- 7-:-_I:.‘. v-7.
’m- f '11:"?- ,-".’,’; ;

.».—--

 

- v t“ w,
i“ 14:33."? .3
: ~ - ‘ ‘- \z. ‘5},9‘“
‘ ‘9': ':‘"‘ WJ‘I.‘I .“fl‘il
x ’ ~‘ 1 7
L ii-K‘K 1;” "g, : .,‘.~l
- . '- " " -\--’...‘;.i’.‘.
z" - ' ' ( . .. _. If“ M 5
.’ ;
‘ - T‘ .< _ .
V ‘ " ‘ - ‘- $9» 523‘ - 3’." A .

   

ox. ,
.

.. a?!“
’t .‘
y ..

..
.e'

cw

flaw»— m- ”4.4,

.. . I
W‘; ..
f K. .I .
"J a"! 1 1f"
-‘/ F3}. «,2
I... ‘. '
,_.r’_-:fl.
diapfpf J
< _ .3.

P . ‘ ,
ion.
.1‘

 

VH..'

'.. Jrr'y '. ..
yi'w‘fiufu'lfu‘gf ,p’, 5
. ".41 1....“

' .1
up; ,f 1U" AI).}
7.2% .A : V ' - I - a? ~v .,
J' ’ -
1.1.: , ‘ ‘ "7
153" 'JIr: Fifi-£52: "

N

 

 

 

 

J "

r—r
Fix-7,.
I:

n
*‘

 

’~".'Vl’o

Z7} .
ll

 

9" ‘ m.
‘ «3%; '35 , . &
'é/yx K. ’2‘... g y
17 ‘f.‘,.’ wit (:1 r
”1'1 nigh}? (“we
P :5"sz ,',, iffy
.r .. 739'. . j -
‘ ”'6'; float“; ,
~ L 'fd'fi’?,:;,?'v
‘1’!»1

'
.4 "I

 

 

' - ‘ a .
,.. . ‘ ' ’
" .1’. o ,Ht'
Ir'lv r“;-..J‘
’ 3 3"1'. .V.‘l

     

 

VEF‘S‘TY UBRAR‘ ES

llllllll ML lll ll \H H‘llll

LIERARY 3 1293 00063 6989
Michigan State
University

 

 

 

 

 

 

 

 

 

 

This is to certify that the

dissertation entitled

TESTING THE ELABORATION LIKELIHOOD MODEL OF PERSUASION:

EFFECTS OF RECIPIENT INVOLVEMENT ON SOURCE AND

MESSAGE PROCESSING
presented by

Paul A. Mongeau

s; 3.
has been accepted towards fulfillment
of the requirements for

Doctor oLEhilemzhldegree in Communicat ion

am 920%

Major professor

Date 2/1CH88

MSU i: an Aflimmtiw Action/Equal Opportunity Institution 0-12771

 

 

MSU RETURNING MATERIALS:
Place in book drop to
LlaaARJES remove this checkout from
.—_ your record. FINES will
~ » be charged if book is
returned after the date
stamped below.

 

 

 

V $039" 5 _ ‘
, Amt” '8 _ a'. m .., a
I '3 , {Elg'é‘t‘fiim
. K17 ~ 171 $41., , s j?
Ell-BMW 001' 011999
W»-
2 ~19 5,. ARR} Mates
"Km 1 0 1993; A3“? 0 11929.05
9°83 “’3‘?

 

 

 

TESTING THE ELABORATION LIXELIHOOD MODEL OF PERSUASION:
EFFECTS OF RECIPIENT INVOLVEMENT 0N SOURCE AND NEESAGE PROCESSING

By

Paul A. Nongeau

A DISSERTATION

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

DOCTOR OF PHILOSOPHY

Department of Communication

1988

I”!
e

v/V/" J!

ABSTRACT
TESTING THE ELABORATION LIKELIHOOD MODEL OF PERSUASION:
EFFECTS OF RECIPIENT INVOLVEMENT 0N SOURCE AND MESSAGE PROCESSING

By
Paul A. Mongeau

Although the Elaboration Likelihood Model of persuasion (ELM) has been
extensively researched. the causal processes alleged to be occurring
within the model have neither been thoroughly specified nor tested.
The present investigation attempts both to specify the causal
relationships between variables within the model and to test that
model. Two alternative interpretations of the ELM are presented that
differ primarily in the hypothesized capacity of human information
processors. From these two alternative interpretations. five causal
models were developed and discussed. A study was performed to
determine the adequacy of these models. None of the five models were
consistent with the data. but a modified model coincided with the
data. Results of the causal modeling and individual difference data
are discussed in terms of the two alternative interpretations of the
ELM. Although highly involved persons processed both source and
message information. they relied on only message information as a basis
for their attitudes. Respondent processing of message information was
consistent with the central route to persuasion. Limitations of the

present study and directions for future research are discussed.

Copyright by PAUL nurses! uoucsan

1988

ACKNOWLEDGMENTS

I would like to acknowledge those persons who assisted me in completing
this document and at the same time helping me remain sane. My primary
thanks and gratitude goes to my committee. especially Dr. James B.
Stiff (Dissertation Director) and Dr. Gerald R. Miller (Chair) without
whose help this dissertation would never have left the ground. let
alone been completed. I would also like to thank the remaining members
of my graduate committee; Dr. Franklin J. Boster. Dr. Kathy Kellermann.
Dr. Katherine 1. Miller. and Dr. John E. Hunter for their help and
support at all stages of my career. More importantly. however. I would
like to thank them for being examples of collegiality and
professionalism that I hope to take with me and model throughout my
career. Since I did not have a chance at the time. I would like to
thank the members of my Naster's committee at Arizona State University;
Dr. Franklin J. Boater (again). Dr. William K. Perrill. and Dr. William

Stinnett for their help assistance and guidance.

In addition. I would like to thank my family for supporting me through
this ordeal. Hy parents. Fred and Martha Taylor. my sister Celeste
Crouch and her family. and my brother Sam have helped immeasurably
through their unswerving interest. support. and uncountable free. home-
cooked. meals. Finally. I would like to thank Arta Damnjanovic for
making the last year of my graduate career my best year.

iv

TABLE OF CONTENTS

List of Tables. . . . . . . . . . . . . . . . . . . . . . . . . . . vii
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . xi
Theoretical Rationale . . . . . . . . . . . . . . . . . . . . . . l
The Elaboration Likelihood Model of Persuasion . . . . . . . . 2
The central route to persuasion . . . . . . . . . . . . . 4
The peripheral route to persuasion . . . . . . . . . . . 5
Criticisms of the Elaboration Likelihood Model . . . . . . . . 8
Predictions Generated from the ELM . . . . . . . . . . . . . . 11
Limitations of Previous Research . . . . . . . . . . . . . . . 17
Modeling the Elaboration Likelihood Model. . . . . . . . . . . 22
Model 1: A.serial model of perceptual processes . . . . . 24
Model 2: A parallel model of perceptual processes . . . . 26
Model 3: A serial model of message processing . . . . . . 29
Model 4: A parallel model of message processing . . . . . 31
Model 5: Three-way model of message processing. . . . . . 33
Individual Differences . . . . . . . . . . . . . . . . . . . . 38
Attitude-Behavior Consistency. . . . . . . . . . . . . . . . . 40
Individual Differences and Attitude-Behavior Consistency . . . 42
Method. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
Participants . . . . . . . . . . . . . . . . . . . . . . . . . 43
Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
Issue involvement . . . . . . . . . . . . . . . . . . . . 43
Source expertise. . . . . . . . . . . . . . . . . . . . . 44
Argument quality. . . . . . . . . . . . . . . . . . . . . 44
Instrumentation. . . . . . . . . . . . . . . . . . . . . . . . 45
Cognitive responses . . . . . . . . . . . . . . . . . . . 45
Issue involvement . . . . . . . . . . . . . . . . . . . . 46
Argument quality. . . . . . . . . . . . . . . . . . . . . 47
Source expertise. . . . . . . . . . . . . . . . . . . . . 48
Attitudes . . . . . . . . . . . . . . . . . . . . . . . . 49
Behavioral intentions . . . . . . . . . . . . . . . . . . 49
Need for Cognition. . . . . . . . . . . . . . . . . . . . 50
Argumentativeness . . . . . . . . . . . . . . . . . . . . 50
Procedures . . . . . . . . . . . . . . . . . . . . . . . . . . 51
Pilot testing . . . . . . . . . . . . . . . . . . . . . . 51
Experimental sessions . . . . . . . . . . . . . . . . . . 52
Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
Pilot Tests. . . . . . . . . . . . . . . . . . . . . . . . . . 53
Argument Strength . . . . . . . . . . . . . . . . . . . . 53
Involvement . . . . . . . . . . . . . . . . . . . . . . . 53
Source expertise. . . . . . . . . . . . . . . . . . . . . 54
Experimental Results: Manipulation Checks. . . . . . . . . . . 54

Argument Strength . . . . . . . . . . . . . . . . .
Involvement . . . . . . . . . . . . . . . . . . . . .
Source expertise. . . . . . . . . . . . . . . . . . .
Experimental Results: Pretest-Attitudes. . . . . . . . . .
Experimental Results: Influences on Post-Message Attitudes
Experimental Results: Source and Message Interactions. . .
Model 1: A serial model of perceptual processes .
Model 2: A parallel model of perceptual processes
Model 3: A serial model of message processing .
Model 4: A parallel model of message processing
Model 5: Three-way model of message processing.
Tests of models: Conclusions. . . . . . . . . .
Experimental Results: Path Analyses. . . . . . . . .
Additional Analyses: Attitude-Behavior Consistency .
Additional Analyses: Individual Differences. . . .
Additional Analyses: Individual Differences and Att t
Behavior Consistency. . . . . . .
Discussion. . . . . . . . . . . . . . . . .
Replication of the Stiff Meta-analysis
Results of Causal Modeling . . . . . .
Conclusions. . . . . . . . . . . . . .
Methodological Differences and Study Li

de-

Directions for Future Research . .
References. . . . . . . . .
Appendix. . . .
Notes . . . . .
Tables. . . . .
Figure Captions
Figures . . . .

a s e s e e e e e e a m e c e e e m e m a

mi

0 e e e "a m e e e
D

e m e s e m e "a m e e e
a...
E

1

vi

54
56
57
58
58
60
61
63
64
66
67
69
7O
76
77

83
85
85
87
92
97
99
102
112
115
117
173
174

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

1.

2.

3.

4.

6a.

6b.

6c.

7a.

7b.

70.

LIST OF TABLES

Contrasts used to describe the serial interaction
between perceived involvement and the source expertise

manipulation on perceived source expertise in model 1. . .117
Contrasts used to test the serial interaction between
perceived involvement and the argument strength

manipulation on perceived argument strength in model 1. . 118

Contrasts used to test the parallel interaction

between perceived involvement and the source expertise

manipulation on perceived source expertise in model 2. . .119

Contrasts used to test the parallel interaction

between perceived involvement and the argument strength

manipulation on perceived argument strength in model 2. .

Contrasts used to test the interaction between

perceived involvement. perceived source expertise. and

perceived argument strength on source cognitions. . . . .
A table of means. standard deviations. and number of

respondents for each of the four perceived involvement
items (factor 501). . . . .
Test of internal consistency for the perceived

involvement measure (factor 501). Correlations

between items in the factor appear below the diagonal.

communalities appear on the diagonal. and deviations

from predicted correlations generated by factor loadings

appear above the diagonal. . . . . . . . . . . . . . . . .123
A test of parallelism for the perceived involvement

measure (factor 501). Item-factor correlations for

factor 501 on all other factors are presented. . . . . . .124
A table of means. standard deviations. number of

observations. and factor loadings for each of the

four perceived argument strength items (factor 502). . . .125
Test of internal consistency for the perceived

argument strength measure (factor 502). Correlations

between items in the factor appear below the diagonal.

communalities appear on the diagonal. and deviations

from predicted correlations generated by factor loadings

appear above the diagonal. . . . . . . . . . . . . . . . .126
A test of parallelism for the perceived argument

strength measure (factor 502). Item-factor correlations

for factor 502 on all other factors are presented. . . . .127
A table of means. standard deviations. number of

observations. and factor loadings for each of the

eight perceived source expertise items (factor 503). . . .128

120

121

.122

vii

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

8b. Test of internal consistency for the perceived
source expertise measure (factor 503). Correlations
between items in the factor appear below the diagonal.
communalities appear on the diagonal. and deviations
from predicted correlations generated by factor
loadings appear above the diagonal. . . . . . . . . . .
8c. A test of parallelism for the perceived source
expertise measure (factor 503). Item-factor
correlations for factor 503 on all other factors
are presented. . . . . . . . . . . . . . . . . . . .
9a. A table of means. standard deviations. number of
observations. and factor loadings for each of the five
pro-message attitude items (factor 504). . . . . . . .
9b. Test of internal consistency for the pre-message
attitude measure (factor 504). Correlations between
items in the factor appear below the diagonal.
communalities appear on the diagonal. and deviations
from predicted correlations generated by factor
loadings appear above the diagonal. . . . . . . . . . .
90. A test of parallelism for the pre-message attitude
measure (factor 504). Item-factor correlations for
factor 504 on all other factors are presented. . . . .
10a. A table of means. standard deviations. number of
observations. and factor loadings for each of the
five post-message attitude items (factor 505). . . . .
10b. Test of internal consistency for the post-message
attitude measure (factor 505). Correlations between
items in the factor appear below the diagonal.
communalities appear on the diagonal. and deviations
from predicted correlations generated by factor
loadings appear above the diagonal. . . . . . . . . . .
10c. A test of parallelism for the post-message attitude
measure (factor 505). Item-factor correlations
for factor 505 on all other factors are presented. . .
lla. A table of means. standard deviations. number of
observations. and factor lodings for each of the
three behavioral intention items (factor 506). . . . .
llb. Test of internal consistency for the behavioral
intention measure (factor 506). Correlations between
items in the factor appear below the diagonal.
communalities appear on the diagonal. and deviations
from predicted correlations generated by factor
loadings appear above the diagonal. . . . . . . . . . .
11c. A test of parallelism for the behavioral intention
measure (factor 506). Item-factor correlations for
factor 506 on all other factors are presented. . . . .
12a. A table of means. standard deviations. number of
observations. and factor loadings for each of the
six (reduced) need for cognition items (factor 507). .

viii

129

.130

.131

132

.133

.134

135

.136

.137

138

.139

.140

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

12b. Test of internal consistency for the (reduced)
need for cognition measure (factor 507). Correlations
between items in the factor appear below the diagonal.
communalities appear on the diagonal. and deviations
from predicted correlations generated by factor
loadings appear above the diagonal. . . . . . . . . . .
A test of parallelism for the (reduced) need for
cognition measure (factor 507). Item-factor
correlations for factor 507 on all other factors are
presented. . . . . . . . . . . . . . . . . . . . . . . . .142
A table of means. standard deviations. number of
observations. and factor loadings for each of the
five avoidance subscale items of the argumentativeness
measure (factor 508).. . . . . . . . . . . . . . . . . . .143
Test of internal consistency for the argument
avoidance measure (factor 508). Correlations between
items in the factor appear below the diagonal.
communalities appear on the diagonal. and deviations
from predicted correlations generated by factor
loadings appear above the diagonal. . . . . . . . . . . .
A test of parallelism for the argument avoidance
measure (factor 508). Item-factor correlations for
factor 508 on all other factors are presented. . . . . .
A table of means. standard deviations. number of
observations. and factor loadings for each of the
three enjoyment subscale items of the
argumentativeness measure (factor 509). . . .
Test of internal consistency for the argument
enjoyment measure (factor 509). Correlations between
items in the factor appear below the diagonal.
communalities appear on the diagonal. and deviations
from predicted correlations generated by factor
loadings appear above the diagonal. . . . . . . . . . . .
A test of parallelism for the argument enjoyment
measure (factor 509). Item-factor correlations for
factor 509 on all other factors are presented. . . . . .
The impact of manipulated argument strength.
manipulated involvement. and manipulated source
expertise on perceived argument strength. . . . . . . . .
The impact of manipulated involvement. manipulated
source expertise. and manipulated argument strength
on perceived involvement. . . . . . . . . . . . . . . . .
The impact of manipulated source expertise.
manipulated involvement. and manipulated argument
strength on perceived source expertise. . . . . . . . . .
The impact of manipulated involvement. manipulated
source expertise. and manipulated argument strength
on post-message attitudes. . . . . . . . . . . . . . . .
The impact of perceived involvement and manipulated
source expertise on perceived source expertise. . . . . .

141
12c.

13a.

13b.

144
13c.

.145
14a.

146

14b.

147
14c.

.148
15.

149
16.

150
17.

151
18.

.152

19.
153

ix

Table
Table
Table

Table

Table

Table

Table

Table

Table

Table

Table
Table
Table
Table
Table
Table
Table
Table

Table

20.
21.
22.

23.

24.

25.

26.

27.

28.

29.

30.
31.
32.
33.
34.
35.
36.
37.

38.

The impact of perceived involvement and manipulated

argument strength on perceived argument strength. . . .
Impact of perceived involvement and perceived

argument strength on message processing. . . . . . . . .
Impact of perceived involvement and perceived

source expertise on source cognitions. . . . . . . . .
The impact of perceived involvement. perceived

source expertise. and perceived argument strength

on message cognitions. . .
The impact of perceived involvement. perceived

source expertise. and perceived argument strength

on source cognitions.
Correlations between variables (below the diagonal)

and deviations predicted coefficients (above the

diagonal) for Model 6 (decimal points have been

omitted). . . . . . . . . . . . . . . . . . . . . . . . .
The impact of the interaction between perceived
involvement and perceived argument strength on

behavioral intention. . . . . . . . .
The impact of the interaction between perceived
involvement and post-message attitudes on behavioral
intention. . . . . . . . . . . . . . . . . . . . . . . .

Contrasts used to test the interaction between

perceived involvement and post-message attitudes on

behavioral intentions. . . . . . . . . . . . . . . . . . .162

Correlations between variables (above the diagonal)

and deviations from predicted correlations (below

the diagonal) for Model number 7 (decimals have been

.157

158

159
160

.161

omitted). . . . . . . . . . . . . . . . . . . . . . . . . 163
The impact of need for cognition and perceived

argument strength on attitudes. . . . . . . .. . . . . . .164
The impact of need for cognition and perceived

source expertise on attitudes. . . . . . . . . . . . . . .165
The impact of argument enjoyment and perceived

argument strength on attitudes. . . . . . . . . . . . . . 166
The impact of argument enjoyment and perceived

source expertise on attitudes. . . . . . . . . . . . . . .167
The impact of argument avoidance and perceived

argument strength on attitudes. . . . . . . . . . . . . . 168
The impact of argument avoidance and perceived

source expertise on attitudes. . . . . . . . . . . . . . .169
The impact of need for cognition and the argument

strength manipulation on perceived argument strength. . . 170
The impact of argument enjoyment and the argument

strength manipulation on perceived argument strength. . . 171
The impact of argument avoidance and the argument

strength manipulation on perceived argument strength. . . 172

CHAPTER 1
THEORETICAL RATIONALE

Few areas in the social sciences emphasize cognitive processes
more than persuasion (Eagly 8 Chaiken. 1984). and as a result. several
cognitive approaches to the study of persuasion have been developed.
The most prolific has been what Eagly and Chaiken (1984) refer to as
the cognitive response approach. The cognitive response approach.
fathered by Hovland. Janis. and Kelley (1953) and McGuire (1972).
”asserts that the persuasive impact of a communication is primarily
determined by the nature of the idiosyncratic thoughts or 'cognitive
responses' that participants presumably generate . . . as they
anticipate. receive. or subsequently reflect upon a persuasive
communication” (Eagly 8 Chaiken. 1984. p. 280).

The purpose of this study is to discuss and to test one cognitive
response model. Petty and Cacioppo's (1981. 1986) Elaboration
Likelihood Model (ELM). Like other cognitive response models. the ELM
assumes that the nature of cognitive responses generated within message
recipients determines the extent of attitude and behavior change. The
ELM's predictions are based upon the cognitive processes recipients
experience when responding to a persuasive communication.
Specifically. the ELM asserts that active cognitive processing of
massage information is not a necessary condition for attitude and
behavior change.

Although considerable research has supported the ELM's‘
predictions. the causal processes at the core of the ELM have never
been adequately specified nor tested. Therefore. this study has two

1

major goals. The first goal is to discuss the ELM with special
attention paid to the causal processes posited by it. This goal will
be realized by discussing two alternative perspectives of the ELM.
generating causal models from each perspective. and reviewing the
relevant literature for each perspective. The second goal is to
generate data that test the causal models. The ultimate goal of this
study is a more clearly delineated conceptualization of how the
variables within the ELM act to affect attitudes and behaviors.

T E re n Likelihood Model of Persuasion

The ELM assumes that people want to hold correct attitudes (Petty I
8 Cacioppo. 1986; c.f. Festinger. 1950). Rhen receiving a persuasive
message. then. persons will process that message in such a way as to
maximise the correctness of their existing attitude or to establish a
new. correct. attitude. The ELM asserts. moreover. that there are two
ways of arriving at a ”correct” attitude.

The principal construct in the ELM is message elaboration. Petty
and Cacioppo (1986) define message elaboration as "the extent to which
a person scrutinixes the issue-related arguments contained in the
persuasive communication” (p. 7). Elaboration refers not only to
simple cognitive message repetition. but also to information from
existing schemes or knowledge structures that recipients add to the
message.

Petty and Cacioppo (1986) argue that the degree of issue-relevant
elaboration performed in response to a persuasive communication can be
placed on a continuum: The amount of ggggggg_g1gbgggtign can range

from vigilant scrutiny of to mindlessly attention to a message.

Moreover. several authors have suggested that existing theories of
attitude change can be placed on such a continuum (Chaiken 8 Stangor.
1987; Petty 8 Cacioppo.,1986). For example. several theories presume
that persons exert considerable cognitive effort in evaluating a
persuasive message. Theories falling at this mindful end of the
continuum include inoculation (McGuire. 1972) and information

integration (Anderson. 1974). Similarly. several theories presume that

 

persons do not vigorously evaluate message content. Such ”mindless”
theories include self-presentation (Ram. 1972) and social judgment

(Sheri! 6 Sherif. 1967).

Petty and Cacioppo argue that all attitude change processes can be "P

t
.‘

encompassed by two distinct routes to persuasion. As such. Petty and
Cacioppo claim that the elaboration continuum can be adequately
represented by looking only at its extremes. Thus. despite the
potential utility of the elaboration continuum Petty and Cacioppo
(1986) abandon its center to concentrate on its extremes. In their
conceptualization. the peripheral route to persuasion represents the
end of the continuum corresponding to a lack of cognitive effort in
evaluating message content. and the central route to persuasion
represents the and of the continuum characterized by a great deal of
cognitive effort. Given the assumptions of dual persuasion processes.
what occurs at the moderate portion of the continuum (i.e.. the area
between the extremes) is unclear and is apparently of little

theoretical importance.

The central route to persuasion.

When using the central route to persuasion persons exert
considerable cognitive effort in evaluating persuasive messages.
According to the ELM. this cognitive effort takes the form of message
elaboration. As noted above. message elaboration involves accessing
attitude-object schema(s) to evaluate message arguments. i.e..
recipients recall issue-relevant information from memory in order to
evaluate carefully the message and arguments presented.

To perform elevated levels of message elaboration. recipients must
be both motivated to process and able to elaborate upon the message.
Mgtivagion to elaborate refers to the extent to which recipients want
to evaluate the message and component arguments:on the other hand.
gbilitz to elaborate refers to the extent to which audience members
possess the knowledge and skills required to evaluate message arguments
(Chaiken. 1980; Chaiken 8 Stangor. 1987: Petty 8 Cacioppo. 1986).

When persons elaborate on message content. attitude change is
predicted to be a function of the direction of elaboration: Thoughts
supporting message recommendations enhance persuasion while thoughts
opposing message recommendations (i.e.. counterarguments) inhibit
persuasion (Eagly 6 Chaiken. 1984). Petty and Cacioppo (1981. 1986)
argue that the direction of elaboration is determined in large part by
the strength of message arguments. Strong message arguments generate
supportive thoughts and enhance persuasion. but weak message arguments

generate counterarguments and inhibit persuasion.

The pggipheralirogtggto pggsuasion.
According to the ELM. when either (or both) ability or motivation

 

to elaborate are low. persons will use the peripheral route to
persuasion. According to Chaiken and Stangor (1987) ”the peripheral
route refers to a family of attitude change theories that specify
factors or motives that produce attitude change without engendering
active message and issue-relevant thinking” ((p. 593). These
peripheral processes include a number of cognitive processes. such as
cognitive heuristics1 (Chaiken. 1980. 1987: Chaiken 6 Stangor. 1987).
attributions (Ram. 1972). or affective mechanisms in attitude change
such as classical or operant conditioning (Staats 8 Staats. 1958; see
Chaiken 8 Stangor. 1987).

