod- factors and effect sizes were calculated between pairs of
mean, RespAmp-mean, RRsd/RespCycle-mean, and DeltaRR- emotions to control for type 2 error (Cohen, 1988). The
rms) because more then 90% of their variance (R 2  0.90) was possibility to categorize the four emotions based on the factors
accounted for by another dependent variable (i.e. RespPeriod- extracted from the PCA was further tested statistically in a series

Table 3
Univariate comparisons of the effect of the four emotions
Dependent variable F (df a) P P-valuesb for pair-wise comparisons (effect size)
A vs. F A vs. H A vs. S F vs. H F vs. S H vs. S
RespPeriod-mean 5.01 0.004 0.02 Ns Ns LSD 0.003 Ns
(3, 56) (1.07) (0.60) ( 0.02) ( 0.75) ( 0.93) ( 0.42)
RespPeriod-median 3.63 0.018 Ns Ns Ns LSD 0.01 Ns
(3, 57) (0.65) ( 0.10) ( 0.27) ( 0.84) ( 0.89) ( 0.24)
RespPeriod-sd 4.10 0.01 Ns LSD Ns Ns 0.04 0.02
(3, 58) (0.73) (0.70) ( 0.27) (0.10) ( 0.96) ( 0.92)
RespAmp-sd 1.66 0.2 Ns LSD Ns Ns Ns Ns
(3, 57) (0.75) (0.99) (0.52) (0.40) ( 0.13) ( 0.43)
RRmean/RespCycle-mean 7.30 0.0003 0.02 Ns Ns 0.001 0.0006 Ns
(3, 54) (1.04) ( 0.43) ( 0.38) ( 1.48) ( 1.35) (0.00)
RRsd/RespCycle-mean 6.33 0.0009 0.002 LSD Ns Ns 0.004 Ns
(3, 55) (1.25) (0.79) (0.16) ( 0.65) ( 1.26) ( 0.73)
RRmax-min/RespCycle 7.16 0.0004 0.001 LSD Ns Ns 0.001 Ns
(3, 54) (1.37) (0.82) (0.09) ( 0.70) ( 1.34) ( 0.76)
RR-mean 5.38 0.003 0.03 Ns Ns 0.009 0.004 Ns
(3, 53) (1.00) ( 0.27) ( 0.25) ( 1.19) ( 1.04) ( 0.07)
DeltaRR-mean 1.33 0.3 Ns Ns Ns Ns Ns Ns
(3, 54) (0.50) (0.14) (0.05) ( 0.48) ( 0.60) ( 0.16)
DeltaRR-sd 1.96 0.13 LSD Ns Ns Ns Ns Ns
(3, 54) (0.65) (0.40) (0.24) ( 0.41) ( 0.61) ( 0,32)
DeltaRR-rms 1.69 0.18 LSD Ns Ns Ns Ns Ns
(3, 54) (0.60) (0.31) (0.19) ( 0.43) ( 0.59) ( 0.23)
RR-HighFreq 5.34 0.003 0.002 LSD Ns Ns LSD Ns
(3, 50) (1.40) (1.15) (0.58) ( 0.45) ( 0.83) ( 0.46)
a
Note that df error varies because of missing data for some dependent variables.
b
A: Anger, F: Fear, H: Happiness, S: Sadness; p-values are given for significant contrasts based on Tukey HSD; LSD: p-uncorrected < .05 using the more lenient
least-square difference method; Ns: not significant; effect sizes are considered moderate to large (in bold) for standardized differences >0.50 and >0.80, respectively
(standardized effect size calculated using Hedges’ bias correction; Cohen, 1988).
10 P. Rainville et al. / International Journal of Psychophysiology 61 (2006) 5 – 18

of stepwise discriminant analyses (SPSS v 11.0; see Friedman, 3. Results

1989 and Duda et al., 2001, for the method and applications of
discriminant analyses). This approach allows characterizing
small samples in high-dimensional space. Correct classification
rates were obtained using a standard and a cross-validation
procedure in which each case is classified successively by the
functions (decision tree) derived from all other cases (leave-one-
out classification or U-method). This method leads to more
conservative estimates of the correct classification rate. A
comparable performance between the standard and the cross-
validation procedure generally attests of the stability of the
classification model. Correct classification rates were evaluated
using Chi-square tests.
3.1. Self-reports of emotions
Subjects reported experiencing the target emotion more
intensely than any other emotion in all trials. The mean (sd)
Likert-scale rating of the target emotion was 2.29 (0.64) for fear,
3.03 (0.86) for anger, 3.25 (0.70) for sadness, and 3.00 (0.67)
for happiness. The target emotion was always felt more intensely
than any emotion reported in the neutral condition for which the
pooled mean rating of all the emotions reported was 0.40 (0.71)
with a median and a mode of 0 (paired t-tests comparing each
emotion with neutral: all p’s < .000001). Unpaired t-test

A B
0.20
0.25

RRsd/RespCycle-sd
0.00
0.00
RR-HighFreq

-0.20 -0.25

-0.50
-0.40
-0.75
0.60
-0.60 0.40
0.20 0.20
0.00 0.00
-0.50 -0.25 0.00 0.25 RespAmp-sd -0.20 -0.20
RespPeriod-mean RespPeriod-median

C D
1.00
0.20

0.00
RR-HighFreq

RespPeriod-sd

0.50
-0.20

-0.40 0.00

-0.60 G
-0.50 -0.50
-0.25 0.00 0.25
-0.03 0.00
RespPeriod-mean -0.05 -0.10
0.00 -0.08 RespPeriod-mean -0.25 -0.05 -0.08
0.25 -0.10 RRmean/RespCycle-mean -0.50 0.00 -0.03
RRmean/RespCycle-mean

E
0.40 Emotion
Anger
RespAmp-sd

0.20 Fear
Happiness
0.00 Sadness
-0.20

0.40
0.20
0.00 -0.45
-0.30
RespPeriod-mean -0.15
-0.20
0.00
RR-HighFreq

Fig. 1. Separation of pairs of emotions in the multidimensional space defined by subsets of dependent variables (defined in Table 1). Each point represents one
observation connected by a line to the subgroup centroı̈d. Contrasts are shown for (A) anger, fear and happiness, (B) anger and sadness, (C) fear and happiness, (D)
fear and sadness, and (E) happiness and sadness. Note that all variables are transformed using the Log(Emotion/Neutral) so that the origin of each axis corresponds to
the physiological activity measured in the neutral control condition.
 P. Rainville et al. / International Journal of Psychophysiology 61 (2006) 5 – 18 11

