Before exploring the implications of the results outlined above, it would be well to consider any weak nesses apparent in the experiment. The only real problem area involves the measuring instruments used in the experiment. As has already been pointed out, the pre test attitude measure accounted for only 37% of the sum of squares in the post-test attitude measure. The problem here is not serious enough to threaten the experiment, but it urges caution in interpreting results. This caution, in the form of multiple tests of significance and consideration of corollary evidence, has been exercised throughout the experiment and the discussion of results. The pre-test/post-test discrepancy is not the only source of error in the experiment9 however. The measures of message counter and consonant argument are, at best, only reflections of a subject's thoughts during the manipulation. Because of variations in delivery, message duration cannot be considered an entirely accurate measure either. These potential sources measurement error can be expected to multiply in the composite measures of per suasion and message counter and consonant argument, resulting in no less than eight sources of measurement error.
The focus of this error is persuasion, the difference measure constructed by subtracting the already standardized measures of initial attitude from final attitude. The measure is subject to three sources of error in the reliability of the pre-test value scales and the post-test questionnaire, as well as the less than ideal correlation between the two measures. Among the measures regressed on persuasion are initial attitude, duration, and the composite measures of message consonant and counter-argument, all of which are potential sources of error. The focus of this potential error is persuasion and it is there that we see just how potential this error is.
Between reward, initial attitude, reward x initial attitude, duration, consonant minus counter-argument and consonant x counter-argument, 70% of the variance in persuasion can be accounted for. If we make the unlikely assumption that as much as 5% of this variance is spurious, that leaves 35% of the variance unaccounted for. The conclusion that all of this variance is due to error would compliment to the model but ignores factors which are already known to affect persuasibility, including personality, message salience, and problems of history, including selective pre-experimental inoculation due to to provide maximum separation between tasks. Thus pre testing was done in one room, the manipulation performed in a second, and the post-testing done out of sight in a hallway. Subjects were drawn from a community which could be expected to be representative of the entire university rather than communications majors. Finally, the conditions of the experiment were, for the most part, rather relaxed.
The small separation between rewards is a potential weakness of the manipulation. Certainly, as both rewards are low when compared to those offered in many experiments, a certain degree of caution is warranted in comparing these results with those of other experiments. But the success of the manipulation despite the small difference between a low reward of $.50 and a high reward of $1.50 speaks for the strength of the theoretical construct on which the experiment was based, especially when one considers that the differences attributable to reward were often as great or greater than those attributable to initial attitude.
The validity of the test measures was established in the control sample. Although the 150 subjects comprise a relatively small sample for use in a 36 variable step- wise multiple regression procedure, about four subjects for each input variable, the final equation used only 12 variables, leaving a respectable ratio of about 12 subjects per variable. Of particular note was the performance of the Rokeach Value Scales. The multiple correlation with the final attitude measure, .61, is not. perhaps, as high as one might desire, but it is substantial, especially when the indirect nature of the measurement is considered.
Comparability of the results with previous experiments was established using analysis of variance statistics. In many ways the analysis duplicated the multiple regression statistics used later in the results. Analysis of variance is really a generalized version of the multiple regression statistic. But besides providing a reference point for those unfamiliar with multiple regression and path analysis, the statistics served as a check against measurement error and regression effects in the data.
The heart of the manipulation was the experimental model. It was here that all the diverse measurements employed in controlling the experiment were brought together into a cohesive framework. The model itself was based on the research set forward in chapters 2 and 3, and formalized in the fourth and fifth chapter. The larger model, outlined in chapter 4, described a broad portion of the persuasive process. Although it was based on research in distraction and counter-attitudinal advocacy, it may well apply to a much larger portion of the persuasive process than that represented by its forebears. The experimental model was much more limited, testing critical portions of the larger model. Both models, of course, are purely theoretical, but they are based on and make sense in terms of existing research.
The post-experimental version of the smaller model is depicted in Figure 9. The figure is a graphical presentation of Table 9 and is explained in detail in the results. The flow of the variables from independent to dependent is clearly visible. The connections represent the meaningful direct effects discerned in the study. Although the normal cutoff point for direct effects in a path analysis is .r>.05, this criterion of meaningfulness originates in the large scale observational studies which are characteristic of research in sociology rather than the more tightly controlled small to medium sample studies which are characteristic of communication research. For the purposes of this experiment, r>.20 is a more practical cutoff for judging the meaningful ness of direct effects. Thus in Figure 9 arrows connect variables only if the direct effect exceeds .20. In direct effects can be calculated from the diagram by multiplying the direct effect of the affective variable on the mediating variable by the direct effect of the mediating variable on the affected variable.
The model seen in Figure 9 is not the same model that was presented in chapter 5. One of the results of the experiment is an increase in our knowledge of the persuasive process. This knowledge has allowed and to a certain degree, demanded changes in the theoretical framework. It should not be concluded, however, that these changes were statistically motivated. The path diagram is not a statistical construct, but a theoretical construct governed by such rules as are necessary to allow controlled statistical testing and, where the model is adequate, prediction.