In peripheral processing. message recipients exert considerably
less cognitive effort evaluating the persuasive message. Put another
way. when taking the peripheral route to persuasion recipients will
expand less cognitive capacity processing the message than when taking
the central route. When taking this peripheral route. recipients are
predicted to concentrate primarily on characteristics of the source
“(e.g.. credibility and/or attractiveness) and/or of the persuasive
situation (e.g.. presence of pleasant music or audience response).
When processing peripherally. the attitude schema is accessed only
‘insofar as it incorporates the inference or affect invoked by the
peripheral cue (Petty 8 Cacioppo. 1986).

Attitude change generated through the peripheral route is directed
by attributions made regarding source or message characteristics upon

which heuristics or associations are based. For example. Chaiken and

Stangor (1987) assert that while taking the peripheral route. persons I\
might use a decision rule such as "expert’s statements can be trusted” *5
(p. 599). If recipients are using such a decision rule. a source
perceived as expert should increase persuasion while a source perceived
as inexpert should inhibit persuasion. Similarly. if recipients are
using the peripheral route. an attractive model (by a positive
association with the persuasive message) should increase persuasion
while an unattractive model should inhibit persuasion. In both cases.
attitude change occurs without active consideration of message
arguments.

The difference between the central and peripheral routes to
persuasion ”has to do with the extent to which the attitude change that
results from a message is due to active thinking about either the issue
or object-related information provided by the message” (Petty 8
Cacioppo. 1981. p. 256). When travelling the central route. message
recipients exert considerable cognitive effort elaborating upon the
message. but peripheral route recipients exert comparatively less
effort.

Not surprisingly. considerable research has sought to identify
those characteristics that affect recipient motivation and/or ability
to elaborate message arguments. Variables that have been identified as
influencing gb111;1_include message repetition (Cacioppo 8 Petty. 1979;
McCullough 8 Ostrom. 1974). recipient posture (Petty. Wells. Heesacker.
Brock. 6 Cacioppo. 1983). message complexity or comprehensibility
(Eagly. 1974: Regan & Chang. 1973; Yalch 8 Elmore-Yalch. 1984). message

modality (Chaiken 6 Eagly. 1976: Wright. 1981). recipient heart rate

(Cacioppo. 1979). and distraction (Allyn 8 Festinger. 1961; Petty.
Wells. 8 Brock. 1976). Variables that have been identified as
influencing motivation include issue involvement or personal relevance
(Miller. 1965: Petty 8 Cacioppo. 1979a.b; Sherif 6 Hovland. 1961)
personal responsibility (Markins 8 Petty. 1982; Petty. Markins.
Williams. 6 Latane. 1977). multiple sources (Markins 8 Petty. 1981).
need for cognition (Cacioppo. Petty & Morris. 1983). audience reactions
(Axsom. Yates. 8 Chaiken. 1987). and the presence of pleasant music
(Park 8 Young. 1986).

The ELM posits that when motivation and ability to elaborate are
high (for example. when involvement is high) recipients will take the
central route to persuasion. In such cases. message cues. such as
argument strength. will exert a persuasive impact. When recipients are
unable and/or unmotivated to elaborate (for example. at low levels of
involvement) the ELM posits that persons will take the peripheral route

to persuasion. In this case. cues such as source expertise will be 5%

persuasive. j/

As noted earlier. however. relatively little research or

hypothesizing has been performed on the impacts of message and

persuasion cues at ggdgrate levels of involvement (i.e.. at the centra1<ijfl
portion of the elaboration continuum).2 The absence of specific
predictions at moderate level of involvement stems from Petty and
Cacioppo's interests in the extremes of the elaboration continuum.
Recently (Petty 6 Cacioppo. 1986). however. they have advanced a
hypothesis regarding the impact of central. peripheral. and moderator

cues on attitude change at moderate involvement. Specifically. persons

are predicted to use source information as an indicator of whether *9 <1 i
message content is worth processing. When the source is perceived as
expert. recipients decide it is worth the cognitive effort to evaluate
message arguments. Attitude change. in this case. is a function of
message strength (as discussed above). When the source is not
perceived as expert. on the other hand. no elaboration will occur. and
attitudes will be based simply on the negative source attributions.
According to Petty and Cacioppo. then. the source must be perceived as
credible gng_arguments judged as strong to generate attitude change in
the direction of the message recommendations at moderate levels of
involvement. \,J
Cr t c s o 1 be 0 Mo e

One important and controversial issue is whether the ELM assumes
humans are serial or parallel information processors (Petty. Cacioppo. ’/
Kasmer. 8 Maugtvedt. 1987: Petty. Kasmer. Maugtvedt. 8 Cacioppo. 1987;
Stiff. 1986: Stiff and Boster. 1987). Stiff (1986) argues that the ELM
depicts message recipients as being serial processors. who are ”forced
to decide which set of cues they will focus on processing” (p. 76). In
other words. Stiff (1986) argues that the ELM depicts recipients as
having limited processing capacity. and paying attention to either
central 9;,peripheral cues. but not both. Petty. Kasmer. Maugtvedt.
and Cacioppo (1987) respond to Stiff‘s comments by equivocating that
”there is nothing about the ELM that pggglgggg processing of both
arguments and peripheral cues” (p. 247; emphasis in original).
Moreover. Petty. Easmer. Maugtvedt. and Cacioppo (1987) allege that

their discussion of the ELM ”does not mean that people are incapable of

processing bgth_arguments and cues”(p. 238; emphasis in original).
Petty. Kasmer. Maugtvedt. and Cacioppo (1987) never claim that the ELM
i;_a parallel processing model. rather they argue that it is apt a
serial processing model.

Obviously. Petty and Cacioppo are indecisive as to whether the ELM
represents a parallel or serial information processing model. In their
recent article (Petty. Kasmer. Maugtvedt. and Cacioppo. 1987) they
admit that ”Although we have endorsed the idea that in forming opinions
people may process both source and message information. we have not
addressed the issue of whether this processing of source and message
information occurs sequentially or in parallel" (p. 247). Although
Petty. Kasmer. Haugtvedt. and Cacioppo (1987) advocate a parallel view
of the ELM. there is considerable evidence that previous discussions of
the model suggest a serial mode of information processing.

First. in the flowchart representing the cognitive processes
performed by recipients. Petty and Cacioppo clearly depict processing
of githg;_centra1 cues g; peripheral cues but not both simultaneously
(Cacioppo 6 Petty. 1987. p. 42: Petty 6 Cacioppo. 1981. p. 264: 1987.
p. 4; Petty. Kasmer. Haugtvedt. and Cacioppo 1987. p. 235: for
additional discussion of this point. see Stiff 8 Boster. 1987).
Although this representation has changed over the ensuing years. one
characterisation has remained constant. That is. all of Petty and
Cacioppo's schematic depictions of the ELM pggglggg_parallel processing
of both source and message information.

Second. additional evidence of the serial processing assumptions

of the ELM comes when Petty and Cacioppo (1986) attempt to distinguish

10

it from the dissociative cue hypothesis as competing explanations of
attitudinal persistence. Specifically. the dissociative cue
hypothesis. according to Petty and Cacioppo ”presumes that in a typical
persuasion situation. attitudes are affected jointly (additively) by
salient cues and message processing” (1986. p. 181). The ELM in the
same persuasion situations predicts that "attitudes may be affected
mostly by cues or by message processing (elaboration)” (p. 181). Thus.
according to the dissociative cue hypothesis. attitudes persist because
of the gg;;_processing of both central and peripheral cues. while the
ELM posits that attitudinal persistence stems from the sepgggtg
processing of source or persuasion cues. Obviously. then. the
difference between the ELM and the dissociative cue hypothesis is the
difference between serial and parallel information processing models.
Finally. Petty and Cacioppo's predictions relating the effect of
central and peripheral cues on attitude change suggest serial
information processing. As discussed above. when involvement is low
the ELM predicts that attitude change will result from peripheral cues
such as source expertise. Central cues. such as argument strength.
have no impact at low levels of involvement. At high levels of
involvement. Petty and Cacioppo suggest that central cues will have an
impact on attitude change while peripheral cues will have little
impact. Moderate levels of involvement may bring about processing of
both central and peripheral cues. but not in parallel with one
another. Specifically. Petty and Cacioppo posit when moderate levels
of involvement are generated recipients will question whether it is

worth the cognitive effort to evaluate the message. Therefore.

11

recipients will evaluate message content only if the source is
perceived as credible. Thus. it is clear that the processing of source
and message information occurs serially rather than in parallel.

Pre ictions erated f as the M.

This serial versus parallel dispute has important implications for
the ELM's predictions. The principal proposition of the ELM delineates
the predicted impacts of central. peripheral. and moderator variables
on immediate attitude change. Specifically. Petty and Cacioppo predict
that ”As motivation and/or ability to process arguments is decreased.
peripheral cues become relatively more important determinants of
persuasion. Conversely. as argument scrutiny is increased. peripheral
cues become relatively less important determinants of persuasion”
(1986. p. 20).

Petty. Cacioppo. Kasmer. and Maugtvedt (1987) clarify this
prediction by asserting that central cues. peripheral cues and the
variable used to mediate their impact (e.g. involvement) will interact
to affect attitude and behavior change. This proposition yields
different predictions and different interactions depending on whether
one interprets the ELM as a serial or parallel information processing
model. As discussed above. Stiff (1986) interprets the ELM as a serial
processing model where message recipients are forced to choose between
processing central and peripheral cues. Therefore. Stiff interprets
the ELM to predict that as involvement increases recipients will be
more likely to process central cues and lg§§_likely to process
peripheral cues (see Figure l). The ELM predicts. then. that as

involvement increases the impact of central cues (e.g.. argument

12

strength) steadily increases. Similarly. as involvement increases the
impact of peripheral cues steadily decreases. Such a pattern is
suggested by comments made by Petty and Cacioppo (1986) regarding the
ELM. In particular. Petty and Cacioppo assert that "In short. we
postulate a trade-off between argument elaboration and the operation of

peripheral cues” (1986. p. 21).

 

Figure 1 about here

 

Stiff argues that the ELM. as a limited capacity serial
information processing model. is an inaccurate representation of
recipients' processing of persuasive messages. Since considerable
research has clearly identified humans as parallel information
processors (e.g.. Lindsay. 1970: Lindsay. Cudding. 8 Tulving. 1965;
Treisman. 1970; Treisman 8 Fearnley. 1971; Treisman 6 Geffen. 1967;
Treisman 6 Riley. 1969: Tulving 8 Lindsay. 1967). Stiff (1986) contends
that Kahneman's (1973) Elastic Capacity Model (ECM) more accurately
represents recipients' processing of persuasive messages. Kahneman
(1973) states that the amount of cognitive effort gxgrtgg on a task is
a function of the amount of cognitive effort dgggnggg_by that task (see
Figure 2). Moreover. the model presumes that persons can perform two
tasks simultaneously (labelled the primary and secondary tasks). At
low levels of cognitive demand. a good deal of processing capacity is
available. but since motivation is low. little processing occurs. At

moderate levels of cognitive demand. available cognitive capacity

 

13

increases to such an extent that bpth_primary and secondary tasks can
be processed. Finally. at high levels of cognitive demand. the level
of available processing capacity asymptotes. As a response to the
restricted level of processing capacity. recipients will focus
attention almost exclusively on the primary task. Thus. to adjust to
this restricted level of processing capacity. a vast majority of the

processing capacity is focused to the primary task.

 

Figure 2 about here

 

The application of Kahneman‘s ECM to persuasion provides
alternative predictions of the relationship between central and
peripheral cues and attitude change (assuming the primary and secondary
tasks are synonymous with processing central and peripheral cues; see
Stiff. 1986). Moreover. according to Stiff. the processing of a set of
_ persuasive cues implies that those cues have persuasive impact. At low
levels of involvement. since neither central nor peripheral cues will
be processed. neither will influence attitude change (see Figure 3).

At moderate levels of involvement. both central and peripheral cues
will be processed. thus. both sets of cues will have an impact upon
attitude change. Finally. at high levels of involvement. since central
‘ cues will be processed predominantly. they are the only cues likely to
have a significant impact upon attitude change. Moreover. the impact
of central cues on attitude change should asymptote. Specifically. the
impact of central cues on attitude change should remain relatively

constant between moderate and high levels of involvement.

14

 

Figure 3 about here

 

Note that the predictions depicted in Figures 1 and 3 are very
similar in the moderate to high ranges of involvement. In these higher
ranges of involvement both models predict that increasing involvement
attenuates the impact of peripheral cues. The models differ slightly
in their predictions regarding the impact of increasing involvement an
argument strength-attitude effect sizes. The serial processing model
predicts that increasing involvement from moderate to high should
increase the impact of argument strength while Stiff's interpretation
of the ECM asserts that the impact of argument should not change.

It is in the moderate to low ranges of involvement that the two
interpretations differ drastically. Specifically. the serial
processing model predicts that as involvement increases from low to
moderate the impact of peripheral cues will be attenuated while the
impact of central cues will increase (see Figure 1). whereas the serial
processing model posits that the impact of both central and peripheral
cues will grow stronger as involvement increases from low to moderate
(see Figure 3).

Although Petty and Cacioppo (1986) have collected an impressive
collection of results to support their view of the ELM. few of these
data are relevant to the current question of serial versus parallel
processing. There are several reasons for making the claim that
previous ELM research is inadequate distinguishing between the ELM and

the ECM. First. the hypotheses depicted in Figure 3 predict that as

15

involvement increases from low to high the impact of source expertise
on attitudes will be curvilinear. Specifically. the impact of source
expertise on attitude change will be an inverted U-shaped function.
Thus. to differentiate between the serial and parallel positions. a
minimum of three levels of issue involvement must be created.
Therefore. previous work on the ELM (e.g.. Cialdini. Levy. Herman.
Kozlowski 8 Petty. 1976; Cialdini 6 Petty. 1981; Petty 8 Cacioppo. 1979
a. b; Petty 8 Cacioppo. 1984; Petty. Cacioppo. 8 Goldman. 1981; Fatty.
Cacioppo. 8 Beesacker. 1981; Petty. Cacioppo. 8 Schuman. 1983) cannot
distinguish between these hypotheses since their research employs only
two levels of involvement. Moreover. it is unclear where on the
elaboration continuum a particular set of respondents fell.

Second. it would be difficult to determine. due to measurement
limitations. at what levels of involvement these studies fall. Many
studies employ single item measures of perceived issue involvement
(e.g.. Cacioppo et al.. 1982; Fatty 6 Cacioppo. 1984; Petty. Cacioppo.
8 Goldman. 1981). Moreover. this item typically consists of a question
relating to the manipulation rather than the perceived level of
interest in the message generated by the manipulation.

A study by Stiff (1986) presented data pertinent to testing the
validity of the models presented in Figures 1 and 3. Stiff (1986) meta-
analyxed studies that investigated the impact of supporting evidence in
a persuasive message on attitude change. Several of those studies also
investigated the impact of source expertise on attitude change. Stiff
divided studies in the meta-analysis into three levels of issue

involvement. The placement of studies into involvement categories was

16

performed by asking five expert raters to estimate "the extent to which
the participants in the study (all college students) were involved with
the topic of the message" (Stiff. 1986. p. 83). This procedure is
adequate to distinguish between the models presented in Figures 1 and 3
if one accepts the assumption that supporting information is a centralw“3v
cue. source credibility is a peripheral cue. and the variable that
moderates their influence is involvement.

Results of the Stiff meta-analysis indicate that as involvement
increases. evidence has an increasingly strong impact upon attitude
change. These results are consistent with the predictions of both the
ECM and the ELM. Moreover. in a more detailed analysis of these meta-l
analytic data. Stiff and Boater (1987) found that the type of evidence
used had no impact on persuasion; i.e.. whether evidence was
operationalixed as the number of arguments or the strength of arguments
had no impact on the size of the evidence-attitude correlation.

The results for source credibility. on the other hand. suggest a
curvilinear relationship. At both low and high levels of involvement.
the impact of source credibility on attitudes was small (gfs=.12 and
.05 respectively). At moderate involvement. on the other hand.
credibility exerted a stronger impact on attitude change (;;.24). In
short. the impact of source credibility was greater at moderate
involvement than it was at either low or high involvement.

Stiff and Easter (1987) present additional data consistent with
the inverted-U shaped function. They performed a one-way analysis of
variance with involvement as the independent variable and the

credibility-attitude change effect sixes as the dependent variable.

_ -m‘m fm'ubfi

\

17

The quadratic trend of involvement on effect size was quite substantial
(5;.75).

Of the two models presented in Figures 1 and 3. the results of
Stiff's meta-analysis are most consistent with the predictions
generated by the parallel interpretation. generated from Kahneman's
(1973) ECM. As noted above. Stiff (1986) proposed this model as an
alternative to the ELM. Therefore. these results are consistent with
the predictions suggested by parallel interpretation generated from
Kahneman's ECM and are inconsistgnt with the pre4dictions generated from
the ELM.

Limi ati s of vi us a

The primary limitation of Stiff's meta-analysis lies in the post-
hoc nature of the involvement ratings (see. Petty. Kasmer. Haugtvedt.
and Cacioppo. 1987). Without a clearer indication and measurement of
the levels of involvement. the question of the adequacy of the ELM is
likely to remain unanswered. Therefore. this study sought to replicate
the results of Stiff's (1986) meta-analysis in a single study.

An additional limitation of the cognitive response literature

- (including all of the ELM studies as well as Stiff's meta-analysis) is

the use of analysis of variance (ANOVA) as the primary. and in most
cases. exclusive analytical tool. The ANOVA.procedure is inappropriate
as the sole analytical tool in evaluating the ELM for several reasons.
some of which are generic to ANOVA while others of which relate
specifically to the use of ANOVA to test the ELM.

One assumption implicit in,the interpretation of ANOVA.is that the‘

\independent variable is manipulated perfectly (Hunter. 1982); i.e..

18

ANOVA presumes that all respondents in a particular condition react to
that manipulation in precisely the same way. Moreover. the respondents )
in one condition are assumed to differ from those in other conditions l
in uniform degrees. i
The assumption of perfect manipulations is frequently violated in '
persuasion research. That is. manipulations of independent variables '
in persuasion have been found to be far from perfect. Easter and
Mongeau (1984) in meta-analysing the fear-appeal literature. found that-
the correlation between manipulated fear and perceived fear was .36
across 40 studies. The extent of this problem is apparent in the
manipulation check data provided in ELM studies. For example. Petty.
Cacioppo. and Goldman (1981) report a a relationship between their
involvement manipulation and perceived involvement of approximately
.19. The source expertise manipulation check correlation (g;.18) is
equally unimpressive. while the argument strength manipulation check
correlation (r=.51). although stronger. is far from 1.00. Petty and
Cacioppo (1979) report a correlation between the involvement j
manipulation and perceived involvement of .42.3 These manipulation I
check correlations are a far cry from the 1.0 assumed by ANOVA. Put
another way. it is likely that many studies investigating the nature of
the ELM have violated the assumptions of ANOVA regarding the l
manipulations of the independent variable. i
It is clear. then. that the manipulation of independent variables
in ELM studies (as well as in persuasion studies in general) is not

strong. Since manipulations are imperfect. the importance of good

measurement of manipulation checks becomes paramount. When the

19

independent variable is manipulated perfectly. one need only to worry
about manipulation checks in order to test for confounds. The less
reliable the manipulation. the more important measurement of
manipulation checks becomes. Second. since it is likely that variables
are not manipulated perfectly. it is important to look not only at the
manipulated level of a variable but also at respondents' perceived
level of the variable.

A second problem with the use of ANOVA is lack of power.
Specifically. the ANOVA has low power to detect either linear or
nonlinear patterns of differences. More important. the ANOVA
concentrates exclusively on the sire of mean differences across
conditions and ignores whether a set of means is linearly or
curvilinearly arranged. Thus. the omnibus F-test alone is particularly
inappropriate for distinguishing between the models presented by Stiff
(1986) and discussed above.

Given the assumptions and nature of the technique. ANOVA is
strongly limited as the g91g,analytical tool for analyzing data
relevant to the ELM. In addition. however. there are reasons specific
to the ELM that make ANOVA inappropriate as a sole analytical tool.
Specifically. the use of ANOVA overemphasizes interest in the ggtggggg_
of cognitive processes (generally attitude change) and de-emphasizes
the nggggg s leading to these outcomes.

The ELM is one of several cognitive response models of
persuasion. These models. as discussed above. presume that a \
persuasive message produces a series of cognitive responses and that l

these responses. in turn. bring about attitude and behavior change. 1

_fi.... -”.—...__ ......__,__ .HAV

20

Therefore. cognitive responses play a key role in mediating the (
relationship between manipulated. independent. variables and attitude
and behavior change. At times. data generated from cognitive response

measures (typically the thought-listing technique) are not reported at

I :‘ZT’Dh—umfi— __ .,

all (e.g.. Petty. Cacioppo. 6 Goldman. 1981). When results of
cognitive response measures are reported they are most typically
discussed in terms of a check on the involvement manipulation. In

addition. little data linking cognitive responses with attitudes are

.‘ V—"-'-‘-W---"\- N
\ n”-- H-..”

reported. At best. within cell correlations between cognitive

responses and attitudes are reported (e.g.. Petty 8 Cacioppo. 1979;

Petty. Cacioppo. 8 Meesacker. 1981). In short. the use of ANOVA has
caused people to overlook the important causal sequences cognitive ;
responses play in the persuasion processes.
Finally. it is likely that the perceived level of a manipulated 5
variable (and not the manipulation itself) influences cognitive I
responses. For example. it may be that pgrceived argument strength is {
a better predictor of message cognitions than the argument strength l W
manipulation. Studies relying solely on ANOVA as a method of analysis l 5?}
have inadequately analyzed the cognitive processes occurring within the
ELM. E
Failure to analyze fully the role of cognitive response variables
is important for several reasons. Data analyzed with ANOVA conceivably ;
can generate results on outcome variables that are consistent with the I

ELM. These same data. on the other hand. if analyzed by means more

_.
- 4"
-.a- t
4"
a.‘

consistent with the theoretical nature of these processes. could

indicate that the intervening cognitive processes are clearly

, --.. .w—f

21

inconsistent with the ELM. Second. and more important. one way to
distinguish between alternative interpretations of the EL! provided by
Stiff (1986; i.e.. serial vs. parallel processing) is through the
measurement and full analysis of cognitive response data. The EL! and
the EC! make alternative predictions regarding the nature of source and
message processes. alternatives that can be distinguished if
investigated. Finally. the excessive interest in outcome variables has
left several unanswered questions regarding cognitive processes.
Specifically. Petty and Cacioppo (Petty. Kaamer. Haugtvedt. 8 Cacioppo.
1987) now predict a three-way interaction between central. peripheral.
and the variables expected to mediate their impacts on attitudes. One
unanswered question is whether the process generating this interaction
occurs at the stage of perceiving message content or at the stage of
processing message content; i.e.. does the level of involvement in the
message influence the perceptions of message variables or does
involvement impact upon the actual processing of that information?
AlthOugh the ANOVA has several limitations. rarely (if ever) has ’
the technique been used to its fullest potential to test the adequacy l
of the 8L6. Specifically. ANOVA. through the use of trend analysis.
can be used to test specific predicted relationships between variables ,
(Ferguson. 1982) with higher power. As such. ANOVA has an important
role to play in determining the adequacy of the ELM. In this study.
several competing models will be specified. each suggesting a different
interaction term. In evaluating the adequacy of these competing 3
models. ANOVA will be used to determine the relative adequacy of them.

Eyen though trend analyses can be used to test specific

22

relationships between variables. it is also necessary to evaluate the
ELK using a technique that can simultaneously evaluate the impact of
;L_ variables within the model. Given the causal nature of the ELH's
predictions. path analysis provides such an analytical tool. Pedhazur
(1982) refers to path analysis as ”a method for studying patterns of
causation among a set of variables” (p. 577). Path analysis is
designed to determine the extent to which data are consistent with a
set of causal predictions. As such. path analysis avoids the
limitations of ANOVA discussed above.

First. path analysis makes no assumptions regarding the quality of
the experimental manipulations. nor the quality of the manipulation
check measures. Second. since path analysis tests specific causal
models. the lack of power inherent in the omnibus F-test is avoided.
That is. ANOVA.does not test specific relationships while path analysis
tests specific sets of causal predictions. Finally. since path
analysis compares data to an entire causal model. it gives the
researcher a degree of flexibility in testing relationships between
variables. The second goal of this study. then. is to analyze data
path analytically in order to investigate the interrelationships of all
the variables within the FLU.

Ho “ la r t on he iho d de

Given the vague nature of the specific causal processes within the
ELM. a large number of causal models could be suggested. From this
universe of potential models. five models have been developed. These
models differ in their assumptions about the nature of recipients'

processing of message information. First. these models differ in

«ad’hnmwafim‘dflha- v s «.0
‘0‘“
‘M.