Table 4 cycle (RRmax-min/RespCycle), and in the high frequency range

Rotated component matrix of the exploratory principal component analysisa (RR-HighFreq), were robust in fear and significant but weaker in
Dependent Factor (% variance explained) happiness. Additional decreases in HRV were suggested by the
Variable 1 (34%) 2 (21%) 3 (17%) 4 (11%) 5 (8%) beat-by-beat changes in RR intervals in sadness (DeltaRR-mean
RespPeriod-median 0.06 0.94 0.14 0.04 0.16 and DeltaRR-sd). There was no indication of changes in HRV in
RespPeriod-sd 0.10 0.49 0.02 0.10 0.79 anger. These change observed during emotions compared to the
RespAmp-median 0.06 0.64 0.11 0.69 0.09 neutral condition confirmed the efficacy of the self-induction
RespAmp-sd 0.26 0.02 0.13 0.40 0.81
method to evoke robust physiological activity associated with
RRmean/RespCycle-mean 0.10 0.15 0.96 0.06 0.03
RRmax-min/RespCycle 0.58 0.69 0.08 0.04 0.27 the experience of emotion. The observed patterns were also
RR-mean 0.11 0.02 0.97 0.05 0.10 suggestive of some distinctive features between the four
DeltaRR-mean 0.93 0.09 0.17 0.02 0.07 emotions tested.
DeltaRR-sd 0.93 0.09 0.05 0.05 0.02 Contrasts between the four emotions were performed
RR-HighFreq 0.84 0.07 0.02 0.18 0.24
using log transformed dependent variables (see Methods).
Resp-HighFreq 0.10 0.02 0.04 0.94 0.19
a
Table 3 summarizes the results of the univariate tests
Loading > .75 are indicated in bold and values > .50 in bold and italic.
contrasting the four emotions on each dependent variable.
indicated that fear was felt as less intense that the other emotions Several dependent variables were sensitive to emotion-related
( p’s < .001) and all other pair-wise comparison between effects. Notably, indices of HRV (linear and spectral), and
emotions did not reach significance (all p’s > .20). These results respiratory activity showed highly significant effects of
indicated that subjects were successful at evoking and emotions. Pair-wise comparison between emotions revealed
experiencing the target emotions. significant effects with moderate to large effect sizes.
Exploration of those effects across several dependent vari-
3.2. Effects of emotion on physiological responses ables suggested some patterns of activity associated with the
(univariate tests) different emotions.
The possibility to discriminate the four emotions was further
Each emotion was first contrasted to the neutral condition examined using graphical representations contrasting pairs of
separately for each dependent variable as an exploratory emotions as shown in Fig. 1. Anger was clearly separated from
measure to describe the data set and identify variables of both fear and happiness based on the combination of
interest. As listed in Table 2, highly significant effects of RespPeriod-mean and RR-HighFreq (Fig. 1A). The contrast
emotion were observed in a number of dependent variables. between anger and fear confirmed the significant univariate
Respiratory period (RespPeriod-median) decreased in fear and effects for both dependent variables (Table 3). However, it is
happiness and less consistently in anger, while the variability in noteworthy that happiness did not differ significantly from
respiratory period (RespPeriod-sd) increased in sadness. In anger on either RespPeriod-mean or RR-HighFreq (Table 3),
contrast, no significant changes were observed in central yet the scatterplot shown in Fig. 1A clearly demonstrates the
tendency measures of respiratory amplitude (RespAmp-median) separation between anger and happiness when considering this
but robust increases in respiratory variability (RespAmp-sd) bi-dimensional space. This illustrates the limits of univariate
were noted in anger and, to a lesser extent, in fear. Linear RR approaches. Graphical exploration of the other pair-wise
indices (RR-mean and RRmean/RespCycle-mean) confirmed contrasts using subsets of 3-dependent variables led to similar
the robust decreases in the RR interval expected in all emotions conclusions. In those representations, anger, fear and happiness
(increase in heart rate). Decreases in HRV within respiratory were easily discriminated from each other (Fig. 1A and C)

Table 5
Effect of emotions (MANOVA) on the factors extracted from the exploratory PCA
Factor Physiological Interpretation Fa P P-valuesb for pair-wise comparisons (effect size)
A vs. F A vs. H A vs. S F vs. H F vs. S H vs. S
1 HRV (parasympathetic, respiration-uncoupled) 3.41 0.025 0.02 LSD Ns Ns Ns Ns
(0.95) (1.15) (0.80) ( 0.22) ( 0.41) ( 0.32)
2 RSA and respiratory period (parasympathetic, respiration-coupled) 2.99 0.04 LSD Ns Ns LSD 0.04 Ns
(0.84) ( 0.09) ( 0.20) ( 0.99) ( 0.89) ( 0.16)
3 RR mean (sympathetic) 4.14 0.01 Ns Ns Ns 0.02 0.02 Ns
(0.70) ( 0.58) ( 0.47) ( 1.21) ( 1.07) (0.02)
4 Respiration amplitude 0.05 0.98 Ns Ns Ns Ns Ns Ns
(0.05) (0.08) (0.12) (0.03) ( 0.08) ( 0.07)
5 Respiration variability 2.92 0.04 Ns 0.04 Ns Ns Ns LSD
(0.59) (1.03) (0.16) (0.81) (0.40) ( 0.87)
a
df = 3, 45.
b
A: Anger, F: Fear, H: Happiness, S: Sadness; p-values are given for significant contrasts based on Tukey HSD; LSD: p-uncorrected < .05 using the more lenient
least-square difference method; Ns: not significant; effect sizes are considered moderate to large (in bold) for standardized differences >0.50 and >0.80, respectively
(standardized effect size calculated using Hedges’ bias correction; Cohen, 1988).
12 P. Rainville et al. / International Journal of Psychophysiology 61 (2006) 5 – 18

