AuthoonomicCluster.pdf

Biological Psychology 84 (2010) 463–473

Contents lists available at ScienceDirect

Biological Psychology

journal homepage: www.elsevier.com/locate/biopsycho

Autonomic speciﬁcity of basic emotions: Evidence from pattern classiﬁcation and

cluster analysis

Chad L. Stephens a , Israel C. Christie b , Bruce H. Friedman a , ∗

a Virginia Polytechnic Institute and State University, Blacksburg, VA 24061-0436, United States

b University of Pittsburgh, Pittsburgh, PA 15213-3313, United States

article info

abstract

Article history:

Received 19 May 2009

Received in revised form 9 March 2010

Accepted 12 March 2010

Available online 23 March 2010

Autonomic nervous system (ANS) speciﬁcity of emotion remains controversial in contemporary emo-

tion research, and has received mixed support over decades of investigation. This study was designed

to replicate and extend psychophysiological research, which has used multivariate pattern classiﬁca-

tion analysis (PCA) in support of ANS speciﬁcity. Forty-nine undergraduates (27 women) listened to

emotion-inducing music and viewed affective ﬁlms while a montage of ANS variables, including heart

rate variability indices, peripheral vascular activity, systolic time intervals, and electrodermal activity,

were recorded. Evidence for ANS discrimination of emotion was found via PCA with 44.6% of overall obser-

vations correctly classiﬁed into the predicted emotion conditions, using ANS variables (z = 16.05, p < .001).

Cluster analysis of these data indicated a lack of distinct clusters, which suggests that ANS responses to

the stimuli were nomothetic and stimulus-speciﬁc rather than idiosyncratic and individual-speciﬁc. Col-

lectively these results further conﬁrm and extend support for the notion that basic emotions have distinct

ANS signatures.

Keywords:

Autonomic nervous system

Emotions

Pattern classiﬁcation analysis

Cluster analysis

A core area of investigation historically derived from James’

(1884) inquiry into the nature and source of emotions is the degree

to which different emotional states might be characterized by

unique patterns of autonomic nervous system (ANS) activity (see

Friedman, this issue for a review of this subject). The research

literature on ANS speciﬁcity for basic emotions has been character-

ized as both supportive (see Ekman et al., 2003 ) and inconsistent

( Cacioppo et al., 2000; Barrett, 2006 ). However, recent research

using the multivariate technique of pattern classiﬁcation analysis

(PCA) has consistently reported patterning of autonomic variables

allowing for differentiation among emotion conditions ( Christie

and Friedman, 2004; Kreibig et al., 2007; Nyklicek et al., 1997;

Rainville et al., 2006 ). The current study is speciﬁcally directed at

replicating and extending the ﬁndings of Christie and Friedman

(2004) and Nyklicek et al. (1997) , which represent a line of con-

tinuity in our research group (the former was based largely on the

latter). Nyklicek et al. (1997) utilized music excerpts to induce emo-

tion; Christie and Friedman (2004) used ﬁlm clips for this purpose.

Both induction techniques were used and compared in the present

study.

A secondary goal of the present study was to examine

individual-speciﬁc ANS response patterns to emotion inductions

by the use of cluster analysis. This approach was modeled after a

study that sought to quantify distinct clusters of individual-speciﬁc

cardiovascular response patterns to various laboratory stressors

( Allen et al., 1991 ). By analogy, this strategy was adopted to

assess individual-speciﬁc response patterns across diverse emotion

induction conditions. Methodological and analytic issues central to

the study of autonomic speciﬁcity are also discussed.

1. Methodological considerations

Recent research on ANS differentiation of emotions has

addressed discrepancies between studies by adopting relevant

suggestions proposed by Stemmler (1989) and Cacioppo et al.

(1993) . For example, distinguishing physiological responses to

emotion inductions from those arising from the experimental

context requires a neutral “context without emotion” condition

( Stemmler, 1989 ). Accordingly, the core studies upon which this

paper is based, as well as the present investigation, included such a

control ( Christie and Friedman, 2004; Nyklicek et al., 1997; Kreibig

et al., 2007 ). Such methods enhance the comparability of ANS speci-

ﬁcity studies across different induction techniques. Furthermore,

Portions of these data presented were in S. Kreibig (Chair), William James’ legacy:

The present state of autonomic response speciﬁcity of emotion, a symposium held at

the Annual Meeting of the Society for Psychophysiological Research, Savannah, GA,

in October 2007.

