Spectral Analysis of Candidates' Nonverbal Vocal Communication: Predicting U.S. Presidential Election Outcomes
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Social Psychology Quarterly
2002, Vol. 65, No. 3,298-308
Spectral Analysis of Candidates' Nonverbal Vocal
Communication: Predicting U.S. Presidential Election
Outcomes*
STANFORD W. GREGORY JR.
TIMOTHY J. GALLAGHER
Kent State University
Past Pourier Transform acoustic analysis of the fundamental frequency of candidates'
voices in 19 nationally televised U.S. presidential debates from the eight elections
including debates held since 1960, in conjunction with subsequent factor analysis,
shows that this nonverbal frequency, below .5 kHz, can reveal the debating candidates'
relative social dominance. Purther analysis presents evidence that the candidates' non-
verbal vocalizations offer a precise metric of their relative dominance or commanding
presence in the presidential campaign: when this metric is compared statistically with
the candidates' popular vote percentages for the U.S. presidency, it accurately predicts
the popular vote outcomes in all of those eight elections.
In previous research by the senior author beneath .5 kHz, is unique in the vocal adapta-
and colleagues (Gregory 1983, 1986), inter- tion literature.
view partners were found to adapt frequen- The Fg is a critical component of human
cies of their voices to one another in the vocalization. When the voice is filtered elec-
course of their interaction. Subsequent work tronically, however, allowing only the F^ to
has refined the theory and method in this line pass, the resulting sound is perceived as a
of research by showing that the acoustic low-pitched and segmented hum absent of
adaptation phenomenon resides beneath .5 any clearly discernible verbal content. When
kHz, otherwise known as the fundamental experimental subjects' vocal frequencies
frequency of phonation (Fg) and perceived as beneath F^ are filtered from the normal con-
pitch in the vocal spectrum (Gregory 1994; versational signal (leaving behind a crisp,
Gregory, Dagan, and Webster 1997; Gregory clear verbal signal), the perceived quality of
and Webster 1996; Gregory, Webster, and interaction, as evaluated by outside judges
Huang 1993). Although the adaptation phe- hearing an unfiltered version of the experi-
nomenon in conversations and interviews has mental subjects' conversational exchange, is
been reported widely by linguistic, social psy- diminished significantly (Gregory et al. 2000;
chological, and communication researchers Gregory et al. 1997). Thus it is evident from
(Burgoon, Dillman, and Stern 1993; Burgoon, previous studies that the low-frequency band
Stern, and Dillman 1995; Capella 1981; beneath F^ is crucial for communicating criti-
Chappie 1940; Jaffe and Feldstein 1970;
cal social information.
Natale 1975), the method used, employing
Analysis of the FQ in interacting partners'
spectral analysis of the frequency band
voices also indicates partners' relative social
status (Gregory et al. 2000; Gregory and
* This research was funded by the Department of Webster 1996). This observation fits well with
Sociology, Kent State University, by the Office of Giles and Coupland's (1991a, 1991b) commu-
Research and Graduate Programs, Kent State
University, and by an equipment grant from Zonic nication accommodation theory, known as
Corporation, New Milford, OH. We are grateful to CAT, whereby persons of lower social status
Mr. Paul M. Gherman of the Vanderbilt University accommodate their nonverbal vocal patterns
Television News Archive for supplying the video to persons of higher status; the latter modify
tapes of the presidential debates used in this research.
Direct correspondence to Stanford W. Gregory Jr., their vocal patterns relatively little. CAT was
Department of Sociology, Kent State University, used in previous research (Gregory and
Kent, OH 44242; sgregory@kent.edu. Webster 1996) to explain the acoustic accom-
298NONVERBAL VOCAL COMMUNICATION 299
modation patterns and relative social status researchers studying emotion in human com-
of 25 celebrity guests on the Larry King Live munications often echo the observation
television talk show. In that study, Mr. King made by ethologists and others that vocal
vocally accommodated persons of higher encoding and decoding behavior in primates
social status than himself, and persons of (for example) is remarkably similar in some
social status lower than Mr. King accommo- respects to that of humans. In keeping with
dated their vocal patterns to him. the subject of this paper, such behavior is
The general point of GAT in connection most often connected with threats and domi-
with spectral analysis of Fg is that persons nance displays (Mazur 1985; Sapolsky 1990).
who occupy higher social status adjust their In connection with human vocalics, Mazur
overall vocal patterns less over time, while (1985:392) supports CAT in noting that "the
those of lower social status adjust their vocal high status person sets the pace and mood of
patterns more to accommodate persons of the conversation, and the low status person
higher social status. In this paper, on the basis follows."