1.2

0.8

0.4
Factor value
Anger
0 Fear
Happiness
-0.4 Sadness

-0.8

-1.2
Factor 1 Factor 2 Factor 3 Factor 4 Factor 5

Fig. 2. Mean T SEM of each emotion on the five PCA Factors. Each emotion is associated with a specific pattern of response across the Factors 1 – 3 and 5 (see
Table 5). Note that 0 on the y-axis corresponds to the mean of all observations and that effects relative to the neutral condition should be interpreted based on Table 2.

while sadness could be separated most clearly from fear
(Fig. 1D) but only partially from anger (Fig. 1B) and happiness
(Fig. 1E).
3.3. Multivariate exploratory approach
The exploratory PCA, applied to reduce the dimensionality
of the dependent variables, converged after five iterations onto
a five factors structure that explained 91% of the overall
variance (Table 4). Factor 1 explained most of the variance in
linear and spectral measures of HRV (DeltaRR-mean, Del-
taRR-sd, RR-HighFreq) and only a smaller part of the variance
in HRV measured within respiratory cycles (RRmax-min/
RespCycle). No respiratory variable loaded noticeably on
Factor 1. This suggests that Factor 1 captured mainly HRV
independent of respiration. In contrast, Factor 2 mainly
reflected variance in respiratory period (RespPeriod-median),
part of the variance respiratory amplitude (RespAmp-median),
and some variance in the variability of respiratory period
(RespPeriod-sd). Factor 2 also captured the variance in HRV
measured within the respiratory cycle (RRmax-min/
RespCycle) and left out from Factor 1. Taken together, Factor
2 appeared to capture most of the variance in HRV coupled
with respiration. Factor 3 specifically and almost exclusively
captured the variance in the mean RR level (RRmean/
RespCycle-mean and RR-mean) independent of HRV and
respiration. Factor 4 explained the variance in the frequency
index of respiration variability (Resp-HighFreq) and some of
the remaining variance in respiratory amplitude (RespAmp-
median). No noticeable variance in HRV was captured by this
Factor. Finally, Factor 5 reflected the remaining variance in the
variability of the respiration period (RespPeriod-sd) and
amplitude (RespAmp-sd). It is also noteworthy to mention
that the inclusion of skin conductance indices in the PCA
produced additional factors with minimal overlap with the
cardiorespiratory variables suggesting that only a negligible
amount of variance was shared between electrodermal and
cardiorespiratory variables.

