127
020
THS

 

"WINNIEIII” W

Michigan State
University

 

 

This is to certify that the

dissertation entitled

COGNITIVE PROCESSING OF PERSUASIVE MESSAGE CUES:
A META-ANALYTIC REVIEW OF THE EFFECTS OF
SUPPORTING INFORMATION ON ATTITUDES

presented by 1

JAMES B. STIFF

has been accepted towards fu‘lfillment
ofthe requirements for

Ph.D. Communication .

 

 

degree in

flfidxg i7? QZaZI/I/w/

M ajor professor

Gerald R. Miller
Date 2 May 85

 

MSU u an Affirmative .4( (ion 'Equal Opportunity Insmurmn ' 012771

RETURNING MATERIALS:
Piece in book drop to
remove this checkout from
your record. FINES will
be charged if book is
returned after the date
stamped below.,

     

MSU

LIBRARIES
—_

 
   
 

    

.rru n 5 y?3"
ti“. sf

{3’

COGNITIVE PROCESSING OF PERSUASIVE MESSAGE CUES:
A HETA-ANALYTIC REVIEW OF THE EFFECTS OF
SUPPORTING INFORMATION ON ATTITUDES

By

James Brian Stiff

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Communication

Abstract

COGNITIVE PROCESSING OF PERSUASIVE MESSAGE CUES:
A META-ANALYTIC REVIEW OF THE EFFECTS OF

SUPPORTING INFORMATION ON ATTITUDES

By

James B. Stiff

This thesis reviews recent theoretical developments on
the effects of source and message cues in persuasion. The
Elaboration Likelihood Model (Petty a Cacioppo, 1981) is
analyzed and the limitations of the model are identified. A
review of recent research on information processing 1
identifies an alternate view of the persuasion process.
Kahneman's (1973) Elastic Capacity Model is offered as-a
more comprehensive model. Consistent with the Kahneman
model, predictions concerning the moderating effects of both
involvement and knowledge on the persuasiveness of central

and peripheral cues are offered.

A meta-analytic review of studies investigating the
effect of supporting information on attitudes provided
support for two of these hypotheses. Specifically, there

was a positive linear relationship between the level of

subject involvement and the effect that central message cues
had on attitudes. In addition, there was a curvilinear
relationship between subject involvement and the effect of
source credibility on attitudes such that, as involvement
increased, the effect of source credibility on attitudes

increased up to a point, beyond which further increases in
involvement produced decrements in attitude change. The two
hypotheses concerning the moderating effects of knowledge
were not supported. Subject knowledge did not moderate the
effect of supporting information on attitudes. Although
knowledge moderated the effect of source credibility on
attitudes, the nature of this moderating effect was not the
linear function that was hypothesized. Several explanations

for these results are discussed.

The combined support for the hypotheses concerning the
moderating effects of involvement is evidence for the
validity of the Kahneman model. Implications of all the

results and suggestions for future research are offered.

For my parents, Marlene and John Stiff

ACKNOWLEDGEMENTS

First and foremost, the efforts of my parents, Marlene
and John Stiff, must be acknowledged. For more than 26
years, they have always been there when I needed them. In
addition to providing me with my innate abilities, they have
been more supportive, both emotionally and financially, than
any two parents ought to have been. It was only through
their leadership, guidance, and encouragement that I learned
to attack life one day at a time. Without their concern for
the value of a quality education and their complete support
of this three year project, my degree would never have

become a reality.

I would also like to acknowledge the efforts of three
people who made the greatest contributions to my education.
Gerald Miller, my major professor, always provided a
challenging and lively learning environment. His propensity
to ask tough questions and pick great horses provided the
foundation for a truly liberal education. Jack Hunter and
Frank Boster, who kept the sort of hours best suited for
graduate students and bats, were always available for
lengthy work sessions and interesting conversations. These
three individuals offered the guidance and friendship

necessary for an effective education.

I am greatful to Kathy for her love and support and to

Jeraldo and Jim for their stabilizing influence on me. The

intellectual and social support of these three has been
quite impressive. I would also like to thank the Miller
family for their hospitality and fine meals and the Hunter

family for their interesting conversations and refreshing

pool.

TABLE OF CONTENTS

LIST OF TABLES
LIST OF FIGURES .

SECTION
I. Introduction
The Elaboration Likelihood Model
Approaches to Persuasion .
Determinants of Central/Peripheral Cue Processing .
Predictions of the Elaboration Likelinood Model
Limitations of the Model . .
Models of Human Information Processing
Single Channel Processing Models
Multi- Channel Processing Models
Determinants of Cue Focusing and Capacity Allocation
Modification of the Elaboration Likelinood Model
Predictions
Empirical Support

II. Method
Data
Procedure . .
Instrumentation .

III. Results

Effects of Supporting Information on Attitudes

Moderator Variables, Supporting Information and Attitudes
Involvement, Supporting Information, and Attitudes
Knowledge, Supporting Information, and Attitudes

Effect of Source Credibility on Attitudes
Involvement, Source Credibility, and Attitudes
Knowledge, Source Credibility, and Attitudes

IV. Discussion
Limitations .
Directions for Future Research

REFERENCES

APPENDIX A

Summary of the
of Evidence on

Summary of the

List of Tables

Overall Analysis of the Effect
Attitudes . . . . . . . .

Linear Effect of Supporting Information

on Attitudes Broken Down by Involvement

Summary of the
Broken Down by

Summary of the
on Attitudes

Summary of the
Broken Down by

Summary of the
Broken Down by

Effect of Evidence on Attitudes
Subject Knowledge

Linear Effect of Credibility
Effect of Credibility on Attitudes
Involvement . .

Effect of Credibility on Attitudes
Subject Knowledge . .

I'O

ICD
U0
[(0

Q
\0

80

82

83

84‘

1.

List of Figures

Kahneman's model of elastic capacity allocation

A model of capacity allocation for persuasive
cue processing

PO

:0)
(K:
m)

(D
m

86

Much research has focused on the cognitive responses
individuals experience when they are exposed to persuasive
messages (Greenwald, 1968, Petty a cacioppo, 1981; Petty,
Ostrom a Brock, 1981). One of the most recent and well
publicized of these cognitive response models is The
Elaboration Likelihood Model (Petty a Cacioppo, 1981).
Although Petty and Cacioppo clearly outline the types of
processing strategies available to individuals, highlight
the factors which influence the use of these strategies.
identify the cognitive responses which result from these
strategies, and forecast the persuasive impact of these
cognitive responses, they do little to document and explain
the information processing mechanisms associated with each

of the major components if their model.

The purpose of this manuscript is to evaluate the
Elaboration Likelihood Model in terms of its consistency
with relevant information processing models. This
evaluation will attempt to integrate the existing research
on information processing with research on cognitive
responses to persuasion. This integration will constitute
elaboration of the Petty and Cacioppo (1981) model and
provide a set of clearly testable hypotheses concerning the
relationship between the use of both central (message) cues
and peripheral (situational) cues on the attitudes of

persuasive message recipients.

The Elaboration Likelihood Model

Approaches to Persusaies

Historically, there have been many approaches to the
process of persuasion. Although labeled differently, these
many approaches can be classified in one of two categories;
(a) approaches which focus on the content of persuasive
messages and (b) approaches which focus on characteristics
of persuasive situations (e.g., source credibility, social

comparison with similar and dissimilar others).

Chaiken (1980) identifies the systematic approach to
persuasion as one which involves the cognitive evaluation of
message content. According to this view, when individuals
receive a persuasive message they exert considerable
cognitive effort attempting to comprehend and evaluate the
message's arguments and assess the arguments' validity in
relation to the message's conclusion. Petty and Cacioppo
(1981) label this the Central Route to persuasion. Since
the cognitive evaluation of message content is central to
this view of persuasion, the quality of the information
supporting the message recommendation is a critical

determinant of attitude change.

Both Chaiken (1980) and Petty and Cacioppo (1981) cite

considerable research to document the capacity of

individuals to cognitively process information contained in
persuasive messages. Research on counterarguing (McGuire,
1961; Miller & Burgoon, 1979; Osterhouse & Brock, 1971) and
the generation of persuasive arguments (Vinkour a Burnstein,
19?“) suggests that individuals are capable of cognitively
processing the content of persuasive messages. Moreover,
research on message comprehension (Chaiken a Eagly, 1976;
Eagly, 1979; Eagly a Warren, 1976) suggests that the
cognitive evaluation of persuasive arguments is central to
attitude change. When poor message comprehensibility
prevented individuals in these studies from cognitively
evaluating the content of these persuasive messages,
differences in the amount of attitude change (relative to
the subjects in the good message comprehension conditions)

were observed.

A second general approach to persuasion focuses on
characteristics of the persuasive situation rather than on
the quality of the arguments in the persuasive message.
Chaiken identifies this heuristic approach to persuasion as
one which requires relatively little cognitive effort of
message recipients. Although message recipients are not
passive, they rely on easily accessible information when
deciding to accept or reject message conclusions. Petty &
Cacioppo label this approach the EQEIEDQEQL Beeee to
persuasion. Like Chaiken (1980), Petty & Cacioppo suggest

that individuals often rely on non-content cues to make

judgments about persuasive messages. Most often, these cues
are defined as characteristics of the source and the
situation. Theoretically, however, they can also be defined
to include basic psychological and physiological cues.

Petty and Cacioppo (1981) argue that when individuals
"associate" (assimilate) the attitude object with positive
source or situational cues they form positive impressions of
the attitude object. When message recipients assimilate the
attitude object with negative source or situation cues they
form negative impressions of the attitude object. According
to both Chaiken and Petty and Cacioppo, these non-content
cues allow individuals to decide whether to accept or reject
the attitude object without the need for processing issue

relevant information.

A plethora of research exists to document the effects
of heuristic persuasion. Research on source attractiveness
(Chaiken, 1979; Dion a Stein, 1978; Eagly a Chaiken, 1975;
Eagly, Wood, & Chaiken, 1978; Mills & Aronson, 1965; Snyder
a Rothbart, 1971), source-recipient similarity (Berscheid,
1966; Burnstein, Stotland, & Zander, 1961; Goethals, 1972;
Goethals a Nelson, 1973), and source expertise (Chaiken,
1980; Norman, 1976, Petty, Cacioppo & Goldman, 1981)
demonstrates the persuasive effects of heuristic cue

processing on attitude change.

Detscmiaeets sf Qectrel/Beciehecel Que Eceeessies

Petty and Cacioppo (1981) argue that upon receipt of a
persuasive message individuals decide which set of cues they
will focus on when processing aspects of the persuasive
message. There are at least two general determinants of cue
processing strategy selection. The first of these is the
individual's motivation to accurately choose between
accepting and rejecting message recommendations. The second
is an individual's ability to systematically process the
content of the persuasive message. Each of these processing

strategies will be discussed below.

Petty and Cacioppo (1981) argue that information
processing capacity limits are the determinant of an
individual's decision to process information systematically
or heuristically. They argue that individuals are often
bombarded with more messages than they can possibly
process. As a coping mechanism, individuals choose to
systematically process messages of great import and not to
systematically process messages of lesser import. Hence,
messages high in personal relevance receive more scrutiny
than messages low in personal relevance. Chaiken (1980)
offers a somewhat different decision mechanism. She argues
that systematic processing requires more effort and results
in fewer decision errors than heuristic cue processing.

Because of these differences, when individuals are forced to

make a choice between these two strategies they conduct an
economic analysis. Presumably, individuals weigh the
advantages of increased decision accuracy against the costs
associated with the cognitive effort required for systematic
cue processing. If the advantages outweigh the costs, than
the individual chooses a systematic strategy over a
heuristic strategy. Chaiken argues that accuracy concerns
are paramount when individuals are highly involved with the
message issue. Thus, both approaches posit that under
conditions of high involvement individuals are motivated to
process the content of persuasive messages. When
involvement is low, individuals prefer the less effortful

heuristic cue processing strategy.

