The Persuasive Effects of Metaphor: A Meta‐Analysis

Abstract

Empirical investigations of metaphor's persuasive effects have produced mixed results. In an effort to integrate the literature, we present a review and meta-analytic summary of existing studies. Six explanations for the potential suasory advantage of metaphor over literal language were reviewed: (a) pleasure or relief, (b) communicator credibility, (c) reduced counterarguments, (d) resource-matching, (e) stimulated elaboration, and (f) superior organization. Next, a meta-analysis was conducted and the impact of seven moderator variables was tested. The overall effect for the metaphor-literal comparison for attitude change was r = .07, which supported the claim that metaphors enhance persuasion. The effect rose to r = .42 under optimal conditions, when a single, nonextended metaphor was novel, had a familiar target, and was used early in a message. Metaphor appeared to exert a small effect on perceptions of source dynamism (r = .06), but showed no demonstrable impact on competence (r =−.01) or character (r =−.02). Of the six theories considered, the superior organization explanation of metaphor's persuasive impact was most supported by the results.

Full text

382 HUMAN COMMUNICATION RESEARCH / July 2002
Human Communication Research, Vol. 28 No. 3, July 2002 382–419
© 2002 International Communication Association
The Persuasive Effects of Metaphor
A Meta-Analysis
PRADEEP SOPORY
University of Memphis
JAMES PRICE DILLARD
University of Wisconsin-Madison
Empirical investigations of metaphor’s persuasive effects have produced mixed results. In an
effort to integrate the literature, we present a review and meta-analytic summary of existing
studies. Six explanations for the potential suasory advantage of metaphor over literal lan-
guage were reviewed: (a) pleasure or relief, (b) communicator credibility, (c) reduced
counterarguments, (d) resource-matching, (e) stimulated elaboration, and (f) superior orga-
nization. Next, a meta-analysis was conducted and the impact of seven moderator variables
was tested. The overall effect for the metaphor-literal comparison for attitude change was r =
.07, which supported the claim that metaphors enhance persuasion. The effect rose to r = .42
under optimal conditions, when a single, nonextended metaphor was novel, had a familiar
target, and was used early in a message. Metaphor appeared to exert a small effect on
perceptions of source dynamism (r = .06), but showed no demonstrable impact on compe-
tence (r = -.01) or character (r = -.02). Of the six theories considered, the superior organiza-
tion explanation of metaphor’s persuasive impact was most supported by the results.
Derived from the Greek words “meta,” meaning “over,” and
“pherein,” meaning “to carry,” metaphor is traditionally defined
as an implied comparison between two dissimilar objects, such
that the comparison results in aspects that normally apply to one object
being transferred or carried over to the second object. Metaphor is cred-
ited with the capacity to structure, transform, and create new knowledge,
as well as evoke emotions, and influence evaluations (Aristotle, trans.
1952a, 1952b; Black, 1962; Crocker, 1977; Lakoff & Turner, 1989;
MacCormac, 1985; Osborn & Ehninger, 1962; Richards, 1936). Perhaps as
Pradeep Sopory (Ph.D., University of Wisconsin-Madison, 1999) is assistant professor of com-
munication at the University of Memphis. James Price Dillard (Ph.D., Michigan State Univer-
sity, 1983) is professor of communication at the University of Wisconsin-Madison. An ear-
lier version of this article was presented at the annual conference of the International Com-
munication Association, Chicago, 1996. This article is part of the doctoral dissertation of the
first author under the supervision of the second author. Thanks to Edward Fink and two
anonymous reviewers for their comments. Correspondence concerning this article should be
addressed to Pradeep Sopory, Department of Communication, The University of Memphis, 143
Theatre/Communication Bldg., Memphis, TN 38152; email: psopory@cc.memphis.edu.

Sopory, Dillard / METAPHOR AND PERSUASION 383
a consequence, metaphor is ubiquitous in persuasion contexts ranging
from politics (Graesser, Mio, & Millis, 1989; Voss, Kennet, Wiley, & Engstler-
Schooler, 1992) to consumer advertising (Boozer, Wyld, & Grant, 1991;
Stern, 1988).
Are metaphors really effective persuasion devices? The results of many
investigations are mixed. Consequently, the question of the superiority of
metaphor over literal language remains open. Moreover, the process by
which such effects may arise is the focus of several competing theories.
Our goals in this study were twofold: (a) to provide an answer regarding
metaphor’s effectiveness and (b) to suggest a framework for addressing
the process. In the service of these aims, the existing literature was sur-
veyed to identify theoretical perspectives that bore on the metaphor-per-
suasion relationship. Using these theories as a guide, we derived hypoth-
eses and tested them via meta-analysis. Before proceeding to the review
and meta-analysis, however, it is important to locate metaphor in a
context provided by the terminology and the theories of metaphor
comprehension.
BACKGROUND
Terminology
Metaphors are linguistic comparisons of the form “A is B.” For example,
the expression “Television is poison” consists of two parts A (television)
and B (poison). A and B are different concepts or conceptual domains;
metaphor links one to the other. The terminology associated with A and B
varies from theorist to theorist (e.g., Black, 1962, 1979; Richards, 1936). In
keeping with more recent usage, we call A the target and B the base
(Gentner, 1982). These terms capture a fundamental feature of metaphor,
the notion that meaning is passed from B to A. Three other tropes, simile,
analogy, and personification are similar to metaphor in that they also in-
volve comparison of concepts or systems of concepts.1 Although clearly
distinct as linguistic devices, these tropes very likely instantiate cognitive
processes similar to those induced by metaphor. Thus, for present pur-
poses we do not distinguish among them.
Theories of Metaphor Comprehension
To explain how metaphor may achieve its persuasive impact, it is nec-
essary to describe the theories that deal with its understanding. Three
such theories are especially germane.2 The literal-primacy view (Beardsley,
1962, 1976; MacCormac, 1985; Searle, 1979) sees metaphor as exceptional
language that is literally false. That is, metaphors are semantic anoma-
lies, which consequently require three stages in the process of understand-

384 HUMAN COMMUNICATION RESEARCH / July 2002
ing: (a) derive the literal meaning of a metaphor, (b) test whether the lit-
eral meaning makes sense and detect an anomaly or a violation of seman-
tic rules, and (c) seek an alternative meaning (i.e., the metaphorical mean-
ing) when the literal meaning fails to make sense (see Gibbs, 1994, for an
elaborated discussion). According to one variation of this view (e.g.,
MacCormac, 1985), when an interpreter confronts an anomaly (i.e., an
expression that is literally false), cognitive tension is generated along with
a desire to reduce it. By finding the nonliteral meaning of the literally
false statement, the anomaly is resolved for the interpreter, and the ten-
sion is dissipated.
While the literal-primacy view treats metaphorical language as devi-
ant and exceptional, two other positions reject the notion of metaphor as
anomalous. Unlike the literal-primacy view, these theories assume that
metaphoricity and literalness of language is a matter of degree, and that
the same general psychological mechanism underlies the processing of
both forms of language.
Ortony’s (1979; Ortony, Vondruska, Foss, & Jones, 1985) salience-imbal-
ance theory uses the notion of attribute salience to explain how metaphors
are comprehended. Salience is defined as the relative importance of an
attribute. Empirically, the first attribute that comes to mind is the most
salient, and so on. A metaphorical expression of the type “A is B” is un-
derstood by constructing the set of shared attributes, then selecting those
attributes that have low salience for the target and high salience for the
base. For example, “Encyclopedias are goldmines” is understood by iden-
tifying attributes such as “valuable nuggets” and “dig,” which have a
high salience for “goldmines” and a low salience for “encyclopedias.” If
the two terms are reversed (i.e., “Goldmines are encyclopedias”) then a
different set of the shared attributes would be chosen, because the attributes
that would be highly salient for “encyclopedias” would be different.
Gentner’s (1982, 1983, 1989; Gentner & Clements, 1988) structure-map-
ping theory builds from associative network models of memory. With re-
gard to comprehension, instead of comparing lists of attributes, it pro-
poses that the relations among the attributes themselves are compared
for similarities. Gentner (1983) linked metaphor explicitly to analogy and
defined a metaphor as “an assertion that a relational structure that nor-
mally applies in one domain can be applied in another domain” (p. 156).
This view posits that metaphors convey a system of connected knowl-
edge, not a mere collection of independent facts. In interpreting a meta-
phor people strive for maximum structural match between target and
base by seeking a relational mapping. For example, the metaphor “Ency-
clopedias are goldmines” is interpreted by noting the common relation
“valuable nuggets found by digging” rather than the independent simi-
lar attributes “valuable nuggets” and “dig.”

Sopory, Dillard / METAPHOR AND PERSUASION 385
PERSPECTIVES ON METAPHOR AND PERSUASION
Metaphor and persuasion research prior to the 1980s utilized the lit-
eral-primacy view of metaphor comprehension. More recent perspectives
have drawn from the salience-imbalance and structure-mapping theories
of metaphor comprehension. Six perspectives on metaphor and persua-
sion are presented below in rough chronological order.
Pleasure or Relief
There are two strands to this view, pleasure or relief, however both
argue that a metaphorical expression is a semantic anomaly, the recogni-
tion of which leads to negative tension (Bowers & Osborn, 1966; Reinsch,
1970, 1973; Tudman, 1971). When the metaphorical meaning is finally
understood the negative tension is relieved. Three stages are involved,
perception of error (or novelty), conflict (or recoil), and resolution. In the first
strand, resolving the metaphorical meaning and thus finding the “unex-
pected similarities” between target and base is pleasurable. According to
the second strand, finding the metaphorical meaning dissipates the nega-
tive tension, leading to relief. The reward of pleasure or relief leads to
reinforcement of the metaphorical meaning and the evaluation associ-
ated with it. In contrast, literal language does not pose any linguistic puzzle
to resolve and consequently yields neither pleasure nor relief. The prin-
ciple of the pleasure or relief view is that the dissipation of negative ten-
sion reinforces the metaphorical meaning and evaluation, thereby increas-
ing persuasion.
Communicator Credibility
The communicator credibility view proposes that communicators who
use metaphors are judged more credible than ones who use literal lan-
guage (Bowers & Osborn, 1966; McCroskey & Combs, 1969; Osborn &
Ehninger, 1962; Reinsch, 1970). Credibility enhancement may occur for
two related reasons. First, as Aristotle (trans. 1952a) argued in the Poetics,
“The greatest thing by far is to be a master of metaphor. It is the one thing
that cannot be learnt from others; and it is also a sign of genius” (p. 255).
Thus, speakers who use metaphors should be judged quite positively.
The second reason (Bowers & Osborn, 1966; Osborn & Ehninger, 1962),
derived from the literal-primacy view, proposes that metaphors may point
out previously unknown similarities between entities. This newfound ap-
preciation of commonalties is a source of interest and pleasure to the
comprehender. Impressed by the message source, the receiver returns a
favorable judgment of communicator credibility.

