An examination of the literature concerning the effect of message variables on attitude change shows little research or theory dealing with sustained attitude change. Most theory and research considers only immediate effects of message variables on attitude. An exception is the work of McGuire and his associates on belief immunization.¹ Their research suggests that building an initial message which takes into account what will transpire subsequent to exposure to that message can increase long-term effect. Specifically, refutation of arguments that are the same or similar to arguments to which a receiver will be exposed later has been found to reduce the impact of subsequent arguments on the receiver's attitudes. Hence, we can advise the communicator who wishes to have a long-range impact to use the so-called "two-sided" message, the type which includes refutation of counter-arguments.

Unfortunately, this is about all the advice, based on research findings, that we are in a position to give such a communicator. While considerable information has accumulated over the past two decades which suggests the effect of various message elements in producing immediate attitude change, almost no research has considered the long-range impact of those variables. Previous research which I reported suggests that the variable of "evidence" usage is important in determining the long-range impact of a communicative effort.² In these studies, "evidence" frequently was found to have no immediate impact, depending upon other circumstances in the communication transaction, but consistently was found to have a positive impact on sustained attitude change over a period of three to seven weeks. No explanation of these long-range effects was provided by the previous research, primarily because those studies were not designed to provide such an explanation. In each case the research merely involved presenting a message to subjects, measuring their attitude, and measuring the attitude again three to seven weeks later. Subsequent to initial exposure to the message, no manipulation was introduced in any of the studies. There was no systematic attempt to introduce counterpersuasion into the experimental environment. However, counterpersuasion may have been introduced in the nature environment in which the experimental subject found himself during the three- to seven-week delay period. Thus, on the basis of the previous studies we do not know whether the evidence served as an immunizing agent against counterpersuasion, or whether the effect was produced in some other way. The study reported below is the first of a planned series of studies designed to determine why "evidence" has sustained impact on attitude change.

At least two credible explanations may be advanced to account for the effect of evidence on sustained attitude change. The first of these has been suggested by Hovland, Janis, and Kelley. In an attempt to explain an observed "sleeper effect," Hovland et al. hypothesized that a source with low credibility may have more impact over time than he does initially because receivers tend to dissociate the source from the message over time.³ Thus, while a source's low credibility can interfere with the production of attitude change while he is remembered, when he is forgotten, his message may have its intended impact to a greater extent. In the words of Hovland et al.,

This dissociation tendency would be maximal when the communication contains arguments and evidence which can be evaluated on their own merits and are likely to be recalled without bringing the source to mind.⁴

This explanation suggests that whether a source is highly credible or not, if he is forgotten the long-range impact of his message will tend to depend upon the arguments and the evidence which he advanced. Thus, if no evidence were advanced, it would be reasonable to hypothesize that the effect on sustained attitude change would be less than if evidence were included, and that this would be true for all sources regardless of their credibility. This hypothesis, in fact, has been supported extensively but by pervious research.

A basic problem with this explanation, however, is that it does not take into account the other communication behaviors of the receiver between the time that he is exposed to the message and the time of delayed attitude measurement. If a person has been exposed to a persuasive attempt of any magnitude, he is not likely simply to ignore that fact for a period of seven weeks. Rather, he is very likely to interact with his peers about the persuasive attempt. In addition, it is common for a person subsequently to be exposed to a persuader taking a position contrary to that taken by the initial communicator. In short, a variety of communication transactions may occur during the post-exposure period before delayed measurement, and the probability is high that one or more relevant events will occur.

The work of McGuire et al. indicates that refutation of counter-arguments is central to the development of belief immunization. This finding suggests another explanation for evidence's long-range impact. When a receiver is exposed to an influence attempt by a communicator, he goes through a mental process of testing what the communicator says against what he already knows or believes. This is akin to mental refutation of the communicator. In an interpersonal communication transaction, the receiver may actually become a source and verbalize his refutation of the other communicator's positions. But whether the refutation remains mental or becomes verbal may not be central to its effect on the refuter. If the refutation appears inadequate to him, he is likely to alter his beliefs in the direction advocated by the communicator. On the other hand, if he perceives his mental or verbal refutation to be adequate, he is likely to reject the views of the communicator. This rationale leads to the hypothesis that receivers will be less influenced by a counterpersuader if the previous communicator has provided evidence for the receiver to use in support of refutation of the counterpersuader. If this hypothesis is correct, inclusion of evidence by an initial communicator will enhance the probability that the receiver will maintain his attitude rather than change it in the direction advocated by a counterpersuader.

The purpose of the study reported in this paper was to test the validity of the second explanation for the long-range impact of evidence. In this study, the time lapse between the initial persuasive message and final measurement of attitude was only a few minutes. If the Hovland et al. explanation is valid, no effect on attitude attributable to inclusion of evidence would be expected because the time lapse would be insufficient for the receivers to dissociate source and content. However, if the explanation based on the work of McGuire et al. is valid, there should be an immunization effect. The latter explanation is not dependent on a time lapse between persuasive attempt and attitude measurement. Specifically, it was hypothesized that subjects will be less affected by counterpersuasion from a second speaker if the first speaker's message contains evidence than they will be if the first speaker's message does not contain evidence.