Another determinant of systematic/heuristic cue
processing strategy selection is an individual's ability to
process the persuasive message. Generally, two types of
ability have been identified as relevant factors. Petty and
Cacioppo (1981) argue that physiological ability is
necessary before an individual can be expected to process
the content in a message. Petty and Cacioppo cite message
comprehensibility and external distractors as two factors
which affect an individual's physical ability to process the
content of persuasive messages. In addition to these
physical concerns there is a cognitive component to an
individual's ability to process the content of persuasive

messages. Individuals who do not possess issue relevant

knowledge may be precluded from evaluating the content of a
persuasive message even though they are motivated to do so.
Gregg (1967) highlights the nature of this problem. He
suggests that the problem of evidential evaluation is to
make sure that reliable criteria are used and to possess
enough knowledge about the subject matter to avoid a

superficial analysis.

Predictions of the Elaboration Likelihood Model

Petty and Cacioppo (1981) consider these determinants
of systematic and heuristic cue processing to be central to
the predictions which their model offers concerning the
relative effectiveness of central (message) and peripheral
(situational) cues in persuasive situations. Specifically,
Petty and Cacioppo argue that individuals who are motivated
and have the ability to process the content of persuasive
messages will evaluate the validity of message content
before deciding whether to accept or reject message
recommendations. Hence, they predict that when message
recipients are involved in the persuasive situation, they
will be influenced more by characteristics of the message
(central cues) than by characteristics of the persuasive
situation (peripheral cues). Thus, ability and involvement
will be positively related to the effects that message
arguments and the evidence used to support them have on

attitudes of the message recipient following exposure to the

persuasive appeal.

Conversely, Petty and Cacioppo (1981) argue that
subjects who do not have the ability to process messages
systematically or who are insufficiently involved to process
messages systematically will focus on characteristics of the
persuasive situation (peripheral cues) when making judgments
about whether to accept or reject message recommendations.
Since low involvement and low ability message recipients
focus on peripheral cues rather than on central message cues
they are likely to be influenced more by aspects of the
persuasive situation than by the content of persuasive
messages. Thus, Petty and Cacioppo predict that ability and
involvement will be negatively related to the effects that
aspects of the persuasive situation have on the attitudes of
message recipients following exposure to a persuasive

appeal.

The Elaboration Likelihood Model of cognitive responses
to persuasive information is, in essence, a model of human
information processing. The thrust of the model centers
around the strategies individuals select to process
information and the cognitive responses which these
strategies produce. While the model clearly identifies the
determinants and outcomes of persuasive information

processing, Petty and Cacioppo fail to directly address the

processes individuals engage in before responding
cognitively to persuasive information. In fact, the
validity of the Elaboration Likelihood Model is dependent
upon the accuracy of a number of unchecked assumptions about

individuals' abilities to process information.

Initially, Petty and Cacioppo assume that message
recipients are forced with a choice between one of two
information processing strategies. Presumably, individuals
who are low in involvement choose to process the peripheral
cues of a persuasive message because this strategy requires
less effort. High involvement individuals choose to
systematically process the content of the persuasive message
because decision accuracy is paramount and worth the added
processing effort. The limitation of the Petty and Cacioppo
model arises from their assumption that individuals must
choose between one of these two processing strategies. Why
can't individuals who are highly involved with a message
exert more cognitive effort and choose to process both types
of persuasive information before deciding whether to accept
or reject message recommendations? Moreover, why is it that
individuals who are low in involvement and predominately
concerned with conserving cognitive energy choose to process

any information at all?

A second assumption that Petty and Cacioppo make in
their model is that individuals are capable of selecting one

processing strategy over another. Unfortunately, this

10

assumption is never directly tested by Petty and Cacioppo.
Petty, Cacioppo a Goldman (1981) report findings which
suggest that highly involved message recipients process more
message relevant information than low involvement message
recipients, however, they offer no data which suggest that
low involvement message recipients process more peripheral
information than high involvement message recipients.
Without these corresponding data all we know is that high
involvement message recipients process more message content
than low involvement message recipients, and we know nothing
about the ability of message recipients to choose between
one of these two information processing strategies. In sum,
Petty and Cacioppo offer no data which suggest that
individuals are capable of making the type of cognitive

choices which constitute the basic premise of their model.

Before proceeding to discuss the empirical support
which exists for the Elaboration Likelihood Model, a
discussion of research on information processing will be
presented in order to assess the accuracy of the unchecked
assumptions of the Petty and Cacioppo model. This research

is presented below.

Meeele er Heman Inrormetion Processing

One of the central premises of the Petty and Cacioppo
model is the notion that individuals have a limited capacity

for processing information. A recent review of research on

 

11

this topic suggests that the limited processing capacity
hypothesis is correct (Kantowitz, 1982). Although most
information processing theorists agree that individuals have
limited capacities for processing information, there is
considerable disagreement about how this processing capacity

is allocated to perform perceptual and cognitive tasks.

Single Channel Eceeeasins Models. The Elaboration
Likelihood Model (Petty & Cacioppo, 1981) suggests that
individuals must choose between processing message content
and processing peripheral message cues. Although Petty and
Cacioppo do not formally acknowledge this, the crux of this
information processing strategy decision stems from an
implicit assumption that individuals are unable to engage in
parallel processing of two different types of stimuli. This
implicit assumption likens the Petty and Cacioppo model to
the information processing models advanced by Broadbent
(1957) and Deutsch and Deutsch (1963). The basic premise of
these two theories is that individuals are unable to process

different stimuli simultaneously.

The Broadbent "filter" theory suggests that human
processors are able to register and store multiple incoming
stimuli but that a "bottleneck" occurs when these stimuli
are ready for perceptual analysis. Broadbent (1957) argues
that individuals are only able to attend to one set of
perceptual stimuli at a time. The Broadbent theory posits

that analysis of unattended items will be impaired, if not

12

precluded, by the limitation that human processors can only
process information which is attended to. A variation of
this basic model was introduced by Deutsch and Deutsch
(1963). Like Broadbent, Deutsch and Deutsch argue that
"bottlenecks” occur in the processing of information.

Unlike Broadbent, Deutsch and Deutsch (1963) argue that the
focus of attention does not affect the ability of
individuals to perceptually analyze stimuli. That is,
individuals should be able to perceptually analyze different
stimuli even if the stimuli were not consciously attended
to. According to Deutsch and Deutsch, the bottleneck in the
processing cycle does not occur until the decision to
respond to a particular set of stimuli occurs. At the
response stage, Deutsch and Deutsch argue that only a single

set of stimuli can be handled effectively.

Although the Petty and Cacioppo Elaboration Likelihood
Model is similar to both the Broadbent (1957) and the
Deutsch and Deutsch (1963) models of information processing,
it is not clear from the presentation of the model which of
these information processing models best describes the
processes outlined by Petty and Cacioppo (1981). Like both
of these information processing models, the Petty and
Cacioppo model suggests that individuals are able to
register and store incoming stimuli from different sources
(message content cues and peripheral cues). In addition,

the Petty and Cacioppo model suggests that individuals are

13

only able to cognitively respond to, and hence be influenced
by, stimuli from one of these sources. Unfortunately, Petty
and Cacioppo are not explicit in describing where
individuals begin processing a single set of cues.
Presumably, this ”bottleneck" occurs at the perceptual
analysis stage, making their model more like the Broadbent
model than the Deutsch and Deutsch model. Only if the
"bottleneck" occurred at the perceptual analysis stage would
.individuals be relatively uninfluenced by stimuli which they
were not attending to during the receipt of a persuasive
message. According to Petty and Cacioppo, individuals
choose to process either message content or peripheral cues
and they are relatively unaffected by the unchosen set of

informative cues.

Although the Petty and Cacioppo Elaboration Likelihood
model is similar to the models of human information
processing outlined by Broadbent (1957) and Deutsch and
Deutsch (1963), these models to information processing have

been by the results of several experiments. These data are

presented below.

Perhaps the most convincing disconfirmation of the
Deutsch and Deutsch model of information processing came
from a study conducted by Treisman a Riley (1969). In this
experiment subjects were asked to continually attend to and
"shadow" (mimic) a list of digits presented through one of

two auditory channels (either the right ear or the left

14

ear). At the same time, subjects were asked to acknowledge
the presence of occasional letters presented in either ear.
Broadbent's model predicted that individuals would be able
to accurately detect occasional letters in the same ear that
the list of digits was being presented because they were
attending to the stimuli received by that ear. Conversely,
Broadbent's filter theory predicted that detection of the
occasional letters in the unattended ear would be low
because stimuli must be attended to before they can be
processed. On the other hand, Deutsch and Deutsch's model
predicted that recognition of the occasional letters would
be equally accurate in the attended and the unattended ear
because attention is not necessary for perceptual
processing. The results of this experiment were very
clear. The subjects detected 76 percent of the letters in
the ear that they were attending and only 33 percent of the
letters in the unattended ear. These results clearly
demonstrate a decrement in information processing when
stimuli are unattended. This finding provides clear
evidence against the Deutsch and Deutsch model of
information processing. Similar findings were reported in a

study conducted by Treisman and Geffen (1967).

Although these data are convincing, they do not provide
a fair test of the Petty and Cacioppo model of information
processing if we are to assume that the Petty and Cacioppo

model more closely resembles the Broadbent (1957) model than

15

the Deutsch and Deutsch (1963) model. However, considerable

data exist to reject the Broadbent model as well.

Although the early research on this issue supported the
Broadbent (1957) filter model of information processing
(Mowbray, 1953; 196A; Moray, 1970a; 1970b; Moray & O'Brien,
1967; Treisman s Geffen, 1968; Webster a Thompson, 1959),
this support predominately took the form of identifying
experimental procedures which failed to document the ability
of individuals to process different stimuli simultaneously.
Although these studies failed to reject the Broadbent (1957)
model, recent research has clearly documented the ability of

individuals to engage in parallel information processing.

Treisman (1970) presented subjects with auditory
stimuli consisting of: (1) a pair of nonsense syllables, (2)
a digit and a nonsense syllable, (3) a single nonsense
syllable, and (u) a single digit. Subjects were exposed to
each of these four types of stimuli randomly ordered and
presented one at a time and were asked to detect whether or
not a digit was presented in the particular stimulus.
Subjects did not know in advance which of the four types of
stimuli they would receive. Subject response time was the
critical dependent measure. After completing several trials
the experimenter exposed the subjects to similar sets of
stimuli. This time however, the subjects were cued to
listen for a specific digit in the stimulus. Subjects were

told which digit, if any, would appear in the following

16

stimulus. This cueing procedure has been demonstrated to
reduce response time because it facilitates the search for

critical aspects of the stimulus message.

Treisman hypothesized that if Broadbent was correct and
individuals process different stimuli independently, then
the "cueing" effects on response time should be doubled for
pairs of stimuli compared to single stimuli. Thus,
Broadbent's model predicts that the effect of additional
information (cues) should produce greater reductions in
response time when subjects respond to either pairs of
nonsense syllables or a nonsense syllable and a digit than
when subjects respond to either a nonsense syllable or a
digit. If, on the other hand, advance cueing produces
similar reductions in response time to both single and
paired stimuli, then this similarity would be evidence of
parallel processing (Treisman, 1970, p. 192).