386 HUMAN COMMUNICATION RESEARCH / July 2002
Reduced Counterarguments
According to the reduced counterarguments view (Guthrie, 1972), the
process of comprehending a metaphor generates a great number of asso-
ciations that result in “an overload in the receiver’s mental circuitry” (p.
4). A high proportion of the cognitive resources of a comprehender are
used up when encountering a metaphorical persuasive message and as a
result fewer resources are left to “derogate or exclude the message con-
tent or the source” (p. 4). The outcome is greater agreement with what is
advocated by the message. Implicitly, this view assumes that all message
recipients are predisposed to counterargue a message regardless of what
it is advocating. The key concept of this view is that processing of meta-
phorical messages consumes more cognitive resources than literal ones,
which disrupts counterarguing, leading to increased persuasion.
Resource Matching
A more sophisticated use of the cognitive resources idea is offered by
the resource matching view (Jaffe, 1988). This perspective proposes that
deriving meaning of a metaphorical expression demands cognitive elabo-
ration (Ortony, 1979), which ensures a better integration of (high quality)
message arguments in memory and ultimately a greater persuasion rela-
tive to the literal message. However, elaboration also requires greater
mobilization of cognitive resources. When there is a match between the
resources required to understand the metaphorical message and the re-
sources available to the interpreter, maximum elaboration and thus maxi-
mum comprehension occurs. When there is a mismatch in the form of too
few cognitive resources, less comprehension occurs and persuasion is in-
hibited. On the other hand, if excess resources are available (for example,
clichéd metaphors), irrelevant idiosyncratic thoughts are generated that
dilute the persuasive impact of the message content. In this view meta-
phors have a persuasive advantage over literal language only under re-
source-enhanced conditions, such as message repetition, where the fa-
miliarity generated by repetition allows a match of resources for a novel
metaphorical message, and excess resources for a literal message.
Stimulated Elaboration
The stimulated elaboration view is linked to two different metaphor
processing theories. Whaley (1991) used structure-mapping theory
(Gentner, 1982, 1989) to propose that understanding metaphors stimu-
lates thought by focusing on a similar relational structure (rather than
simple features) between target and base, and hence evoking a richer set
of associations in semantic memory as compared to literal language. This
greater number of semantic connections produces greater elaboration of

Sopory, Dillard / METAPHOR AND PERSUASION 387
message content, which in turn leads to increased persuasion given suit-
able processing conditions. Whaley proposed that certain types of meta-
phors (what he called explanatory analogies, based on Gentner, 1982) func-
tion as high-quality arguments, whose processing results in more elabo-
ration than that of literal messages. Therefore, if both motivation and abil-
ity are high and the message is compelling, the outcome is a greater num-
ber of thoughts agreeing with message advocacy and thereby greater per-
suasion (cf. Chaiken, Liberman, & Eagly, 1989; Petty & Cacioppo, 1986).
Using concepts of salience-imbalance theory (Ortony, 1979), Hitchon
(1991) proposed that when the common features of target and base are
assembled together to comprehend a metaphor (the assembly is called
the ground), the evaluation associated with the features is also part of the
ground (see also Ottati, Rhoads, & Graesser, 1999). In this view, forma-
tion of the ground requires elaboration of the ground-relevant attributes
as well as their associated valence. Thus, elaboration leads to a greater
number of valenced thoughts, which (when in the appropriate direction)
lead to greater persuasion. In contrast, to extract the meaning of a literal
expression, there is no need to find the ground and hence no need to
elaborate the message content. To summarize, the persuasion principle of
the stimulated elaboration view is that metaphors lead to enhanced per-
suasion by inducing a greater amount of appropriately valenced (agree-
ment with message) thinking than literal-only messages.
Superior Organization
The superior organization view, also derived from Gentner’s (1982,
1989) structure-mapping theory, proposes that a metaphor helps to struc-
ture and organize the arguments of a message better than literal language
(Read, Cesa, Jones, & Collins, 1990; see also Mio, 1996, for a similar view).
A metaphor evokes a greater number of semantic associations. When these
associations are consistent with the metaphor, the different arguments
are connected more coherently via the many available semantic pathways.
Additionally, the links to the metaphor “highlight” the arguments mak-
ing them more salient. Consequently, interpreters are apt to find it easier
to relate the arguments to each other, and because they are highlighted,
are more likely to process them. This view implies that more coherent
organization and the resulting highlighting of the arguments improves
the comprehension of message arguments. Then, as McGuire (1972, 1985)
has suggested, enhanced comprehension should increase persuasion. The
key idea here is that metaphor provides relevant semantic associations
by which arguments of a message become linked, leading to higher
persuasion. Literal language does not so readily afford this orga-
nizing function.

388 HUMAN COMMUNICATION RESEARCH / July 2002
PREDICTED EFFECTS OF METAPHOR ON ATTITUDE AND
COMMUNICATOR JUDGMENTS
A variety of hypotheses can be derived from the above theories, and as
might be expected, many of these hypotheses stand in contradiction to
one another. This is desirable in that establishing the empirical veracity
of various hypotheses may produce a pattern of evidence that favors one
or more theories over the others. In the paragraphs below, we present a
series of hypotheses that follow from the six theories. Space constraints
prohibit a detailed explication of the derivation of the predictions from
the theories, but the logical relationship of each hypothesis to each of the
six theories is summarized in Table 1.
The Effect of Metaphor on Attitude and Communicator Credibility
For different reasons, each of the six theories discussed previously pre-
dicts that metaphors should show suasory superiority over literal lan-
guage. Thus, all of the theories point to the following:
H1: Metaphorical messages are more persuasive than literal messages.
Message receivers can be asked to judge the credibility of the message
communicator at two points during message reception: once, before they
begin processing the message, and second, after they finish processing
the message. We call the premessage perception of message source initial
credibility and the postmessage perception terminal credibility. Many writ-
ers have asserted that communicators who use metaphors are judged more
favorably than those who use literal language (e.g., Aristotle, trans. 1952a;
Bowers & Osborn, 1966; McCroskey & Combs, 1969; Osborn & Ehninger,
1962; Reinsch, 1970). Thus, use of metaphor should enhance the terminal
credibility of the speaker. However, credibility is not a unitary construct
(e.g., Berlo, Lemert, & Mertz, 1969; Bowers & Phillips, 1967; McCroskey,
1966; McCroskey & Young, 1981). Three concepts are common across stud-
ies: competence, character, and dynamism.3 In-line with the argument that
metaphor enhances credibility, we hypothesize that metaphor use will
increase favorable postmessage judgments of the three subcomponents
of credibility:
H2: Use of metaphors should produce more favorable source judgments for
all three aspects of terminal credibility: (a) competence, (b) character, and
(c) dynamism.

Sopory, Dillard / METAPHOR AND PERSUASION 389
Variables Expected to Influence the Magnitude of the Metaphor Effect on
Attitude
Initial communicator credibility. Message receivers can judge the
premessage or initial credibility of a communicator as high or low. All
metaphor and persuasion views imply that metaphor-using communica-
tors with high initial credibility should have greater impact on attitude
than ones having low credibility. Thus, we have:
H3: The effectiveness of metaphor usage is positively related to initial commu-
nicator credibility.
Extendedness of metaphor. Metaphors work by comparing the target do-
main with the base domain. An extended metaphor uses one base to con-
struct a number of different submetaphors with the same target. For ex-
ample, the base “death” may be used for the following metaphors, all
with the target “legalizing marijuana”: “To argue that marijuana is no
worse than alcohol is to argue that premeditated homicide is no worse
than accidental homicide,” “legalizing marijuana [is] . . . inviting the kiss
of death,” and “to try the death weed, marijuana” (from Siltanen, 1981).
Resource matching and superior organization views suggest extended
metaphors should be more persuasive, while pleasure or relief, credibil-
ity, reduced counterarguing, and stimulated elaboration views imply the
opposite expectation. Thus, two hypotheses are offered:
H4a: Extended metaphors are more persuasive than nonextended metaphors.
H4b: Extended metaphors are less persuasive than nonextended metaphors.
Number of metaphors. According to the pleasure/relief, credibility, re-
duced counterarguing, and stimulated elaboration views, use of more
rather than fewer metaphors in a message (all trying to change attitudes
towards the same target) should lead to more persuasion. Resource-match-
ing and superior organization make the opposite prediction. Therefore,
two hypotheses are offered:
H5a: Persuasion is a positive function of the number of metaphors contained
in a message.
H5b: Persuasion is a negative function of the number of metaphors contained
in a message.
Position of metaphors. In principle, metaphors may be placed at the be-
ginning, middle, or end of a message. Placement in the introduction should

390 HUMAN COMMUNICATION RESEARCH / July 2002
lead to enhanced persuasion according to superior organization view,
while the rest of the views suggest a greater impact with a conclusion
positioning. Thus, the following predictions are proposed:
H6a: Metaphors in the introduction of a message are more persuasive than in
any other position.
H6b: Metaphors in the conclusion of a message are more persuasive than in
any other position.
Familiarity of target. The target and base of a metaphor may have vary-
ing degrees of familiarity for a message recipient. To facilitate transport
of information from base to target, the familiarity of base is customarily
high. On the other hand, the target term of a metaphor may be familiar or
unfamiliar to the subjects in a particular study. For example (from Read
et al., 1990), “Aid to Zaire is like . . . ” is a low familiar target for North
American undergraduate audiences, while “Seat belt use is like . . . ” is
probably a high familiar target. A higher familiarity of target should lead
to greater persuasion according to resource-matching and superior orga-
nization views, while a lower familiarity should lend itself to increased
persuasion according to pleasure/relief, credibility, reduced counterar-
guing, and stimulated elaboration views. Thus, we have:
H7a: Persuasion is directly related to familiarity of target.
H7b: Persuasion is inversely related to familiarity of target.
Novelty of metaphors. A distinction is usually made between novel and
conventionalized (Lakoff & Johnson, 1980) or “dead” (Black, 1962) meta-
phors. Novel metaphors are expressions that create new information about
the target. For example, the expression “Don’t worry about John’s future.
He is a caterpillar—and will metamorphose into a butterfly soon” may
provide novel information about John. Conventionalized metaphors are
metaphors that were once novel, but with repeated usage have been com-
pletely absorbed into the conventions of everyday language (e.g., “The
leg of a table”). Such metaphors are not immediately recognized as meta-
phors. Clichés are also overused and not novel (e.g., “Magic Johnson flies
like an eagle on the basketball court”), but are immediately recognized as
metaphors. All but the resource matching perspective suggest that nov-
elty should enhance persuasive effectiveness. Thus, we have:
H8a: Persuasion is directly related to metaphor novelty.
H8b: Persuasion is inversely related to metaphor novelty.

Sopory, Dillard / METAPHOR AND PERSUASION 391
Modality of presentation. Persuasive messages may be encountered
through a variety of media. The differences between reading and listen-
ing include the fact that reading allows for more processing time as well
as multiple reviews of the message (van Dijk, 1987). In contrast, when
listening in real time a person can process a message only once. Pleasure
or relief, resource matching, stimulated elaboration, and superior organi-
zation perspectives predict superiority for the written mode, while com-
municator credibility and reduced counterarguing suggest the opposite.
Accordingly, the following predictions are presented:
H9a: Metaphors presented in the written modality should be more persuasive
than metaphors presented in the audio modality.
H9b: Metaphors presented in the audio modality should be more persuasive
than metaphors presented in the written modality.
METHOD
Sample of Studies
The literature search was directed towards data-based sources. First,
we conducted a computer search of CommSearch, PsychLit, PsychInfo,
WorldCat, Social Science Abstracts, and Dissertation Abstracts using the search
string (metaphor, analogy, simile, figurative, or trope) and (attitude, cred-
ibility, or persuasion). We covered all years for which electronic search
was possible in any given database (e.g., 1983–2000 for Social Science Ab-
stracts). A wildcard symbol was used at an appropriate place for each key
word to account for plurals and other variants of the search terms. Sec-
ond, the reference/bibliography section of articles found through the
search as well as of other relevant articles was surveyed. The search un-
covered 41 data-based studies.
All the studies were examined for suitability for inclusion in the meta-
analysis. The criteria for selection were: (a) the studies must permit quan-
titative assessment of the effects of (b) metaphorical versus literal mes-
sages on (c) attitude toward target or credibility of message source. Using
these criteria, 17 studies were excluded.4 For example, both Peracchio and
Luna (1998) and Zaltman and Coulter (1995) were excluded as the stud-
ies examined persuasive effects of metaphor qualitatively. One article
(Edwards & Clevenger, 1990) was excluded because the study was con-
ducted on the production rather than effects of metaphor, and another
(Johnson & Taylor, 1981) was excluded because the design did not in-
clude a metaphor-literal comparison. Most studies that were excluded
examined dependent variables other than attitude toward the attitude
target, such as attitude towards the message/advertisement (Dingena,

