Distraction and Dissonance: A model of the persuasive process

Chapter VII - Results

As the analysis of the experimental data was dependent on the results of the control manipulation, the control data was analyzed first. The most immediate concern was over the effect questionnaire order had on attitude. It was felt that any effect could have pro found implications on both the control and experimental data. No significant or near significant effects of order were found, however. This result bolstered the manipulations by demonstrating several things. First, it showed that the pre-test and post-test instruments had no effect on each other. Second, it demonstrated that the list of arguments given subjects during the manipulations was not persuasive in and of itself, that persuasion in the experiment could not be attributed to this list of arguments. Third, it was shown that the measure of counter-argument had no persuasive impact of its own. This result was later confirmed in the experiment, where a similar manipulation yielded the same result.

Satisfied that the experiment was shown reasonably pure of confounding effects, the experimenter turned to the concern of developing a good predictor of pre-test attitude. Here, the first responsibility involved increasing sensitivity by developing a composite attitude measure. It was felt that five questions on the post-test instrument related directly to the topic of the persuasive message. To clarify these relationships the questions were compared using principle component analysis. The analysis yielded two components which were then obliquely rotated. Examination of these components indicated that the second was not meaningful as a post-test measure. The first component, as can be seen in Table 1, was very meaningful.

Component one involves three questions from the post-test measure. Two, positively related to the component, advocate increasing spending on social pro grams. The third, negatively related to the component would seem to involve spending priorities. Subjects who advocate increased military spending do not advocate increased social spending, and vice versa. This interpretation is, of course, highly consistent with the topic used in the experiment, that 'military spending should be decreased in order to allow greater spending on social programs.' The second component involved two questions, both of which would appear to support the increased social spending perspective. It can be argued, though, that educational quality can be improved and welfare systems shifted to jobs systems, without a major spending increase. Indeed, such shifts could involve a decrease in costs. Component two appears to reflect this perspective.

On the basis of this analysis, component scores were computed for the first component. The scores, the composite measure of attitude served as the basis for the final phase of analysis on the control data. The purpose of this analysis was to devise a system for re creating these component scores, first from the post- test and, second, from the pre-test. The method used in formulating both predictive equations was forward (stepwise) multiple regression.

As the component scores were derived from the post-test survey, the correlation between the factor scores and the post-test attitude regression predictions were assured of identity. As shown in Table 2, this was the case. The three principle questions correlate .99 with the component scores, indicating that the two questions not attributed to the component have, in fact, little impact. Predicting the component scores from the post- test proved more difficult. According to Rokeach (1968) an attitude is a system of beliefs and values. Theoretically, by tapping the values belonging to an attitude's belief-value system, the attitude can be predicted. The value component of one such system, one surrounding social and military spending priorities, is recounted in Table 3. Here the system is presented in the form of a multiple regression equation.

The table actually shows two equations. The first, titled standardized partial regression coefficients (betas) shows the relative importance of the variables to the equation. Higher standardized betas indicate the greater importance of a variable as a predictor of the dependent variable. The second equation, titled unstandardized partial regression coefficients, is the actual prediction equation.

The 11 values which form the value system represent about one third of the Rokeach Values. The entire value scales can account for about 40% of the variance in attitude on the post-test measure. The 11 variables in the equation correlate almost as well; their .61 multiple correlation indicating a 37% prediction accuracy. The 3% difference in prediction accuracy is small, especially when it is considered that the additional variance is accounted for by 25 variables. The 11 variable equation has, moreover, a lower standard error than other competing equations.

The components of the equation make sense as the value system surrounding attitudes on social and military spending priorities. The meaningfulness of the first value, national security, should be rather obvious. It should be expected that a person concerned about national security would also favor military spending. A predisposition toward military spending does not, however, preclude a favorable attitude toward social spending. This can be clarified, however, by consideration of the next two variables, equality and freedom. Rokeach (1973) demonstrates the importance of these two values in determining political ideology. In terms of American politics, liberalism and conservatism can be predicted on the basis of the ranking of equality, liberalism increasing as equality becomes more important. A liberal, in American politics, would be expected to favor social priorities while a conservative would be expected to favor military priorities. Rankings of freedom should further clarify this liberal/conservative breakdown. Low rankings of freedom should indicate extreme feelings, leanings past liberalism toward communism and past conservatism toward fascism.