Treisman reported results consistent with the
hypothesis that humans are capable of parallel information
processing. She found that advanced cueing reduced response
time by about 115 milliseconds for both single and paired
stimuli when compared to the no cue conditions. These data
suggest that individuals are capable of parallel processing
and are not limited to single cue processing as the
Broadbent (1957) model suggests. Results from several
'related studies (Lindsay, 1970; Lindsay, Cudding, a Tulving,

1965; Treisman a Fearnely, 1971; Tulving & Lindsay, 1967)

17

support this conclusion that humans are multi-channel

information processors.

Given that the two models (Broadbent, 1957; Deutsch &
Deutsch, 1963) which best resemble the model of information
processing implied by the Petty and Cacioppo Elaboration
Likelihood Model (1981) are inaccurate representations of
human information processing, the model specified by Petty
and Cacioppo is also an inaccurate representation of the
processes individuals engage in when they evaluate
persuasive messages. Specifically, the assumption that
individuals are always forced to choose between
systematically processing central message cues and
processing heuristic peripheral cues (Petty a Cacioppo,
1881) depicts humans as single channel information
processors. Since the present data clearly suggest that
individuals are multi-channel information processors, a
modification of the Petty and Cacioppo model which explains
the occurrence of both single channel processing and
parallel cue processing is warranted. This modification
must begin with an understanding of the factors which
facilitate and inhibit an individual's ability to process
different stimuli simultaneously. One information
processing theory which is particularly relevant to this
quest was developed by Kahneman (1973). A discussion of

this model is presented below.

18

Belti-Qheaeel Eeeeessias Medals. In 1973. Kahneman
introduced his Elastic Capacity View of information
processing. This model suggests that humans are
multi-channel limited capacity processors. Kahneman
supported the view that humans are capable of parallel
stimulus processing which evolved from and is supported by
the research discussed in the preceding paragraphs.
Kahneman's position that humans have a limited processing
capacity was not new either. What is novel about this
approach was the concept of an elastic capacity to process

information.

Kahneman argued that within the bounds of a fixed upper
limit, the amount of processing capacity available to an
individual at any one time is a function of the difficulty
of the processing task (effort) and to a lesser extent, the
level of arousal of the individual. Presumably, individuals
aroused and processing difficult tasks exert more cognitive
effort, and hence have more processing capacity available to
them than individuals who are unaroused and processing
simple tasks. Rahneman, Pearler, and Onuska (1968)
supported this claim. They asked subjects to engage in
either simple short term memory tasks or more difficult
addition tasks. For each task condition, they varied the
amount of monetary rewards (punishments) associated with
performance. The dependent measure, cognitive effort, was

assessed by measuring pupil dilation during task

19

performance. Kahneman er elr reported a main effect for
task difficulty indicating that more effort was exerted in
the difficult task than in the simple task. There was also
a small incentive effect, but this difference was limited to
the simple memory task condition. Thus, the Kahneman et

el. findings are consistent with the notion that cognitive
capacity (effort) expands as a function of task difficulty
(capacity demand), and in some cases, involvement produced

by monetary incentives.

Kahneman describes an individual's capacity to process
information as a negatively increasing function of capacity
demanded by the primary task (Figure 1). Implicit in this
description is the notion that there is always some capacity
available to process secondary tasks. Inspection of Figure
1 reveals the nature of this relationship. At low levels of
demand, the amount of capacity supplied to the primary task
is much less that the total capacity, leaving considerable
"spare" capacity for simultaneous secondary information
processing. At higher levels of primary task demand, the
amount of capacity supplied to the primary task approaches
total capacity, leaving little "spare" capacity for
secondary information processing (Kahneman, 1973). One of
the major advantages of Kahneman's model is this unique
ability to account for research findings supporting both

single channel and multi-channel hypotheses.

Kahneman (1973) argues that serial (sequential or

20

single channel) information processing and simultaneous
(parallel or multi-channel) information processing are both
predicted by his model. At low levels of primary task
capacity demand enough spare capacity exists to permit
secondary tasks to be processed in parallel with the primary
task. At high levels of primary task effort, the level of
spare processing capacity is diminished, forcing the
processor to focus on a single task. This analysis is
parsimonious with existing research on motivation and
attention which demonstrates that high levels of arousal
cause attention to be focused on a few central aspects of
the situation at the expense of less central, extraneous

aspects. This research is presented below.

Deteemieeate e: Que focusing and Qeeeeitx Allegation-
In 1959, Easterbrook reviewed research on the effects of
motivation and emotion on cue utilization in information
processing. From this review Easterbrook hypothesized that
increased drive tends to interfere with the use of
incidental cues, perhaps diminishing their value and
delaying reaction time to them. Meanwhile, however,
increased drive also tends to sharpen or concentrate action,
and perhaps enhance the importance of central cues and
expediate reaction to them (1959, p.192). This hypothesized
attention process is consistent with Kahneman's (1973)
suggestion that as processing intensity increases, the

amount of capacity available to process secondary tasks

21

decreases.

Easterbrook (1959) cites considerable empirical support
for this hypothesis. Bahrick, Pitts, and Rankin (1952) had
subjects engage in a primary visual tracking task and asked
them to also conduct a secondary light recognition task.

All subjects performed these two tasks simultaneously with
and without the presence of monetary incentives. Results
indicated that as expected, scores on central tasks were
improved with the use of incentives (high arousal
condition), however the scores on the secondary tasks were
not improved through the use of incentives for primary task
processing. In fact, when incentives were offered for
performance on the primary task, there was a marked
decrement in performance on the secondary task. These
findings suggest that increases in arousal produce a
focusing of attention toward the primary task and away from
the secondary task. This effect of arousal on attention
allocation is not limited to the use of monetary
incentives. This same effect was produced when "heat
blasts" (Bursill, 1958), electric shocks (Cornsweet, 1969),
noise (Hockey, 1970a), and sleep deprivation (Hockey, 1970b)

were used to manipulate subject arousal.

It is interesting to note that subjects can be induced
to allocate more processing capacity toward primary
processing tasks and away from secondary processing tasks of

stimuli from the same perceptual channel (visual stimuli)

22

(Kahneman, 1967). They can also be induced to allocate more
capacity to process stimuli from one sensory channel (visual
or auditory) and less capacity to process stimuli from
another sensory channel (Lane, 1979). More important,
Kahneman (1973) cites the work of Kantowitz and Knight
(197A, 1976) and suggests that humans can consciously adopt
and apply a capacity allocation rule for processing

information.

In sum these data provide considerable support for the
hypothesis that increases in arousal cause individuals
processing multi-channel stimuli to focus attention on a
primary task and away from secondary tasks. Moreover,
humans seem able to devise and implement capacity allocation
rules for processing information. These findings are
consistent with the Kahneman's model of information

processing outlined above.

Msdifieetiea of the Eleeecetiec Liseliheee Eeeel

Given that Petty and Cacioppo's view of humans as
single channel information processors is inconsistent with
existing research on information processing, modification of
the Elaboration Likelihood Model seems necessary for it to
have utility in accurately predicting human processing of
persuasive messages. Because the major source of this
inconsistency stems from Petty and Cacioppo's assumption

that humans cannot process different stimuli simultaneously

23

while information processing research suggests that they

can; modifications of the Petty and Cacioppo model should
include the prediction of parallel information processing.
Although a number of different modifications can produce a

reconciliation between the proposed theory and existing

research, perhaps the most advantageous modification
includes adapting the Petty and Cacioppo model to Kahneman's

model of information human processing.

Adapting the Petty and Cacioppo model to Kahneman's
model is advantageous for two reasons. First, as noted
above, considerable empirical evidence supports Kahneman's
model of human information processing. Thus, one can be
confident that it is accurate. Second, modification of the
Elaboration Likelihood Model to include Kahneman's notions
of elastic processing capacity and processing capacity
allocation represents an extension rather than a redirection
of Petty and Cacioppo's original thinking. Application of
the Kahneman model explains the capacity allocation
mechanisms implied, but never directly addressed, by the
Elaboration Likelihood Model. In addition, application of
the Kahneman model extends the predictive range of the
Elaboration Likelihood Model by predicting situations where
individuals engage in parallel processing and are influenced
by both central and peripheral cues. Integration of these
two models not only provides a clear description of the

factors which influence the processing of central and

24

peripheral cues, but it also identifies the mechanisms which
affect these factors. The end result is a more parsimonious
representation of the effects of central and peripheral cues
on the attitudes of individuals following exposure to a
persuasive message. A sketch of this integrated model is

presented below (Figure 2).

During exposure to persuasive messages individuals are
bombarded with a number of different cues. When these cues
are directly related to the content of the persuasive
message and its recommendations they are referred to as
central cues. When the cues are related to non-message
aspects of the persuasive situation (i.e., source
characteristics) they are referred to as peripheral cues
(Petty and Cacioppo, 1981). During persuasive message
evaluation individuals tend to focus primarily on one type
of cue (central or peripheral) and secondarily on the
other. As a result, attention and processing capacity is
directed toward the primary cues. Processing capacity which
remains after the primary cues have received their

allocation is then available for secondary cue processing.

At low levels of involvement, individuals are not
motivated to process many persuasive cues. Hence,
information processing capacity demanded by the primary task
(either central or peripheral cues) is minimal, leaving
considerable processing capacity for secondary cues.

Although enough processing capacity exists to meet demand

25

and individuals are able to process both sets of cues
simultaneously, individuals are not motivated to process
either type of persuasive cue to a great extent. Thus, both
types of cues will be processed to some minor extent and
both types of cues will have a minimal influence on

attitudes.

As involvement increases, individuals become more
attentive to aspects of the persuasive attempt. This
increase in attentiveness causes individuals to demand more
capacity to process the primary cues. As demand for
capacity increases, total capacity also increases because
capacity is a negatively increasing function of primary task
demand. Thus, although the primary task demands more
capacity, the increase in total capacity permits allocation
of capacity to secondary processing tasks. Thus, as
involvement increases to moderate levels, the amount of
capacity allocated to process primary and secondary cues
increases. In addition, because demand has not yet exceeded
capacity, individuals are capable of processing both sets of
cues simultaneously. This increased attention to both
central and peripheral cues causes them to be more effective

in influencing attitudes.

Only when individuals are highly involved with the
persuasive message will they be motivated to process
information beyond their capacity. At high levels of

involvement, the processing capacity demanded by the primary

26

task approaches total capacity leaving little spare capacity
for secondary cue processing. It is at this point that
individuals are forced to choose between processing central
and peripheral cues. This cue selection process has, as
Petty and Cacioppo predicted, direct effects on the role of
central and peripheral cues in persuasion. Chaiken (1980)
argues that accuracy concerns are paramount when message
recipients are highly involved with the message issue. As a
result, highly involved individuals, faced with a choice
between processing central message cues and peripheral cues,
will choose to process central message cues because of their
increased utility for making accurate decisions. Thus,
highly involved message recipients should be influenced by
the central message cues of a persuasive message and not by

the peripheral cues.

Bcedietieas

This reformulated model presents two testable
hypotheses concerning the moderating effect of involvement
on the relationship between both central and peripheral cues

on attitudes. These hypotheses are presented below.

27

Hypothesis 1

There will be a positive linear relationship between the level of
message recipient involvement and the effect that central message

cues have on attitudes.

There will be a curvilinear relationship between the level of
message recipient involvement and the effect that peripheral cues
have on attitudes, such that, as message recipient involvement
increases the effect of peripheral cues on attitudes increases up
to a point, beyond which further increases in recipient

involvement will produce decreases in attitudes.