392 HUMAN COMMUNICATION RESEARCH / July 2002
TABLE 1
Logical Relationships of Hypotheses to Theoretical Perspectives
Hypotheses for variables expected
to moderate the effect of metaphor
on attitude Theoretical relationship
Initial source credibility
H3: High > Low
Pleasure or relief If source less credible, metaphor labeled
anomalous or absurd, so more easily dismissed
Communicator credibility Metaphor judged as more creative if source more
credible
Reduced counterargument Less counterarguing against message from high
credibility source
Resource matching Message from high credibility source processed
more readily, so greater chance of resource match
Stimulated elaboration More thinking about metaphor when message
from high credibility source
Superior organization More coherent organization of information around
metaphor when message from high credibility
source
Metaphor extendedness
H4a: Extended > Nonextended*
Resource matching Use of one base only reduces resource demands,
so increase in resource match
Superior organization Use of one base only increases coherent
organization of information
H4b: Nonextended > Extended
Pleasure or relief Use of one base only, so diminished pleasure or
relief and reinforcement
Communicator credibility Use of one base only, so source judged less
creatively versatile
Reduced counterargument Use of one base only reduces resource demands
thus increasing counterarguing
Stimulated elaboration Use of one base only, so reduction in elaboration
Number of metaphors
H5a: Higher > Less
Pleasure or relief More pleasure or relief with more metaphors, so
more reinforcement
Communicator credibility More metaphors indicate greater creativity of
source
Reduced counterargument More metaphors require more resources, so less
counterarguing
Stimulated elaboration Increased elaboration with more metaphors
H5b: Less > Higher*
Resource matching Fewer metaphors require fewer resources, so
increase in resource match
Superior organization Less interference to coherent organization of
information with fewer metaphors

Sopory, Dillard / METAPHOR AND PERSUASION 393
Metaphor position
H6a: Introduction > Conclusion**
Superior organization Position at message start acts as “title” to facilitate
coherent organization of remaining message
H6b: Conclusion > Introduction
Pleasure or relief Pleasure or relief not dissipated and more readily
available for reinforcement at message end
Communicator credibility Positive source judgment at message end
amplifies advocacy; at start discourages deep
processing of message
Reduced counterargument Increased cognitive load at message end
suppresses availability of counterarguments to
entire message
Resource matching Consumption of resources at message start hinders
resource match for comprehension of remaining
message
Stimulated elaboration Increased elaborations at message start interferes
with comprehension of remaining message
Metaphor target familiarity
H7a: High > Low*
Resource matching More resources available for metaphor processing,
so resource match likely
Superior organization Facilitates coherent organization of target-base
linkages with message arguments
H7b: Low > High
Pleasure or relief Metaphor more puzzling with low familiarity, so
increased pleasure or relief when meaning derived
Communicator credibility Target-base commonalities more striking when low
familiarity, so increased source appreciation
Reduced counterargument Increased resource demands for metaphor
processing with low familiarity, so less
counterarguing
Stimulated elaboration Low familiarity calls for increased elaboration to
construct ground and meaning
Metaphor novelty
H8a: High > Low**
Pleasure or relief Novel metaphor more puzzling, so increased
pleasure or relief when meaning derived
Communicator credibility A greater “sign of genius;” increased appreciation
of message source
Reduced counterargument More effortful to process, so less counterarguing
Stimulated elaboration Greater elaboration required to construct the
ground and derive meaning
TABLE 1 Continued
Logical Relationships of Hypotheses to Theoretical Perspectives
Hypotheses for variables expected
to moderate the effect of metaphor
on attitude Theoretical relationship

394 HUMAN COMMUNICATION RESEARCH / July 2002
Superior organization Unique target-base linkages affect greater
convergence of message arguments around metaphor
H8b: Low > High
Resource matching More resource intensive to process, so greater
chance of resource mismatch
Modality of presentation
H9a: Written > Audio
Pleasure or relief Greater time required to process metaphor, which
is afforded by written presentation
Resource matching More time for multiple passes over message allows
more chance of resource match
Stimulated elaboration More time to process, so greater elaboration
Superior organization More time to process, so greater development of
coherent message organization
H9b: Audio > Written*
Communicator credibility Less time available to process message, so source
judgment has greater impact
Reduced counterargument Difficulty of metaphor understanding increases
with limited time, inhibiting counterarguing
NOTE: See text for details of the theoretical perspectives and derivation of H1 and H2. Of
the 6 theoretical relationships, only those directly following a hypothesis are linked to it.
* indicates directional support for the hypothesis; ** indicates directional support and sig-
nificant difference between groups p < .05.
TABLE 1 Continued
Logical Relationships of Hypotheses to Theoretical Perspectives
Hypotheses for variables expected
to moderate the effect of metaphor
on attitude Theoretical relationship
1994; McQuarrie & Mick, 1999; Phillips, 1997), logical inferences (Antrim,
1979; Bosman, 1987; Bosman & Hagendoorn, 1991), metaphor identifica-
tion (Graesser, Mio, & Millis, 1989), and comprehension (Mitchell,
Badzinski, & Pawlowski, 1994).
Two unpublished master’s theses (Mangum, 1994; Shannon, 1981) fit
the three criteria but had to be excluded as we were unable to extract an
effect size given the available statistical information. Of the published
articles, Reinsch (1971) and Siltanen (1981) were based on a master’s the-
sis and Reinsch (1974) on a Ph.D. dissertation. More data were reported
in Reinsch’s actual master’s thesis (1970) and dissertation (1973), which
consequently replaced the articles (Reinsch, 1971, and 1974, respectively).
Siltanen (1981) was retained as no additional data were found in the thesis.
All told, 24 data-based sources (indicated by an asterisk in the refer-
ences section) were used in the meta-analysis.5 Some of the sources con-
tained multiple experiments. We considered each experiment as a sepa-
rate study, and thus the total number of studies was 29, which were all

Sopory, Dillard / METAPHOR AND PERSUASION 395
used for the attitude analysis, with the total number of subjects N = 3945.
These numbers were somewhat smaller for the credibility analyses: for
competence, number of studies = 12, N = 1938; for character, number of
studies = 12, N = 1936; and for dynamism, number of studies = 7 and N =
1252. The distribution of number of studies and subjects across type of
publication (published versus unpublished), year of publication (pre-1980
versus post-1980), and study design (pretest-posttest versus posttest only)
was generally consistent.6
Characteristics of Studies
The studies used a wide variety of attitude objects as their persuasion
targets. The metaphors used in the messages also exhibited great vari-
ability in terms of bases and targets. In most cases, the attitude object was
also the metaphor target, but for a few studies the metaphor targets were
simply related (albeit closely) to the attitude object. The messages in
many studies were attributed to a specific person, but some studies
had nonperson sources for the message (e.g., a newspaper editorial
or an advertisement).7
Coding. The moderator variables of interest were identified in the studies
based on their potential for influencing the persuasion process. Variables
were manipulated in some studies, but in the majority of cases, we made
a judgment of the value of the variable. See Tables 2 and 3 for the modera-
tor variables and their values.
The values of familiarity of target, novelty of metaphor, and source
credibility were coded independently by the two authors.8 To determine
these author-coded values, decision rules were formed and then the two
coders independently assigned a value to the variables for a study. The
intercoder reliability using Cohen’s kappa was .90 for target familiarity,
.88 for metaphor novelty, and .96 for initial communicator credibility. Dis-
agreements over judgments were resolved by refining the rules. The judg-
ment criteria are given below.
To judge the familiarity of target as low or high, it was necessary to see
the target from the perspective and in the context of the subjects in the
studies (e.g., keeping the year in mind when the experiment was done).
For example, the target of “AZT effects on the AIDS virus” (Whaley, 1991)
was more likely to be high than low familiarity for undergraduate stu-
dents in the early 1990s. To the best of our ability, we coded familiarity
from the perspective of the study participants.
The novelty of metaphors was defined in terms of the extent of knowl-
edge of similar connections between the target and base of a metaphor
that recipients (in this case subjects in a particular study) would have
prior to encountering the metaphorical message. If the similarities exist,
novelty is low; if the similarities do not exist, novelty is high. Similar to

396 HUMAN COMMUNICATION RESEARCH / July 2002
TABLE 2
Effect Sizes of Attitude for Metaphor-Literal Comparison
Author(s), date, & Initial
experiment # (if needed) credibility Extendedness Number Position Familiarity Novelty Modality r N weight
Bowers & Osborn (1966) High Extended 8 Conclusion ? Low Audio 0.185 132
McCroskey & Combs (1969) High Extended ? Body Low High Written 0.149 264
Low Extended ? Body Low High Written 0.062 264
Reinsch (1970) High Nonextended 4 Body Low Low Audio 0.161 66
Tudman (1971) High Extended 11 Introduction High Low Written -0.035 43
& Conclusion
High Extended 4 Introduction High Low Written -0.023 43
High Extended 7 Conclusion High Low Written 0.012 43
Guthrie (1972) High Nonextended 30 Introduction, Low ? Written 0.044 65
Body, & Conclusion
Low Nonextended 30 Introduction, Low ? Written 0.005 65
Body, & Conclusion
Reinsch (1973) # 1 High Extended 3 Conclusion Low Low Written -0.153 66
High Nonextended 1 Conclusion Low Low Written -0.156 168
Reinsch (1973) # 2 High Extended 3 Body Low Low Written -0.063 43
High Nonextended 1 Body Low Low Written -0.079 42
Shelley (1974) High Nonextended 9 Conclusion High Low Written -0.254 30
Siltanen (1981) ? Extended 13 Conclusion High Low Audio 0.332 29
Jaffe (1988) # 1 Low Nonextended 1 Conclusion High Low Written -0.112 122
Jaffe (1988) # 3 Low Nonextended 1 Conclusion High Low Written 0.222 71
Jaffe (1988) # 4 Low Nonextended 1 Conclusion High Low Written 0.086 71
Read, Cesa, Jones, & ? Nonextended 1 Introduction & High High Audio 0.237 45
Collins (1990) # 2 Conclusion
Hitchon (1991) # 1 Low Nonextended 1 Introduction High High Written 0.400 72

Sopory, Dillard / METAPHOR AND PERSUASION 397
Hitchon (1991) # 2 Low Nonextended 1 Introduction High High Written 0.547 80
Whaley (1991) # 1 High Nonextended 6 Body & Conclusion High High Written 0.187 44
Whaley (1991) # 2 High Nonextended 2 Body High High Written -0.219 53
High Nonextended 6 Body & Conclusion High High Written -0.124 88
Brennan (1992) Low Extended 8 Introduction, Low High Written 0.216 192
Body & Conclusion
Mio, Thompson, & High Extended 1 Introduction High High Written 0.176 177
Givens (1993)
Nelson & Hitchon (1995) Low Nonextended 3 Introduction & Body High Low Written 0.217 68
Mio (1996) # 1 ? Nonextended 1 Introduction Low High Audio 0.133 106
Mio (1996) # 2 ? Extended ? Introduction & Body Low High Audio 0.035 190
Morgan (1997) High Nonextended 1 Introduction High Low Written 0.027 179
Fitzgerald (1998) # 1 Low Nonextended 1 Introduction Low High Written 0.201 135
Low Nonextended 1 Introduction Low Low Written -0.079 75
Fitzgerald (1998) # 2 Low Nonextended 1 Introduction Low High Written 0.103 135
Low Nonextended 1 Introduction Low Low Written -0.133 75
Pawlowski, Badzinski, & Low Nonextended 2 Introduction High High Written -0.108 62
Mitchell (1998)
Ottati, Rhoads, & High Extended 17 Introduction, High Low Audio -0.051 198
Graesser (1999) # 1 Body & Conclusion
Ottati, Rhoads, & High Extended 17 Introduction, High Low Audio 0.114 144
Graesser (1999) # 2 Body & Conclusion
Whaley & Wagner (2000) High Nonextended 1 Conclusion High High Written -0.024 200
NOTE: ? means that the value for the variable could not be judged.
TABLE 2 Continued
Effect Sizes of Attitude for Metaphor-Literal Comparison
Author(s), date, & Initial
experiment # (if needed) credibility Extendedness Number Position Familiarity Novelty Modality r N weight