of 19 televised presidential debates held Burgoon, in examining the channels of
between 1960 to 2000, we present analyses of nonverbal communication in relation to
data testing the association between the pop- dominance and social status, maintains that
ular vote, other conventional measures of "the voice is more influential than visual cues
candidate preference, and the relative social in determinations of dominance. This may be
status of U.S. presidential candidates, as mea- because it is an innately used and recognized
sured by Fg analysis. These results can offer signal" (Burgoon et al. 1996:316). More
an objective measure of dominance in the specifically, and most compelling for the pre-
presidential election that is not biased by sent research, Burgoon et al. (1996:399-403),
semantic/verbal or partisan interpretation. in their survey of literatures on nonverbal
persuasive behaviors and strategies that pro-
THEORETICAL CONNECTION ject power and credibility, cite the quality of
BETWEEN NONVERBAL VOCAL "more variety/intonation" as a key factor in
COMMUNICATION AND POLITICAL "actually achieving persuasion." This obser-
DECISIONS vation has been supported in previous FQ
analysis (Gregory and Webster 1996):
Nonverbal Vocal Communication
celebrities with the highest social status
In early work on nonverbal communica- showed the greatest variation in intonation
tion, Ekman and colleagues (Ekman 1965; (variation in amplitude between frequencies
Ekman and Friesen 1969) used the plausible within the Fg band), to which less dominant
working assumption that nonverbal commu- celebrities accommodated.^
nication channels operate under less con-
scious control and thus will reveal the ' An anonymous reviewer for this article comment-
speaker's true message or feeling to a greater ed on the question of "transiency versus constancy" in
extent than more consciously controlled the quality of a speaker's vocal interaction. In the pre-
sent article we address transiency in that one candi-
channels. This assumption, connecting non- date may adapt or accommodate his vocal qualities to
verbal communications with autonomic func- the more dominant partner. Constancy was addressed
tions and "truth," carries through in Ekman's in a previous study (Gregory and Webster 1996), in
later work on deceit and continues in later which the authors, using a factor analysis technique,
works by others on this subject. Burgoon, ranked the Fg vocal attributes in a sample of 25
celebrity guests on the talk show Larry King Live into
Buller, and Woodall (1996), for example, in an order of factor loads from more dominant to less
their popular text on nonverbal behavior, dominant guests. This rank ordering of factor loads
maintain the nonverbal/emotional and ver- was compared with a ranking of the same celebrities
bal/conscious connections while updating by a sample of 596 undergraduate students; the result
was a strong Pearson correlation between the F^
Ekman's perspective, whereby nonverbal results and the students' rankings. The higher factor
communications encode (produce) and loads for dominant guests showed a constancy quality
decode (interpret) the verbal message. of F(| that was not present for less dominant guests. In
The procedures of encoding and decod- other words, that research provided evidence that
more dominant guests produced similar F,, spectra, or
ing, of course, embrace content, and300 SOCIAL PSYCHOLOGY QUARTERLY
In addition, in summarizing the strate- these perceptions are constructed in the
gies that project power and credibility, course of the campaign by a number of con-
Burgoon et al. (1996:399) write that "'confi- ventional attributes such as partisanship,
dent' voices evoke more compliance from issue proximity, personality and character
others. Such vocal patterns have proved to be assessments, campaign funding, and media
persuasive in public and interpersonal con- coverage (Rahn, Aldrich, and Borgida 1994).
texts." This observation has been supported These attributes culminate and are signified
in two Fg studies conducted by the senior by the polls taken at various points before
author and colleagues. In these studies, the the election;^ it is obvious that a strong show-
subjects' voice signal was altered so that their ing in the polls influences the candidate's
interaction partners did not, in all cases, hear self-esteem and his perceptions of his person-
the subjects' voice signal in its acoustic al status. The significance of these numerous
entirety. An acoustically unaltered recording conventional attributes, when positive, fills
of these subjects' voices was made to use as a candidates with confidence in their chances
stimulus for other subjects to hear. for election.
In the first study (Gregory et al. 1997), The presidential debates thus showcase a
some conversations included the complete candidate's status in the campaign. If he can
acoustic signal; others were filtered by elimi- maintain a solid standing throughout the
nating the FQ frequency and leaving the debates, success is more likely in the election.
remainder of the spectrum above .5 kHz; still Although the debate is structured differently
others included the FQ plus another .5 kHz, from one election year to the next, it is an
creating a 1.0 kHz band. Subjects listening to interactive event in which opponents com-
these three types of conversation, but with- pete to gain the upper hand. Of course, man-
out any filtering, rated them; the results agement of a dominant presence is
showed that the presence of the F^ was influenced by the candidates' many experi-
important in conveying a social dominance ences before the debates, as noted above, as
signal. well as by each candidate's personal attribut-
In the second study, the researchers es (Rosenberg and McCafferty 1987). The
(Gregory et al. 2000) used a similar method: debate, however, also offers a unique display
outside subjects evaluated conversations that of other features that are less clearly defined
had been altered acoustically. The results and are generated interactively by the candi-
showed the primacy of the vocal over the dates. Therefore other less conventional and
visual signal in the communication of social unconsciously derived attributes also can
status and dominance via the .5 kHz band. come into play.
