The Persuasive Effect of Source
Credibility: Tests of
Cognitive Response
BRIAN STERNTHAL
RUBY DHOLAKIA
CLARK LEAVITT*
Two experiments are reported identifying the circumstances In which
high credibility either facilitates, inhibits, or has no effect on the communica-
tor's persuasiveness in relation to a less credible source. These data pro-
vide support for the cognitive response view of information processing
and suggest the importance of message recipient's initial opinion as a
determinant of persuasion.
I
n experimental investigations of the persuasive ef-
fect of source credibility, it has been frequently
demonstrated that highly trustworthy and expert
spokespeople induce a greater positive attitude toward
the position they advocate than do communicators
with less credibility (cf. Stemthal, Phillips, and
Dholakia in press). This finding can be explained in
terms of cognitive response (cf. Greenwald 1968,
1970;
Petty, Ostrom, and Brock 1978). According to
this formulation, a message recipient's initial opinion
is an important determinant of influence. In response
to a persuasive appeal, individuals rehearse their issue-
relevant thoughts, as well as those presented to them.
Message rejection occurs when people opposed to the
communicator's advocacy review counterarguments to
assertions made to the message. If a highly credible
source inhibits counterarguing. whereas a less credible
source does not, cognitive response predicts the su-
perior persuasive power of a highly credible com-
municator. Consistent with this interpretation. Cook
(1969) reported less counterargumentation in response
to a competent source than to an incompetent source.
Despite the substantial number of studies indicating
that a highly credible source is more persuasive than a
low credibility source, this
finding
is less than univocal.
• Brian Stemthal is Assistant Professor, Depanment of Marketing,
Nonhwestem University, Evanston, IL
60201.
Ruby Dholakia
is Assistant Professor of Marketing, Indian Institute of Manage-
ment, Calcutta, India. Clark Leavitt is Professor. Department of
Marketing, Ohio State University, Columbus, OH 43210. The au-
thors wish to thank Bobby Calder, Anthony Greenwald, Alice
Tybout. and Peter Wright for their comments on drafts of
this
paper.
McGinnies (1973) reported that a highly credible source
was more persuasive than a less credible communica-
tor when a message recipient's initial opinion toward
the advocacy was very negative, but no credibility
efTect emerged when they held a less negative initial
opinion. Dholakia and Sternthal (1977) found no
systematic credibility effect, even though the highly
credible source was perceived to be more expert and
trustworthy than the low credibility communicator.
Dean, Austin, and Watts (1971) reported that their
highly credible source induced greater persuasion when
message recipients were adamantly opposed to the
communicator's position, but the less credible source
was more influential when the issue was one toward
which individuals were likely to have had a favorable
predisposition (e.g., counseling against an annual
x-ray). Similarly, Bochner and Insko (1966) observed
that a highly credible source was more persuasive
than a moderately credible source when the advocacy
was highly discrepant, but that the moderately credible
source induced somewhat greater persuasion when the
position advocated was relatively close to their initial
opinion. Finally, Bock and Saine (1975) found that a
low credibility source was more persuasive than a
highly credible source when research participants
favored the advocacy.
Although the majority of these investigations were
not conceived as tests of cognitive response (an ex-
ception is Dholakia and Sternthal 1977), their find-
ings can be interpreted in terms of this formulation.
Specifically, the finding that a highly credible source
exhibits greater persuasive power than one of lower
credibility is consistent with the cognitive response
JOURNAL OF CONSUMER RESEARCH • Vol. 4 • March 1978
EFFECT OF SOURCE CREDIBILITY
253
prediction If it is assumed that message recipients had
a negative predisposition toward the advocacy. In
this situation, a credible source is more likely to inhibit
counterargumentation than a source whose credibility
is moderate or low. In turn, the reduction of counter-
argumentation stimulates persuasion.
The finding that a source who lacks credibility is
more persuasive than a highly credible communicator
is explained by cognitive response if it is assumed that
individuals have a positive predisposition toward the
message issue. In cognitive response terms, those
favoring the advocacy will feel a greater need to
insure that a position with which they agree is being
adequately represented when the source's credibility
is low than when the source's credibility is high.