∗ Corresponding author at: Department of Psychology (0436), Virginia Polytechnic

Institute and State University, Blacksburg, VA 24061-0436, USA.

Tel.: +1 540 231 9611; fax: +1 540 231 3652.

E-mail address: bhfriedm@vt.edu (B.H. Friedman).

doi: 10.1016/j.biopsycho.2010.03.014

464

C.L. Stephens et al. / Biological Psychology 84 (2010) 463–473

a common set of guidelines for psychophysiological research on

emotion promotes coherence in the literature. These principles

generally have been followed in the present study, as well as in

similar research ( Christie and Friedman, 2004; Kreibig et al., 2007;

Nyklicek et al., 1997 ). Hence, the methodology and results are

highly comparable across these studies.

ate multivariate statistical technique to investigate the presence

of distinct patterns of physiology for discrete emotional experi-

ences.

The PCA applied in the current study is an internal classiﬁca-

tion analysis (ICA, Huberty, 1994 ). ICA classiﬁes units whose data

are used to determine the classiﬁcation statistics. The alternative

to this is an external classiﬁcation analysis, in which data on the

cases to be classiﬁed are not used in constructing the classiﬁca-

tion function ( Huberty, 1994 ). Instead, these cases are classiﬁed

by classiﬁcation statistics obtained on another set of cases. It

has been demonstrated that the total hit rate obtained from an

internal classiﬁcation analysis will generally be biased and over-

estimate the true hit rate ( Frank et al., 1965; Hora and Wilcox,

1982; Huberty, 1984 ). In order to minimize this bias the number

of predictors was kept small relative to the size of the sample.

This step addresses the issue of tests of signiﬁcance of difference

between proportions accurately reﬂecting that the discriminant

analysis classiﬁcation rule results in a better-than-chance hit rate

( Huberty, 1994; Panel on Discriminant Analysis, Classiﬁcation, and

Clustering, 1989 ).

2. Induction technique selection

The selection of affect induction method is of crucial impor-

tance. Seven dimensions salient to this choice have been outlined

( Rottenberg et al., 2007 ). These considerations include: ecological

validity, intensity, demand characteristics, standardization, tem-

poral considerations, attentional capture, and complexity. Careful

reﬂection on these dimensions is key when determining which

technique best matches the aims of a study. Additionally, the emo-

tional framework upon which the induction method was developed

is of utmost importance, because this will limit interpretation of the

results. For example, emotion inductions based upon a dimensional

model of affect (e.g., Russell, 1980 ) are unlikely to yield information

easily interpretable in a discrete emotion framework (e.g., Izard,

1977 ).

Both music and ﬁlm affect induction techniques were used in the

present study. In part, this choice was based on studies by Nyklicek

et al. (1997) which employed music, and Christie and Friedman

(2004) which used ﬁlm inductions. Selection of emotion types in

these two studies was based on a hybrid discrete-dimensional

model of affect, which posits a hierarchical relation between lower-

order discrete emotions and higher order affective dimensions.

Inclusion of both music and ﬁlm also allows for a direct compari-

son of these two induction techniques, which to date has only been

accomplished indirectly through meta-analysis ( Gerrards-Hesse et

al., 1994; Westermann et al., 1996 ).

4. Stimulus-response speciﬁcity and individual-response

speciﬁcity

The concept of autonomic speciﬁcity of emotional states can

be viewed in the more general context of two basic princi-

ples in psychophysiology: stimulus-response speciﬁcity (SRS) and

individual-response sterotypy (IRS). Simply stated, SRS asserts that

certain stimulus contexts are associated with speciﬁc patterns

of physiological responding ( Lacey, 1959, 1967 ). For example,

a well-known SRS pattern is the orienting response to novelty,

which consists of heart rate deceleration, delayed respiration

followed by increased amplitude and decreased frequency, periph-

eral vasoconstriction, and increased skin conductance ( Sokolov,

1963 ). The notion of SRS is predicated on there being an adap-

tive match of the response pattern to the situation; e.g., the

physiological responses observed in orienting subserve attentional

and behavioral responses to novel stimuli. As such, SRS patterns

have presumably evolved and possess some degree of universality

within and across species.