23

whether humans are assumed to be serial or parallel information
processors. Models 1 and 3 assume that humans are serial information
processors while the remaining models presume that humans are parallel
processors.

The second difference between models is whether perceived
involvement influences the perception or processing of message and
source variables. The first two models suggest that perception of
source and message variables will be influenced by recipients’ level of
perceived involvement in the message. In the case of the final three
models. source and message cggnitions are predicted to be a function.
in part. of recipients' perceptions of their involvement in the
message.4

Before discussing the models at length. four assumptions
underlying the processing of persuasive messages need be explicated.
First. source and message controls used in experimental studies are apt
predicted to have a direct influence on message and/or source
cognitions. attitudes. or behavioral intentions. Put another way. the
impact of message and source manipulations on source and message
cognitions is predicted to be mediated by the perception of those
variables. Second. perceptions of these message and source variables
are predicted to influence the valence of source and message
cognitions. Third. the valence of source and message processing is
predicted to influence post-message attitudes. Finally. post-message
attitudes are likely to be consistent. to a certain extent. with pre-

message attitudes.

24

Model 1: A serial model of perceptual processes.
The first causal model generated from the ELM (see. e.g.. Petty 8

Cacioppo. 1981; Petty. Cacioppo. 8 Goldman. 1981). asserts that
perceptions of source expertise and argument strength are. in part. a
function of recipients’ perception of involvement in the message.
Moreover. as a serial processing model. respondents are predicted to be
forced to choose whether to concentrate upon either the source
expertise cues or to the argument strength cues. Therefore. as
perceived involvement increases. respondents will become increasingly
motivated to attend to message arguments and decreasingly motivated to
attend to source expertise.

In this model (see Figure 4). perceived issue involvement is
predicted to be a linear function only of the involvement control;
i.e.. as the message becomes more personally relevant. recipients'
perceptions of involvement are predicted to increase in a linear

fashion.

 

Figure 4 about here

 

Perceived source expertise is predicted to be a function of both
perceived involvement and the source expertise control. At low levels
of perceived involvement. perceived source expertise will be a strong
function of the source expertise control. At moderate levels of
perceived involvement. perceived source expertise will be a weaker
function of the source expertise control. Finally. at high levels of

involvement. source expertise will pp; be a function of the source

25

expertise control. Thus. perception of source expertise is posited to
be consistent with an interaction between perceived involvement and the
source expertise control where recipients are most likely to
differentiate between high and low expert sources under conditions of
low involvement. Moreover. as perceived involvement increases.
perceived source expertise is predicted to be a decreasing function of
the source expertise control. The contrasts employed to describe this

interaction are presented in Table l.

 

Table 1 about here

 

Recipients' perceptions of source expertise. in turn. are
predicted to have an impact on the nature of the cognitions regarding
the source. Specifically. if the source of the message is considered
expert. thoughts regarding the source will be positive. if the source
is perceived as inexpert. source cognitions will be negative.

Perceived argument strength in this model is predicted to be a
function of both perceived involvement and the argument strength
control. At low levels of perceived involvement. recipients are not
predicted to differentiate between strong and weak arguments since they
are insufficiently motivated to process or to evaluate message
content. As perceived involvement increases. perceived argument
strength is predicted to be an increasingly strong function of the
argument strength control. At moderate involvement. then. perceived
argument strength is predicted to be a weak function of the argument

strength control. while at high levels of perceived involvement

26

perceived argument strength is predicted to be a strong function of the
argument strength control. Thus. perception of argument strength is
predicted to reflect an interaction between perceived involvement and
the argument strength control. The contrasts used to represent this

interaction are presented in Table 2.

 

Table 2 about here

 

Recipients‘ perceptions of argument strength. in turn. are
predicted to have an impact on the nature of message cognitions. i.e..
cognitions relating to the message itself or the arguments contained in
the message. If message arguments are perceived as strong. message
cognitions will be generally pro-message. If message arguments are
perceived as weak message cognitions will take the form of
counterarguments.

Both source and message processing. in turn. are predicted to
influence attitudes. The extent to which source and message processing
support the source and/or the message determines the extent to which
attitudes will be consistent with message recommendations. To the
extent that source or message cognitions are negative (e.g..
counterarguments) attitudes are likely to be opposed. or counter to.
message recommendations. Finally. attitudes are presumed to predict
behavioral intentions.

el ' ar 1 mod 1 of ce tual oce e .
An alternative interpretation of Model 1 can be developed using

Stiff's interpretation of the ECM. This model is similar to Model I in

27

that perceived involvement is again predicted to influence the
perception of source expertise and argument strength. These models
differ. however. since the perception of variables can occur
simultaneously rather than serially; recipients are not forced to
choose sets of variables to process. Thus. this model does not predict
a different causal sequence of variables. however. it does suggest an
alternative set of cognitive processes. Since Model 2 is identical to
Model 1 in all ways except for the difference between serial and
parallel perception of source and message variables. only the
differences in cognitive processes between Models 1 and 2 will be
discussed.

Employing Stiff's interpretation of the ECM. at low levels of
involvement. respondents are predicted to be unmotivated to process
githg; source or message cues. At moderate involvement. cognitive

processing capacity increases to such an extent that both source and

 

message cues may be perceived simultaneously. At high levels of
involvement. however. available cognitive capacity asymptotes. such
that only message cues will be processed.

Using the ECM as a guide. the perception of source expertise is
predicted to be a function of both perceived involvement and the source
expertise control. At low levels of perceived involvement. the ECM
asserts that persons are unmotivated to process any cues. therefore.
perceived source expertise will not vary as a function of the source
expertise control. At moderate levels of perceived involvement.

persons will be sufficiently motivated to evaluate the expertise of the

 

source; therefore. perceived source expertise will vary as a function

28

of the source control. Finally at high levels of perceived
involvement. the ECM asserts that the amount of available cognitive
processing asymptotes such that recipients will be Eggplg to evaluate
source expertise. with the bulk of processing directed toward
processing message content. Thus. at high levels of perceived
involvement. perceived source expertise will not vary as a function of
the source expertise control.

These perceptual processes are predicted to be consistent with a
perceived involvement by source expertise interaction on perceived
source expertise. Specifically. at extreme levels of perceived
involvement. perceived source expertise will not vary as a function of
the source expertise control. Only at moderate levels of perceived
involvement will the source expertise control have an impact on
perceived expertise ratings. The contrasts used to represent this

interaction are presented in Table 3.

 

Table 3 about here

 

Perceived argument strength. on the other hand. is predicted to
vary as a function of both perceived involvement and the argument
strength control. Specifically. at low levels of perceived
involvement. as noted above. recipients will be unmotivated to process
any cues relevant to the message; therefore. perceived argument
strength is not likely to vary as a function of the argument strength
control. At moderate levels of perceived involvement. recipients are

predicted to be sufficiently motivated to process both argument

29

strength and source expertise cues; hence. perceived argument strength
should vary as a function of the argument strength control. Finally.
at high perceived involvement. available processing capacity
asymptotes. Available processing capacity. according to Stiff (1986).
is predicted to be allocated to the processing of message arguments.
Consequently . the impact of the argument strength control on perceived
argument strength should be similar at high perceived involvement and
moderate perceived involvement.

Thus. information processing related to the perception of argument
strength is predicted to be consistent with an interaction between
perceived involvement and the argument strength control. The argument
strength control should have stronger impacts on perceived argument
strength at the moderate and high perceived levels of involvement
conditions than at low perceived involvement. The contrasts used to

represent this interaction are presented in Table 4.

 

Table 4 about here

 

 

Models 3. 4. and 5 will assert that the perception of source
expertise and argument strength are pg; dependent upon perception of
involvement in the message. According to the remaining models. then.
perception of source expertise is a function solely of the source
expertise control; perception of argument strength is a function solely
of the argument strength control; and perception of message involvement

is solely a function of the involvement control.5

30

Model 3 predicts that valence of message cognitions will be a
function of perceived involvement and perceived argument strength.
Valence of source cognitions. according to this model. is predicted to
be a function of perceived involvement and perceived source expertise.
Moreover. as perceived involvement increases. recipients will become
increasingly likely to process message information and decreasingly

likely to process source information (see Figure 5).

 

Figure 5 about here

 

The valence of source cognitions is predicted to be a function of
both perceived involvement and perceived source expertise. As
perceived involvement increases. recipients are decreasingly likely to
process source information; thus. valence of source cognitions is most
likely to vary as a function of perceived source expertise at low
levels of perceived involvement. Moreover. as perceived involvement
increases. the valence of source cognitions is less likely to vary as a
function of perceived source expertise. and processes leading to the
valence of source cognitions are consistent with an interaction between
perceived involvement and perceived source expertise. The contrasts
used to represent this interaction are identical to those presented in
Table 1 except that these contrasts represent the impact of perceived
involvement and perceived source expertise on the valence of source
cognitions.

The valence of message cognitions is predicted to be a function of

both perceived involvement and perceived argument strength. This

31

serial processing model predicts that at low levels of perceived
involvement. recipients will not be sufficiently motivated to process
message arguments and will instead process source information. At low
levels of perceived involvement. valence of message cognitions will not
differ as a function of perceived argument strength. As perceived
involvement increases. persons are increasingly likely to process
message information; therefore. as perceived involvement increases. the
valence of message cognitions is predicted to be an increasingly strong
function of perceived argument strength. The valence of message
cognitions is predicted to be consistent with an interaction between
perceived involvement and perceived argument strength. The contrasts
employed to characterize this interaction are identical to those
presented in Table 2. however. in this case. they represent the impact
of perceived involvement and perceived argument strength on message
cognitions.

The valence of source and message cognitions is predicted to
influence attitudes. As source and message cognitions become more
positive. attitudes are predicted to be more consistent with message
recommendations. If source and message cognitions are negative. on the
other hand. attitudes are predicted to be opposed to message
recommendations. Post-message attitudes. in turn. are predicted to
influence behavioral intentions.

del 4: A r 1 el model 0 me sa roc si .

Model 4 is identical to Model 3 except that the processing of

message and source information is predicted to occur in parallel.

rather than serially. Based on the predictions of the ECM. at low

32

levels of perceived involvement. recipients are predicted to be
insufficiently motivated to process either source or message cues. At
moderate perceived involvement. available processing capacity increases
such that recipients are able to process both source and message
information. Finally. at high levels of perceived involvement.
recipients' available processing capacity asymptotes. such that they
will concentrate exclusively on message cues.

Valence of source cognitions is posited to be a function of both
perceived involvement and perceived source expertise. At low levels of
perceived involvement. respondents are predicted to be insufficiently
motivated to process source information. Therefore. the valence of
source cognitions is not predicted to differ as a function of perceived
source expertise. At moderate levels of involvement. since available
processing capacity increases. the valence of source cognitions y;
predicted to vary as a function of perceived source expertise:
Specifically. if the source is evaluated as expert. source cognitions
are predicted to be positive. whereas if the source is evaluated as
being inexpert. source cognitions are predicted to be negative. At
high perceived involvement. the valence of source cognitions is not
expected to vary as a function of perceived source expertise since
respondents are focusing primarily on the message. These processes are
consistent with an interaction between perceived involvement and
perceived argument strength on the valence of message cognitions. The
contrasts used to represent such an interaction are the same as
presented in Table 3. although the contrasts in the present case relate

perceived involvement and perceived source expertise on source

33

cognitions.

Valence of message cognitions is predicted to be a function of
both perceived involvement and perceived argument strength. At low
levels of perceived involvement. the valence of message cognitions is
not predicted to vary as a function of perceived argument strength
since recipients are unmotivated to process such information. At
moderate levels of perceived involvement. valence of message cognitions
is predicted to vary as a function of perceived argument strength:
Specifically. if message arguments are perceived as weak. message
cognitions will be negative. while if message arguments are perceived
as being strong. message cognitions will be positive. Finally. at high
levels of perceived involvement. the impact of perceived argument
strength on message cognitions is predicted to be the same as at the
moderate level of perceived involvement. Therefore. processing of
message information is predicted be consistent with an interaction
between perceived involvement and perceived argument strength. The
contrasts used to represent such a predicted interaction are identical
to those presented in Table 4. although in the present case. the
contrasts relate perceived involvement and perceived argument strength
to message cognitions.

: -wa model of me sa e roces in .

In the most recent discussion of the ELM. Petty. Kasmer.
Haugtvedt. and Cacioppo (1987) predict that manipulations of argument
strength. involvement. and source expertise will interact to affect
attitudes. In addition. they claim that cognitions generated by

message recipients determine the direction and extent of attitude

34

change. It is difficult. however. to understand how pggg these
predictions could be an accurate representation of the processes
recipients employ in evaluating a persuasive message. First. the three
message and source manipulations interacting to directly influence
attitudes implies a lack of important information processing on the
part of recipients. The importance of cognitive processing. on the
other hand. implies that the relationship between manipulated variables
and attitude change is mediated by several cognitive variables not
explicitly specified by the ELM. It is difficult to present a model
capturing both these positions.

Can a model be developed where the three manipulated variables
directly influence attitudes and. at the same time. identify the
important role played by cognitions? If Petty. Kasmer. Maugtvedt. and
Cacioppo's predictions are identified as a shorthand description of the
processes alleged to be performed in processing message and source
information. this description generates more questions than it
answers. For example. are the message and source manipulations meant
to be synonymous with message and source cognitions? Are message and
source cognitions generated gipggtly from the manipulated variables?

De recipients accurately perceive source and message controls? Do the
manipulations and message and source processing have separate
influences? These questions are not addressed directly by the ELM.

Model 5 (see Figure 6) predicts that perception of involvement is
a direct function of the involvement control; perception of source
expertise is a direct function of the source expertise control. and

perception of argument strength is a direct function of the argument

35

strength control; i.e.. perception of source expertise and argument

strength in this model does not depend on perceptions of involvement.

 

Figure 6 about here

 

The valence of source cognitions is a function of both perceived
involvement and perceived source expertise (although the most recent
version of the ELM predicts a three-way interaction. valence of source
cognitions can only be described as a function of two variables). The
valence of message cognitions. on the other hand. is a function of
perceived involvement. perceived argument strength. and perceived
source expertise.

According to Petty. Kasmer. Haugtvedt. and Cacioppo (1987) at low
levels of perceived involvement respondents are insufficiently
motivated to process message cues but are motivated to process source
cues. At moderate perceived involvement. respondents will not be able
to decide whether it is worth the cognitive effort required to evaluate
the message; hence. message processing will occur only if the source is
evaluated as being expert. If the source is evaluated as being
inexpert. respondents will decide it is not worth the cognitive effort
to process the message. Finally. at high perceived involvement.
recipients will be motivated to process primarily message content.

The valence of source cognitions is posited to be a function of
perceived involvement and perceived source expertise. At low levels of
perceived involvement. valence of source cognitions is posited to be a

function of perceived source expertise. If the source is perceived as

36

expert. valence of source cognitions will be positive. while if the
source is perceived as inexpert. valence of source cognitions will be
negative. At moderate perceived involvement. recipients are predicted
to process source information so as to decide whether it is worth
processing message information; consequently. at moderate involvement
the valence of source cognitions is posited to be a weaker function of
perceived source expertise. when the source is perceived as expert.
valence of source cognitions will be positive. and when the source is
perceived as inexpert. the valence of source cognitions will be
negative. Finally. at high levels of perceived involvement. recipients
are predicted to be preoccupied with message factors such that valence
of source cognitions is not predicted to vary as a function of
perceived source expertise.

The valence of source cognitions is predicted to be a function of
both perceived involvement and perceived source expertise. As such.
valence of source processing is predicted to be consistent with an
interaction between perceived involvement and perceived source
expertise. The contrasts used to represent this interaction are
identical to those presented in Table 1. although the contrasts
represent the impact of perceived involvement and perceived source
expertise on source cognitions.

The valence of message cognitions is predicted to be a function of
perceived involvement. perceived source expertise. and perceived
argument strength. At low levels of perceived involvement respondents
will not be motivated to process message content; thus. the valence of

message cognitions is not predicted to differ across levels of

37

perceived argument strength or perceived source expertise. At moderate
perceived involvement. valence of message arguments is predicted to
vary as a function of perceived argument strength 9911 if the source is
perceived as expert. At moderate levels of involvement Petty. Kasmer.
Haugtvedt. and Cacioppo (1987) assert that recipients ”may examine the
persuasion context for indications (e.g.. is the source expert?) as to
whether or not they should attempt to process the message” (p. 245).
Therefore. only under conditions of high perceived source expertise
will valence of message cognitions vary as a function of perceived
argument strength. Finally. at high perceived involvement. valence of
message cognitions is predicted to vary strongly as a function of
perceived argument strength. The valence of message cognitions is
predicted to occur in a way consistent with a three-way interaction of
perceived involvement. perceived argument strength. and perceived
source expertise. The nature of this interaction on the valence of

message cognitions is presented in Table 5.

 

Table 5 about here

 

Attitudes in Model 5 are predicted to be a function of the valence
of both message cognitions and source cognitions. Specifically. when
the cognitions evaluate the source positively. or support a message
argument. attitudes are predicted to be consistent with message
recommendations. When source or message cognitions are negatively
disposed toward the source or the message (i.e. counterarguments)

attitudes are predicted to be counter to message recommendations.

38

Post-message attitudes. in turn. are predicted to influence behavioral
intentions.
Individual Differgnces

One individual difference characteristic used within the context
of the ELM is need for cognition. Originally developed by Cohen
(Cohen. Stotland. 8 Rolfe. 1955). Petty and Cacioppo (1986) have
reconceptualized this construct. defining it as a person's general
motivation to engage in ”effortful cognitive endeavors” (p. 47). This
characteristic has obvious relevance for the ELM. Persons high in need
for cognition are more likely to take the central persuasive route than
persons low in need for cognition. who are more likely to take the
peripheral route.

In two separate studies. Cacioppo et al. (1983) found that high
need for cognition recipients discriminated more strongly between
strong and weak arguments than did low need for cognition recipients.
Strong arguments created more attitude change and weak arguments less
attitude change for high need for cognition recipients. Furthermore.
in both studies Cacioppo et a1. (1983) present evidence that these
attitude change data were mediated by greater cognitive evaluation of
message content for high need for cognition recipients. In the first
experiment. high need for cognition recipients felt that the message
with strong arguments was more important than the message with weak
arguments. In addition. high need for cognition recipients recalled
more arguments than low need for cognition recipients. In the second
study. the correlations between message evaluation and attitudes were

much higher for high (;§.70) than low need for cognition recipients

39

(p;.22).

A second individual difference characteristic of interest to
persuasion researchers is argumentativeness. Infants and Rancer (1982)
defined argumentativeness as a ”generally stable trait which
predisposes the individual in communication situations to advocate
positions on controversial issues and to attack verbally the positions
which other people take on these issues” (p. 72).

Differences in argumentativeness have important implications for
the ELM. Specifically. persons high in argumentativeness are likely to
expend considerable cognitive energy evaluating message content
regardless of level of involvement and are thus likely to be persuaded
by the strength of message arguments. Persons low in argumentativeness
are not likely to exert considerable energy evaluating the message.
These persons are likely to be persuaded by peripheral cues such as
source expertise. Thus. persons high in argumentativeness. need for
cognition. or both are more likely to take the central route to
persuasion than persons scoring low on these two variables.

Just as the models in Figures 4 and 5 differ in whether
involvement interacts at the perceptual or processing stage. the
question arises as to whether need for cognition and argumentativeness
influence the perception or the processing of experimental variables.
Specifically. regardless of the level of perceived involvement. persons
high in need for cognition and argumentativeness may exhibit less
source and more message cognitive processing than those low in these
characteristics. The second possibility is that need for cognition and
argumentativeness may directly affect perceived involvement. with

persons high in need for cognition and argumentativeness finding almost

40

all messages equally interesting and involving. It is this interest in
messages. then. that may create more strenuous message processing.

A final individual difference characteristic relevant to the
present study is respondents' pre-message attitudes. In this case.
attitudes toward the topic are likely influence message processing and
attitudes. Respondents' post-message attitudes are likely to be a
function of both the message and source cognitions. as predicted by the
ELM. as well as of previous attitudes. In other words. where one ends
up depends on where one begins. Pre-message attitudes will be included
in the analyses of Models 1 through 5 presented above. It is predicted
that pre-message attitudes will have an impact only on post-message
attitudes.

ti u - vio ist nc

Although not central to the issue of serial or parallel
information processing. nor relevant to the adequacy of Models 1
through 5. the methods performed in the present study allow the
investigation of a further question relevant to the ELM. Specifically.
the distinction between the central and peripheral routes to persuasion
is important not only because of their impacts on immediate attitude
change. but also for their differential effects on the consistency of
attitudes and behavior. Changes generated by the central route. when
compared to changes generated by the peripheral route. ”are postulated
to be relatively persistent. predictive of behavior. and resistant to
change" (Petty. Kasmer. Maugtvedt. and Cacioppo. 1987. p. 223) due to
the interconnections within the attitude-object schema provided by

message processing. Relatively little research. however. has been

41

performed linking the peripheral routes to attitude-behavior
consistency.

Evidence for greater attitude-behavior consistency at high levels
of issue-involvement can be found in Sivacek and Crano (1982). They
investigated the consistency between students' attitudes and behaviors
regarding a statewide referendum designed to raise the legal drinking
age from 18 to 21. Respondents were divided into three "vested
interest” groups by their age at the time the increase in the drinking
age would be enacted: High vested interest respondents would be unable
to drink legally for a long period; moderate vested interest
respondents would not be able to drink legally for a short period. and
low interest respondents would not be affected by the law. Ten days to
two weeks after the measurement of attitudes. respondents were phoned
and asked if they would like to assist a student organization working
against the referendum. The behavioral measure was the number of calls
the respondent was willing to make. Results indicated that as vested
interest (i.e.. involvement) increased. attitude-behavior consistency
became stronger. Specifically. attitude-behavior correlations were
stronger in the high vested interest condition <g;.6l) than either in
the moderate (g;.40) or in the low (g;.l6) vested interest conditions.

These results are consistent with the ELM. Proponents of the ELM
would assert that greater vested interest is likely to be related to
the amount of issue-related thinking the person is willing to exert in
evaluating the message. This greater amount of issue-related thinking.

according to advocates of the ELM. leads to greater attitude-behavior

42

consistency. Thus. in the present study recipients are likely to have
a greater amount of attitude-relevant information on which to base
their behavior (c.f. Davidson. Yantis. Morwood. 6 Montano. 1985).
Individual Differences and Attitude-Behavior Consistgncy

Greater involvement in the persuasive message should lead to
greater message elaboration. and this increased message elaboration
should lead to greater consistency between attitudes and behaviors.
Furthermore. Petty and Cacioppo (Cacioppo. et al.. 1983; Petty 8
Cacioppo. 1986) assert that individuals high in need for cognition
exert greater cognitive effort in evaluating persuasive messages; i.e..
need for cognition motivates respondents to elaborate upon the message
in a manner similar to issue involvement. If need for cognition and
argumentativeness lead to greater motivation to elaborate on messages.
respondents scoring high on these variables should display greater
attitude-behavior consistency than respondents low in these traits.
Therefore. an additional question to be answered in the present study
is whether persons high in need for cognition. argumentativeness. or
both will exhibit greater degrees of attitude-behavior consistency than

their low scoring counterparts.

CHAPTER 2
METHOD
Masts

Male and female undergraduates <M=325> at Michigan State
University participated in the study to earn extra credit for their
business communication. introductory communication. and small group
communication courses. Of the 325 participants. 227 read and responded
to the experimental stimuli. Ninety-eight participants completed the
pretest attitude measure. however. did not attend the experimental
session.

Esme.

The study was a 3 (issue involvement: high. medium. low) 8 2
(source expertise: high. low) X 2 (argument quality: high. low)
factorial. independent groups design. Participants were nested within
groups.

Issue involvement. Issue involvement refers to the extent to ”3
which a message has personal consequences for audience members. This
variable was controlled by modifying experimental instructions.
Participants in the high involvement condition were told that Michigan
State University was considering a policy requiring undergraduates to
take and pass comprehensive examinations as a requirement for
graduation for the following school year. Participants in the moderate
involvement condition were told that students at Northern Arizona
University would be required to take and pass comprehensive
examinations as a graduation requirement beginning in 10 years.
Participants in the low involvement conditions were told that Honors
students in Swedish universities would be required to take and pass

43 )

44

comprehensive examinations in order to graduate beginning in 10 years.

Source expertise. Source expertise was defined as the extent to
which participants perceive a message source as possessing skills and
specialized knowledge in an area relevant to the message. This
variable was also controlled by modifying experimental instructions.
Respondents in the high expertise conditions were told that the author
of the message was a Harvard professor nationally renowned for his
research on educational testing and a leading expert on comprehensive
examinations. Respondents in the low expertise conditions were told
that the author was a man who had recently purchased a considerable
amount of stock in a company that develops and sells comprehensive
examinations. Further. they were told the author knew little about the
examinations themselves and that he would make a large profit if
comprehensive examinations were adopted at more universities.