In addition to these hypotheses concerning the
moderating effects of involvement on the use of central and
peripheral persuasive cues, the reformulated model presented
above also extends the predictions offered by the original
Petty and Cacioppo model concerning the effect of subject
knowledge on the use of central and peripheral cues. Petty
and Cacioppo's Elaboration Likelihood model argued that
knowledge about the message topic provided message
recipients with the ability to process central and
peripheral cues. According to an information processing

view, knowledge about the message topic should increase the

28

efficiency with which message recipients process central and
peripheral cues. Thus, knowledge about the message topic
should increase the number of central and peripheral cues
which individuals can process, and facilitate the effect of
these cues on attitudes. The reformulated model offers the

following hypotheses.

Bxeetbeais 1

There will be a positive linear relationship between the level of
message recipient knowledge about the persuasive message topic

and the effect of central message cues on attitudes.

Bxeetbeais 5

There will be a positive linear relationship between the level of
recipient knowledge about the message topic and the effect of

peripheral cues on attitudes.

Emeicieal Suspect

Although no experiments have been specifically designed
to test these hypothesized effects of involvement and
knowledge on the use of central and peripheral cues, a
number of experiments offer sufficient evidence to permit
indirect evaluation of these hypotheses. Specifically, a

review of current research on the effectiveness of evidence

29

in persuasive messages provides such an evaluation.

Several researchers (Bettinghaus, 1953; Bostrom a
Tucker, 1969; Kline, 1969; McCroskey, 1969) report that the
use of evidence increases the effectiveness of persuasive
messages. However, research by several authors (Cook, 1969;
Dresser, 1963; Gilkinson, Paulson, a Sikkink, 1956;
McCroskey a Combs, 1969; McCroskey, Young a Scott, 1971)
found minimal effects for evidence. Although several
conceptual reviews of_this literature provide a general
description of the effect of evidence in persuasive messages
(Kellerman, 1981; McCroskey, 1969; Reynolds & Burgoon,
1989), these reviews do not contain a discussion of the
extent to which both involvement and knowledge moderate the

effectiveness of evidence in persuasive messages.

Recently, several studies have investigated the
moderating effect of involvement on the use of central and
peripheral cues in persuasion. These studies generally
manipulate the level of recipient involvement with the
message, the use of evidence to support the message claim,
and the perceived credibility of the message source. The
first of these studies was conducted by Harte (1976). Harte
presented subjects with a persuasive message on a topic that
was determined in a pre-test to be either highly involving
(balanced federal budget) or low in involvement (British

Royalty). These messages were presented with either maximal

or minimal evidence from either high or low credible

30

sources. Harte found no immediate effect for either
credibility or evidence on either of the two message
topics. Delayed post-test measures indicated that, in the
high involvement conditions, the evidence manipulation
produced significant changes in the attitudes of message
recipients. There was no delayed effect for the evidence
manipulation in the low involvement conditions. Source
credibility had no delayed effect on attitudes in either of
the involvement conditions. Chaiken (1980) presented
message recipients with messages containing either one or
six arguments from either a likable or unlikable
communicator. The message issue was involving for half of
the message recipients and uninvolving for the half the
message recipients. Results indicated that the high
involvement message recipients were persuaded more when the
message contained six arguments that when it contained one
argument. These message recipients were relatively
unaffected by the likability of the communicator.
Conversely, low involvement message recipients were
persuaded more by a likable source than an unlikable source
and were relatively unaffected by the number of arguments

presented in the message.

Petty, Cacioppo, and Goldman (1981) argued that
manipulating the number of arguments presented in the
persuasive message might constitute a peripheral one as well

as representing obvious central message cues. Petty er elr

31

suggested that when the number of arguments serves as the
evidence manipulation, message recipients who are not
cognitively processing the content of the message might
process the knowledge that the claim had "many" (few)
supporting arguments and base their attitudes accordingly.
Petty and Cacioppo write, "if people are unmotivated or
unable to think about the message, and no other salient cues
are available, they might invoke the simple, but reasonable
decision rule; the more arguments the better, and their
attitudes might change in the absence of thinking about or
scrutinizing the arguments" (1989, p.70). To alleviate this
potential manipulation problem, Petty EL el. (1981)
replicated the Chaiken (1980) experiment with one minor
change. Instead of manipulating central message cues by
altering the number of supporting arguments (as Chaiken
did), Petty QL el. held the number of supporting arguments

constant and varied their quality.

The results of the Petty QE el. (1981) experiment
replicated the results of the Chaiken (1980) study. Further
support was found for the moderating effects of involvement
on the use of both central and peripheral message cues in
persuasion. Since this replication, Petty and his
colleagues have produced several additional experiments
which replicate the same process (Harkins a Petty, 1981;

Petty a Cacioppo, 1983; Petty, Cacioppo, & Schumann, 1983).

Together, these data suggest that involvement moderates

32

the effects of both central message cues and peripheral cues
on attitudes. Unfortunately, because each of these studies
involves the manipulation of only two levels of involvement,
data from these studies are insufficient to assess the
empirical validity of the hypothesized curvilinear effects
of involvement on the use of central and peripheral cues in

persuasion.

In addition to offering further support for involvement
as an important moderating variable, Harte (1971) provided
evidence to suggest that message recipient knowledge also
effects the processing of central message cues. Harte
studied the ability of individuals to apply tests of
evidence and reported that individuals who were highly or
moderately informed on an issue were more accurate when
applying tests of evidence in a message than were poorly
informed individuals. Harte reported similar findings for

subject involvement.

Although no experimental data exist to directly test
the hypothesized relationships presented above, a plethora
of empirical data on the use of supporting information exist
to permit an indirect assessment of the moderating effects
of both involvement and knowledge on the use of central
message cues (and to a lesser extent peripheral cues) on
attitudes. Although many of these previous studies have
conceptual and methodological flaws, data from these studies

are relevant to the above hypotheses and can provide useful

33

insights about the persuasion process. Before continuing
further in this line of research it would behoove us to
organize and review existing knowledge about these issues.
One useful way to organize previous research findings is
through the use of an empirical cumulation technique known

as meta-analysis (Hunter, Schmidt, a Jackson, 1982).

Meta—analysis is the preferred technique for these data
because it separates effects of methodological flaws from
effects due to theoretical processes. By empirically
cumulating research findings into a single set of sample
statistics, we can identify the support which each of the
hypothesized relationships has received from the research on
the use of supporting information in persuasive messages.
The study presented below represents a meta-analytic
investigation of the effects of supporting information in
this study identifies the role of both involvement and
issue-relevant knowledge as moderators of the supporting

information-attitude relationship.

34

Method

A review of relevant research on the effects of
supporting information in persuasive messages was
conducted. This review included a census of relevant
journals in communication, social psychology, marketing,
advertising, and the use of several social science indexes.
The unavailability of unpublished dissertations, theses and
convention papers restricted the literature search to
published articles and book chapters which reported original

data.

Research was included in the review if it contained a
quantifiable estimate of the relationship between the amount
of supporting information presented in the persuasive
message and the attitudes that recipients had toward the
message recommendation. The nature of the supporting
information presented in the persuasive message
manipulations in previous research varied considerably. A
number of studies manipulated the amount of supporting
information by either including or excluding evidence in the
persuasive message (Bostrom a Tucker, 1969; Harte, 1976;
Holtzman, 1966; Kline, 1969; McCroskey, 1970; McCroskey,
Young & Scott, 1972). Although the conceptual definitions of

evidence varied greatly among these studies, messages that

35

included evidence exposed recipients to more central cues
than messages that included no supporting evidence. For
example, McCroskey argued that his evidence manipulations
affected recipients' abilities to go through the mental
process of testing what the communicator says against what

they already know or believe (1970, p.189).

Other studies altered the number of arguments offered
in support of the message recommendation (Calder, Insko &
Yandell, 1979; Chaiken, 1980; Cook, 1969, Lashbrook, 1977;
Ostermeir, 1971). In these studies, the number of arguments
in the message was intended to effect central cue exposure.
For example, Calder et el. designed their study to
demonstrate empirically the relationship between information
and attitudes and to explore the form of this relationship
and the various cognitive and memorial processes involved
(1979, p.63). Still other studies manipulated supporting
information by including either specific or general data in
the message (Cathcart, 1955; McCroskey, 1967). Messages
containing specific data should have produced more central
cue processing than messages containing only general
statements. Specific data should be easier for message
recipients to recognize and process than general statements

of support.

McCroskey and Combs (1969) manipulated supporting
information by including analogies in the message. In this

study, the messages containing analogies provided more

36

central message cues for recipients than the messages
without analogies. Finally, several studies (Harkins and
Petty, 1981; Petty et el. 1981; Petty er el. 1983)
included central cue processing as a manipulated variable in
the experiment. In sum, all of these studies manipulated

recipient exposure to central message cues.

The literature search found 29 published articles and
book chapters. Of these 29 publications, six were unusable
because relevant effect sizes could not be computed from the
results. Four additional studies (Chen, 1933; Dresser,

1963; Luchok a McCroskey, 1978; Norman, 1976) were
eliminated from further consideration because of severe
methodological anomolies. The Chen (1933) study involved
the use of extremely strong and lengthy manipulations. This
study involved the presentation of "pro- and anti-Chinese
propaganda lectures" to entire classes of college students.
The length of this manipulation exceeded one hour and was
many times longer than the next longest message manipulation
in the meta-analysis. Not surprisingly, the effect size for
the supporting information manipulation in this study was
larger than the average effect size of the other studies by
a factor of greater than three. The Dresser study was
eliminated because the correlation between the manipulation
and the manipulation check was negative (r = -.13).
Manipulation problems also eliminated the Luchok a McCroskey

(1978) and the Norman (1976) articles. Luchok & McCroskey's

37

manipulations attempted to vary the relevance of the
supporting information and the quality of the source
citation. However, conceptual ambiguities with these
definitions produced results which were difficult to
interpret. Finally, the Norman study confounded the
supporting information manipulation with the source
credibility manipulation. The high supporting information
condition involved a high source expertise manipulation.
The low supporting information condition included a high
source attractiveness manipulation. Because the attitude
change effects due to the supporting information
manipulation could not be separated from the differing
credibility manipulations, the data in this study were
excluded from further analyses. The remaining 19 data-based
publications produced a total of 31 estimates of the effect
of supporting information on attitudes. A list of these

studies and their sample sizes is presented in Appendix 1.

Precedeee

The meta-analytic procedures used in the present study
are discussed by Hunter, Schmidt, and Jackson (1982).
Essentially, these procedures involve the estimation of
effect sizes between the variables of interest for each
individual study. Once all of the relevant effect sizes are
identified, they are cumulated into a set of summary

statistics which describe the entire body of research.

38

The effect size statistic selected for use in the
present study was the correlation coefficient. This
statistic was selected over the e statistic preferred by
Glass, McGaw, and Smith (1981). Although the r statistic is
a direct transformation of the e statistic, r was selected
because it has several unique advantages. Initially, r is

has a commonly known finite metric ranging from +1.00 to

-1.00; e does not. Second, the use of r is applicable in
related analyses i.e., multiple regression and path
analysis. Third, instead of accepting the variance in the e

statistic as a given state of affairs, as Glass er elk do,
the use of r permits identification of possible sources of
variance in rr Hunter er elk (1982) discuss techniques for
separating the variance in r due to statistical artifacts
i.e., sampling error, measurement error, restriction in
range, from variance in r due to unknown determinants. Once
these statistical artifacts have been identified, the
estimate of the true variance in r provides an estimate of

the stability of the estimated effect size.

Of the 19 studies which were retained for use in the
meta-analysis, only seven of the studies provided an
estimate of the reliability of the dependent measure. The
use of manipulation checks was even more sporadic. When
manipulation checks were used, authors failed to report the
reliability of the manipulation check measure. Thus, the

data in the present analysis were difficult to correct for

39

error in measurement.