It is in this framework of adaptation that the principle of SRS can

be applied to the autonomic speciﬁcity question. A functional view

of affect holds that basic emotions evolved for their adaptive value

in speciﬁc contexts (e.g., Ekman, 1992; Frijda, 1994; Levenson,

1994a; Plutchik, 2000 ). It logically follows that distinct ANS pat-

terns would be associated with these situations (and accompanying

emotions), since these patterns subserve appropriate behavioral

responses to the situation. PCA presents a technique for detecting

SRS ANS patterns evoked under basic emotions.

The existence of universal response patterns (SRS) does not pre-

clude individual differences in the expression of these patterns

(IRS). In fact, the construct of IRS developed prior to that of SRS: it

was hypothesized that idiosyncratic stress responses contributed

to “psychosomatic” disorders such as tension headaches or essen-

tial hypertension ( Malmo and Shagass, 1949 ). It was later shown

that healthy subjects also display IRS, particularly regarding the rel-

ative contribution of various response components to the overall

response patterns ( Lacey et al., 1953 ). In general, some degree of

SRS and IRS inﬂuences all psychophysiological response patterns

( Engel, 1960 ).

The current study operates from the perspective of SRS and IRS

as the two main sources of variability in physiological respond-

ing, namely situation driven responses and individual differences

in response style. Thus, manifest physiological responses represent

the combined effect of these two sources of variability.

3. Multivariate studies

The question of autonomic speciﬁcity of emotions hinges on

the presence of consistent response patterns between persons. To

detect such activity, it is necessary to utilize analytic methods that

are sensitive to multiple response systems ( Fridlund and Izard,

1983 ). Univariate statistics disrupt the continuity of the physiologi-

cal response patterns by parsing physiological events into separate

non-representative pieces ( Thayer and Friedman, 2000 ). Nonethe-

less, the one extant meta-analysis of ANS speciﬁcity research was

based soley on univariate studies, presumably because that liter-

ature has historically been dominated by univariate methods (cf.

Cacioppo et al., 2000 ). In contrast, PCA represents a multivariate

approach with unique advantages for detecting psychophysiolog-

ical patterns in affective contexts ( Christie and Friedman, 2004;

Fridlund et al., 1984; Kreibig et al., 2007; Nyklicek et al., 1997;

Rainville et al., 2006 ).

Pattern classiﬁcation analysis is ideal for this research question

because it affords simultaneous consideration of multiple response

variables ( Fridlund and Izard, 1983; Fridlund et al., 1984 ). In this

technique, classiﬁcation functions are generated for each of several

output classes (e.g., emotions). Cases are then assigned to classes

based on a vector of input elements (e.g., ANS responses). This

assignment involves generation of posterior probabilities for each

emotion condition for each observation; subjects are then assigned

to the emotion condition with the highest probability regardless of

the condition of origin. The success of the classiﬁcation functions,

and thus the discriminability of the classes based upon respective

input elements, is evaluated by testing the percentage of correct

classiﬁcations against chance level using a standardized normal test

statistic ( Huberty, 1994 ). This approach represents the appropri-

C.L. Stephens et al. / Biological Psychology 84 (2010) 463–473

465

In the same sense that a multivariate technique like PCA is nec-

essary to fully capture the presence of ANS response patterns during

emotional states, an instantiation of SRS, we believe the illustration

of IRS likewise requires an analytic tool sensitive to response pro-

ﬁles. Cluster analysis is one such multivariate technique. Cluster

analysis is a multivariate technique that has been used to quantify

IRS in cardiovascular responses to laboratory stressors ( Allen et al.,

1991 ). This use of cluster analysis is based on the premise that, even

with a modest number of physiological measures, a relatively small

number of hierarchical response proﬁles (i.e., individual response

styles) are likely to exist upon which subgroups of subjects can be

formed. If a subset of individuals exhibit similar response patterns,

consistently distinguishing them from other individuals across con-

ditions, then it can be said that the individuals exhibit similar

patterns of IRS. A strong tendency for subjects to form distinct car-

diovascular response clusters across diverse stressors was found by

Allen et al. (1991) , thus illustrating IRS. The particular relevance of

this study is that it demonstrates the utility of a cluster analysis for

identifying the degree of IRS in ANS responding across diverse sit-

uations. As such, the cluster analytic strategy of Allen et al. (1991)

was modeled in the present study.