398 HUMAN COMMUNICATION RESEARCH / July 2002
TABLE 3
Effect Sizes of Facets of Credibility for Metaphor-Literal Comparison
Author(s), date , & Facet of Initial N
experiment # (if needed) credibility* credibility Extendedness Number Position Familiarity Novelty Modality r weight
Bowers & Osborn (1966) Competence High Extended 8 Introduction ? Low Audio -0.094 132
Character High Extended 8 Conclusion ? Low Audio -0.086 132
McCroskey & Combs (1969) Competence High Extended ? Body Low High Written 0.011 264
Competence Low Extended ? Body Low High Written -0.008 264
Character High Extended ? Body Low High Written -0.008 264
Character Low Extended ? Body Low High Written -0.012 264
Dynamism High Extended ? Body Low High Written 0.027 264
Dynamism Low Extended ? Body Low High Written 0.027 264
Reinsch (1970) Competence High Nonextended 4 Body Low Low Audio 0.236 66
Character High Nonextended 4 Body Low Low Audio 0.118 66
Dynamism High Nonextended 4 Body Low Low Audio 0.146 66
Tudman (1971) Competence High Extended 11 Introduction High Low Written 0.000 43
& Conclusion
Competence High Extended 4 Introduction High Low Written -0.074 43
Competence High Extended 7 Conclusion High Low Written -0.044 43
Character High Extended 11 Introduction High Low Written -0.015 43
& Conclusion
Character High Extended 4 Introduction High Low Written -0.044 43
Character High Extended 7 Conclusion High Low Written -0.029 43
Guthrie (1972) Competence High Nonextended 30 Introduction, Body, Low ? Written 0.055 65
& Conclusion
Competence Low Nonextended 30 Introduction, Body, Low ? Written -0.150 65
& Conclusion
Character High Nonextended 30 Introduction, Body, Low ? Written 0.072 65
& Conclusion

Sopory, Dillard / METAPHOR AND PERSUASION 399
Character Low Nonextended 30 Introduction, Body, Low ? Written -0.227 65
& Conclusion
Dynamism High Nonextended 30 Introduction, Body, Low ? Written 0.017 65
& Conclusion
Dynamism Low Nonextended 30 Introduction, Body, Low ? Written 0.173 65
& Conclusion
Reinsch (1973) # 1 Competence High Extended 3 Conclusion Low Low Written -0.124 66
Competence High Nonextended 1 Conclusion Low Low Written 0.078 168
Character High Extended 3 Conclusion Low Low Written -0.235 66
Character High Nonextended 1 Conclusion Low Low Written -0.055 168
Dynamism High Extended 3 Conclusion Low Low Written 0.015 66
Dynamism High Nonextended 1 Conclusion Low Low Written 0.127 168
Reinsch (1973) # 2 Competence High Extended 3 Body Low Low Written 0.409 42
Competence High Nonextended 1 Body Low Low Written 0.427 42
Character High Extended 3 Body Low Low Written 0.207 41
Character High Nonextended 1 Body Low Low Written 0.211 41
Dynamism High Extended 3 Body Low Low Written 0.125 43
Dynamism High Nonextended 1 Body Low Low Written 0.366 42
Shelley (1974) Competence High Nonextended 9 Conclusion High Low Written -0.058 30
Character High Nonextended 9 Conclusion High Low Written -0.251 30
Dynamism High Nonextended 9 Conclusion High Low Written -0.118 30
Suit & Paradise (1985) Competence High Extended 1 Conclusion High High Audio -0.064 60
Competence High Nonextended 3 Conclusion High Low Audio -0.453 40
Character High Extended 1 Conclusion High High Audio 0.080 60
Character High Nonextended 3 Conclusion High Low Audio -0.199 40
TABLE 3 Continued
Effect Sizes of Facets of Credibility for Metaphor-Literal Comparison
Author(s), date , & Facet of Initial N
experiment # (if needed) credibility* credibility Extendedness Number Position Familiarity Novelty Modality r weight

400 HUMAN COMMUNICATION RESEARCH / July 2002
Read, Cesa, Jones, Competence Low Nonextended 1 Introduction Low High Written -0.192 63
& Collins (1990) # 1 Competence Low Nonextended 1 Introduction Low High Audio 0.197 63
Character Low Nonextended 1 Introduction Low High Written -0.150 63
Character Low Nonextended 1 Introduction Low High Audio 0.439 63
Rinalducci (1994) Competence Low Nonextended 10 Introduction, Body Low Low Video 0.046 179
& Conclusion
Character Low Nonextended 10 Introduction, Body Low Low Video 0.046 179
& Conclusion
Dynamism Low Nonextended 10 Introduction, Body Low Low Video 0.064 179
& Conclusion
Whaley (1998) Competence High Nonextended 1 Conclusion High High Written -0.200 200
Character High Nonextended 1 Conclusion High High Written -0.200 200
NOTE: ? means that the value for the variable could not be judged.
* Authoritativeness and qualification coded as competence; trustworthiness and ethical coded as character; extroversion coded as dynamism.
TABLE 3 Continued
Effect Sizes of Facets of Credibility for Metaphor-Literal Comparison
Author(s), date , & Facet of Initial N
experiment # (if needed) credibility* credibility Extendedness Number Position Familiarity Novelty Modality r weight

Sopory, Dillard / METAPHOR AND PERSUASION 401
target familiarity, we judged novelty from the perspective of the research
participants based on this definition of the newness of the similarity
relations.
Some studies used a number of metaphors, and the only solution was
to judge an “average” familiarity and novelty. For matters of convention
and simplicity, familiarity and novelty were assessed as low or high, but
for some studies the average was closer to a medium. Thus, the use of
“average” for these studies with multiple metaphors embodies our un-
certainty about their familiarity and novelty judgment.
For some studies, an estimate of initial communicator credibility (high or
low) was obtained from data on pretest credibility mentioned in the re-
sults section. For other studies, the actual pretest numbers were computed
and compared to the range of values for the measurement scales to deter-
mine initial credibility. For the rest of the studies, especially the ones in
which message sources were not persons (e.g., excerpts from political
speeches), the judgment was made based on whether the source stood to
gain from the message advocacy. For example, an ad for a consumer prod-
uct was judged to have low credibility because the message source clearly
benefited if the message advocacy was accepted by message recipients.
Criteria for determining the number of metaphors (1, 2–8, 9 or more),
extendedness of metaphors (extended or nonextended), and modality of pre-
sentation (written, audio, or video) were straightforward. Similarly, the
position of metaphors required no inference and was coded as “introduc-
tion only,” “conclusion only,” “body only” (i.e., middle of a message),
“introduction & conclusion,” “introduction, body, & conclusion,” and
“body & conclusion.” To further analyze whether placing metaphors in
the introduction facilitated persuasion, the metaphor position was also
coded according to where a metaphor first appeared in a message. Two
categories were created. The “first appearance in introduction” category
included the following metaphor placements: “introduction only,” “in-
troduction & conclusion,” and “introduction, body, & conclusion.” The
second category “first appearance in body or conclusion” included: “body
only,” “conclusion only,” and “body & conclusion.” The two authors in-
dependently rated each study and there were no disagreements for these
four moderator variables.
Data Management
Two groups of effect sizes were created, one for attitudes and the other
for credibility; the credibility data points were subdivided into compe-
tence, character, and dynamism. Some of the studies had multiple experi-
ments, each of which was treated as an independent study.
Unit of analysis. For the data from each study, an effect size for each
different value of the moderator variables was calculated. That is, we es-

402 HUMAN COMMUNICATION RESEARCH / July 2002
timated the effects based on cell-to-cell comparisons within studies. Each
estimate was treated as an independent data point. For example, Guthrie’s
(1972) study, with its 2 × 2 design produced two estimates. The first was
for metaphor versus literal in the high credibility condition and the other
was for metaphor versus literal in the low credibility condition.
Estimating effect size. The results of the studies were typically given in
terms of the t or F statistics, which we converted to r. Tables 2 and 3 pro-
vide the effect sizes for attitude and the three aspects of credibility re-
spectively. There were 38 estimates for attitude from 29 studies. For com-
petence there were 20 data points from 12 studies, for character 20 data
points from 12 studies, and for dynamism 11 data points from 7 studies.
The effect size estimates were constructed so that positive values indi-
cated that the metaphor was superior to the literal message and negative
values indicated the opposite.
Calculation of sample weights. The effect size of each cell was weighted
in accordance with the size of the sample that was used in its calculation.
In most cases these N weights were calculated by simply pooling the num-
ber of subjects in the two groups that were being compared. However, for
some studies the control group (usually the no-metaphor or the literal
group) was used for more than one comparison. For example, in Whaley
(1991) the literal group was first compared to the 2-analogy group and
then to the 4/6-analogy group. In order not to overweight the effect sizes
in such cases, the N weight was calculated by first dividing the N of the
control group by number of comparisons and then pooling the resulting
value with the N of the comparison group. The N weights appear in Tables
2 and 3 for attitude and credibility respectively.
Correcting for measurement error. Measurement error depresses the mag-
nitude of the correlation coefficient, thereby obscuring the true relation-
ship between variables. To account for measurement error in the depen-
dent variables (attitude, competence, character, and dynamism), the reli-
ability of the measurement scales was used as a correction within indi-
vidual studies. For studies that did not report reliabilities the following
procedure was used. First, the reliability and the number of scale items
for each study was noted. Second, from each study that reported this in-
formation, the average reliability and the average number of scale items
was calculated. Thus, using the artifact distribution, the average interitem
correlation was found. Then, for studies that did not report reliability, but
did provide information about the number of scale items, the reliability
was estimated using the average interitem correlation and the number of
scale items given for the study. For studies that did not give any informa-
tion about reliability, or number of scale items, the overall average reli-
ability calculated above was substituted. The average reliability for atti-
tude was .88, competence .85, character .92, and dynamism .84.