10
9
8 Predicted
Absolute frequency

7 Emotion

6 Anger
Fear
5
Happiness
4
Sadness
3
2
1
0
Anger Fear Happiness Sadness
Target Emotion
Fig. 3. Distribution of the observations in the four categories derived from the discriminant analysis (color bars) as a function of the target emotion experienced by
the subject (x-axis). The horizontal broken lines represent chance levels. The overall correct prediction rate (65.3%) was significantly higher than chance (v 2(9) = 48.2,
p < .001).
 P. Rainville et al. / International Journal of Psychophysiology 61 (2006) 5 – 18
The emotions were further contrasted using a multivariate
analysis of variance applied on the five factors extracted from
the exploratory PCA. The analysis indicated an overall
significant effect of the emotion (MANOVA — Pillai’s Trace,
F (15, 129) = 2.78, p < 0.001) that was further examined using
univariate tests (Table 5). Significant main effects and pair-
wise contrasts were confirmed on Factors 1– 3 and 5, and
moderate to large effect sizes (Cohen, 1988) were observed,
indicating that the four emotions could be distinguished from
each others based on the factors extracted from the PCA. This
differentiation is illustrated in Fig. 2 by the unique pattern of
response of each emotion in the multi-dimensional space
defined by those factors. Anger differed from the other three
emotions mainly on Factor 1. This indicated that the experience
of anger was not associated with a decrease in HRV relative to
the neutral condition, contrary to fear, happiness and sadness
(see Table 2 for direction of changes relative to the neutral
condition). Fear was clearly distinguished from the other
emotions on Factors 2 and 3 (Fig. 2), indicating the strongest
decrease in respiratory period and HRV within respiratory
cycles and the largest decrease in RR. Happiness was
associated with less variability in respiration (period and
amplitude) than the other emotions as indicated by the
moderate to large effect sizes observed on Factor 5. Happiness
also distinguished itself from anger on Factor 1, and from fear
on Factors 2 and 3. Finally, sadness distinguished itself on a
combination of factors. It was associated with less HRV than
anger (Factor 1), a longer respiratory period, larger respira-
tory sinus arrhythmia, and a smaller RR decrease than fear
(Factors 2 and 3), and more respiratory variability than
happiness (Factor 5). The inclusion of electrodermal indices
in the PCA and of the corresponding factors in subsequent
analyses did not reveal additional differences between the
emotions (not shown).
3.4. Discriminant analysis
The possibility to categorize the four emotions was
further tested in a series of stepwise discriminant analyses
using all five factors extracted in the PCA. The output of the
analysis indicated that the four emotions could be adequately
classified using Factors 1, 2, 3 and 5, as shown in Fig. 3.
Factor 4 did not contribute to the discriminant solution,
consistent with the results of the MANOVA (see Table 5 and
Fig. 2). The overall successful classification rate of 65.3%
was significantly superior than chance (chance = 25%; Chi-
square(df=9) = 48.2, p < 0.001). The sensitivity (hit rate) of the
classification results ranged from 61.5% (anger) to 72.7%
(fear) while specificity (correct rejection rate) ranged from
74.3% (sadness) to 94.7% (happiness). The result of the
more conservative cross-validation procedure (leave-one-out
or U-method) led to a correct classification rate of 49.0%, a
level that remained significantly higher than chance (Chi-
square(df=9) = 27.4, p = 0.001) and attested of the stability of
the classification. Pair-wise discriminant analyses further
confirmed that the emotions could be distinguished from
each other based on different combinations of factors
13
Table 6
Pair-wise correct classification rates based on the stepwise discriminant
analyses
Correct classification Fear Happiness Sadness
ratea (cross-validated
rate; factors
includedb)
Anger 79.2% - 83.3%- 77.8%-
(79.2%-; 1, 2, 5) (83.3%-; 1, 5) (74,1%.; 1, 2, 3)
Fear 81.8%- 76.0%-
(81.8%-; 3, 2, 5) (76.0%; 3, 2)
Happiness 72.0%c
(60% ns; 5)
a
Assessment of significance is based on Chi-square tests with df = 1; pair-
wise chance rate = 50%.
b
PCA factors contributing to the discriminant classification model.
c
p < .05; .p < .01; -p < .005; ns: not significant ( p > .05).
extracted from the PCA (Table 6). The more conservative
cross-validation procedure led to identical correct classifica-
tion rates for most pair-wise contrasts, attesting of the
stability of the classification models. Only the contrasts
between sadness and anger, and between sadness and
happiness, showed some decrease following the cross-
validation procedure. This is consistent with the lower level
of specificity observed for sadness and with the less
consistent contrasts obtained between sadness and both
happiness and anger in the MANOVA (Table 5).
4. Discussion
This study provides some evidence that basic emotions are
associated with distinctive patterns of cardiorespiratory activ-
ity. Different emotions were distinguished from the neutral
control condition based on different subsets of dependent
variables and multi-dimensional exploration of the data
revealed complex patterns of activity that characterized each
emotion. The exploratory PCA suggested that the variance in
cardiorespiratory activity varied along five dimensions and
discriminative analyses based on those factors allowed for a
prediction of the emotion felt at a rate significantly higher than
chance. The functional significance of the specific patterns of
activity observed is discussed in relation to the previous reports
indicating a distinct contribution of parasympathetic and
sympathetic regulatory processes to emotions. The methodo-
logical limitations of the study to demonstrate the distinctive
patterns of physiological activity are discussed and future
research directions are identified.
4.1. Physiological dimensions of cardiorespiratory activity
The variance in the multiple chronometric indices of
cardiorespiratory processes was structured along five physio-
logically meaningful dimensions. The multivariate approach
allowed us to separate the relative contribution of high
frequency HRV associated with vagal outflow to the heart
(Factors 1 and 2) from changes in heart rate independent of
HRV and possibly reflecting sympathetic activity (Factor 3).
14 P. Rainville et al. / International Journal of Psychophysiology 61 (2006) 5 – 18
