See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/275667505

The Sound of Intellect

Article in Psychological Science · April 2015

DOI: 10.1177/0956797615572906 · Source: PubMed

CITATIONS READS

59 2,819

2 authors:

Juliana Schroeder Nicholas Epley

University of California, Berkeley University of Chicago
63 PUBLICATIONS 2,148 CITATIONS 116 PUBLICATIONS 19,922 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Juliana Schroeder on 17 September 2015.

The user has requested enhancement of the downloaded file.

572906
research-article2015
PSSXXX10.1177/0956797615572906Schroeder, EpleySpeech Conveys Intellect

Research Article

Psychological Science

The Sound of Intellect: Speech Reveals 2015, Vol. 26(6) 877­–891

© The Author(s) 2015
Reprints and permissions:
a Thoughtful Mind, Increasing a Job sagepub.com/journalsPermissions.nav
DOI: 10.1177/0956797615572906

Candidate’s Appeal pss.sagepub.com

Juliana Schroeder and Nicholas Epley

Booth School of Business, University of Chicago

Abstract
A person’s mental capacities, such as intellect, cannot be observed directly and so are instead inferred from indirect
cues. We predicted that a person’s intellect would be conveyed most strongly through a cue closely tied to actual
thinking: his or her voice. Hypothetical employers (Experiments 1–3b) and professional recruiters (Experiment 4)
watched, listened to, or read job candidates’ pitches about why they should be hired. These evaluators rated a
candidate as more competent, thoughtful, and intelligent when they heard a pitch rather than read it and, as a result,
had a more favorable impression of the candidate and were more interested in hiring the candidate. Adding voice to
written pitches, by having trained actors (Experiment 3a) or untrained adults (Experiment 3b) read them, produced
the same results. Adding visual cues to audio pitches did not alter evaluations of the candidates. For conveying one’s
intellect, it is important that one’s voice, quite literally, be heard.

Keywords
communication, voice, speech, mind perception, social cognition, decision making, open data

Received 8/29/14; Revision accepted 1/25/15

Some of the most important decisions in life are based on understanding the contents of his or her thoughts.
inferences about another person’s mental capacities: Is Because of the paralinguistic cues in voice, such as into-
this person trustworthy or deceptive? Was the perpetrator nation, cadence, and amplitude, observers who hear
capable of judging right from wrong? Will this job candi- communicators guess their actual thoughts and feelings
date be smart enough to succeed here? Such inferences more accurately than observers who read the exact same
require sophisticated social cognition about invisible words in text (Hall & Schmid Mast, 2007; Kruger, Epley,
mental processes that go beyond observable behavior, Parker, & Ng, 2005). Adding visual information to verbal
guided by top-down mechanisms of egocentric projec- information does not appear to increase this accuracy
tion (O’Brien & Ellsworth, 2012; Van Boven & Loewenstein, (Hall & Schmid Mast, 2007; Gesn & Ickes, 1999), which
2003) and stereotype application (Fiske, Cuddy, & Glick, suggests that visual information may be redundant with
2007) as well as by bottom-up interpretations of another speech or at least less informative for mental-capacity
person’s behavior (Gilbert & Malone, 1995). inferences.
Here we examine how judgments of another person’s We predicted that a person’s speech, beyond commu-
mental capacity—specifically, the capacity for reasoning nicating the contents of that person’s mind (his or her
and intellect—is affected by a cue directly linked to the specific thoughts and beliefs), also conveys the person’s
person’s ongoing mental experience: his or her voice. A fundamental capacity to think (his or her capacity for
person’s voice, after all, is a conduit for expressing reasoning, thoughtfulness, and intellect). Changes in the
sophisticated thoughts, beliefs, and knowledge using the
semantic and paralinguistic cues available in language
Corresponding Author:
(Pinker & Bloom, 1990). Research has revealed the Nicholas Epley, University of Chicago, Booth School of Business, 5807
unique importance of a person’s voice, over and above South Woodlawn Ave., University of Chicago GSB, Chicago, IL 60637
the semantic content of the person’s language, for E-mail: epley@chicagobooth.edu

Downloaded from pss.sagepub.com at UNIV CALIFORNIA BERKELEY LIB on July 20, 2015
878 Schroeder, Epley

tone, cadence, and pitch of an individual’s voice, for putting “little feeling or life into the words” spoke in a
example, may reveal the process of thinking and reason- relatively monotone voice and were subsequently rated
ing while it is happening, thereby conveying the presence as less mindful by observers, compared with actors who
of mental capacity more clearly than would the semantic were asked to read a speech as if they “were the real
content of language alone. Just as variability in motion speaker.” If people who read a person’s speech do not
serves as a cue for biological life, so too may variability in spontaneously compensate for the lack of paralinguistic
voice serve as a cue for a lively, active, and capable mind. cues in text, then their impressions of the speaker’s men-
If so, then a person should appear to have greater mental tal capacities could be systematically diminished com-
capacity—to be more thoughtful, rational, and intelli- pared with the impressions of observers who hear the
gent—when observers hear what the person has to say person’s speech.
than when they read what the person has to say. We tested the importance of a person’s voice for com-
Inferences about another person’s mental capacity are municating intellect in a domain where judgments of a
important in social life because the capacity for thinking, person’s mental capacities are both common and critical:
reasoning, and rationality is a defining feature of person- hiring decisions. We asked M.B.A. students to provide
hood according to both philosophers and laypeople spoken and written “elevator pitches”—short descrip-
(Demoulin et al., 2004; Dennett, 1987; Farah & Heberlein, tions of their qualifications—that they might use with
2007; Gray, Gray, & Wegner, 2007; Haslam, Bain, Douge, potential employers. Across five experiments, either
Lee, & Bastian, 2005; Kant, 1781/2007; Locke, 1841/1997). hypothetical employers (Experiments 1–3b) or profes-
People are perceived as more capable of reasoning than sional job recruiters (Experiment 4) watched, listened to,
are animals and robots. Failing to recognize another per- or read these candidates’ pitches and then evaluated the
son’s capacity for thinking, reasoning, and rationality is candidates’ intellect, reported their general impressions
therefore a subtle form of dehumanization (Harris & of the candidates, and indicated their interest in hiring
Fiske, 2006; Haslam & Loughnan, 2014; Waytz, Schroeder, the candidates. We predicted that job candidates would
& Epley, 2014). By predicting that a person’s speech seem more competent, thoughtful, and intelligent when
reveals his or her capacity for intellect, we are also pre- evaluators heard them explain their qualifications than
dicting that speech is humanizing. when evaluators read the text of the very same speeches,
Our hypotheses are based on three existing empirical or read written descriptions of the candidates’ qualifica-
results. First, observers more accurately predict another tions. Because intellect is essential for many jobs, we also
person’s thoughts and feelings when they can hear that predicted that potential employers would have more
person speak than when they read the same content favorable impressions of the candidates and be more
(Hall & Schmid Mast, 2007; Ickes, 2003; Kruger et al., interested in hiring the candidates when they heard the
2005; Mehrabian & Wiener, 1967; Zaki, Bolger, & Ochsner, candidates’ speech.
2009). Communicators themselves do not seem to recog- To ease presentation of the results, we report all analy-
nize this, as they expect to communicate equally well ses of evaluators’ ratings at the level of the individual
across media (Kruger et al., 2005). Second, giving a evaluators. However, we also analyzed the data in all of
machine a human voice increases observers’ tendency to our studies using hierarchical linear models (HLMs) to
anthropomorphize the otherwise mindless machine, account for the nesting of evaluators within candidates
attributing to it a mind capable of thinking and feeling (e.g., as recommended by Judd, Westfall, & Kenny, 2012).
(Nass & Brave, 2005; Takayama & Nass, 2008; Waytz, To do this, we created multilevel random-intercept, ran-
Heafner, & Epley, 2014). Third, paralinguistic characteris- dom-slope models with evaluators (Level 1) nested
tics of a person’s voice (e.g., pitch level) alter observers’ within candidates (Level 2), treating experimental condi-
trait-based impressions of that person (Gregory & tion as a fixed effect and the candidate being evaluated
Webster, 1996; Hughes, Mogilski, & Harrison, 2014; Jones, as a random effect. These analyses all yielded results that
Feinberg, DeBruine, Little, & Vukovic, 2010; Laplante & were as strong as or stronger than the results of the more
Ambady, 2003). In a series of experiments most relevant conservative tests we report here. We report the results
to our hypotheses (Schroeder & Epley, 2015), speakers for the HLMs in Supplemental Results, in the Supplemental
were rated as less mindful—for example, as less thought- Material available online.
ful and reasonable—when observers read a transcript of
their speech than when observers heard the very same
Experiment 1: Voice Versus Transcript
speech. Likewise, adding an actor’s voice to written text
led observers to rate the original author as more mindful. We videotaped M.B.A. students giving spoken elevator
In these experiments, pitch variance (i.e., intonation) pitches to their top potential employers. Evaluators then
conveyed the capacity for thinking most strongly. Actors watched, listened to, or read transcripts of the videos. We
who were instructed to read the words of a speech predicted that candidates would seem more competent,