The fourth and fifth variables in the equation, true friendship, and honesty, although not directly associated with the dependent measure, are meaningful predictors. Their effect is felt through the suppressor relationships they have with the variables already in the equation, notably national security. True friendship and national security, considered together, would seem to form an index of defensiveness or trust, with security increasing as friendship decreases. The nature of this suppressor relationship can be examined in Table 4. True friendship is correlated only .04 with the dependent measure, but is correlated -.26 with national security, the most important variable in the equation. The result, as shown in Table 3, is its standardized beta, a direct effect equivalent to a .18 correlation.

Honesty's suppressor relationship is also with national security. The importance of security increases as the importance of honesty decreases. It would appear that people who feel honesty unimportant probably think others feel the same way. Ultimately, then, a way of defending oneself from the dishonesty of others (whether the other is another person or another nation) is through a strong national security. The next variable, independence, can be grouped with honesty because, again, the variables suppress one another. Together, these variables measure part of what may be the Machiavellianism, Dogmatism, Intolerance of Ambiguity personality variable group (Rokeach, 1973). It would be expected that as dogmatism increased the prevailing national thought pattern, which, in the United States at this point in time, seems to favor military over social spending, would take priority.

Pleasure is related to freedom and equality, particularly equality, and draws additional strength from these variables. Pleasure appears to be related to political ideology, as is a world at peace. This effect is apparent not only in this study but in Rokeach's (1973) study of the effect of value change on political ideology. The effect of these variables is to clarify the attitude prediction due to political ideology.

A world of beauty is perhaps the least correlated variable in the entire value system. It can work alone as a predictor. A world where military spending could be easily reduced so that social spending could be increased would indeed be a more beautiful world. As such, the prediction due to a world of beauty could be said to be a measure of idealism, of a utopian ethic. Rokeach (1973) shows that concern with this value declines with age, a hint that the generation gap of the late 1960's and early 1970's, which made itself most clearly in terms of attitude toward the military, may have been due in part to differences in idealism rather than to differences in political ideology.

The three remaining variables in the equation, obedient, cheerful, and clean, might be referred to as boy scout values. Their function appears to be that of clarifying variables which are already in the system. obedient and cheerful are both suppressed by independent. As such they can be interpreted as contributing to Machiavellianism/Dogmatism prediction. This interpretation is consistent with the findings summarized by Rokeach (1973), where both obedience and cheerful were predictors of these and other related personality measures. Cheerful and clean seemingly are suppressed by value rankings of a world of beauty. From a logical standpoint, as both of these variables involve modes of beauty, they may act to reinforce the utopian concept discussed earlier.

At any rate, cheer and cleanliness are ideal end states. The value system, considered as a whole, can be said to reflect four components of attitudes on social and military spending priorities. The first rotates around political ideology, with liberalism associated with social spending priorities. The second involves trust and the need for security against the supposed dishonesty of others. The third involves idealism, orientation toward a better world. The last reflects personality as measured in dogmatism, machiavellianism, authoritarianism, intolerance of ambiguity, and other related personality measures. The 37% accuracy with which this values system can measure attitudes on social and military spending priorities is not, perhaps, optimal. One would prefer such a measure to predict better than this does. It is, however, sufficient to the task of making a pre-test prediction. On balance, when one considers that the value scales measure only 36 values out of a pool of over 100 different values, when one remembers that the value system is only a part of a much larger belief system, when one recalls that the opinions of friends and family can also affect attitude, the predictive accuracy of these few values is rather remarkable.

With the analysis of the control survey data complete, it was possible to turn to the task of analyzing the actual experimental data. The first step in this analysis involved determining whether either initial attitude or the amount of money offered influenced subjects' decision to encode the message. It had been expected that both reward and pre-test attitude would affect the subject's decision to encode the message, with those in the more dissonant conditions, low reward and negative attitude, less likely to encode the message. No support was generated for these expectations. The F-ratios for both main effects and the interaction were less than one.

Using the data from the 39 subjects who actually encoded the message, the effects of the actual manipulation were tested. Although regression analysis would be used in subsequent analysis and would have yielded more sensitive measurements in this case, the test was made using analysis of variance with initial attitude split into two groups at the mean. The resulting 2 x 2 (initial attitude x reward) analysis of variance tested for differences in post-test attitude. If there had been no persuasion, a main effect would have been expected for initial attitude. Developing expectations of what the analysis would look like were there no persuasion was more difficult. In fact, no main effect for initial attitude would have been a meaningful indicator of attitude change. The prime expectations were reserved for reward, and the interaction. The theoretical construct would predict, for subjects with a positive initial attitude (message consonant subjects) a slightly higher post test attitude in the high reward condition. subjects with negative initial attitudes (message dissonant subjects) on the other hand, should have much higher post- test attitudes in the low reward than would be found in the high reward condition. This was expected to show itself in a near significant effect for reward and a highly significant interaction effect.