Since most studies measured the dependent variable with
semantic differential or Likert response format using
several items; the dependent measures in all 19 studies were
assumed to be homogeneous in quality with the dependent
measures in the seven studies that reported reliability
data. Hunter er elk (1982) provide methods for correcting
population correlation estimates for error of measurement
when there is sporadic reporting of measurement quality.
These methods were used in the present study. Because there
was an extremely limited amount of information regarding the
quality of the manipulation check measures, the present
study only corrected the population correlation estimates

for error of measurement in the dependent variable.

lastcumentetiea

In order to test the hypotheses that involvement and
knowledge moderate the supporting information-attitude
relationship, estimates of the involvement and knowledge
levels of research participants in previous studies had to
be identified. Because very few studies manipulated
involvement (Chaiken, 1980; Harte, 1976; Petty 2E elk 1981)
and none manipulated knowledge, post hoc analyses served as

estimates of subject involvement and knowledge.

In order to obtain these estimates, five raters made

4O

evaluations of the topics used in the 19 studies which were
retained for use in the meta-analysis. All five raters were
Ph.D. students in communication. Because some of the 19
studies made use of two or more persuasive message topic
manipulations, 30 topics were evaluated by these raters.

For each study, raters were asked to consider the topic and
the date that the study was published before determining the
extent to which the subjects in the study (subjects in all
studies were college students) were involved with and
knowledgeable about the the topic used in the persuasive
message. Ratings of involvement for all of the topics were
completed before ratings of knowledge were made.

Evaluations of topic involvement were made using four
semantic differential scales. Bi-polar adjectives were used
as anchors for each scale. The responses of these four
scales were summed to form a single estimate of topic
involvement. The reliability of this four item measure was
quite high (alpha = .96). Evaluations of topic knowledge
were also made using four semantic differential scales.
Responses to these four items were summed to form a highly

reliable measure of subject knowledge (alpha = .99).

The five judges were were consistent in their
assignment of involvement and knowledge levels for each of
the studies. The inter-rater reliability estimate for the
involvement ratings was moderately high (alpha = .79). The

inter-rater reliability estimate for knowledge ratings was

41

somewhat higher (alpha = .80). The estimates of involvement
and knowledge made by each of the five judges ranged from a
low value of 9 to a high value of 28. These ratings of
involvement and knowledge were later grouped into three

general categories; high, medium, and low.

42

Results

The present meta-analysis was divided into several
major sections. Initially, the linear, quadratic and power
function effects of the use of supporting information in
persuasive messages were identified. Next, the hypothesized
moderating effects of involvement and subject knowledge on
the supporting information-attitude relationship were
tested. Finally, although the primary intent of this study
was to assess the effect of supporting information in
persuasive messages, data from several of these studies lent
themselves to a preliminary analysis of the effect of source

credibility on attitude change.

These analyses were conducted because the hypothesized
effect of involvement on the selection of heuristic versus
systematic information processing strategies suggests not
only that involvement and knowledge should moderate the
supporting information-attitude relationship, but also that
involvement and knowledge should moderate the source
credibility-attitude relationship (Chaiken, 1980; Petty er
e;h 1981). Since some data were available to provide a
partial test of this hypothesized effect, these data were
analyzed. The author wishes to acknowledge, however, that a
more stringent test of this effect (one which is free of
sampling error) would require a meta-analysis which includes

a census of all the literature on the effect of source

credibility on attitudes.

43

Effects of suspecting iefetmetiee ea attitudes

A summary of the overall analysis of the linear,
quadratic, and power function effects of supporting
information on attitudes is presented in Table 1. Inspection
of this table reveals that the average linear effect of the
use of supporting information following a persuasive message
was r = .20 (r = .21 when corrected for attenuation due to
error in measurement). There is a substantial amount of
variance in the distribution of these effect size
estimates. Only 53% of this variance is due to sampling
error and error in measurement, leaving 97% due to

unspecified determinants.

Six studies provided an estimate of the quadratic

effect of supporting information on attitudes. The average
correlation among these six estimates was r = .15 (r = .16
when corrected for error in measurement). Because the

variance in the distribution of effect size estimates was
smaller than the sampling error and measurement error
variance, the true variance in the corrected population

estimate was zero.’

Five studies provided data to test the power function
effect of supporting information on attitudes. Data fitting
a power function indicate that messages with low levels of

supporting information have a negligible effect on attitudes

44

while messages with moderate and high levels of supporting

information have an equally strong effect on attitudes. The
average correlation among these five estimates was r = .21
(r = .22 when corrected for error in measurement). Once

again, the variance in the distribution of effect size
estimates was smaller than the sampling error and
measurement error variance. Thus, the true variance in the

corrected population estimate was zero.

In sum, there was considerable unspecified variance in

the estimate of the linear effect of supporting information

on attitudes. This relationship warrants further
investigation. There was no unidentified variance in the
quadratic and power function effect size estimates. These

two estimates required no further investigation.

dddststet Estistlsss Suenettius lstetmstisu and Attitudes

Previous theory and research suggests two variables
which may moderate the effect of supporting information on
attitudes: level of subject involvement and level of subject
knowledge. Each of these variables is addressed separately

below.

luzelzsueuts sueeettius iutesustieus sud attitudes-
The evaluations made by each of the five raters served as
estimates for the level of involvement of subjects with the

persuasive message. Level of involvement was divided into

45

three categories (high, moderate, low). Separate
meta-analyses were conducted on the studies within each of
these categories. Involvement ratings for the persuasive
message topics ranged from a low of 9.6 to a high of 25.2
(possible range was 9-28). These scores were divided into
three equal categories (low a 9-11, moderate a 11-19, high =
19-26). With four exceptions all studies with message topics
that fell into one of these three involvement categories

received the corresponding involvement label.

Studies which manipulated involvement and provided
separate effect size estimates for the manipulated levels
(Chaiken, 1980; Harte, 1976; Petty er elr1983) were not
categorized according to topic involvement ratings. Rather,
effect sizes from low involvement conditions of these
studies were placed in the low involvement category and
effect sizes from the high involvement conditions of these
studies were placed in the high involvement category. The
involvement level of one other set of studies was not
categorized according to topic involvement. In the Calder
er eta (1979) experiments, where subjects participated in
"simulated jury situations", subject involvement was higher
than one would infer given the topic involvement ratings.
Because the procedural task in these experiments was much
more involving than the typical experimental procedure used
in all other studies, the results of these experiments were

placed in the high involvement category.

46

In sum, unless experimental procedures dictated
otherwise, the involvement level of subjects in each of
these studies was categorized according to the topic
involvement ratings. When experimental procedures specified
a general level of subject involvement, effect sizes were
categorized accordingly. The effect size estimates within

each of these three categories were analyzed separately.

The results of these analyses are summarized in Table
2. Inspection of this table reveals that involvement
moderates the effect of supporting information on attitudes
following persuasive messages. Six estimates of this effect
among low involvement subjects produced a mean correlation
of r = .12 (r = .12 when corrected for error in
measurement). The variance in the distribution of these
effect size estimates that could be attributable to sampling
and measurement error was larger than the total observed
variance in these estimates. Hence, the true variation in
the population correlation estimate was zero. Thus, no

further analyses of this subgroup were warranted.

Fourteen estimates of the effect of supporting
information on attitudes among moderately involved subjects
produced an average correlation of r = .18 (r = .19 when
corrected for error in measurement). The variance in the
distribution of these 1? effect size estimates was
relatively large (.0076). However, the variance

attributable to sampling error and error in measurement

47

accounted for a sizable portion of this variance leaving
only 30% of the total observed variance (.0023) due to
unspecified factors. Because the variance due to
unspecified factors was trivial, the estimate was considered
to be stable and no further analyses were computed on this

subgroup.

Eleven estimates of the effect of supporting
information on attitudes among highly involved subjects
produced an average correlation of r = .30 (r = .32 when
corrected for measurement error). Once again, most of the
observed variance in this distribution of effect sizes
(66%), was attributable to sampling error. The amount of
variance due to unspecified factors was relatively small.
Thus, the population correlation estimate was stable and no

further analyses were computed on this subgroup.

In sum, these three subgroup analyses provide strong
evidence for the claim that level of subject involvement
moderates the supporting information-attitude relationship.
Hunter er elr suggest that moderator variables will show
themselves in two ways: (1) the average correlation will
vary from sub-group to sub-group and (2) the average
corrected variance (true variance) for the effect size
estimate will be lower for the subgroups than for the data
as a whole (1982, p. 98). Both of these characteristics are
apparent in the present data. Comparison of the average

correlations within each of the three subgroups reveals that

48

the effect sizes vary with levels of involvement. Moreover,
comparison of the true variance in the population estimate
(Table 1) is larger than the true variance in any of the

subgroup analyses (Table 2).

In sum, analysis of the subgroup data suggests that
involvement is an important predictor in the supporting
information-attitude relationship. The nature of this
relationship is such that as involvement increases, the
strength of the effect of supporting information on

attitudes increases.

tusslsdssc suspecting istesmstieus sud sttitudsss
Since involvement was shown to moderate the supporting
information-attitude relationship (accounting for almost all
of the non-artifactual variance in the distribution of
effect sizes), the search for additional moderator variables
was unwarranted by the data. However, because knowledge was
theoretically hypothesized to moderate the supporting
information-attitude relationship, the utility of this
variable as a moderating factor was tested. As before,
subgroup analyses were computed for each of three levels of

subject knowledge.

The evaluations of subject knowledge made by each of
the five raters served as the estimated level of knowledge
of subjects for each of the message topics. Ratings of

subject knowledge ranged from 9.8 to 26.8 (possible range

49

was 9-28). Once again, these scores were divided into three
equal-interval categories (low = 9-11, moderate = 11-19,
high = 19-27). Study effect sizes were categorized
accordingly. Meta-analyses were computed for each of the
three subgroups. The results of these analyses are

summarized in Table 3.

Inspection of this table reveals that the average
supporting information-attitude correlation among subjects
in the five low knowledge studies was r = .21 (r = .22 when
corrected for error in measurement). A considerable amount
of the total observed variance in the distribution of these
effect sizes was due to unidentified determinants.

Seventeen effect sizes provided an estimate of the effect of
supporting information on attitudes among moderately
knowledgeable subjects (r 8 .19; r = .20 when corrected for
error in measurement). Only 19% of the total observed
variance in these 17 effect sizes could not be attributed to

-sampling error or measurement error.

Finally, the average correlation among subjects in the
high knowledge category was r = .25 (r = .26 when corrected
for error in measurement). Most of the variance (86%) in
this distribution of effect sizes was due to unspecified

factors.

In general, the data in Table 3 provide no support for

subject knowledge as a moderator variable in the supporting

50

information-attitude relationship. There was very little
variance in the effect size estimates of the three
subgroups. Moreover, the average amount of true variance
(variance attributable to unidentified, non-artifactual
factors) in the three subgroup estimates was only slightly
smaller (.0057) than the true variance in the overall
analysis (.0058){ Thus, the level of subject knowledge was
not a moderator of the supporting information-attitude

relationship.

In summary, the data on the effect of supporting
information on attitudes following a persuasive message
suggest that subject involvement is an important moderator
in the supporting information-attitude relationship. As
involvement increases, the effect of supporting information
on post-message attitudes becomes stronger. On the other
hand, level of subject knowledge about the message topic was
not an important moderator in the supporting

information-attitude relationship.

Eifsst st Ssutss stditilitx ed Attitudes

In addition to the data on the supporting
information-attitude relationship, the present study
permitted preliminary analyses of the effect of source
credibility on attitudes. With the understanding that these
data represent only a minute portion of the total literature

on the effects of source credibility, these analyses are

51

presented below.