Downloaded from pss.sagepub.com at UNIV CALIFORNIA BERKELEY LIB on July 20, 2015
Speech Conveys Intellect 879

thoughtful, and intelligent when their pitches were heard were still typing at the 10-min mark, we told them to fin-
rather than read, and that evaluators would consequently ish their thought and stop typing.
be more interested in hiring candidates whose pitches After finishing their spoken and written pitches, the
were heard rather than read. candidates completed a short survey asking them to pre-
Including the video condition provided a test between dict (a) how positively someone would evaluate their
our proposed mechanism—that speech conveys intellect written pitch (0 = not at all positively, 6 = very positively),
through paralinguistic cues in voice—and an alternative (b) how interested someone would be in hiring them
explanation—that speech conveys intellect through indi- after reading their written pitch (0 = not at all interested,
viduation. If additional individuating information con- 6 = very interested), (c) how positively someone would
veys intellect, then video should make a person appear evaluate their spoken pitch (0 = not at all positively, 6 =
even more mentally capable than audio alone. If, as we very positively), (d) how interested someone would be in
predicted, mental capacity is revealed primarily through hiring them after listening to their spoken pitch (0 = not
a person’s voice, then evaluators’ impressions should be at all interested, 6 = very interested), and (e) how many
similar whether they watch a video or listen to an audio times they had given their pitch before. We collected
recording. these predictions in order to examine how the candidates
expected they would be judged. Theoretically, such
expectations matter because they indicate whether the
Method
cues that convey mental capacities in social interaction
Participants. Eighteen M.B.A. students at the University are obvious to those in the midst of the interaction.
of Chicago Booth School of Business (mean age = 28.2 Practically, such expectations matter because they could
years, SD = 2.07; 11 males) responded to our request for guide how candidates approach potential employers.
research assistance, serving as job candidates in exchange Candidates who believe their spoken pitch will be judged
for a $5 Starbucks gift card. We then recruited 162 people exactly the same as their written pitch may see no reason
(mean age = 36.86 years, SD = 15.01; 80 males) visiting to seek voice time with a potential employer.
the Museum of Science and Industry in Chicago to evalu- We used only the spoken pitches from the candidates
ate the candidates in exchange for a food item. The sam- to create the stimuli for the hypothetical employers (evalu-
ple size for evaluators was predetermined by our goal to ators) in this experiment. The evaluators were assigned to
have 3 people evaluate each of the 18 candidates in each one of three conditions: Those in the video condition
of the three conditions (i.e., slightly more than 50 evalu- watched and listened to a candidate’s spoken pitch, those
ators per condition). We did not know what effect size to in the audio condition only listened to a spoken pitch, and
predict in this first experiment, but we arrived at this those in the transcript condition read a transcribed pitch.2
number because it was feasible at our laboratory, offered Each evaluator therefore observed only one candidate’s
multiple evaluators for each target, and was our best pitch in one medium. After seeing, hearing, or reading a
guess of the sample size needed to detect an effect of candidate’s pitch, evaluators completed a survey. The sur-
interest. Fifty participants per condition yields 80% power vey first explained,
to detect a medium-sized effect.
You just [watched/listened to/read the transcript of]
Procedure. We recruited the M.B.A. students to partici- an MBA student from the University of Chicago
pate as job candidates in a 20-min study on how people Booth School of Business talking about why he or
make hiring decisions. The candidates first named the she should be hired for his or her ideal job. This job
company for which they would most like to work and is in the service sector and it requires a highly
then considered (for 1 min) the pitch they would make to competent, thoughtful, and intelligent employee.
encourage this company to evaluate them positively and Your role in this study is to pretend that you are the
to hire them. Each candidate provided both a spoken and employer who is considering this candidate for the
a written pitch to the prospective employer (order coun- job. Based on the [clip/transcript] that you just
terbalanced). In the spoken-pitch condition, we told can- [watched/listened to/read], please let us know your
didates that we would videotape them as they gave their impressions of the candidate. You must evaluate
pitch and that they should speak directly to the camera. this candidate against all of the other candidates
We told them that they had 2 min to talk, although we who are also applying for the same job—you can
allowed them to reach the natural conclusion of their assume these other candidates are also high
pitch (actual video durations ranged from 49 s to 2 min achieving MBA students.
30 s). In the written-pitch condition, we told candidates to
compose a letter to the prospective employer. Candidates Evaluators then answered three questions about the
had 10 min1 to type their letter on a computer; if they candidate’s intellect: They rated (a) how competent the

Downloaded from pss.sagepub.com at UNIV CALIFORNIA BERKELEY LIB on July 20, 2015
880 Schroeder, Epley