Therefore, it is expected that these individuals will
generate more support arguments and will be more
persuaded by a source lacking in credibility. The find-
ing that credibility has no systematic persuasive
effect is consistent with cognitive response if it is
assumed that individuals had a neutral initial opinion
toward the issue. In this case, they are likely to
generate a similar number of support arguments and
counterarguments in response to the appeal.
In sum. by making reasonable assumptions about a
message recipient's initial opinions, cognitive response
can order the persuasive main effect of credibility
reported in previous investigations. A more compelling
test of the cognitive response explanation of the credi-
bility-persuasion relationship requires:
1.
the systematic manipulation of source credibility.
2.
a knowledge of the individual s initial opinion
toward the communication issue, and
3.
the measurement of thoughts as well as attitudes
in response to an appeal.
The purpose of the present investigation was to pro-
vide such a test. To this end, two experiments were
conducted. In Experiment I, the persuasive impact of
a
high and a moderate credibility source were examined
in a context where research participants were likely
to have a favorable initial opinion toward the message.
In Experiment II, this test of cognitive response was
extended by investigating the source credibility effect
for subjects who had a negative initial opinion. The
persuasive effect of credibility for all subjects, regard-
less of their initial opinion, was also examined in
Experiment II.
This analysis was motivated by the desire to deter-
mine whether Dholakia and Sternthal's (1977) failure
to observe a credibility main effect was attributable
to the aggregation of responses from individuals
varying in initial opinion within a credibility treat-
ment. In both experiments, measures of attitudes and
behavior as well as measures of support arguments
and counterarguments were administered.
EXPERIMENT I: METHOD
Overview
Experiment I employed a 2 x 2 factorial design.
Subjects who were likely to have a favorable initial
opinion toward the issue were presented an appeal
that was attributed to either a high or moderate
credibility source. For half the subjects in each treat-
ment the communicator was identified prior to the mes-
sage,
while for the remainder his identification was de-
ferred until the end of the communication. After
reading the appeal, its persuasive impact was meas-
ured on five indices: attitudes, intentions, support
arguments, counterarguments, and behavioral re-
sponse.
On the basis of cognitive response, it is hypothesized
that the interaction between source credibility and
timing of the source's identification will be signifi-
cant. Specifically, it is expected that a moderately
credible source will induce greater persuasion than a
highly credible source when communicator identifica-
tion precedes the message. As noted earlier, in this
situation the moderately credible source is expected
to stimulate greater support argumentation and, there-
fore,
greater acceptance than the highly credible source.
In contrast, when source identification follows the
message, credibility cues are made available too late
to affect the message recipient's thought generation
process. As a result, it is predicted that the message
will serve as the only influence cue, and source
credibility will have no systematic persuasive effect.
The timing variable was chosen because it provides a
way of determining the reliability of the credibility
effects obtained in this experiment. Previous investiga-
tions have demonstrated that there is no systematic
effect of credibility when source identification follows
the message (cf. Ward and McGinnies 1974). If this
finding is replicated, then confirmation of the cogni-
tive response prediction that a source who lacks
credibility will induce greater persuasion than a highly
credible communicator when source identification
precedes the message is unlikely to be due to some
unknown factor peculiar to the present study.
Issue
Given the objective of the present research, the com-
munication issue had to be one toward which message
recipients evinced a positive initial disposition. At the
same time, it was important that subject's predisposi-
tion was not so favorable as to preclude attitude
change because of a ceiling effect. In a pretest, it
was found that passage of the Consumer Protection
Agency Bill (CPAB) fulfilled these criteria. Forty sub-
jects,
drawn from the same pool as those who partici-
pated in the main study, exhibited a favorable predis-
position toward the bill on two 7-point Likert-type
scales. Specifically, subjects indicated agreement with
254
THE JOURNAL OF CONSUMER RESEARCH
"freedom of private business results in the exploita-
tion of consumers,"
(AT
= 4.73. 5.D. =
1.39),
and dis-
agreement with "consumer interests are most ef-
fectively protected by business," {X = 1.40, S.D.
= .65), two of the items employed in the main study
to identify the attitudinal effects of the independent
variables. For both these initial opinion items, mean
attitudes were significantly more positive than the mid-
point of the scale."
Subjects
The 56 participants in this study were recruited from
two sections of an undergraduate management course.