Table 5 shows the results of this analysis of variance. The effect of initial attitude is non-significant, indicating that there was substantial persuasion as a result of the manipulation. In contrast, the interaction is highly significant. This indicates that the persuasion occurred in the predicted directions. But reward makes no approach of significance. This indicates an unexpected balance in persuasion, with high reward equally different from low reward, although in opposite directions, regardless of initial position. The non significant difference on initial attitude, however, erases the possibility of this effect being due to persuasion in the positive initial attitude conditions.

Examination of the means, shown in Table 6, confirms this interpretation. Expected means for post-test attitude in both the positive and negative initial attitude conditions were determined from the control sample.

Comparison of these means with the experimental means shows, first, that there was very little attitude change for high reward in either initial attitude condition, and
second, that there was a great deal of attitude change for low reward in both attitude conditions. For the negative initial attitude low reward condition, this change indicated persuasion. But for subjects with positive initial attitudes, low reward provides a major and unexpected surprise. Subjects were actually dissuaded.

It was tempting to view this result as spurious, so the result was analyzed more closely. Means for persuasion, which was operationally defined as final attitude minus initial attitude were therefore calculated for the four cells of the experiment and the two control conditions. The control conditions in Table 7 show means equivalent to no change in both conditions. Both high reward conditions showed a non-significant amount of persuasion. Negative initial attitude low reward subjects were highly persuaded. Both of these results are highly consistent with expectations. Once again, however, the positive initial attitude low reward condition is, contrary to expectations, highly dissuaded. The analysis of variance for this cell breakdown, shown in Table 8, confirms the significance of the persuasion. The interaction is once again significant. This is no surprise and is meaningful only in that it underlines the fact that most movement is in the low reward conditions. But, while there are no significant differences in reward, there are significant differences in initial attitude conditions, indicating first that attitude movement is balanced in reward conditions, and second that this movement is in opposite directions.

Although these results seemed to support the dissuasion phenomena, it was possible that the effect was due to the extreme scores of a few individuals in the experimental group. Examination of the individual scores does not support this interpretation. None of 11 subjects in the positive initial attitude group were dissuaded. Neither of the persuaded subjects were highly persuaded, but eight of the dissuaded subjects were.

The best conclusion that can be reached concerning this dissuasion effect is that it is real. The effect holds true both in comparisons with non-experimental control subjects and with the subjects' own initial attitude. It cannot, moreover, be localized to only a portion of the experimental group. Ultimately these results leave us with an exclamation point and a question mark. First, the hypothesis that persuasion in the negative initial attitude conditions could be explained in terms of traditional dissonance hypothesis was supported. But the second hypothesis, that persuasion in positive initial attitude conditions could be explained by standard persuasion variables (an incentive hypothesis), while sup ported, leaves us with a question. What could possibly explain dissuasion in an encoding situation? An interesting answer forwards itself, but it would be better to hold the problems of the question mark until the implications of the exclamation point have been explored more fully.

The overall success of the basic experimental manipulation allows exploration of the experimental model. This model, as presented in chapter 5 and summarized in Figure 8 is a four level causal structure flowing from independent (exogenous) to dependent (endogenous) variables. The experimental model involves two causal and two non-causal (control) exogenous variables. The first of these variables, which was labeled post-test order effects was significantly correlated with only one variable, duration, a fellow non-causal variable. It can be said without question that this is a spurious effect as the message (and its duration) preceded the randomly assigned post-test in time.

Duration, the second non-causal exogenous variable was significantly related to both initial and final attitude. Because of this relationship, message duration was re-assigned to the second level of the causal structure as an endogenous variable. The effect, although not hypothesized, is not surprising. The duration of a persuasive message should be expected to influence the effectiveness of that message. But the nature of the effect is not what one would expect. Discussion of this effect will be reserved briefly.

Implicit to the manipulation of reward and initial attitude is the manipulation of a third variable, their interaction. In path analysis, treatment of a variable interaction as an element of the causal flow violates the techniques assumption that the relationships between the variables are additive. As interactions are essentially multiplicative, the effect of this assumption is to prohibit attempts to predict interactions additively from the interacting variables. Such an attempt would make little sense, as interactions are generally relatively orthogonal to the interactees. This orthogonality should preclude additive prediction in any case. The interaction remains a manipulated variable, however, and as such may have importance as a predictor of subsequent variables. This is especially true in the encoding situation.