6.2. Apparatus

6.2.1. Music clips

Music clips that were piloted during the selection phase were

used to elicit the discrete emotions: amusement, anger, con-

tentment, fear, sadness, surprise, and a relatively neutral state

( Stephens et al., 2007 ). The selection criteria for the musical pieces

were (a) maximal response on the discrete emotion item from self-

report, (b) a small standard deviation in self-reported affect, and

activation. Two musical pieces for each emotion were employed

during the experimental phase of the study; six of these pieces

were the same as those used in Nyklicek et al., 1997 (see Appendix

A for list of music stimuli). The clips varied in length ranging

from 69 to 149 s, with an average length of 113 s with loudness

within the music clips varying between 50 and 90 dB and 70 dB

for the “white” noise clip (125 s). Music clips and “white” noise

were presented through noise-canceling headphones (Sony model

#: MDR-NC6)

6.2.2. Film clips

Standardized ﬁlm clips were presented to elicit the discrete

emotions amusement, anger, contentment, fear, sadness, surprise

and a relatively neutral state ( Fredrickson and Levenson, 1998;

Gross and Levenson, 1995; Rottenberg et al., 2007 ; see Appendix

A for list of ﬁlm stimuli). The clips varied in length ranging from

61 to 247 s, with an average length of 145 s and were presented

on a 17-in. desktop computer monitor approximately 1.5 ft from

the subject. The washout audio/video piece presented before each

music and ﬁlm clip consisted of 65 s of repeating colored vertical

“screen test” bars. This was created using the neutral noncommer-

cial clip standardized by Gross and Levenson (1995) .

5. The present study

The primary aim of the present study was to utilize two mul-

tivariate techniques, PCA and cluster analysis, to assess SRS and

IRS, respectively, in regard to autonomic speciﬁcity of emotion. A

secondary aim was to include two emotion induction techniques

to address the issue of whether ANS patterns are a function of

the emotion elicitation context ( Stemmler, 1989 ). It was predicted

that PCA would reveal statistically signiﬁcant hit rates for each

induced emotion across both techniques, providing further sup-

port for autonomic discrimination of basic emotions, and directly

replicating the results of Christie and Friedman (2004) and Nyklicek

et al. (1997) . An exploratory cluster analysis was also performed to

examine the degree of IRS in response to the emotion inductions.

Toward these ends, a montage of ANS responses were acquired

in subjects who viewed affective ﬁlm clips and listened to music

excerpts selected for their emotional qualities.

6.2.3. Self-report questionnaire

A 23-item affect self-report scale was completed electronically

via MediaLab questionnaire presentation software immediately

following each emotion elicitation (ASR; modiﬁed from Christie and

Friedman, 2004 , and Nyklicek et al., 1997 ). Subjects were instructed

to “Select the number on the scale that best describes how you felt

during the music/ﬁlm clip that you just listened to/viewed. If the

word does not at all describe how you felt during the clip, select

1. If the word very accurately describes how you felt, select 7, for

an intermediate amount, select 4, etc.” This computerized question-

naire was completed using the right-handed mouse of the stimulus

presentation computer. The ASR contained items in accord with

both discrete (content, amused, fearful, angry, sad, and neutral) and

dimensional (good, bad, positive, negative, calm, agitated, pleasant,

unpleasant, passive, active, relaxed, excited, and indifferent) mod-

els of affect, and so it matches well with a hybrid model of affective

space. The ﬁnal three questions on each ASR were regarding level

of intensity and enjoyment as rated on a 7-point Likert scale and

familiarity rated either ‘yes’ or ‘no’.

6. Method

6.1. Subjects

Fifty undergraduates (27 women and 22 men, M= 19.3, SD = 1.3

years, range = 18-26 years) were recruited using an online exper-

iment management system and received course credit for their

participation. One subject was excluded from the analysis due to

technical difﬁculties and loss of physiological and self-report data.

It has been suggested that depression and/or alexithymia may dis-

tort responses to emotional stimuli ( Christie and Friedman, 2004 ),

so subjects were screened using the Beck Depression Inventory-

II (BDI-II; Beck et al., 1996 ) and the Toronto Alexithymia Scale

(TAS-20; Bagby et al., 1994a,b ). Cutoffs of 19 on the BDI-II a, the

upper limit for mildly depressive symptomatology ( Beck et al.,

1996 ; sample mean = 4.5, SD = 4.8; range: 0–18), and 51 on the

TAS-20, the lower limit for identiﬁcation of alexithymia ( Bagby

et al., 1994a,b ; sample mean = 47.8, SD = 6.9; range: 30–50) were

employed. Subjects were likewise excluded if they indicated a his-

tory of cardiovascular or neurological problems, were currently

taking medication for hypertension, depression, or anxiety, or were

smokers. Subjects were instructed to abstain from consuming caf-

feine and/or alcohol for at least 12 h prior to the study. This study

received approval from the Institutional Review Board at Virginia

Tech.