candidate seemed compared with an average candidate years, SD = 9.03; 28 males) currently searching for jobs
for an M.B.A.-level position (−5 = much less competent, created written and spoken pitches for their preferred
5 = much more competent), (b) how thoughtful the can- employers. These participants then made the same pre-
didate seemed compared with an average candidate for dictions as the M.B.A. job candidates.
an M.B.A.-level position (−5 = much less thoughtful, 5 = As with the M.B.A. students, neither sample expected
much more thoughtful), and (c) how intelligent the can- to be evaluated more favorably by employers who lis-
didate seemed compared with an average candidate for tened to their pitches than by those who read their
an M.B.A.-level position (−5 = much less intelligent, 5 = pitches. The M.A. students did not predict that they would
much more intelligent). Evaluators then reported their be evaluated differently by employers who listened to
general impressions of the candidate: how much they their pitches (M = 3.88, SD = 1.20) than by those who
liked the candidate (0 = did not like at all, 10 = extremely read their pitches (M = 3.94, SD = 1.18), paired t(15) =
liked), how positive their overall impression of the candi- 0.15, p = .88, nor did they predict that their likelihood of
date was (0 = not at all positive, 10 = extremely positive), getting hired would differ depending on whether employ-
and how negative their overall impression of the candi- ers heard their pitches (M = 3.94, SD = 1.34) or read their
date was (0 = not at all negative, 10 = extremely negative; pitches (M = 4.06, SD = 1.00), paired t(15) = 0.36, p = .73,
reverse-scored). Finally, participants rated how likely d = 0.10. The community members predicted that they
they would be to hire the candidate for the job (0 = not would be evaluated significantly more positively by
at all likely, 10 = extremely likely). We averaged the rat- employers who read their pitches (M = 4.53, SD = 1.26)
ings of intelligence, competence, and thoughtfulness to than by employers who listened to their pitches (M =
form a composite measure of intellect (α = .91) and the 3.52, SD = 1.38), paired t(39) = 4.47, p < .01, d = 0.76.
ratings of liking, positive impression, and negative They also predicted that they would have a greater likeli-
impression to form a composite measure of general hood of being hired by employers who read their pitches
impressions (α = .89). (M = 4.35, SD = 1.29) than by employers who listened to
their pitches (M = 3.53, SD = 1.41), paired t(39) = 3.69,
p < .01, d = 0.61. Without further data, we are reluctant to
Results speculate about why the community sample predicted
Degrees of freedom in the statistical tests reported vary that they would be perceived more positively in writing
slightly because some participants failed to answer every than in speaking. We simply note that none of the three
survey item. samples expected to be seen as more mindful, or more
employable, when employers heard their voice.
Job candidates’ predictions. The M.B.A. students
were somewhat experienced in giving their spoken Hypothetical employers’ evaluations. As predicted,
pitches. On average, they had already given their pitches evaluators’ beliefs about job candidates’ intellect—their
1.44 (SD = 1.58) times. These participants did not predict competence, thoughtfulness, and intelligence—depended
that they would be evaluated differently when employers on the communication medium, F(2, 157) = 10.81, p <
listened to their spoken pitches (M = 3.61, SD = 0.78) .01, η2 = .12. As indicated by the standardized scores
than when employers read their written pitches (M = shown in Figure 1, evaluators who heard pitches rated
3.22, SD = 0.94), paired t(17) = 1.20, p = .25, d = 0.45. the candidates’ intellect more highly (M = 0.91, SD =
They also did not expect any difference in their likeli- 1.79) than did evaluators who read transcripts of pitches
hood of getting hired depending on whether employers (M = −0.70, SD = 2.81), t(157) = 3.79, p < .01, 95% confi-
listened to their spoken pitches (M = 3.28, SD = 0.89) or dence interval (CI) of the difference = [0.70, 2.51], d =
read their written pitches (M = 3.00, SD = 1.08), paired 0.60. Evaluators who watched pitches did not evaluate
t(17) = 0.80, p = .44, d = 0.29. In short, the students did the candidates’ intellect (M = 1.09, SD = 1.80) differently
not have strong expectations that other people’s evalua- than evaluators who listened to pitches, t(157) < 1. Simply
tions would depend on whether their pitches were heard adding more individuating information about a candidate
or read. We directly tested the veracity of these candi- through visual cues, such as physical appearance and
dates’ predictions in Experiment 2. nonverbal mannerisms, had no measurable impact on
Because these predictions were underpowered given evaluations of the candidate’s mind. Candidates’ intellect
the sample size of only 18 candidates, we collected data was conveyed primarily through their voice.
from two more samples of job candidates in an effort to Perhaps more important, evaluators who heard pitches
better understand candidates’ intuitions (for details, see also reported more favorable impressions of the candi-
Supplemental Results in the Supplemental Material). dates—liked the candidates more and had more positive
Sixteen M.A. students (mean age = 29.3 years, SD = 5.94; and less negative impressions of the candidates—than
11 males) and 40 community members (mean age = 28.6 did evaluators who read pitches (M = 5.69, SD = 1.96, vs.

Downloaded from pss.sagepub.com at UNIV CALIFORNIA BERKELEY LIB on July 20, 2015
Speech Conveys Intellect 881

Video
Audio
Transcript

Employers’ z-Scored Evaluation of Candidates

0.8

0.6

0.4

0.2

–0.2

–0.4

–0.6

–0.8
Intellect General Impression Hiring Likelihood
Fig. 1. Results from Experiment 1: evaluators’ standardized ratings of the job candidates’
intellect, their general impressions of the candidates, and their likelihood of hiring the
candidates. Results are shown separately for the video, audio, and transcript conditions.
Error bars represent ±1 SEM.

M = 4.78, SD = 2.64), t(159) = 2.16, p = .03, 95% CI of the candidates (M = 4.46, SD = 2.43) compared with evalua-
difference = [0.02, 1.80], d = 0.34 (see Fig. 1). Evaluators tors who heard pitches, ts < 1.
who heard pitches also reported being significantly more We predicted that a candidate’s voice would make
likely to hire the candidates (M = 4.34, SD = 2.26) than him or her seem more competent, thoughtful, and intel-
did evaluators who read exactly the same pitches (M = ligent, which in turn would lead potential employers to
3.06, SD = 3.15), t(156) = 2.49, p = .01, 95% CI of the dif- a more favorable general impression, and increase their
ference = [0.22, 2.34], d = 0.40 (see Fig. 1). These results perception of how likely they were to hire the candidate.
again did not appear to stem simply from having more A mediation analysis (Fig. 2) supported this hypothesis:
individuating information about the candidates in the Evaluators’ perceptions of a candidate’s intellect and
audio condition, because evaluators who watched pitches their general impressions of the candidate sequentially
did not report more favorable impressions (M = 5.98, mediated the effect of hearing the candidate’s voice
SD = 1.91) or an increased likelihood of hiring the (audio condition, coded as 1), rather than reading the

β = 0.77, SE =
Intellect 0.06, p < .01 Impressions
(α = .91) (α = .89)

β = 1.71, SE = β = –0.35, SE = β = 0.47, SE = β = 0.59, SE =

0.46, p < .01 0.31, p = .26 0.10, p < .01 0.10, p < .01

Experimental
Hiring
Condition:
β = –0.17, SE = 0.30, p = .57 Likelihood
Audio vs. Transcript

Fig. 2. Results from Experiment 1: mediation model testing the effect of experimental condition on reported likeli-
hood of hiring a job candidate, as mediated by perceived intellect of the candidate and general impressions of the
candidate.

Downloaded from pss.sagepub.com at UNIV CALIFORNIA BERKELEY LIB on July 20, 2015
882 Schroeder, Epley

candidate’s pitch (transcript condition, coded as 0), on hir- content (transcribed spoken pitch vs. written pitch).
ing likelihood. The effect of condition on hiring likelihood Therefore, we increased our targeted sample size to 4
was significant before perceived intellect and general evaluators per speaker per condition (total of 216 evalu-
impressions were included in the model, β = 1.21, SE = 0.54, ators). Because we did not know what effect size to
p = .03, but this effect became nonsignificant when the expect in this experiment, this was our best estimate of
mediators were included (β = −0.17, SE = 0.30, p = .57). the sample size we would need to detect an effect of
A 5,000-sample bootstrap test estimated that perceived interest. Our final sample was 218 visitors to the Museum
intellect had a significant indirect effect of 0.80 (SE = of Science and Industry in Chicago (mean age = 35.0
0.28, 95% CI = [0.33, 1.44]), that general impressions did years, SD = 12.8; 106 males), who evaluated candidates
not have a significant indirect effect (indirect effect = in exchange for a food item. We collected data from
−0.21, SE = 0.18, 95% CI = [−0.59, 0.13]), and that the two more than our targeted number of evaluators because we
mediators had a significant combined indirect effect of continued running the experiment until the end of a
0.79 (SE = 0.25, 95% CI = [0.33, 1.33]; MacKinnon, scheduled room reservation.
Fairchild, & Fritz, 2007).
These analyses suggest that a potential job candidate’s Procedure. Both the spoken and the written pitches
voice conveyed intellect, which led to a more positive from the 18 M.B.A. students in Experiment 1 served as
impression of the candidate and increased hypothetical our stimuli. We assigned participants serving as hypo-
interest in hiring the candidate among evaluators. These thetical employers (evaluators) to one of three condi-
results are consistent with our hypothesis that speech, tions: Those in the audio condition listened to a
because of the natural paralinguistic cues in voice that candidate’s spoken pitch, those in the transcript condi-
are particularly well equipped to express thought, can tion read the transcript of a candidate’s spoken pitch, and
reveal a person’s mental capacities. There was no system- those in the writing condition read a candidate’s own
atic evidence that being able to see someone in addition written pitch. After evaluators listened to the speech,
to hearing him or her affected inferences about mental read the transcribed speech, or read the written pitch,
capacity. This suggests that it is not merely the addition they completed a survey that included the same items as
of individuating information that reveals a person’s mind. in Experiment 1. As in that experiment, we averaged rat-
Rather, job candidates’ intellect seemed greater when ings to create composite measures of intellect (α = .88)
observers heard (rather than read) their speeches, regard- and general impressions (α = .87).
less of the total amount of information in the communi-
cation medium.
Results
Experiment 2: Speaking Versus Degrees of freedom in the statistical tests reported vary
slightly because some participants failed to answer every
Writing survey item.
In Experiment 1, we tested the importance of a person’s Evaluators’ beliefs about job candidates’ intellect—their
voice in observers’ evaluations of that person by transcrib- competence, thoughtfulness, and intelligence—again var-
ing candidates’ speech to ensure that semantic content ied by communication medium, F(2, 215) = 3.07, p = .05,
was identical in the audio and transcript conditions. In η2 = .03. As indicated by the standardized scores shown
Experiment 2, we conducted a replication test of the main in Figure 3, evaluators who heard pitches rated the can-
results from Experiment 1, using the same spoken pitches, didates’ intellect more highly (M = 1.12, SD = 1.85) than
and also added a critical third condition in which evalua- did evaluators who read transcripts of pitches (M = 0.35,
tors read candidates’ written pitches. If a written pitch is SD = 2.41), t(215) = 2.09, p = .04, 95% CI of the differ-
evaluated like a spoken pitch, then candidates’ voices are ence = [0.06, 1.47], d = 0.29. They also rated the candi-
not necessary to convey intellect, but some aspect of eval- dates’ intellect more highly than did evaluators who read
uating a transcribed speech would apparently explain our written pitches (M = 0.31, SD = 2.34), t(215) = 2.20, p =
prior results. If a written pitch is evaluated like a tran- .03, 95% CI of the difference = [0.12, 1.50], d = 0.30.
script, then this would provide stronger evidence that a Evaluations of the candidates’ intellect did not differ
person’s voice conveys his or her mental capacities. between the transcript and writing conditions, t(215) < 1.
Evaluators’ general impressions of the job candi-
dates—liking, positive impression, and negative impres-
Method sion of the candidates—also varied by condition, F(2,
Participants. We anticipated additional variance in this 214) = 4.72, p = .01, η2 = .04. Specifically, evaluators who
experiment because we varied not only the communica- heard pitches reported more favorable impressions of the
tion medium (speech vs. writing) but also the semantic candidates (M = 6.30, SD = 1.78) than did evaluators who