Where the interaction is placed at the same inclusion level as the interacting variables, however, there is no implicit violation of the path assumptions. First, there is no attempt to additively predict the interaction from the interacting variables. Second, there is no problem of shared residuals, as the variables are reasonably orthogonal to one another. Third, there is no violation of the assumption that changes in system variables always occur as a linear function of changes in other variables. Rather, this assumption is strengthened with the corollary that variables at the same inclusion level are linear functions only of those variables at preceding inclusion levels, an assumption already enunciated in the path assumption of a systems having no reciprocal causations or feedback loops. Finally, it should be reasonably undebatable" that there is no causal relation ship between a variable and its interaction. Rather, they are separate variables which co-vary only insofar as a change in one implies a change in the other. The changes, although mutually occurring, do not imply correlation between the variables.

Two interaction effects will be examined in the experimental model. As both interactions, the interaction of reward and initial attitude, and the interaction of message counter and consonant argument, are both examined at the same inclusion level as and are reasonably orthogonal to, their components, there is no weakening of the path assumptions. The inclusion of these two interactions, as well as the aforementioned shift of duration from exogenous to endogenous status, should be a good hint that the experimental model did not survive intact. This is indeed the case, and before presenting the model as supported in the data, it would be well to examine the reasons for the changes that were made.

The interaction of reward and initial attitude was added because the principle effect of these two variables on persuasion is in their joint effect. The combination of dissonance and incentive predictions gives maximum persuasion to the cross-products, not the main effects. This is not to say that reward and attitude do not have individual effects. As was seen in the analysis of variance and will be seen below, they do. Rather, the prediction is such that the interaction cannot be ignored.

The next set of changes resulted from the reciprocal effects of counter and consonant argument. The correlations between the two variables were such that it was impossible to either reject causation between the variables or establish a non-reciprocal causal flow between the variables. The only solution to this problem was to create a composite message argument variable. This difference variable (message counter-argument was subtracted from message consonant argument) was highly successful. It correlated highly with both component variables as well as accounting for nearly all of the variance accounted for by the two variables in multiple regressions on persuasion and final attitude. The interaction of the two variables accounted for additional variance and, as we shall see later, provides a key explanation of the question mark effect noted above.

The only other change in the model was, as has been explained, message durations shift from exogenous to endogenous status. This leaves three variables at the first (exogenous) inclusion level. The relationships between these variables, as given in the system, need not be explained causally. There are, however, relationships among these variables. Reward and the interaction are correlated at the .02 level, which is hardly cause for concern, and the interaction is non-significantly correlated .21 with initial attitude. Reward and initial attitude, however, are correlated -.31. This correlation is not only significant, (L=4, df=l/37, p<.O5), but significant in the wrong direction. Apparently there was a tendency for those in the high reward condition to have somewhat lower initial attitudes than those in the low reward condition. This correspondence could have had an effect in obscuring significance in the variables. Low reward subjects were expected to move to a higher final attitude, especially where their initial attitudes were low. Higher initial attitudes in the low reward condition could have obscured this persuasion, an effect that could be expected to evidence itself in suppressing reward and initial attitude. This suppression, although evident in the analysis, never proved to be a difficult problem.

Manipulation of the exogenous variable system should effect changes in the rest of the model, The most immediate effects in this system should be on the actual act of encoding, the second level of the experimental model. The second level, as originally outlined was composed of counter and consonant argument. These components, for reasons outlined above, were converted into interaction and difference variables. In addition, the variable message duration was moved to a position between the first and second inclusion levels. In Table 9, a summary of the entire path analysis, message duration is the first dependent variable analyzed. Three variables, reward, initial attitude, and reward x initial attitude, pre cede it in the system. Of these, only one, initial attitude, is a significant predictor (f=5.17, df=l/35, p=.05) of message duration. The direct effect of initial V attitude is represented by its standardized partial regression co-efficient, 137, a figure which compares well with its zero order correlation of .42, indicating that subjects with positive initial attitudes spoke longer. As message duration is the first endogenous variable in the system, there are no indirect effects at its level. The multiple correlation of the three exogenous variables with duration is .44.

The next level involves the interaction and difference components of counter and consonant argument. The two variables are correlated -.01, underscoring the lack of causal relationship. The interaction of message counter and consonant argument is significantly predicted y only one variable, message duration, (f=4, 15, df=l/34, p<.05) but the indirect effect of initial attitude through duration allows the otherwise weak effect of initial attitude to become a trending direct effect of .30 (f=2.69, df=l/34, p=.11). The multiple correlation of the pre ceding system variables on message counter x consonant argument is only .37, an indication that these variables account for only a small portion of the variance in the interaction.