This confirms that changes in HRV are not simply secondary to
changes in heart rate. Similar separation of sympathetic and
parasympathetic influences has been performed using the
relative low and high frequency components of HRV extracted
from spectral analyses applied to longer recording epochs
(Task force of the European Society of Cardiology and the
North American Society of Pacing and Electrophysiology,
1996), from the temporal coupling of RR changes with changes
in blood pressure (Hughson et al., 1993), or from fractal
dimensionality measurements (Butler et al., 1993; Yeragani et
al., 1993; Toichi et al., 1997). The observed partial segregation
of variance associated with sympathetic and parasympathetic
influences is highly consistent with a multi-dimensional model
of autonomic regulation of the heart (Berntson et al., 1994).
The present results further indicate an additional separation
between HRV changes coupled with respiration (Factor 2:
respiratory sinus arrhythmia) and uncoupled with respiration
(Factor 1). Respiration-related variables describing both the
level and variability in period and amplitude also contributed
unique variance (Factors 4 – 5) independent of changes in heart
rate or HRV.
The physiological mechanisms underlying the changes
associated with each Factor extracted in the PCA can be
inferred from the specific variable loading pattern reported in
Table 3. The baroreceptor reflex mediates respiratory sinus
arrhythmia (Factor 2) and involves vagal afferents to the
solitary nucleus, which sends efferent projections to the
cardiomotor preganglionic neurons in the motor nucleus of
the vagus and the nucleus ambiguus (Standish et al., 1994).
Those output nuclei are also under the influence of mesence-
phalic and diencephalic structures (e.g. periacqueductal grey
area, hypothalamus, amygdala) that contribute to the genera-
tion and coordination of complex patterns of somato-visceral
responses that characterize emotion (Bandler and Shipley,
1994; Saper, 2002). Here, Factor 1 may best capture the top-
down regulatory control of vagal output by higher-order brain
structures independent of respiration, while Factor 2 reflects
the changes in HRV driven by respiration and mediated by
the baroreceptor reflex and vagal output (cf. loading of linear
HRV index and respiration measures on Factor 2). This is
consistent with a hierarchical organisation of sympato-vagal
regulatory mechanisms previously proposed to characterize the
somatic expression of emotion (Porges, 1997). Importantly,
Factors 1 and 2 accounted for 43% and 21% of the variance,
respectively. This underscores the contribution of top-down
regulatory mechanisms and the critical importance of indexing
respiration-dependent and respiration-independent parasympa-
thetic activity to describe the somatic states associated with
emotions.
Global changes in heart rate generally reflect changes in
vagal and/or sympathetic influences. However, in the PCA
analysis, the variance in mean RR (RR-mean and RRmean/
RespCycle-mean) was almost exclusively explained by Factor
3, and was almost completely segregated from changes in HRV
associated with Factors 1 –2. This implies that the global
changes in mean RR observed here reflect sympathetic
mechanisms and/or slow changes in vagal activity, that do
not contribute to the linear and spectral indices of high
frequency HRV. Factor 3 may therefore reflect the descending
excitatory influence of supra-spinal mechanisms on the
excitation of pre-ganglionic neurons of the intermediolateral
column of the thoracic spinal cord (sympathetic output), the
changes in circulating catecholamines affecting the heart, and/
or the (slow) tonic release in vagal tone. This activity is
typically captured by the low frequency component of HRV
that could not be assessed directly in the Fast Fourier
Transform analysis performed here, due to the short duration
of the recordings. Still, the multivariate method used here
allowed us to disentangle those various components of heart
rate regulation.
It is noteworthy that variations in skin conductance
activity did not overlap significantly with changes in
cardiorespiratory activity. Based on the pattern of cardiore-
spiratory activity, it might have been expected that Factor 3
(global changes in heart rate) would also explain changes in
skin conductance because it reflects, at least in part,
sympathetic activity. However, consistent with the distinctive
regulatory mechanisms involved in cardiac and skin sympa-
thetic activity, those factors were segregated. Furthermore,
the inclusion of skin conductance in the PCA and in the
discriminant analysis did not improve the discrimination of
emotions.
4.2. Characterization of emotions in a multidimensional
cardiorespiratory space
The patterns we report suggest that the robust increases in
heart rate observed in fear, anger, happiness, and sadness,
relative to the neutral condition (see Table 2) reflect varied
combinations of sympathetic and parasympathetic changes, the
latter of which may be coupled or uncoupled with respiration.
The separation of the four emotions was evident along many
dimensions and no single factor could account for all pair-wise
segregations. This finding is clearly not compatible with the
pervasive view that one single dimension, that of arousal,
suffices to characterize the somatic patterns of different
emotions. Instead, the present results confirm the multi-
dimensional nature of the somatic states which characterize
basic emotions.
Based on the results of the analyses of variance on the
PCA factors, the effect size estimates, and the discriminant
analyses, we propose a heuristic decision tree to distinguish
the emotions based on cardiorespiratory criteria (Fig. 4). At
the first node, anger is discriminated from the other emotions
by an increase in heart rate without noticeable changes in
high frequency HRV (see Tables 2, 3 and 5, and Fig. 2).
This is consistent with the relative dominance of sympathetic
activation previously reported in anger (McCraty et al.,
1995). Notably, the significant increase in heart rate in anger
occurred in the absence of changes in high frequency HRV.
By contrast, the increase in heart rate observed in fear,
happiness, and sadness was clearly associated with changes
in vagal activity indexed by high frequency HRV. Stemmler
et al. (2001) reported similar effects in fear and anger where
 P. Rainville et al. / International Journal of Psychophysiology 61 (2006) 5 – 18 15