Previous research on the effect of supporting
information on attitudes produced 10 studies which provided
estimates of the effect of source credibility on attitudes
following a persuasive message. All but one of these 10
studies were retained and used in the present analysis. One
study (McCroskey, 1971) was excluded because manipulated
source credibility had a negative effect on attitudes. All
other studies produced positive effects for source
credibility. Because McCroskey (1970) did not provide
manipulation checks, and because the results of his source
credibility manipulations were inexplicable, the study was
eliminated from further analyses. The nine remaining
studies produced 10 estimates of the effect of source
credibility on attitudes. These 10 effect size estimates

were subjected to a meta-analysis.

A summary of the overall analysis of the effect of
source credibility on attitudes is presented in Table 9.
Inspection of this table reveals that the 10 estimates
produced an average correlation of r = .10 (r = .10 when
corrected for measurement error). There was considerable
variance in the distribution of effect sizes (.0082), 80% of

which was due to sampling error.

Involvement and subject knowledge were hypothesized to

influence individuals' decisions to process information

52

hueristically or systematically. Tests of both hypothesized

moderator variables are presented below.

luxelusmsutc sauces staditilitx aud attitudes- As
before, subgroup analyses were computed to test the utility
of involvement as a moderator variable in the source
credibility-attitude relationship. Source credibility
effect sizes were categorized into one of three levels of
involvement using procedures discussed earlier. Two studies
(Chaiken, 1980; Petty er elk 1983) manipulated involvement
and permitted the estimation of.separate effect sizes for
high and low involvement subjects. For these studies,
credibility effect sizes were categorized according to the
manipulations not the source credibility ratings. Each of
these subgroups was subjected to a separate meta-analysis.

The results of these separate subgroup analyses are

summarized in Table 5.

Involvement moderates the source credibility-attitude
relationship. The average effect size in the low
involvement sub-group was r = .12 (r = .12 when corrected
for measurement error). Because the variance attributable
to sampling error was larger than the observed variance in
the distribution of estimate effect sizes the true variance

was zero and the estimate was considered to be stable.

The average effect size of the four estimates in the

moderate involvement subgroup was r = .29 (r = .25 when

53

corrected for measurement error). Finally, the average
correlation in the high involvement sub-group was r = .05 (r
= .06 when corrected for measurement error). The true

variance in this estimate was also zero.

In summary, analysis of the three subgroups suggests
that involvement moderates the effect of source credibility
on attitude change. There is considerable variance in the
average correlations of the three subgroups and the true
variance of each of the three subgroups is zero. Moreover,
the nature of the moderating effect of source involvement on
source credibility suggests that there is a curvilinear
relationship between the level of involvement and the source
credibility effect size. At low levels of involvement, the
effect of source credibility on attitudes is negligible. At
moderate levels of involvement, the effect of source
credibility on attitudes becomes more pronounced. Finally,
at high levels of involvement, the effect of source

credibility on attitudes decreases considerably.

tusulsdsaa sauces essditilitz aud attitudsss Although
the data presented in the previous section did not warrant
continued search for additional moderator variables, the
theoretical review suggested that knowledge should moderate
the source credibility-attitude relationship. This

hypothesis was tested. Data resulting from a test of this

hypothesis are presented below.

54

Effect sizes for each of the studies in the analysis
were categorized into one of three knowledge subgroups using
the procedures outlined earlier. Meta-analyses were
computed on each of these three subgroups. The results of
these analyses are summarized in Table 6. The average effect
of source credibility on attitudes among low knowledge
subjects was small (r = .08; r = .08 when corrected for
measurement error). All observed variance in the
distribution of effect sizes within this subgroup was

attributable to sampling error.

Further inspection of Table 6 reveals that the average
effect size among moderately knowledgeable subjects was r =
.29 (r = . 25 when corrected for measurement error). The
average effect size among subjects in the high knowledge
subgroup was similar to that of the moderate knowledge group
(r = .21; r = .22 when corrected for measurement error).

All of the observed variance in the distribution of effect
sizes for both subgroups was attributable to sampling

error.

These data suggest that knowledge moderates the source
credibility-attitude relationship. There is considerable
variance in the effect size estimates for these subgroups.
Moreover, all of the observed variance in the distribution
of effect size estimates in each of these subgroups was
attributable to sampling error. However, the observed

moderating effect of knowledge resembles a power function

55

and not the linear effect predicted in Hypothesis 9. At low
levels of subject knowledge the effect of source credibility
on attitudes was negligable. At moderate and high levels of
knowledge source credibility has a similar, more pronounced

effect on subject attitudes.

In sum, these data suggest that both involvement and
knowledge moderate the effect of source credibility on
attitudes. For each analysis, the variance in each of the
three subgroup estimates was considerable. Each moderator
variable also accounted for all of the unidentified variance

in the overall analysis.

56

Discussion

The results from the present study provide strong
support for the Kahneman (1973) model of human information
processing. Consistent with Hypothesis 1, involvement
moderated the effect of supporting information (central
message cues) on attitudes. As involvement (processing
capacity in the Kahneman model) increased, the effect of
supporting information on attitudes also increased.
Involvement also moderated the effect of source credibility
(peripheral cues) on attitudes. The observed relationship
between involvement and the effect of source credibility on
attitudes was such that as involvement increased to moderate
levels, the effect of source credibility on attitudes also
increased. Further increases in involvement, beyond a
moderate level, produced a decrement in the effect of source
credibility on attitudes. This observed curvilinear
relationship between involvement and the effect of source

credibility on attitudes fully supports Hypothesis 2.

This conjoint support for Hypotheses 1 and 2 has
important implications for persuasion research. These data
not only support Kahneman's (1973) view of humans as
parallel information processors, but they also refute
directly the basic assumptions of Petty and Cacioppo's
Elaboration Likelihood Model. Petty and Cacioppo (1981)
assumed that individuals are incapable of parallel cue

processing and hence, are forced to choose between

57

processing central and peripheral cues. They further
assumed that uninvolved message recipients would process
only peripheral ones while involved recipients would choose
to process only central message cues. The results of the
present study suggest that these assumptions are incorrect

and paint a different picture of the persuasion process.

The present findings suggest that at low levels of
involvement, individuals process neither central message
cues nor peripheral cues. Hence, supporting information and
source credibility cues had a negligible effect on message
recipient attitudes. At moderate levels of involvement,
however, message recipients were motivated and able to
process both central message cues and peripheral cues.
Among these message recipients, both supporting information
and source credibility cues affected attitudes. The
similarity between the supporting information effect size
(rho = .19) and the source credibility effect size (rho =
.25) suggests that moderately involved message recipients
are dividing their processing capacity and processing both

sets of cues to a similar extent.

The finding that highly involved message recipients are
influenced primarily by central message cues and are
relatively unaffected by peripheral cues is consistent with
Easterbrook's (1959) hypothesis that increased drive tends
to interfere with the processing of incidental cues.

Presumably, highly involved message recipients focused most

58

of their capacity on processing central message cues leaving
little capacity available to process secondary peripheral

cues.

Taken as a whole, these findings suggest that
Kahneman's Elastic Capacity Model is a defensible
representation of information processing during persuasive
encounters. Humans are capable of parallel cue processing
and apparently are able to divide their attention between

processing central message cues and peripheral cues.

Hypothesis 3 was not supported by the data.
Knowledgeable message recipients were hypothesized to be
more efficient processors of central message cues and hence,
influenced more by them, then unknowledgeable message
recipients. Results indicated that the effect of central
message cues on attitudes was similar across all three

subgroups.

There are two possible explanations for this outcome.
First, although capable of more efficient cue processing,
highly knowledgeable recipients may not have been any more
motivated than moderate or low knowledge recipients to
process central message cues. If there were no differences
in one processing across the three knowledge subgroups, then
one would not expect any differences in the effects of
supporting information on attitudes among these three

subgroups. Since only one of the studies where subjects

59

were coded as being highly knowledgeable was also a study
where they were coded as being highly involved, this

explanation is a distinct possibility.

Second, although highly knowledgeable message
recipients may be more efficient processors of central
message cues, they may also hold more firmly established
beliefs about the persuasive message issue. Indeed, the
processing capacity of knowledgeable message recipients may
enable them to counterargue the message more effectively.
Thus, the hypothesized effect of increased processing
efficiency on attitudes may have been negated by increases
in counterarguing. If this occurred, then one would expect
the effect of supporting information on attitudes to be

similar among the three knowledge subgroups.

Although subject knowledge moderated the effect of
source credibility on attitudes, the nature of this observed
relationship was not the linear relationship predicted in
Hypothesis 9. Rather, the observed relationship represented
a power function where the effect of source credibility cues
on the attitudes of low knowledge recipients was negligible
while the effect of source cues was equally pronounced among
moderately and highly knowledgeable message recipients.

Once again, it may be the case that highly knowledgeable
message recipients although more capable, are no more
motivated to process peripheral cues than moderately

knowledgeable message recipients.

6O

timitatisus

Perhaps the most significant limitation of this study
stems from the post hoc estimates of subject involvement and
knowledge. Because few studies manipulated involvement and
knowledge, these post hoc estimates were necessary in order
to apply the results of these research studies as a formal

test of the hypotheses developed above.

Although the reliability of each measure was
sufficiently large, formal evidence for the validity of
these measures could not be obtained. However, without
these post hoc estimates, data from many of these studies
could not have been included in the analysis. Ideally,
future research on these effects will follow the lead of
Harte (1976), Chaiken (1980), and Petty and Cacioppo (1981)
and incorporate both subject involvement and knowledge as

variables in the research design.’

Several additional limitations of the present study
stem from limitations in previous research. Most of the
studies in the present analysis failed to estimate the
reliability of their dependent measures. Moreover, few
studies estimated the relationship between the manipulated
variable and subjects' perceptions of the manipulations.
Without these reliability and manipulation check data,

corrections for errors in measurement were difficult. In

61

addition to problems in data collection and reporting, a
great majority of the studies in the present analysis did
not incorporate source credibility manipulations in their
design. As a result, analysis of the effects of both
supporting information and source credibility on attitudes
was limited to a small number of studies with a restricted

sample size.

The present findings provide several directions for
future research. Initially, the Kahneman model must be
subjected to further empirical scrutiny before we can fully
establish the utility of this model for persuasion
research. Specifically, an experiment on persuasive
messages that manipulates three levels of subject
involvement, two levels of supporting information, and two
levels of source credibility should be conducted. Such an
experiment should measure the effects of these three
independent variables on attitudes toward the message issue,
impressions of source credibility and amount of supporting
information, and the extent to which message recipients

engage in both central and peripheral cue processing.

Three levels of manipulated subject involvement permit
a test of the hypothesized linear effect between involvement
and the effect of central message cues on attitudes as well

as the hypothesized curvilinear (inverted-U) relationship

62

between involvement and the effect of source credibility on
attitudes. This study permits direct assessment of the cue
processing implied but not tested in previous research
(Petty and Cacioppo only measured the amount of central cue
processing and the present meta-analysis assumed persuasive
cue processing). Only after this assessment has been made
can the processing of both central and peripheral cues be
related directly to the observed differences in attitudes at
different levels of involvement. This experiment would
represent an extension of the Petty and Cacioppo (1981)
model and provide a direct test of Kahneman's (1973) model

in a persuasive context.

In addition, an investigation of the effect of subject
knowledge on attitudes should be conducted. Petty and
Cacioppo argue that subject knowledge affects Binformation
processing. Though this hypothesis is consistent with the
Kahneman model, there are no empirical data to suggest that
knowledge influences processing ability. A study that holds
involvement constant and manipulates levels of knowledge,
while measuring cognitive processing, could provide a test
of this hypothesis. The meta-analysis found that recipient
knowledge of the message issue was not related to the effect
of supporting information on attitudes. It was suggested
that the effects of increased processing efficiency among
highly knowledgeable message recipients may have been

negated by increases in counterarguing. Another

63

investigation that holds involvement constant and
manipulates knowledge could provide a test of these
hypotheses. If measures of processing efficiency and
counterarguing both increase as knowledge increases, then
support would exist for the combined null effect of these

variables on attitude change.