Downloaded from pss.sagepub.com at UNIV CALIFORNIA BERKELEY LIB on July 20, 2015
Speech Conveys Intellect 883

Writing
Transcript
Audio
0.8

Employers’ z-Scored Evaluation of Candidates

0.6

0.4

0.2

–0.2

–0.4

–0.6

–0.8
Intellect General Impression Hiring Likelihood
Fig. 3. Results from Experiment 2: evaluators’ standardized ratings of the job candidates’
intellect, their general impressions of the candidates, and their likelihood of hiring the
candidates. Results are shown separately for the writing, transcript, and audio conditions.
Error bars represent ±1 SEM.

read transcripts of pitches (M = 5.44, SD = 2.39), t(214) = mediators: β = 1.08, SE = 0.46, p = .02; with the media-
2.31, p = .02, 95% CI of the difference = [0.16, 1.54], d = tors: β = 0.08, SE = 0.24, p = .74). A 5,000-sample boot-
0.32, and evaluators who read the candidates’ written strap test estimated that perceived intellect had a
pitches (M = 5.23, SD = 2.46), t(214) = 2.90, p < .01, 95% significant indirect effect of 0.42 (SE = 0.20, 95% CI =
CI of the difference = [0.37, 1.77], d = 0.40 (see Fig. 3). [0.05, 0.83]), that general impressions did not have a sig-
Evaluators who heard pitches also reported being more nificant indirect effect (indirect effect = 0.14, SE = 0.14,
likely to hire the candidates (M = 4.83, SD = 2.53) than 95% CI = [−0.11, 0.43]), and that the two mediators had a
did evaluators who read transcripts (M = 3.77, SD = 2.88), significant combined indirect effect of 0.41 (SE = 0.19,
t(208) = 2.30, p = .02, 95% CI of the difference = [0.15, 95% CI = [0.06, 0.79]; MacKinnon et al., 2007).
1.97], d = 0.32, and marginally more likely to hire the As shown in Figure 4, a sequential mediation analysis
candidates than did evaluators who read written pitches comparing voice (audio condition, coded as 1) against
(M = 3.99, SD = 2.73), t(208) = 1.84, p = .07, 95% CI of the text (transcript and writing conditions combined, coded
difference = [−0.03, 1.73], d = 0.26 (see Fig. 3). General as 0) yielded the same conclusions. Including perceived
impressions of the candidates and hiring likelihood did intellect and general impressions in the model made the
not vary between the transcript and writing conditions, effect of communication medium nonsignificant (without
ts < 1. the mediators: β = 0.95, SE = 0.40, p = .02; with the media-
Finally, we again tested whether evaluators’ percep- tors: β = −0.04, SE = 0.21, p = .85). A 5,000-sample boot-
tions of the candidates’ intellect and general impressions strap test estimated that perceived intellect had a significant
of the candidates sequentially mediated the effect of indirect effect of 0.41 (SE = 0.17, 95% CI = [0.10, 0.76]),
hearing pitches (rather than reading them) on likelihood that general impressions did not have an indirect effect
of hiring the candidates. Because the contrast between (indirect effect = 0.21, SE = 0.12, 95% CI = [−0.01, 0.46]),
the audio and writing conditions on hiring likelihood was and that the two mediators had a significant combined
only marginally significant with a two-tailed test (p = .07), indirect effect of 0.37 (SE = 0.15, 95% CI = [0.09, 0.67];
a sequential mediation analysis comparing these two MacKinnon et al., 2007).
conditions was technically unjustified. In a sequential These results both replicate and extend those of
mediation test comparing the audio (coded as 1) and Experiment 1. A person’s voice again seemed to commu-
transcript (coded as 0) conditions alone, adding per- nicate a more thoughtful mind than written text did, and
ceived intellect and general impressions to the model this effect emerged both when the semantic content of a
reduced the effect of communication medium on hiring speech was held constant by transcribing it and when the
likelihood so that it became nonsignificant (without the speaker was allowed to craft a written pitch him- or

Downloaded from pss.sagepub.com at UNIV CALIFORNIA BERKELEY LIB on July 20, 2015
884 Schroeder, Epley

β = 0.78, SE =
Intellect 0.04, p < .01 Impressions
(α = .88) (α = .87)

β = 0.81, SE = β = 0.36, SE = β = 0.51, SE = β = 0.59, SE =

0.33, p = .01 0.21, p = .09 0.07, p < .01 0.07, p < .01

Experimental
Hiring
Condition:
β = –0.04, SE = 0.21, p = .85 Likelihood
Audio vs. Text

Fig. 4. Results from Experiment 2: mediation model testing the effect of experimental condition (audio condition
vs. transcript and writing conditions combined) on reported likelihood of hiring a job candidate, as mediated by
perceived intellect of the candidate and general impressions of the candidate.