During one of the regular class hours, the students
were asked by their class instructor (who was not the
experimenter) to participate in a study of federal
legislation. Participation was voluntary. The sponsors
of the study were described as an external group
who had sought the assistance of the course instructor
to administer the study. All the students agreed to
participate.
The use of a convenience sample comprised of stu-
dents is appropriate, given that the principal aim of the
present research is to detect the relation between vari-
ables of theoretical significance. In such theory centered
research, there are three requirements for external va-
lidity (Kruglanski 1975). First, the theoretical variables
must be captured in the experimental setting (construct
validity). Second, the operationalizations of the inde-
pendent variables must be causally related to the
operationalizations of the dependent variables (internal
validity). Third, the causal relationships observed
should not occur by chance (statistical conclusion
validity). In contrast, those aspects of the situation
that are theoretically irrelevant, in this instance sample
representativeness, need not be considered. In fact,
given that the present study entails theory-oriented
research where individual differences are not of
theoretical interest, between-subject variance is treated
as random error that
is
appropriately kept to
a
minimum
by selecting homogeneous samples (Kruglanski 1975).
Therefore, the selection of a student sample for the
present study in no way undermines the generation of
inferences regarding the persuasive impact of source
credibility or the adequacy of cognitive response in
ordering the observed effects.
Procedure
Subjects were given booklets containing the inde-
pendent and dependent variables by their instructor.
The experimental task involved reading a one-page
message dealing with the CPAB (S.707) that was pend-
ing before the United States Senate. The message pre-
U = 2.80, df = 38. p < .01 for statement one and t = 21.67.
df
=
38, p < .001 for statement two. All t tests are one-tailed
unless otherwise noted.
sented arguments in favor of the bill and stated the
functions of the proposed agency. Depending upon the
treatment to which subjects were assigned, the message
was attributed to either a high or moderate credibility
source, who was identified either before or after the
message. These experimental conditions were ad-
ministered in both classes to avoid confounding the
effects of the independent variables. Once they had
read the communication, subjects were asked to com-
plete a questionnaire that included the dependent
measures.
Independent Variables
Two independent variables were manipulated in the
study: source credibility and timing of source identifica-
tion. The source was described as a supporter of the
bill and the reasons for his advocacy were given.
In the high credibility condition, the source was de-
scribed as a Harvard-trained lawyer with extensive
experience in the area of consumer issues and a
recognized expert whose advice was widely sought.
His support of this particular bill was made distinctive
by describing him as a person who generally did not
favor government controls, but who supported the
CPAB.
The less credible source was portrayed as an
individual with no special expertise, but one who
was interested in consumer protection because of a
job opportunity as consumer lobbyist. He was de-
scribed as holding socialistic views with strong opin-
ions in favor of all government controls.
The second independent variable involved manipu-
lating the timing of the source's identification. For
half the subjects, the source description appeared
before the written communication, while for the re-
maining subjects the source description was deferred
until the end of the communication.
Measuring Instruments
Five measures of persuasion, as well as a credibility
manipulation check, served as the dependent variables.
Message recipient's attitude toward the bill was deter-
mined first by administering the two attitudinal items
used in the pretest, plus two items that focused more
explicitly on the position advocated in the appeal:
"Passage of the Consumer Protection Agency Bill
would make me feel good" and "It is my belief that
consumers do not need protection." Responses to these
questions were measured on a seven-point scale rang-
ing from strongly agree to strongly disagree. The four
attitude questions were summed lo form an overall
score because they were highly interrelated (Cronbaeh
a = .64). These attitude measures were followed by a
behavioral intention item where subjects indicated
their likelihood of supporting the bill on a lOO-point
scale.
The next measure tapped message recipients' sup-
port arguments and counterarguments. Argumentation
EFFECT OF SOURCE CREDIBILITY
255
was determined
by
having subjects list all the thoughts
that came
to
mind
in
response
to the
appeal. Three
minutes were given
to
complete this task. Subjects
then categorized their thoughts
as
either favorable
or
unfavorable
to
passage
of the
bill.
For
each subject,
the thoughts listed that were consistent with
the
posi-
tion taken
in the
message were summed
to
yield
a
support argument score, while
the
thoughts listed that
opposed
the
message position were summed
to
yield
a counterargument score.