One such important and less consciously
Dominance in the Presidential Debates derived attribute resides in the candidates'
nonverbal vocalizations. Studies on the influ-
The U.S. presidential debates are sched- ence of the voice in power relations have
uled to be held from one to six weeks before shown that well-known political leaders
the election. By the time the debates are con- "exude power" in their voices (Ng and
ducted, the candidates usually have been
campaigning for at least a year; as a result,
both the candidates and the electorate have
formed moderately solidified perceptions ^ The debate polls used in this research were
derived from a number of sources for the eight elec-
about the participants' relative statuses in the tion years analyzed. Final statistical means from these
campaign. Candidates' and voters' percep- sources were used in the analysis and are cited in
tions about participants in the presidential Table 2 of this report. The polls in summary form are
campaign include notions about the relative found in The Atlanta Constitution, October 17 1966;
dominance of one candidate over another; The Gallup Organization, www.gallup.com/trends/
phistory.asp, 1936-1996; September 28, October 2,14,
18, November 1,1976; The New York Times, October
what may be called a dominance spectrum. In the pre- 16, 21, 25, 1984; September 27, October 15, 1988;
sent article, however, we compare only one candidate October 12,13,1992; The San Diego Tribune, October
with another, and thus consider a transiency rather 10, 1996; USA Today, October 1, 10, 17, 1996; U.S.
than a constancy quality of the F^. News and World Report, October 31,1960.NONVERBAL VOCAL COMMUNICATION 301
Bradac 1993:12). This quality of confidence (Gregory and Webster 1996), and, second, to
or social dominance, as noted above in show how this metric is related to another
Burgoon's survey of nonverbal persuasive indication of a candidate's social dominance,
behaviors (Burgoon et al. 1996), is communi- namely the popular vote. Thus if a candidate
cated openly in candidates' voices. The rela- demonstrates a commanding presence in the
tive amounts of dominance or social status debates on the basis of Fg analysis, this simply
shown in candidates' vocalic cues are negoti- means that this candidate, possibly because
ated in the context of the debate: according of confidence due to previous campaign suc-
to CAT, a less dominant candidate will cess, an opponent's blunder, or even some
accommodate vocalic cues to a more domi- innate charismatic personal quality, has a cer-
nant candidate without conscious awareness tain voice characteristic that distinguishes
of doing so. him from his opponent. It does not mean that
Although this social status negotiation this voice quality, as associated with the can-
between actors was most commonly associat- didate's possible success in the debate, will
ed with conversations and interviews in pre- lead to success with the electorate by causing
vious FQ analysis research, it is an even more observers of the debates to vote for him. The
important part of debates because the debate Fg analysis of vocalic cues is simply and only
is conditioned by interaction between partic- a metric of something found in previous
ipants concerned with contentiously held research, which proclaims one participant's
claims about critical issues. The debate is an social dominance over another. If this metric
interactive contest in which opponents are is associated with another indicator of social
influenced by one another. Ultimately, as dominance, such as a higher percentage of
shown in previous acoustic studies (Gregory the popular vote, such a finding is further val-
and Webster 1996), the less dominant partner idation of Fg analysis as a robust metric of
will accommodate vocalic cues to the more social dominance.
dominant partner.
Previous spectral analysis of interacting METHOD
partners has shown that the more dominant
partner can be distinguished from the less For this study we analyzed 19 debates
dominant. We now apply the appropriate from eight U.S. presidential elections:^
techniques to the debate context to deter- Kennedy/Nixon (1960), Carter/Ford (1976),
mine, first, whether the dominant partner in Reagan/Carter (1980), Reagan/Mondale
the debate can be distinguished from the less (1984), Bush/Dukakis (1988), Clinton/Bush
dominant partner, and, second, whether the (1992), Clinton/Dole (1996), and Bush/Gore
dominant partner is associated with a higher (2000).'' The audio portion of videotaped ver-
percentage of the popular vote in the elec-
tion. Thus the spectral analysis of Fg in the
3 The 1996 and 2000 presidential debates were
debate interaction can serve as a metric of recorded by the author; the others were rented from
the debating candidates' relative dominance. the film library at Vanderbilt University. Our sample
Then the validity of this metric can be cor- includes all the presidential elections since the incep-
roborated further by showing a solid rela- tion of televised presidential debates in 1960. No
tionship with the final test of social presidential debates were held in 1964,1968, or 1972.