A test of the model in its original form (see Figure 8, chapter five) showed certain flaws in its construction. Message counter and consonant argument were obviously far too highly correlated to exist as separate variables in a non-causal sequence. The interaction of counter and consonant argument proved too important an indicator of dissonance to be left out of the model. As an indicator of biased scanning, the interaction of counter and consonant argument proved too important to ignore. Message duration proved to be more than a simple control variable. Adjustments were made in the model to account for these factors and the revised model in Figure 9 is the result.
The revised model differs from its original in two ways. First, it is more complete, accounting for the effects of selected interactions. This accounting would have been made on a non-selective basis in an analysis of variance. Second, it is more parsimonious, with only those variables which could truly be called exogenous enjoying that designation and the joint and spurious effects of the variables reduced to zero.. Neither model could be said to be more in accordance with the theoretical construct presented in chapters 2, 3,. and 4, although the revised model represents . the concepts underlying the theoretical construct more fully.
This last point is important for, although there undoubtedly is a unique best model of the process of persuasion, the experimental model is only of limited scope and accuracy. Indeed, although most of the causal flow depicted in Figure 9 will probably stand up to the test of time, there are questions that need to be answered before it can be wholly accepted. Of particular concern is the location of duration in the causal flow. The path analysis seemed to support the flow indicated in the figure, but the simple correlation of group means for consonant argument and duration argues that duration may follow consonant argument in the causal flow. Certainly, it can be safely assumed that increased duration does not cause increased message consonant argument.
Although the model is consistent with theory in persuasion it does not fully accept any of the traditional views of the counter-attitudinal process. The central hypothesis of incentive theory, a positive relationship between reward, effort, and persuasion, received no support in the presence of dissonance. In the non-dissonant conditions the direction of the persuasion was right, but was due not to persuasion in the high reward condition, but to dissuasion in the low reward condition. The biased scanning hypothesis, by comparison, fared quite well, although the workings of the mechanism are described somewhat differently than they were originally. The hypothesis of cognitive dissonance, that reward was negatively related to persuasion in the presence of dissonance, received considerable support, but the reason for the persuasion, a search for a justification of action and attitude, was not supported. Only initial attitude significantly affected consonant argument. Dissonance affected counter-argument, creating support for cognitive distraction and biased scanning as the mechanisms underlying persuasion due to unrelieved cognitive dissonance. It can be said, then, that the model combines the best components of several theoretical viewpoints into a larger and more comprehensive theory.
In reality, the only experimental result that was not totally in harmony with the larger construct outlined in the early chapters was the dissuasion in the low reward, positive initial attitude group. The construct did not allow for dissuasion in an attitudinal encoding situation. It appears, however, that poorly rewarding an encoder with a positive initial attitude can result in poor biased scanning. It should be imagined that this can only occur where personal commitment is not high. Using reward to dissuade a missionary is unlikely to have the desired effect. The missionary likely sees other rewards in attitudinal encoding. But where there is neither a missionary type commitment nor a dissonant aversion to a message, low reward may well have a deleterious effect on attitude.
The evidence supporting the poor biased scanning explanation is not the best offered by this experiment. It is however, very consistent and highly significant. Although the dissuasion is, at least in terms of this study, very real, the phenomenon demands further re search in which depth of conviction is measured along with the reward and initial attitude variables. Despite the chapter three claims to the contrary, the dissuasion in this experiment is not an isolated incident, even among the experiments reported. Nuttin (1966) notes a similar dissuasion pattern in his experiment. Close examination of Janis and Gilmore (1965) indicates, moreover, that the persuasion reported in their positive sponsor conditions may actually have been dissuasion due to poor biased scanning in their negative sponsor conditions. This evidence serves notice not only of the reality of dissuasion due to poor biased scanning, but of the dangers of tunnel vision in the early stages of research.
The defeat of the hypothesis that any message encoding situation will yield persuasion in the direction of argument is not a major defeat for the theoretical construct. In fact, it tightens the theory. In the presence of the hypothesis it was impossible to say whether the cognitive distraction was due to the effort of encoding the message or the dissonance caused by the disparity between action and attitude.
With the hypothesis discarded the distraction can be attributed to the dissonance. It is noteworthy that biased scanning may have two edges, that one can scan poorly and be affected by that act.
Overall, the results give a little more insight into how attitudes fluctuate and can be controlled. It was a little scary to think that an employer would get the most loyalty from the employee he pays least. Now one can be encouraged in the knowledge that after winning loyalty with low salary, an employer will have to pay to keep that loyalty. The results also point up the near perfection of the Judeo-Christian, Moslem message of salvation. To the non-believer, the message holds a low reward, allowing persuasion through unrelieved dissonance, Upon conversion, however, the message, salvation, becomes a high reward, encouraging maintenance.
This interpretation is supported in the pre-study results forwarded in chapter
6. Respondents, when filling out the Rokeach Value Scales, almost invariably
rank salvation either first or last. The correlation between this ranking and
church attendance exceeds r>.90. Apparently, then, the converted rank religion's
message high while the unconverted do the opposite. Thus the message for those
who wish to remain agnostic is 'Beware of he who comes bearing gifts of salvation.
Foulger, Davis A. (2005). Discussion. From the Hypermedia
Edition of Foulger, Davis A. (1977). Distraction and Dissonance: A model
of the persuasive process. Master's Thesis. University
of Central Florida. Retrieved from http://foulger.info/davis/mastersThesis/chapter8.htm.