Argument quality. Argument quality refers to the extent to which i;
the arguments used to support the advocacy are sound and compelling. it
Although all participants read a message advocating institution of a
policy requiring comprehensive examinations for undergraduates. the
supporting arguments were varied. One-half of the participants read a
message containing strong arguments. i.e.. arguments designed to
produce thoughts supporting message recommendations. The remaining
participants read a message containing weak arguments. i.e.. arguments
designed to generate thoughts opposing message recommendations. The
arguments used in the study were taken from the strong and weak
arguments provided by Petty and Cacioppo (1986). The three strongest

and three weakest of these arguments were used (see Appendix).

45

Instrumentation

Cognitive respgnses. After reading the message. participants were
instructed to list the thoughts occurring to them while reading the
message. Eighteen. six-inch horizontal lines were provided for
participants to record their thoughts. one per line. After recording
their thoughts. participants were asked to rate their thoughts as
either positive. negative. or neutral toward the policy of
comprehensive examinations for undergraduates (see Petty. Cacioppo. &
Heesacker. 1981). In addition. trained judges rated participants'
thoughts on the dimensions of valence (pro-message or anti-message or
neutral) and target (source-related. message-related. or neutral; see
Cacioppo. Harkins. 8 Petty. I981; Chaiken. 1980).

Seven categories of cognitions were generated. There were three
categories of message cognitions: positive (e.g.. ”Passing the test
with a high score (can) help attain a job.” ”Exams would be a good
idea.'). negative ("undocumented.” "Profs would become lazy because
they would have to administer a test.") and neutral (e.g.. ”In France.
they have comprehensive examinations." "About how long will they
be?”). Three categories of source cognitions were generated: positive
(”Roger Smith is credible because he is from Harvard.” ”Written by an
intelligent man (Harvard).”). negative (“Roger Smith is out for
himself.” ”The author is greedy."). and neutral (”The fact that Roger
Smith has a financial risk in this study.” ”Visualize Mr. Smith.”).
In addition. there was one category of irrelevant responses. i.e..
responses not relating to the message or the source of the message

(e.g.. ”I feel awful. I've been sick all night.” ”The girl next to me

46

has big boobs.” ”Mike is already engaged. but I still want him.”).

Two trained raters coded each of the cognitive responses for both
valence and target. Inter-coder reliability was acceptable for both
categories. Coder agreement was 85% on valence (kappa=.70; _;.68) and
958 on target (kappa=.86; r=.78). In the case of disagreement on
valence. the respondent's own rating of valence was used. In the case
of disagreement on target. a third rater. blind to the experimental
condition. was asked to break the tie. Scores for the variables of
valence of source cognitions and valence of message cognitions were
computed by subtracting the number of negative thoughts from the number
of positive thoughts within each target category.

Issue involvement. Perceived issue involvement was measured by
responses to four. five-point. Likert items. They included ”The
outcome of this proposal is important to me;” and ”This proposal is
relevant to pg graduation from MSU."

Dimensionality of measurement was tested by confirmatory factor
analysis (Hunter. Cohen. 6 Nicol. 1982). All multi-item measures were
tested for unidimensionality. Hunter (1980) suggests three criteria
for testing the unidimensionality of a set of items: first. that the
items are homogeneous in content. so that all items measuring a single
cluster have similar semantic content; second. that the items are
internally consistent. so that items correlate with one another to
within sampling error of the product of their factor loadings; third.
that items in a scale are parallel. i.e.. that items in a scale
correlate similarly with other scales.

The involvement measure was found to be both internally consistent

47

(see Table 68). parallel with other measures (see Table 6C), and have
moderate reliability (a=.75). Means. standard deviations. and factor '
loadings for each of the items in the measure are presented in Table

6A.

 

Tables 6 A. B. and C About Here

 

In several instances. respondents were divided into three groups
by their scores on the perceived involvement scale. Respondents
reporting perceived involvement scores between five and nine were
placed into the low perceived involvement group. Respondents reporting
scores between 10 and 12 were placed into the moderate perceived
involvement group. Finally. respondents reporting scores between 13
and 20 were placed into the high perceived involvement group.

Argument guality. Perceived argument quality was measured by
responses to four. five point. Likert items. These items were modified
from Poster and Mayer (1984) where they were found to be
unidimensional. Items used to measure argument strength included "The
arguments contained in the message were convincing." and ”The arguments
contained in the message were strong."

The perceived argument quality measure was found to be internally
consistent (see Table 78). parallel with other measures (see Table 70).
and highly reliable (m=.88). Means. standard deviations. and factor
loadings for each of the argument quality items are presented in Table

7A.

48

 

Tables 7 A. B. and C About Here

 

In several instances. respondents were placed into one of two
groups by their score on the perceived argument quality scale.
Respondents reporting scores between five and 11 were placed into the
low perceived involvement group. while those reporting scores between
12 and 20 were placed into the high perceived argument strength group.

Source expgrtise. Perceived source expertise was measured by
responses to eight. five point. Likert-type items. Items used to
measure perceived source expertise included "The author of this message
is unawape of the issues surrounding comprehensive examinations." and
”The author is qualified to speak on this issue.”

The perceived source expertise measure was found to be internally
consistent (see Table 88). parallel with other measures (see Table 80).
and highly reliable (a:.95). The means. standard deviations. and

factor loadings of all items in this measure are presented in Table 8A.

 

Tables 8 A. B. and C About Here

 

In several instances. respondents were placed into one of two
groups by their score on the perceived source expertise scale.
Respondents reporting scores between eight and 24 were placed into the
low perceived source expertise group. while those reporting scores
between 25 and 40 were placed into the high perceived source expertise

group.

49

Attitudes. Respondents’ attitudes toward the topic of
comprehensive examinations for undergraduates were measured twice.

Each time. attitudes were tapped with five semantic differentials
anchored by good/bad. favorable/unfavorable. wise/foolish.
beneficial/harmful and satisfactory/unsetisfactory.

Both pre-message and post-message attitude scales were found to be
internally consistent (see Tables 98 and 108). parallel with other
measures (see Tables 90 and 100). and highly reliable (pre-
message:u=.95; post-message: a=.93). Means. standard deviations. and
factor loadings of all items in these measures are presented in Tables

9A and 10A. respectively.

 

Tables 9 A. B. and C About Here

 

 

Tables 10 A. B. and C About Here

 

Behavioral intention, Respondents were informed that the
proponents of the proposal outlined in the message wished to receive
feedback from interested students and that a (fictitious) meeting would
be held two weeks hence to receive such feedback. Students were also
told several student groups were working both for and against the
proposals. Students were asked. first. their intention to attend the
meeting. Second. respondents were asked whether they wished to work
for a group that supported the proposal. Opposed the proposal or

whether they did not wish to work for either group. If respondents

50

indicated that they wished to work for one of the student groups they
were also asked how many hours they wanted to work for those groups.
The components comprising the behavioral measure were found to be
internally consistent (see Table 118). parallel (see Table 110). and of
moderate reliability for such a measure (s=.59). Means. standard
deviations. and factor loadings for all items comprising this measure

are presented in Table 11A.

 

Tables 11 A. B. and C About Here

 

Qggd for cogniiion. Need for cognition was measured with the 18-
item Need for Cognition scale developed by Cacioppo and Petty (1982)
and reported in Petty and Cacioppo (1986). Confirmatory factor
analysis indicated a lack of unidimensionality in the scale. Six of
the original items were retained and found to fit a unidimensional
model.

The reduced Mead for Cognition scale was found to be internally
consistent (see Table 128). parallel (see Table 120). and of moderate
reliability (m=.77). Means. standard deviations. and factor loadings

for all items comprising this measure are presented in Table 12A.

 

Tables 12 A. B. and C About Here

 

Agggggntatiggness. Argumentativeness was measured by the 20 item
Argumentativeness scale developed by Infante and Rancer (1982).

Initial confirmatory factor analysis indicated multidimensionality in

51

the measurement. Following Infante and Rancer (1982). the scale was
divided into approach and avoidance dimensions. The final avoidance
dimension was comprised of five items. These items were found to be
internally consistent (see Table 138). parallel (see Table 130). and of
moderate reliability (m=.82). The final approach dimension was
comprised of three items. These three items were found to be
internally consistent (see Table 148). parallel with other measures
(see Table 140). and of moderately low reliability (m=.63). Means.
standard deviations. and factor loadings for all items comprising the
avoidance measure are presented in Table 13A. while this information

for the items in the argument enjoyment measure are presented in Table

 

 

 

 

14A.
Tables 13 A. 8. and C About Here
Tables 14 A. 8. and C About Here
Procedures

Piiot iesting. .A separate group of 60 respondents were asked to
read and rate the strength of the arguments presented by Petty and
Cacioppo (1986). Each respondent was given six arguments (three strong
and three weak). Approximately 20 respondents rated each individual
argument.

An additional separate group of 66 respondents was asked to

evaluate one involvement and one expertise condition. Respondents

52

first read the involvement manipulation (i.e.. the description of the
comprehensive examination policy) and completed the involvement
measure. Respondents then read the source expertise manipulation
(i.e.. the description of the author) and completed the expertise
measures.

Expgrimental sesgions. Data were collected in two waves. In the
initial wave. recipients' pretest opinions were measured on the focal
issue. compulsory comprehensive examinations for undergraduates. This
measurement was disguised in the form of a student opinion poll. Six
issues of local. regional. or national interest were tapped in this
opinion poll. The second topic in each packet was the experimental
issue of comprehensive examinations for undergraduates.

One week later. the experimental sessions occurred. After
completing research consent forms. participants were given a packet of
materials. Packet instructions contained both the involvement and
source expertise manipulations. Following the instructions was an
approximately 350-word message regarding the adoption of comprehensive
examinations for undergraduate students. After reading the message.
participants completed the cognitive response. manipulation check.
attitude. behavioral intention. and individual difference scales.

Upon collection of the final individual difference measurement
scale respondents were fully debriefed. The nature of the study. the
nature of the deception performed in the study. and the reasons for the

deception were discussed at length.

CHAPTER 3
RESULTS
W

Argument strength. Eighteen arguments (nine strong and nine weak)
provided by Petty and Cacioppo (1986) were pilot tested for perceived
argument strength. Results of analyses of perceived argument strength
ratings indicated that on a scale which ranged from four to 20 strong
arguments (M;l3.66) were rated stronger than weak arguments (M;9.30;
3535)=3.79; g;.54 p(.OOl; all statistical tests are two-tailed).

To maximize the differences between strong and weak messages. the
three strongest and weakest arguments were chosen for the experimental
messages. Results indicate that the three strongest arguments
(arguments S1. S2. and 86 from Petty and Cacioppo. 1986. p. 54-59) were
substantially stronger (M;14.36) than the three weakest arguments
(arguments 83. H7. and R8; §;7.98; i535)=5.85; §;.69; p(.OOl).
Therefore. there is a substantial difference between the strong and
weak argument messages.

Invplvement. Respondents were asked to report the extent of their
perceived involvement to one of the three involvement conditions.
Results indicate that on a scale ranging from five to 25 the low
involvement condition (!;11.90) was rated lower than the moderate
condition (M;14.13) which. in turn. was rated lower than the high
involvement condition (M;21.82). A one-way AMOVA.indicated that the
involvement manipulation had a strong impact on perceived involvement
ratings (EKZ. 63)=32.33; p(.001; gtg3;.51; gt;;.68). Moreover. 91
percent of the sums of squares for the main effect for involvement can
be explained by the linear trend (F

'lin
53

(1.63)=58.73; §£F1. 63; 25.001;

54

p;.68). The quadratic trend was also significant; however. this effect
was substantially smaller than the linear trend (Fquad(1' 63)=5.93;
13.01; 3.22).6

§gugce ggpgrtise. Ratings of perceived source expertise indicate
that on a scale which ranges from eight to 40 respondents perceived the
highly expert source (M;31.03) as substantially and significantly more
expert than the low expertise source (M;l7.41). Since respondents
completed the expertise manipulation check after completing the
involvement manipulation check a two-way AMOVA was performed on these
data. Results of these analyses indicate that the expertise
manipulation had a strong impact on ratings of perceived expertise
(2(1. 58)=96.30; p(.OOl; 9:11;.61; p;.78. gf=.80 when corrected for
attenuation due to measurement error). Involvement had a trivial and
statistically nonsignificant impact on ratings of perceived expertise
both as a main effect and when interacting with source expertise.

ment 1 u t : i u atio e k

Agggpgnt strength. Data indicating the impact of involvement.
source expertise and argument strength on perceived argument strength
are arrayed in Table 15. The argument strength manipulation exerted a
strong impact on respondents' ratings of perceived argument strength.
On a scale ranging from four to 20. respondents in the high argument
strength conditions rated the arguments in the message as stronger
(M;l3.42) than the arguments read by respondents in the low argument
strength conditions (M;9.02; 231$ 211)=152.01 p(.001; =.63; when cell

sizes vary. unweighted means analysis of variance is performed).

55

 

Table 15 About Here

 

In addition. the source expertise manipulation exerted a
significant. but much weaker impact on respondents' ratings of
perceived argument strength. Messages allegedly written by the highly
expert source were rated as containing stronger arguments (!;11.76)
than the message written by the inexpert source (Ms9.02; E31.
211)=11.41; p(.001; r;.l7).

The involvement and argument strength manipulations also
interacted to affect perceived argument strength.(§32. 211)=4.74;
p;.01; g;.16). As involvement increased. respondents differentiated
more strongly between strong and weak arguments.

Finally. the argument strength and source expertise manipulations
interacted to affect ratings of perceived argument strength (3 1.
211)=4.43; 2?.04; ;;.11). Specifically. in the strong argument
conditions. respondents' ratings of perceived argument strength did not
differ across levels of source expertise. In the weak argument
conditions. respondents' ratings of perceived argument strength
differed. with the high expertise source's messages being rated as
stronger than the low expertise source's messages.

For several analyses. respondents were collapsed into groups of
high and low perceived argument strength. based on their score on the
perceived argument strength measure. Respondents in the low perceived
argument strength group (M;8.0l) perceived message arguments as

significantly and substantially weaker than respondents in the high

56

perceived argument strength group (M;14.06; i(225)=23.73; p(.001).

Involvement. Data indicating the impact of involvement. source
expertise. and argument strength on perceived argument strength are
presented in Table 16. The only manipulated variable to have a
significant effect on perceived issue-involvement was controlled issue-
involvement (£32. 215)=17.9l; p(.OOl; _;.38). On a scale ranging from
four to 20 respondents in the low (!;11.13) and moderate (!;10.35)
conditions did not differ from one another (t(144)=1.49;§;-.12 p).lO).
Both groups reported lower levels of perceived involvement than those
respondents in the high involvement conditions (5:13.60). The
difference between the high and low involvement (i(144)=4.05; p(.OOl;
p;.32) and between the high and moderate involvement conditions

(3(144)=5.51; p(.OOl; g;.42) were significant and substantial.

 

Table 16 About Here

 

Although it would have been ideal to have the low. moderate and
high involvement groups at equidistant points on the involvement
continuum. the lack of such ordering of groups in the present study is
not a fatal flaw. The goal of any manipulation is to generate
different levels of a variable within respondents. Although the
differences in the present study are not exactly as intended. it is
clear that varying levels of involvement 33;; generated. First. central
tendency measures clustered around the midpoint of the scale of 12
(pggp;11.73. pggigp§11.31). Moreover. the distribution of perceived

involvement scores is not strongly skewed (skewness=.301) and at the

57

same time is somewhat flatter than the normal distribution (kurtosis=-
.581). Moreover. the standard deviation of 3.71 indicated that there
are a substantial number of respondents at each extreme of perceived
involvement.

Since the data were analyzed with causal modeling techniques. the
less than ideal manipulation of involvement is not of great concern.
since the models suggest that the involvement manipulation has an
impact on the pgrceived level of involvement. It is this perceived
involvement that produces the important effects in the proposed models.

In addition. for several analyses respondents were divided into
groups of low. moderate. and high pggggiggg involvement by their scores
on the perceived involvement measure. Respondents in the low perceived
involvement (M;7.52) reported significantly and substantially less
involvement in the message than respondents in the moderate perceived
involvement group (!;11.00). In turn. respondents in the moderate
perceived involvement group reported significantly and substantially
less involvement than respondents in the high perceived involvement
group (MF15.88; §;524.70; p(.001)

§gpgggiexpgrtise. Data indicating the impact of involvement.
source expertise. and argument strength on perceived source expertise
are presented in Table 17. The source expertise manipulation exerted a
strong impact on respondents' ratings of perceived source expertise.

On a scale ranging from eight to 40. respondents rated the high expert
source as more expert (M;26.97) than the low expertise source (M;l7.63;

B1! 210’=148a64; 2(0001; £=e61)a

58

 

Table 17 About Here

 

In addition. the argument strength manipulation exerted a
significant. but much weaker impact on ratings of perceived expertise.
Respondents in the strong argument conditions rated the source as more
expert (M;24.12; than respondents who read the weak argument messages
(!;20.56; 231. 210)=3.58; p;.03; 5;.13).

Again. in several sets of analyses. groups of high and low
perceived source expertise were created by using respondents’ scores on
the source expertise measure. Respondents placed in the low perceived
expertise group (M;16.58) reported significantly and substantially more
negative evaluations of the source than respondents in the high
perceived source expertise group (§;29.25; §;23.97; p(.001).
Expgrimental Results: Pretest Attitu es

Pretest attitudes. on the whole. were neutral. On a scale ranging
from five to 35. the average of all respondents completing the pretest
was near the midpoint of the scale of 20 (§;21.98). The distribution
of pretest attitude scores had a standard deviation of 6.55. and was
neither strongly skewed (skewness=-.262) nor strongly peaked or flat
(kprtogip=-.137). Pretest attitude scores did not differ across the 12
experimental conditions.

r me al u ' Influen e o ost-Messa e tti udes

Impact of the involvement. source expertise. and argument strength

manipulations is summarized in Table 18. Both the involvement and

argument strength manipulations exerted significant impacts on

59

post-message attitudes. As involvement increased. attitudes became
more strongly opposed to message recommendations. On a scale ranging
from five (negative) to 35 (positive). respondents in the low
involvement condition (M;22.88) felt most positively toward the
proposal while respondents in the moderate (!;21.63) and low (M;l9.76)
conditions held increasingly negative attitudes (£32. 215)=4.02;

p?.019; p;.-.18).

 

Table 18 About Here

 

The effect of argument strength on post-message attitudes
indicated that respondents in the strong argument conditions (M;23.27)
felt more positively toward the proposal than did respondents in the
low argument strength conditions (M;l9.69; EKl.215)=l5.40; p(.001;
;;.25). Source expertise had a trivial and nonsignificant impact on
attitudes. In addition. ANOVA results indicated that none of the
potential interactions between manipulated variables in this design
exerted a substantively strong nor statistically significant impact on
attitudes.

Additional analyses revealed two variables that influenced post-
message attitudes. First was message cognitions: as message cognition
become more positive. attitudes were more consistent with message
recommendations (p;.57). Second. pro-message attitudes exerted a
significant impact on post-message attitudes (§;.38). i.e.. post-
message attitudes tended to be consistent with pre-message attitudes.

Finally. completing the attitude pretest had no impact on

60

respondents' reports of post-message attitudes. Those respondents who
completed both the pretest and post-message attitude measurements did
not differ in post-message attitudes (M;21.46) from those who completed
only the post-message attitude measure (!;21.21; 5(225)=.l9; r=.01;
p).20).

Expgrimental figsults: Source and Message Ipteractigns

0f the several causal models tested Models 1 through 5 differed in
the nature of cognitive processing presumed to occur during message
reception. Models 1 and 2 suggest that perceptions of source expertise
and perceptions of argument strength are partially a function of
perceived involvement. Models 3. 4. and 5. on the other hand. predict
that source cognitions and message cognitions are a function.
partially. of perceived involvement.

The analysis of variance is inadequate to compare the adequacy of
these models since it investigates only the impacts of the three
manipulations. not the variables as they are perceived by
participants. Moreover. the AMOVA tests only for the standard
disordinal interactions and not the specific nature of the predicted
interactions. Therefore. the specific interactions predicted in Models
1 through 5 must be tested. These tests include the use of perceived
involvement. and the other perceived variables when applicable. If the
predicted interactions are not observed. then the model predicting that
interaction is incorrect. Therefore. the testing of the interactions

represents a necessary. but not sufficient. test of the models.

61

Model 1: A serial model of perceptual processes . Model 1

predicts that recipients will perceive source expertise information
consistent with a serial interaction between perceived involvement and
the source expertise manipulation on perceived source expertise (see
Table l). The observed impact of perceived involvement and source
expertise on perceived source expertise is presented in Table 19.
These data suggest that perceived source expertise is a function of the
main effect for manipulated source expertise. Since differences in
perceived expertise scores between high and low source expertise
conditions do not vary dramatically across the three perceived
involvement conditions. perceived involvement and source expertise do
not interact. as predicted by Model 1.

The interaction of perceived involvement and source expertise on
perceived source expertise developed from serial processing does.
however. correlate moderately with the obtained perceived expertise
scores. Specifically. the predictions of the serial interaction
correlate .30 with obtained perceived source expertise. The predicted
interaction correlates with obtained perceived source expertise scores
because the contrasts used to test this interaction are not orthogonal
with a main effect for the source expertise manipulation. The
correlation for the interaction is. however. substantially smaller
than the main effect of the source expertise manipulation on perceived

source expertise (;;.61; §(204)=8.92; p(.001).

62

 

Table 19 About Here

 

Model 1 also predicts that recipients will process message
information consistent with an interaction between perceived
involvement and the argument strength manipulation on perceived
argument strength. The observed impact of perceived involvement and
the argument strength manipulation on perceived argument strength is
presented in Table 20. The main effect for the argument strength
manipulation on perceived argument strength is .63. It is apparent
that these factors do not interact to affect perceived argument
strength. The predicted serial interaction correlates moderately with
the perceived source argument strength (g;.42). The main effect for
perceived argument strength is significantly larger than the effect
size for the predicted interaction (i5203)=5.68; p(.001). The
interaction effect correlates with observed perceived argument strength
scores because the contrasts used to test the interaction are not
orthogonal with the observed main effect for the argument strength

manipulation.

 

Table 20 About Here

 

In sum. message recipients do not process message and source
information consistent with interactions between perceived involvement
and argument strength on perceived argument strength and between

perceived involvement and source expertise on perceived source

63

expertise. The perceptions of source expertise and argument strength
do not depend on recipients' level of perceived involvement. Put
another way. recipients perceived both source and message
characteristics at all levels of perceived involvement.

Mggel 2: A pgggilgl modei of pgrceptual processes. Model 2
predicts that recipients perceive source expertise consistent with a
perceived involvement by source expertise manipulation interaction (see
Table 3). Table 19 shows there is no such interaction; i.e..
perception of source expertise does not depend on perceived
involvement. Since perceived involvement and source expertise did not
interact to influence perceived source expertise. Model 2 is. by
definition. inconsistent with the data. The source interaction
predicted by Model 2 correlated moderately with perceived argument
strength scores (5;.48). This correlation is not zero because the
predicted interaction is not an orthogonal contrast to the main effect
for source expertise. Again. the main effect for the source expertise
manipulation on perceived source expertise was significantly stronger
(§;.61) than the interaction effect ($5205)=2.32; p(.02).

In addition. Model 2 predicted an interaction between perceived
involvement and the argument strength manipulation on perceived
argument strength. The parallel interaction (p;.38) correlated
significantly less strongly with observed perceived argument scores
than did the main effect for argument strength (g;.63; i5205)=6.04;
p(.001). The data presented in Table 20 clearly indicate that
perceived argument strength scores vary solely as a function of the

argument strength manipulation; i.e.. perceived involvement does not

64

influence how person perceive the strength of message arguments. The
interaction correlates with perceived argument strength scores because
the interaction is not orthogonal to the observed main effect.

In analyzing the perception of message and source variables. the
only effects found were the main effects. Thus. it is clear that
Models 1 and 2. which specify serial or parallel interactions at the
perceptual stage. are inconsistent with the data. The results clearly
indicate that perceived involvement does not influence respondents'
perceptions of argument strength and source expertise as predicted by
the models.

Model 3: A serial model of message processing. The serial
processing model (see Figure 5) predicts that recipients process
message information consistent with a serial interaction between
perceived involvement and perceived argument strength on message
cognitions. Data relevant to the predicted interaction are presented
in Table 21. These data indicate that message cognitions are solely a
function of perceived argument strength (;;.45). The serial
interaction (represented in Table 2) correlated with message cognitions
significantly less strongly <;=.33; g_<204>=2.31; 1.5.01).7 The
correlation between the interaction of perceived involvement and
perceived argument strength on message cognitions is not zero because
the contrasts used to test this interaction are not orthogonal to the

observed main effect.