Sopory, Dillard / METAPHOR AND PERSUASION 403
Analysis
We took two different but complementary approaches to analysis of
the data. First, we conducted tests of hypotheses 1–9. These focused tests
allowed us to assess the degree of support for the particular claims ad-
vanced by each theory. Though informative, this approach to the data
suffers from one significant drawback: Unless the data are evenly distrib-
uted across all of the moderator variables, the hypothesis tests are not
orthogonal. In fact, correlations among the predictors (i.e., moderators)
ranged in size from .01 to .59 for attitude. This prompted us to pursue a
second means of summarizing the data: model fitting. This approach
emphasizes the creation of groups of studies where effect sizes are homo-
geneous within groups. The output of this procedure is a more parsimo-
nious representation of the data.
The relationship between hypothesis testing and model fitting may be
viewed as analogous to the relationship between bivariate and multiple
regression. Insofar as the hypothesis tests consider the relationships be-
tween two variables, they are akin to examining bivariate correlations in
primary data. As used here, model fitting more closely resembles stepwise
multiple regression. Each approach has virtues and limitations that are
essentially the same in primary and meta-analytic data analysis.
Hypothesis tests. After the effect sizes were calculated and weighted by
their sample sizes, the average r was computed for our four dependent
variables: attitude, competence, character, and dynamism. To test the hy-
potheses a subset of effect sizes was formed for the values of each inde-
pendent variable. That is, effect sizes were grouped corresponding to high
target familiarity versus low target familiarity, high novelty versus low
novelty, and so on. Next, 95% confidence intervals were computed. If the
intervals overlapped, we concluded that the difference between the mean
effect sizes was not statistically significant.
Model fitting. The results of the meta-analysis could be used to repre-
sent many different models of effects based on different moderator vari-
ables. It was necessary therefore to choose an appropriate model of the
influence of moderator variables a priori, and then incorporate into it any
relevant variables that emerged as predictive of metaphor effect after the
meta-analysis.
To do this, the distributions of r were tested for homogeneity of vari-
ance. This test assesses whether the variation in the distribution of the
effect sizes may be plausibly attributed to sampling error or, alternatively,
due to the presence of one or more moderator variables. First, a χ2 test
was computed. A nonsignificant χ2 statistic indicates that the variance in
the distribution may be attributed to sampling error (Hunter, Schmidt, &
Jackson, 1982), whereas a significant result implies the presence of a mod-
erator variable. In addition, we considered the ratio of expected variance

404 HUMAN COMMUNICATION RESEARCH / July 2002
to observed variance (EV/OV). Hunter et al. (1982) suggest that, as a rule
of thumb, the variation may be attributed to sampling error if the ratio is
75% or greater.
RESULTS
Tests of Main Effect Hypotheses
H1: The effect of metaphor on attitude. The overall sample size weighted
mean effect size (r) for the metaphor-literal comparison was .066 (k = 38,
N = 3945). The confidence interval did not include zero. After correction
for measurement error the effect size (r´) rose to .073. Thus, the hypoth-
esis that metaphoric language is more persuasive than literal, was sup-
ported. Table 4 provides summary information for this and subsequent
hypothesis tests.
H2: The effect of metaphor on credibility. H2 proposed that use of meta-
phors would encourage favorable postmessage communicator judgments
for all three components of credibility. For H2a (competence), the overall
r was -.006 (r´ = -.088) and the confidence interval included zero. For H2b
(character), r was -.021 (r´ = -.024) and the confidence interval included
zero. For H2c (dynamism), r was .059 (r´ was .064) and the confidence
interval did not include zero. Thus, while there was no evidence that
metaphor enhanced perceptions of competence or character (H2a and H2b
respectively), the data did suggest a small but reliable relationship be-
tween metaphor and dynamism (H2c).9
Tests of Moderator Hypotheses for Attitude
H3: Initial communicator credibility. The mean effect size for low cred-
ibility sources was larger (r = .116; r´ = .123) than that for high credibility
sources (r = .023; r´ = .027). The confidence intervals did not overlap (but
touched, see Table 4). Thus, contrary to H3, the results indicated that use
of metaphor benefited low initial credibility communicators more than
high initial credibility ones.
H4: Extendedness of metaphors. The hypothesis that extended metaphors
are more persuasive was borne out to some extent. The average effect
size of the extended metaphors (r = .088; r´ = .095) was larger than that for
nonextended metaphors (r = .048; r´ = .054). However, the confidence in-
tervals overlapped thereby providing only directional support for the
hypothesis.
H5: Number of metaphors. The pattern of average effect sizes suggested
that the use of 1 metaphor only (r = .076; r´ = .086) in a persuasive mes-
sage was a better strategy than the use of 2–8 (r = .063; r´ = .066) or 9 or

Sopory, Dillard / METAPHOR AND PERSUASION 405
TABLE 4
Summary of Results and Confidence Intervals for Mean Effect Sizes
95% confidence interval
|----.----|----.----|----.----|----.----|
Variables r k N -.2 -.1 0 .1 .2
Metaphor-literal main effect
H1: Attitude .07* 38 3945 .04 |-----| .10
H2: Terminal credibility
Competence -.01* 20 1938 -.05 |--------| .04
Character -.02 20 1936 -.07 |--------| .02
Dynamism .06 11 1252 .00 |---------| .11
Moderator analysis for metaphor effect on attitude
H3: Initial credibility
High .02* 20 2088 -.02 |--------| .07
Low .12* 14 1487 .07 |---------| .17
H4: Extendedness
Extended .09 14 1828 .04 |--------| .13
Nonextended .05* 24 2117 .01 |-------| .09
H5: Number
1 .08* 16 1753 .03 |--------| .12
2-8 .06* 12 900 -.00 |------------| .13
9 > .02 7 574 -.07 |-----------------| .10
H6: Position
Introduction .12* 11 1139 .07 |-----------| .18
Body .07 6 732 -.01 |-------------| .14
Conclusion -.01* 10 932 -.07 |------------| .06
H7: Familiarity
High .07* 21 1795 .02 |--------| .11
Low .06* 16 1951 .02 |-------| .10
H8: Novelty
High .12* 16 2107 .08 |--------|.16
Low .01* 20 1708 -.04|--------| .05
H9: Modality
Written .06* 30 3035 .02 |-------| .10
Audio .09 8 910 .03 |-------------|.16
NOTE: N = number of subjects, k = number of data-points. Effect sizes and confidence inter-
vals are not corrected for measurement error. Nonhomogenous subsets are indicated by an *
following the effect size. Homogeneity determined by the Chi-square test and the expected
variance/observed variance > 75% rule.

406 HUMAN COMMUNICATION RESEARCH / July 2002
more metaphors (r = .017; r´ = .018). Thus, H5b, that more metaphors
would lead to more persuasion, was not supported. Instead, the results
suggested the opposite, that fewer metaphors are more persuasive (H5a).
However, analysis of the confidence intervals for the means of the three
groups suggested that the differences among them were not significant,
thus qualifying this conclusion.
To further look at the effect of number of metaphors, separate analyses
for extended and nonextended metaphors were conducted. The results
followed the above pattern: extended metaphors with 1 only r = .176 (k =
1, N = 177), 2–8 r = .101 (k = 6, N = 519), and with 9 or more r = .035 (k = 4,
N = 414); nonextended metaphors with 1 only r = .064 (k = 15, N = 1576),
2–8 r =.012 (k = 6, N = 381), and with 9 or more r = -.028 (k = 3, N = 160). At
-.043, the correlation between the number of metaphors and the corre-
sponding effect size also implied support for the conclusion that increased
number of metaphors leads to decreased persuasiveness.
H6: Position of metaphors. Metaphors can be located in the introduction,
body, or conclusion of a message. The average effect size was much higher
for introduction only (r = .123; r´ = .138) than that for metaphors in the
conclusion only (r = -.063; r´ = -.064) and body only (r = .067; r´ = .071).
The confidence interval for the introduction group overlapped with the
body group, but not with the conclusion group.
The data were also analyzed with regard to the position at which a
metaphor was first used in a message. Results for first appearance in in-
troduction were r = .103, r´ = .113, k = 20, N = 2149, and for first appear-
ance in body or conclusion were r = .022, r´ = .024, k = 18, N = 1796. The
95% confidence interval for the first appearance in introduction group
(.062 to .145) just barely overlapped with the confidence interval of the
first appearance in body or conclusion group (-.024 to .068). The position
effect was more pronounced for nonextended metaphors, with first ap-
pearance in introduction r = .119 (k = 13, N = 1162) and first appearance in
body or conclusion r = -.039 (k = 11, N = 955), but slightly reversed for
extended metaphors, with first appearance in introduction r = .084 (k = 7,
N = 987) and first appearance in body or conclusion r = .092 (k = 7, N =
841). Thus, the above sets of results generally indicated that use of meta-
phor in the introduction was more persuasive than in other parts of a
message. H6a was supported and H6b was not.
H7: Familiarity of target. As H7a proposed, people who were more
knowledgeable about the target were more persuaded by meta-
phors as compared to literal language. Subjects in the high knowl-
edge group showed greater persuasion (r = .065; r´ = .071) than their
counterparts in the low knowledge groups (r = .059; r´ = .065). But,
because the confidence intervals overlapped there was only directional
support for H7a.

Sopory, Dillard / METAPHOR AND PERSUASION 407
H8: Novelty. The mean effect size for high novel metaphors (r = .118; r´
= .129) was larger than that for less novel metaphors (r = .006; r´ = .008).
The confidence intervals around the mean effect sizes indicated that
the difference was significant. Thus, H8a, that novel metaphors are
more persuasive than not novel ones, received support, while its op-
posite H8b did not.
H9: Modality of presentation. The mean effect size in the written group (r
= .059; r´ = .065) was smaller than that for the audio group (r = .091; r´ =
.101). However, the confidence intervals overlapped. Thus, H9a, which
proposed greater persuasion in the written modality group, was not sup-
ported; instead, the reverse (i.e., audio > written) received directional
support, a result consistent with H9b.
Model Fitting
The results of the overall metaphor-literal comparison for attitude sug-
gested the presence of moderator variables in the data [χ2 (37) = 88.69, p <
.05, EV/OV = 43%].10 Our efforts to model the data were guided by the
desire to optimize parsimony, empirical clarity, and theoretical discrimi-
nation among the theories. Thus, we partitioned the data set on a single
variable, tested both resulting groups for homogeneity of variance, then
repeated this process until homogeneity was attained or there were too
few data points in a group to permit such a test. After many attempts to
structure the data set in different ways, ultimately, we could fit only the
model that appears in Figure 1. It shows that when a metaphor is
nonextended, single, placed early in a suasory message, novel, and with
a familiar target, its impact can be substantial (r = .42). The only oddity in
the analysis was for the nonextended group. Here the two homogeneity
indices were in conflict. For such cases the χ2 statistic was taken as the
more powerful indicator (Hunter & Schmidt, 1990) and the group was
considered homogenous.
In sum, the results of the model fitting reinforced many of the conclu-
sions drawn from the hypothesis tests. They complemented and extended
those findings by allowing a better understanding of the relationships
between particular variables after controlling for others and thereby re-
vealing more clearly the effect of certain moderators on the metaphor-
attitude relationship. The most dramatic example appears at the bottom
of the tree structure (Figure 1) between extended and nonextended meta-
phors. There was directional support for extended over nonextended
metaphors in the bivariate analysis, but a reversed effect emerged in model
fitting after controlling for number and position. Other important differ-
ences can also be seen in branches above that point, such as a much stron-
ger effect emerging for number and position.