In 1964, Lyndon Johnson's poll position in the presi-
dominance in the campaign: the popular dential race eliminated the need for a debate with
vote. Barry Goldwater. In 1968, Richard Nixon believed he
The debates themselves and the nonver- could avoid debating with Hubert Humphrey
bal vocal cues given off by candidates are not because in early September he was leading his oppo-
nent by as much as 12 points in the Gallup poll. In
presented here as a causal agent prompting 1972, Nixon felt no pressure to debate with George
observers of the debates to eventually vote McGovern, who was "politically wounded by his mis-
for one candidate over the other. We simply steps" (Germond and Witcover 1979:193). All debates
offer this acoustic analysis of candidates' inciuded in the sample for this paper took place one
to six weeks before election day.
vocalic cues: first, as an indicator of a candi-
'' At this writing, with nearly 100 percent of the
date's dominating presence in the debate, as popular vote counted, former Vice-President Al Gore
shown in previous FQ analysis research had a 539,947 margin in the popular vote over his302 SOCIAL PSYCHOLOGY OUARTERLY
sions of these 19 debates were analyzed as ties (on the Y-axis). We used identical FFT
follows. Nine voice samples, distributed settings for all samples.^
equally over the course of the debate, were The LTAS were transferred from the
taken from each candidate. We analyzed FFT analyzer to a PC for SPSS analysis. To
these samples, which lasted approximately six simplify analysis, we concentrated the nine
seconds each, using a dedicated instrument, LTAS samples for each candidate into three
the Fast Fourier Transform analyzer (here- mean LTAS representing the beginning, the
after called the FFT analyzer, a dual-channel middle, and the end of each debate for each
Zonic Medallion FFT analyzer). candidate. A factor analysis under the com-
Spectral samples produced by FFT ponent model was performed on the LTAS.
analysis are called long-term averaged spec- We chose the unrotated principal-factors
tra (LTAS) and represent an extended aver- solution because the results are interpretable
aged speech utterance (Pittam 1987, without further rotation. We averaged factor
1994:52-58; Pittam, Galois, and Callan 1990; loads to produce the factor analysis value for
Scherer 1985).^ LTAS indicate the character- each candidate. When an election included
istic distribution of energy over the frequen- more than one debate, we averaged the fac-
cy range in a speaker's voice for a particular tor analysis results from each debate to pro-
utterance; they represent the more stable duce the acoustic analysis results for that
characteristics of a speaker's voice, indepen- election. All available videotapes of debates
dent of short-term variations such as specific for all televised elections between 1960 and
words or pauses, that could influence a result 2000 were analyzed for this research. This
too strongly. Thus we averaged unique ener- research was guided by prior theory and
gy levels over a relatively long expanse of results, which thus allow for a direct interpre-
speech, creating a result that represented a tation of the principal factors (Gorsuch
spectral spread or curve for the voice sample. 1983:175-76).
For this research, each LTAS sample is a Means of factor loads from factor analy-
spectrum composed of 300 frequency values sis of the LTAS, hereafter called acoustic
(on the X-axis) with their respective intensi- analyses, for each of the 16 candidates' in
each of the eight presidential elections were
compared with poll results preceding the
opponent, former governor of Texas (and now debates, mean poll results for each of the
President) George W. Bush. debates, the final pre-election Gallup poll
' FFT analysis is an algorithm for transforming results, and the popular vote results. We per-
data from the time domain, or the variation in ampli-
tude versus time, to the frequency domain, or the formed these correlation comparisons in the
variation in amplitude in a system versus frequency. following manner:
The FFT analyzer is an instrument used in industry, The popular vote contains a strong
for example, to detect the sounds of a faulty bearing degree of nonindependence of observation,
or gear in a noisy transmission, and in medicine to
detect a faulty heart valve among the sounds of ven-
and the polls contain some degree of non-
tricular and atrial contractions. These examples of
FFT analysis show the instrument's ability to detect a * Settings for the FFT analyzer were (1) frequency
small wave in the presence of much more powerful range between .126 and .5 kHz; (2) A/D converter 32
signals. When the components are separated through bit; (3) anti-aliasing filter set for .5 kHz; (4) spectral
FFT frequency domain analysis, the small compo- averaging set at 8 spectra per vocal sample, generating
nents are easy to see because they are not masked by 300 spectral lines in the frequency domain; (5)
the larger ones. R.N. Bracewell (1989:86), an electri- Hanning window; and (6) sampling points set at 1024
cal engineer at Stanford University, relates Fourier per frame. More highly detailed technical information
analysis to the human auditory function in his com- is available from the senior author.
ment "To calculate a transform, just listen. The ear ' Acoustic analysis values for Ross Perot were not
automatically performs the calculation, which the included for the 1992 and 1996 debates for two rea-
intellect can execute only after years of mathematical sons. First, third-party or independent candidates are
education. The ear formulates a transform by con- generally inconsequential in U.S. elections: Eugene
verting sound—the waves of pressure traveling McCarthy received only .9 percent of the popular
through time and the atmosphere—into a spectrum, a vote in 1976, John Anderson 6.6 percent in 1980, and
description of the sound as a series of values at dis- Ross Perot 18.9 percent in 1992 and 8.5 percent in
tinct pitches. The brain turns this information into a 1996. Second, none of these candidates marshaled
perceived sound." enough support to gain any electoral college votes.NONVERBAL VOCAL COMMUNICATION 303
independence: for example, if one member of tions, using Fisher's Z transformation. This
a pair of two candidates receives 45 percent transformation corrects for the nonnormal
of the popular vote, the other will receive sampling distribution of r that results in the
very close to 55 percent. Thus we chose a absence of such a correction. The conse-
resampling method for estimating the corre- quence of Fisher's correction is a normal
lation between all paired combinations of sampling distribution, from which one then
these five variables: acoustic analysis results can compute confidence intervals. Using a
(AAR), Gallup poll results before the debate sample size of 19 (i.e., 19 debates), we com-
(PRED), polls taken just after the debates puted 95 percent confidence intervals, as
(DP), the Gallup poll immediately before the reported in Table 1. Because 95 percent of all
election (GAL), and the popular vote such intervals contain the true value for p, we
(VOTE). For each of these combinations of
can reasonably conclude that p is greater
two variables, we computed all 256 (2^) possi-
than zero in three of the four measures of
ble correlations of sample size 8 in which the
two candidates from the same election were association, since zero is not contained in
not included together in the same sample. In those intervals.
this manner we eliminated the nonindepen- In Table 2 we present all of the data used
dence described above. We then computed to compute the values in Table 1. According
the estimate of correlation by calculating the to Table 2, the AAR metric predicted the
mean of the 256 correlation coefficients for popular vote outcomes in all eight elections.
each pair of variables. This metric predicted the popular vote in the
2000 election, but not the presidency.
RESULTS Although former Vice-President Gore's pop-
ular vote was .5 percent higher than that of
Comparisons of the outcomes for esti- former Texas Governor Bush, his opponent
mates of correlation between AAR and the was designated president-elect because he
four other variables cited above are present- eventually received a total of 271 electoral
ed in Table 1. Because the distributions of
votes. (According to the U.S. Constitution, a
correlation coefficients for all four compar-
isons were skewed highly negatively, as candidate receiving 270 or more votes in the
expected, we use the median values from the electoral college becomes president-elect.)
distribution of correlation coefficients (r) as This outcome was historically unusual^ and
the most accurate estimates of the true popu- could not have been predicted by the tech-
lation values p. As shown in Table 1, the esti- niques discussed in this paper.
mates of correlation for the four comparisons
are AAR by PRED, .80; AAR by DP, .37; * A winning presidential candidate received fewer
AAR by GAL, .77; and AAR by VOTE, .80. popular votes than his opponent on only one other
occasion in U.S. History: in 1876, Rutherford B. Hayes
Next we computed 95 percent confi- became president by receiving 185 electoral votes,
dence intervals for each of the four correla- while his opponent, Samuel Tilden, received 184.
Table 1. Measures of Central Tendency, Variability, and 95 Percent Confidence Interval for the Pearson
Correlation Coefficient: Acoustic Analysis Results by Other Variables'
AAR by AAR by AAR by AAR by
Statistic PRED DP GAL VOTE
Mean .77 .33 .73 .79
Median .80 .37 .77 .80
Standard Deviation .13 .29 .16 .11
Minimum .24 -.30 .10 .45
Maximum .95 .82 .91 .96
95% CI for p (.50,.94) (-.22,.74) (.45,.93) (.48,.94)
Note: AAR = acoustic analysis results: PRED = Gallup poll results before the debate: DP = polls taken just after
the debates: GAL = Gallup poll immediately before the election: VOTE = popular vote.
" Based on all possible samples of 7 where the independence assumption is not violated.304 SOCIAL PSYCHOLOGY QUARTERLY
Table 2. AAR Data. Candidate Preference Data, and Predictive Value of AAR With DP and VOTE
AAR Predictions
Debate DP VOTE
Year Candidate AAR PRED DP GAL VOTE Outcome Outcome
2000
Bush .66 .41 .45 .48 .479
Gore .74 .49 .43 .46 .484 No Yes
1996
Clinton .74 .49 .55 .52 .492
Dole .09 .36 .31 .41 .409 Yes Yes
1992
Clinton .65 .42 .40 .49 .430
Bush .11 .31 .22 .37 .375 Yes Yes
1988
Bush .78 .48 .46 .56 .534
Dukakis .71 .42 .37 .44 .457 Yes Yes
1984
Reagan .78 .58 .39 .59 .588
Mondale .15 .38 .47 .41 .406 No Yes
1980
Reagan .11 .47 .44 .47 .508
Carter .53 .39 .26 .44 .410 Yes Yes
1976
Carter .86 .48 .36 .48 .501
Ford .81 .43 .31 .49 .480 Yes Yes
1960
Kennedy .66 .49 .43 .51 .499
Nixon .62 .46 .27 .49 .498 Yes Yes
Note: AAR = acoustic analysis results; PRED = Gallup poll results before the debate; DP = polls taken just after
the debates; GAL = Gallup poll just before the election.
AAR also predicted the winners of six of win the debate contest: his mean percentage
the eight debate polls (Reagan/Mondale, for all three debates was slightly lower than
1984 and Bush/Gore, 2000 were not predict- his opponent's because of a sharp decline in
ed). It is interesting that the AAR metric cor- the second debate after winning the first, and
relates closely with the PRED results but not because he won the third by only a small mar-
so closely with the DP results. This indicates gin.
that the AAR measures follow poll results According to the conventional wisdom
based on traditional political phenomena expressed in the media before the debate.
such as partisanship and platform issues, and Gore was better at debating than Bush.
less so on respondents' ancillary verbal and Respondents tended to favor Bush, however,
consciously determined replies to pollsters' as shown by results from the polls in response
questions about who won the debates. Table 2 to the question "Regardless of which candi-
shows, for example, that Mondale clearly won date you happen to support, who do you
over a less articulate Reagan in the 1984 think did a better job in the debate?" (Moore
debate, but Reagan showed a stronger AAR 2000). According to media accounts, this
result, a stronger poll trend before the result stemmed from Bush's popular "down-
debate, and a popular vote that eventually home," sincere debating style, in contrast to
won him the electoral college vote and the Gore's less popular "smooth" style. The
election. The 2000 election is somewhat simi- important point in this connection is that
lar in this regard: Gore's AAR result was these poll responses to a candidate's perfor-
higher than Bush's as were his poll results mance in the debate draw on a respondent's
before the debate, and he won the popular conscious verbal reply to a specific question,
vote. Yet, like Reagan in 1984, Gore did not and do not show so much of an emotionalNONVERBAL VOCAL COMMUNICATION 305
response toward a candidate. The results dis- more conventional conscious and rational
played in Tables 1 and 2, however, reporting influences of the campaign, such as verbal
on the acoustic analysis of the Fg in presiden- statements, political affiliation and platform
tial candidates' voices, present outcomes of issues, demeanor, and physical attractiveness.
emotionally based dominance contests that Although our earlier theoretical expla-
have little bearing on the verbal signal. nation can explain only the relation between
In this study we do not examine whether the AAR metric and a commanding presence
this emotionally based signal influences an in the debates, we offer a conjecture on the
observer's voting behavior because no data link between the AAR metric and voting
are presented here to support such a claim. behavior. The acoustic analysis of the F^ sig-
Evidence from this study and from previous nal in candidates' voices produces a nonver-
investigations, however, supports the claim bal, unconscious measure of social
that evaluations based on the consciously dominance relations. This evidence of social
derived verbal signal do not necessarily dominance may be communicated to
relate to those derived from the emotional observers of the debate, and the resulting
domain. Thus we find that the association perception of one candidate's social domi-
between AAR and DP is not as strong in nance over the other ultimately may be
Table 1 as are other results because the expressed through the observers' voting
debate polls tend to tap more conscious and behavior. In a rational model of voting
verbally conditioned responses, whereas the behavior, observers of a presidential debate
AAR metric is derived from emotional and presumably gain information on a candidate
interactive resources produced within the in order to consciously reinforce or alter their
debate. ultimate political decision. This straight-
forward, rational approach, however, may
DISCUSSION AND CONCLUSIONS not apply to all or even to most voters.
The notion that overt reasoning does not
A past study (Gregory and Webster underlie all important decisions is not new to
1996) showed that social dominance could be social psychological literature. Nineteenth-
measured through a combination of FFT and century philosophers such as Friedrich
factor analysis; a significantly altered theoret- Nietzsche, for example, asserted that "the
ical and methodological version of this work greater part of our intellectual activity goes
was applied in the present study of national on unconsciously and unfelt by us"
political debates. Use of this altered version (Nietzsche [1882] 1964:333). Later writers
can predict voting behavior in U.S. presiden- such as Michael Polanyi, renowned for
tial elections. Some authors (Sigelman and declaring "We can know more than we can
Sigelman 1984) contend that the debates tell" (Polanyi 1967:4), reminded us that
tend to reinforce preexisting preferences; and "[o]ur body is the ultimate instrument of all
this was shown, to some extent, in the Fg our external knowledge, whether intellectual
results as presented here. or practical" (1967:15).
Another position, however, challenges Current research supports this connec-
the view that the only important effect of the tion. In a number of works based on cognitive
debates is to reinforce prior preferences. This neurology, Bechara, Damasio, and colleagues
"erosion" effect, as it has been called, obvi- (Bechara et al. 1997; Damasio 1994) pro-
ously can affect undecided voters, particular- duced important evidence contrary to the
ly those who possess relatively little political rational model whereby persons decide
information (Lanoue 1992). In close elections advantageously in complex situations by
involving large numbers of undecided and ill- using overt reasoning founded on declarative
informed voters, an unconsciously encoded knowledge. In an important experiment in
perception of one candidate's dominance cognitive neurology, Bechara et al.
over another, as derived from vocalic cues (1997:1293) suggest "that overt reasoning is
from the debate, may create a singularly preceded by a nonconscious biasing step that
influential bias that overrules the effects of uses neural systems other than those that306 SOCIAL PSYCHOLOGY OUARTERLY
support declarative knowledge." In the A past study (Gregory and Webster
experiment embodying these conclusions, for 1996) showed that social dominance could be
example, Bechara and colleagues used nor- measured through a combination of FFT and
mal participants and patients with prefrontal CFA analysis; we applied a theoretical and
brain damage and decision-making defects to methodological version of this work in the
perform a gambling task. The experiment present study of national political debates.
showed that the "normals" began to choose Use of this version has produced an accurate
advantageously even before realizing the metric of debate dominance and prediction
strategy that worked best, but the patients of popular vote outcomes. Thus the results of
continued to choose disadvantageous^ even the FQ analysis presented here offer a
when they knew the correct strategy. In addi- straightforward and objective measure of an
tion, normals showed anticipatory autonomic otherwise elusive feature of a candidate's
skin conductance responses when consider- presentation of self, which predicts the candi-
ing a choice involving risk, even before they date's relative social status and the popular
knew explicitly that it was risky. In contrast, vote.
the patients never produced the anticipatory Perhaps our finding—that analysis of a
autonomic responses, although some eventu- near-insignificant low-frequency humming
ally realized which of the choices were risky. sound can predict the results of important
This experiment suggests that nonconscious contests for the most powerful political posi-
biases in normal persons can guide behavior tion in the world—suggests that anthropolo-
before conscious awareness. These findings gist Ray Birdwhistell, one of the earliest and
are useful in interpreting voters' decisions in most eminent researchers of nonverbal inter-
national elections. action, may not be far off the mark in his
roughly estimated but empirically observed
In relating these cognitive neuroscience assumption that "65 percent of the social
results to Giles's work on CAT and to the meaning" in human interactions is conveyed
literature on vocalic nonverbal behavior by by nonverbal cues "that can make use of
Burgoon and others, we reach the theoreti- information received acoustically, visually, by
cal conclusion that actors autonomically touch, by smell, and so on" (Birdwhistell
code and decode nonverbal vocalic signals 1974:213).
infused with dominance and social status
content, such as was gained from the REFERENCES
debates. In the absence of conscious aware-
ness, according to CAT, a less dominant Bechara, Antoine, Hanna Damasio, Daniel Tranel,
actor will accommodate vocalic cues to a and Antonio R. Damasio. 1997. "Deciding
more dominant actor. This encoded signal in Advantageously Before Knowing the
Advantageous Strategy." Science
debaters' voices may be detected and 275:1293-95.
decoded without conscious intervention by Birdwhistell, Ray L. 1974. "The Language of the
an audience of observers. Subsequently the Body: The Natural Environment of Words."
decoded content could play an influential Pp. 27-52 in Human Communication:
role in observers' encoding decisions affect- Theoretical Explorations, edited by Albert
ing behavior, such as the vote. Silverstein. New York: Wiley.
Bracewell, Ronald N. 1989. "The Fourier
If this conjectural pattern of uncon- Transform." Scientific American 260:86-95.
scious perceptions and behavior operates in Burgoon, Judee K., David B. Buller, and W. Gill
national elections, the F^ may exert a strong Woodall. 1996. Nonverbal Communication:
influence over the popular vote in very close The Unspoken Dialogue. New York:
elections involving many undecided and less McGraw-Hill.
highly informed voters. Also, if the poll Burgoon, Judee K., Leesa Dillman, and Lesa A.
results before the debates are very close, the Stern. 1993. "Adaptation in Dyadic
Interaction: Defining and Operationalizing
influence of the Fp may be considerable in Patterns of Reciprocity and Compensation."
causing a vote erosion favoring the candi- Communication Theory 4:293-316.
date with a more commanding presence. Burgoon, Judee K., Lesa A. Stern and LeesaNONVERBAL VOCAL COMMUNICATION 307
Dillman. 1995. Interpersonal Adaptation. Interview Partners Effectively Predicts
New York: Cambridge University Press. Communication Accommodation and Social
Cappella, Joseph N. 1981. "Mutual Influence in Status Perceptions." Journal of Personality
Expressive Behavior: Adult-Adult and and Social Psychology 70:1231-40.
Infant-Infant Dyadic Interaction." Gregory, Stanford W., Jr., Stephen Webster, and
Psychological Bulletin 89:101-32. Gary Huang. 1993. "Voice Pitch and
Chappie, Eliot D. 1940. "Measuring Human Amplitude Convergence as a Metric of
Relations." General Psychology Quality in Dyadic Interviews." Language
Monographs 23:3-147. and Communication 13:195-217.
Damasio, Antonio R. 1994. Descartes' Error. New Jaffe, Joseph and Stanley Feldstein. 1970. Rhythms
York: Avon. of Dialogue. New York: Academic Press.
Ekman, Paul. 1965. "Differential Communication Lanoue, David J. 1992. "Qne That Made a
of Affect by Head and Body Cues." Journal Difference: Cognitive Consistency, Political
of Personality and Social Psychology Knowledge, and the 1980 Presidential
2:726-35. Debate." Public Opinion Quarterly
Ekman, Paul and Wallance V. Friesen. 1969. "Non- 56:168-84.