6.2.4. Physiological recording equipment

Physiological signals were acquired using pre-gelled electrodes

(Surtrace; ConMed Co., Utica, NY) and attachment sites were pre-

pared using 70% isopropyl alcohol. Electrocardiogram (ECG) was

recorded using a BIOPAC ampliﬁer (ECG100C; BIOPAC Systems

Inc., Goleta, CA) with thoracic electrodes placed in a Lead II con-

ﬁguration. Impedance cardiogram (ICG) was recorded using the

Minnesota Impedance Cardiograph (Instrumentation for Medicine,

Minneapolis, MN) with a four spot electrode array as per Sherwood

et al. (1990) . Skin conductance level (SCL) was recorded using an

isolated SC coupler (V71-23; Coulbourn Instruments, Allentown,

PA) with electrodes placed on the thenar and hypothenar emi-

nences of the left palm ( Dawson et al., 2007 ). Respiratory signals

466

C.L. Stephens et al. / Biological Psychology 84 (2010) 463–473

were recorded from thoracic and abdominal sites using two aneroid

chest bellows connected to resistive bridge strain gage couplers

(V94-19 and V71-23; Coulbourn Instruments, Allentown, PA). All

signals were digitized at 1000 Hz using a Biopac MP100 Data Acqui-

sition System (12-bit resolution). An IBS SD-700A automated BP

monitor (Industrial and Biomedical Sensors Corp., Waltham, MA)

was used to measure systolic (SBP) and diastolic blood pressures

(DBP) approximately halfway into each epoch via an automated

auscultatory method at the upper left arm (level with the subject’s

heart). Subjects were instructed to limit the movement of their left

arm during the experiment to minimize movement artifact in SCL

and BP measures.

analysis software. As mentioned previously all efforts to correctly

score dZ/dt min amplitude and measured-LVET were made. One lim-

itation to be noted was the use of spot electrodes which pick up only

half of the impedance measured by band electrodes, thus the SVs

(and COs) were monitored for supraphysiological values.

Respiratory signal for each subject was analyzed using the

AMSRES program, a component of the VU-AMS software package.

This software was used to view and analyze data recorded with

BIOPAC hardware/software: ECG, and thoracic respiration strain-

gauge recording. The AMSRES program computed respiration rate

(RR), inspiration time (TI), and expiration time (TE) via an automatic

scoring algorithm for detecting the beginning and end of inspiration

and expiration. The AMSRES program used the R-wave to R-wave

intervals time series of the ECG, to extract two IBIs per breath: the

shortest IBI and the longest IBI. The former during a prolonged

inspiration interval, starting at the beginning of inspiration and

ending 1000 ms after the end of inspiration and the latter during

a prolonged expiration interval, starting at the begin of expiration

and ending 1000 ms after the end of expiration.

Respiratory Sinus Arrhythmia (RSA) or heart period variability

associated with respiration was derived according to the peak-

trough method ( Grossman et al., 1990 ). RSA was calculated by

the subtraction of the shortest from longest IBI, provided that the

shortest IBI (highest HR) was part of an accelerating series and the

longest IBI (lowest HR) was part of a decelerating series. If either the

longest or shortest IBI was missing for a breath cycle, or a negative

value was obtained on subtraction, the RSA value was disregarded

(denoted as NA) from further analyses.

From the ECG, ICG, BP, respiration, and SCL tracings mean values

were taken of the last 60-s periods of each condition (washout and

stimulus) for each of the following autonomic variables: IBI, mean

heart rate (mHR), peak-trough respiratory sinus arrhythmia (RSA),

square root of the mean squared successive differences in heart

period (rMSSD), cardiac pre-ejection period (PEP), LVET, SV, CO,

SBP, DBP, mean arterial pressure (MAP), total peripheral resistance

(TPR), RR, TI, TE, and SCL.