408 HUMAN COMMUNICATION RESEARCH / July 2002
DISCUSSION
Two questions motivated this project. First, do metaphors possess a
suasory advantage over literal language? The data suggest an affirmative
response. Second, of the six theories considered, which best explains the
persuasive effects of metaphor? Our response here is necessarily more
complicated, but the brief version is that the superior organization ac-
count is best supported by the data. Our case for the pair of conclusions is
built on two distinct types of knowledge claims. Relationship-present claims
assert the existence of a reliable association between two variables; rela-
tionship-absent claims are advanced when a researcher contends that one
variable does not reliably covary with another (Dillard, 1998). Both types
of claims are important to theory building and application. In the follow-
ing sections, we consider the evidence for each type of claim as it relates
to the theories and hypotheses under scrutiny.
Metaphor-Literal Main Effects
The impact of metaphor on attitude. Relative to their literal counterparts,
metaphorical messages produced greater attitude change. First, the con-
fidence interval around the positive correlation coefficient (r = .07) did
Figure 1. Model Fitting: Effect Sizes for Attitude Divided by Familiarity, Novelty,
Position, Number, and Extendedness Moderator Variables
NOTE: Clusters in thicker lines represent homogenous groupings. k = number of data-points,
N = number of participants.
* p < .05.
First Appearance Body/Conclusion
r = -.049 k = 4 N = 385
χ
2
(3) = 4.63 EV/OV = 87%
Number of Metaphors > 1
r = -.108 k = 1 N = 62
Number of Metaphors = 1
r = .306 k = 4 N = 374
χ
2
(3) = 10.33* EV/OV = 39%
Nonextended
r = .423 k = 3 N = 197
χ
2
(2) = 4.19 EV/OV = 73%
Extended
r = .176 k = 1 N = 177
Low Familiarity
r = .059 k = 16 N = 1951
χ
2
(15) = 27.99* EV/OV = 58%
Entire Data Set
r = .066 k = 38 N = 3945
χ
2
(37) = 88.69* EV/OV = 43%
High Familiarity
r = .065 k = 21 N =1795
χ
2
(20) = 58.46* EV/OV = 36%
High Novelty
r = .124 k = 7 N = 1286
χ
2
(6) = 5.34 EV/OV = 100%
Low Novelty
r = -.089 k = 7 N = 535
χ
2
(6) = 5.42 EV/OV = 100%
First Appearance Introduction
r = .247 k = 5 N = 436
χ
2
(4) = 19.96* EV/OV = 25%
High Novelty
r = .108 k = 9 N = 821
χ
2
(8) = 41.13* EV/OV = 22%
Low Novelty
r = .033 k = 12 N = 1041
χ
2
(11) = 15.40 EV/OV = 79%

Sopory, Dillard / METAPHOR AND PERSUASION 409
not include zero.11 In addition, we uncovered only 2 instances out of 14 in
which the use of metaphor may be detrimental to the goal of generating
agreement with the message advocacy. The effect for metaphor was mar-
ginally negative for the conclusion position and null for low meta-
phor novelty.
Additional confidence in this relationship-present claim can be derived
from application of Rosenthal’s (1991) file drawer statistic (FDS), which
estimates the number of “hidden” studies (i.e., studies not included in
the meta-analysis showing no effects) that would be required to overturn
the observed effect size. The FDS analysis showed that to overturn the
proposition that a metaphor-using message is more persuasive than a
literal-only one, 164 new studies would be required, all of which would
have to have a no-relationship finding.
Finally, the binomial effect size display (BESD; Rosenthal, 1991;
Rosenthal & Rubin, 1982) is a statistic that estimates the advantage that
would accrue if the claim is used to construct a persuasive message. In
other words, it presents the effect of a given variable as a change in suc-
cess rate for a dichotomous dependent variable. The BESD analysis
showed that, for the overall data set, metaphor was roughly 6% more
persuasive than literal language. If such a small effect seems to be of little
pragmatic consequence, it is revealing to consider that the popular vote
in the 2000 U.S. presidential election was reportedly won by a difference
of only .05% (48.4% vs. 47.9%), the 1996 election 8.5% (49.2% vs. 40.7%),
the 1992 election 5.6% (43% vs. 37.4%) and the 1988 election 7.8% (45.6%
vs. 53.4%; U.S. Census Bureau, 2000; U.S. National Archives and Records
Administration Federal Register, 2001). The metaphor-literal difference
is comparable to these voting choice differences.
The relatively high FDS and BESD statistics suggest a high degree of
confidence about metaphor’s positive effect on attitude. However, we
warn against the conclusion that any metaphor can be used under any
condition to create potent suasory impact. Messages contained in the meta-
analysis were developed by researchers with advanced training and
typically subjected to pretesting prior to data gathering. Moreover, in spite
of this fine tuning of messages, 15 of the 38 effect sizes were negative,
indicating that metaphor use diminished persuasiveness in almost
40% of the cases.
The impact of metaphor on terminal credibility. The hypothesis tests for
communicator credibility revealed that use of metaphor confers a suasory
advantage by enhancing the dynamism component, but not the compe-
tence and character components of terminal credibility. The effect size for
metaphor use on the dynamism aspect of terminal credibility was .06.
The confidence interval did not contain zero (when computed beyond
the second decimal). The analysis of the moderator variable of initial (low

410 HUMAN COMMUNICATION RESEARCH / July 2002
and high) communicator credibility showed that use of metaphors in-
creased judgments of dynamism for both low and high credible commu-
nicators. The file-drawer statistic for dynamism was 12 studies and the
BESD estimate for increase in success rate was 6%.
In contrast, effect sizes for the competence and character aspects of
credibility were, for all practical purposes, zero. It is tempting to advance
a relationship-absent claim (i.e., that metaphor has no effect on compe-
tence or character). However, a power analysis using a two-tailed alpha
of .05 and assuming a true effect size of .05 (see Cohen, 1988), indicated
that our data possessed power of only .60. While we can accurately claim
a lack of support for the hypothesized association between metaphor and
credibility, the low power of that test does not permit confidence in the
claim that metaphor does not influence judgments of credibility.
Theories of the Persuasive Effects of Metaphor
The various theoretical explanations of metaphor’s persuasive effects
can be differentiated with the help of the meta-analysis results. Although
it was not possible to test all aspects of the theories given the limitations
of the data set, the results do permit us to draw some substantive conclu-
sions about their explanatory viability. Table 1 displays the theoretical
perspective connected with each hypothesis tested in the meta-analysis
and the results of the tests (see the first column). From this framework,
we see that the superior organization theory received the most support.12
Superior organization. Table 1 shows that almost all hypotheses guided
by the superior organization view (Read et al., 1990) were supported. The
results showed that a single metaphor was more persuasive than greater
numbers. As the superior organization view implies, it is only a single
metaphor (and not more) that should provide the optimum opportunity
for enhanced organization of the message information. Metaphors were
also most persuasive when extended and when placed in the introduc-
tion position of a message. This suggests that persuasion occurred due to
the organizing potential of metaphor-as-theme, which facilitated selec-
tion and integration of information from the message and prior knowl-
edge. Similarly, the persuasive superiority of high familiar target meta-
phors over low familiar ones suggests that higher prior knowledge al-
lowed recipients to better organize the target-base linkages.
Only the result that metaphor was more persuasive in the audio than
the written modality was contrary to prediction. In retrospect this appar-
ently disconfirming evidence may follow more from our derivation of
the hypothesis than from the theory itself. It is possible that the opportu-
nity for multiple passes in the written format permits the formation of
semantic linkages that are irrelevant to the message topic and that inter-
fere with the coherent organization of the message. Because the audio

Sopory, Dillard / METAPHOR AND PERSUASION 411
modality demands immediate processing, these extraneous linkages may
be inhibited, thereby allowing the organizing power of a metaphor to be
fully realized.
Pleasure or relief. The data did not speak directly to the reinforcement
principle of the pleasure or relief view (e.g., Bowers & Osborn, 1966).
However, of all the hypotheses guided by this view, only the one pertain-
ing to novelty of metaphor was borne out. The result that high novel meta-
phors are more persuasive than low novel ones can be easily explained
by other views and so does not confer much support to the literal-pri-
macy view. In addition, the assumptions of literal-primacy theory that
underlie this view were strongly tested in the hypothesis for modality of
presentation. Literal-primacy view (e.g., Beardsley, 1962, 1976) suggests
that the literal meaning of an expression is obligatorily understood be-
fore the metaphorical meaning. As such, the comprehension of a meta-
phor should take longer than the comprehension of an equivalent literal
language. This view supports an advantage for written modality by en-
suring that cognizers have enough time to comprehend a message,
whereas the likelihood of pleasure or relief in the audio modality should
be depressed. The results, however, showed that audio modality was more
persuasive, contradicting the prediction from literal-primacy view. There-
fore, the support for the pleasure or relief view of metaphor’s persuasive
advantage is weak.
Communicator credibility. Two hypotheses were supported by the com-
municator credibility explanation (e.g., Bowers & Osborn, 1966), novelty
of metaphor and modality of presentation. However, the finding that high
novel metaphors are more persuasive than low novel ones can be de-
rived from other theories and so does not offer strong support for this
view. On the other hand, the modality result, that audio format is more
persuasive than the written format, is somewhat supportive. Still, a strong
test of this view was available through the metaphor effect on terminal
communicator credibility which showed that individuals do not judge
metaphor-using communicators more favorably than sources who use
literal language. Thus, the notion that use of metaphor prompts a source
heuristic based on competence or character, thereby leading to greater
persuasion, is probably not the best explanation for the persuasive effects
of metaphor.
Reduced counterargument. The central assumption of the reduced
counterargument view (Guthrie, 1972), that metaphor demands more
cognitive resources or effort than literal language to process, was tested
through a self-report measure only in Whaley (1991). The results showed
no statistical difference between metaphor and literal conditions. Simi-
larly, data on counterarguments, measured as number of thoughts gener-
ated opposing the message arguments, were available from Hitchon (1991),

412 HUMAN COMMUNICATION RESEARCH / July 2002
Ottati et al. (1999), and Whaley (1991). Neither study showed a signifi-
cant difference between metaphor and literal conditions.
Two predictions guided by the counterarguing view, related to meta-
phor novelty and modality of presentation, were borne out. The novelty
result is explained by many other views and so is not strong evidence of
support. The modality result offers some support for this view, though
again, the result can also be interpreted to support the alternative source
credibility view. The strongest test of reduced counterargumentation was
offered by the number of metaphors hypothesis: More metaphors should
demand more resources, leading to reduced counterargumentation, and
hence increased persuasion. The opposite was found. Thus, the reduced
counterargument view does not garner much support from the
meta-analysis.
Resource matching. The resource matching explanation (Jaffe, 1988) also
makes the assumption that metaphor processing requires more cognitive
resources than literal language, and proposes that metaphors should be
persuasive only under resource enhancing conditions, such as message
repetition. In addition to the result of null differences in cognitive effort
from Whaley (1991), a look at the meta-analysis studies also contradicts
this proposition. All studies in this analysis (barring the one using the
resource matching perspective) presented messages only once, and the
overall results do show that metaphors led to more attitude change than
literal language. Similarly, multiple processing of a message in the writ-
ten modality may be seen as equivalent to message repetition, thus facili-
tating resource matching. However, it was the audio modality, which al-
lows only a single pass through a message, that was more persuasive.
Along the same lines, novel metaphors, which would demand more re-
sources than not novel ones, were impressively persuasive in our dataset.
The three hypotheses flowing from the resource matching view that were
supported (metaphor number, extendedness, and target familiarity) have
an alternative and more plausible explanation provided by the superior
organization view. Therefore, this view is not an ideal candidate for a
theoretical explanation of metaphor’s persuasive effects.
Stimulated elaboration. The key variable in the stimulated elaboration
account is the number of thoughts generated in response to a metaphori-
cal language message as compared to a literal one. Studies which investi-
gated metaphor’s persuasive effects and measured this type of elabora-
tion (Hitchon, 1991; Mitchell et al., 1994; Ottati et al., 1999; Whaley, 1991)
did not find any difference between metaphor and literal language con-
ditions. Similarly, the only hypothesis guided by this view that was sup-
ported was novelty of metaphor, a result that is easily explained by the
other views. Thus, elaboration, currently conceived as number of linguis-
tic expressions, may not be the right variable for explaining metaphor’s
greater persuasive capacity.

Sopory, Dillard / METAPHOR AND PERSUASION 413
FUTURE RESEARCH AND CONCLUSIONS
Theorists since Aristotle have proposed that metaphor could be fruit-
fully used for persuasion. The meta-analytic summary of existing empiri-
cal studies affirms this supposition regarding metaphor’s suasory effec-
tiveness over literal counterparts. The identification of seven moderator
variables further distinguishes between conditions that enhance or weaken
this effectiveness. The results afford an initial set of directions for use (or
avoidance) of metaphors in a variety of applied contexts. Importantly,
our efforts to model the optimal conditions for metaphor effectiveness
produced a clear set of recommendations for practitioners. Specifically,
the data suggest that the persuasive impact of metaphor is maximized
when the audience is familiar with the metaphor target, the metaphor is
novel, is used at the start of a message, is single, and nonextended. Of
course, readers should bear in mind the relatively modest number of per-
sons and messages on which most of these summary claims rest. Though
the results reported here represent a synthesis of available data, they
should be viewed as a platform from which to launch further inquiry
rather than as the final word on the metaphor-persuasion relationship.
Future research should investigate other variables that have the potential
to moderate the effectiveness of metaphorical messages. Two such vari-
ables, initial attitude towards the metaphor target and involvement with
the message topic (Ottati et al., 1999; see also Johnson & Eagly, 1990), may
play an important role in the derivation of metaphorical meaning as well
as the extent to which this meaning impacts evaluations.
Of the six explanations of metaphor effect, the moderator analyses pro-
vide most support for the superior organization explanation of metaphor’s
suasory impact. However, our results do not allow us to fully disentangle
the processes suggested by the other metaphor and persuasion theories,
so our conclusion of support for the superior organization view should not
be interpreted to mean that the other theories are incorrect. The develop-
ment of a more comprehensive theory of the suasory functions of meta-
phor represents an important challenge for future inquiry. Future research
should investigate persuasion processes engendered by metaphorical
messages versus messages with intense and vivid language. Language
intensity (Bowers, 1964; Bowers & Osborn, 1966; Hamilton, Hunter, &
Burgoon, 1990) and vividness (Frey & Eagly, 1993) views of message attributes
have attempted to subsume metaphorical language, and research can shed
light on any processing continuities and commonalities. Along the same
lines, investigations should also try to uncover any parallels among per-
suasive effects of linguistic metaphor and pictorial or visual metaphors
(McQuarrie & Mick, 1999), other types of figurative language such as irony
(McGuire, 2000), and other message features that seek to persuade through
implication such as rhetorical questions (Burnkrant & Howard, 1984).

414 HUMAN COMMUNICATION RESEARCH / July 2002
Finally, the role of initial communicator credibility in judgments of ter-
minal credibility should be studied more systematically, especially keep-
ing in mind that use of metaphor only enhanced positive judgments of
the dynamism component of credibility. It may be that the perception of
dynamism induced by metaphorical language leads to improved atten-
tion to a message, thereby increasing comprehension and persuasion (see
McGuire, 1969). To further illuminate this process, studies should look at
research on the use of analogy as counterargument, often called rebuttal
analogy (Whaley, 1998), which has shown that analogy use compromises
evaluations of communicators as well as quality of message arguments.
Metaphor has a long history as a persuasion device. Continued research
will provide greater specification of the process as well as contribute to
its more effective use to influence evaluations.
NOTES
1. Metaphor is considered generally distinct from idioms, metonymy and synecdoche.
Idioms are expressions that have become fixed or frozen, for example, “Keep a stiff upper
lip.” Metonymy are expressions that use an entity in place of another entity associated with
it, for example, “The White House issued a statement yesterday” instead of “The President
issued a statement yesterday.” Synecdoche are expressions that use a part to stand for the
whole or the whole to stand for a part, for example, “Look at his new wheels” instead of
“Look at his new car.”
2. Other theories of metaphor comprehension include interaction (Richards, 1936), class
inclusion (Glucksberg & Keysar, 1990), parallel constraint satisfaction (Holyoak & Thagard,
1989, 1995), conceptual metaphor (Lakoff, 1987, 1993; Lakoff & Johnson, 1980), and domains
interaction (Tourangeau & Sternberg, 1981, 1982).
3. There are other dimensions of credibility also, such as attractiveness and sociability.
However, the studies in our database did not assess other credibility components in a man-
ner consistent enough to have a sufficient pool of studies to conduct an analysis.
4. The list of excluded studies can be obtained from the first author.
5. Our thanks to Jacqueline Hitchon, Jeffery Mio, and Victor Ottati for supplying rel-
evant information about their studies.
6. The distribution for published vs. unpublished studies was as follows. Attitude: num-
ber of studies = 16 vs. 13 , N = 2344 vs. 1601, mean effect size = .07 vs. .06; Competence:
number of studies = 5 vs. 7 , N = 1449 vs. 489, mean effect size = .02 vs. -.11; Character:
number of studies = 5 vs. 7 , N = 1468 vs. 468, mean effect size = -.03 vs. -.03; Dynamism:
number of studies = 4 vs. 3 , N = 913 vs. 339, mean effect size = .06 vs. .06.
7. The listing of characteristics of metaphors and message sources for each study can be
obtained from the first author.
8. A sample of students could have done the coding instead of the authors. However, the
authors, with their better understanding of history and context, were more suited to situate
themselves in the shoes of the study participants than the typical undergraduate student.
9. This pattern remained the same when data were partitioned using the initial commu-
nicator credibility (low or high) as a moderator: competence (low = -.005, high = -.015),
character (low = .013, high = -.045), and dynamism (low = .033, high = .077). The confidence
intervals in all cases overlapped.
10. A meaningful model could not be produced for the three credibility facets due to the
lack of sufficient data-points.

Sopory, Dillard / METAPHOR AND PERSUASION 415
11. The small effect size found here is not unlike the magnitude of effects obtained meta-
analytically for other message variables in persuasion research. For example, a two-sided
message containing refutation of counterarguments is superior to a one-sided message by
roughly the same effect size of .07 (Allen, 1998). Moreover, the effect of metaphor becomes
more pronounced when particular moderator variables are taken into account.
12. This line of interpretation of the results is supported by evidence from nonpersuasion
research. See Sopory and Dillard (in press) for details.
REFERENCES
References marked with an asterisk indicate studies included in the meta-analysis.
Allen, M. (1998). One- and two-sided messages. In M. Allen & R. W. Preiss (Eds.), Persua-
sion: Advances through meta-analysis (pp. 87–98). Cresskill, NJ: Hampton Press.
Antrim, Z. D. (1979). Do creative metaphors change attitudes toward contraceptive information?
Unpublished master’s thesis, Virginia Commonwealth University, Richmond.
Aristotle. (1952a). Poetics. (I. Bywater, Trans.). In W. D. Ross (Ed.), The works of Aristotle:
Rhetorica, de rhetorica ad Alexandrum, poetica. Oxford: Clarendon Press.
Aristotle. (1952b). Rhetoric. (W. R. Roberts, Trans.). In W. D. Ross (Ed.), The works of Aristotle:
Rhetorica, de rhetorica ad Alexandrum, poetica. Oxford: Clarendon Press.
Beardsley, M. (1962). The metaphorical twist. Philosophy and Phenomenological Research, 22, 293–307.
Beardsley, M. (1976). Metaphor and falsity. Journal of Aesthetics and Art Criticism, 35, 218–222.
Berlo, D. K., Lemert, J. B., & Mertz, R. J. (1969). Dimensions for evaluating the acceptability
of message sources. Public Opinion Quarterly, 33, 563–576.
Black, M. (1962). Models and metaphors. Ithaca, NY: Cornell University Press.
Black, M. (1979). More about metaphor. In A. Ortony (Ed.), Metaphor and thought (pp. 19–43).
Cambridge, UK: Cambridge University Press.
Boozer, R. W., Wyld, D. C., & Grant, J. (1991). Using metaphor to create more effective sales
messages. Journal of Consumer Marketing, 8, 59–67.
Bosman, J. (1987). Persuasive effects of political metaphors. Metaphor and Sym-
bolic Activity, 2, 97–113.
Bosman, J., & Hagendoorn, L. (1991). Effects of literal and metaphorical persuasive mes-
sages. Metaphor and Symbolic Activity, 6, 271–292.
Bowers, J. W. (1964). Some correlates of language intensity. Quarterly Journal of Speech, 50, 415–420.
*Bowers, J. W., & Osborn, M. M. (1966). Attitudinal effects of selected types of concluding
metaphors in persuasive speeches. Speech Monographs, 33, 147–155.
Bowers, J. W., & Phillips, W. A. (1967). A note on the generality of source-credibility scales.
Speech Monographs, 34, 185–186.
*Brennan, I. (1992). Moderators of the complex allegoric advertising execution-brand atti-
tude relationship (Doctoral dissertation, The University of Texas at Arlington, 1992).
Dissertation Abstracts International, 54, 598.
Burnkrant, R. E., & Howard, D. J. (1984). Effects of introductory rhetorical questions versus
statements on information processing. Journal of Personality and Social Psychology, 47, 1218–1230.
Chaiken, S., Liberman, A., & Eagly, A. H. (1989). Heuristic and systematic processing within
and beyond the persuasion context. In J. S. Uleman & J. A. Bargh (Eds.), Unintended
thought (pp. 212–252). New York: Guilford Press.
Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Hillsdale, NJ: Erlbaum.
Crocker, J. C. (1977). The social functions of rhetorical forms. In J. D. Sapir & J. C. Crocker
(Eds.), The social use of metaphor: Essays on the anthropology of rhetoric (pp. 33–66). Philadel-
phia: University of Pennsylvania Press.
Dillard, J. P. (1998). Evaluating and using meta-analytic knowledge claims. In M. Allen & R.
W. Preiss. (Eds.), Persuasion: Advances through meta-analysis (pp. 257–270). Cresskill, NJ:
Hampton Press.