Heuristic Decision Tree

Increase in HR with Increase in HR

NO change in high and decrease in high
frequency (HF) HRV frequency (HF) HRV

Decrease in HF HRV Decrease in HF HRV

Un-coupled with Coupled with changes
changes in respiration in respiration (RSA)

Increase in respiration No change in respiration

variability variability

ANGER SADNESS HAPPINESS FEAR

Fig. 4. Heuristic decision tree derived from the observations of distinct patterns of cardiorespiratory activity during basic emotions (see text).

both emotions were found to reduce the heart period but
only fear was associated with decreases in heart period
variability. This is also consistent with the conclusions
reached by Ekman et al. (1983) and Levenson et al.
(1990) who derived a decision tree based on measures of
heart rate and skin temperature recorded during a directed
facial action task and in a relived emotion task. Both tasks
showed increases in heart rate in anger and fear but larger
increases were found in skin temperature in anger. Taken in
isolation, the increase in heart rate could be interpreted as a
stronger non-specific arousal response. However, the increase
in skin temperature may relate to our findings that the heart
rate increase in anger reflects mainly sympathetic activation
while the increase found in other emotions may also involve a
decrease in parasympathetic activity (i.e. here a decrease in
HF heart rate variability; see Fig. 4). This clearly demon-
strates that similar changes in heart rate induced by various
emotions may reflect different combinations of sympathethic
and parasympathetic changes, consistent with modern multi-
dimensional conception of chronometric regulation of cardiac
function (Task force of the European Society of Cardiology
and the North American Society of Pacing and Electrophys-
iology, 1996; Berntson et al., 1993a; Malliani et al., 1991;
Pagani et al., 1997). This is consistent with the notion that
basic emotions are associated with distinct patterns of auto-
nomic activity.
At the next node of the decision tree (Fig. 4), fear is
distinguished from happiness and sadness by a robust
decrease in high frequency HRV which is strongly coupled
with respiration (see Tables 2, 3 and 5 and Fig. 2, Factor 2).
Decreases in HRV were also noted in happiness and to a
lesser extent in sadness (Table 2). However, the decrease in
HRV noted in fear was mainly the result of reduced
respiratory sinus arrhythmia (Factor 2), while the effects
observed in happiness and sadness were not coupled with
respiration (Factor 1). This important reduction in vagal tone
during fear is consistent with previous reports in patients
suffering from panic attacks (Yeragani et al., 1991; Friedman
and Thayer, 1998; Rao and Yeragani, 2001). This suggests
that the reduction in HRV in fear may be mainly mediated
by the baroreceptor reflex and secondary to the decrease in
respiratory period. We interpret the experimental studies
using lactate infusion and hyperventilation to imply that the
modification of respiration and the reduction of vagal
activity may be critical to the induction of fear states and
that these changes may precede the subjective experience of
fear.
By contrast, the reduction in HRV we observed during
happiness and also reported in the brief summary of another
study (Lane et al., 2001), may reflect additional influences
from brain structures affecting pre-ganglionic vagal efferent
neurons in the nucleus ambiguus or the vagal motor nucleus,
independent from changes in respiration. The function of this
decrease in vagal tone during happiness remains to be
elucidated. Finally, the larger variability in respiration
observed in sadness than in happiness provided the final
criterion in the decision tree to discriminate the emotions.
Taken together, these results support the hypothesis that basic
emotions are associated with distinct patterns of chronomet-
ric cardiorespiratory activity.
16 P. Rainville et al. / International Journal of Psychophysiology 61 (2006) 5 – 18
4.3. Theoretical and methodological considerations
The variety of the emotional situations and reactions that
were recalled and relived in the laboratory is likely to have
reduced our ability to discriminate the emotions. For example,
we did not control for the potential effects of anger expression/
suppression, or for the different behavioral responses (e.g.
fight, flight, or freeze) mentally evoked by the fear experienced
in the specific scenario recalled. The mental evocation of motor
activity has been shown to affect autonomic activity (Decety,
1996) and this may have contributed to the variability within
each emotion condition. Similarly, we relied on mental imagery
and reliving of an emotional episode, a paradigm that may
reduce the ability to detect specific patterns across emotions
compared to real-life emotional episodes (Stemmler et al.,
2001). However, those effects would have worked against our
hypothesis that basic emotions are associated with distinct
patterns of cardiorespiratory activity by introducing uncon-
trolled variance in the data. Nevertheless, the specification of
those effects in emotion research will most certainly contribute
to improve our ability to discriminate emotions based on
physiological activity.
Our method specifically capitalized on subjective emotional
feelings as indicators of emotional states and our results
confirmed the association with a distinct somatic state. In spite
of differences in the specific situation and behavioural
response(s) remembered, the patterns of physiological activity
associated with the emotions converged into specific areas of the
multi-dimensional physiological space suggesting that the
consensus on the emotion felt was associated with consistent
patterns of physiological activity. This implies that the experi-
ence we associate with basic emotions have a distinct somato-
visceral correspondence in spite of the different scenarios and
the potentially different behavioral responses evoked mentally.
This does not negate the influence of specific behavioral
responses on somato-visceral activity, but rather suggests that
those responses may produce shifts in physiological activity
within the sub-space that characterizes a certain emotion.
It is plausible that the specific behavioural response
associated with the expression of those emotions and the
experimental context may explain part of the variability
observed within the emotional conditions. For example,
physiological activity characterizing anger varied along mea-
sures of respiratory period (Factor 2) and high frequency HRV
(RR-HighFreq; Factor 1). Changes along those dimensions
could relate to outward expression. A positive relation has been
reported between the low frequency component of HRV and
anger expression (Ramaekers et al., 1998). It is therefore possible
that the suppression of anger expression may produce shifts
towards relative increases in vagal tone paired with faster
breathing, while the facilitation of anger expression may be
associated with shifts towards decreases in vagal tone (relative
decreased RR-HighFreq) and slower breathing. According to
our hypothesis, however, those shifts would be expected to
occur within the sector of the multi-dimensional physiological
space that characterizes anger. A shift outside of that sector
would be expected to produce a change in emotional experience.
The context of the emotional induction has also been shown
to influence the pattern of response and thereby limit the
external validity of any paradigm. Stemmler et al. (2001) have
shown that the somato-visceral specificity of fear compared to
anger interacts with the experimental context (real life vs.
recalled and imagined situation). However, this study did find a
decrease in HRV during fear but not anger in both contexts, in
support of our interpretation that some aspects of emotional
responses may consistently characterize specific emotional
states in different contexts.
One possible way to increase our ability to discriminate
between emotions would be to better control for individual
differences in physiological activity. Here, individual subjects
were their own control as each emotion observation was
normalized based on the neutral condition. However, subjects
did not contribute to all four emotion conditions and we relied
on a between subject design to contrast the four emotions. One