The primary independent variable in this study was the inclusion or exclusion of evidence in the initial message. However, in order to increase generalizability, several additional variables were introduced into the design. Because credibility had been found in studies of immediate impact to interact with evidence, credibility of the sources of both the initial and the counterpersuasive message was manipulated. However, because credibility was not of primary concern in this investigation, the credibility for the two sources in any one treatment condition was approximately the same, i.e., both high or both low. Since conflicting messages were necessary to the design of this study, one pro and one con, it was possible that the effects could vary depending upon which message was presented first. Thus, the order of presentation was counterbalanced in the design of the study. Finally, although the primary concern of the present study was the effect of the evidence included in the initial message, the use of evidence in the counterpersuasive message was also manipulated to determine whether an interaction was present. Thus, the design of the study included four independent variables, each with two levels: evidence usage of first speaker (included or not included), credibility of first and second speakers (high or low), order of presentation of the experimental messages (pro first or con first), and use of evidence by second speaker (used or not used).

The topic chosen for the experimental messages was "federal control of education." Messages which advocate federal control of education and in which evidence was manipulated had been developed for previous studies.⁵ New messages opposing federal control of education were developed. The new messages were somewhat shorter than the original messages, but included the same proportion of evidence to total content. The "evidence" version of the message was created first, and then that evidence was generalized (i.e. "56%" became "a majority") and all citations of sources were omitted in order to create the "no evidence" version.

The subjects were 264 college students enrolled in a basic public speaking course at Michigan State University during the fall term, 1968. They were distributed approximately equally across the 16 possible conditions. The subjects were only available as intact classes. Thus, each class represented one experimental condition. The classes were assigned to experimental conditions randomly.

Attitude toward the concept was measured by six semantic differential type scales that had been developed previously for this purpose.⁶ source credibility was measured by six semantic differential type scales for each of two dimensions, authoritativeness and character.⁷

Each experimental condition was administered during a single class period. Students were given individual packets which included instructions, several semantic differential type instruments including the federal control of education pretest, a credibility induction for the first speaker, credibility scales as a pretest for the first speaker, attitude and credibility scales to be completed after the first speech, a credibility induction for the second speaker, credibility scales to serve as a pretest for the second speaker, and attitude and credibility scales to serve as a posttest following exposure to the second speech. The students were not asked to put their names on the packets. After the subjects had completed the pretest, an audio-tape of the first message was played. After completion of this tape, the subjects were instructed to complete the second set of tests, read the second credibility induction, and respond to the scales appropriate to that induction. When this was completed, the second taped message was played, followed by the completion of the final scales in the packet.

Analysis of the pretest attitude data showed significant differences among the various experimental groups. To control for these differences, analysis of covariance was employed for the posttest attitude data with the pretest attitude serving as the covariate. Analysis of the pretest credibility data for both the first and second speakers indicated that the credibility manipulation succeeded for both pro and con messages. Therefore, pretest-posttest credibility change scores were analyzed by analysis of variance. The attitude scores for the first posttest were subjected to two separate analyses of covariance with two factors (credibility, evidence usage): one analysis for the subjects who heard the pro message first and one for those who heard the con message first. The second attitude posttest scores, separated into pro and con conditions were subjected to analysis of covariance with three factors (the same as above plus evidence usage of second speaker). The credibility change scores for the first speaker were subjected to three-factor analyses of variance (credibility, evidence usage, pro-con message). The credibility change scores for the second speaker were analyzed by four-factor analyses of variance (the same factors as above plus evidence usage of second speaker). The .05 level was set for significance on all tests.

Although the data obtained on the measures taken immediately after exposure to the first speaker were not relevant to the primary hypothesis of this study, these data were analyzed in order to determine comparability of results in this study with results of earlier studies and to compare the effectiveness of the pro and the con messages. In the earlier studies, the pro messages, in combination with the credibility inductions employed, generally produced an interaction between credibility and evidence. Specifically, evidence was found to increase immediate attitude change in the low credibility condition but to have no significant effect in the high credibility condition. Similarly, the evidence version of the message was found to increase credibility in the low credibility condition but to have no impact in the high credibility condition.

The analysis of the attitude data in the present study showed a significant interaction for both the pro message and the con message between initial credibility and evidence usage (pro F = 4.22, p < .05; con F = 9.82, p < .05). The nature of the interaction was comparable to that obtained in earlier studies. (See Table 1.)

*Range of possible scores was 6-42. Grand mean on pretest was 27.6.

The results of the analysis of the credibility ratings of the first speaker were less comparable to earlier studies. The three-factor analysis of the change scores on the authoritativeness dimension showed two significant min effects, on initial credibility (F = 37.6, p < .05) and evidence usage (F = 10.9, p < .05). The expected interaction between initial credibility and evidence usage was not significant (F < 1). The significant effect for the initial credibility was attributable to the highly credible source's tendency to lose credibility and the less credible source's tendency to increase credibility as a result of subjects' exposure to the message. The significant evidence effect is attributable to the highly credible source's tendency to lose less credibility when including evidence and the less credible source's tendency to increase credibility more by including evidence. (See Table 2.)