Verbal Leakage and Clues to Deception." Mazur, Allan. 1985. "A Biosocial Model of Status
Psychiatry 32:88-106. in Face-to-Face Primate Groups." Social
Germond, Jack W. and Jules Witcover. 1979. Forces 64:377^02.
"Presidential Debates: An Overview." Pp. Moore, David W. 2000. "Gore's Support Increases in
191-205 in The Past and Future of Last Poll Before Debate." Retrieved Qctober
Presidential Debates, edited by Austin 4, 2000 (www.Gallup.com/poll/releases/
Ranney. New York: Plenum. pr001004.asp).
Giles, Howard and Nikolas Coupland. 1991a. Natale, Michael. 1975. "Convergence of Mean
Language: Contexts and Consequences. Vocal Intensity in Dyadic Communication as
Pacific Grove, CA: Brooks/Cole. a Function of Social Desirability." Journal of
, eds. 1991b. Contexts of Accommodation. Personality and Social Psychology
Cambridge, UK: Cambridge University 32:790-804.
Press. Ng, Sik H. and James J. Bradac. 1993. Power in
Gorsuch, Richard L. 1983. Factor Analysis. 2nd ed. Language: Verbal Communication and Social
Hillsdale, NJ: Erlbaum. Influence. Newbury Park, CA: Sage.
Gregory, Stanford W., Jr. 1983. "A Quantitative Nietzsche, Friedrich W. [1882] 1964. The Joyful
Analysis of Temporal Symmetry in Wisdom, translated by Thomas Common.
Microsocial Relations." American New York: Russell and Russell.
Sociological Review 48:129-35. Pittam, Jeffery. 1987. "A Long-Term Spectral
. 1986. "Social Psychological Implications Measurement of Voice Quality as a Social
of Voice Frequency Correlations: Analyzing and Personality Marker: A Review."
Conversation Partner Adaptation by Language and Speech 30:1-12.
Computer." Social Psychology Quarterly . 1994. Voice in Social Interaction: An
49:237^6. Interdisciplinary Approach. Thousand Qaks,
. 1994. "Sounds of Power and Deference: CA: Sage.
Acoustic Analysis of Macro Social Pittam, Jeffery, Cindy Gallois, and Victor Callan.
Constraints on Micro Interaction." 1990. "The Long-Term Spectrum and
Sociological Perspectives 37.497-526. Perceived Emotion." Speech Communi-
Gregory, Stanford W., Jr., Kelly Dagan, and cation 9:177-87.
Stephen Webster. 1997. "Evaluating the Polanyi, Michael. 1967. The Tacit Dimension.
Relation of Vocal Accommodation in Garden City, NY: Anchor.
Conversation Partners' Fundamental Rahn, Wendy M., John H. Aldrich, and Eugene
Frequencies to Perceptions of Borgida. 1994. "Individual and Contextual
Communication Quality." Journal of Variations in Political Candidate Appraisal."
Nonverbal Behavior 21:23-43. American Political Science Review
Gregory, Stanford W., Jr., Brian E. Green, Robert 88:193-99.
M. Carrothers, Kelly A. Dagan, and Stephen Rosenberg, Shawn W. and Paul McCafferty. 1987.
Webster. 2000. "Verifying the Primacy of "The Image and the Vote." Public Opinion
Voice Fundamental Frequency in Social Quarterly 51:31^1.
Status Accommodation." Language and Sapolsky, Robert M. 1990. "Stress in the Wild."
Communication 21:37-60. Scientific American 264:116-23.
Gregory, Stanford W., Jr. and Stephen Webster. Scherer, Klaus R. 1985. "Methods of Research on
1996. "A Nonverbal Signal in Voices of Vocal Communication." Pp. 136-98 in308 SOCIAL PSYCHOLOGY QUARTERLY
Handbook of Methods in Nonverbal Sigelman, Lee and Carol K. Sigelman. 1984.
Behavior Research, edited by Klaus R. "Judgments of the Carter-Reagan Debate:
Scherer and Paul Ekman. Cambridge, UK: The Eyes of the Beholders." Public Opinion
Cambridge University Press. Quarterly 48:624-28.
Stanford W. Gregory Jr. is professor in the department of sociology at Kent State University. His
areas of interest generally include the social psychology of communication and language and
more specifically include objective means of electronically examining qualitative social compo-
nents within language and communication. He has published articles on these subjects in
Americati Sociological Review and Journal of Personality and Social Psychology. He is cur-
rently applying his spectral analysis techniques to examination of doctor-patient interaction.
Timothy J. Gattagher is assistant professor in the department of sociology at Kent State
University. His areas of research include doctor-patient interaction and access to health care, as
well as the more general area of nonverbal and emotional aspects of social organization. He has
published on the nonverbal aspects of doctor-patient interaction (Patient Education and
Counseling, 2001) and on emotional intelligence in medical school applicants (Academic
Medicine, 2000).You can also read