alternative would have been to use a full within-subject design
by asking all subjects to recall one autobiographical episode
for each of the four emotions. However, this approach could
reduce the strength of the emotion experienced if some
emotional episodes are not recalled as reliably and if the
emotions are not relived as vividly. Here, we chose to recruit
subjects based on their self-reported ability to relive at least
one emotion based on autobiographical recall. This compro-
mise illustrates the challenge of emotion research in psycho-
physiology, constantly struggling to control adequately for
baseline physiological differences and to induce strong
emotions reliably.
Several other aspects of this study should be considered
with caution before a definite conclusion is reached. An
important limitation concerns the small sample size. The
investigation of multi-dimensional physiological space requires
the assessment of several physiological responses. Here the
number of dependent variables was very high relative to the
number of observations but several variables were expected to
be highly overlapping and an exploratory PCA was applied to
reduce the dimensionality of the data set. This analysis requires
a large number of independent observations to lead to
interpretable and stable factors. Here we applied the PCA on
a small sample with partly dependent observations (i.e. some of
our subjects contributed to two emotion conditions). Never-
theless, the output from the PCA was easily interpretable based
on the known physiology of cardiorespiratory regulation, an
observation that supports the validity of the factor structure
obtained. However, we must emphasize that the results of the
PCA should be considered preliminary in view of the important
limitations and because the sample did not allow us to perform
a confirmatory analysis to test for the stability of the factors
structure observed.
The discriminant analysis may not suffer as much from such
limitation as it is better suited to characterize small sample
sizes in high dimensional space (Friedman, 1989 and Duda et
al., 2001). Indeed, the leave-one-out method (U-method or
cross-validation) involves the repeated calculation of a
discriminant function based on a sample of n 1, as each
observation is successively left out in the successive reiteration
 P. Rainville et al. / International Journal of Psychophysiology 61 (2006) 5 – 18
of the analysis (SPSS v 11.0). The overall cross-validation
performance is therefore based on the ability of the model to
correctly classify each observation based on a function derived
from all the other observations. The significant and relatively
high correct classification rate observed using this method
suggests good internal stability of the results. External
validation of the discriminant function and of the heuristic
algorithm proposed in Fig. 4 will be required to confirm the
generalizability of the findings.
4.4. Future research directions
The increased sensitivity and specificity of non-invasive
methods of measurement and analysis of autonomic activity in
humans allow the concurrent evaluation of several regulatory
systems. Future assessment of patterns of peripheral activity
during emotions should include refined measurements of both
cardiac and respiratory activity (e.g. Boiten et al., 1994;
Wientjes, 1992), as well as concurrent measurements of other
systems, including multiple autonomic and endocrine
responses, in order to provide a more comprehensive assess-
ment of somatic states. These studies must further consider the
non-linear organization of physiological systems, their tempo-
ral dynamics, and the interactions among systems, all of which
may contribute to the unique experience of emotions. The
distinctive changes associated with specific emotions might
explain the association between certain disorders of emotions
(e.g. anxiety disorders) and specific somatic symptoms or
diseases of the cardiovascular system (e.g. Grippo and
Johnson, 2002; Kubzansky et al., 1997; Sheps and Sheffield,
2001), although more research is required to establish the
significance of this relation (Fleet et al., 2000).
Modern functional brain imaging methods further provide
another perspective on the problem of peripheral regulation
during emotion. Brain systems involved in autonomic
regulation are essential for the production of the coordinated
patterns of somatic activity that are part and parcel of emo-
tions (Damasio, 1994; Damasio et al., 2000; Saper, 2002).
Recent evidence further emphasize the role of the anterior
cingulate cortex and the insula in autonomic regulation and in
the representation of somatic states characterizing emotions
and associated with cognition (Critchley et al., 2000a, b;
Critchley et al., 2001; Lane et al., 2001). The inclusion of
multidimensional peripheral physiological measures in those
functional brain imaging studies may contribute to the
refinement of theoretical models describing higher-order
mechanisms of autonomic regulation in emotion. For exam-
ple, the five independent factors describing cardiorespiratory
activity in the present study may relate to different
components of the brain network involved in emotion.
Examining the temporal dynamics of the activation of brain
networks associated with emotions might further contribute to
the understanding of the role of those different regions in the
efferent and afferent components of autonomic regulation and
their relative role in the subjective experience of emotion
(Critchley et al., 2000a, b; Critchley et al., 2004; Bechara and
Naqvi, 2004).
17
5. Conclusion
The hypothesis that the feeling of basic emotions is
consistently associated with distinct patterns of somatic activity
is receiving increasing support from psychophysiological
research. The implication of our findings validates the
recommendation of a growing number of independent inves-
tigators, according to which a multidimensional assessment of
physiological activity is necessary to describe somatic states
associated with basic emotions. Our results further indicate a
contribution of dynamic changes in cardiorespiratory activity to
the production of distinct somatic states that normal individuals
experience as anger, fear, happiness, and sadness. The iden-
tification of distinctive physiological patterns characterizing
emotional states will foster the development of automated
systems for the recognition of emotions using real-time
multivariate analysis procedures (Picard et al., 2001). In
addition to the exciting but challenging possibility to imple-
ment such system in the design of emotionally competent
machines, these systems may contribute to the clinical
evaluation of affective disorders and to research in psychoso-
matic and affective neurosciences.
Acknowledgments
We thank Daniel Tranel for his advice on psychophysiolog-
ical methods and analysis, Anne Virasith and Martin Bilodeau
for their guidance in the application and interpretation of
discriminant analyses, and Don C. Fowles for his constructive
comments on the manuscript. Supported in part by grants from
the Mathers Foundation and NIH (NINDS) Grant 5 PO 1
NS19632-23, by the Human Frontier Science Program (long-
term fellowship to PR) and by the Quebec FRSQ (PR).
References
Bandler, R., Shipley, M.T., 1994. Columnar organization in the midbrain
periaqueductal gray: modules for emotional expression? Trends Neurosci.
17, 379 – 389.
Bechara, A., Naqvi, N., 2004. Listening to your heart: interoceptive awareness
as a gateway to feeling. Nat. Neurosci. 7, 102 – 103.
Berntson, G.G., Stowell, J.R., 1998. ECG artifacts and heart period variability:
don’t miss a beat. Psychophysiology 35, 127 – 132.
Berntson, G.G., Cacioppo, J.T., Quigley, K.S., 1993a. Respiratory sinus
arrhythmia: autonomic origins, physiological mechanisms, and psycho-
physiological implications. Psychophysiology 30, 183 – 196.
Berntson, G.G., Cacioppo, J.T., Quigley, K.S., 1993b. Respiratory sinus
arrhythmia: autonomic origins, physiological mechanisms, and psycho-
physiological implications. Psychophysiology 30, 183 – 196.
Berntson, G.G., Cacioppo, J.T., Quigley, K.S., Fabro, V.T., 1994. Autonomic
space and psychophysiological response. Psychophysiology 31, 44 – 61.
Boiten, F.A., Frijda, N.H., Wientjes, C.J.E., 1994. Emotions and respiratory
patterns: Review and critical analysis. Int. J. Psychophysiol. 17, 103 – 128.