65

 

Table 21 about here

 

The serial processing model also predicts that recipients process
source information consistent with an interaction between perceived
involvement and perceived source expertise on source cognitions. This
serial interaction (represented in Table 1) predicts that at low levels
of involvement recipients’ source cognitions will very strongly as a
function of perceived source expertise. As involvement increases the
serial interaction predicts that perceived source expertise will exert
increasingly less influence on source cognitions. Data relevant to
this interaction are presented in Table 22. These data indicate that
perceived source expertise and perceived involvement interacted to
affect source processing but not in a way consistent with the contrasts
established in Table 1. Specifically. the valence of respondents'
source processing differed significantly across levels of perceived
source expertise at both the low (1(66)=4.36; g;.47; p(.001) and high
($585)=3.80; 5;.38; p(.001) levels of involvement. At the moderate
level of involvement. valence of source processing did not differ

across levels of perceived source expertise (g; )=.72; p;.19; p).lO).

 

Table 22 about here

 

Although perceived involvement and perceived source expertise

interact to affect source processing. the form of this interaction is

66

inconsistent with the interaction predicted by Model 3. The predicted
serial source interaction correlated moderately with observed source
processing scores (g;.27). The main effect for perceived source
expertise (§;.32) was not significantly stronger than the serial
interaction effect (i(205)=l.13; p).20).

In sum. the predicted serial interactions between perceived
involvement and perceived source expertise on source cognitions and
between perceived involvement and perceived argument strength on
message cognitions were inconsistent with the data. The valence of
cognitions regarding the source and message do not depend strongly on
recipients' level of perceived involvement. Stated differently.
valence of recipients’ source and message cognitions did not differ
across levels of perceived involvement.

Model 4: A.pgrallel model of message processing. The parallel
processing model presented predicts that recipients will process
message information consistent with an interaction between perceived
involvement and perceived argument strength on message cognitions. Data
presented in Table 21 indicate that the predicted parallel interaction
correlates moderately (;;.34) with message cognitions. This
correlation is significantly smaller than the main effect for perceived
argument strength (p;.45; i(220)=2.68; p(.01). The correlation between
the predicted interaction and message cognitions is not zero because
the interaction is not orthogonal to the simple main effect.

The parallel processing model predicts that source cognitions will
reflect perceived source expertise only at moderate levels of
involvement; i.e.. at both low and high levels of involvement. source

cognitions are not expected to differ across levels of perceived source

67

expertise. This prediction is inconsistent with obtained data. As
presented in Table 22. the moderate level of perceived involvement is

the only level where source cognitions fail to strongly reflect

 

perceived source expertise. The parallel source interaction does not
correlate strongly with the obtained data (p;.07).

In sum. the predicted parallel interactions between perceived
involvement and perceived argument strength on message cognitions and
between perceived involvement and perceived source expertise on source
cognitions were not supported by the results. The valence of source
and message cognitions does not depend on the recipients' level of
perceived involvement. Put another way. the valence of source and
message were consistent at all levels of perceived involvement.

Modei 5: Three-way model of message ppgcessing. The three-way
processing model predicts that recipients will process source and
message information consistent with a perceived involvement by'
perceived source expertise by perceived argument strength interaction
on both source and message processing. Data relevant to the three-way»
interaction of perceived involvement. perceived argument strength. and
perceived source expertise on message processing are presented in Table
23. These data indicate that at low levels of perceived involvement.
message processing is primarily a function of argument strength. At
moderate levels of perceived involvement. message processing is
negative in the low perceived expertise-low perceived argument strength
condition. and relatively positive in the other conditions. At the
high level of perceived involvement. message cognitions are relatively

positive in the high perceived expertise source-strong perceived

68

argument condition. In all other conditions. message cognitions are

opposed to message recommendations.

 

Table 23 About Here

 

Data presented in Table 23 reflect a strong main effect for
perceived argument strength (;;.35). The three-way interaction
predicted by Petty et al. (1987) correlates significantly less strongly
with the obtained pattern of means (5;.14; t5216)=4.12; p(.001).
Therefore. the model presenting the three-way interaction between
perceived involvement. perceived source expertise. and perceived
argument strength on message cognitions is not supported by the data.

Model 5 also predicts that recipients will process source
information consistent with an interaction between perceived
involvement. perceived source expertise. and perceived argument
strength on source cognitions (see Table 5). This model predicts that
perceived source expertise will exert the strongest influence on source
cognitions at low perceived involvement. with weak influence at
moderate levels of involvement. and no influence at high levels of
perceived involvement. Data relevant to this predicted interaction is

presented in Table 24.

 

Table 24 About Here

 

Results of the interaction between perceived involvement.

perceived source expertise. and perceived argument strength on source

69

cognitions are inconsistent with predictions. These data indicate that
source cognitions are primarily a function of perceived source
expertise (p;.35). The predicted interaction presented in Table 5 is
not consistent with these data (g;.28). This correlation. again. is
not zero since it does not provide an orthogonal contrast to the main
effect for perceived source expertise.

Tegts of models: Conciusions. Hone of the predicted models were
consistent with the present findings. Results pertinent to Models 1
and 2 indicate that interactions do not occur at the perceptual stage;
i.e.. perceived source expertise is a direct function of the source
expertise control. while perceived argument strength is a direct
function of the argument strength control. Results pertinent to Models
3. 4. and 5 indicate that perceptions of source and message variables
did not interact to affect source and message cognitions. Message
cognitions were found to be a direct function of perceived argument
strength while source cognitions were found to be a function.
primarily. of perceived source expertise.

The ELM predicts that source and message variables generate
positive and negative cognitions about the message and the source.
These cognitions. in turn. are predicted to influence attitudes. At
the same time. the ELM predicts that perceived involvement in the
message will mediate this effect of message and source factors on
attitudes. These results support the former but not the latter
assumption. Involvement failed to interact with perceived argument
strength to influence message cognitions. nor did perceived involvement

interact with perceived source expertise to influence source

70

cognitions. In fact. perceived involvement had no important and
substantial impacts in the analyses outlined above.

It is possible. however. to test the proposed models (e.g. Figure
4) without the predicted interactions. The only modifications
necessary to test such a model would be to remove the argument strength
by perceived involvement and the perceived involvement by source
expertise interactions. In this version of Figure 4. then. the
argument strength control is predicted to have a direct impact on
perceived argument strength. while the source expertise control is
predicted to have a direct impact on perceived source expertise.
Perceived involvement. in this model. relates to no other variable.

Such a model is presented in Figure 7.

 

Figure 7 About here

 

Expgrimental Results: Path Analyges

The adequacy of a causal model can be evaluated with two classes
of tests. micro and macro. Micro tests attempt to determine the fit of
individual paths within a causal model by investigating the
significance of individual paths and whether the paths differ
significantly from predictions. Macro tests indicate the overall
amount of error of prediction in the model by investigating overall
goodness-of-fit measures and other measures of the amount of error of
prediction in the model. Finally. the adequacy of a causal model may
be evaluated by the extent to which the model explains variation in

exogenous variables of interest (in the present case posttest attitudes

71

and behavioral intention).

- Initial analyses of the model in Figure 7 were performed using
both LISREL VI (Jareskog 8 Sarbom. 1984) and the PC-version of
PACKAGE. Correlations between variables in this model are presented
below the diagonal in Table 25. while deviations from predicted
correlations appear above the diagonal. Overall. the results indicate
this model is not an adequate representation of the data. The Chi-
square value is statistically significant (Chi-sguare(40)=285.88).
Since Chi-square represents a very powerful test. Bentler and Bonett
(1980) have suggested that the Chi-squared/degrees of freedom ratio be
used as a test of the fit of a model. with any ratio of less than five
indicating an adequate fit. The Chi-square/degrees of freedom ratio
for the present data is 7.15. The goodness of fit index for this model
is .840 while the adjusted goodness of fit index is .736. Errors of
prediction are commen in this model. The sum of the squared error for
Model M is 1.269. The average squared deviation is .023. while the

average absolute deviation is .100.

 

Table 25 About Here

 

Taken as a whole. these results indicate that Model 6 does not
adequately fit the data. Therefore. the model should be revised reduce
the errors of prediction. There are several aspects of the present
model that require attention. First. one path was found to be
nonsignificant. Specifically. source cognitions exerted no influence

on attitudes (path coefficient=-.01). and this path can be omitted from

72

further analyses of this model.

Analyses of Model 6 also uncovered 17 deviations significantly
larger than would be expected by sampling error alone. Several of
these deviations deserve mention. First. there is a large deviation
between perceived argument strength and perceived source expertise.
The large correlation between these variables (p;.58) indicates that a
path need be inserted here. with perceived argument strength the
antecedent and perceived source expertise the consequent variable.
Substantively. this indicates that recipients' perceptions of argument
strength influence their perceptions of the expertise of the source.

In addition. there is a significant deviation between pre-message
attitudes and message cognitions. Again. a direct path needs to be
inserted. indicating that pre-message attitudes are one of the factors
inducing the valence of message cognitions. Substantively. this
indicates that respondents' message processing is biased toward their
pre-message attitude.

A large number of deviations were found for paths including the
variable behavioral intentions. Specifically. significant deviations
were discovered for paths between manipulated involvement and
behavioral intention. perceived argument strength and behavioral
intention. manipulated argument strength and behavioral intention.
perceived argument strength and behavioral intention. and message
cognitions and behavioral intention. Since all these deviations
involve either involvement or argument strength variables. it seems
advisable to investigate the impact of perceived involvement and

perceived argument strength on behavioral intention.

73

The nature of the impact of perceived involvement and perceived
argument strength on behavioral intentions is presented in Table 26.
From this table. it can be seen that perceived involvement and
perceived argument strength interact to influence behavioral
intention. The main effects for perceived involvement (2(2. 198)=5.27;
p(.01) and perceived argument strength (Eil. 198)=7.58; p(.01) exert
statistically significant impacts on behavioral intentions. It is
clear. however. that these main effects can be more clearly described
by the interaction between these factors where only at high perceived
involvement will perceived argument strength exert a strong influence

on behavioral intention (F(2.l98)=5.11; p(.Ol).

 

Table 26 About Here

 

An alternative explanation for the perceived involvement by
perceived argument strength interaction on behavioral intention is that
the influence of perceived argument strength may be funneled through
attitudes. Such a suggestion is reasonable since perceived argument
strength exerts such a strong impact on attitudes. In the present
experiment. for example. it is likely that the only time a respondent
was motivated to behave was when the change in policy was imminent
(i.e. the high involvement condition) and when the respondent was
opposed to the proposal. Thus. given this logic. there should be a
strong perceived involvement by post-test attitude interaction on
behavioral intention. The data investigating such an interaction are

presented in Table 27.

74

 

Table 27 About Here

 

From Table 27. it should be clear that perceived involvement and
post-test attitude interact strongly to influence behavioral intention
(the post-test data were dichotomized using a median split). Only when
perceived involvement is high and post-test attitudes opposed to
message recommendations do respondents intend to behave. The contrasts
used to represent this interaction are presented in Table 28. In this
case. the interaction was found to exert a strong impact on behavioral

intentions (F(2.217)=42.58; p(.001; £L;54).

 

Table 28 About Here

 

A model incorporating the changes suggested by the deviations in
Model 6 is presented in Figure 8. This model differs from its
immediate predecessor in several respects. First. the perceived
involvement by post-test attitude interaction term has been included as
the only determinant of behavioral intentions. Second. the path from
source cognitions to post-test attitudes has been deleted. Finally.
several paths have been added to the model. Specifically. paths from
the source expertise manipulation to perceived argument strength. from
perceived argument strength to perceived source expertise. from
perceived argument strength to post-test attitudes. and from pre-

message attitude to message cognition have been added to the present

75

model in an attempt to attenuate the number and size of deviant paths.

 

Figure 8 About Here

 

Table 29 contains the path coefficients above the diagonal and
deviations between obtained path coefficients and predicted
coefficients. Given 55 correlations. one can expect 2.8 deviations to
be larger than would be expected due to sampling error; thus the five
deviations observed are larger than would be expected by sampling error
alone. These include deviations from the involvement manipulation to
message cognitions. from the involvement manipulation to post-message
attitudes. from the source expertise manipulation to source cognitions.
from the argument strength manipulation to behavioral intention. and
from perceived argument strength to behavioral intention are greater '

than would be expected given chance factors.

 

Table 29 About Here

 

 

All of the path coefficients in this model are significantly
different from zero (the smallest grvalue is 3.59; p(.05). Thus. the
micro indicators reflect an adequate fit between the obtained data and
the model's predictions. The macro indicators of fit include the sum
of squared deviations. the average squared deviation. and the average
absolute deviation. The sum of squared deviations in the model was

.29. The average squared deviation was .005. The average absolute

76

deviation was .055. The Chi-square value for the overall fit of the
model was 102.16 (df=46; p(.0001). The Chi—square value may be
significant because it provides an overly powerful test of fit. The
present model fulfills the criterion of the ratio of Chi-square to
degrees of freedom suggested by Bentler and Bonett (1980; Chi-
squareldf=2.22). Finally. the goodness of fit index for the model is
.932. while the adjusted goodness of fit index is .885.

The final macro test of a causal model is the squared multiple
correlation. The squared multiple correlation gives an indication of
the strength of effect variables in the model exert on endogenous
variables. In the present case. the endogenous variables of interest
are post-message attitudes and behavioral intention. The multiple R?
on post-message attitudes given the structural equations generated from
Model 7 is .479. The multiple R2 for behavioral intention is .504.

Thus. in conclusion. both the micro and macro tests indicate that
Model 7 provides an adequate fit to the present data. All the paths
predicted to be significant are significant; the variables in the model
exert strong impact on endogenous variables of interest. and the model
does not contain large errors in predictions.

s: Att tu -8ehavior stenc

The ELM predicts that the relationship between message elaboration
and the consistency of attitudes and behaviors is linear and positive.
To test this prediction. involvement groups were compared in the extent
to which their attitudes were consistent with behaviors. Overall.
posttest attitudes were significantly related to behavioral intention

(;;.33; p(.001). Respondents were divided into three groups by their

77

scores on perceived involvement. Attitude-behavior correlations were
.26 for the low perceived involvement group. .11 for the moderate
perceived involvement group. and .47 for the high perceived involvement
group.

Analyses indicate that the low and moderate perceived involvement
groups do not differ from one another in attitude-behavior consistency
(n;.89; p).05). The high perceived involvement group did differ
significantly from the moderate involvement group (n;2.40; p(.05). The
high involvement group did not differ from the low involvement group in
the degree of attitude-behavior consistency (n;l.45; p).05). These
results imply that attitudes do not become more consistent with
behaviors in a linear fashion as the amount of perceived involvement
increases.

Additional analyges: Individual Differences

An additional prediction related the impact of need for cognition
and argumentativeness as moderators of the relationship between source
and message manipulations and attitudes. Petty and Cacioppo (1986)
predict that low need for cognition/argumentativeness respondents will
take the peripheral route to persuasion and be persuaded by source
factors (i.e.. source expertise). High need for
cognition/argumentativeness respondents are expected to take the
central route to persuasion and should thus be persuaded by message
cues (i.e.. argument strength). If such a prediction were consistent
with the data. need for cognition/argumentativeness by perceived
argument strength and need for cognition/argumentativeness by perceived

source expertise interactions on attitudes would be substantial in size

78

and statistically significant.

The need for cognition by perceived argument strength interaction
on attitudes is depicted in Table 30. Table 30 clearly depicts a main
effect for perceived argument strength (p;.37). The predicted
interaction between need for cognition and perceived argument strength
on attitudes is statistically significant (p;.l8; p;.01). but small in
size and significantly smaller than the main effect for argument

strength (1(216)=3.97; p(.001).

 

Table 30 About Here

 

The need for cognition by source expertise interaction is depicted
in Table 31. In this case. source expertise is predicted to have a
greater impact on low need for cognition than high need for cognition
respondents. The data presented in Table 31 are clearly inconsistent
with the hypothesis that need for cognition influences perception of
source expertise. This effect is significant (n;.22) only because the
contrasts used to create this interaction are not orthogonal to the
simple main effect. The main effect for source expertise (;;.37) is
significantly and substantially larger (£5216)=2.92; p(.Ol). These
results are inconsistent with the proposition that need for cognition
acts as a moderator of the relationship between source and message

variables and attitudes.

79

 

Table 31 About Here

 

The enjoyment subscale of the argumentativeness scale by perceived
argument strength interaction on attitudes is depicted in Table 32.
Persons high in argument enjoyment are predicted to base their
attitudes more strongly on perceived argument strength than persons low
in argument enjoyment. Again. the interaction is significant (;;.24)
only because interaction is not orthogonal to a simple main effect for
argument strength. The main effect for argument strength (5;.38) exerts
a significantly and substantially larger influence on attitudes than

does the interaction (p(216)=2.92; p(.Ol).

 

Table 32 About Here

 

The enjoyment by perceived source expertise interaction on
attitudes is reported in Table 33. Although the impact of perceived
source expertise on attitudes is stronger for high argument enjoyment
respondents.this effect is not strong. for the interaction does not
exert a strong impact on attitudes (5?.14). Moreover. the interaction
exerts a significantly and substantially weaker impact on attitudes
than the main effect for source expertise (g§.3l; g5215)=3.46;
p(.001). This implies that argument enjoyment does not act as a
moderator between perception of source and message variables and

attitude change.

80

 

Table 33 About Here

 

Since the items for the avoidance scale were reflected. high
scores on the avoidance subscale reflect high scores for
argumentativeness. Therefore. argument strength is expected to have a
stronger impact on high avoidance respondents while source expertise is
expected to have a greater impact on low avoidance respondents. The
avoidance by argument strength interaction is presented in Table 34.
The predicted interaction (;;.29) and the main effect for argument
strength correlate approximately equally with post-message attitudes

<t_<217>=o.41; p_>.20).

 

Table 34 About Here

The avoidance by perceived source expertise interaction on
attitudes is found in Table 35. Source expertise is predicted to have
a stronger impact on attitudes for low avoidance respondents. The
results indicate that the opposite occurred; i.e.. source expertise
exerted a stronger impact for high. rather than low avoidance
respondents. The predicted interaction exerts a significant impact on
attitudes (5;.14) only because it is not an orthogonal contrast to the
main effect for source expertise. The interaction. moreover. exerts
significantly less impact on attitudes than does the main effect for
source expertise (g;.32; i5216)=3.72; p(.001). In sum. argument

avoidance is not a strong moderator of the relationship between source

81

and message factors and attitudes.

Table 35 About Here

 

Evidence for the moderating impact of need for cognition is also
presented by Cacioppo. et al.. (1983) by indicating that the
correlation between message cognitions and posttest attitudes was
greater for high than low need for cognition respondents. Similar
analyses were performed for the present data. Overall. there was
consistency between the valence of message cognition and post-message
attitudes (g;.57). Individual differences. in all cases. failed to
moderate the consistency between message cognitions and post-message
attitudes. High need for cognition respondents (5;.57) did not differ
from low need for cognition respondents (p;.52) in the degree to which
the nature of message processing was related to posttest attitudes
(5;.52; p).05). Similar results were uncovered for the enjoyment and
avoidance sub-scales of the argumentativeness scales. High enjoyment
respondents (g;.50) exhibited less cognition-attitude consistency than
did low enjoyment respondents (;;.58; n;.88; p).05). High avoidance
respondents exhibited more cognition-attitude consistency (n;.58) than
did low avoidance respondents (5;.51; 5;.88; p).05).

Finally. Cacioppo. et al. (1983) report that high need for
cognition respondents differentiated more strongly between strong and
weak arguments than did low need for cognition respondents. Thus. need
for cognition and the argumentativeness subscales are predicted to

interact separately with the argument strength manipulation to affect

82

judgements of perceived argument strength. Results for the interaction
of need for cognition and the argument strength manipulation on
perceived argument strength are reported in Table 36. These data
reveal a strong main effect for the argument strength manipulation
(;;.62). In addition. however. the interaction predicted by Cacioppo
et al. (1983) also exerted a significant impact on perceived argument
strength (n;.49). The difference between these correlations is
significant (g(216>=3.25; p(.Ol); i.e.. the main effect for the
argument strength manipulation is stronger than the predicted

interactions.

 

Table 36 About Here

 

The impact of the enjoyment by argument strength manipulation
interaction on perceived argument strength is presented in Table 7.
Again. the interaction between enjoyment and the argument strength
manipulation is significant (p;.49). but it is significantly weaker

than the main effect for argument strength (5;.62; t5216)=3.25; p(.Ol).

 

Table 37 About Here

 

Finally. the impact of the avoidance subscale of the
argumentativeness scale by argument strength manipulation interaction
on perceived argument strength is presented in Table 38. Once again.
the interaction exerts a significantly weaker impact (p;.48) on

perceived argument strength than does the main effect for argument

83

strength (§;.62; £5217)=3.51; p(.001).

 

Table 38 About Here

 

ngditional Analyges: Individual Differences and Attitudg-Behavior
Consistency

Petty and Cacioppo (1981; 1986) posit that when message recipients
exert considerable cognitive effort in evaluating messages. attitudes
will be consistent with behaviors. One of the factors hypothesized to
generate cognitive effort was need for cognition and/or
argumentativeness. Consequently. persons high in need for cognition
and argumentativeness should exhibit greater amounts of attitude-
behavior consistency than persons low in these traits. To examine this
possibility. need for cognition and the sub-scales of the
argumentativeness scale were compared for the extent to which they
moderated the consistency between attitudes and behaviors.

Results reveal a nonsignificant difference between low (p;.32) and
high (5;.35) need for cognition respondents (5;.25; p).05) in the
consistency between attitudes and behaviors. Similar results were
obtained for the enjoyment trait; specifically. there was a
nonsignificant difference between low (g;.24) and high (p;.40)
enjoyment respondents (n§1.28; p).05).

High avoidance respondents (g;.06). on the other hand. exhibited
significantly less attitude-behavioral consistency than did low
avoidance respondents (;;.45; n;3.06; p(.Ol) These results indicate

that persons who tend to avoid arguments exhibit less

84

attitude-behavioral consistency than those respondents who tend not to
avoid arguments. The implication is that those persons who avoid
arguments (in the present case those in! in avoidance because of the
reflection of items) will not exert as much cognitive effort in

evaluating message arguments.

Chapter 4
DISCUSSION

This study had two goals. First. it attempted to replicate the
findings of Stiff's (1986) metaanalysis. Since the manipulations and
methods employed were very similar to previous ELM studies. the study
could also be considered a replication of much of Petty and CaciOppo's
work. The second goal was to analyze the present data using causal
modeling techniques to gain a fuller understanding of the processes and
procedures underlying message reception and processing. The ultimate
outcome was thus a more clearly delineated conceptualization of how the
variables within the ELM function to affect attitudes and behavioral
intentions. This chapter will discuss the extent to which these goals
were reached. the limitations of the present study. and the direction
of future research in this area.
Replicatign of the Stiff finia-analygin

Results of Stiff's (1986) meta-analysis indicated that recipient
processing of message information resulted in an issue involvement by
evidence interaction on attitudes. Specifically. Stiff (1986) found
that as involvement increased argument strength exerted an increasingly
strong impact on attitudes. It was assumed that the impact of argument
strength on attitudes was mediated by both perceived argument strength
and message cognitions. Therefore. the present study investigated the
extent to which perceived involvement and perceived argument strength
interact to influence message cognitions. The results are inconsistent
with those reported by Stiff. In the present investigation. the impact
of perceived argument strength on message cognitions did not differ
across levels of perceived involvement.

85

86

The preceding statement emphasizes one of the differences between
the present study and Stiff's meta-analysis. Specifically. the impact
of pgrception of issue involvement and pgrception of argument strength
were investigated here. The impact of manipulated argument strength on
attitudes was found to be mediated by both perceived argument strength
and message cognitions. The present investigation thus assumes that
the impact of message and source factors on attitudes is indirect.
while previous ELM studies have been unable to differentiate between
direct and indirect models.

Stiff (1986) found that source expertise exerted a stronger impact '*’
on attitudes at moderate involvement than at either low or high
involvement. Again. the present investigation presumed that the impact
of the source manipulation on attitudes was mediated by both perceived
source expertise and source cognitions. Thus. the present
investigation investigated the impact of perceived involvement and
perceived source expertise on source cognitions. In fact. source ;;~»
cognitions did not differ as a function of perceived involvement.

Taken as a whole. then. results of this investigation are
inconsistent with the results of the Stiff metaanalysis. This study.
moreover. can be considered a replication of previous work on the ELM.

In fact. the present study is a particularly close replication of

Petty. Cacioppo. and Goldman (1981). who found that involvement

interacted separately with source expertise and argument strength to 4*
affect attitudes. Specifically. Petty. Cacioppo. and Goldman (1981)
found. as the ELM predicted. that argument strength exerted the

strongest impact on attitudes at high involvement while source

87

expertise exerted the greatest impact on attitudes at low involvement.afif’

Results of the present investigation are inconsistent with Petty.
Cacioppo. and Goldman (1981). Although the argument strength
manipulation exerted an effect on attitudes it did not differ across
levels of manipulated involvement. Moreover. the causal modeling
analyses indicate that the impact of the argument strength manipulation
on attitudes is mediated by both perceived argument strength and
message cognitions. The source expertise manipulation exerted no &;*”
impact on attitudes. Finally. neither perceived source expertise nor
source cognitions exerted a significant or substantial impact on
attitudes.

