IIJH.

      
   
    

 

I‘M

   

 

     

hum

 

 

 

 

     

u'x.

   

 
 
     

   

 

   

 

 

   

     

     

   

 

 
   

   

             

 

 

 

 

   

   

     

THESHS

Illi'llllllllllllllllllllll lllllll'lllll

3 1293 00896 9721

S

 

 

 

 

 

 

 

 

This is to certify that the
thesis entitled

FACTORS AFFECTING THE SUASORY FORCE OF IMPLICIT AND EXPLICIT
CONCLUSIONS IN PERSUASIVE MESSAGES

presented by

Michael Gerard Cruz

has been accepted towards fulfillment
of the requirements for

Master of Arts degfiwin Communication

 

 

 

Majo rofessor

0-7639 MS U is an Affirmative Action/Equal Opportunity Institution

 

  
     

  

LIBRARY
Michigan State =
University

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.

 

DATE DUE DATE DUE DATE DUE

 

m

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

MSU Is An Affirmative Action/Equal Opportunity Institution

c:\clrc\dmoduepm3-n-[

FACTORS AFFECTING THE SUASORY FORCE OF IMPLICIT AND EXPLICIT

CONCLUSIONS IN PERSUASIVE MESSAGES

BY

Michael Gerard Cruz

A THESIS

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

MASTER OF ARTS

Department of Communication

1991

6 SC? “ 6192/?

ABSTRACT

FACTORS AFFECTING THE SUASORY FORCE OF IMPLICIT AND EXPLICIT
CONCLUSIONS IN PERSUASIVE MESSAGES

BY

Michael Gerard Cruz

Body of Abstract

A meta-analysis found that persuasive messages with
explicit conclusions produced slightly more attitude change
than persuasive messages with implicit conclusions. A study
then investigated the existence of two possible mediating
variables, conclusion comprehension, and perceived position
of the source. Data were obtained from 176 participants on
two persuasive messages. Analyses indicated that the data
were consistent with a causal chain in which conclusion
drawing affected conclusion comprehension which affected the
perceived position of the source, which in turn affected
attitude change. Moreover, the data were consistent with a
similar causal model in which attitude change was replaced
with post-test attitude, and pre-test attitude was included
as a predictor of post-test attitude. Implications of these
findings for the construction of persuasive messages are

discussed.

 

ACKNOWLEDGEMENTS

Thanks to Linda Marshall for being very supportive and
for leading the way, and to Drs. Frank Boster, Steve Wilson,
and Jack Hunter for making this thesis much better than it

otherwise would have been.

ii

 

TABLE OF CONTENTS

List of Tables ........................................ v
List of Figures ...................................... vi
Introduction ...... ....... ............... ... ...... .....1
Review.................... ........ . ................. ..8
I. Statistical Artifacts ...................... 11
II. Moderators .............................. ...12
III. Source Credibility ......................... 13
IV. Conclusion Comprehension ................... 15
V. Intelligence.. .......... . ...... ............15
VI. Involvement......... ....................... 16
VII. Initial Position......... ...... . ..... ......18
VIII.Commitment.... ............................. 20
IX. Message Sidedness .............. . ....... ....20
Hypotheses ........................................... 22
Method ...... .... ................................... ..26
I. Subjects............... .................. ..26
II. Design................ .................. ...26
III. Instrumentation ............................ 26
IV. Messages. .................................. 27
V. Procedures ................................. 28

iii

Resu1tSOOOOO ........ ......OOOOOOOO. 000000000000 O ..... 30

I. Preliminary Analyses.. ..................... 30
II. Tests of Hypotheses ........................ 31
Discussion ............................... .... ........ 44
Conclusion ........................................... 48
Endnotes.. ........................................... 50
Appendix A................................. .......... 51
I. Legalization of Marijuana Message. ....... ..51
II. Change to Semesters Message.. .............. 52
Appendix B....... ..... ............ ............... ....53
I. Measure of Involvement .......... . ...... ....53

II. Measure of Source Credibility..............54
III. Measure of Attitude: Marijuana Topic......55
IV. Measure of Comprehension: Marijuana Topic.56
V. Measure of Attitide: Semester Topic ....... 57

VI. Measure of Comprehension: Semester Topic..58

Appendix C ..... . ..................... . ............... 59
I. Message Topic Pre-test Items ............... 59
Bibliography ............... .................... ...... 61

iv

T.“

LIST OF TABLES

Studies in the Meta-analysis ..................... 8
Cell Means and ANOVA (Tubbs, 1968) ..... .........10
Cell Means, Contrasts, and ANOVA (Kardes, 1988).17
Effect of Conclusion Type and Initial Position..19
Mean, Standard Deviation, And Cronbach’s Alpha..32

Regression of Comprehension Onto Involvement....33

LIST OF FIGURES

Path Model Using Attitude Change Scores ......... 37
Path Model Using Post-test Attitude Scores ...... 38
First Path Model With Involvement Removed. ...... 40
Second Path model With Involvement Removed ...... 41
Exploratory Path Model .......................... 43

vi

INTRODUCTION

Constructing and presenting a persuasive message is
presently an inexact undertaking. Although a number of
alternatives are available when organizing a persuasive
message, which alternatives produce the greatest attitude
change are not always known. One such alternative is
whether to state explicitly the conclusion of an argument or
leave the conclusion implicit. Few studies have examined
whether a message is more persuasive when audience members
are left to draw their own conclusions, or when the message
includes a specific statement of the conclusion.
Furthermore, past research stalled when conflicting and
inconclusive results were obtained, and no theoretical
framework was available to reconcile the findings.

Previous researchers defined explicit persuasive
messages as including a specific statement of the final
conclusion at the end of the message. Implicit persuasive
messages were then defined as explicit messages with the
statement of the final conclusion removed. Extant arguments
predict that both implicit and explicit messages will
produce greater attitude change. Favoring implicit
conclusions is the nondirective school of psychotherapy.
Rogers (1947) states:

... we have frequently observed that when the

individual has been authoritatively told that he is
governed by certain factors or conditions beyond his

1

2
control, it makes therapy more difficult, and it is

only when the individual discovers for himself that he

can organize his perceptions that change is possible.

(p. 361)

In the present context Rogers’ argument suggests that the
targets of a persuasive message will be influenced more if
they arrive at their own conclusions than if a conclusion is
provided.

Alternatively, Brehm (1966) theorized that restriction
or elimination of a person’s perceived free behaviors will
arouse a state of psychological reactance. Reactance was
thought to result in attempts to restore freedom, and to
increased valuation of restricted behaviors. If holding an
opinion is a free behavior, then a persuasive message with
an explicit conclusion can be viewed as an attempt to
restrict that person's freedom. A source’s explicit attempt
to restrict the target’s range of opinions produces
psychological reactance, and hence little attitude change.
Conversely, an implicit message allows the audience to draw
their own conclusions, producing no reactance. In the
absence of psychological reactance, attitude change in the
desired direction is more likely.

Linder and Worchel (1970) offered a third reason to
expect that implicit messages would be more persuasive than
explicit messages. They argued that stating the conclusion
of one’s message reduces credibility. The explicit
communicator appears to have a vested interest in persuading

the audience; whereas an implicit communicator appears

dispassionate or nonpartisan. Contrary to Linder and

Worchel, Hovland and Mandell (1952) argued that a
communicator who offers an implicit conclusion is seen as
having something to conceal, and a communicator who states a
conclusion explicitly is seen as frank and forthright.

Hovland and Mandell also argued that audience members
may require an explicit conclusion to understand the message
clearly. Through insufficient intelligence, motivation, or
prior knowledge, audience members may not comprehend an
implied conclusion. Were the recipients of an implicit
message to reach no conclusion, a conclusion unrelated to
that intended, or a conclusion in the direction opposite
that intended, less attitude change would occur in the
advocated direction than for a corresponding explicit
message. Thus, conclusion comprehension was proposed as a
mediator of the conclusion drawing-attitude change
relationship. Moreover, Hovland and Mandell proposed
audience intelligence as a moderating variable, presuming
that less intelligent receivers need an explicit statement
of the conclusion to understand the message, but more
intelligent receivers do not.