herself. The capacity for intellect, it appears, is more pick their ideal job and write a pitch to the employer
readily conveyed through one’s voice than through one’s about why they would be a good fit for the job. We
writing, even when the semantic content is identical. The want you to pretend that you are the MBA student
results for the writing condition, of course, do not indi- who wrote the pitch. We want you imbue your
cate that it is impossible for a talented writer to overcome words with all of the thoughts, emotions, and
the limitations of text alone; they indicate only that our substance that the writer him/herself felt. We want
M.B.A. students in Experiment 1 did not predict that they you to read it as if you were actually coming up with
needed to overcome these limitations and did not do so the lines naturally off the top of your head, as in a
spontaneously. real conversation, rather than reading from a script.
We want you to speak as naturally as you would if
you were making a real pitch to an employer right
Experiments 3a and 3b: Giving Voice
now.
to Text
Experiments 1 and 2 suggest that removing a speaker’s We designed these instructions in order to maintain
voice from his or her spoken pitch can make the speaker the natural paralinguistic cues in readers’ voices, so that
seem less mindful (less thoughtful, rational, and intelli- the readings would not sound artificial or strange to lis-
gent). In Experiments 3a and 3b, we conducted a more teners. Each actor read all 18 written pitches3 from
comprehensive test of our hypothesis by examining Experiment 1 out loud, and we later separated each
whether adding voice to text likewise affects evaluations recorded pitch into its own sound file.
of the author’s intellect. In Experiment 3a, we recruited Evaluators were 265 visitors to the Museum of Science
four trained stage actors to read all 18 pitches. To ensure and Industry in Chicago (mean age = 35.03 years, SD =
that our results were not due to some aspect unique to 14.40; 124 males), who agreed to participate in exchange
actors’ voices, we conducted a replication with a more for a food item. We targeted a sample of approximately
representative sample of readers in Experiment 3b. 270 participants to serve as evaluators, so that approxi-
mately 3 participants would evaluate each version (four
actors’ renditions plus the candidate’s original written
Experiment 3a version) of each of 18 job pitches. We collected as much
Method. The 4 most experienced stage actors who data as possible until the end of a scheduled room
responded to our request for assistance received $25 reservation.
each for their participation (mean age = 20 years; 2 males; We randomly assigned participants serving as poten-
selected from a pool of 12). Actors came to a recording tial employers (evaluators) to one of three conditions:
booth, where an experimenter gave them the following Those in the writing condition read a written pitch, those
instructions: in the female-speaker condition listened to one of the
female actors reading a written pitch, and those in the
Today, you will be reading 18 different “elevator male-speaker condition listened to one of the male actors
pitches” from the University of Chicago Booth School reading a written pitch. Because prior research suggests
of Business students. These 18 students were told to that people may evaluate the voice of a female differently

Downloaded from pss.sagepub.com at UNIV CALIFORNIA BERKELEY LIB on July 20, 2015
Speech Conveys Intellect 885

Writing
Male Speaker
Female Speaker
0.8

Employers’ z-Scored Evaluation of Candidates

0.6

0.4

0.2

–0.2

–0.4

–0.6

–0.8
Intellect General Impression Hiring Likelihood
Fig. 5. Results from Experiment 3a: evaluators’ standardized ratings of the job candidates’
intellect, their general impressions of the candidates, and their likelihood of hiring the
candidates. Results are shown separately for the writing, male-speaker, and female-speaker
conditions. Error bars represent ±1 SEM.

than the voice of a male (Brooks, Huang, Kearney, & 3.21, ps < .01, 95% CIs of the difference = [0.38, 1.59] and
Murray, 2014; Eagly & Mladinic, 1989), although we had [0.33, 1.59], ds = 0.41 and 0.40. We observed weaker
observed no effects of the candidate’s gender in effects of experimental condition on evaluators’ general
Experiments 1 and 2, we randomly assigned evaluators to impressions of the candidates, F(2, 262) = 2.76, p = .07,
separate gender conditions. We therefore unconfounded η2 = .02 (see Fig. 5). Evaluators had marginally more pos-
the speaker’s gender from the pitch’s content so that we itive impressions when they listened to the female speak-
could test for gender effects holding semantic content ers (M = 6.33, SD = 1.82) than when they read the same
constant. After reading or listening to a pitch, evaluators pitches (M = 5.77, SD = 2.14), t(262) = 1.80, p = .07, 95%
answered the same survey items used in Experiment 1. CI of the difference = [−0.07, 1.21], d = 0.22. Evaluators
We averaged ratings to create composite measures of had more negative impressions of male speakers (M =
intellect (α = .84) and general impressions (α = .80). 5.79, SD = 1.78) than of female speakers, t(262) = −2.12,
p = .04, 95% CI of the difference = [−1.03, −0.06], d = 0.26,
Results. Degrees of freedom in the statistical tests but evaluations of male speakers did not differ from eval-
reported vary slightly because some participants failed to uations of candidates whose written pitches were read,
answer every survey item. t(262) < 1. Evaluators who listened to female speakers
Evaluators’ judgments of the candidates’ intellect, their also reported being more likely to hire the candidates
general impressions of the candidates, and their reported (M = 6.31, SD = 2.06) than did evaluators who read
likelihood of hiring the candidates did not differ signifi- pitches (M = 4.96, SD = 2.86), t(259) = 3.43, p < .01, 95%
cantly depending on whether the transcripts were read by CI of the difference = [0.56, 2.13], d = 0.42, and were
the first or second female actor, or by the first or second marginally more likely to hire the candidates than were
male actor,4 ts < 1.70, ps > .09. We therefore collapsed across evaluators who listened to male speakers (M = 5.69, SD =
the two actors of each gender in the following analyses. 2.27), t(259) = 1.91, p = .06, 95% CI of the difference =
As predicted, evaluators’ beliefs about the candidates’ [0.02, 1.20], d = 0.24 (see Fig. 5). Evaluators who listened
intellect varied significantly by experimental condition, to male speakers were also marginally more likely to hire
F(2, 262) = 6.34, p < .01, η2 = .05. As indicated by the the candidates than were those who read the same
standardized scores shown in Figure 5, evaluators judged pitches, t(259) = 1.86, p = .06, 95% CI of the difference =
candidates to have greater intellect when they listened to [−0.10, 1.56], d = 0.23.
the female and male speakers (M = 2.36, SD = 1.59, and When we combined the data from the male-speaker
M = 2.33, SD = 1.72, respectively) than when they read and female-speaker conditions and compared evalua-
the same pitches (M = 1.37, SD = 2.19), ts(262) = 3.29 and tions of written versus spoken pitches, we observed that

Downloaded from pss.sagepub.com at UNIV CALIFORNIA BERKELEY LIB on July 20, 2015
886 Schroeder, Epley

β = 0.70, SE =
0.05, p < .01 Impressions
Intellect
(α = .84) (α = .80)

β = 0.95, SE = β = –0.38, SE = β = 0.49, SE = β = 0.67, SE =

0.27, p < .01 0.22, p = .08 0.06, p < .01 0.06, p < .01

Experimental
Hiring
Condition:
β = 0.39, SE = 0.21, p = .07 Likelihood
Audio vs. Writing

Fig. 6. Results from Experiment 3a: mediation model testing the effect of experimental condition on reported
likelihood of hiring a job candidate, as mediated by perceived intellect of the candidate and general impressions
of the candidate.

evaluators who listened to the pitches (n = 212) believed author’s gender constant, or simply random variance of a
the candidates had greater intellect than did evaluators moderately sized effect across multiple experiments.
who read the same pitches (n = 53), t(262) = 3.57, p < What is clear is that listening to pitches—even pitches
.01, 95% CI of the difference = [0.44, 1.51], d = 0.44. They spoken by actors and not by the candidates who created
also reported being more likely to hire the candidates, them—rather than reading the very same pitches affects
t(259) = 2.89, p < .01, 95% CI of the difference = [0.33, evaluations of the authors’ mental capacity. Once again,
1.75], d = 0.36. However, evaluators who listened to the we found that a person’s voice seems to reveal a mind
pitches did not have significantly more positive impres- capable of thinking and reasoning.
sions of the candidates than did evaluators who read the Although the results of Experiment 3a are consistent
pitches t(262) = 1.02, p = .31. with our hypotheses, it is possible that they are accounted
As in the prior experiments, and as shown in Figure 6, for by some unique aspect of actors’ voices, such as
evaluators’ beliefs about the candidates’ intellect and being particularly attractive (Zuckerman & Driver, 1989).
their general impressions of the candidates sequentially We therefore conducted a replication test in Experiment
mediated the effect of condition—audio condition (male- 3b, using speakers not uniquely selected for being actors.
and female-speaker conditions combined, coded as 1)
versus writing condition (coded as 0)—on hiring judg-
Experiment 3b
ments. Including perceived intellect and general impres-
sions in the model made the effect of communication Method. Eighteen visitors to the Museum of Science and
medium nonsignificant (without the mediators: β = 1.04, Industry in Chicago (11 males) each agreed to read one
SE = 0.36, p < .01; with the mediators: β = 0.39, SE = 0.21, job candidate’s written pitch in exchange for a food item.
p = .07). A 5,000-sample bootstrap test estimated that Because we could not ask each visitor to read all 18
intellect had a significant indirect effect of 0.46 (SE = pitches, we were unable to orthogonally manipulate
0.17, 95% CI = [0.15, 0.79]), that general impressions did reader’s gender as we did in Experiment 3a. Instead, we
not have an indirect effect (indirect effect = −0.26, SE = matched each reader’s gender to the actual candidate’s
0.15, 95% CI = [−0.57, 0.03]), and that the two mediators gender. Each reader received one hard copy of a job
had a significant combined indirect effect of 0.45 (SE = candidate’s pitch. We gave the readers the same reading
0.16, 95% CI = [0.15, 0.77]; MacKinnon et al., 2007). instructions as in Experiment 3a and allowed them to
Adding a human voice—whether male or female—to practice reading the pitch as many times as they wanted
written pitches made the job candidates seem to have before we recorded them.
greater intellect, and increased reported interest in hiring The procedure for evaluators was similar to that in the
the candidates. Results for general impressions of the audio-pitch and written-pitch conditions in Experiment
candidates were less consistent, varying by the speaker’s 3a. We predetermined that we needed a sample size of at
gender. We did not predict this variability, nor did we least 108 evaluators, approximately 3 participants for
observe it in any other experiment. It is therefore unclear each job candidate in each of the two conditions. Our
whether this result reflects something systematic about final sample was 135 online workers on Amazon
variability in what the voice conveys, something about Mechanical Turk (mean age = 31.9 years, SD = 9.9; 87
manipulating a speaker’s voice while holding the original males), who evaluated candidates in exchange for $0.30.