In conclusion. the data from the present study indicate that.
regardless of the level of manipulated or perceived involvement in the
message. respondents processed message content in a way consistent with
the central route to persuasion. At all levels of manipulated and
perceived involvement. attitudes were a function of cues generally
associated with central processing (e.g.. perceived argument strength
and message cognitions). Furthermore. at all levels of manipulated and
perceived involvement. attitudes gig_nng_vary as a function of
variables generally associated with peripheral processing of message

+5

content (e.g. perceived source expertise and source cognitions).

Results of Causai Modeiing

Predictions based on all five causal models tested were found to
be inconsistent with observed data. Perception and processing of
message or source information did not vary as a function of perceived

involvement.

88

Given the absence of source and message interactions. a model
identical to Figure 4 without interactions was tested. This model (see
Figure 7) was also found to be inconsistent with the data. One path
was deleted and several paths added to increase the fit of the model.

The final causal model (see Figure 8) generated from the present
data is instructive in several respects. First. it is clear that the
manipulations of the source and message variables have no direct impact
on attitudes and/or behavioral intentions. Inspection of Figure 8
reveals clearly that the argument strength manipulation exerts a direct
impact on pgrceived argument strength. which. in turn. influences
message cognitions. Both message cognitions and perceived argument
strength influence attitudes. A different pattern emerges for source
expertise. with the source manipulation exerting a direct impact on
perceived source expertise. Perceived source expertise exerts a direct
impact on source cognitions. but neither perceived source expertise nor
source cognition exerts a significant nor substantial impact on
attitudes. These results indicate that the impact of message and
source manipulations on attitudes is indirect.

This description of the final causal model underscores one of the
important findings of the present study. A central assumption of the
ELM is that source and message cognitions are important determinants of
attitudes. The present data support this assumption only for mgssagg
gggniiigng: Rhile message cognitions exert a strong impact on
attitudes. source cognitions exert no impact. either direct or
indirect. on attitudes.

A second important finding of this study is that perceived

89

involvement did ng; interact with perceived source expertise or
perceived argument strength to influence attitudes. As reported above.
these results are consistent with the hypothesis that respondents were
processing message information thought the central route to persuasion
at all levels of perceived involvement.

This discussion emphasizes several important points regarding the
role of cognitive processes in the ELM. First. from the causal
modeling analyses. it is clear that message cognitions exert a strong
impact on post-message attitudes. Thus. in analyzing cognitive
response data. it is clearly important to consider message and source
cognitions as more than simply a check for the involvement
manipulation.

Second. it is important to note the valence rather than the number
of message and source cognitions. The ELM makes clear predictions as
to the impact of message and source variables on the gnigngn of message
cognitions. Analyses in many previous ELM studies have investigated
simply the nnnnn; of cognitions. The operational difference between
the number and valence of cognitive responses represent an important
difference between the present study and previous ELM studies.

Finally. it is important to consider message cognitions and source
cognitions separately. Again. the ELM makes clear predictions
regarding the separate role of both message and source processing.
Moreover. Figure 8 indicates that source and message cognitions play
different roles in message processing. The summation of source and
message cognitions into one measure of cognitive processing is clearly

inappropriate.

90

Another interesting finding generated from the causal models is
the reciprocal relationship between perceptions of source expertise and
perceptions of argument strength. In the present study. the source
expertise manipulation exerted a direct impact on perceptions of
argument strength. Perceptions of argument strength. in turn. exerted
a strong impact on respondents' perceptions of perceived source
expertise. This multicollinearity of source and message perceptions
can be seen in other ELM studies. An extreme case occurs in Petty.
Cacioppo. and Goldman (1981) where the argument strength manipulation
exerted a stronger impact on perceived source expertise than did the
source expertise manipulation. This reciprocal relationship between
perceived argument strength and perceived source expertise is troubling
when one considers that these two variables are generally considered to
be predictive of attitudes under different sets of conditions.
Specifically. message cues and source cues. alleged to be predictive of
attitudes under conditions of central and peripheral processing. may
not be independent factors.

In addition to the causal modeling results. the present study
investigated attitude-behavior consistency and the role of individual
differences in message processing. Petty and Cacioppo (1981. 1986)
posit that the increased cognitive processing performed in the central
route to persuasion should lead to greater consistency between
attitudes and behaviors. Results of relevant analyses were. generally.
not consistent with this position. Increasing perceived involvement
did not increase attitude-behavior consistency. While the high

perceived involvement group exhibited the greatest amount of

91

attitude-behavior consistency. the low perceived and moderate perceived
involvement groups did not differ significantly from one another.

These findings are similar to those reported by Sivacek and Crano
(1982) where low and moderate vested interest groups did not differ
from one another in attitude-behavior consistency and were less
consistent than the high vested interest groups. Sivacek and Crano
(1982) posit that vested interest may act as a threshold: When
respondents exceed the threshold. attitudes predict behaviors; when
respondents are below the threshold. attitudes do not predict
behaviors.

Attitude-behavior consistency was also investigated in terms of
individual differences. Petty and Cacioppo (1986) argue that when
attitude change is generated by considerable message elaboration.
attitudes should be consistent with behaviors. Involvement in the
topic of the message is one variable expected to generate message
elaboration. but other personality variables have been identified as
performing a similar function. Two of these. need for cognition and
argumentativeness. were investigated here. In these analyses. only the
avoidance dimension of the argumentativeness scale moderated the
relationship between attitudes and behaviors. High avoidance
respondents (i.e.. persons who tend to avoid getting into arguments)
exhibited less consistency between their attitudes and behaviors than
did low avoidance respondents.

Taken as a whole. the findings concerning the role of need for
cognition in message processing are not consistent with predictions

(e.g.. Cacioppo et al.. 1983). First. the need for cognition scale

92

suffered from measurement error. The original 18-item scale failed to
fit a unidimensional model. and when factor analyzed. a reduced scale
of only six items was found to be unidimensional. The reliability of
the reduced scale (a=.76) proved to be approximately equal to the
reliability of the original. l8-item. scale (m:.82).

Need for cognition has been used primarily to predict the extent
of message processing a person will exert in evaluating a persuasive
message. In the present investigation. need for cognition. as well as
the subscales of the argumentativeness scale. failed to discriminate
between central and peripheral processing. Moreover. as noted above.
need for cognition did not mediate attitude-behavior consistency.

Critics may suggest that reduction of the 18-item Need for
Cognition scale to six items changes the scale to such an extent that
it is now measuring a construct far from what was initially intended.
In anticipation of such a criticism. all analyses were performed on
both the revised scale and on the original 18-item original. In no
case were the differences in results substantive or significant.
Conclusion;

This study began with a summary of Stiff's discussion of the ELM
as either a serial or parallel information processing model. It was
noted that the ELM makes markedly different predictions depending on
whether one interprets it as a serial or parallel model. Although not
adequate in distinguishing humans as serial or parallel information
processors. the results of this study are relevant to the overall
explanatory adequacy of the serial or parallel positions.

Results on perception of message information clearly indicate that

93

persons perceive both central and peripheral cues across all levels of
involvement. Thus. it is clear that persons are able to accurately
perceive both source and message cues although it cannot be determined
whether these perceptions occurred serially or in parallel.

Findings concerning the processing of message information indicate
that persons use exclusively argument strength information in
determining their post-message attitudes. Across all levels of
perceived involvement. perceived argument strength exerted an impact on
attitudes. Moreover. the causal model presented in Figure 8 indicates
that perceived argument strength has both a direct and indirect impact
on attitudes. Since the impact ofperceived argument strength on
attitudes is mediated by message cognitions. it seems clear that only
message strength cues were being processed when it came time to
determine attitudes.

The present results clearly indicate that while respondents
accurately pgrceived source and message information only message
processing was performed. Thus. message processing at all levels of
involvement occurs consistent with Petty and Cacioppo's (1981. 1986)
central route to persuasion. with perceived argument strength and
message cognitions exerting an influence on attitudes but with neither
perceived source expertise nor source cognitions impacting attitudes.
The primary question remaining is 28! these results are consistent with
central processing. rather than replicating the findings of Petty.
Cacioppo. and Goldman (1981).

The manipulations of involvement. source expertise. and argument

strength were developed so as to be consistent with previous studies.

94

specifically Petty. Cacioppo. and Goldman (1981). The ELM provides no
assistance in attempting to describe 22! respondents should be
processing exclusively via the central route in the present
investigation and via both the central and peripheral routes in
previous investigations. In this sense. the ELM is merely a
descriptive. rather than predictive model..

One variable that can explain the message processing differences
observed here is message modality or channel. Chaiken (1987) argues
that ”any factor in a persuasion setting that increases either the
salience or vividness of extrinsic [i.e.. peripheral] cues may enhance
their persuasive impact” (p. 23). Thus. any manipulation that makes a
set of persuasion cues more salient should increase the persuasive
impact of that cue. In an audiotape or videotape presentation. the
message source should be the more salient cue. leading to an increase
in source (vs. message) effects. If the message were complex. the
salience of the source could interfere with the processing of the
persuasive message. particularly under conditions of low motivation
(Petty 8 Cacioppo. 1986).

When the message is written. on the other hand. the source is a
much less salient persuasive force. Furthermore. if it is complex. the
written message allows recipients sufficient time to digest arguments
at their own pace. Consequently. one would expect the written message
containing the more salient set of cues to have a strong persuasive
impact. Petty and Cacioppo (1986) agree that ”given sufficient
motivation to process an advocacy. print should generally enhance

agreement to strong but reduce agreement to weak messages” (p. 77).

95

Thus. if one considers the EL! as a competition model--i.e.. a
model where sets of cues compete with one another for their cognitive
"time”--message modality or channel should be a potentially strong
moderator of the route to persuasion. Under conditions of high
involvement. respondents should be motivated to process central cues.
to 'fight through' the salient source information (if any) and to
concentrate on message arguments. Under conditions of low involvement.
on the other hand. respondents should be motivated to process only the
most salient set of cues. If the message is video or audio.
characteristics of the source (e.g.. credibility. attractiveness.
similarity) will be most salient; if the message is written. message
arguments will be the most salient cues.

Such an analysis coincides with prior studies investigating the
ELM. The vast majority of these studies have used either audio or
video taped presentations (e.g.. Petty. Cacioppo. and Goldman. 1981;
Petty. Cacioppo. and Heesacker. 1981; Petty and Cacioppo. 1979a; Petty
et al.. 1976; Petty. Hells. Heesacker. Brook. 8 Cacioppo. 1983). where
source effects at low levels of involvement are expected. Moreover.
studies that have failed to support the ELH's hypotheses have typically
used written messages (e.g.. Burnkrant 8 Howard. 1984).

A final conclusion relates to the extent to which results of the
present study support or refute the ELM. Stiff (1986) used Kahneman's
EC! to develop an alternative interpretation of the ELM. Both Stiff's
and Petty and Cacioppo’s interpretations were used to develop specific
predictions regarding the interaction of perceived involvement.

perceived source expertise. and perceived argument strength on source

96

and message cognitions. None of the predicted interactions were
consistent with the obtained data. The implication of these results is
that the predictions of the BLH. including the most recent predictions
made by Petty. Kasmer. Haugtvedt. and Cacioppo (1987). are inconsistent
with the present results.

The model most consistent with the data was the one presented in
Figure 8. In this model. perceived argument strength was the only
predictor of message cognitions while perceived source expertise was
the only predictor of source cognitions. Both message cognitions and
perceived argument strength. in turn. influenced behavior. The only
important impact of perceived involvement in the entire model consisted
of an interaction with post-test attitudes on behavioral intentions:

To report an intention to behave. respondents must be highly involved
in the message and be opposed to the message recommendation.

Results on individual differences. both in terms of affecting
message processing and in terms of attitude-behavior consistency. were
not consistent with the BLH's predictions. First. need for cognition
did not influence how persons processed message information. More
specifically. need for cognition did not mediate the perception of
argument strength or source expertise nor the impact that argument
strength or source expertise exerted on attitudes. Finally. need for
cognition did not influence the consistency between the nature of
message cognitions and post-message attitudes. nor did it mediate the

consistency between attitudes and behaviors.

97

Methodological Differences and Study Limitations

 

Results of the present investigation are inconsistent with the
accumulated data from a large body of literature. There are several
significant differences between the present and previous investigation
that may partially account for these discrepancies. Several of these
differences bear repeating. In the present study. perceptions of
source and message variables were investigated while previous ELM
investigations analyzed the impacts of the manipulations themselves.
There are two important reasons for analysing perceptions of
manipulated variables. First. in a majority of investigations.
including the present study. the manipulation of variables has been
weak. Thus. there may not be strong systematic differences in the
perceptions of message and source variables across experimental
groups. Second. even if the interactions were strong. it is
theoretically more interesting to investigate the perceptions of source
and message variables instilled in the audience member by the
persuasive message.

A second related difference between the present study and previous
ELM investigations concerns the measurement and analysis of
cognitions. While the present investigation analyzed the valence
rather than the number of message cognitions. previous investigations
have simply analysed number of cognitions. even though the ELM makes
clear predictions regarding the valence of source and message
processing. Moreover. previous ELM investigations have failed to
separate source cognitions from message cognitions. The present study

analysed these factors separately. Again. this is an important

98

difference since the ELM makes specific predictions regarding the
influence on bgth_source and message cognitions. The combination of
source and message cognitions into a single measure simply does not
allow an adequate test of the model.

The final difference between the present investigation and
previous ELM studies lies in the use of causal modeling techniques to
analyze data. Although the ELM presents moderately clear predictions
of the causal relationships between variables in the model. previous
investigations have been inadequate to test these predicted
relationships. The use of causal modeling in the present study has
eliminated this inadequacy.

An important limitations of the present study relates to the
involvement manipulation. Although it is clear that varying levels of
perceived involvement were generated. the variation in involvement was
not systematic. Despite the fact that pilot test data indicated
differences between groups. the low and moderate experimental
involvement groups did not differ from each other. The weakness of the
involvement manipulation may have been a function of an unexpectedly
large percentage of seniors (68$) in the final sample. since it is
unlikely that seniors would feel as involved in a policy change that
would probably not occur until they had graduated.

A second limitation of the present study lies in the use of a
behavioral intention measure. There are specifiable conditions where
it is known that behavioral intentions are consistent with actual
behavior. It would be instructive. however. to build a model that

includes attitudes. intention to act. and respondents' actual behavior.

99

Directions for Future Research.

'Several directions for future research are suggested by the
present outcomes. First. the causal model developed here should be
replicated. and the extent to which the model bears up over repeated
tests in different settings and with different respondent populations
should be examined. Put differently. the extent to which results of
the present study are unique to the specific circumstances of the study
need be investigated.

In addition. different central. peripheral. and moderator cues
should be investigated and placed into the causal models. The ELM
should be capable of describing the impact of several different
variables within a single causal framework. If different variables
create grossly different causal processes. this could be taken as
evidence against the applicability of the ELM.

The importance of analyzing future data using causal modeling
techniques cannot be overstated. Several interesting results. not
obtainable by traditional analytical techniques were were uncovered
here. In addition. the most parsimonious means of testing the
predictions of a cognitive response model is to observe and
simultaneously analyze all relevant cognitive responses.

Second. as noted above. future studies should include measures of
both behavioral intention agg_actual behavior. Given the ELM. the
relationship between the amount of cognitive processing and attitude-
behavior consistency is an important one. and very few studies have
been performed using actual behavioral measures.

Similarly. the ELM posits that attitude change generated through

100

the central route is more enduring and resistant to counter-persuasion
than change generated through the peripheral route. Therefore. future
studies should measure attitudes both immediately after message
presentation and after some period of time. If such a design were
implemented. a resistance to counter-persuasion manipulation could be
added to the delayed measurement of attitudes.

Finally. the nature of involvement as a variable should be further
examined. Considerable difficulty has been encountered in attempting
to manipulate more than two levels of perceived involvement. both in
the present investigation and elsewhere (J.P. Dillard. personal
communication. 31 December. 1987). Little is known regarding the
nature of the variable of involvement even though it is a popular entry
in the persuasion literature. Alternative methods of generating
varying levels of perceived involvement should be developed. One
possibility would be to manipulate the external rewards and punishments
associated with particular messages. This approach eliminates the
present problem of having messages of different levels of involvement
on different topics.

Rhatever method is used to develop different levels of
involvement. manipulations that systematically create three or more
levels of involvement need be developed. There are two reasons for
this claim. As stated above. Stiff's (1986) data indicated a
curvilinear relationship between source credibility and attitude change
across levels of involvement. Testing such a curvilinear prediction
requires at least three levels of involvement. The second reason for

such a manipulation is to test the difference between truly moderate

101

involvement and ”moderate" involvement where respondents are told
nothing about the relevance of the message (Heesacker. Petty. and
Cacioppo. 1983; Puckett. Petty. Cacioppo. and Fisher. 1983). It seems
as if there may be important differences between a situation where a
respondent has no basis to make a judgement of message relevance and a
situation where respondents feel the message is neither highly relevant
nor highly irrelevant.

Finally. the role played by message modality on the elaboration
process should be investigated further. The explanation for the
present findings that information processing reflects the central route
regardless of the level of involvement due to the modality of the
message generates several intriguing research questions. Are written
messages always processed centrally? How can strong source effects be
generated using print messages? How can strong message effects be
generated using a video presentation under conditions of low
involvement? These and several other questions are left unanswered

given our current state of knowledge in this area.

REFERENCES

REFERENCES

Allyn. J. 8 Festinger. L. (196l). The effectiveness of unanticipated
persuasive communication. Journal of Abnormal and Social
Psychology. §§, 35-40.

Anderson. M. H. (1974). Information integration theory: A brief
survey. In D. Krantz. R. Atkinson. 8 P. Suppes. (Eds.).
Contempgrary developments in mathematical pgychology (Vol. 2. pp.
236-305). San Fransisco: M. M.

Freeman.

Axsom. D.. Yates. S.. 6 Chaiken. S. (1987). Audience response as a

heuristic cue in persuasion. Journal of Personality and Social
Psychology. §§, 30-40.

Ben. D. J. (1972). Self-perception theory. In L. Berkowitz (Ed.).
Advances in expgrimental social pgyghology (Vbl. 6; pp. 1-62).

MY: Academic Press.

Bentler. P. M. 8 Bonett. D. G. (1980). Significance tests and goodness
of fit in analysis of covariance structures. Psyghological
Bulletin. 88, 588-606.

Boster. F. J. 8 Mayer. M. E. (1984). Choice shifts: Argument
qualities or social comparisons. In R. Bostrom (Ed.).

Communication yearbook 8 (pp. 393-410). Beverly Hills. CA: Sage.

102

103

Burnkrant. R. E. 8 Howard. D. J. (1984). Effects of the use if
introductory rhetorical questions versus statements on information
processing. Jgggpgl of Personality and Social Psychology. 11.
1218-1230.

Cacioppo. J. T. (1979). The effects of exogenous changes in heart rate
on the cognitive responses and their relationships to behavior.
WM: 31- 480-489-

Cacioppo. J. T.. Markins. 8. G.. 8 Petty. R. E. (1981). The nature of

attitudes and cognitive responses and their relationship to

behavior. In R. Petty. T. Ostrom. 8 T. Brock (Eds.). t ve
pg;pgpggg_1p_pgggpggigp (pp. 31-54). Millsdale. NJ: Lawrence
Erlbaum.

Cacioppo. J. T. 8 Petty. R. E. (1979). Effects of message repetition
and position on cognitive responses. recall. and persuasion.

J C it 1 l a a: 97-109.
Cacioppo. J. T. 8 Petty. R. E. (1982). The need for cognition.

nggggl g; 2gpggnglity and §gcig1 Psycpglggy. 1;. 116-131.
Cacioppo. J. T. 8 Petty. R. E. (1985). Central and peripheral routes
to persuasion: The role of message repetition. In A. Mitchell 8
L. Alwitt (Eds.). 0 ice cesse a a v
(pp. 91-111). Millsdale. NJ: Erlbaum.
Cacioppo. J. T. 8 Petty. R. E. (1987). Stalking rudimentary processes
in social influence: A psychophysiological approach. In M. Zanna.
J. Olson. 8 C. Hendricks (Eds.). So : Ontario

gygpgglgg (Vol. 5; pp. 41-74). Millsdale. MJ.: Erlbaum.

104

Cacioppo. J. T.. Petty. R. E.. 8 Morris. K. (1983). Effects of need
for cognition on message evaluation. recall. and persuasion.
Journai of Persogglity ppd Social Psycholggy. 1;, 805-818.

Chaiken. 8. (1980). Heuristic versus systematic information processing
and the use of source versus message cues in persuasion. Jourppl
of Pergopgiity and Sogial Psygpglggy. 33, 752-766.

Chaiken. S. (1987). The heuristic model of persuasion. In M. Zanna.
J. Olson. 8 C. Herman (Eds.). al nfluenc : The O o
gygpggipg (Vol. 5; pp. 3-39). Millsdale. NJ: Erlbaum.

Chaiken. S. 8 Eagly. A. M. (1976). Communication modality as a
determinant of message persuasiveness and message
comprehensibility. Journal of Eeysopglity gng Social ngghology.
§§, 605-614.

Chaiken. S. 8 Stangor. C. (1987). Attitudes and attitude change.

In M. R. Rosenzweig 8 L. R. Porter (Eds.). Apnual Rgvigw g;
ggyghgig y (Vol. 38: pp.575-630). Palo Alto. CA: Annual Reviews.
Inc..

Cialdini. R. B.. Levy. A.. Herman. P.. Koslowski. L.. 8 Petty. R. E.
(1976). Elastic shifts in opinion: Determinants of direction and
dunbilitr- W: as. 563-
672.

Cialdini. R. B. 8 Petty. R. E. (1981). Anticipatory opinion effects.

In R. Petty. T. Ostrom. 8 T. Brock (Eds.). Cogniiive rggpgpggg ip
pggguagign (pp. 217-235). Millsdale. NJ: Erlbaum.

105

Cohen. A. R.. Stotland. E.. 8 Rolfe. D. M. (1955). An experimental
investigation of need for cognition. Journai of Abnormal and
§ggigl Psychoiogy. 5;. 291-294.

Davidson. A. R. Yantis. 8.. Morwood. M.. 8 Montano. D. E. (1985).

Amount of information about the attitude object and attitude-

behavior consistency. Jgurnal oi Eggsonality egg §gcial
ngghgiggy. 12. 1184-1198.

Eagly. A. H. (1974). Comprehensibility of persuasive arguments as a
determinant of opinion change. Jourpgi g; Ppgggnality gpg §gcial

211211212512: 2.9.: 758-773-
Eagly. A. M. 8 Chaiken. S. (1984). Cognitive theories in persuasion.

In L. Berkowits (Ed.). van e i nta s a 010
(Vol. 17. pp. 267-359). MY: Academic Press.
Ferguson. 6. A. (1982). tat sti l n s h n
_ggpgg1igp,(5th ed). MY: McGraw-Mill.
Festinger. L. (1950). A theory of social comparison. flgpgn Rglations.
_1. 117-140.

Markins. S. G. 8 Petty. R. E. (1981). The multiple source effect in
persuasion: The effects of distraction. Personality ang §ggigi
{gygggiggy_flyiigiigp 2, 627-635.

Harkins. S. G. 8 Petty. R. E. (1982). The effects of task difficulty

and task uniqueness on social loafing. our of Per n

W: 43.: 1214-1229-

106

Heesacker. M.. Petty. R. E. 8 Cacioppo. J. T. (1983). Field dependence

and attitude change: Source credibility can alter persuasion by

affecting message-relevant thinking. Jgggppi of Personality. gi.

653-666.
Hovland. C. I.. Janis. I. L.. and Kelley. H. H. (1953). Commupigpiion
r ua i n: Ps cal s dies 0 in s. New

Haven. CT: Yale University Press.
Hunter. J. E. (1980). Factor analysis. In P. R. Monge 8 J. M.

Cappella (Eds.). Multivpgiatg teghnigugg in hypgp coggynigatigp
resegrch. MY: Academic Press.

Hunter. J. E. (1982). Error pi pgasurement: An oygrvigy. Unpublished
manuscript. Department of Psychology. Michigan State University.
East Lansing. MI..
Hunter. J. R.. Cohen. S. M.. 8 Micol. T. S. (1982). mm
o n to do co re 1 nal anal is nclu n t n 1
a t e t r a nal is.
Unpublished manuscript. Department of Psychology. Michigan State
University. East Lansing.
Infante. D. A. 8 Rancor. A. S. (1982). A conceptualization and measure
of argumentativeness. Jgggpal gi gggggnality Agseggpgpt. 59. 72-
80.
Jareskog. K. G. 8 Sorbom. D. (1984). I V : An
t i nshi he .
Mooresville. IM: Scientific Software.
Kahneman. D. (1973). Attention ggg giggri. Englewood Cliffs. NJ:

Prentice-Hall.