Another variable proposed to affect conclusion
comprehension was involvement (Kardes, 1988). The type of
involvement discussed by Kardes was termed outcome-relevant
involvement by Johnson and Eagly (1989). Kardes argued that
persons who are highly involved in an issue would
spontaneously generate the conclusion of an implicit message

more frequently than would weakly involved persons. The

THI

self-generated conclusions of more involved persons were
expected to produce greater attitude change than the
incomplete information available to less involved persons.
Thus Kardes predicted that for highly involved persons,
implicit messages would produce as much attitude change as
explicit messages. For less involved persons, Kardes
predicted that implicit messages would produce less attitude
change than explicit messages. Interestingly, Kardes
expected no main effect for outcome-relevant involvement,
despite research to the contrary (for a review, see Johnson
& Eagly, 1989).

Thistlethwaite, de Haan and Kamenetzky (1955) also
discussed conclusion comprehension, but as a moderating
variable. Specifically, they argued that implicit messages
are more effective when audience members are able to
comprehend the intended conclusion. When audience members
are unable to comprehend the conclusion to an implicit
message, e.g. when intellectual ability is low, explicit
messages are more effective.

A second variable thought to interact with conclusion
drawing is initial position (Fine, 1957; Linder & Worchel,
1970; Thistlethwaite, de Haan, & Kamenetzky, 1955; Weiss &
Steenbock, 1965). Initial position was defined as a
person’s opinion on an issue prior to hearing a relevant
persuasive message. Weiss and Steenbock (1965) argued that
the principles of the nondirective school of psychotherapy

would apply only when the intended conclusion was

TH!

disagreeable to the audience. When audience members hold
initially negative opinions, an explicit message would be
resisted, but an implicit message would be accepted. Only
when the conclusion being advocated is agreeable to the
audience will explicit messages be more effective.

Linder and Worchel (1970) predicted a three-Way
interaction between initial position, conclusion drawing,
and effort. Invoking Festinger's theory of cognitive
dissonance (1957), they argued that exerting effort to
understand a message with which one disagrees is dissonance
arousing. To reduce dissonance persons would change their
opinion in the direction advocated by the message. Because
implicit messages are more difficult to comprehend, greater
dissonance will arise. Explicit messages require less
effort to comprehend, and produce less dissonance. Thus, if
the explicit and implicit conclusions were comprehended
equally, implicit messages would produce greater attitude
change for initially unfavorable individuals. For initially
favorable persons, expending effort to understand an
implicit message is not dissonance arousing, and hence does
not produce greater attitude change than the corresponding
explicit message.

The final variable previously proposed to interact with
conclusion drawing is commitment. Tubbs claimed that high
levels of commitment (termed value-relevant involvement by
Johnson and Eagly, 1989) "will render audiences receptive to

congenial communications and resistant to uncongenial" (p.

TH:

14). Therefore, Tubbs predicted that commitment would
interact with conclusion type to affect attitude change
although he is unclear about the nature of the interaction.

The applicability of the above hypotheses to past
research is difficult to assess because prior studies did
not give exact descriptions of the persuasive messages used.
Messages were described as stating the final conclusion
explicitly or omitting the final conclusion, but it is
unclear whether other conclusions than the final one were
drawn within the message. That is, the messages may have
contained any number of arguments and conclusions that all
led to a final conclusion.r One may wonder whether implicit
messages actually did draw some explicit conclusions
although omitting the final conclusion. If it is the case
that conclusions were drawn within implicit messages, the
arguments of reactance theory and the nondirective school of
psychotherapy may not apply. Thus past research may have
provided only a weak test of their applicability. In fact,
doubt is cast on all prior research because of the ambiguity
in the construction of the implicit messages.

This study addressed this criticism through the use of
carefully crafted persuasive messages (Appendix A). Only
one final conclusion was drawn in the explicit messages, and
no conclusions were drawn in the implicit messages. Also
different from previous studies, two types of implicit
message were used--a moderately implicit message and a

strongly implicit message. Moderately implicit messages

were constructed by removing the conclusion from the
corresponding explicit messages. Strongly implicit messages
were constructed by removing all explicit statements of the

topic from the corresponding moderately implicit messages.

i |

 

REVIEW

To assess the viability of the preceding explanations,
a meta-analysis was conducted on six studies (listed in
Table 1). These are the only studies found that had
examined the impact of conclusion type on attitude change.
Although a seventh study (Linder & Worchel, 1970) discussed
the impact of conclusion drawing, an implicit message
condition was not included in the design. The six studies
included a total sample of 1423 subjects. Thistlethwaite et
al. (1955) used Air Force recruits; the other five studies

used college undergraduates.

Table 1

Studies in the Meta-analysis
95% Confidence

 

Author r N Interval
1. Fine .092 375 -.01 < r < .19
2. Hovland & Mandell .236 235 .12 < r < .36
3. Kardes .038 192 -.03 < r < .11
4. Thistlethwaite et al. .011 428 -.08 < r < .10
5. Tubbs .376 52 .14 < r < .61
6. Weiss & Steenbock .000 141 -.17 < r < .17

 

Pearson’s r was chosen as the measure of effect size.
The sign of r was chosen to be positive when explicit
messages produced greater attitude change than implicit
messages. In each case, the effect size was calculated by
taking the square root of etaZ‘with one degree of freedom.
Effect sizes were weighted by sample size and cumulated.

For Tubbs (1968), Fine (1957), and Kardes (1988) this
calculation was not straightforward. Tubbs reported an
incorrect analysis of variance, but also included his data
in their entirety so the reader could redo the analysis
(Table 2). Fine reported cell means and p values, but not
within-cell standard deviations or E values. Fortunately,
he reported 9 values precisely. From 9, E could be
determined, and the analysis of variance table reproduced.1
Finally, Kardes did not report a test of the main effect for
conclusion drawing, only the results of an a priori contrast
analysis. Because a table of means was also included, the
analysis of variance table could be reconstructed, and the
effect size calculated.

Overall, messages with explicit conclusions were
superior to messages with implicit conclusions (; = .082,

s? = .0064). However, the observed variance in
correlations was greater than expected due to sampling error
(expected sampling error §2== .0042, or 66 percent of the

observed variance), indicating the presence of additional

artifacts or moderators.

TH!

Table 2

10

Cell Means and ANOVA (Tubbs, 1968)

 

 

 

 

 

 

 

 

 

(Higher scores indicate a more attitude change in the
direction advocated by the message.)

Explicit Implicit

Conclusion Conclusion
High 7.50 -4.17
Commitment N=6 N=6
Medium 5.58 1.50
Commitment N=12 N=12
Low 5.13 0.25
Commitment N=8 N=8
Source of Sums of Mean
Variance Squares df Squares F p eta
Conclusion 480.08 1 480.08 9.79 .007 .376
Type
Commitment 28.71 2 14.35 .29 .789 .092
Level
Interaction 123.36 2 61.68 1.03 .366 .191
Within 2764.63 46 60.10 - - -
Total 3396.78 51 - - - -

2 |

 

 

11

Statistical Artifacts

Sampling error produces an observed variance effect
sizes, but other statistical artifacts can also increase the
observed variance in effect sizes. One such artifact is
unequal reliability of measurement in the dependent
variable. None of the six studies presented measurement
reliability data, but all indicated the number of items used
to measure the dependent variable. Because the number of
items in a measure associates positively with measurement
reliability, correlating effect size with number of items
provides an indirect assessment of the impact of
unreliability. Curiously, number of items correlates
negatively with effect size (; = -.64, p = .09). That is,
the more items used to measure the dependent variable, the
smaller the effect size. Not statistically significant with
only six data points, the correlation is probably due to
sampling error.

Another artifact that may increase the variance in
effect sizes is restriction in range for the independent
variable. At first glance, range variation should not be a
problem because each study included an explicit conclusion
condition and an implicit conclusion condition.
Nevertheless, one can speculate about the degree to which
the conclusion is implicit or explicit.

Implicit and explicit may be considered end points of a
continuum along which messages vary. For example, a

particular message may only weakly imply a conclusion,

 

12

necessitating an explicit statement of the conclusion for
comprehensibility. On the other hand, a message may very
strongly imply a conclusion, making an explicit statement of
the conclusion virtually redundant. If the implicit
messages in these studies imply differing conclusions
relative to the explicit messages, range variation may be
producing variation in effect sizes. Because only
Thistlethwaite et a1. (1955) reported data on the ability of
subjects to state the conclusion of the message, the impact
of range variation cannot be assessed.

In sum, the variance in effect sizes obtained in past
research cannot be attributed to the statistical artifacts
discussed previously. Hence moderating variables must be
investigated as potential sources of the variance in effect
sizes.