Downloaded from pss.sagepub.com at UNIV CALIFORNIA BERKELEY LIB on July 20, 2015
Speech Conveys Intellect 887

Evaluators either listened to a reader’s recording of a companies voluntarily agreed to evaluate pitches of job
pitch or read a candidate’s written pitch. Evaluators then candidates from the University of Chicago Booth School
evaluated the candidates using the same survey used in of Business. These recruiters had attended a conference
Experiment 1. We averaged ratings to create composite at the University of Chicago and were e-mailed afterward
measures of intellect (α = .92) and general impressions to request their participation. We initially e-mailed all 66
(α = .85). recruiters who had attended the conference and then
extended a second, personal invitation to those who did
Results. As in Experiment 3a, evaluators who listened not respond to the first e-mail. We did not continue our
to the pitches (n = 67) believed that the candidates had unsolicited request for participation beyond the second
greater intellect (M = 8.21, SD = 1.62), compared with e-mail out of professional courtesy. Our target sample
those who read the pitches (n = 68; M = 7.38, SD = 2.06), was every participant at the conference. We included all
t(133) = 2.60, p = .01, 95% CI of the difference = [0.20, recruiters who completed the survey in the following
1.46], d = 0.45. They also had more positive impressions analyses.
of the candidates (M = 8.45, SD = 1.82, vs. M = 7.33, SD =
2.45), t(133) = 3.00, p < .01, 95% CI of the difference = Procedure. Because we knew that our sample of evalu-
[0.38, 1.85], d = 0.52, and reported being more likely to ators would be smaller than in the previous experiments,
hire them (M = 7.85, SD = 2.25, vs. M = 6.74, SD = 3.00), we randomly selected three job candidates’ spoken
t(133) = 2.44, p = .02, 95% CI of the difference = [0.21, pitches from Experiment 1 to use as stimuli. In an online
2.02], d = 0.42. As in Experiments 1, 2, and 3a, beliefs survey, we randomly assigned recruiters to either listen to
about the candidates’ intellect and general impressions of one of the spoken pitches (audio condition) or read the
them sequentially mediated the effect of condition (hear- transcription of one of those pitches (transcript condi-
ing pitches vs. reading pitches) on hiring judgments. tion). We recorded how long each recruiter spent on the
Including perceived intellect and general impressions in survey page with the stimulus. The recruiters then
the model made the effect of communication medium answered the same survey items as in Experiment 1, with
nonsignificant (without the mediators: β = 1.11, SE = 0.46, one change: All responses were recorded on Likert scales
p = .02; with the mediators: β = –0.10, SE = 0.23, p = .66). labeled from 0 to 10. Finally, because this experiment
A 5,000-sample bootstrap test estimated that perceived was conducted online, we also asked the recruiters to
intellect had a significant indirect effect of 0.41 (SE = complete a memory test in which they reported “every-
0.39, 95% CI = [0.10, 0.75]), that general impressions did thing you can remember about the pitch the MBA stu-
not have an indirect effect (indirect effect = 0.30, SE = dent gave.” This memory test allowed us to evaluate a
0.16, 95% CI = [−0.09, 0.65]), and that the two mediators possible alternative interpretation of our observed results
had a significant combined indirect effect of 0.51 (SE = based on the amount of information remembered when
0.23, 95% CI = [0.12, 1.00]; MacKinnon et al., 2007). Thus, listening compared with reading. We averaged ratings to
even untrained readers convey a more capable mind create composite measures of intellect (α = .92) and gen-
through their voices, making candidates seem more eral impressions (α = .93).
appealing and employable.
Results
Experiment 4: Professional Recruiters The pattern of evaluations by professional recruiters rep-
Replicating our experiments by manipulating information licated the pattern observed in Experiments 1 through 3b
used in actual hiring decisions in real firms would be (see Fig. 7). In particular, the recruiters believed that the
unethical, but we got closer to an ecologically valid test job candidates had greater intellect—were more compe-
of our hypotheses in Experiment 4 by examining whether tent, thoughtful, and intelligent—when they listened to
the same patterns observed in Experiments 1 through 3b pitches (M = 5.63, SD = 1.61) than when they read pitches
would be replicated if the evaluators were expert recruit- (M = 3.65, SD = 1.91), t(37) = 3.53, p < .01, 95% CI of the
ers. In their jobs, the professional recruiters in Experiment difference = [0.85, 3.13], d = 1.16. The recruiters also
4 were charged with actually hiring from the very same formed more positive impressions of the candidates—
sample of M.B.A. students who made job pitches in our rated them as more likeable and had a more positive and
experiment. less negative impression of them—when they listened to
pitches (M = 5.97, SD = 1.92) than when they read pitches
(M = 4.07, SD = 2.23), t(37) = 2.85, p < .01, 95% CI of the
Method difference = [0.55, 3.24], d = 0.94. Finally, they also
Participants. Thirty-nine professional recruiters (mean reported being more likely to hire the candidates when
age = 30.85 years, SD = 6.24; 9 males) from Fortune 500 they listened to pitches (M = 4.71, SD = 2.26) than when

Downloaded from pss.sagepub.com at UNIV CALIFORNIA BERKELEY LIB on July 20, 2015
888 Schroeder, Epley

Audio
Transcript
10
9

Recruiters’ Evaluation of Candidates

8
7
6
5
4
3
2
1
0
Intellect General Impression Hiring Likelihood
Fig. 7. Results from Experiment 4: professional recruiters’ ratings of the job candi-
dates’ intellect, their general impressions of the candidates, and their likelihood of
hiring the candidates. Results are shown separately for the audio and transcript condi-
tions. Error bars represent ±1 SEM.

they read the same pitches (M = 2.89, SD = 2.06), t(37) = [0.37, 2.47]), but that general impressions (95% CI =
2.62, p < .01, 95% CI of the difference = [0.41, 3.24], d = [−0.30, 0.40]) and the two mediators combined (95% CI =
0.86. [−0.14, 1.40]) did not have an indirect effect (MacKinnon
Unlike in the prior experiments, however, the evalua- et al., 2007). It appears that our professional recruiters, in
tors’ beliefs about the candidates’ intellect and their gen- contrast to the evaluators in Experiments 1 through 3b,
eral impressions of the candidates only partially mediated based their interest in hiring more heavily on their per-
the effect of communication medium (audio condition ceptions of intellect alone than on joint considerations of
vs. transcript condition, coded 1 and 0, respectively) on intellect and general impressions. But just as in the previ-
hiring decisions (Fig. 8). When intellect and general ous experiments, those perceptions were influenced by
impression were included in the model, the effect of the presence or absence of a candidate’s voice.
communication medium became nonsignificant (without Finally, the recruiters did not spend significantly differ-
the mediators: β = 1.83, SE = 0.70, p = .01; with the medi- ent amounts of time engaging with the stimuli in the
ators: β = −0.04, SE = 0.52, p = .94). A 5,000-sample boot- audio condition (M = 173.85 s, SD = 145.97) and the
strap test estimated that perceived intellect had a transcript condition (M = 137.86 s, SD = 197.49), t(37) =
significant indirect effect of 1.30 (SE = 0.54, 95% CI = 0.65, p = .52, d = 0.21. Although 6 recruiters did not