In addition to further tests of the Kahneman model,
future research should attempt to integrate this model with
important aspects of the persuasion process. Specifically,
investigations should measure attitude change over time and
attempt to identify combinations of cue processing that
produce persistent attitudes. It may be that individuals
who choose to process both central and peripheral cues have
a broader foundation of support for their attitudes, and
hence, have a more diverse pool of arguments with which they
can resist future counterpersuasive attempts, than
individuals who initially process only one set of persuasive
cues. A study that manipulates different combinations of
central and peripheral cue processing and then measures
resistance to counterpersuasion will provide an answer to

this question.

Analyses in the present study were limited to central
cue information that was provided by the source of a
persuasive message. In addition, the only peripheral cues
that were analyzed were those relating specifically to

source credibility. Identification of additional sources of

64

central and peripheral cues and the relative ability of
individuals to process these cues, will provide social
scientists with a more complete picture of the persuasion
process. Finally, because the present analysis was limited
to a cumulation of source credibility effects from nine
studies, a more comprehensive meta-analysis on the effects
of source credibility on attitudes would be enlightening.
Not only would a project of this type provide additional
data to test the validity of the Kahneman and Petty and
Cacioppo models, it would also constitute a synthesis of the

source credibility literature.

65

References

Bahrick, H.P., Fitts, P.M., & Rankin, R.E. (1952). Effect
of incentives upon reactions to peripheral stimuli.

deutual oi Exustimautal Esxstslesx. it. 900-906.

Bostrom, R. N. and Tucker, R. K. (1969). Evidence,
personality, and attitude change. QQQEQQLQQLLQQ

Eeusstaeus. 1t. 22-27.

Berscheid, E. (1966). Opinion change and communicator
communicates similarity dissimilarity. geerrel er

Estseualitx add Seeial Esxehelesx. 9. 670-680.

Broadbent, D.E. (1957). A mechanical model for human
attention and immediate memory. Peyerelegreel Beyree,

Q9, 205-215.

Burnstein, E., Stotland, E. & Zander,_A. (1961). Similarity
to a model and self evaluation. Jeergel ef Aerermel

add Sadial Esxsbelesx. t2. 257-269.

Bursill, A.E. (1958). The restriction of peripheral vision
during exposure to hot and humid conditions. geerterly

Journal of Experimental Psycholegy, 1g, 113-129.

Cacioppo, J.T., Petty, R.E., & Morris, K.J. (1983). Effects
of need for cognition on message evaluation, recall and
persuasion. deutual of Eatssualitx aud Sssial

Psychology. 95. 805-818.

66

Calder, B J., Insko, C A.. & Yandell, B. (1979). The
relation of cognitive and memorial processes to
persuasion in a simulated jury trial. Journal of

Applied Sesial Esxshelesx. 9. 62-93.

Cathcart, R. S. (1955). An experimental study of the
relative effectiveness of four methods of presenting

evidence. Speech Monographs. 22. 227-233.

Chaiken, S. (1979). Communicator physical attractiveness
and persuasion. deutual of Batseaalitx and desist

Bsxehslosx. 31. 1387—1397.

Chaiken, S. (1980). Heuristic versus systematic cue
processing and the use of source versus message cues in
persuasion. deutual of Batssaalitx and assist

Bsxstolssx. 31. 977-109.

\Chaiken, S & Eagly, A.H. (1976). Communication modality as
a determinant of message persuasiveness and message
comprehension. deusual at Eatseualitx and Seeial

Bsxshslssx. it. 605-619.

Chen. W.K.C. (1935). The iatlususa of seal Bcepasauda

meterrel eper ereeeere' attitudes. New York: Archives

of Psychology.

Cook, T.D. (1969). Competence, counterarguing, and attitude

change. deutual of Eatseualitx. 11. 392-358.

67

Cornsweet, D.J. (1969). Use of cues in the visual periphery
under conditions of arousal. degree; of Egrerimeetel

Bsxcnolosx. to. 19-18.

Daines, J.E., Hunter, J.E. a Woelfel, J. (1978). Mass
communication and belief change: A test of three
mathematical models. Human Communication Bcsoancn. a.

293-252.

Deutsch, J.A. a Deutsch, D. (1963). Attention: Some
theoretical considerations. Eeyeeelegreel Berger, 19,

80-90.

Dion, K. a Stein, S. (1978). Physical attractiveness and
interpersonal influence. dounnal of Eonsonalitx and

Social Esxcnolosx. 19. 97-108.

Dresser, W. R. (1963). Effect of "satisfactory" and
"unsatisfactory" evidence in a speech of advocacy.

Communication Eonosnapns. 3o. 302-306.

Eagly, A. H. (1979). Comprehensibility of persuasion
arguments as a determinant of opinion change. leereel

of Bonsonalitx and Social Esxcnolosx. 29. 758~773.

68

Eagly, A.H. & Chaiken, S. (1975). An attributional analysis

of the effect of communicator characteristics on

opinion change: The case of communicator

attractiveness. leg

Esxcholosx. 32. 136-199.

Eagly, A. H. and Warren, R. (1976). Intelligence,

comprehension, and opinion change. leereel 92

Bonsonalitx. at. 226-292.

Eagly, A.H., Wood, W. a Chaiken, S. (1978). Causal

inferences and their effect on opinion change. Journal

of Bansonalitx and Social Bsxcnolosx. at. 925-935.

Easterbrook, J.A. (1959). The effect of emotion on cue

utilization and the organization of behavior.

Bsxcnolosical Boxisn. so. 183-201.

Gilkinson, H., Paulson, S. F

., and Sikkink, D. E. (1959).

Effects of order and authority in an argumentative

speech. Quantonlx lounnal of Season. to. 183-192.

Glass, G.V., McGaw, B. and Smith, M.L. (1981).

Esta-analxsis in Social Bosoancn. Beverly Hills. CA.:

Sage.

69

Goethals, G. (1972). Consensus and modality in the
attribution process: The role of similarity of
information. loacaal of Eccaoaaliix aaa Social

Eaxcbolosx. a1. 3n-92.

Goethals, G. a Nelson, R. (1973). Similarity in the
influence process: The belief value distinction.
loacaal oi Ecaaoaaliix aac Social Eaxcbolosy. SS.

117-122.

Greg, R. B. (1967). The rhetoric of evidence. HSSLSEQ

loacaal of Secccb communication. 31. 180-189.

Greenwald, A.G. (1968). Cognitive learning, cognitive
response to persuasion, and attitude change. In A.
Greenwald, T. Brock, a T. Ostrom (Bds.) Eoyonologioo;

foondaoiono of ottitnooo. New York: Academic Press.

Harkins, S.G. & Petty, R.E. (1981). Effects of source
magnification of cognitive effort on attitudes: An
information processing view. loonno; of goooonoligy

aac Social Baxcbolosx. lg. uo1-u13.

Harte, T. (1971). Audience ability to apply tests of
evidence. loacaal oi the imccicaa Eocaaaic

iaaociaiioa. S. 109-115.

70

Harte, T. (1976). The effects of evidence in persuasive
communication. gonogol Stotoo Speech Journal, g1,

42-u6.

Hockey, G.R.J. (1970a). Effect of loud noise on
attentional selectivity. Qnonyonly loonno; of

Saoccimcaial Eaxcbolosx. 2;. 28-36.

Hockey, G.R.J. (1970b). Single probability and spatial
location as possible bases for increased selectivity in
noise. Qaacicclx loacaal oi Sanacimcaial Eaxcholosy.

g2, 37"‘2.

Holtzman, P.D. (1966). Confirmation of ethos as a

confounding element in communication research.

Sommaaicaiioa Boaoscaoba. 33. “fin-“66.

Hunter, J.E., Schmidt, F.L. a Jackson, 6.8. (1982). nogo
aaalxaia: Samalaiias ccacaccb iiaaiasa accoaa aiaaica.

Beverly Hills, Ca.: Sage.

Kahneman, D. (1973). Attention and Effort. Englewood

Cliffs, N.J.: Prentice Hall.

Kahneman, D., Beatty, J., a Pollock, I. (1967). Perceptual

deficit during a mental task. Soionoo, 151, 218-219.

71

Kahneman, D. Pearler, W.S., a Onuska, L. (1968). Effects of
verbalization and incentive on the pupil response to

mental activity. Canadian loucnal of Eaxcbolosv. 22.

186-196.
Kantowitz, B.H. (1982). Interfacing human information
processing and engineering psychology. In W.C. Howell

a E.A. Fleishman (Eds.) Homon nongogmonoo ono

EEQQQQLleSX. Hillsdale, N.J.: Earlbaum.

Rantowitz, B.H. a Knight, J.L. (197a). Testing tapping and
time sharing. loacaal oi Sanccimcaial Baxcboloax. 193.

331-336.

Rantowitz, H.B. a Knight, J.L. (1976). Testing tapping and
time sharing II: Auditory secondary task. noon

Eavcbolosica. i9. 3u3-362.

Kellerman, K. (1980) The concept of evidence: A critical
review. loacaal oi the imccicaa Eoccaaic iaaociaiioa.

1Q. 159-172.

Kline, J. A. (1969). Interaction of evidence and readers'
intelligence on the effect of short messages.

Qaacioclx loacaal oi Socccb. SS. ”07-413.

Lane, D.H. (1979). Developmental changes in
attention-deployment skills. goonnol of Egoonimonoo;

Sailc Bavcbolosx. SS. 16-29.

72

Lashbrook, W.B., Snavely, W.B., & Sulliven, D.L. (1977).
The effects of source credibility and message
information quality on attitude change in apathetics.

communication uoaoscaoha. ii. 252-262.

Lindsay, P.H. (1970). Multichannel processing in
perception. In D.I. Hostofsky (Ed.) Attonoion:

Soaicmooaacy ibaocy and analysis. New York:

Appleton-Century-Crofts.

Lindsay, P.H., Cuddy, L.L., & Tulving, E. (1965). Absolute
judgments of simultaneously presented visual and

auditory stimuli. ancboaomic Science. a. 211-212.

Luchok, J.A. & McCroskey, J.C. (1978). The effect of
quality of evidence on attitude change and source
credibility. Sociacaa Sococb Sommaaicaiioa loacaal.

9.3.9 371’383.

McCroskey, J.C. (1967). 'The effects of evidence in

persuasive communication. Western Soc on, 31, 189-199.

McCroskey, J. C. (1969). A summary of experimental research
on the effects of evidence in persuasive
communication. Qaaciccly loacaal of Soccca. 55.

169-176.

McCroskey, J.C. (1970). The effects of evidence as an
inhibitor of counterpersuasion. Speech Monogooono, 31,

188-19”.

73

McCroskey, J.C. & Combs, W.H. (1969). The effects of the
use of analogy on attitude change and source

credibility. loacaal of Sommanicaiioa. 12. 333-339.

McCroskey, J.C., Young, T.J., & Scott, M.D. (1972). Special
reports: The effects of message sidedness and evidence
on inoculation against counterpersuasion in small group

communication. S_mmuaicatioa Boaoscaoba. 39. 205-212.

McGuire, W.J. (1961). Resistance to persuasion conferred by
active and passive refutation of the same and alternate
arguments. loacaal of ioaocmal Social Eazchology. o3.

326-332.

Miller, H.D. & Burgoon, M. (1979). The relationship between
violations of expectations and the induction of
resistance to persuasion. Homon QQEEHBLQSLLQQ

Bcacaccb. S. 302-313.

Mills, J. & Aronson, E. (1965). Opinion changes as a
function of the communicator's attractiveness and
desire to influence. loucaal oi Eccaoaaliiy and Social

ancbolosy. 1. 173-177.

Mills, J. and Harvey, J. (1972). Opinion change as a
function of when information about the communicator is
received and if he is attractive or expert. loonnol of

Eccacaaliiy and Social Baycaolosy. 21. 52-55.

74

Moray, N. (1970a). Time sharing in auditory perception:
Effect of stimulus duration. Jonnnol of tno Aoooooioo;

Sociciy oi imccica. 51. 660-661.

Moray, N. (1970b). Introductory experiments in time
sharing: Detection of intensity and frequency
increments. louaaal oi the Acoustical Society oi

imccica. £1. 1071-1073.

Moray, N. a O'Brien, T. (1967). Signal detection theory
applied to selective listening. gongnol of tho

Acoustical Society oi Amccica. ii. 765—772.

Mowbray, G.H. (1953). Simultaneous vision and audition: The
comprehension of prose messages with varying levels of

difficulty. louaaal oi iaoccimcaial Baxcbology. aa.

365-372.
Mowbray, G.H. (196a). Perception and retention of verbal
information presented during shadowing. gooono; of EDS

icoaaiical Society oi imacica. 3o. 1u59-1u65.

Norman, R. (1976). When what is said is important: A
comparison of expert and attractive sources. goonno;

oi iaoccimcaial Social iaycbolosy. 12. 29u-300.

75

Osterhouse, R.A. & Brock, T.C. (1970). Distraction
increases yielding to propaganda by inhibiting
counterarguing. loucaal oi iccsoaaliiy and Social

Esycbolosy. 15. sun—358.

Ostermeir, T.M. (1967). The effects of type and frequency
of reference upon perceived source credibility and

attitude change. _ pooo on gono gr roono, 3g, 137-1un.

Petty, R.E. and Cacioppo, J.T. (1979). Issue involvement

can increase or decrease persuasion by enhancing

2

IO

message relevant cognitive responses. Jour rn_;

Eccsoaaliiy and Social Esycbolosy. 31.1915-1926.

Petty, R.E. & Cacioppo, J.T. (1981). Attitoooo and

Id
0
'1
m
C
9
(ll
H
O

{5
If}
”—I

0 m

ID
ID
In-
K)
IO

' I3

I"
I0
B
IO
IO
l"!
III
I"!
K
[U
M

'1
0
IO
0
3'
O
m

Dubuque, IA.: Wm.

Petty, R.E. & Cacioppo, J.T. (1983). Central and peripheral
routes to persuasion: Application to advertising. In
L. Percy & A. Woodside (Eds.) ndgo noioin g ono oononmon

ooyonology. Lexington, MA.: D.C. Health.

Petty. R.E. & Cacioppo, J.T. (1989). The effects of
involvement on responses to argument quantity and
quality: Central and peripheral routes to persuasion.
loo coal oi Bacsooality and Social Psychology. lo.

69-

76

Petty, R.E., Cacioppo, J.T., and Goldman, R. (1981).

Personal involvement as a determinant of argument based
persuasion. Journal oi iacsoaaliiy aao Social

isycboloSy. 11. 8&7-855.

Petty, R.E., Cacioppo, J.T., & Schumann, D. (1983). Central
and peripheral routes to advertising effectiveness: The
moderating role of involvement. gonnnol of Qonoomoo

Bcscaccb. 19. 13u-1u8.

Petty, R.E., Ostrom, T.M., & Brock, T.C. (1981). Historical
foundations of the cognitive response approach to
attitudes and persuasion. In R.E. Petty, T.M. Ostrom,
a T.C. Brock (Eds.) Sosniiiya ccsooasca is

ponoooogon. Hillsdale, N.J.: Earlbaum.

Pokorny, G.E. & Gruder, C.R. (1969). An experimental study
of the effect of satire used as support in a persuasive
speech. flasicca loacaal oi Sooccb Sommoaicaiioa. 33.

204-211.

Reynolds, R.A. a Burgoon, M. (1983). Belief processing,
reasoning and evidence. In R.N. Bostrom (Ed.)

Qommnnioooion yearbook o. Beverly Hills: Sage.

Snyder, M. a Rothbart, M. (1971). Communication
attractiveness and opinion change. Qanooion Joonnol of

Bobsvioaal Scic_cc. 3. 337-387.

77

Treisman, A.M. (1970). Perception and recall of
simultaneous speech stimuli. Logo Eoyonoiogioo, 33,
132-1h8.

Treisman, A.M. a Fearnley, J.S. (1971). Can simultaneous

speech be classified in parallel? Eonoontion ono

Treisman, A.M. a Geffen, G. (1967). Selective attention on
cerebral dominance in perceiving and responding to
speech messages. Qoaciccly loacaal oi Saoccimcaial

isycbolosy. 19. 1-17.

Treisman, A.M. & Geffen, G. (1968). Selective attention and
cerebral dominance in perceiving and responding to
speech messages. Qaaciaclx loucaal oi Ssoccimcaial

isycbolosy. 29. 139-151.

Treisman, A.M. & Riley, J.G.A. (1969). Is selective
attention selective perception or selective response?
A further test. loacaal oi Saoccimcaial isycbolosy.

19.) 27.3“.

Tulving, E. & Lindsay, P.H. (1967). Identification of
simultaneously presented simple visual and auditory

stimuli. iota EsycboloSioa. 21. 101-109.

78

Vinkour, A. & Burnstein, E. (1979). The effects of
partially shared persuasive arguments on group induced
shifts. logical oi Eacsoaality and Social Esycbolosy.

Wall, V.D. (1972). Evidential attitudes and attitude
change. Hostess loacaal oi Soacch iommaaicatioa. 39.

115-123.

Webster, J.C. a Thompson, P.O. (1959). Responding to both
of two overlapping messages. ioacaal oi the Acoustical

Society oi America. 29. 396-u02.

Whitehead, J.L. (1971). Effects of authority-based
assertion on attitude and credibility. Qommnniootion

Boaoscaobs. 39. 311-315.

79

Table 1
Summary of the Overall Analysis of

the Effect of Evidence on Attitudes

Effects
Linear Quadratic Power Function
Number of estimates 31 6 5
Number of subjects #703 979 879
Average r .20 .15 .21
Variance in r .0115 .0095 .0049
Sampling error variance .0061 .0058 .0052
True variance .0059 -.0013 .0003
True standard deviation .073 .00 .00
Average corrected r .21 .16 .22
True variance in corrected r .0055 -.0016 .0009
True corrected standard .075 .00 .00

deviation in corrected r

80

Table 2
Summary of the Linear Effect of Supporting

Information on Attitudes Broken Down by Involvement

Involvement

Low Moderate High
Number of estimates 6 1a 11
Number of subjects 929 2383 1387
Average r .12 .18 .30
Variance in r .0012 .0076 .0098
Sampling error variance .0063 .0055 .0066
True variance -.0051 .0021 .0032
True standard deviation .00 .096 .057
Average corrected r .12 .19 .32
True variance in corrected r -.0057 .0023 .0033
True standard deviation in .00 .098 .058

corrected r

81

Table 3
Summary of the Effect of Evidence on

Attitudes Broken Down by Subject Knowledge

Knowledge

Low Moderate Hig
Number of estimates 5 17 5
Number of subjects 1323 2545 719
Average r .21 .19 .25
Variance in r .0081 .0072 .0165
Sampling error variance .0035 .0063 .0062
True variance .0096 .001 .0103
True standard deviation .068 .032 .101
Average corrected r .22 .20 .26
True variance in corrected r .0050 .001 .0112
True standard deviation in .071 .032 .106

corrected r

82

Table 9

Summary of the Linear Effect of Credibility on Attitudes

Linear Effect

Number of estimates 10
Number of subjects 1395
Average r .10
Variance in r .0082
Sampling error variance .0066
True variance .0016
True standard deviation .09
Average corrected r .10
True variance in corrected r .0018

True standard deviation in corrected r .092

83

Table 5
Summary of the Effect of Credibility

on Attitudes Broken Down by Involvement

Involvement

Low Moderate High
Number of estimates 5 3 2
Number of subjects 891 328 159
Average r .12 .29 .05
Variance in r .0096 .0030 .0020
Sampling error variance .0056 .0081 .0125
True variance -.0010 -.0051 -.O105
Average corrected r .12 .25 .06
True variance in corrected r -.0011 -.0055 -.0116
True standard deviation in .00 .00 .00

corrected r

84

Table 6

Summary of the Effect of Credibility on

Attitudes Broken Down by Subject Knowledge

Number of estimates

Number of subjects

Average r

Variance in r

Sampling error variance

True variance

True standard deviation
Average corrected r

True variance in corrected r

True standard deviation in
corrected r

Low

692

.08

.0000
.0030

.0030

.00

.08

.0033

.00

Subject Knowledge

Moderate

5
935

.29

.0057
.0103

-.0096

.00

.25

-.0052

.00

High

398

.21

.0036
.0053

.0017

.00

.22

.0020

.00

85

Figure 1

Kahneman's model of elastic capacity allocation

Capacity.
Supplied

I

 

Supply = Demand

Total ////
Capacit)\ '/__ __ _ _,

1:... Spare

- /// Capacity
I

./ //’
//, ‘K\ Capacity Supplied to
Primary Task

/

’

Capacity Demanded by Primary Task

Figure 2

A model of capacity allocation for persuasive cue processing

Capacity
Supplied

/’

 

Capacity Available

Total for Peripheral Cue
. Processin
Ca ac1t g
”
/'
’-

Capacity Allocated for
Central Cue Processing

 

Level of Processing Capacity

87

Appendix A

Studies Providing a Quantifiable Estimate of

the Effect of Supporting Information on Attitudes

Authors Estimate

Bostrom a Tucker (1969)
Cacioppo, Petty a Morris (1983)
Calder, Insko a Yandell (1979)
Calder, Insko & Yandell (1979)
Calder, Insko & Yandell (1979)
Calder, Insko & Yandell (1979)
Cathcart (1955)

Chaiken (1980)

Chaiken (1980)

Cook (1969)

Cook (1969)

Harkins a Petty (1981)

Harkins a Petty (1981)

Harte (1976)

Harte (1976)

Holtzman (1966)

1

1

Total N

158
7a
315
105
168
206
31s
79
79
188
17
so
100
no
no

100

Evid.
.26
.53
.30
.20
.39
.20
.15
.26
.13
.09
.51
.23
.26
.21
.29

.11

Effect Size

Cred.

.16
.16
.27

.30

.25
.31

.16

88

Effect Size

Authors Study Total N Evid. Cred.
Kline (1969) 1 150 .21 -
Kline (1969) 2 150 .22 -
Kline (1969) 3 150 .39 -
Lashbrook (1977) 1 119 .10 .07
McCroskey (1967) 1 120 .16 -
McCroskey (1967) 2 120 .25 -
McCroskey (1970) 1 132 .15 -
McCroskey (1970) 2 132 .28 -
McCroskey & Combs (1969) 1 528 .11 .08
McCroskey, Young a Scott (1972) 1 518 .08 -
Ostermier (1971) 1 100 .25 -
Petty & Cacioppo (1989) 1 B3 .55 -
Petty & Cacioppo (1989) 2 83 .06 -
Petty, Cacioppo & Schumann (1983) 1 160 .21 .15

Pokorny & Gruder (1969) 1 92 .32 -

"llllllllllllllll“