Due to differential lengths of music clips and ﬁlm clips, change

scores were calculated by subtracting the mean scores from the

last 60 s of the preceding washout period from the mean scores of

the last 60 s of the music/ﬁlm stimulus for each autonomic vari-

able (emotion induction minus baseline). Absolute values of SV,

CO, and TPR are unreliable for inter-subject comparisons ( Smith

and Kampine, 1990 ). Therefore, responses for these variables were

deﬁned as percent-change scores between stimulus condition and

the preceding washout condition. The following variables were

used for the PCA: IBI (ms), RSA (ms), PEP (ms), LVET (ms), SV (%

6.2.5. Quantiﬁcation of physiological data

ECG, ICG, and respiration data were processed using the

Vrije Universiteit-Ambulatory Monitoring System software suite

(VU-AMS, Vrije Universiteit, Department of Psychophysiology,

Amsterdam, The Netherlands) and heart rate variability (HRV)

was processed using HRV Analysis Software v1.1 (The Biomedical

Signal Analysis Group , Department of Applied Physics, University

of Kuopio, Finland). SCL data were analyzed using the BIOPAC Acq-

Knowledge software (BIOPAC Systems Inc, Goleta, CA) ( BIOPAC

Systems Inc, 2005 ).

Artifact detection was performed by software-assisted visual

inspection of physiological signals, corrections were made using

BIOPAC AcqKnowledge software when necessary. For heart rate

variability analysis, an R-wave detection function converted the

ECG to a time series of inter-beat-intervals (IBIs) using detection of

the R-waves in the ECG. A graphical display of the IBI time series by

the HRV Analysis Software permitted detection of possible ectopic

beats or other deviations from physiologically plausible values. All

beats deviating more than 30% from the previous or subsequent

beat were considered suspect and ECG was examined ( deGeus et

al., 1995 ). Excessively short/long beats were corrected by removing

movement artifact from the ECG.

Impedance cardiogram for each subject was analyzed using the

AMSIMP program, a component of the VU-AMS software package.

This software was used to view and analyze ICG recorded with

BIOPAC hardware/software. The program contains an automatic

scoring algorithm for detecting the upstroke (B-point), dZ/dt min ,

and incisura (X-point) in each ICG complex. The B-point or upstroke

was deﬁned as a ﬁrst or second order zero-crossing in the dZ/dt sig-

nal, close to the dZ/dt isoelectric line, and the starting point of the

longest uphill slope before the dZ/dt min point. The dZ/dt min was

deﬁned as the highest point of the ICG complex (y-axis inverted)

between the B- and the X-point. The X-point or incisura was deﬁned

as a local minimum after the dZ/dt min ; this is often but not necessar-

ily, the lowest point in the entire signal. Due to the nature of the ICG

signal quality, the automatic scoring algorithm was used to provide

preliminary values. When artifact was present the automatic scor-

ing algorithm provided a guideline for choosing the components of

interest manually via the software interface.

Every effort was made to ensure that ICG recording, and scoring

was performed in accordance with the standards in the literature

( Sherwood et al., 1990; Turner, 2000 ). The cardiac pre-ejection

period (PEP) was deﬁned as the time in milliseconds (ms) between

ECG Q-wave onset and B-point in the ICG. Left ventricular ejection

time (LVET) was deﬁned as the time in ms between B- and X-points

in the ICG. It should be noted that the reliability of stroke volume

(SV) and cardiac output (CO) computed from impedance cardiog-

raphy remains a debated issue (cf. Sherwood et al., 1990; Turner,

2000 ). Steps were taken to limit the error attributed to parame-

ters that inﬂuence the calculation of SV using the Kubicek formula

( Kubicek et al., 1966 ). A constant of 135

CO (%

), SBP (mmHg), DBP (mmHg), MAP (mmHg), TPR (%

), RR

(bpm), TI (ms), TE (ms), and SCL (

S). A smaller subset of vari-

ables was used for the exploratory cluster analysis in line with

previous research ( Allen et al., 1991 ): mHR (bpm), rMSSD (ms), PEP

(ms), SV (%

), SBP (mmHg), DBP (mmHg), MAP (mmHg), and TPR

6.2.6. Procedure

All experimental sessions were conducted in a sound atten-

uated room with subjects seated in a comfortable chair and the

experimenter in a separate room observing via a two-way mirror.

Experimenter-subject communication was possible via an audio

intercom. Stimuli were presented in one of two sequences with

music pieces and ﬁlm clips alternating in a partially counter-

balanced fashion. Consistent with Christie and Friedman (2004) ,

negatively valenced ﬁlm clips alternated with positively or neu-

trally valenced ﬁlm clips. The two sequences of presentation were

the reverse of each other. Each emotion was evenly distributed

throughout the entire experiment and the same emotion induction

was not experienced two times in a row.