Moderators

Hunter and Schmidt (1990) argued that the effects of
moderators can be assessed in two ways. The first method
entails coding study features and correlating those features
with effect sizes. The second method is to create subsets
of subjects based on study features and calculate effect
sizes for each subset. Hunter and Schmidt argued that the
first method is more desirable, but the second method was
chosen because only six studies were available. The
following moderators were included from past research:
source credibility, conclusion comprehension, intelligence,

involvement, initial position, and commitment. Another

13

moderating variable, message Sidedness, was also analyzed.
Message Sidedness, whether a message includes only arguments
in support of a position (one-sided), or includes both pro
and con arguments (two-sided), was not discussed by past
researchers, but was included in this analysis because prior
studies included both one- and two-sided messages. Jackson
and Allen (1987) and Allen et al. (1991) have argued that
message Sidedness affects attitude change.
Source Credibility

Hovland and Mandell (1952) and Fine (1957) controlled
the credibility of the source in their studies. Hovland and
Mandell attributed their messages to either an economist
from a leading university (high credibility) or the head of
a large importing firm (low credibility). Fine attributed
his messages to the New York Times (high credibility), or
the Daily Worker, official newspaper of the Communist Party
in the United States (low credibility). Neither study
assessed the effectiveness of their controls. The other
four studies neither controlled nor measured source
credibility. Moreover, the credibility of the source used
in the other four studies could not be assessed. Thus, data
on source credibility were available from just two studies.

Hovland and Mandell presented sufficient data for
separate effect sizes to be calculated for each level of
source credibility, but Fine did not report within-cell
variances. Because the mean within-cell variance could be

calculated from the analysis of variance table, calculations

14

were based on the assumption that within-cell variances were

equal.

These two studies produced a combined r .156, sf =
.0054, N = 574. For these studies, the variance expected
from sampling error is §?== .0033 or 61 percent of the
observed variance. With a more credible source, r = .170,
s3 = .0076, N = 282, and with a less credible source, r =
.116, s3 = .0050, N = 292. The difference between the two
correlations is not statistically significant, although the
effect size for credible sources is 1.5 times the effect
size for less credible sources. To three decimal places the
variance expected from sampling error was §2== .067 for both
more credible and less credible sources. Thus, partitioning
the sample by levels of source credibility did not reduce
the variance in effect sizes.

These results suggest that source credibility does not
moderate the relationship between conclusion type and
attitude change, although this hypothesis cannot be
eliminated given the very low power of this analysis. In
addition, these data do not address the hypothesis that
source credibility acts as a mediator as postulated by
Linder and Worchel (1970) and Hovland and Mandell (1952).

As yet, researchers have not measured the impact of
conclusion drawing on credibility. Hence, the argument

stated previously, that conclusion type affects credibility

which in turn affects attitude change, remains to be tested.

 

15

Conelusien Cemprehension
Thistlethwaite et al. (1955) argued that implicit

messages are more effective only when audience members are
able to comprehend the intended conclusion. When audience
members are unable to comprehend the conclusion to an
implicit message, eXplicit messages would be more effective.

Strangely, they do not test their hypothesis. They
find a main effect for conclusion type on comprehension such
that explicit conclusions were better comprehended than
implicit conclusions, and they report that conclusion
comprehension did not affect attitude change. They never
test the interaction between conclusion type and
comprehension on attitude change, however. Rather, they
merely report that comprehension did not have an effect on
attitude change. Furthermore, they do not report cell
means, preventing the reader from testing the hypothesis.
Thus the hypothesized impact of conclusion comprehension
remains to be tested.
Intelligenee

Thistlethwaite et al. (1955) and Hovland and Mandell
(1952) assessed intelligence using the Technician Specialty
Aptitude Indices and the ACE intelligence test respectively.
Neither report an interaction between intelligence and
conclusion type. Thistlethwaite et al. found a main effect
for intelligence on attitude change; Hovland and Mandell
observed no effect for intelligence. Hovland and Mandell

argued that their subjects, all college students, showed

'Hl

16

little variance in intellectual ability, precluding an
observable impact. Unfortunately, the data for low and high
intelligence subjects could not be cumulated because cell
means were not reported in either study.

Invo vement

Kardes (1988) used an advertisement for a compact disc
player to study conclusion drawing. High outcome-relevant
involvement was induced by heading one advertisement with
two statements, "You Will Probably Own a Compact Disc Player
Sooner Than You Think," and "Some CD Players are Very Bad
and Some are Very Good," (p. 228) in bold-faced type. The
low outcome-relevant involvement advertisement was headed,
"Compact Disc Players."

The cell means and analysis of variance table from
Kardes (1988) are shown in Table 3. Kardes reported a
significant E-value (F = 5.87, p = .05) for his own a priori
contrasts. Based on the results of the contrast analysis,
Kardes argued that the least attitude change occurred for
less involved subjects who received an implicit message, and
that subjects in the other three cells did not differ
significantly; however, the residual explained variance
approached significance (F = 1.28, p = .07). Moreover, the
complete analysis of variance table shows that involvement
has a main effect that is stronger than the interaction
proposed by Kardes (F = 7.02, p < .01). Finally, a post hoc

contrast analysis (the contrasts are shown in Table 3)

TH;

17
Table 3

Gel Mea 5 Contra ts nd NOV Kar es 988
(Higher scores indicate a more positive attitude. Numbers

in () are the contrasts indicated by Kardes, and numbers in
[] are the post hoc contrasts.)

 

 

 

 

 

 

 

Explicit Implicit

Conclusion Conclusion
High 5.54 (1) 5.65 (1)
Involvement N=48 [1] N=48 [1]
Low 5.21 (1) 4.88 (-3)
Involvement N=48 [0] N=48 [-2]
Source of Sums of Mean
Variance Squares df Squares F p eta
Conclusion 0.59 1 0.59 0.22 >.05 .038
Drawing
Involvement 14.82 1 14.82 7.02 .01 .189
Interaction 2.37 1 2.37 1.12 >.05 .076
Contrasts 12.39 1 12.39 5.87 .05 .173
(Kardes)
Contrasts 16.36 1 16.36 7.75 .01 .199
(post hoc)
Within 396.83 188 2.11 - - -

 

Total 414.62 191 - - - -

 

18

produced a significant interaction (E = 7.75, p < .01)
slightly different from the interaction proposed by Kardes.

Specifically, the data are consistent with an
interaction in which the high involvement cells produce the
greatest attitude change, the low involvement/explicit
conclusion cell produces intermediate levels of attitude
change, and the low involvement/implicit conclusion cell
produces the least attitude change. Given the present
findings, Kardes' proposed interaction may not be the best
explanation for the data. Without additional data, however,
no strong conclusions can be drawn.
Initiel Besigien

Three studies (Fine, 1957; Thistlethwaite et al., 1955;
Weiss & Steenbock, 1965) have examined the interaction
between conclusion type and initial position. Fine (1957)
found that for persons initially favorable to the message,
explicit conclusions aremore persuasive than implicit
conclusions, but for those initially unfavorable no
differences between explicit and implicit conclusions were
observed. Weiss and Steenbock (1965) reported the opposite.
For persons initially favorable no differences were found;
for persons initially unfavorable explicit conclusions were
superior to implicit conclusions. Interestingly, a post hoc
contrast analysis of the Weiss and Steenbock data indicates
otherwise (see Table 4 for the contrasts used in this

analysis). For persons initially favorable implicit

19

conclusions produced more attitude change than explicit
conclusions. Finally, Thistlethwaite et al. (1965) report
no interaction between initial position and conclusion type

(Table 4).

Table 4

Effect of Conclusion Type and Initial Poeition

Thistlethwaite et al., 1955 (higher scores indicate a more
positive attitude).

 

 

Explicit Implicit
Conclusion Conclusion
Favorable 4.30 4.48
N=44 N=44
Unfavorable 2.91 2.82
N=l70 N=l70

 

 

 

 

Weiss and Steenbock, 1965 (Higher scores indicate a more
negative attitude).

 

 

Explicit Implicit
Conclusion Conclusion
Favorable 10.50 8.77
N=l6 N=36
Unfavorable 12.39 14.12
N=22 N=25

 

 

 

 

As with intelligence, the results for initial position
could not be cumulated. Because neither Fine nor
Thistlethwaite et al. reported cell means, effect sizes
could not be calculated. Furthermore, data were not

available on the initial attitudes of subjects in Hovland

20

and Mandell (1952), Kardes (1988), and Tubbs (1968). The
topics used were such that no determination could be made of
subjects’ initial attitudes. Results conflict, but these
may or may not be within sampling error of each other.
Again, further study is necessary.

mum;

Tubbs (1968) reported an interaction between commitment
and conclusion type on attitudes, but his analysis of
variance is incorrect. A re-analysis of his data indicates
that commitment did not have a main effect (F = .29), nor
did it interact with conclusion type (E = 1.03, df = 2, 46,
p = .366). No other authors investigated commitment.
W

Although not considered by previous authors, the type
of argument used appears to have varied across studies.

From the descriptions provided, it is clear that three of
the studies (Hovland and Mandell, 1952; Thistlethwaite et
al, 1955; Tubbs, 1968) used two-sided persuasive messages.
The other studies did not explicitly describe the messages
used. If one assumes that one-sided messages were used, one
can speculate about a possible interaction between
conclusion drawing and message Sidedness.

When two-sided messages were used, the mean effect for
conclusion type was r = .111, e3 = .0162, N = 715. For the
studies not specifying message Sidedness, r = .058, e3 =
.00136, N = 708. The variance in effect sizes for

unspecified messages is well within the expected variance

T?-

21

due to sampling error (e?== .00423), but for two-sided
messages, the expected variance due to sampling error (e ==
.00413) is only 25 percent of the observed variance.
Clearly, any conclusions about an interaction between
message sidedness and conclusion drawing must rest on an
unsubstantiated assumption--that one-sided messages were
used in the studies that did not specify two-sided messages.
Therefore, any conclusions drawn from the analysis remain
tentative and speculative. If the assumption is valid, an
interaction between message sidedness and conclusion drawing
is suggested (the observed correlations do not differ
significantly from each other, but statistical power is low
with data from only six studies). For both unspecified and
two-sided messages explicit messages were more persuasive
than implicit messages, but the difference was more
pronounced when two-sided messages were used. The
difference in effect sizes is small, but not necessarily
unimportant (Abelson, 1985; Rosenthal & Rubin, 1979).
Specifically, the observed difference may be an indication
that conclusion comprehension mediates the relationship
between conclusion drawing and attitude change. Because
two-sided messages present conflicting viewpoints, explicit
conclusions may be necessary for the audience to interpret

the message.

HYPOTHESES

As is apparent from the meta-analysis few conclusions
can be drawn from previous research. No prior study has
tested a model that links conclusion drawing to attitude
change through specific mediating variables. Although
conclusion comprehension has been studied as one such
mediating variable, results are inconclusive. For example,
Thistlethwaite et al. (1955) found an effect of conclusion
drawing on conclusion comprehension such that explicit
conclusions were better comprehended, but report an
insignificant effect of conclusion comprehension on attitude
change. The apparent absence of a relationship between
conclusion comprehension and attitude change may be due to
the presence of additional mediators or moderators.

In particular, conclusion comprehension is predicted to
have a causal impact on the perceived position advocated.
Typically, a persuasive message takes a stand on an issue
rather than arguing a neutral position. When message
recipients understand the conclusions to such messages, they
naturally perceive that the source of the message holds a
non-neutral opinion. In contrast, when message recipients
do not understand the conclusions, they have no reason for
attributing a non-neutral opinion to the source. Hence,
lack of conclusion comprehension would produce an overall

more moderate perception of the source’s position.

22

23

Next, perceived position of the source is predicted to
impact attitude change as specified in the linear
discrepancy model (see Hunter, Danes & Cohen, 1984). Linear
discrepancy states that attitude change can be predicted
from attitude and messages values by the equation

Aa = a(m - a).

That is, attitude change is proportional to the discrepancy
between the position advocated in the message and the
message recipient’s attitude. This equation can also be
written as

atfl = (1-a)at-+ am.
That is, one’s subsequent attitude is a positively-sloped
linear function of one’s prior attitude and the position
advocated in a persuasive message weighted by the values a
and 1 - a.

In summary, I hypothesize that more explicit messages
will produce greater conclusion comprehension
(Thistlethwaite et al., 1955), greater conclusion
comprehension will produce more extreme perceptions of the
source’s position (as argued above), and more extreme
perceptions of the source’s position will produce greater
attitude change (Hunter, Danes & Cohen, 1984).

The impact of involvement on conclusion comprehension
also requires further examination. Kardes (1988) reported
an interaction between outcome-relevant involvement and
conclusion drawing in which conclusion comprehension was

said to be a mediating variable. Unfortunately, Kardes did

24

not provide pertinent data. Although subjects answered
questions about the conclusion of the message, Kardes did
not report the accuracy of those responses, only the
latencies exhibited by subjects while answering. Subjects
in the low outcome-relevant involvement, implicit message
condition did exhibit greater response latencies compared to
subjects in the high outcome-relevant involvement or
explicit message conditions, but whether or not Kardes'
results generalize to conclusion comprehension will be
investigated in this study.

If the effects on response latencies generalize to
conclusion comprehension, then outcome-relevant involvement
should interact with conclusion drawing such that the
regression of conclusion comprehension onto outcome-relevant
involvement has: (a) a steep positive slope for the
strongly implicit message, (b) a moderately positive slope
for the moderately implicit message, and (c) a zero slope
for the explicit message.

Outcome-relevant involvement is also expected to
influence attitude change. Johnson and Eagly (1989)
reported that persons higher in outcome-relevant involvement
exhibit less attitude change than persons lower in outcome—
relevant involvement.

The last issue to be considered in this study is the
role of source credibility. Source credibility was
previously proposed as a mediating variable (Hovland &

Mandell, 1955; Linder & Worchel, 1978), but the impact of

25

conclusion drawing on credibility has not been measured.
Different authors predicted that compared to explicit
messages, implicit messages will either reduce and increase
source credibility. Because conflicting proposals were
advanced concerning this relationship, this study will
investigate whether either of competing explanations is

consistent with data.

METHOD

Subjects

Subjects were 176 undergraduates drawn from
communication classes at a midwestern university. Subjects
received course extra credit for participating.

Des' n

A 3 x 2 X 2 mixed design with random assignment to
treatments was used. The independent variables were
conclusion type, involvement and topic, and the dependent
variables were pre-test and post-test attitude. Each
subject responded to two topics, and within topics, subjects
were randomly assigned to one of three conclusion-drawing
conditions and one of two involvement conditions.
Instrumentetion

Attitude change was calculated as the difference
between pre-test and post-test measures of attitude. The
same 6-item, Likert-type scale was used on both the pre- and
post-test. Four other variables were assessed: (a)
perceived position advocated by the source, (b) pre-test and
post-test source credibility, (c) outcome-relevant
involvement with the topic, and (d) conclusion
comprehension. Subjects’ perceptions of the position
advocated by the source were measured using the same scale
that was used to measure their own attitudes (thus higher

scores indicated perceptions of a more positive position

26

27

advocated by the source). Their perceptions of the
credibility of the source, and their involvement with the
topic were assessed using 8- and 9-item semantic
differential scales respectively. Comprehension of the
conclusion was assessed by eight true/false items testing
their comprehension of the conclusion of the message (see
Appendix B for the items in each scale).

Messages

Desired were topics on which initial attitudes varied
widely and represented the entire range of opinions from
extremely positive to extremely negative. Based on pre-test
data, the topics chosen were the legalization of marijuana,
and the change from a quarter system to a semester system at
Michigan State University. For each topic 31 pre-test
subjects responded on 5-items (Appendix C); the possible
scores ranged from 5 to 25. Responses to the legalization
of marijuana issue ranged from 5 to 25, with a variance of
33.8. For the semester issue responses ranged from to 5 to
24, with a variance of 32.4.

For each topic an explicit message was constructed that
contained three different arguments leading to the same
explicitly-stated conclusion, and a final restatement of the
conclusion. Moderately implicit messages were produced by
removing each statement of the conclusion from the explicit
message. Strongly implicit messages were constructed by
removing each explicit statement of the topic from the

moderately implicit messages (Appendix A).

28

The three messages about legalizing marijuana were
described as editorials in the newspaper USA Todey. In the
low involvement condition the editorials were said to
concern a political debate in California in 1986. In the
high involvement condition, the editorials were said to
concern a political debate in Michigan in 1990. The three
messages about the semester change were described as
editorials in the Jeurnal of Education. In the low
involvement condition the editorials were said to concern
policies at the University of Southern California in 1986.
In the high involvement condition the editorials were said
to concern policies at Michigan State University in 1990.
Proeeduges

At the start of each experimental session, the
researcher read the following:

We are interested whether college students are well

informed about some current issues. We will assess how

well informed you are by first asking you some
questions about an issue. Next, you will read some
background information related to the issue. Finally,
you will again answer some questions about the issue.

Please read all the materials carefully, and answer the

questions truthfully.

The researcher then distributed a packet containing a
warm-up task (an authoritarianism scale (Altemeyer, 1988)
and a self esteem scale (Rosenberg, 1965)) and the following
which were included once for each topic: (a) a pre-test
measure of attitude embedded in items unrelated to the

study, (b) a pre-test measure of source credibility, (c) a

message, (d) a post-test measure of attitude, (e) a measure

29

of involvement, (f) a measure of conclusion comprehension,
and (g) a post-test measure of source credibility. Topic
order was reversed across the questionnaires so that some
subjects read the legalization of marijuana message first,
and other subjects read the semester change message first.
In addition, involvement was crossed with topic order, so
that the high and low involvement conditions were divided
between each topic and each order of presentation.
Subjects were allowed as much time as necessary to

complete the materials. When all the questionnaires were

completed, subjects were debriefed and dismissed.

RESULTS

Preliminary Analyses

Prior to testing the hypotheses, three analyses were
conducted. First, a confirmatory factor analysis was
employed to assess the content validity of all the scales
used in the study. The analysis included the deletion of
items for three reasons: (a) failing the criterion of
internal consistency, (b) failing the criterion of
parallelism, and (c) reducing the reliability of the scale
(see Appendix B for deleted items). For scales that
occurred more than once in the questionnaire, data from the
first responses to the items were factor analyzed, and
subsequent responses to the items were examined to assess if
the same factor structure fit each set of responses.

With one exception repeated scales produced the same
factor structure for each set of responses. The exception
was the source credibility measure; three of the four
occurrences of the measure fit the same factor structure,
but the post-test measure of source credibility for the
semester change issue generated inconsistent results. Items
5, 6, and 7 were significantly negatively correlated with
the other items and were removed from the measures of post-
test source credibility and source credibility change. All
scales produced unidimensional solutions, and means,

standard deviations, and Cronbach’s alpha for each are

30

31

listed in Table 5. Also listed are attitude change and
source credibility change. Both were computed by
subtracting pre-test scores from post-test scores.

The second analysis employed t-tests to determine if
the order of topic presentation had an impact. In no case
were statistically significant differences found for any of
the variables associated with either topic.

Third, the effect of the involvement induction on
perceived involvement was assessed. The involvement
induction had no impact on perceived involvement (; = -.046,
p > .05) for the legalization of marijuana issue. For the
change to semesters issue, however, there was a small, but
significant impact (n = .128, p < .05). Because the
involvement induction was weak or unsuccessful, subsequent
analyses were conducted using the measure of perceived
involvement. For both topics perceived involvement was
similarly distributed, with scores ranging from 6 to 42 and
a small negative skew. For the legalization of marijuana
topic perceived involvement had a mean of 30.9 and a
standard deviation of 6.7, for the semester change topic
perceived involvement had a mean of 31.2 and a standard
deviation of 8.2.

T s t s's

The impacts of conclusion drawing and involvement on
conclusion comprehension were tested using analysis of
variance and linear regression respectively. A significant

effect was found for conclusion drawing (for the marijuana

32

Table 5

Meen, Standard Deviation, And Cronnech’s Alphe

(#1 indicates a scale used for the legalization of marijuana

issue, and a #2 indicates a scale used for the change to

semesters issue.)

 

Scale Items Mean SD Alpha
Involvement #1 30.9 6.7 .85
Involvement #2 31.2 8.2 .89
Pre-test Attitude #1 22.6 10.1 .87
Pre-test Attitude #2 22.2 8.4 .84
Post-test Attitude #1 23.6 10.0 .90
Post-test Attitude #2 23.0 9.2 .93
Conclusion Comprehension #1 2.0 1.0 .51
Conclusion Comprehension #2 3.4 .9 .56
Perceived position of source #1 32.6 8.8 .86
Perceived position of source #2 36.3 6.7 .78
Attitude Change #1 .7 3.1 .32
Attitude Change #2 .8 4.6 .49
Pre-test Source Credibility #1 34.5 7.2 .89
Pre-test Source Credibility #2 32.5 6.1 .87
Post-test Source Credibility #1 34.3 6.6 .89
Post—test Source Credibility #2 19.4 4.3 .88
Source Credibility Change #1 -.3 5.6 .76
Source Credibility Change #2 .5 3.1 .68

33

topic, E = 38.75, df = 2, 170, p = .00; for the semester
topic F = 5.26, df = 2, 167, p = .01), but not for perceived
involvement (for the marijuana topic, n = .04, p = .31; for
the semester change topic, ; = -.07, p = .19). To test the
interaction between perceived involvement and conclusion
drawing, conclusion comprehension was regressed onto
involvement at each level of conclusion drawing (Table 6).
The slopes of these regression equations did not differ
significantly from zero, or from each other (p > .05)
indicating that the hypothesized interaction between

involvement and conclusion drawing did not occur.

Table 6

Regression of Comprehension Onto Involvement

(Separate regressions were computed at each level of
conclusion drawing. N = 169; bs are unstandardized
regression coefficients. No ps were statistically
significant at p < .05.)

 

Standard

Conclusion Drawing b Error of b Intercept
Strongly Implicit

Marijuana Topic .025 .018 .50

Semester Topic .005 .017 3.03
Moderately Implicit

Marijuana Topic -.002 .015 2.10

Semester Topic .006 .015 3.29
Explicit

Marijuana Topic .020 .014 1.93

Semester Topic -.012 .011 4.05

34

To test the causal models hypothesized previously a
path analysis was conducted. Two models were analyzed, one
which used attitude change scores and one which used pre-
test and post-test attitude scores.2 Because involvement
was found not to affect conclusion comprehension in the
previous analysis, the hypothesized interaction between
involvement and conclusion drawing was not included in the
models. Thus, one model specified a causal chain which
included, in order: conclusion drawing, conclusion
comprehension, perceived position of the source and attitude
change. Involvement was included in the first model also as
a predictor of attitude change. The second model replaced
attitude change with post-test attitude, and included pre-
test attitude as a predictor of post-test attitude.

Prior to computing zero-order correlations scatterplots
were examined to see if each pair of variables was linearly
related for each topic. Because no nonlinearities were
observed, correlations (Tables 16 and 17) and path
coefficients (Figures 1) were computed. Based on goodness
of fit indices (eni-sguare = 4.33, df = 5, p > .10, and
chi-square = 3.67, df = 5, p > .10 for the legalization and
semester topics respectively) the first model was consistent
with the data. For the legalization of marijuana topic each
of the path coefficients was significant at p = .05 except
for the path from involvement to attitude change; for the
semester change topic neither the path from perceived

position of the source to attitude change nor the

35

Table 16

Correletion Matrix: Legalization Of Marijuana Issue

(An asterisk (*) indicates a correlation significant at
p < .05, N = 165, one-tailed test.)

1 2 3 4 5 6 7 8 9

(1) PERCENED 1.00
INVOLVEMENT
(2) PRE-TEST .26* 1.00
ATTITUDE
PRE-TEST
(3) SOURCE .13* .02 1.00
CREDIBILITY
(4) CONCLUSION -.05 .08 -.13*1.oo
DRAWING
(5) CONCLUSION .04 .10 -.06 .55*1.oo
CONPREIIENSIoN
PERCEIVED
(6) POSITION 0F .17* .23* -.09 .46* .66* 1.00
msouncr
(7) POST-TEST .22* .94*-.01 .10 .13* .27*1.oo
ATTITUDE
POST-TEST
(8) SOURCE .27* -.02 .67* -.14* -.04 -.05 -.03 1.00
CREDIBILITY
(9) ATTITUDE -.09 -.11 .01 .03 .13* .12 .18* .02 1.00

GMME

36
Table 17

gerreletion Metnix: Cnenge To Semesters Issue

(An asterisk (*) indicates a correlation significant at
p < .05, N = 165, one-tailed test.)

1 2 3 4 5 6 7 8 9

(1) PERCEIVED 1.00
INVOLVENENT
(2) PRE-TEST -.11 1.00
ATTITUDE
PRE-TEST
(3) SOURCE .15* .12 1.00
CREDIBILITY
(4) CONCLUSION -.03 .12 .11 1.00
DRAWING
(5) CONCLUSION -.07 -.05 .07 .55*1.00
OONPREEENSION
PERCEIVED
(6) POSITION 0F .13*-.04 .214: .26* .57* 1.00
THE SOURCE
(7) POST-TEST -.06 .374. .204: .13*-.01 -.02 1.00
ATTITUDE
POST-TEST
(3) SOURCE .24* .224 .7o* .04 -.03 .09 .35*1.00
CREDIBILITY
(9) ATTITUDE .11 -.08 .19* .04 .03 .07 .41* .3o*1.oo

GENE

37
Figure 1

Path Model Using Attitude Chenge Scores

The first path coefficients shown are for the legalization
of marijuana issue, the second for the change to semesters
issue. An asterisk (*) indicates a path coefficient
significant at p = .05, N = 169.

 

 

 

 

 

 

 

 

 

Conclusion

Drawing

0 55*

.22*

L—————> Conclusion
Comprehension

.66*
.57*

 

 

> Perceived Position
of the source

 

 

 

.14*
.02

 

 

 

> Attitude
> Change

 

 

 

Involvement ~——————1 -.11

.10

 

 

 

 

 

38

Figure 2

Path Model Using Post-test Attitude Scores

The first path coefficients shown are for the legalization
of marijuana issue, the second for the change to semesters
issue. An asterisk (*) indicates a path coefficient
significant at p = .05.

 

 

 

 

 

 

 

 

 

 

Conclusion
Drawing
.55*
.22*
> Conclusion
Comprehension
.66*
.57*
> Perceived Position

 

of the source

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

.06*
.01
Pre-test .93* ——————-> Post-test
Attitude > Attitude
.88* >
-.03
Involvement
.04

 

 

 

39

path from involvement to attitude change was statistically
significant. Finally, none of the correlations predicted by
the model differed significantly from those observed in the
data (p = .05).

The second model (Figure 2) also fit the data (for the
marijuana topic, chi-sgnane = 8.06, df = 7, p > .10; for the
semester topic, cni-sguane = 4.08, df = 7, p > .10).
Consistent with the first model all path coefficients were
significant at p < .05 except the path from involvement to
post-test attitude for both topics, and the path from
perceived position of the source to attitude for the
semester topic. Again, none of the correlations predicted
by the model differed significantly (p > .05) from those
observed in the data.

Because involvement did not have the anticipated impact
on attitude, analyses were conducted with involvement
removed from the models (Figures 3 and 4). With attitude
change as the final dependent variable, cni-sguare = 1.32,
df = 3, p > .10 for the marijuana topic, and
eniesgnene = 1.86, df = 3, p > .10 for the semester topic.
With post-test attitude as the final dependent variable,
gnizsgnene = 4.94, df = 5, p > .10 for the marijuana topic,
and cni-sguare = 2.40, df = 5, p > .10 for the semester
topic.

In each of the previous analyses no correlation
predicted by a model differed significantly from the

observed correlation, however, for the marijuana topic the

 

40
Figure 3

First Path Medel With Invelvement Removed

The first path coefficients shown are for the legalization
of marijuana issue, the second for the change to semesters
issue. An asterisk (*) indicates a path coefficient
significant at p = .05, M = 169.

 

 

 

 

 

 

 

 

 

 

Conclusion

Drawing

.55*

.22*

e—————> Conclusion
Comprehension

.66*
.57*

 

> Perceived Position
of the source

 

 

 

.12*
.03

 

 

> Attitude
Change

 

 

 

 

 

41

Figure 4

Secend Peth nedel With Involvement Removed

The first path coefficients shown are for the legalization
of marijuana issue, the second for the change to semesters
issue. An asterisk (*) indicates a path coefficient
significant at p = .05, M = 169.

 

 

 

 

 

 

 

 

 

 

Conclusion
Drawing
.55*
.22*
> Conclusion
Comprehension
.66*
O 57*
> Perceived Position

 

of the source

 

 

 

.06*
.01

 

 

 

Pre-test .92* -——————> Post—test
Attitude > Attitude
.88*

 

 

 

 

 

 

 

 

42

observed correlation between pre-test attitude and perceived
position of the source differed from the predicted
correlation by .20, approaching statistical significance

(e = 1.84, p = .06). A causal link from pre-test attitude
to perceived position of the source was not predicted, but
an exploratory analysis seemed warranted because of the
nearly significant difference statistically discussed
previously and because of a significant zero-order
correlation (n = .23, p < .01). Hence, a fourth causal
model, the model in Figure 4 save an added causal path from
pre-test attitude to perceived position of the source, was
tested (Figure 5). This model also fit the data
(cni-sgnare = 1.50, df = 4, p > .10, and chi-sgnane = 2.16,
df = 4, p >. 10 for the legalization and semester topics
respectively), and, more importantly, the greatest
difference between a predicted and observed correlation was
.09 (e = 1.01, p = .32) for the legalization of marijuana

1.21, p = .23) for the semester change

topic and .13 (5
topic. In each case the largest error was for the link
between conclusion drawing and perceived position of the
author.

Left to be investigated was the role of source
credibility. Conclusion drawing was observed to have no
association with source credibility change (for the
legalization of marijuana topic, n = -.01, p > .05; for the
semester change topic, n = —.07, p > .05). The impact of

conclusion drawing on post-test source credibility was also

examined with pre-test source credibility as a covariate.

43

Again, no relationship was found (E = .26, p = .61).

Figure 5

Exploratory Path Model

The first path coefficients shown are for the legalization
of marijuana issue, the second for the change to semesters

issue. An asterisk (*) indicates a path coefficient
significant at p = .05, M = 169.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Conclusion
Drawing
.55*
.22*
> Conclusion
Comprehension
.65*
.57*
.16* L—————> Perceived Position
> of the source
-.01
.06*
.01
Pre-test .92* > Post-test
Attitude > Attitude

 

 

 

 

.88*

 

 

 

DISCUSSION

The goal of this study was to clarify the relationship
between conclusion drawing and attitude change. To some
extent this goal was achieved, although the results were not
perfectly consistent across topics. For the legalization of
marijuana topic two causal models were found to be
consistent with the data. One model indicates that two
variables, conclusion comprehension and perceived position
of the source, mediate the relationship between conclusion
drawing and attitude change. A causal chain which includes,
in order: conclusion drawing, conclusion comprehension,
perceived position of the source, and attitude change was
found to be consistent with the data. In detail, the more
explicit the conclusion to a persuasive message, the better
the conclusion is comprehended. Greater conclusion
comprehension produces perceptions that the source of the
message advocates a more extreme position. Last,
perceptions that the source holds a more extreme position
produce more attitude change.

The second causal model, conceptually similar to the
first, was also found to be consistent with the legalization
of marijuana data. The only difference between the models
is that the first predicts attitude change from perceived

position of the source, and the second predicts post-test

44

45

attitude from perceived position of the source and pre—test
attitude.

Although post hoc, a third causal model was also found
to be consistent with the data. Similar to the second
model, the third model added a causal path from pre-test
attitude to perceived position of the source. Such a link
would be predicted by Social Judgment theorists (for a
review see Kiesler, Collins & Miller, 1969), who argue that
messages similar to one’s own attitude will be assimilated,
or perceived as even more similar to one’s own attitude, but
messages which are different from one’s own attitude will be
contrasted, or perceived as even more different from one’s
own attitude. Such a relationship was not observed,
however. Instead, a linear relationship was observed, in
which more positive pre-test attitudes were associated with
perceptions that the source held a more positive attitude.

To reconcile the data with Social Judgment theory would
require the conclusion that the messages were within the
latitude of acceptance for most subjects. Given the
distribution of pre-test scores (ranging from the top to the
bottom of the scale for both topics), such a conclusion is
unlikely. Because the third model was exploratory,
replication is necessary. Further theoretical work is also
necessary to explain the observed causal relationship from
pre-test attitude to perceived position of the source that

was observed here.