107

Lindsay. P. H. (1970). Multi-channel processing in perception. In D.
I. Mostfosky (Ed.). Attention: Contem ra r and anal is
(pp. 149-171). MY: Appleton-Century-Crofts.

Lindsay. P. H.. Cuddy. L. L. 8 Tulving. E. (1965). Absolute judgements

of simultaneously presented visual and auditory stimuli.

W: Z: 211-212-
McCullough. J. L. 8 Ostrom. T. M. (1974). Repetition of highly similar
messages. Journal of Appiied §ggigi ggyghoiggy. §2. 395-397.

McGuire. W. J. (1972). Attitude change: The information-processing
paradigm. In C. G. McClintock (Ed.). 1 nt i
pgygpgiggy,(pp. 108-141). MY: Holt. Rinehart. and Winston.

Miller. M. (1965). Involvement and dogmatism as inhibitors of attitude

change. Jogipgi of Experimental Social Psyghology. 1. 121-132.

Park. C. W. 8 Young. S. M. (1986). Consumer response to television

commercials: The impact of involvement and background music on

brand attitude formation. ggpgppi_gi_§gpkgiipg_§gyggggh. 23. ll-
24.
Pedhasur. E. J. (1982). a in v
W (2nd ed.). MY: Holt.

Rinehart. 8 Winston.
Petty. R. E. 8 Cacioppo. J. T. (1979a). Issue involvement can increase
or decrease persuasion by enhancing message-relevant cognitive

responses. Joginai oi Egypopgliiy gpg Sociai gyyghgiggy. 31. 1915-
1926.

108

Petty. R. E. 8 Cacioppo. J. T. (1979b). Effects of forewarning of

persuasive intent and involvement on cognitive responses and

persuasion. Persogglity and Social Psychology Bulletin. g. 173-

176.

Petty. R. E. 8 Cacioppo. J. T. (1981). Aitituggg ppg pgrggasign:
Qiggyic and contegpggpgy gpproaghes. Dubuque. IA: William C.
Brown.

Petty. R. E. 8 Cacioppo. J. T. (1984). The effects of involvement on

response to argument quantity and quality: Central and peripheral

routes to persuasion. Journal gi Pgiygggiity and figgiai
Psyghglogy. 39, 69-81.
Petty. R. E. 8 Cacioppo. J. T. (1986). Communigpiign apg ppigpppiop.

MY: Springer-Verlag.

Petty. R. E.. Cacioppo. J. T.. 8 Goldman. R. (1981). Personal
involvement as a determinant of argument-based persuasion.
Joggppl oi gggggpgiiiy pnd §gcial ggychgiogy. 11. 847-855.

Petty. R. E.. Cacioppo. J. T.. 8 Heesacker. M. (1981). Effects of
rhetorical questions on persuasion: A cognitive response
approach. Jggpggi gi ggrggpality gpd Sggigi ggyghplogy. 19. 432-
440.

Petty. R. 8.. Cacioppo. J. T.. Kasmer. J. A.. 8 Haugtvedt. C. P.
(1987). A.reply to Stiff and Boster. Qggggpiggpigp_!gpggggpp§,
Q1. 257-263.

Petty. R. B.. Cacioppo. J. T.. 8 Schuman. D. (1983). Central and

peripheral routes to advertising effectiveness: The moderating

role of involvement. Jguygpi gi Consugg; Resggign. 19, 134-148.

109

Petty. R. B.. Harkins. S. G.. Williams. R. D. 8 Latane. B. (1977). The
effects of group size on cognitive effort and evaluation.
Personality ang §gcigl ggyghgiggy fiyiigtip. 3. 579-582.

Petty. R. 8.. Kasmer. J. A.. Haugtvedt. C. P. 8 Cacioppo. J. T. (1987).
Source and message effects in persuasion: A reply to Stiff's
critique of the Elaboration Likelihood Model. Cgmmupicatigp
Mggggggphg. 51. 233-249.

Petty. R. E.. Wells. G. L.. 8 Brock. T. C. (1976). Distraction can
enhance or reduce yielding to propaganda: Thought disrupting
versus effort justification. Jouinal oi Pgipgggiity gnd Sogiai
Psychology. 31. 874-884.

Petty. R. 3.. Wells. 6. L.. Heesacker. M.. Brock. T.. 8 Cacioppo. J. T.
(1983). The effects of recipient posture on persuasion: A
cognitive response analysis. it c
Bulleiin. 2. 209-222.

Puckett. J.. Petty. R. B.. Cacioppo. J. B.. 8 Fisher. D. (1983). The

relative impact of age and attractiveness stereotypes on

persuasion. Jggppgi_gi_§gpggigiggy. 38. 340-343.

Regan. D. T. 8 Cheng. J. B. (1973). Distraction and attitude change: A
resolution. Jopgnpi gi Egpgiimeptal §ggigl ngcholggy. 2. 138-
147.

Sherif. M. 8 Hovland. C. I. (1961). cia ud ° simi t n
OWN- New Haven-

CT: Yale University Press.

110

Sherif. M. 8 Sherif. C. W. (1967). Attitude as the individual's own
categories: The social-judgement involvement approach to attitude
and attitude change. In C. Sherif 8 M. Sherif (Eds.). Atiitude.
ego-invglvement. and changg (pp. 105-139). MY: Wiley.

Sherman. S. J. (1987). Cognitive processes in the formation. change.

and expression of attitudes. In M. Zanna. J. Olson. 8 C. Herman

(Eds.). Social influgpge: Thg Ontgrio sygpgsium (Vol. 5. pp. 75-
106). Hillsdale. NJ: Erlbaum.

Sivacek. J. 8 Crano. W. D. (1982). Vested interest as a moderator of
attitude-behavior consistency. Joyippl of Personglity gpd §ggial
Psychology. 53. 210-221.

Stiff. J. B. (1986). Cognitive processing of persuasive message cues:
A meta-analytic review of the effects of supporting information on
attitudes. Communication Mgnggraphs. 53. 75-89.

Stiff. J. B.. 8 Boster. F. J. (1987). Cognitive processing: Additional
thoughts and a reply to Petty. Kasmer. Haugtvedt. and Cacioppo.
Communicatiop Mopggiaphs. 53. 250-256.

Treisman. A. M. (1970). Perception and recall of simultaneous speech

stimuli. Apia ggygpglogica. 33. 132-148.
Treisman. A. M. 8 Fearnley. J. S. (1971). Can stimulus speech be

classified in parallel? Pgrggption and Psychophygics. i9. 1-7.
Treisman. A. M. 8 Geffen. G. (1967). Selective attention: Perception

or response? anitgrly ggurpgi gi §gpg§imental ggychoiogy. i2. 1-
l7.

111

Treisman. A. M. 8 Riley. J. G. (1969). Is selective attention
selective perception or selective response? A further test.
Journal of Experimental Psychology. i9. 27-34.

Wright. P. L. (1981). Cognitive responses to mass media advocacy. In
R. E. Petty. T. Ostrom. 8 T. Brock. (Eds.). Cognitive responses in
pgrguasion (pp.263-282). Hillsdale. NJ: Erlbaum.

Yalch. R. F. 8 Elmore-Yalch. R. (1984). The effect of numbers on the

route to persuasion. Jgurnal of Copgupgi Researgp. ii. 522-527.

APPENDIX

APPENDIX

Strong argument message

The Mational Scholarship Achievement Board recently revealed the
results of a five-year study conducted on the effectiveness of
comprehensive exams at Duke University. The results of the study
showed that since the comprehensive exam has been introduced at Duke.
the grade point average of undergraduates has increased by 31%. At
comparable schools without the exams. grades increased by only 8% over
the same period. The prospect of a comprehensive exam clearly seems to
be effective in challenging students to work harder and faculty to
teach more effectively. It is likely that the benefits observed at
Duke University could also be observed at other universities that adopt
the exam policy.

One aspect of the comprehensive exam requirement that students at
the schools where it has been tried seem to like is that all regular
final examinations for seniors are typically eliminated. This
elimination of final exams in all courses for seniors allows them to
better integrate and think about the material in their major area just
prior to graduation rather than ”wasting” a lot of time cramming to
pass tests in courses in which they are really not interested.
Students presently have to take too many courses in subject that are
irrelevant to their career plans. The comprehensive exam places
somewhat greater emphasis on the student's major and allows greater
concentration on the material that the student feels is most relevant.

112

113

Graduate schools and law and medical schools are beginning to show
clear and significant preferences for students who received their
undergraduate degrees from institutions with comprehensive exams. As
the Dean of the Harvard Business School said: ”Although Harvard has not
and will not discriminate on the basis of race or sex. we do show a
strong preference for applicants who have demonstrated their expertise
in an area of study by passing a comprehensive exam at the
undergraduate level.” Admissions officers of law. medical. and
graduate schools have also endorsed the comprehensive policy and
indicated that students at schools without exams would be a a
significant disadvantage in the very near future. Thus. the
institution of comprehensive exams will be an aid to those who seek

admission to graduate and professional schools after graduation.

Weak argument message

A member of the Board of Regents has stated publicly that his
brother had to take a comprehensive exam while in college and now he is
manager of a large restaurant. He indicated that he realised the value
of the exams since their father was a migrant worker who didn't even
finish high school. He also indicated that the university has received
several letters from parents in support of the exam. In fact. 4 of the
6 parents who wrote in thought that the exams were an excellent idea.
Also. the prestigious Mational Accrediting Board of Higher Educations

seeks input from parents as well as students. faculty. and

114

administrators when evaluating a university. Since most parents
contribute financially to their child's education and also favor the
exams. the university should institute them. This would show that the
university is willing to listen to and follow the parents' wishes over
those of students and faculty who may simply fear the work involved in
comprehensive exams.

Faculty members at universities with the comprehensive exams who
were interviewed by researchers from the Carnegie Commission on Higher
Education revealed that they liked the exams because it reduced the
number of test they felt they had to give in the classes knowing that
students would still face one ultimate test of their knowledge in the
comprehensive exam. A study at Motre Dame showed that this reduction
in regular course test saved enough paper to cover the cost of painting
two classrooms.

Data from the University of Virginia show that some students favor
the senior comprehensive exam policy. For example. one faculty member
asked his son to survey his fellow students at the school since it
recently instituted the exams. Over 558 of his son's friends agreed
that in principle. the exams would be beneficial. Of course. they
didn't all agree but the fact that most did proves that undergraduates
want the exams. As Saul Siegel. a student whose father is a vice-
president of IBM wrote in the school newspaper: ”The history of the
exams can be traced to the ancient Greeks. If comprehensive exams were
to be instituted. we could feel the pleasure at following traditions
begun by Plato and Aristotle. Even if there were no other benefits of

the exams. it would be worth it just to follow tradition."

NOTES

NOTES
1 Chaiken (1987; Axsom. Yates. 8 Chaiken. 1987; Chaiken 8 Stangor.
1987; Eagly 8 Chaiken. 1984) argues that the term 'heuristics' should
not be used synonymously with term 'peripheral processing' asserting
that heuristics represent a more specific construct. Data attempting
to identify the specific use of heuristics (e.g.. Chaiken. 1980.
experiment 2; Chaiken. 1987. pp. 23-30). however. have been
disappointing.
2 A few studies have been performed. however. with 'moderate'
involvement conditions. where recipients are simply told nothing about
the nature of the relevance of the message (e.g.. Heesacker. Petty. 8
Cacioppo. 1983; Puckett. Petty. Cacioppo. 8 Fisher. 1983). It is not
clear. however. whether recipients in such a condition would react in a
manner similar to those in a 'truly' moderate involvement condition
(one which attempts to generate truly modgrate levels of involvement).
3 There is no evidence of correcting the manipulation check
correlations for attenuation due to measurement error in either the
fear appeal literature meta-analyzed by Boster and Mongeau (1984) or in
the ELM investigations. Therefore. the manipulation check correlations
reported are likely to be underestimates of the parameters. After

being corrected for attenuation. however. these correlations are still

likely to be well under 1.0.

115

116

4 Although Petty and Cacioppo have laboriously identified several

variables which may fit in the the categories of central cue.
peripheral cue. or moderator cue the variables argument strength.
source expertise. and issue involvement will be used as exemplars of
these cues so that the description of the models will be consistent
with the testing of these models.

5 The models depicted in Figures 4 and 5 give a visual representation
of the central and peripheral routes to persuasion. Given such a
presentation. they are similar to the model of heuristic and central
processing presented by Sherman (1987).

6 When three outliers are eliminated from the moderate involvement
condition. the linear trend becomes stronger (F=62.89; df=1.60; p(.OOl;
r=.71 and .77 when corrected for attenuation due to measurement

error). The linear trend in this case describes 96 percent of the sums
of squared for the involvement manipulation.

7 These analyses were performed on cell sample sizes which varied
greatly. The variation in cell sizes in this case is a function of the

covariation of perceived argument strength and perceived source

expertise (r=.63).

TABLES

117

Table 1. Contrasts used to describe the serial interaction between
perceived involvement and the source expertise manipulation
on perceived source expertise in model 1.

Perceived Involvement

Low Moderate High
High +2 +1 0
Manipulated
Source
Expertise

Low -2 -l 0

118

Table 2. Contrasts used to test the serial interaction between
perceived involvement and the argument strength
manipulation on perceived argument strength in model 1.

Perceived Involvement

Low Moderate High

High 0 +1 +2
Manipulated
Argument
Strength

119

Table 3. Contrasts used to test the parallel interaction between
perceived involvement and the source expertise manipulation
on perceived source expertise in model 2.

Perceived Involvement
Low Moderate High

High -1 +5 -1
Manipulated
Source
Expertise

Low -1 -1 -1

120

Table 4. Contrasts used to test the parallel interaction between
perceived involvement and the argument strength manipulation on
perceived argument strength in model 2.

Perceived Involvement
Low Moderate High

High -1 +2 +2
Manipulated
Argument
Strength
Low -1 -1 -l

121

Contrasts used to test the interaction between perceived

involvement. perceived source expertise. and perceived
argument strength on source cognitions.

Table 5.
Low
Perceived
Source
Expertise
Low High
High -1 +1
Perceived
Argument
Strength

Low -1 +1

Moderate

Perceived

Source

Expertise
Low High

-1/2 +1/2

-1/2 +1/2

Perceived Involvement

High

Perceived

Source

Expertise
Low High

0 0

122

A table of means. standard deviations. and number of

respondents for each of the four perceived involvement

Table 6a.
items (factor 501).
Item

14. The outcome of this proposal is
important to me.

15. This proposal represents a
significant change for students
currently enrolled at MSU.

16. This proposal is relevant to gy
graduation at MSU.

17. This proposal will have pg impact

on my academic life at MSU.

Mean

3.29

3.49

2.34

3.39

S. D.

1.07

1.32

1.28

1.22

227

227

227

227

Factor
Loadings

.64

.52

.89

.59

Table 6b.

123

Test of internal consistency for the perceived involvement
measure (factor 501). Correlations between items in the
factor appear below the diagonal. communalities appear on
the diagonal. and deviations from predicted correlations
generated by factor loadings appear above the diagonal.

1 2 3 4
1 .42 .05 -.02 -.02
2 .38 28 -.01 -.04
3 .55 .45 .78 .05

124

Table 6c. A test of parallelism for the perceived involvement measure
(factor 501). Item-factor correlations for factor 501 on
all other factors are presented.

501
1 2 3 4
502 -.04 .10 .10 .10
503 .01 .03 .10 .10
504 .03 .00 .09 .08
505 -.06 -.06 -.01 .02
506 .24 .26 .33 .30
507 .05 -.11 .03 .08
508 -.04 .03 -.08 -.06
509 .09 .10 .04 .02

Where:

501: Perceived Involvement

502: Perceived Argument Strength
503: Perceived Source Expertise
504: Pre-Message Attitude

505: Post-Message Attitudes

506= Behavioral Intention

507: Meed for Cognition

508: Argument Avoidance

509: Argument Enjoyment

125

A table of means. standard deviations. number of

observations. and factor loadings for each of the four
perceived argument strength items (factor 502).

Table 73.
Item

1. The arguments contained in the
message were strong.

2. The arguments contained in the
message were convincing.

3. The arguments contained in the
message were persuasive.

4. The arguments contained in the

message were reasonable.

Mean

2.77

2.67

2.74

3.04

S. D.

1.07

1.03

0.99

1.04

223

223

223

223

Factor
Loadings

.77

.88

.84

.74

Table 7b.

126

Test of internal consistency for the perceived argument
strength measure (factor 502). Correlations between items
in the factor appear below the diagonal. communalities
appear on the diagonal. and deviations from predicted
correlations generated by factor loadings appear above the
diagonal.

1 2 3 4
1 .59 .00 .02 -.03
2 .68 .78 -.02 .03
3 .67 .72 .71 00

127

Table 7c. A test of parallelism for the perceived argument strength
measure (factor 502). Item-factor correlations for factor
502 on all other factors are presented.

502
1 2 3 4

501 .09 .11 .08 .04
503 .41 .49 .46 .52
504 -.03 .06 .02 .08
505 .26* .48 .38 .47
506 -.27 -.24 -.2o —.29
507 -.14 -.11 --.12 -.oa
508 -.12 -.03 —.13 .05
509 -.23 -.16 -.22 -.08

Where:

501: Perceived Involvement

502: Perceived Argument Strength
503: Perceived Source Expertise
504: Pre-Message Attitude

505: Post-Message Attitudes

506: Behavioral Intention

507: Need for Cognition

508: Argument Avoidance

509= Argument Enjoyment

* This correlation is not parallel with other correlations on this
factor.

Table 8a.

1.

4.

5.

8.

128

A table of means. standard deviations. number of

observations. and factor loadings for each of the eight

perceived source expertise items (factor 503).

Item

The author is highly informed on
the merits of comprehensive

examinations for college students.

The author of this message is

unaware of the issues surrounding

comprehensive examinations.

The author is an expert on this
topic.

The author is knowledgeable on
this topic.

The author has experience with
comprehensive examinations.

The author is qualified to speak
on this issue.

The author of the message is
incompgtent with regards to
comprehensive examinations.

The author is well read on this
issue.

Mean

2.63

3.26

2.35

3.02

2.77

2.76

2.89

2.93

S. D.

1.18

1.07

1.12

1.14

1.12

1.15

1.05

1.04

227

227

227

226

226

226

226

226

Factor
Loadings

.82

.75

.85

.88

.82

.89

.81

.82

Table 8b.

x
This correlation

factor.

Test of internal consistency for the perceived source
expertise measure (factor 503).

129

Correlations between items

in the factor appear below the diagonal. communalities

appear on the diagonal. and deviations from predicted

correlations generated by factor loadings appear above the

diagonal.

1 2 3 4 5 6 7
.66 .03 .05 .03 -.03* -.03 -.04
.65 .56 .oo .00 -.04 -.03 .00
.75 .64 .73 -.02 .02 .00 -.07
.75 .66 .73 .77 -.02 -.04 .04
.59 .59 .72 .70 .66 .06 .04
.70 .64 .76 .74 .79 .79 .03
.62 .61 .62 .75 .70 .75 .66
.67 .61 .70 .70 .65 .72 .69

is not consistent with other correlations on

e00

.00

.00

-.02

-.02

-.01

.03

.67

this

130

Table 8a. A test of parallelism for the perceived source expertise
measure (factor 503). Item-factor correlations for factor
503 on all other factors are presented.

503
1 2 3 4 5 6 7 a

501 .11 .05 .07 .13 -.02 .04 .13 .03
502 .57 .50 .46 .53 .35* .47 .47 .49
504 -.06 .02 .01 .07 .03 .03 .09 -.02
505 .27 .32 .22 .29 .29 .25 .23 .19
506 -. 21 -. 15 —. 12 —. 1o -. 06 -. 13 -.09 -. 13
507 -.14 -.06 -.1o -.09 -.01 -.04 -.04 -.01
508 -.19 .oo -.14 -.11 -.06 —.09 .01 .00
509 -.27 -.15 -.19 -.19 -.13 -.23 -.13 -.05

Where:

501= Perceived Involvement

502: Perceived Argument Strength
503: Perceived Source Expertise
504: Pre-Message Attitude

505: Post-Message Attitudes

506: Behavioral Intention

507: Need for Cognition

508: Argument Avoidance

509: Argument Enjoyment

* This correlation is not parallel with other correlations on this
factor.

131

Table 9a. A table of means. standard deviations. number of
observations. and factor loadings for each of the five pre-
message attitude items (factor 504).

Item Mean S.D. M Factor
Loading
1. Favorable-Unfavorable 4.44 1.57 179 .89
2. Good-Bad 4.13 1.70 177 .88
3. Wise-Foolish 4.53 1.31 179 .85
4. Beneficial-Harmful 4.74 1.30 179 .78

5. Satisfactory-Onsetisfactory 4.31 1.47 180 .82

Table 9b.

132

Test of internal consistency for the pre-messaqe attitude
measure (factor 504). Correlations between items in the
factor appear below the diagonal. communalities appear on
the diagonal. and deviations from predicted correlations
generated by factor loadings appear above the diagonal.

504

1 2 3' 4 5
.79 .06 .02 -.05 -.03
.84 .78 -.04 -.03 .02
.78 .71 .72 .04 -.02
.64 .66 .70 .61 .04
.70 .74 .68 .68 .68

133

Table 9c. A test of parallelism for the pre-message attitude measure
(factor 504). Item-factor correlations for factor 504 on
all other factors are presented.

1 2 3 4 5
501 .09 .09 -.01 .11 .05
502 .04 .06 -.10 .06 .11
503 .00 .06 -.05 .01 .09
505 .35 .31 .29 .39 .25
506 -.03 -.02 .01 .03 -.06
507 .02 .13 .05 .03 .14
508 .06 .15 .14 .03 .13
509 -.09 .02 .07 .05 -.11

Where:

501= Perceived Involvement

502: Perceived Argument Strength
503= Perceived Source Expertise
504: Pre-Message Attitude

505: Post-Message Attitudes

506: Behavioral Intention

507: Need for Cognition

508: Argument Avoidance

509: Argument Enjoyment

134

Table 10a. A table of means. standard deviations. number of
observations. and factor loadings for each of the five post-

message attitude items (factor 505).

Factor
Item Mean S.D. M Loadings
1. Favorable-Unfavorable 3.98 1.72 227 .89
2. Good-Bad 4.42 1.61 227 .90
3. Wise-Foolish 4.34 1.50 227 .85
4. Beneficial-Harmful 4.63 1.48 227 .81
5. Satisfactory-Unsatisfactory 4.03 1.62 227 .83

Table

* This deviation is greater than

alone.

10b.

135

Test of internal consistency for the post-message attitude
Correlations between items in the
factor appear below the diagonal. communalities appear on
the diagonal. and deviations from predicted correlations
generated by factor loadings appear above the diagonal.

measure (factor 505).

5

1

.79

.83

.70

.70

.77

2

.03

.80

.75

.70

.76

3 4

-.O6 -.02

-.01 -. 03
.72 .07”
.76 .66
.70 .65

would be expected

.03
.01
-.01
-.02
.69

by sampling error

136

Table 10c. A test of parallelism for the post-message attitude

measure (factor 505). Item-factor correlations for factor
505 on all other factors are presented.

505
l 2 3 4 5

501 -. 10 -. 12 . 12* -. 02 -. 05
502 .40 .45 .48 .35 .42
503 .23 .19 .31 .26 .27
504 .33 .27 .32 .36 .33
506 -.29 -.29 -.22 -.23 -.30
507 e 12 e 05 -e 06 -0 05 e 07
508 .06 -.08 -.01 .08 .06
509 e03 -e 06 -e 04 -0 03 -e 03

Where:

501= Perceived Involvement

502: Perceived Argument Strength

503: Perceived Source Expertise

504: Pre-Message Attitude

505: Post-Message Attitudes

506: Behavioral Intention

507: Mead for Cognition

508= Argument Avoidance

509: Argument Enjoyment

* This correlation is not parallel with other correlations on this
factor.

137

Table 11a. A table of means. standard deviations.

number of

observations. and factor loadings for each of the three
behavioral intention items (factor 506).

Item Mean
1. Intention to attend informational 0.08
meeting.

2. Intention to work for a group 0.06
supporting or opposing proposal.

3. Mumber of hours of work promised. .432

8. D.

O 264

.30

2.11

227

227

227

Factor
Loadings

.50

.57

.63

138

Table 11b. Test of internal consistency for the behavioral intention
measure (factor 506). Correlations between items in the
factor appear below the diagonal. communalities appear on
the diagonal. and deviations from predicted correlations
generated by factor loadings appear above the diagonal.