*cm was used for speciﬁc

blood resistance (rho b ), and the distance between the measuring

ICG electrodes was recorded for each subject and input into the

C.L. Stephens et al. / Biological Psychology 84 (2010) 463–473

467

Subjects were instructed to pay close attention to how they felt

as they watched the ﬁlm clips and listened to the music pieces, and

were told that following each clip they would be asked to describe

how they felt. After a 10-min laboratory adaptation period, a 1-min

baseline recording, during which subjects were presented with the

1-min and 5-s washout clip, conﬁrmed proper equipment function-

ing and subjects completed a baseline ASR followed by the ﬁrst clip.

After each music/ﬁlm stimulus, subjects completed the ASR scale

to assess affective response during the presentation. Upon comple-

tion of the scale, subjects were presented with the same 1-min and

5-s washout clip, during which they were instructed to sit quietly

and clear their mind of all thoughts, feelings, and memories. The

next stimulus presentation then commenced. This procedure was

repeated for the remaining stimuli conditions in a manner closely

approximating that of both Nyklicek et al. (1997) and Christie and

Friedman (2004) .

Pattern classiﬁcation using ANS variables revealed that overall

37.6% of observations were correctly classiﬁed into the predicted

emotion conditions during music stimuli, whereas 40.2% of obser-

vations were correctly classiﬁed during ﬁlm induction. As depicted

in the diagonals of Table 2 , classiﬁcation hit rates for individual

emotion conditions ranged from 22.4% to 59.2%. The overall clas-

siﬁcation hit rate, indicating overall classiﬁcation success, was sig-

niﬁcantly greater than chance for both music induction (z = 12.34,

p < .0.001) and ﬁlm induction (z = 13.73, p < .001). The 95% conﬁ-

dence interval for the probability of correct classiﬁcation for music

and ﬁlm induction is 0.325 to 0.427 and 0.350 to 0.454 respectively.

Pattern classiﬁcation using ASR variables revealed that overall

55.1% of observations were correctly classiﬁed into the predicted

emotion conditions during music stimuli, whereas 59.5% of obser-

vations were correctly classiﬁed during ﬁlm induction. As depicted

in the diagonals of Table 3 , classiﬁcation hit rates for individual

emotion conditions ranged from 26.5% to 89.8%. The overall clas-

siﬁcation hit rate, indicating overall classiﬁcation success, was sig-

niﬁcantly greater than chance for both music induction (z = 21.60,

p < .001) and ﬁlm induction (z = 23.92, p < .001). The 95% conﬁdence

interval for the probability of correct classiﬁcation for music and

ﬁlm induction is 0.498–0.604 and 0.542–0.647 respectively.

7. Results

7.1. Univariate analysis of discrete ASR variables—manipulation

check

Paired t-tests were used to explore the effectiveness of the

affect manipulations with respect to the discrete ASR items. For

the amusement, contentment, fear, sadness, and surprise emotion

conditions, ratings on discrete item for the respective conditions

were signiﬁcantly greater than all other discrete items (all ps < .05).

Means and standard deviations for all discrete ASR variables can be

found in Table 1 . No statistically signiﬁcant differences were found

for responses on the discrete or dimensional ASR items with regard

to the subject gender or the presentation sequence of the emotion

inductions.

7.3. Pattern classiﬁcation using variables from aggregated music

and ﬁlm inductions

Overall, 44.6% of observations were correctly classiﬁed into the

predicted emotion conditions. Hit rates for individual emotion con-

ditions ranged from 32.7% for sadness to 63.3% for surprise. The

overall classiﬁcation hit rate occurred at signiﬁcantly greater than

chance levels (z = 16.05, p < .001). Individual emotion conditions

were successfully classiﬁed at signiﬁcantly greater than chance

levels: amusement (z = 6.94, p = .001), anger (z = 4.49, p < .001), con-

tentment (z = 6.94, p < .001), fear (z = 4.90, p < .001), neutral (z = 5.72,

p < .001), sadness (z = 3.67, p < .001), and surprise (z = 9.80, p < .001).

Distinct patterning of physiology can be seen across emotion con-

ditions ( Fig. 1 ).