46

Considering the semester change topic, the same models
were found to fit the data, but the path coefficients from
perceived position of the source to attitude change and
post-test attitude were not statistically significant.
Finding an impact, however, was precluded by a ceiling
effect in the measurement of perceived position of the
source. The author’s position was perceived as
significantly more positive for the semester change issue at
each level of conclusion drawing (gs range from 3.32 to
7.00, ps < .05) with 56 of 169 subjects scoring 42 on a
scale from 6 to 42. Moreover, the distribution exhibited a
significant negative skew (p (-1.76 < skew < -1.04) = .95).
It may be that the subjects, all undergraduates, were better
informed about the semester change topic and hence better
able to discern the conclusion.

The hypothesized effects of involvement were not
observed. Consistent with Kardes (1988), involvement was
predicted to interact with conclusion drawing to affect
conclusion comprehension, and to interact with perceived
position of the source to affect attitude change. Neither
prediction was consistent with the results. The difference
between these results and Kardes’ data may be due to the
topic and the involvement induction used. Perhaps few of
the subjects in Kardes’ study had strong, pre-existing
opinions on the topic, purchasing a compact disc player. If
weaker pre-existing opinions were held, the involvement

control would have been more successful and had an impact on

47

outcome variables. In this study the involvement control
was unsuccessful, and data were analyzed using perceived
involvement. Because perceived involvement, unlike
experimentally controlled involvement, is susceptible to
confounds, the effects of involvement may have been
obscured. Once again, further study is necessary to
determine if, or when, involvement interacts with conclusion
drawing.

Finally, despite much conjecture by previous authors,
conclusion drawing was not found to affect perceived source
credibility. Although strong conclusions can never be drawn
from just one study, indications are that the relationship
between conclusion drawing and attitude change is not

mediated by source credibility.

CONCLUSION

In sum, although some of the hypotheses were consistent
with the data, others were not. As hypothesized, the
relationship between conclusion drawing and attitudes was
mediated by conclusion comprehension and perceived position
of the message. This finding is consistent with the meta-
analysis that found only a weak association between
conclusion drawing and attitude change (in the meta-
analysis, n = .082; for the marijuana topic, ; = .08; for
the semester topic, ; = .04). The results of this study
suggest that writers of persuasive messages should use
explicit messages, or, at least, be very concerned with
writing comprehensible ones.

The previous conjectures that implicit messages would
be more persuasive were not consistent with the meta-
analysis or the current study, but because conclusion
drawing had such a strong impact on conclusion
comprehension, implicit messages were virtually precluded
from producing significant attitude change. If conclusion
could be controlled independently of conclusion drawing,
perhaps through the use of very simple messages, the
hypothesized advantages of implicit messages might realized.
Such a scenario is suggested by Linder and Worchel (1970)
whose subjects were informed of the intended conclusions to

implicit persuasive messages after having attempted to

48

49

discern the conclusion for themselves. Under these
conditions, initially implicit but ultimately explicit
messages were more persuasive than initially explicit
messages.

Implicit messages might also hold a persuasive
advantage when situational factors enhance the
comprehensibility of the message conclusion. For example,
when the speaker is known by audience members to hold a
particular position or when audience members are very well
informed on the topic, implicit messages might not suffer
from reduced comprehension and might, therefore, produce
greater attitude change. Investigation of the impact of
conclusion drawing when no differences in conclusion

comprehension exist is left to further study.

ENDNOTES

1Because Fine (1957) reports p values such as .02, .11,
.06, etc., I assumed that p < .10 (for the effect of
conclusion type on attitude change) implied p > .09. Two E
values were used to reproduce the ANOVA table, one
corresponding to p = .10 and one to p = .09. Because these
two Es differed by little and produced an even smaller
difference in ;, they were averaged. This value was then
used to calculate the effect size.

2The second model was included because of the
unreliability of measurement of attitude change (for the
marijuana topic, a = .32; for the semester change topic,

a = .49).

50

APPENDICES

Appendix A

Legalization of Marijuana Message

(Double-underlined sections were removed from the explicit
message for both the moderately and strongly implicit
messages. Single-underlined sections were removed for the
strongly implicit message only.)

I disagree with my opponents current position. The
following are my reasons for disagreeing.

During prohibition, organized crime grew and profited
through the illegal sale of alcohol. The illegal sale of
alcohol provided crime syndicates with huge amounts of
income. Today, organized crime remains a problem even
though alcohol is now legal. One reason that organized
crime still exists is that other sources of illegal income
are available. That is, organized crime still profits from
the sale of illegal substances. One such substance is

nexijnene. Therefore, to reduce she pewer of onganized
crimeI marijuana should be legalized.

Americans value their personal freedom highly. In
fact, our constitution guarantees our freedom and insures
that the government cannot restrict our freedom unless it is
for the good of all. Americans can even act in ways which
are dangerous or harmful to themselves. For example,
alcohol and tobacco use are legal despite the known dangers
associated with both. On the other hand, other less
dangerous substances are illegal. One sugn snpstence is
ma "u . Since marijuana is iess dangerous tnan aicehoi
and tebecco, marijuena use sheuid also be legei.

Some people have argued that many immoral actions or
"bad influences" often lead to illegal actions. For
example, pornography has been said to lead to violent,
sexual crimes. For this reason, those people have argued
that all immoral actions should be outlawed. Usually, these
people can give numerous actions which they consider
immoral, and then argue that these immoral actions lead to
illegal actions. However, many immoral actions do not lead

illegal actions and should remain legal. One sucn innenai
as; is negijnene use, wnign nas peen incenrectly said te

e d ' coc 'n . Since manijuana use dees
not 1 - '_°_-_’____:_r__1__d__._:!_____ e a __'____.___. - h_-____.___-_,-
I292;-

 

51

52

Change to Semesters Message

(Double-underlined sections were removed from the explicit
message for both the moderately and strongly implicit
messages. Single-underlined sections were removed for the
strongly implicit message only.)

I believe that a change in the current system is
warranted. Here are my reasons for wanting a change.

One of the major functions of our institution is to
give a quality education. In order to fulfill this
function, professors should cover material in depth,
providing a more rigorous and advanced viewpoint than given
to high school students. However, time limitations often
prevent professors from providing such extensive coverage of
subjects. In many cases, professors are only able to skim
over the necessary material and are unable to spend time on
more interesting topics. The major source of the time
limitation is the guarter system. Changing from the guarter
system to the semester system would give professors time to
teach better classes. Therefore, a change should be made

from guarters to semesters.
Many students complain that tuition is too high. These

students are having trouble affording college. One target
of their complaints is the administration. Students claim
that administrators waste too much money on irrelevant
expenses. For example, high registration fees are much
maligned by angry students. One of the main contributors to
high administrative and registration costs is the guarter
system. Changing from the guarter system to the semester
system would reduce registration costs. Again, a calendar
change should be made from gparters to semesters.

A third issue which I will address is academic
progress. Increasingly, students are taking five and six
years to graduate. However, the system was designed to
allow students to graduate in four years. The complexity of
the current system plays a part in delaying graduation.
Students have a multitude of requirements and options which
must be chosen with inadequate advising. To improve
academic progress, a less complex system should be chosen.
The guarter system contributes to complexity. Converting
from the guarter system to the semester system would reduce
complexity and improve academic progress. Hence, we should
convert to a semester calendar.

In conclusion, I have given three arguments why a
semester system is superior to a guarter system. Now is the
time to change from a guarter system to a semester system.

 

Appendix B

a u e f vo v ment

(An asterisk (*) indicates an item deleted based on
confirmatory factor analysis.)

1. involving 1 2 3 4 5 6 7 uninvolving
2. unimportant 1 2 3 4 5 6 7 important

3. relevant to 1 2 3 4 5 6 7 not relevant

you to you
*4. does not 1 2 3 4 5 6 7 affects you
affect you
5. matters 1 2 3 4 5 6 7 does not
matter

*6. impact on 1 2 3 4 5 6 7 no impact on
your life life

7. a concern 1 2 3 4 5 6 7 no concern
of yours of yours

8. significant 1 2 3 4 5 6 7 insignificant

*9. worrisome 1 2 3 4 5 6 7 not worrisome

53

54

Measune of Source Credipility

(An asterisk (*) indicates an item deleted based on
confirmatory factor analysis. Items 5, 6, and 7 were also
deleted from the post-test measure of source credibility for
the semester change issue.)