1 2 3
1 26 .00 .00
2 .28 .34 .00

139

Table 11c. A test of parallelism for the behavioral intention measure
(factor 506). Item-factor correlations for factor 506 on
all other factors are presented.

506
1 2 3
501 .27 .24 .23
502 -.22 -.21 -.11
503 -.14 -.07 -.04
504 -.01 -.03 .01
505 -.22 -.21 -.11
507 .11 .04 .09
508 .01 .00 .05
509 .11 .09 .13

Where:

501= Perceived Involvement

502: Perceived Argument Strength
503= Perceived Source Expertise
504: Pre-Message Attitude

505: Post-Message Attitudes

506: Behavioral Intention

507: Meed for Cognition

508= Argument Avoidance

509: Argument Enjoyment

Table

140

123. A table of means. standard deviations.

number of

observations. and factor loadings for each of the six
(reduced) need for cognition items (factor 507).

Item

1. Thinking is not my idea of fun.

2.

4. I

6.

I would rather do something that
requires little thought than
something that is sure to
challenge my thinking abilities.

I try to anticipate and avoid
situations where there is a
likely chance I will have to
think in depth about something.

only think as hard as I have to.

I like tasks that require little
thought once I've learned them.

The idea of relying on thought
to make my way to the top
appeals to me.

Mean

2.27

2.09

2.12

2.63

2.59

3.85

S. D.

.73

.73

.68

.93

.97

.73

226

226

225

223

226

226

Factor
Loadings

.55

.66

.70

.61

.53

.57

141

Table 12b. Test of internal consistency for the (reduced) need for
cognition measure (factor 507). Correlations between items
in the factor appear below the diagonal. communalities'
appear on the diagonal. and deviations from predicted
correlations generated by factor loadings appear above the

diagonal.
1 2 3 4 5 6
1 .31 -.05 .00 .06 -.04 .03
2 .31 .44 .06 .01 .00 -.01
3 .39 .52 .50 -.05 .04 -.04
4 .40 .41 .38 .37 -.02 .OO
5 .25 .35 .41 .30 .28 .OO

142

Table 12c. A test of parallelism for the (reduced) need for cognition
measure (factor 507). Item-factor correlations for factor
507 on all other factors are presented.

507
1 2 3 4 5 6
501 -.02 -.04 -.07 .08 .05 .09
502 -.05 -.O6 -.06 -.15 -.03 -.15
503 -.03 -.01 -.02 -.13 .03 -.11
504 -.02 -.04 -.07 .08 .05 .09
505 -.03 .08 .04 -.05 .03 .04
506 .12 .06 -.05 .18 .05 .14
508 .25 .10 .16 .18 .25 .11
509 .23 .13 .13 .23 .23 .23

Where:

501= Perceived Involvement

502= Perceived Argument Strength
503= Perceived Source Expertise
504: Pre-Message Attitude

505: Post-Message Attitudes

506= Behavioral Intention

507= Meed for Cognition

508= Argument Avoidance

509= Argument Enjoyment

Table 13s. A table of means. standard deviations.

1.

3.

5.

143

number of

observations. and factor loadings for each of the five
avoidance subscale items of the argumentativeness measure

(factor 508).

Item

I enjoy avoiding arguments.

Arguing with a person creates
more problems for me than it
solves.

I get an unpleasant feeling when
I realize I am about to get
into an argument.

I am happy when I keep an
argument from happening.

I try to avoid getting into
arguments.

Mean

2.72

2.58

2.81

3.29

3.05

S. D.

.92

.86

.90

.83

.87

Factor

M Loadings
226 .71

225 .56

225 .67

225 .65

225 .86

Table 13b.

144

Test of internal consistency for the argument avoidance
Correlations between items in the
factor appear below the diagonal. communalities appear on
the diagonal. and deviations from predicted correlations
generated by factor loadings appear above the diagonal.

measure (factor 508).

.51

.44

.46

.43

.63

.04

.31

.32

.32

.45

-.02
.01
.45
.45

.55

-.03

-.04

.01

.42

.60

.02

-e 03

-.03

.04

.73

145

Table 13c. A test of parallelism for the argument avoidance measure
(factor 508). Item-factor correlations for factor 508 on
all other factors are presented.

508
1 2 3 4 5
501 .04 .02 -.12 -.09 -.03
502 -.01 -.16 .00 -.O6 -.07
503 -.04 -.12 -.02 -.05 -.07
504 .04 .13 , .05 .07 .13
505 .08 .01 .00 .04 -.04
506 .02 .04 -.01 .03 .05
507 .12 .29 .27 .11 .20
509 .56 .41 .56 .32 .52

Where:

501= Perceived Involvement

502: Perceived Argument Strength
503= Perceived Source Expertise
504: Pre-Message Attitude

505: Post-Message Attitudes

506: Behavioral Intention

507: Meed for Cognition

508= Argument Avoidance

509= Argument Enjoyment

146

Table 14a. A table of means. standard deviations. number of
observations. and factor loadings for each of the three
enjoyment subscale items of the argumentativeness measure
(factor 509).

Factor
Item Mean S.D. M Loadings
1. I am energetic and enthusiastic 3.66 .77 225 .51
when I argue.
2. I enjoy a good argument over a 3.49 .95 224 .71
controversial issue.
3. I enjoy defending my point of 4.03 .62 225 .60

view on an issue.

147

Table 14b. Test of internal consistency for the argument enjoyment
measure (factor 509). Correlations between items in the
factor appear below the diagonal. communalities appear on
the diagonal. and deviations from predicted correlations
generated by factor loadings appear above the diagonal.

1 2 3
1 .27 .01 -.01
2 37 .50 01

148

Table 14c. A test of parallelism for the argument enjoyment measure

(factor 509). Item-factor correlations for factor 509 on
all other factors are presented.

509
l 2 3

501 .10 .05 .03
502 -.09 -.18 -.11
503 -.16 -.O7 -.14
504 -.02 .03 -.O4
505 -.04 .Ol -.02
506 .16 .14 .06
507 .22 .19 .19
508 .43 .52 .32

Where:

50 1= Perceived Involvement

502: Perceived Argument Strength

503= Perceived Source Expertise

504= Pre-Message Attitude

505: Post-Message Attitudes

506: Behavioral Intention

507: Meed for Cognition

508= Argument Avoidance

509= Argument Enjoyment

149

Table 15. The impact of manipulated argument strength. manipulated
involvement. and manipulated source expertise on perceived
argument strength.

Low Argument Strength
Involvement
Low Moderate High
M=10.68 M=10.20 M=9.00
High Sx=2.96 Sx=4.05 Sx=2.60
M=19 M=17 M=19
Source
Expertise
M=8.78 M=8.23 M=7.05
Low Sx=3.30 Sx=2.70 Sx=2.09
M=l8 M=l7 M=l9
High Argument Strength
Involvement
Low Moderate High
M=12.94 M=l4. l7 M=13.85
High Sx=2.34 Sx=2.60 Sx=2.36
Source M=18 M=17 M=20
Expertise
M=l2.76 M=13.06 M=13.83
Low Sx=2.55 Sx=2.08 Sx=2.36
M=21 M=l7 M=l8

Table 16. The impact of manipulated involvement. manipulated source
expertise. and manipulated argument strength on perceived

involvement.

High

Argument
Strength

Low

High

Argument
Strength

Low

High
Source
Expertise

Low

Low

M=11.38
Sx=2.62
M=21

M=10.ll
Sx=3.7l
M=l8

150

Low Involvement

Source Expertise

High

M=11.56
Sx=3.47
N=18

M=ll.42
Sx=3.04
M=l9

Moderate Involvement

Low

M=10.58
Sx=3.06

M=17

Source

Expertise

High

M=10.33
Sx=3.36
M=l8

M=10.5O
Sx=2.73
M=l8

Mo

M=10.50
Bx=3.12
M=20

derate Involvement

Source Expertise

Low

M=13.22
Sx=4.35
M=l8

8:13.40
Bx=4.30
M=20

High

M=13.95
Sx=3.36
M=20

M=13.80
Sx=4.29
M=20

151

Table 17. The impact of manipulated source expertise. manipulated
involvement. and manipulated argument strength on perceived
source expertise.

Low Source Expertise

Involvement
Low Moderate High
M=17.62 M=19.76 M=22.71
High Sx=6.84 Sx=5.55 Sx=6.73
Argument M=21 M=17 M=17
Strength
Low M=l6.l7' M=15.11 M=14.95
Sx=4.88 Sx=5.58 Sx=4.31
“=13 M=18 M=l9
High Source Expertise
Involvement
Low Moderate High
M=23.39 M=29.56 M=28.05
High Sx=6.17 Sx=3.36' Sx=3.59
M=18 M=18 M=20
Argument
Strength
M=27.17 M=25.26 M=24.53
Low Sx=6.l9 Sx=6.33 Sx=7.24
M=18 M=19 M=19

152

Table 18. The impact of manipulated involvement. manipulated source
expertise. and manipulated argument strength on post-
message attitudes.

Low Involvement

Source Expertise

Low High
M=22.29 M=24.83
High Sx=6.70 Sx=4.78
M=21 M=18
Argument
Strength
M=22.94 M=21.63
Low Sx=5.83 Sx=7.58
M=18 M=l9
Moderate Involvement
Source Expertise
Low High
M=25.06 M=23.67
High Sx=5.52 Sx=7.49
M=18 M=20
Argument
Strength
M=17.78 M=20.35
Low Sx=5.98 Sx=7.49
M=18 M=20
High Involvement
Source Expertise
Low High
M=21.83 M=21.95
High Sx=5.33 Sx=8.03
M=18 M=20
Source
Expertise
M=17.25 M=18.20
Low Sx=8.23 Sx=6.00
M=20 =20

 

153

Table 19. The impact of perceived involvement and manipulated source
expertise on perceived source expertise.

Perceived Involvement

Low Moderate High
M=27.00 M=26.10 M=27.49
High Sx=6.82 Sx=5.96 Sx=4.91
M=33 M=3O M=49
Source
Expertise
M=l6.86 M=18.35 M=17.63
Low Sx=6.37 Sx=6.55 Sx=5.85

M=35

M=33

8:38

154

Table 20. The impact of perceived involvement and manipulated
argument strength on perceived argument strength.

Perceived Involvement

Low Moderate High
M=13.68 M=13.09 M=13.51
High Sx=2.04 Sx=2.51 Sx=2.3l
M=3l =35 M=45
Argument
Strength E
M=8. 24 M=8. 97 14:9. 76 F
Low Sx=3.31 Sx=3.06 Sx=3.12 '

N=37

M=34

M=41

 

Table 21. Impact of perceived involvement and perceived argument
strength on message processing.

High
Perceived
Argument
Strength

Low

Perceived Involvement

Low

M=-.73
Sx=2.21
M=33

$-30 60
Sx=2.70
M=35

Moderate

M=-.2l
Sx=2.94
M=33

M=-3.39
Sx=3.31
N=32

High

M=-.63
Sx=2.85
M=52

=-3.06
Sx=2.78
8:34

156

Table 22. Impact of perceived involvement and perceived source
expertise on source cognitions.

Perceived Involvement

Low Moderate High
M=0.00 M=-O.18 M=0.04
High Sx=0.48 Sx=0.55 Sx=0.42
M=27 M=28 M=46
Perceived
Source
Expertise
M=-0.63 M=-0.46 =-O.5l
Low Sx=0.94 Sx=0.79 Sx=0.84

M=4l M=39 M=41

 

157

Table 23. The impact of perceived involvement. perceived source
expertise. and perceived argument strength on message cognitions.

High
Perceived
Argument
Strength
Low

High
Perceived
Argument
Strength

Low

High
Perceived
Argument
Strength

Low

Low Perceived Involvement

Perceived Source Expertise

Low

M=-.67
Sx=l.78
M=12

H=-3e71
Sx=2.76
8:28

High

M=-.76
Sx=2.47
M=21

M=-3.5O
Sx=2.66
M=6

Moderate Perceived Involvement

Perceived Source Expertise

Low

“3’.45
Sx=3.24
M=11

Hz-4e11
Sx=3.02
"=28

High

M=-.19
Sx=2.89
M=21

u:-o.29
Sx=2.93
M=7

High Perceived Involvement

Perceived Source Expertise

Low

H=-2040
Sx=2.69
“=15

M=-3.23
Sx=2.96
M=26

High

M=0.14
Sx=2.63
"=36

“3‘2650
8X=2.20

158

Table 24. The impact of perceived involvement. perceived source
expertise. and perceived argument strength on source

cognitions.
Low Perceived Involvement
Perceived Source Expertise
Low High
M=-O.58 M=0.33
High Sx=0.79 Sx=0.52
M=12 M=6
Perceived
Argument
Strength
M=-O.68 M=-O.10
Low Sx=1.02 Sx=.44
M=28 =21
Moderate Perceived Involvement
Perceived Source Expertise
Low High
M=-O.55 M=-0.l9
High Sx=0.93 Sx=0.60
M=11 M=21
Perceived
Argument
Strength
M=-O.43 M=-O.l4
Low Sx=.74 Sx=.38
M=28 M=7
High Perceived Involvement
Perceived Source Expertise
Low High
M=-O.27 M=0.03
High Sx=.46 Sx=.29
M=15 M=36
Perceived
Argument
Strength
M=-O.65 M=0.13
Low Sx=0.98 Sx=0.83

M=26 =8

 

159

Table 25. Correlations between variables (below the diagonal) and
deviations predicted coefficients (above the diagonal) for
Model 6 (decimal points have been omitted).

1 2 3 4 5 6 7 8 9 10 11
1 100 0 0 0 0 -3 3 -15* -8 -19* 14*
2 2 100 0 O 4 18* 0 7 16* 7 -9
3 -3 -1 100 0 6 0 26* -3 7 3 -20*
4 4 -9 9 100 7 -2 8 17* -4 9 6
5 32 5 5 8 100 12 9 3 5 -4 36*
6 -5 17 66 4 11 100 58* 0 20* 16* -18*
7 4 64 25 2 9 58 100 40* 0 31* -12
8 -16 7 37 21 3 61 40 100 22* 5 -4
9 -7 44 7 -6 5 2O 43 22 100 10 -4
10 -19 4 26 38 -4 49 29 57 9 100 0
11 -14 -8 -27 -3 36 -29 -11 -21 -4 -32 100

Where:

l= Involvement Manipulation
2= Source Expertise Manipulation
3= Argument Strength Manipulation
= Pre-Message Attitude

= Perceived Involvement

= Perceived Argument Strength

= Perceived Source Expertise

= Message Cognitions

= Source Cognitions

10= Post-Message Attitudes

ll= Behavioral Intention

* Indicates that deviation from predicted correlation is larger than
would be expected given sampling error alone.

Table 26.

Perceived
Argument
Strength

The impact of the interaction between perceived involvement
and perceived argument strength on behavioral intention.

High

Low

Perceived Involvement

M=0.00
Sx=0.00
M=33

M=0.09
Sx=0.51
N=34

Moderate

M=-0.06
Sx=0.35
M=33

M=O.2O
Sx=0.68
M=35

High

M=0.03
SxO.56
M=40

M=0.84
Sx=1.34
"=31

Table 27. The impact of the interaction between perceived involvement
and poet-message attitudes on behavioral intention.

Positive

Post-Message
Attitudes

Megative

Perceived Involvement

Low

M=0.00
Sx=0.00
M=38

M=O.10
Sx=0.55
=30

Moderate

M=0.08
Sx=0.67
=38

M=0.06
Sx=0.36
N=31

High

M=0.00
SxO.56
M=45

M=O.75
Sx=1.27
M=36

162

Table 28. Contrasts used to test the interaction between perceived
involvement and post-message attitudes on behavioral
intentions.

Perceived Involvement

Low Moderate High

Positive 0 o 0 F
Post-Message
Attitudes

Negative 0 0 +1

 

163

Table 29. Correlations between variables (above the diagonal) and
deviations from predicted correlations (below the diagonal)
for Model number 7 (decimals have been omitted).

1 2 3 4 5 6 7 8 9 10 11 12
1 100 2 -3 32 -5 4 -16 -7 -19 17 14
2 0 100 -1 -9 5 17 64 7 44 4 6 -8
3 0 0 100 9 5 66 25 37 7 26 -13 -27
4 0 0 0 100 8 4 2 21 -6 38 -13 -3
5 0 4 6 7 100 11 9 3 5 -4 54 36
6 -3 0 0 0 12 100 58 61 20 49 -17 -29
7 4 0 -6 5 9 0 100 40 43 29 -6 -11
8 16* -2 -4 0 3 0 6 100 22 57 -17 -21
9 -7 16* -7 -5 5 -5 0 7 100 0 1 -4
10 20* -1 -9 0 - -4 0 2 0 -2 10 -4 -32
11 1 8 3 3 2 5 6 8 6 -2 100 54
12 5 -7 -19* 6 8 -17* -5 -7 -1 -8 0 100

Where:

1= Involvement Manipulation

2= Source Expertise Manipulation

3= Argument Strength Manipulation

4= Pre-Message Attitude

5= Perceived Involvement

6= Perceived Argument Strength

7: Perceived Source Expertise

8= Message Cognitions

9= Source Cognitions

10=Post-Message Attitudes

ll=Perceived Involvement X Post-Message Attitude Interaction on
Behavioral Intention

12=Behavioral Intention

* Indicates that deviation is greater than expected given sampling
error alone.

164

Table 30. The impact of need for cognition and perceived argument
strength on attitudes.

Meed for Cognition

Low High
M=23.05 M=24.72
High Sx=6.46 Sx=5.36
“=66 “=50
Perceived
Argument
Strength
M=18.51 M=18.54
Low Sx=7.17 Sx=7.22
M=-47 M=56

 

165

Table 31. The impact of need for cognition and perceived source
expertise on attitudes.

Meed for Cognition

Low High

M=23.34 M=24.l7

High Sx=6.73 Sx=6.62
M=53 M=47 :
Perceived E
Source 3
Expertise i
M=l9.30 M=19.36 .

Low Sx=l9.30 Sx=6.75

M=61 M=58 3

 

166

Table 32. The impact of argument enjoyment and perceived argument
strength on attitudes.

Argument Enjoyment

Low High
M=23.43 M=24.31
High Sx=6.08 Sx=6.09
M=67 M=49
Perceived
Argument
Strength
M=18.27 M=18.73
Low Sx=6.16 Sx=7.59

M=48 M=55

 

167

Table 33. The impact of argument enjoyment and perceived source
expertise on attitudes.

Argument Enjoyment

Low High
M=22.73 M=24.95
High Sx=6.94 Sx=6.21
M=56 =43
Perceived
Source
Expertise
M=l9.91 M=18.83
Low Sx=6.39 Sx=7.16

M=57 N=62

168

Table 34. The impact of argument avoidance and perceived argument
strength on attitudes.

Argument Avoidance

Low High
M=23.02 M=24.62
High Sx=5.40 Sx=6.58
M=57 M=6O
Perceived
Argument
Strength
M=18.91 M=18.02
Low Sx=6.51 Sx=7.57

M=45 N=58

169

Table 35. The impact of argument avoidance and perceived source
expertise on attitudes.

Argument Avoidance

Low High
M=22.55 M=24.88
High Sx=5.68 Sx=6.67
=49 M=5l
Perceived
Source
Expertise
M=l9.92 M=18.87
Low Sx=5.68 Sx=7.49

M=50 M=69

170

Table 36. The impact of need for cognition and the argument strength
manipulation on perceived argument strength.

Meed for Cognition

Low High
M=13.38 M=13.51
High Sx=2.22 Sx=2.42
M=61 M=49
Argument
Strength
M=9.52 M=8.56
Low Sx=3.18 Sx=3.23

M=52 M=57

 

171

Table 37. The impact of argument enjoyment and the argument strength
manipulation on perceived argument strength.

Argument Enjoyment

Low High

M=13.54 M=13.31

High Sx=2.34 Sx=2.27

M=59 M=49

Argument
Strength

M=9.7l M=8.38

Low Sx=3.25 Sx=3.01

M=56 N=55

Table 38.

Argument
Strength

172

The impact of argument avoidance and the argument strength
manipulation on perceived argument strength.

High

Low

Argument Avoidance

Low

M=13.43
8x=2.l9
M=51

M=9.43
Sx=3.31
M=51

High

M=13.48
Sx=2.42
M=58

M=8.68
Sx=3.12
M=6O

 

FIGURES

Figure Captions

Figure l. The impact of central cues. peripheral cues. and
moderator cues on attitudes assumed by the serial information
processing model.

Figure 2. The impact of cognitive demand on the allocation of
cognitive effort (From Stiff. 1986; used with permission of the
author and publisher).

Figuie 3. The impact of central cues. peripheral cues. and
moderator cues on attitudes assumed by the parallel information
processing model.

Figprg 4. Predicted path model with interaction occurring at the
stage of message perception (i.e. Model 1).

Figure 5. Predicted path model with two two-way interactions
occurring at the stage of message processing (i.e. Model 2).

Figure 6. Predicted path model with three-way interaction occurring
at the stage of message processing (i.e. Model 3).

E1gp;g__. Results of test of Model 6 (path coefficients are placed
onto appropriate paths).

Figuig . Results of revised model 6 (path coefficients are placed
onto appropriate paths).

173

174

I» Central Cues
Impact on

Attitudes

Peripheral Cues
9

 

 

Low High

Level of Moderator

Figure l.

Banach
Supplie

Figure 2.

/

 

175

Capacity Available
for Peripheral Cue

Total Processing
CmpmcituN ___ _____ E:
I fl

/

    

/
/

   

R Capacitg Allocated for
Central Cue Processing

 

Processing Capacitg Demanded

Impact
On
Attitudes

Figure 3.

 

176

‘\

' \\

Central Cues

Peripheral Cues

 

Low

l l
Moderate High

Level of Moderator Variable

3

acouuacmoo enquwmaxu oeuuuoaxm sequsoaxu

 

 

concomeWIllllll ouusom .Amilllllllln ooucom Am\\\\\\\ season
vo>aoouom x
acoeo>uo>cu
ve>« uses
7, couuceucm movcuuuu< nevsuauu< ucome>Ho>cH useleaaoecn
N aeuogezon m emceeuZIueom M ewemaozloum veswoouomT ..i.l. ocean
useeeaao>cn
caucouum vo>ueoeem
occauacmou uceecwu<an x
0933: Alllllmezoouem 59.3mm
ace-sued

suwcouum
ucoesua<

Figure 4.

8
7
1L

coquceucu

Heuoa>ezomAWIlllvwe

noesbquu<
eeezlueom

All

4/

eewuuoawm
woucom
acouuwcwoo vo>uoouem
summon P11 x
useso>~o>cu
vu>ueouom
mvvnu a u u<
ewemaezleum
//// ucase>ao>c
ecouuucwoo vo>uaomom
enemas: A111 x
:uwcouum
ucossmu<
vu>ueouem

sensuous» .
eousomnnn

va>wouuom

uceIo>Ho>cHin

oeuuueaxu
summon

usele>aoacu

 

vo>aooueml/.

guwcouum
acolsuudnmn

ocean

camcouum

 

ve>ueouom

ace-suu<

Figure 5.

179

oeuuuoawu caduceus»
season A11 common

 

vo>qeouem
ocowuqcmoo
mousom
, oeauuocwm
eousom
ve>uoouom
x
acesa>Ho>cH
couuceucn nevcuauu<T eavcuquu< ve>aeouom.Am111111 useloanoacu acumeanoacn
damages—en m swung—Tags? succession.— x vozeouom 1 women
:uwceuum
ucosswu<
\vguoouem
acoauucwou
owemeez
camcomum
ucemcuu<.tr7 cumceuum
poisons.— / ucemsuu<

Figure 6.

180

 

 

 

 

couucoucn A
Heuouaeaen

mm.l

 

 

 

meanness» congenscacex
ecouuucuou A me . season A so . seamen—In
eousom ve>ueuuem season

#0.:
succumbed novcuquu< uceeo>ao>sn souvensnuce:
.I.omeeeeznmm unease: useeH.Am usemeauoaen
ueom um. I cum veaaeouem an. . eseen

Em.

:uwcouum coauensaucez
acouuusmoo m use-awm<.A no . numceuum
ensues: Ho. vesueuuem acomsuu<

Figure 7.

181

oauuueawu couueusawcez

 

 

 

 

 

 

 

 

now—53: ecouuucuooAV 33 03:81 on . caduceus»
ammunozluuom season vo>ueouom summon
x
ucoso>ao>cu
vo>uouuem
em. 55.1
w“.
movsuwuu< eevcuuuu< ucemo>uo>cn couuensnuce:
couscoucn oweouexnm cm..l.omeeeox eaeenamrdeWIllllliusemeaaoacu
Heuou>ecen unom 1 sum voaueouem ocean
mq O .
mm.
.2. S . _
L\ sunsemuw seguensndce:
ecouuacwooA “Essence“ . caucus;
unease: co. vwaueouem co. ucemmuud

Figure 8.

"I11111111111661ES