The overall ASR classiﬁcation hit rate was 60.6% with individual

emotion condition hit rates ranging from 36.7% for fear to 71.4%

for amusement. All hit rates, both overall classiﬁcation (z = 23.31,

p < .001) and each of the emotion conditions were signiﬁcantly

7.2. Comparison of music and ﬁlm induction techniques

The most informative means to compare across music and ﬁlm

emotion inductions techniques is via hit rates from pattern classi-

ﬁcation analysis. An initial component of this study was to monitor

both self-report and autonomic responses to music and ﬁlm within

the same sample to allow for the best possible conditions for com-

parisons.

Table 1

Mean (SD) of ASR variables by emotion condition across emotion induction techniques.

Emotion condition

3.43 (2.02)

2.69 (2.09)

2.13 (1.6)

2.54 (1.81)

1.79 (1.28)

1.81 (1.35)

2.38 (1.76)

1.59 (1.3)

3.5 (2.15)

1.56 (1.15)

2.69 (1.91)

3.22 (2.23)

2.19 (1.62)

2.32 (1.67)

1.16 (0.6)

3.41 (1.87)

1.36 (0.74)

2.82 (1.9)

1.5 (1.05)

2.88 (1.62)

2.00 (1.54)

3.72 (1.62)

2.09 (1.37)

4.68 (1.64)

1.99 (1.36)

3.98 (2.06)

3.23 (1.72)

2.49 (1.61)

3.83 (2.04)

2.56 (2.01)

2.23 (1.66)

2.66 (2.02)

1.52 (1.12)

1.93 (1.39)

2.54 (1.94)

5.01 (1.45)

1.93 (1.2)

4.19 (1.66)

1.78 (1.21)

2.38 (1.55)

2.38 (1.36)

2.11 (1.37)

Dimensional variables

1.31 (0.91)

3.27 (1.88)

1.55 (1.09)

2.8 (1.86)

2.23 (1.6)

2.73 (1.63)

2.27 (1.56)

2.33 (1.54)

2.22 (1.51)

2.72 (1.96)

2.72 (1.74)

2.4 (1.76)

2.68 (1.77)

2.41 (1.68)

4.16 (1.93)

2.05 (1.57)

4.11 (1.8)

2.12 (1.72)

2.16 (1.52)

2.14 (1.45)

1.97 (1.36)

4.27 (1.88)

1.99 (1.53)

3.77 (2.05)

1.98 (1.53)

1.99 (1.44)

1.93 (1.37)

1.95 (1.28)

1.45 (1.07)

3.54 (2.04)

1.73 (1.39)

3.26 (2.05)

2.8 (2.07)

2.96 (1.76)

2.63 (1.79)

1.45 (1.07)

3.54 (2.04)

1.73 (1.39)

3.26 (2.05)

2.8 (2.07)

2.96 (1.76)

2.63 (1.79)

am 4.59 (2.25) 1.15 (0.56) 3.99 (1.78) 1.33 (0.84) 2.41 (1.75) 1.05 (0.25) 2.45 (1.79)

an 2.43 (1.97) 3.12 (1.9) 2.26 (1.52) 2.67 (1.62) 2.26 (1.66) 3.15 (1.4) 2.84 (1.85)

co 2.29 (1.67) 1.31 (0.67) 4.61 (1.9) 1.5 (1.05) 3.34 (1.96) 1.63 (1.08) 1.95 (1.31)

fe 2.16 (1.64) 1.67 (1.11) 2.49 (1.86) 3.59 (2.02) 2.42 (1.73) 1.73 (1.2) 2.68 (1.8)

ne 1.88 (1.29) 1.83 (1.37) 2.31 (1.58) 2.07 (1.9) 3.52 (2.24) 1.07 (0.24) 1.86 (1.54)

sa 2.02 (1.49) 1.68 (1.22) 2.53 (1.62) 2.09 (1.51) 2.56 (1.73) 4.26 (1.63) 1.87 (1.37)

su 2.1 (1.54) 1.51 (1.14) 2.33 (1.46) 3.01 (1.77) 2.77 (1.83) 1.41 (0.88) 4.29 (2.24)

Abbreviations: active = ac, agitated = ag, bad = ba, calm = ca, excited = ex, good = go, negative = neg, passive = pa, pleasant = pl, positive = po, relaxed = re, unpleasant = un, amuse-

ment = am, anger = an, contentment = co, fear = fe, neutral = ne, sadness = sa, surprise = su.

Discrete variables

Plik z chomika:

Inne pliki z tego folderu:

Inne foldery tego chomika: