The Influence of Speech Confidence on Choice and Decision Confidence

1
The Influence of Speech Confidence on Choice and
Decision Confidence

• Dr Briony Pulford
• Psychology Division
• University of Wolverhampton

2
Abstract Objectives The influence generated by
the confidence of a person's speech was
investigated, firstly on the choice of answers
that people make and secondly on the confidence
that they then hold in those answers. The
influence of the gender of the speaker and
listener on persuasive impact were also
investigated.   Design A mixed factorial
design was employed 3 (degree of confidence of
the text's speaker within subjects) x 2
(speakers' gender) x 2 (group). Choice of answer
and confidence in it were the DVs.   Methods 290
undergraduates participated and were asked to
choose the correct definition of 9 unusual words.
In the experimental group the three definitions
were supplemented by confidence cues
(previously developed in a pilot study)
indicating the confidence of the speaker to be
either high, medium or low, which the control
group did not receive. The genders of the three
speakers were either all male or all female.
Participants were also asked to give their
impressions of the three speakers, using 7 point
Likert scales.   Results Significant
interactions showed that high and low speech
confidence had the most aversive impact on the
choices of females who listened to males. Speech
confidence had some influence on choice and
raised listeners' confidence by around 10.
  Conclusions This research shows that the
gender of the speaker and listener does influence
whether confident speech is perceived positively
or negatively. Knowing the confidence of the
speaker raises confidence in one's own choices
and also influences the impressions formed about
the speaker's character.
3
Introduction

• The term confidence describes a persons belief
  that his or her knowledge, judgement or decision
  is correct.
• Both the police and jurors in a courtroom are
  faced with making judgements about how confident
  a person is, for example, a witness to a crime.
  When faced with a confident eyewitness or a
  non-confident one, which one would you believe?
• London, Meldman and Lanckton (1970) studied
  persuasion in simulated two-person juries and
  concluded that the single significant behavioral
  difference between persuaders and persuadees was
  in the expression of confidence (p. 182), and
  that in dyadic interactions persuasion is a
  function neither of intelligence, pre-discussion
  conviction, position, or ability but of the
  expression of confidence during the discussion.
  Whilst discussing an issue words are used that
  communicate confidence and doubt in one's own and
  the other's positions. London et al. questioned
  whether expressed confidence is a means by which
  informational influence is transmitted.

4

• When a group gets together to perform a task it
  is advantageous if they can identify the members
  of that group that have the best judgement, but
  identification of the best members is difficult
  (Einhorn, Hogarth, Klempner, 1977). It is made
  easier if members of the group can assess their
  confidence accurately and communicate it to the
  group (Sniezek Henry, 1989). But people are
  over or underconfident and may communicate their
  confidence poorly - either intentionally or
  unintentionally. Unintentional miscommunication
  may occur because the language used is imprecise
  or ambiguous (Sniezek Henry, 1989).
• There is also the problem of over- and
  under-confident people. Some people are not very
  good at monitoring their levels of confidence and
  may seem very sure they are right and persuade
  you to go along with them, only for you to find
  out later that they were wrong.

5

• Thomas and McFadyen (1995) argue for the presence
  of a confidence heuristic a short-cut to
  arriving at a decision in conditions of
  uncertainty that uses the confidence with which
  information is expressed to assess the
  reliability of the information. This heuristic,
  like many others, can be prone to error.
•  
• Our society may have different expectations about
  the confidence that may be expressed by men and
  women, and as such the impact of hearing
  different levels of confidence coming from men
  and women should differ according to one's own
  gender. In a society where modesty is feminine
  and confidence is masculine the low confidence
  male and the highly confident female may stand
  out from the crowd and as such may not be liked
  or as persuasive.
• That confidence has a role in persuasion in
  groups seems clear, but the way that confidence
  is communicated between group members remains to
  be discovered. Thus this study aims to find out
  how confident and non-confident language is
  perceived by listeners and what impact it has on
  their decisions and confidence.

6
Method

• Participants
• The 290 participants in the experiment were
  British undergraduate students and members of the
  general public who volunteered for the study (178
  women and 112 men, mean age 28.35 years, SD
  11.82 years).
• Materials
• A pilot study was carried out to generate a pool
  of 'confidence cues', for example, phrases such
  as "At a guess", "I'm not sure", "I think this
  is", "It's probably", "I'm certain that",
  "Without doubt" (see Table 1).

7
Table 1 Confidence Cues
8

• Once the perceived confidence of the speaker was
  known the cues were rank ordered and then three
  cues of differing levels of confidence (one from
  the top, middle and bottom of the range) were
  selected and attached as a prefix to three
  definitions of an unusual word, only one of which
  was the correct meaning. This was repeated for 9
  unusual words which were selected from a
  dictionary (Penannular, Tapotement, Meliorism,
  Crewel, Samite, Cheops, Gullah, Dorp, and
  Barysphere) giving 27 definitions with confidence
  cues attached.
• Obscure words were chosen to make the task
  difficult and the observed behaviour
  representative of how people behave when they
  feel uncertain (which is when they turn to other
  people for advice).
• The gender of the speaker was varied, so that
  half of the questionnaires had male speakers
  names only (John, Ben, Mark) and the rest had
  female speakers (Sarah, Liz, Anne). In the
  Experimental condition the highly confident
  speakers were John and Sarah, the medium
  confidence speakers were Ben and Liz, and the low
  confidence speakers were Mark and Anne.
• 1) Barysphere
• John This is easy, its an early type of
  submersible diving bell.
• Ben I think this is a layer of the earths
  interior.
• Mark Im not sure but it could be a type of
  mineral that has a round appearance.

9
Design and Procedure The study was a 2 (group
confidence cues or no confidence cues) x 2
(speakers gender) x 2 (listeners gender) x 3
(speakers confidence level) mixed design with
repeated measures on the last variable. The first
dependent variable was the percentage of the time
that each speakers answer (John/Sarah, Ben/Liz,
Mark/Anne the high, medium and low confidence
speakers respectively in the experimental
condition abbreviated to A, B and C in the
statistical analyses) was chosen out of nine
questions. The second and third dependent
variables were the mean confidence and accuracy
scores of the participants answers when they
agreed with the high, medium and low confidence
speakers. The control condition was used to
determine the percentage of the population that
chose each of the three answers for each
definition (question) when they were not phrased
with any confidence cues attached to them. The
participants were asked to circle which one of
the three speakers gave the answer they thought
was correct, and how confident they felt that
their choice of answer was correct.
10
Finally, participants were also asked to give
their impressions of the three speakers, using 7
point response scales to assess intelligence (1
highly intelligent, 7 quite low
intelligence), friendliness (1 not at all
friendly, 7 extremely friendly),
self-confidence (1 high self-confidence, 7
quite low self-confidence), competence (1 not
very competent, 7 very competent),
optimism/pessimism (1 optimist, 7 pessimist).
They also indicated which person they liked most
and least, and trusted most and least. Each of
the three speakers gave the correct answer the
same number of times so that judgements made
about them would be based upon their portrayed
confidence and not their actual accuracy and
knowledge. Therefore if participants in the
experimental condition tend to agree with one of
the speakers rather than the others then this
must be due to the speaker's confidence level and
not their accuracy.
11
Results

• An ANCOVA with age as a covariate showed that
  accuracy was based on the participants knowledge
  and was not affected by any independent variable
  in the experiment.
• The Influence of Speakers Confidence on Choice
  of Answer
• When the confidence cues were added in the
  experimental condition the participants showed a
  small but significant switch in their preference
  of answer, with 4.43 fewer participants choosing
  the high confidence speaker, 7.46 more choosing
  the medium, and 3.04 (ns) less choosing the low
  confidence speaker, F(2, 572) 7.54, p .001.
• To make understanding the interactions simpler
  the bars in Figures 1 and 2 represent the
  differences between the mean scores for the
  control group and the experimental group, who saw
  the confidence cues. Thus a score near to zero
  indicates that there was no difference between
  the two groups and the independent variables did
  not affect the experimental groups answers.

12
Figure 1 The Effect of Gender, Speakers Gender
and Speakers Confidence on Choice of Answer
13
Low confidence speech seems to have little effect
on the attractiveness of an answer when people
are uncertain of themselves, whereas highly
confident speech seems off-putting to some people
who then choose the answer given by the medium
confidence speaker instead. Figure 1 shows that
males in the experimental condition did not alter
their choice of answer when exposed to the high,
medium and low confidence cues, F(2, 216) .21,
p .81, compared with the control group. The
difference scores for males in Figure 1 are all
close to zero, for all conditions, showing that
males are unaffected by the gender or confidence
of the speaker they are listening to. Females
show a 6.82 move away from the answers which are
given by the highly confident speaker, t(176)
2.77, p .01, a non-significant drop of 2.87
when the answer is spoken by a low confidence
speaker, t(176) 1.16, p .25, and a
significant increase of 9.69 in choosing the
answer given by the medium confidence speaker,
t(176) 3.95, p .0001. Figure 1 show that the
highly confident speaker is off-putting to some
female listeners but the gender of the speaker is
not important here. However, females show a
dislike of the low confidence male speaker (a
6.92 drop in choosing the answer when low
confidence cues are given, t(88) 2.02, p lt .05)
with a large shift towards the medium confidence
male (a 13.31 rise in choosing the answer when
medium confidence cues are given, t(88) 3.72, p
.001), but do not seem to be put off an answer
given by a low confidence female speaker (a
non-significant 1.40 rise).
14
The Influence of Speakers Confidence on
Participants Confidence Males were
significantly more confident than females in the
control condition (41 vs 30) but both showed
the same level of confidence in the experimental
condition (44 and 42) as females confidence
rose but males confidence did not, F(1, 260)
4.45, p .05. The addition of confidence cues
(irrespective of the level of confidence)
significantly raised female participants
confidence by 11.74, F(1, 158) 18.93, p
.001, by an average of 17.44 when listening to
women and by 7.02 when listening to men, this
interaction was approaching significance, F(1,
158) 3.44, p .07. The presence of confidence
cues did not, however, raise the males
confidence levels significantly (2.60, F(1, 98)
1.04, p .31). There is, however, a
significant interaction of group with the
speakers level of confidence, which shows that
highly confident speakers significantly raise
males confidence by 8.78 (t(109) 2.27, p
.03) but that medium and low confidence speakers
did not raise confidence at all for male
listeners (1.22 and 0.25) The situation for
female listeners was different though, there was
a nearly significant interaction between
speakers level of confidence and group, F(2,
316) 2.82, p .06, showing that the high
confidence speaker raised womens confidence by
15.19, the medium confidence speaker raised it
by 9.60 and the low confidence speaker raised it
by 10.05 (all at p lt .001).
15
Figure 2, the Effect of Gender, Speakers Gender
and Spealers Confidence on Confidence
16
In order to make the four way interaction more
easily understood Figure 2 shows the change in
confidence between the control and the
experimental groups that is attributable to the
introduction of the confidence cues. Figure 2
shows how mens confidence is unaffected by
listening to women. When men hear medium and low
confidence female speakers their confidence drops
fractionally (-5.22, and 2.10 respectively,
both non-significant). High confidence women
speakers have more of an influence than medium
and low confidence female speakers but still do
not significantly increase mens confidence
(7.49). When women listen to men they raise
their confidence most when the man is high in
confidence (11.28), less when the man has medium
confidence (8.05) and non-significantly (2.58)
when the man has low confidence. Womens
confidence is boosted the most when they listen
to other women, especially when the speaker has
high or low confidence, but all levels of
confidence spoken by women significantly raise
womens confidence in their answers.
17
Attributions About the Speakers   The control and
experimental groups significantly differed in
which speaker they liked most and it can be
concluded that speaking with high confidence
makes the majority of listeners develop a dislike
for the speaker, and the majority of listeners
like the medium confidence speaker the most.
  The confidence cues attached to the
definitions made participants in the experimental
group like and trust the medium confidence
speaker the most, and dislike and mistrust the