The three-factor analysis of variance of character change scores showed only one significant effect, an initial credibility by evidence by direction-of-message interaction (F = 4.1, p < .05). An examination of the mean change scores for the various conditions (Table 2) indicated that the expected credibility by evidence interaction was present for the pro message but not for the con message.

The data bearing directly on the primary hypothesis for this study were the attitude scores obtained after exposure to the second message. The separate analyses for the subjects exposed in the pro-con order and those exposed in the con-pro order produced nearly identical results. In both cases two significant effects were observed, evidence of first speaker and credibility by evidence of second speaker interaction (pro-con order: evidence of first speaker, F = 7.1, p < .05; credibility by evidence of second speaker, F = 4.0, p < .05; con-pro order: evidence of first speaker, F = 8.8, p < .05; credibility by evidence of second speaker, F = 4.3, p < .05).

In every case the subjects who had been exposed to an initial evidence message followed by a counterpersuasive message recorded attitudes more in line with the intent of the initial speaker than those subjects exposed to an initial no-evidence message. (See Table 3.) These results were consistent with the primary hypothesis of this investigation.

*The higher the score, the more consistent with intent of first speaker.

**The lower the score, the more consistent with intent of first speaker.

The credibility by evidence usage of second speaker interaction was of the same type for the first speaker. That is, inclusion of evidence enhanced the effect of the second speaker who was initially less credible, but had no effect when the second speaker was initially highly credible. While this result was only of secondary interest in the current investigation, it does suggest the increased generalizability of the results found in the first part of this study and in previous studies. That is, it would appear that any speaker with moderate-to-low credibility may benefit by the use of evidence whether he is a first speaker or a second speaker, but use of evidence by a highly credible source would appear to have little or no effect on immediate attitude change.

The final data analyzed were those obtained on the credibility posttest measures for the second speaker. No significant effects due to message order were observed. However, two significant effects were observed on both dimensions of credibility. These effects were evidence of second speaker (authoritativeness F = 5.5, p < .05; character F = 4.0, p < .05) and a credibility by evidence of first speaker by evidence of second speaker interaction (authoritativeness F = 4.8, p < .05; character F = 4.6, p < .05). An examination of the data reported in Table 4 indicates that, in general, inclusion of evidence by a second speaker increases his credibility. However, as is suggested by the significant interaction, the effect of evidence used by the second speaker is more complex than that. Specifically, a less credible source tends to gain credibility by including evidence whether he is following either an evidence or no-evidence first speech. But a highly credible source who includes evidence following a no-evidence speech seems to profit little. The important effect for the highly credible source is that if he does not include evidence following an evidence speech, he tends to lose credibility. More simply, a source with low credibility is perceived as more credible whenever he uses evidence. A highly credible source, who does not use evidence when a preceding speaker has, tends to be derogated.

The most important finding of the current investigation is that evidence appeared to serve as an inhibitor to counterpersuasion, as was hypothesized. In addition, the explanation for this effect of evidence, contrary to the explanation suggested by Hovland, did not depend on a time lapse in which receivers could go through a process of dissociation of source and content. In this study, no time lapse was permitted prior to counterpersuasion other than the time necessary to complete the measures between messages.

Caution should be exercised in generalizing fro the results obtained in this study. Since both the initial and counterpersuasive messages were in the form of public speeches, the study suggests that these results would generalize to the normal debate confrontation. A more typical type of communication transaction subsequent to exposure to an initial message, however, is interaction among peers. Whether the observed results would generalize to that type of circumstance is unknown and remains for later research to discover. The primary conclusion, which would appear appropriate to draw on the basis of the present study in conjunction with earlier reported studies, is that evidence appears to be an important variable in persuasion. While in some cases evidence may not have an impact on immediate attitude change, it does seem to have a predictable impact as an inhibitor of counterpersuasion.

1. See, for example, W. J. McGuire, "The Effectiveness of Supportive and Refutational Defenses in Immunizing and Restoring Beliefs Against Persuasion," Sociometry, XXIV (1961), 184-197.

2. James C. McCroskey, "A Summary of Experimental Research on the Effects of Evidence in Persuasive Communication," QJS, LV (1969), 169-176.

3. Carl I. Hovland, Irving L. Janis, and Harold H. Kelley, Communication and Persuasion (New Haven, 1953), pp. 28-281.

4. Ibid., p. 280.

5. James C. McCroskey, "Experimental Studies of the Effects of Ethos and Evidence in Persuasive Communication" (unpublished D.Ed. dissertation, Pennsylvania State University, 1966).

6. Ibid. Scales employed were harmful-beneficial, good-bad, right-wrong, unfair-fair, negative-positive, and wise-foolish.

7. James C. McCroskey, "Scales for the Measurement of Ethos," SM XXXIII (1966), 65-72.

Click Here To Go Back To PERIODICALS