β = 0.93, SE =
Intellect 0.12, p < .01 Impressions
(α = .92) (α = .93)

β = 1.99, SE = β = 0.06, SE = β = 0.66, SE = β = 0.30, SE =

0.56, p < .01 0.48, p = .91 0.21, p < .01 0.18, p = .11

Experimental
Hiring
Condition:
Likelihood
Audio vs. Transcript β = –0.04, SE = 0.52, p = .94

Fig. 8. Results from Experiment 4: mediation model testing the effect of experimental condition on
reported likelihood of hiring a job candidate, as mediated by perceived intellect of the candidate and gen-
eral impressions of the candidate.

Downloaded from pss.sagepub.com at UNIV CALIFORNIA BERKELEY LIB on July 20, 2015
Speech Conveys Intellect 889

complete the memory test, those who did complete it not identify which paralinguistic cues of speech do so.
wrote similar amounts about the candidates in the audio Prior research using a broader set of mental-capacity
condition (M = 53.94 words, SD = 30.06) and the tran- measures (Schroeder & Epley, 2015) suggests that pitch
script condition (M = 58.67 words, SD = 15.07), t(31) = variance, or intonation, plays a significant role. Just as
−0.55, p = .59, d = 0.20. variance in motion (i.e., movement) is a cue that reveals
If voice affected evaluations only of novice employers, the presence of biological life, variance in pitch may
then our results would be of more theoretical interest reveal the presence of an active and lively mind. Pitch
than practical importance. However, we obtained the variance can convey enthusiasm, interest, and active
same results among the very recruiters whose job it was deliberation, whereas a monotone voice sounds dull and
to hire from the same sample of M.B.A. students who mindless. Indeed, one reason why text may not convey a
provided the elevator pitches, and who were relative person’s intellect is that readers do not spontaneously
experts in evaluating job candidates. Whereas some evi- add pitch variance or other paralinguistic cues into writ-
dence suggests that experience reduces decision-making ten text. Identifying the cues that convey a person’s mind
biases (e.g., List, 2003), the effect size we observed for through speech is essential for understanding moderators
hiring interest was actually larger among professional of the effects we observed.
recruiters (d = 0.86) than among novice recruiters in the Second, Experiment 1 suggests that being able to see
equivalent conditions in Experiments 1 and 2 (ds = 0.40 a candidate may have no additional impact on evalua-
and 0.32). Unfortunately, our sample in Experiment 4 tions of intellect above and beyond the impact of being
contained only a restricted set of randomly selected able to hear the candidate. Although a person’s body lan-
pitches from our larger set, and so these effect sizes are guage (e.g., Imada & Hakel, 1977; Stewart, Dustin,
not perfectly comparable. At the very least, however, Barrick, & Darnold, 2008), demographics (e.g., Bertrand
Experiment 4 demonstrates that the effect of voice does & Mullainathan, 2003; Biernat & Kobrynowicz, 1997;
not appear to disappear among hiring experts. Our Glick, Zion, & Nelson, 1988), and appearance (Borkenau
results may be of practical, and not just theoretical, & Liebler, 1993; Murphy, 2007) can affect evaluations,
importance. these prior experiments did not pit the importance of
visual cues against vocal cues in the same experiment;
instead, visual cues were manipulated in the absence of
General Discussion any vocal cues. Our experiments suggest that a person’s
The words that come out of a person’s mouth convey the voice could be uniquely equipped to communicate other-
presence of a thoughtful mind more clearly than the wise invisible mental capacities, but our experiments
words typed by a person’s hands—even when those were also not designed to test the relative impact of visual
words are identical. Across five experiments, evaluators and vocal cues. We suggest that a person’s mental capaci-
who listened to job pitches were consistently more inter- ties, such as intellect, may be conveyed more strongly
ested in hiring the candidates than were evaluators who through speech than through body cues partly because
read identical pitches. A person’s voice communicates body cues lack the semantic content inherent in spoken
not only the content of his or her thinking, but also the or text-based language. More experiments that directly
humanlike capacity for thinking. compare different communication media are needed to
These results would apparently be surprising to the clearly identify the independent influences of visual and
speakers themselves. In Experiment 1, we asked three auditory cues on inferences about other people’s minds.
separate samples of job candidates to predict how they Finally, our experiments raise practical implications for
would be judged, and none expected that their spoken people who are trying to reveal their thoughtful mind to
pitches would convey significantly greater intellect than others (e.g., job candidates). Not only inexperienced
their written pitches. Practically, such expectations matter evaluators but also professional recruiters were influ-
because they may affect how job candidates approach enced by a candidate’s voice. Although text-based com-
employers. If candidates are unaware of how the com- munication media, such as e-mail, may provide quick
munication medium affects the impression they convey, and easy ways to connect with potential employers, our
then they will be just as likely to write to employers as to experiments suggest that voiceless communication comes
speak with them. Theoretically, these expectations matter with an unexpected inferential cost. A person’s voice, it
because they suggest that the underlying mechanisms seems, carries the sound of intellect.
responsible for conveying a person’s mental capacities
are surprising rather than obvious. Author Contributions
We believe that our experiments raise three interesting J. Schroeder and N. Epley developed the hypotheses and
avenues for further research. First, speech may reveal the experiments to test them. J. Schroeder created materials, over-
presence of a thoughtful mind, but our experiments do saw data collection, and conducted data analyses under the

Downloaded from pss.sagepub.com at UNIV CALIFORNIA BERKELEY LIB on July 20, 2015
890 Schroeder, Epley

supervision of N. Epley. J. Schroeder and N. Epley wrote the 4. Three participants rated their candidate’s intellect and their
manuscript. Both authors approved the final version of the general impressions of the candidate, but not their likelihood
manuscript for submission. of hiring the candidate.