The argumentation measurements were followed
by
two source evaluation measures.
One
measured
sub-
jects'
perception
of
the trustworthiness
and
expertise
of
the
source.
Six
items (three related
to
expertise
and three related
to
trust), each rated
on a
seven-
point semantic differential scale, were used for this pur-
pose.
Items included: expert — not expert; experienced
—not experienced; trained—untrained; trustworthy
—not trustworthy; moral — immoral; good — bad.
A second
set of
dependent variables which
was not
manipulated
in the
present study was related
to
attrac-
tiveness
of the
source. These nonequivalent source
credibility items included: attractive
—
unattractive;
dynamic — not dynamic: aggressive — not aggressive.
These items
are
nonequivalent
in the
sense that
un-
like responses
to the
expertise
and
trust scales,
re-
sponses to the attractiveness items should not be sensi-
tive to the dependent variable. They were administered
to determine whether demand characteristics
ac-
counted
for the
subject's perception
of the
source's
credibility (Orne 1969).
If the
source credibility
in-
duction
was
effective
and not
attributable
to
demand
characteristics, then subjects would perceive
the
highly credible source
to be
more trustworthy
and
expert than
the
moderately credible source,
but
they
should
not
find
one
source more attractive than
the
other.
Behavioral compliance
was
determined
by
whether
subjects signed
a
petition that
was lo be
sent
to
their
senators
in
Washington urging
a
vote
for the
bill.
RESULTS
Manipulation Check
A manipulation check determined whether
the
source credibility induction was effective. Subjects who
received
the
message from
the
highly credible source
perceived
the
communicator
to be
significantly more
trustworthy and expert than did subjects who received
the message from
the
moderately credible person.^
There was no difference between the high and moderate
credibility sources
on
the nonequivalent attractiveness
measures.^ Since the credibility manipulation involved
' High credibility:
X =
28.86.
S.D. = 8.64, n = 28. Low
credi-
bility:
A*
=
20.57.
S.D. =
8.01,
;i =
28,
f =
i.n.df
= 54.
p < .001,
> High credibility:
X =
13.07.
S.D. = 4.52. Low
credibility:
X
= 13.68, 5.Z).
= 3,90.
varying trustworthiness
and
expertise
but not
attrac-
tiveness, these data indicate that the credibility induc-
tion
was
effective
and
unlikely
to be
attributable
to
demand characteristics.
Although
the
means
for the
highly
and
moderately
credible sources differ significantly
on
the trustworthi-
ness and expertise dimensions, the moderately credible
source
was not
perceived
in a
totally negative light.
In fact, there was no difference between the moderately
credible source
and the
scale midpoint
(/ <
1). Given
that
the
experiment entailed having
a
source
ask for
compliance with
a
request
to
sign
a
petition,
it was
necessary that
the
less credible source have some
favorable attributes. Indeed,
in
pretesting
the
credi-
bility induction
it was
found that subjects perceived
the situation
to be
contrived when
the
moderately
credible source's biography was less favorable than the
one used
in
this experiment.
Attitudinal Response
Subjects' attitudina! responses categorized
by
inde-
pendent variables
are
reported
in
Table
1. To
deter-
mine
the
effects ofthe experimental variables
on sub-
jects'
attitudes toward
the
CPAB,
an
analysis
of
variance was performed employing the sum ofthe atti-
tudinal responses
as the
dependent measure.
It was
found that both source credibility
(F < 1) and
timing
of
the
source's identification
(F = 1.34, df- 1,52,
p
>
.20)
did not
have
a
significant effect
on
attitudes.
As predicted, however,
a
significant disordinal interac-
tion between source credibility
and the
time
of the
source identification was found
(f" =
15.97,1^/=
1,52,
p
<
.001).
The
moderately credible source
was
more
persuasive than
the
high credibility communicator
when the credibility cue was presented before the mes-
sage,
whereas
the
highly credible source induced
a
more positive attitude when the credibility cue followed
the communication
as
shown
in the
figure.
Although
a
significant source credibility
x
timing
interaction
is
necessary
to
confirm cognitive response
predictions,
it is not
sufficient.