1. believable 1 2 3 4 5 6 7 unbelievable
*2. inaccurate 1 2 3 4 5 6 7 accurate

3. unconvincing 1 2 3 4 5 6 7 convincing

4. well-informed 1 2 3 4 5 6 7 poorly-informed

5. unbiased l 2 3 4 5 6 7 biased

6. correct 1 2 3 4 5 6 7 incorrect

7. knowledgeable 1 2 3 4 5 6 7 not knowledgeable

8. untrustworthy 1 2 3 4 5 6 7 trustworthy

55

Meesnze e: Afitipnde: Merijuana Topic
1. Marijuana should be legalized.

Strongly 1 2 3 4 5 6 7 Strongly
Disagree Agree

Marijuana should remain illegal.

Strongly 1 2 3 4 5 6 7 Strongly
Disagree Agree

Marijuana is harmful.

Strongly 1 2 3 4 5 6 7 Strongly
Disagree Agree

There is nothing wrong with smoking marijuana.

Strongly 1 2 3 4 5 6 7 Strongly
Disagree Agree

People who smoke marijuana should not be treated like
criminals.

Strongly 1 2 3 4 5 6 7 Strongly
Disagree Agree

Americans have a right to smoke marijuana if they want
to.

Strongly 1 2 3 4 5 6 7 Strongly
Disagree Agree

 

56

Measune pf gonclnsion Comprenensien: Marijuana Topic

(An asterisk (*) indicates an item deleted based on
confirmatory factor analysis.)

*1. The author believes that prostitution should be

*2.

*4.

*6.

*7.

legalized.

a. true b. false

The main topic of the message was civil rights.
a. true b. false

More than anything else, the message was about
marijuana.

a. true b. false

According to the author, alcohol and tobacco should be
illegal.

a. true b. false

The author’s opinion is that marijuana should be
legalized.

a. true b. false

According to the author, pornography is dangerous.

a. true b. false

The author thinks that all drugs should be legalized.
a. true b. false

The author’s main point is that crime is a big problem
in this country.

a. true b. false

eas

57

o tt‘t de: Se ester T ic

(The high involvement items are shown. For the low
involvement items, MSU was replaced by USC. An asterisk (*)
indicates an item deleted based on confirmatory factor
analysis.)

1.

*5.

Changing to semesters at MSU is a good idea.

Strongly 1 2 3 4 5 6 7 Strongly
Disagree Agree

MSU should not be changing to a semester system.

Strongly 1 2 3 4 5 6 7 Strongly
Disagree Agree

MSU has used the quarter system and should continue to
do so.

Strongly 1 2 3 4 5 6 7 Strongly
Disagree Agree

Changing to semesters will increase the quality of
education at MSU

Strongly 1 2 3 4 5 6 7 Strongly
Disagree Agree

I would rather have three 10-week terms than two 15-
week terms every year.

Strongly 1 2 3 4 5 6 7 Strongly
Disagree Agree

The advantages of the semester system outweigh the
advantages of the quarter system.

Strongly 1 2 3 4 5 6 7 Strongly
Disagree Agree

58

Meesnge of Compnenensien: Semester Topic

(An asterisk (*) indicates an item deleted based on
confirmatory factor analysis.)

*1.

*4.

*6.

*7.

The author believes that the quality of education is
low.

a. true b. false
The main topic of the message was college tuition.
a. true b. false

More than anything else, the message was about semester
and quarter systems.

a. true b. false

According to the author, there should be more
uniformity in education.

a. true b. false

The author’s opinion is that semester systems are
better than quarter systems.

a. true b. false

According to the author, education is more important
than money.

a. true b. false

The author thinks that the school year should not be
interrupted by breaks.

a. true I b. false

The author’s mainpoint is that education is a failure
in this country.

a. true b. false

Appendix C

Message Topic Pre-test Items

1.

Marijuana should be legalized.

Strongly 1 2 3 4 5 6 7 Strongly
Disagree Agree

Marijuana should remain illegal.

Strongly 1 2 3 4 5 6 7 Strongly
Disagree Agree

Marijuana is harmful

Strongly 1 2 3 4 5 6 7 Strongly
Disagree Agree

There is nothing wrong with smoking marijuana.

Strongly 1 2 3 4 5 6 7 Strongly
Disagree Agree

People who smoke marijuana should not be treated like
criminals.

Strongly 1 2 3 4 5 6 7 Strongly
Disagree Agree

Changing to semesters at MSU is a good idea.

Strongly 1 2 3 4 5 6 7 Strongly
Disagree Agree

We should not be changing to a semester system.

Strongly 1 2 3 4 5 6 7 Strongly
Disagree Agree

MSU has used the quarter system and should continue to
do so.

Strongly 1 2 3 4 5 6 7 Strongly
Disagree Agree

Changing to semesters will increase the quality of
education at MSU

Strongly 1 2 3 4 5 6 7 Strongly
Disagree Agree

59

10.

60

The advantages of the semester system outweigh the
advantages of the quarter system.

Strongly 1 2 3 4 5 6 7 Strongly
Disagree Agree

 

BIBLIOGRAPHY

 

BIBLIOGRAPHY

Abelson, R. P. (1985). A variance explanation paradox:

When a little is a lot. Esyenelegicei Enlletin, 21, 129-
133.

Allen, M., Hale, J., Mongeau, P., Berkowitz-Stafford, S.,
Stafford, S., Shanahan, W., Agee, P., Dillon, K.,
Jackson, R., & Ray, C. (1990). Testing a model of
message sidedness: Three replications. demmnnieation

Monographs, 51. 275-291-

Altemeyer, B. (1988). Enemies ef fineeden; Mndenstanding
night-wing entnenitarienisn. San Francisco: Jossey-
Bass.

Brehm, J. W. (1966). A t s c o ' al e ctan e.

New York: Academic Press.

Festinger, L. (1957). e0 0 ' 've diss na e.
Stanford: Stanford University Press.

Fine, B. J. (1957). Conclusion-drawing, communicator
credibility, and anxiety as factors in opinion change.

Jounnal of Annennel and Seciel Psycholegy, dd, 369-374.

Hovland, C. I., & Mandell, W. (1952). An experimental
comparison of conclusion-drawing by the communicator and
by the audience. J o 'a

Psycheiegy, 51, 581-588.

Hunter, J. E., Danes, J. E., & Cohen, S. H. (1984).
Matnenenicai nedels of ennitnde ehange, Volune i.

Orlando: Academic Press Inc.

Hunter, J. E. & Schmidt, F. L. (1990). eMethods pf neta-
eneiysis; Cegnegting ezre; end pias in neseangn
findings. New York: Sage.

Jackson, S. & Allen, M. (1987). t - a 5'3 t e
effegtiyeness 9f one-sided and ego-sided ergumentation.

Paper presented to the Annual Meeting of the
International Communication Association, Montreal,
Canada.

61

 

 

62

Johnson, B. T. & Eagly, A. H. (1989). Effects of
involvement on persuasion: A meta—analysis.
Psychological Bulletin, 106, 290-314.

Kardes, F. R. (1988). Spontaneous inference precesses in
advertising: The effects of conclusion omission and
involvement on persuasion. Journal Of Consumer Research,
i5, 225-233.

Kiesler, C. A., Collins, B. E., & Miller, N. (1969).
Attitude change: A critical analysis of theoretical
perspectives. Malabar Florida: Robert E. Krieger
Publishing Company.

Linder, D. E. & Worchel, S. (1970). Opinion change as a
result of effortfully drawing a counterattitudinal
conclusion. Journal of Experimental Social Psychology,

p, 432-448.

Rogers, C. R. (1947). Some observations on the
organization of personality. American Psychologist, 2,
358-368.

Rosenberg, M. (1965). Society and the adolescent self-

image. Princeton, NJ: Princeton University Press.

Rosenthal, R. & Rubin, D. B. (1979). A note on percent
variance explained as a measure of the importance of
effects. Journal of Applied Psychology, 2, 395—396.

Thistlethwaite, D. L., de Haan, H., & Kamenetzky, J.
(1955). The effects of "directive" and "nondirective"
communication procedures on attitudes. Journal of

Abnormal and Social Psychology, pi, 107-113.

Tubbs, S. L. (1968). Explicit versus implicit conclusions
and audience commitment. Speech Monographs, 35, 14-19.

Weiss, W. & Steenbock, S. (1965). The influence on
communication effectiveness of explicitly urging action
and policy consequences. Journal of Experimental Social
Psychology, i, 396-406.