Butler, G.C., Yamamoto, Y., Xing, H.C., Northey, D.R., Hughson, R.L., 1993.
Heart rate variability and fractal dimension during orthostatic challenges. J.
Appl. Physiol. 75, 2602 – 2612.
Cacioppo, J.T., Berntson, G.G., Larsen, J.T., Poehlmann, K.M., Ito, T.A., 2000.
The psychophysiology of emotion. In: Lewis, M., Haviland-Jones, J.M.
(Eds.), The Handbook of Emotion. 2nd Edition. Guilford Press, New York,
pp. 173 – 191.
18 P. Rainville et al. / International Journal of Psychophysiology 61 (2006) 5 – 18
Cannon, W.B., 1931. Again the James – Lange and the thalamic theories of
emotion. Psychol. Rev. 38, 281 – 295.
Cannon, W.B., 1987. The James – Lange theory of emotions: a critical
examination and an alternative theory. By Walter B. Cannon, 1927. Am.
J. Psychol. 100, 567 – 586.
Christie, I.C., Friedman, B.H., 2004. Autonomic specificity of discrete emotion
and dimensions of affective space: a multivariate approach. Int. J.
Psychophysiol. 51, 143 – 153.
Cohen, J., 1988. Statistical power analysis for the behavioral sciences. (second
edition. Lawrence Erlbaum Associates, Hillsdale, NJ.
Collet, C., Vernet-Maury, E., Delhomme, G., Dittmar, A., 1997. Autonomic
nervous system response patterns specificity to basic emotions. J. Auton.
Nerv. Syst. 62, 45 – 57.
Critchley, H.D., Corfield, D.R., Chandler, M.P., Mathias, C.J., Dolan, R.J.,
2000. Cerebral correlates of autonomic cardiovascular arousal: a functional
neuroimaging investigation in humans. J. Physiol. 523 (Pt 1), 259 – 270.
Critchley, H.D., Elliott, R., Mathias, C.J., Dolan, R.J., 2000. Neural activity
relating to generation and representation of galvanic skin conductance
responses: a functional magnetic resonance imaging study. J. Neurosci. 20,
3033 – 3040.
Critchley, H.D., Melmed, R.N., Featherstone, E., Mathias, C.J., Dolan, R.J.,
2001. Brain activity during biofeedback relaxation: a functional neuroima-
ging investigation. Brain 124, 1003 – 1012.
Critchley, H.D., Wiens, S., Rotshtein, P., Ohman, A., Dolan, R.J., 2004. Neural
systems supporting interoceptive awareness. Nat. Neurosci. 7, 189 – 195.
Damasio, A.R., 1994. Descarte’s Error: Emotion, Reason and the Human Brain.
Avon Books, New York, NY.
Damasio, A.R., Grabowski, T.J., Bechara, A., Damasio, H., Ponto, L.L.B.,
Parvizi, J., et al., 2000. Subcortical and cortical brain activity during the
feeling of self-generated emotions. Nat. Neurosci. 3, 1049 – 1056.
Decety, J., 1996. Do imagined and executed actions share the same neural
substrate? Cogn. Brain Res. 3, 87 – 93.
Duda, R.O., Hart, P.E., Stork, D.G., 2001. Pattern Classification, Second
Edition. John Wiley and Sons, New York, NY.
Ekman, P., Levenson, R.W., Friesen, W.V., 1983. Autonomic nervous system
activity distinguishes among emotions. Science 221, 1208 – 1210.
Fleet, R., Lavoie, K., Beitman, B.D., 2000. Is panic disorder associated with
coronary artery disease? A critical review of the literature. J. Psychosom.
Res. 48, 347 – 356.
Friedman, J.H., 1989. Regularized discriminant analysis. J. Am. Stat. Assoc.
84, 165 – 175.
Friedman, B.H., Thayer, J.F., 1998. Autonomic balance revisited: panic anxiety
and heart variability. J. Psychosom. Res. 44, 133 – 151.
George, D.T., Nutt, D.J., Walker, W.V., Porges, S.W., Adinoff, B., Linnoila, M.,
1989. Lactate and hyperventilation substantially attenuates vagal tone in
normal volunteers. Arch. Gen. Psychiatry 46, 153 – 156.
Grippo, A.J., Johnson, A.K., 2002. Biological mechanisms in the
relationship between depression and heart disease. Neurosci. Biobehav.
Rev. 26, 941 – 962.
Grossman, P., van Beek, J., Wientjes, C., 1990. A comparison of three
quantification methods for estimation of respiratory sinus arrhythmia.
Psychophysiology 27, 702 – 714.
Hughson, R.L., Quintin, L., Annat, G., Yamamoto, Y., Gharib, C., 1993.
Spontaneous baroreflex by sequence and power spectral methods in
humans. Clin. Physiol. 13, 663 – 676.
James, W., 1994. The physical bases of emotion. 1894. Psychol. Rev. 101,
205 – 210.
Kubzansky, L.D., Kawachi, I., Spiro III, A., Weiss, S.T., Vokonas, P.S.,
Sparrow, D., 1997. Is worrying bad for your heart? A prospective study of
worry and coronary heart disease in the Normative Aging Study.
Circulation 95, 818 – 824.
Lane, R.D., Reiman, E.M., Ahern, G.L., Thayer, J.F., 2001. Activity in medial
prefrontal cortex correlates with vagal component of heart rate variability
during emotion. Brain Cogn. 47, 97 – 100.
Levenson, R.W., Ekman, P., Friesen, W.V., 1990. Voluntary facial action
generates emotion-specific autonomic nervous system activity. Psycho-
physiology 27, 363 – 384.
Malliani, A., Pagani, M., Lombardi, F., Cerutti, S., 1991. Cardiovascular neural
regulation explored in the frequency domain. Circulation 84, 482 – 492.
McCraty, R., Atkinson, M., Tiller, W.A., Rein, G., Watkins, A.D., 1995. The
effects of emotions on short-term power spectrum analysis of heart rate
variability. Am. J. Cardiol. 76, 1089 – 1093.
Pagani, M., Montano, N., Porta, A., Malliani, A., Abboud, F.M., Birket, C., et
al., 1997. Relationship between spectral components of cardiovascular
variabilities and direct measures of muscle sympathetic nerve activity in
humans. Circulation 95, 1441 – 1448.
Picard, R.W., Vyzas, E., Healey, J., 2001. Toward machine emotional
intelligence: analysis of affective physiological state. IEEE Trans. Pattern
Anal. Mach. Intell. 23, 1175 – 1191.
Porges, S.W., 1997. Emotion: an evolutionary by-product of the neural
regulation of the autonomic nervous system. Ann. N.Y. Acad. Sci. 807,
62 – 77.
Ramaekers, D., Ector, H., Demyttenaere, K., Rubens, A., Van de, W.F., 1998.
Association between cardiac autonomic function and coping style in
healthy subjects. Pacing Clin. Electrophysiol. 21, 1546 – 1552.
Rao, R.K., Yeragani, V.K., 2001. Decreased chaos and increased nonlinearity of
heart rate time series in patients with panic disorder. Auton. Neurosci. 88,
99 – 108.
Saper, C.B., 2002. The central autonomic nervous system: conscious visceral
perception and autonomic pattern generation. Annu. Rev. Neurosci. 25,
433 – 469.
Schacter, S., Singer, J.E., 1962. Cognitive, social, and physiological determi-
nants of emotional state. Psychol. Rev. 69, 379 – 399.
Sheps, D.S., Sheffield, D., 2001. Depression, anxiety, and the cardiovascular
system: the cardiologist’s perspective. J. Clin. Psychiatry 62 (Suppl 8),
12 – 16.
Standish, A., Enquist, L.W., Schwaber, J.S., 1994. Innervation of the heart and
its central medullary origin defined by viral tracing. Science 263, 232 – 234.
Stemmler, G., Heldmann, M., Pauls, C.A., Scherer, T., 2001. Constraints for
emotion specificity in fear and anger: the context counts. Psychophysiology
38, 275 – 291.
Talman, W.T., Kelkar, P., 1993. Neural control of the heart: central and
peripheral. Neurol. Clin. 11, 239 – 256.
Task force of the European Society of Cardiology and the North American
Society of Pacing and Electrophysiology, 1996. Heart rate variability:
standards of measurement, physiological interpretation, and clinical use.
Circulation 93, 1043 – 1065.
Toichi, M., Sugiura, T., Murai, T., Sengoku, A., 1997. A new method of
assessing cardiac autonomic function and its comparison with spectral
analysis and coefficient of variation of R – R interval. J. Auton. Nerv. Syst.
62, 79 – 84.
Vernet-Maury, E., Alaoui-Ismaili, O., Dittmar, A., Delhomme, G., Chanel, J.,
1999. Basic emotions induced by odorants: a new approach based on
autonomic pattern results. J. Auton. Nerv. Syst. 75, 176 – 183.
Wientjes, C.J., 1992. Respiration in psychophysiology: methods and applica-
tions. Biol. Psychol. 34, 179 – 203.
Yeragani, V.K., Pohl, R., Balon, R., Ramesh, C., Glitz, D., Jung, I., et al., 1991.
Heart rate variability in patients with major depression. Psychiatry Res. 37,
35 – 46.
Yeragani, V.K., Srinivasan, K., Vempati, S., Pohl, R., Balon, R., 1993. Fractal
dimension of heart rate time series: an effective measure of autonomic
function. J. Appl. Physiol. 75, 2429 – 2438.
Yeragani, V.K., Srinivasan, K., Pohl, R., Berger, R., Balon, R., Berchou, R.,
1994. Sodium lactate increases sympathovagal ratios in normal control
subjects: spectral analysis of heart rate, blood pressure, and respiration.
Psychiatry Res. 54, 97 – 114.
Yeragani, V.K., Srinivasan, K., Pohl, R., Berger, R., Balon, R., Berchou, R.,
1994. Sodium lactate increases sympathovagal ratios in normal control
subjects: spectral analysis of heart rate, blood pressure, and respiration.
Psychiatry Res. 54, 97 – 114.