It
must also
be
demon-
strated that:
1.
the
moderately credible source
is
significantly
more persuasive than
the
highly credible source
when
the
source credibility cues precede
the
message,
and
2.
the
high
and
moderate credibility sources
do not
differ significantly
in
their persuasive power
when
the
credibility cues appear after
the mes-
sage.
Using
the
Newman-Keuls procedure
to
make these
contrasts, the above predictions were confirmed. When
source identification preceded the message, the moder-
ately credible source induced
a
more positive attitude
than
the
highly credible source
(q =
4.96,
df= 52,
p
<
.05), whereas there
was no
systematic credibility
256
THE JOURNAL OF CONSUMER RESEARCH
TABLE 1
MEAN (X) AND STANDARD DEVIATIONS (S.D.) FOR ATTITUDES, SUPPORT ARGUMENTS COUNTERARGUMENTS AND
INTENTIONS CATEGORIZED BY INDEPENDENT VARIABLES, SOURCE CREDIBlLlPi', AND
TIMING OF THE CREDIBILITY INDUCTION
Timing of the
credibility induction
Source before message (SM)
Source after message (MS)
Dependent
measure
Attitude
Support arguments
Counterarguments
Intention
Attitude
Support arguments
Counterarguments
Intention
X
19,64
0.71
1.43
62.14
22.71
1.29
0.79
78.29
Source
High
S.D.
5.17
1.20
1.34
28.06
2.89
1.82
1.05
15.23
credibility
Moderate
X
25.00
1.86
0.93
89.00
19.43
0.93
1.57
59.64
S.D.
3,04
1.92
1,00
11.90
4.60
1.21
1.87
27.16
effect when the source identification was deferred until
after the message (<? = 2.97, df
-=
52, p > .05). Fur-
thermore, the high credibility source induced a sig-
nificantly more positive attitude toward the bill when
the source was identified after—rather than before
the message {q = 2.78, df = 52, p = .05). However,
the moderately credible source was significantly more
influential when introduced before the message (q
= 5.\5, df= 52, F < .01).
Argumentation-Counterargumentation
A more compelling test of cognitive response re-
quires operationalizing the variables presumed to medi-
FIGURE
SOURCE CREDIBILITY x TIMING OF CREDIBILITY INDUCTION
Attitude
25
24
23
22
21
20
19
Source-Message
Message Source
Timing of the Source
Credibility Induction
• -
High
Credibility Source
A-Moderate Credibility Source
ate persuasion — support argumentation and counter-
argumentation. To address this issue, the effects on
argumentation of source credibility and timing of the
source identification were examined (See Table 1).
The source credibility x timing of source identifica-
tion was marginally significant for support argumenta-
tion {F = 3.19, df = 1,52, p = .08) and counterargu-
mentation (F = 3.13, df= 1,52, p = .08). As pre-
dicted, when source identification preceded the
message, subjects generated more support arguments
in response to the moderately credible source than the
highly credible one (/ = 2.74, df = 54, p < .01). Also
congenial to the cognitive response formulation was the
finding of no differences in support argumentation
attributable to source credibility when source identifica-
tion occurred after the message (/ < 1) and the ob-
servation of no credibility effects on counterargu-
mentation (identification before message, / = 1.58,
df = 54, /? > .10; identification after message, t
= ].95, df= 54, p > .05).-
To evaluate the mediating role of support argu-
ments and counterarguments further, the effect ofthe
treatments on attitude was determined when the num-
ber of each of these thought types was treated as a
covariate. When the number of support arguments
served as a covariate, the independent variables still
had a significant effect on attitudes (F = 4.62, df
= 3,51, p < .01), Nevertheless, support arguments
accounted for a substantial and statistically significant
portion (38 percent) ofthe explained attitude variance
(F = 3.63, df= 3,51, p < .05). Similarly, when the
number of counterarguments was treated as a co-
variate, the treatment-attitude relationship was main-
tained (F = 5.01, df
=
3,51, p < .005), though coun-
terarguments accounted for a significant portion ofthe
explained variance (F = 3.87, df= 3,51,/? < .05).
* Two-tailed tests v^ere used in making these contrasts, because
no source credibility effects were predicted from cognitive re-
sponse.