Acknowledgments References
We thank Kaushal Addanki, Julia Arozena, Jesus Diaz, Janet Bertrand, M., & Mullainathan, S. (2003). Are Emily and Greg
Flores, Jasmine Kwong, Justin Liang, Paul Lou, Sean Alexander more employable than Lakisha and Jamal? A field experi-
McWillie, Rachel Meng, Sarah Molouki, Michael Pang, Megan ment on labor market discrimination (National Bureau of
Porter, Emily Shaw, Lester Tong, Sherry Tseng, and Emily Economic Research Working Paper No. 9873). Cambridge,
Wolodiger for assistance conducting experiments. We are also MA: National Bureau of Economic Research.
grateful to the University of Chicago Career Services for enabling Biernat, M., & Kobrynowicz, D. (1997). Gender- and race-based
us to connect with professional recruiters. standards of competence: Lower minimum standards but
higher ability standards for devalued groups. Journal of
Declaration of Conflicting Interests Personality and Social Psychology, 72, 544–557.
The authors declared that they had no conflicts of interest with Borkenau, P., & Liebler, A. (1993). Convergence of stranger rat-
respect to their authorship or the publication of this article. ings of personality and intelligence with self-ratings, partner
ratings, and measured intelligence. Journal of Personality
and Social Psychology, 65, 546–553.
Funding
Brooks, A. W., Huang, L., Kearney, S. W., & Murray, F. E.
This research was supported by the Booth School of Business. (2014). Investors prefer entrepreneurial ventures pitched
by attractive men. Proceedings of the National Academy of
Supplemental Material Sciences, USA, 111, 4427–4431.
Demoulin, S., Leyens, J., Paladino, M., Rodriguez-Torres, R.,
Additional supporting information can be found at http://pss
Rodriguez-Perez, A., & Dovidio, J. (2004). Dimensions of
.sagepub.com/content/by/supplemental-data
“uniquely” and “non-uniquely” human emotions. Cognition
& Emotion, 18, 71–96.
Open Practices Dennett, D. C. (1987). The intentional stance. Cambridge, MA:
MIT Press.
Eagly, A. H., & Mladinic, A. (1989). Gender stereotypes and
All the data and some of the materials for these experiments attitude toward women and men. Personality and Social
have been made publicly available via Open Science Framework Psychology Bulletin, 15, 543–558.
and can be accessed at https://osf.io/e3afr/. Because the stimuli Farah, M. J., & Heberlein, A. S. (2007). Personhood and neu-
contain identifying information, it would violate participants’ roscience: Naturalizing or nihilating? American Journal of
confidentiality to post them on a public Web site. To obtain the Bioethics, 7, 37–48.
stimuli for use in scientific research, please e-mail either Juliana Fiske, S. T., Cuddy, A. J. C., & Glick, P. (2007). Universal dimen-
Schroeder (jschroeder@chicagobooth.edu) or Nicholas Epley sions of social cognition: Warmth, then competence. Trends
(epley@chicagobooth.edu). The complete Open Practices in Cognitive Sciences, 11, 77–83.
Disclosure for this article can be found at http://pss.sagepub Gesn, P., & Ickes, W. (1999). The development of meaning
.com/content/by/supplemental-data. This article has received and contexts for empathic accuracy: Channel and sequence
the badge for Open Data. More information about the Open effects. Journal of Personality and Social Psychology, 77,
Practices badges can be found at https://osf.io/tvyxz/wiki/ 746–761.
view/ and http://pss.sagepub.com/content/25/1/3.full. Gilbert, D. T., & Malone, P. S. (1995). The correspondence bias.
Psychological Bulletin, 117, 21–38.
Glick, P., Zion, C., & Nelson, C. (1988). What mediates sex dis-
Notes crimination in hiring decisions? Journal of Personality and
1. We arrived at these suggested time limits by asking two Social Psychology, 55, 178–186.
M.B.A. students to create spoken and written pitches without Gray, H. M., Gray, K., & Wegner, D. M. (2007). Dimensions of
any time restrictions and then timing how long it took for them mind perception. Science, 315, 619.
to speak and write these pitches. Gregory, S. W., & Webster, S. (1996). A nonverbal signal in
2. One research assistant transcribed the spoken pitches, and voices of interview partners effectively predicts communica-
a second checked the transcriptions for accuracy. In order to tion accommodation and social status perceptions. Journal
make the transcripts more readable, we removed verbal filler of Personality and Social Psychology, 70, 1231–1240.
words unless their exclusion changed a sentence’s meaning. Hall, J. A., & Schmid Mast, M. (2007). Sources of accuracy in the
3. We slightly modified the pitches for this experiment by empathic accuracy paradigm. Emotion, 7, 438–446.
removing salutations (e.g., “Dear Employer”) and signatures Harris, L. T., & Fiske, S. T. (2006). Dehumanizing the lowest of
(e.g., “Sincerely, X”) from the letters in order to facilitate trans- the low: Neuroimaging responses to extreme out-groups.
lation to voice. Psychological Science, 17, 847–853.

Downloaded from pss.sagepub.com at UNIV CALIFORNIA BERKELEY LIB on July 20, 2015
Speech Conveys Intellect 891

Haslam, N., Bain, P., Douge, L., Lee, M., & Bastian, B. (2005). More Mehrabian, A., & Wiener, M. (1967). Decoding of inconsis-
human than you: Attributing humanness to self and others. tent communications. Journal of Personality and Social
Journal of Personality and Social Psychology, 89, 937–950. Psychology, 6, 109–114.
Haslam, N., & Loughnan, S. (2014). Dehumanization and infra- Murphy, N. A. (2007). Appearing smart: The impression man-
humanization. Annual Review of Psychology, 65, 399–423. agement of intelligence, person perception accuracy,
Hughes, S. M., Mogilski, J. K., & Harrison, M. A. (2014). The and behavior in social interaction. Personality and Social
perception and parameters of intentional voice manipula- Psychology Bulletin, 33, 325–339.
tion. Journal of Nonverbal Behavior, 38, 107–127. Nass, C., & Brave, S. (2005). Wired for speech: How voice acti-
Ickes, W. (2003). Everyday mind reading: Understanding what vates and enhances the human-computer relationship.
other people think and feel. Amherst, NY: Prometheus Cambridge, MA: MIT Press.
Books. O’Brien, E., & Ellsworth, P. C. (2012). More than skin deep:
Imada, A. S., & Hakel, M. D. (1977). Influence of nonverbal Visceral states are not projected onto dissimilar others.
communication and rater proximity on impressions and Psychological Science, 23, 391–396.
decisions in simulated employment interviews. Journal of Pinker, S., & Bloom, P. (1990). Natural language and natural
Applied Psychology, 62, 295–300. selection. Behavioral & Brain Sciences, 13, 707–784.
Jones, B. C., Feinberg, D. R., DeBruine, L. M., Little, A. C., & Schroeder, J., & Epley, N. (2015). The humanizing voice.
Vukovic, J. (2010). A domain-specific opposite-sex bias in Manuscript in preparation.
human preferences for manipulated voice pitch. Animal Stewart, G. L., Dustin, S. L., Barrick, M. R., & Darnold, T. C.
Behaviour, 79, 57–62. (2008). Exploring the handshake in employment inter-
Judd, C. M., Westfall, J., & Kenny, D. A. (2012). Treating stimuli views. Journal of Applied Psychology, 93, 1139–1146.
as a random factor in social psychology: A new and compre- Takayama, L., & Nass, C. (2008). Driver safety and information
hensive solution to a pervasive but largely ignored problem. from afar: An experimental driving simulator study of wire-
Journal of Personality and Social Psychology, 103, 54–69. less vs. in-car information services. International Journal of
Kant, I. (2007). Critique of pure reason. London, England: Human-Computer Studies, 66, 173–184.
Penguin. (Original work published 1781) Van Boven, L., & Loewenstein, G. (2003). Social projection of
Kruger, J., Epley, N., Parker, J., & Ng, Z.-W. (2005). Ego- transient drive states. Personality and Social Psychology
centrism over e-mail: Can we communicate as well as we Bulletin, 9, 1159–1168.
think? Journal of Personality and Social Psychology, 89, Waytz, A., Heafner, J., & Epley, N. (2014). The mind in the
925–936. machine: Anthropomorphism increases trust in an autono-
Laplante, D., & Ambady, N. (2003). Of how things are said: mous vehicle. Journal of Experimental Social Psychology,
Voice tone, voice intensity, verbal content, and perceptions 52, 113–117.
of politeness. Journal of Language and Social Psychology, Waytz, A., Schroeder, J., & Epley, N. (2014). The lesser minds
22, 434–441. problem. In P. G. Bain, J. Vaes, & J.-P. Leyens (Eds.),
List, J. A. (2003). Does market experience eliminate market Humanness and dehumanization (pp. 49–67). New York,
anomalies? The Quarterly Journal of Economics, 118, 41–71. NY: Psychology Press.
Locke, J. (1997). An essay concerning human understanding. Zaki, J., Bolger, N., & Ochsner, K. (2009). Unpacking the infor-
Harmondsworth, England: Penguin Books. (Original work mational bases of empathic accuracy. Emotion, 9, 478–487.
published 1841) Zuckerman, M., & Driver, R. (1989). What sounds beautiful
MacKinnon, D. P., Fairchild, A. J., & Fritz, M. S. (2007). Mediation is good: The vocal attractiveness stereotype. Journal of
analysis. Annual Review of Psychology, 58, 593–614. Nonverbal Behavior, 13, 67–82.

Downloaded from pss.sagepub.com at UNIV CALIFORNIA BERKELEY LIB on July 20, 2015

View publication stats