highly confident speaker. The trend was the same
for male and female listeners, but was more
pronounced for females as about 10 more females
than males switched to the medium confidence
speaker for who they liked and trusted most, and
around 15 more females than males disliked and
mistrusted the high confidence speaker the most.
18
Fig 3 Attributions about Speakers
19
Figure 3 shows that ratings of intelligence and
competence ratings did not change much when
confidence cues were added in the experimental
group. In this experiment confidence cues do not
seem to influence these attributions. Changes
in ratings when confidence cues are added show
that medium and low confidence speakers were
rated as much more friendly than the high
confidence speaker, who in turn was rated as more
self-confident (2) than both of them (the medium
confidence speaker was correctly assigned a score
of 4 which was average confidence and the low
confidence speaker scored 5 which is lower than
average confidence, additional evidence that the
participants had detected the confidence cues).
The ratings of optimism showed that the highly
confident speaker was seen as an optimist, and
the low confidence speaker a pessimist, which the
control group did not perceive them as, and the
medium confidence speakers rating of optimism
was unaffected when compared with the control
group.
20
Discussion
To conclude, the results of this experiment show
that mens confidence is only raised 9 by highly
confident speakers, and their choice of answer is
unaffected by the speakers confidence. Womens
confidence is significantly raised by 12 when
they hear how confident the speaker feels in
their answer, and the high confidence speaker
raises womens confidence the most (15),
although the high confidence speaker can lead to
a 7 drop in the number of women agreeing with
their answer. The medium confidence speaker
attracts women to agree with them 10 more often
and promotes confidence by 10. The low
confidence speaker does not put women off their
answer and also raises confidence by 10.
Within the boundaries of this experimental
situation it appears that if you wish to persuade
a woman that the answer that you give is correct
then speak with medium confidence, especially if
you are a man. If you are trying to persuade a
man then it does not matter how confidently you
speak as all levels of confidence will have no
effect on the choice that he makes, but the only
way to raise his confidence in his answer is to
speak with a high level of confidence yourself
(especially if you are a woman). There may be
gender differences in the ability to detect
confidence cues, meaning that they are used when
they are perceived, but men perceive them less
often. This remains to be investigated.
21
Although the control group had no confidence tags
attached it cannot be claimed that they had no
implicit confidence within them. The absence of
doubtful confidence cues may make the control
statements seem fairly factual and certain. This
needs investigating. The experiment showed that
people liked the medium confidence speaker more
than the high and low conf speaker, and thus
likeability of the speaker may have been a factor
used in choice of answer when they were
completely uncertain. Thomas and McFadyen
(1995) say that the confidence heuristic is
simply to adopt the option which is most
confidently argued for (p. 105). My research
shows that the most confident speaker is not the
one people adopt, women preferred the medium
confidence speaker and men were unaffected by
confident speech, thus disputing their definition
of the confidence heuristic. It may be that
people do recognise that the relationship between
confidence and accuracy is not a strong one, it
was only around r .17 in this experiment, and
this is why they are more persuaded by the medium
than the high confidence speaker, who they may
label as overconfident. This research now needs
extending to see if confidence cues are only used
in a low personal relevance situation where
heuristics are used, or whether in fact they also
have an influence when people are using deeper
processing strategies.
22
References

• Einhorn, H. J., Hogarth, R. M., Klempner, E.
  (1977). Quality of group judgment. Psychological
  Bulletin, 84, 158-172.
• London, H., Meldman, P. J., Lanckton, A. V. C.
  (1970). The jury method How the persuader
  persuades. Public Opinion Quarterly, 34, 171-183.
  
• Sniezek, J. A., Henry, R. A. (1989). Accuracy
  and confidence in group judgment. Organizational
  Behavior and Human Decision Processes 43, 1-28.
• Thomas, J. P., McFadyen, R. G. (1995). The
  confidence heuristic A game-theoretic analysis.
  Journal of Economic Psychology, 16, 97-113.

23
(No Transcript)