416 HUMAN COMMUNICATION RESEARCH / July 2002
Dingena, M. (1994). The creation of meaning in advertising: Interaction of figurative advertising
and individual differences in processing styles (Tinbergen Institute Research Series No. 62).
Amsterdam: Thesis Publishers.
Edwards, R., & Clevenger, T., Jr. (1990). The effects of schematic and affective processes on
metaphorical invention. Journal of Psycholinguistic Research, 19, 91–102.
*Fitzgerald, K. A. (1998). Beyond apples and oranges: Using analogies to catalyze new product
adoption (Doctoral dissertation, University of California at Los Angeles, 1998). Disserta-
tion Abstracts International, 59, 3550.
Frey, K. P., & Eagly, A. H. (1993). Vividness can undermine the persuasiveness of messages.
Journal of Personality and Social Psychology, 65, 32–44.
Gentner, D. (1982). Are scientific analogies metaphors? In D. Miall (Ed.), Metaphor: Problems
and perspectives (pp. 106–132). Brighton, UK: Harvester Press.
Gentner, D. (1983). Structure-mapping: A theoretical framework for analogy. Cognitive
Science, 7, 155–170.
Gentner, D. (1989). The mechanisms of analogical learning. In S. Vosniadou & A. Ortony
(Eds.), Similarity and analogical reasoning (pp. 199–241). Cambridge, UK: Cambridge
University Press.
Gentner, D., & Clements, C. (1988). Evidence for relational selectivity in the interpretation
of analogy and metaphor. In G. Bower (Ed.), The psychology of learning and motivation
(Vol. 22, pp. 307–358). Orlando, FL: Academic Press.
Gibbs, R. W., Jr. (1994). The poetics of mind: Figurative thought, language, and understanding.
Cambridge, UK: Cambridge University Press.
Glucksberg, S., & Keysar, B. (1990). Understanding metaphorical comparisons: Beyond simi-
larity. Psychological Review, 97, 3–18.
Graesser, A. C., Mio, J., & Millis, K. K. (1989). Metaphors in persuasive communication. In
D. Meutsch & R. Viehoff (Eds.), Comprehension of literary discourse: Results and problems of
interdisciplinary approaches (pp. 131–154). Berlin, Germany: Walter de Gruyter.
*Guthrie, M. (1972). Effects of credibility, metaphor, and intensity on comprehension, credibility,
and attitude change. Unpublished master’s thesis, Illinois State University, Normal.
Hamilton, M. A., Hunter, J. E., & Burgoon, M. (1990). An empirical test of an axiomatic
model of the relationship between language intensity and persuasion. Journal of Lan-
guage and Social Psychology, 9, 235–255.
*Hitchon, J. C. (1991). To be or what to be: Metaphorical predication in advertising (Doc-
toral dissertation, University of Wisconsin-Madison, 1991). Dissertation Abstracts Inter-
national, 52, 1438.
Holyoak, K. J., & Thagard, P. (1989). Analogical mapping by constraint satisfaction. Cogni-
tive Science, 13, 295–355.
Holyoak, K. J., & Thagard, P. (1995). Mental leaps: Analogy in creative thought. Cam-
bridge, MA: MIT Press.
Hunter, J. E., & Schmidt, F. L. (1990). Methods of meta-analysis: Correcting error and bias in
research findings. Newbury Park, CA: Sage.
Hunter, J. E., Schmidt, F. L., & Jackson, G. B. (1982). Meta-analysis: Cumulating research find-
ings across studies. Beverly Hills, CA: Sage.
*Jaffe, F. (1988). Metaphors and memory: A study in persuasion (Doctoral Dissertation,
Northwestern University, 1988). Dissertation Abstracts International, 49, 2311.
Johnson, B. T., & Eagly, A. H. (1990). Involvement and persuasion: Types, traditions, and the
evidence. Psychological Bulletin, 107, 375–384.
Johnson, J. T., & Taylor, S. E. (1981). The effect of metaphor on political attitudes. Basic and
Applied Social Psychology, 4, 305–316.
Lakoff, G. (1987). Women, fire, and dangerous things. Chicago: Chicago University Press.
Lakoff, G. (1993). The contemporary theory of metaphor. In A. Ortony (Ed.), Metaphor and
thought (2nd ed., pp. 202–251). New York: Cambridge University Press.
Lakoff, G., & Johnson, M. (1980). Metaphors we live by. Chicago: University of Chicago Press.

Sopory, Dillard / METAPHOR AND PERSUASION 417
Lakoff, G., & Turner, M. (1989). More than cool reason: A field guide to poetic metaphor. Chicago:
University of Chicago Press.
MacCormac, E. (1985). A cognitive theory of metaphor. Cambridge, MA: MIT Press.
Mangum, A. (1994). The effect of metaphor in persuasive appeals. Unpublished master ’s thesis,
West Texas A&M University, Canyon.
McCroskey, J. C. (1966). Scales for the measurement of ethos. Speech Monographs, 33, 65–72.
*McCroskey, J. C., & Combs, W. H. (1969). The effects and the use of analogy on attitude
change and source credibility. Journal of Communication, 19, 333–339.
McCroskey, J. C., & Young, T. J. (1981). Ethos and credibility: The construct and its measure-
ment after three decades. Central States Speech Journal, 32, 24–34.
McGuire, W. J. (1969). The nature of attitudes and attitude change. In G. Lindzey & E. Aronson
(Eds.), The handbook of social psychology (2nd ed., Vol. 3, pp. 136–314). Reading, MA:
Addison-Wesley.
McGuire, W. J. (1972). Attitude change: The information-processing paradigm. In C. G.
McClintock (Ed.), Experimental social psychology (pp. 108–141). New York: Holt,
Rinehart & Winston.
McGuire, W. J. (1985). Attitudes and attitude change. In G. Lindzey & E. Aronson (Eds.),
Handbook of social psychology (3rd ed., Vol. 2, pp. 234–326). New York: Random House.
McGuire, W. J. (2000). Standing on the shoulders of ancients: Consumer research, persua-
sion, and figurative language. Journal of Consumer Research, 27, 109–114.
McQuarrie, E. F., & Mick, D. G. (1999). Visual rhetoric in advertising: Text-interpretive, ex-
perimental, and reader-response analyses. Journal of Consumer Research, 26, 37–54.
*Mio, J. S. (1996). Metaphor, politics, and persuasion. In J. S. Mio & A. N. Katz (Eds.), Meta-
phor: Implications and applications (pp. 127–146). Mahwah, NJ: Erlbaum.
*Mio, J. S., Thompson, S. C., & Givens, G. H. (1993). The commons dilemma as metaphor:
Memory, influence, and implications for environmental conservation. Metaphor and Sym-
bolic Activity, 8, 23–42.
Mitchell, N. A., Badzinski, D. M., & Pawlowski, D. R. (1994). The use of metaphors as vivid
stimuli to enhance comprehension and recall of print advertisements. In K. W. King
(Ed.), Proceedings of the 1994 conference of the American Academy of Advertising (pp. 198–
205). Athens, GA: American Academy of Advertising.
*Morgan, S. E. (1997). Metaphorical messages and the literal minded: Accounting for indi-
vidual cognitive differences in the design of persuasive health messages (Doctoral dis-
sertation, The University of Arizona, 1997). Dissertation Abstracts International, 58, 1264.
*Nelson, M. R., & Hitchon, J. C. (1995). Theory of synesthesia applied to persuasion in print
advertising headlines. Journalism & Mass Communication Quarterly, 72, 346–360.
Ortony, A. (1979). Beyond literal similarity. Psychological Review, 86, 161–180.
Ortony, A., Vondruska, R., Foss, M., & Jones, L. (1985). Salience, similes, and the asymmetry
of similarity. Journal of Memory and Language, 24, 569–594.
Osborn, M. M., & Ehninger, D. (1962). The metaphor in public address. Speech Mono-
graphs, 29, 223–234.
*Ottati, V., Rhoads, S., & Graesser, A. C. (1999). The effect of metaphor on processing style in
a persuasion task: A motivational resonance model. Journal of Personality and Social
Psychology, 77, 688–697.
*Pawlowski, D. R., Badzinski, D. M., & Mitchell, N. (1998). Effects of metaphors on children’s
comprehension and perception of print advertisements. Journal of Advertising, 27, 83–98.
Peracchio, L. A., & Luna, D. (1998). The development of an advertising campaign to dis-
courage smoking initiation among children and youth. Journal of Advertising, 27, 49–56.
Petty, R. E., & Cacioppo, J. T. (1986). The elaboration likelihood model of persuasion. In L.
Berkowitz (Ed.), Advances in experimental social psychology (Vol. 19, pp. 123–205). San Di-
ego, CA: Academic Press.
Phillips, B. J. (1997). Thinking into it: Consumer interpretation of complex advertising im-
ages. Journal of Advertising, 26, 77–87.

418 HUMAN COMMUNICATION RESEARCH / July 2002
Read, S. J. (1985). Metaphor in political rhetoric. Unpublished manuscript, University of South-
ern California, Department of Psychology, Los Angeles.
*Read, S. J., Cesa, I. L., Jones, D. K., & Collins, N. L. (1990). When is the federal budget like
a baby? Metaphor in political rhetoric. Metaphor and Symbolic Activity, 5, 125–149.
*Reinsch, N. L., Jr. (1970). An investigation of the effects of the metaphor and simile in persuasive
discourse. Unpublished master’s thesis, Central Michigan University, Mt. Pleasant.
Reinsch, N. L., Jr. (1971). An investigation of the effects of the metaphor and simile in per-
suasive discourse. Speech Monographs, 38, 142–145.
*Reinsch, N. L., Jr. (1973). Attitudinal effects of brief and extended similes and metaphors
(Doctoral dissertation, University of Kansas, 1973). Dissertation Abstracts International, 34, 7914.
Reinsch, N. L., Jr. (1974). Figurative language and source credibility: A preliminary investi-
gation and reconceptualization. Human Communication Research, 1, 75–80.
Richards, I. A. (1936). The philosophy of rhetoric. London: Oxford University Press.
*Rinalducci, E. J. (1994). The effects of figurative language on the perceived credibility of
political candidates (Master’s thesis, Mississippi State University, 1994). Masters Abstracts
International, 33, 1001.
Rosenthal, R. (1991). Meta-analytic procedures for social research (Rev. ed.). Beverly Hills, CA: Sage.
Rosenthal, R., & Rubin, D. B. (1982). A simple, general purpose display of magnitude of
experimental effect. Journal of Educational Psychology, 74, 166–169.
Searle, J. R. (1979). Metaphor. In A. Ortony (Ed.), Metaphor and thought (pp. 92–123). Cam-
bridge, UK: Cambridge University Press.
Shannon, J. R. (1981). Metaphor and sex of speaker as factors in persuasibility and source credibil-
ity. Unpublished master’s thesis, Central Missouri State University, Warrensburg.
*Shelley, D. B. (1974). An experimental study of the effects of metaphor on audience response. Un-
published master’s thesis, San Francisco State University, San Francisco, CA.
*Siltanen, S. A. (1981). The persuasiveness of metaphor: A replication and extension. South-
ern Speech Communication Journal, 47, 67–83.
Sopory, P., & Dillard, J. P. (in press). Figurative comparison and persuasion. In J. P. Dillard &
M. W. Pfau (Eds.), The persuasion handbook: Developments in theory and practice. Thousand
Oaks, CA: Sage.
Stern, B. B. (1988). How does an ad mean? Language in services advertising. Journal of Ad-
vertising, 17, 3–14.
*Suit, J. L., & Paradise, L. V. (1985). Effects of metaphors and cognitive complexity on per-
ceived counselor characteristics. Journal of Counseling Psychology, 32, 23–28.
Tourangeau, R., & Sternberg, R. (1981). Aptness in metaphor. Cognitive Psychology, 13, 27–55.
Tourangeau, R., & Sternberg, R. (1982). Understanding and appreciating metaphors. Cogni-
tion, 11, 203–244.
*Tudman, R. B. (1971). An experimental study of the effects of thematic metaphor on readers’ atti-
tude toward message topic and message source. Unpublished master’s thesis, Central Mis-
souri State College, Warrensburg.
U.S. Census Bureau. (2000). Statistical abstract of the United States: 2000 (120th ed.). Washing-
ton, DC: U.S. Census Bureau.
U.S. National Archives and Records Administration Federal Register (2001). 2000 presiden-
tial election popular vote totals by state. Retrieved August 15, 2001, from http://
www.nara.gov/fedreg/elctcoll/2000popres.html
van Dijk, T. A. (1987). Episodic models in discourse processing. In R. Horowitz & S. J. Samuels
(Eds.), Comprehending oral and written language (pp. 161–196). New York: Academic Press.
Voss, J., Kennet, J., Wiley, J., & Engstler-Schooler, T. (1992). Experts at debates: The use
of metaphor in the U.S. Senate debate on the Gulf crisis. Metaphor and Symbolic
Activity, 7, 197–214.
*Whaley, B. (1991). Toward a comprehensive model of analogy in persuasion: A test of the
persuasive roles of analogy (Doctoral dissertation, Purdue University, 1991). Disserta-
tion Abstracts International, 53, 20.

Sopory, Dillard / METAPHOR AND PERSUASION 419
*Whaley, B. B. (1998). Evaluations of rebuttal analogy users: Ethical and competence consid-
erations. Argumentation, 12, 351–365.
*Whaley, B. B., & Wagner, L. S. (2000). Rebuttal analogy in persuasive messages: Communica-
tor likability and cognitive responses. Journal of Language and Social Psychology, 19, 66–84.
Zaltman, G., & Coulter, R. H. (1995). Seeing the voice of the customer: Metaphor-based
advertising research. Journal of Advertising Research, 35, 35–51.