The difference measure of counter and consonant argument, called argument in Table 9, is predicted by all of the three exogenous measures, but is not related to the fourth preceding measure, message duration. The strongest predictor of argument is initial attitude, (f=5.65, df=l/35, p<.025) a variable whose total effect of .35 is composed largely of its direct effect. reward x initial attitude is the second significant predictor (f=4.l0, df=l/35, p<.05). The effect of reward alone results in a trend (f=3.2, df=l/35, p=.08) which may be more important than its significance level would indicate.

Persuasion is the major interchange of the model. It is here that the threads of direct and indirect effects take action in modifying attitude. Up to this interchange, indirect effects, the effects variable A has on variable C by way of its effect on variable B, have been minor. Here they take on a great deal of importance. Because of this, the results are rather complex. The reader is directed to Table 9, where these results are clearly summarized. Reward, with a direct effect of .01, has no direct influence on persuasion in the model.
The direct effects of initial attitude and initial attitude x reward, -.60 and .31, respectively, are significant, These results are highly consistent with those obtained in the analysis of variance depicted in Figure 8.

The direct effects of message counter and consonant argument also proved meaningful indicators of persuasion. The difference measure (consonant-counter-argument) was the more important of the two, its .55 partial beta an indicator of its strength in predicting the persuasion. The interaction (consonant x counter-argument) was less important, with a -.24 partial beta, but remains significant. Interpretation of these results in terms of the hypotheses is difficult, and until we can look at these composite measures in terms of their components, it should be remembered that these are composite scores. It can be noted, however, that these measures are effective predictors of persuasion.

Although no hypothesis had been forwarded concerning the effect message duration would have on persuasion, the author was surprised to find duration had a direct effect of -.32. It seemed reasonable to expect that increased message duration would increase both commitment to the message and message consonant argument, thus increasing persuasion. It can be concluded from this result, that if duration increased either commitment or consonant argument, the effect was more than balanced by some other effect. There appear to be two explanations for the negative correlation between duration and persuasion. The first, that message counter-argument increases with duration, may have some influence, but cannot be accepted due to the low correlations between duration and the measures of argument. The second, that message duration is positively related to initial attitude, which in turn is negatively related to persuasion, makes more sense in light of the results.

It has already been noted that initial attitude effects message duration directly. The raw correlation is .42. Its direct effect in the presence of reward is .37. Initial attitude is negatively related to persuasion because only those with anti-message initial attitudes can be expected to experience and attempt to reduce dissonance. As such they are more persuasible than those initially pro-message. Message duration reflects this effect isolating those who initially agree with the message and are likely both to speak longer and to be less persuaded from the rest of the sample. An important aspect of this direct effect of message duration on persuasion is the indirect effect of initial attitude on persuasion.

Table 9 shows an indirect effect of -.12 for initial attitude on persuasion via message duration. Although the effect appears small and is counter-balanced by the indirect effect of initial attitude via the difference measure of message counter and consonant argument, it should not be regarded lightly. Although the direct effect of duration is significant, its total effect takes into account two positive indirect effects of .09 and .06. Although neither effect is large, they reduce the total effect of message duration to a non significant -.19. The most meaningful indirect effects in this portion of the model are those of reward and reward x initial attitude, -.14 and .15, respectively. These effects act first in enhancing the total effect of the interaction and second, in demonstrating that reward has some effect all its own.. Interpretation of these indirect effects is not possible without examining the components of the composite measure of counter and consonant argument and will, as such, be reserved until later.

It car be seen in these results, that self-persuasion in the encoding situation is a highly complex process. No less than five significant main effects have traced their way into the process. It must be noted, however, that these five threads and their intertwinings do a remarkably credible job of explaining the process of persuasion. The multiple correlation achieved in the regression of these variables on persuasion is .84, an accounting of 70% of the variance in the dependent variable. This correlation is extremely significant (F~12.78, df=6/32, p=5.3x10-17 ). Thus while it must be said that the process is a complex one, it should not be inferred that it is too complex to be understood.

The last stage of the model, Final Attitude, is something of an anti-climax after the heavily trafficked intersection that precedes it. There are only two direct effects, initial attitude and persuasion. The multiple correlation is 1.00, a perfectly understandable result when one recalls that persuasion is a linear combination (the change score) of initial and final attitude, The important results here are the indirect effects, all of which occur via persuasion. None but one bear remarking on further than has already been done except to say that a direct effect on persuasion is an indirect effect on attitude. This, of course, is not a result, but a matter of definition. Initial Attitude has both direct and indirect effects on Final Attitude. Interestingly, these effects balance each other, leaving initial attitude with almost no total effect on final attitude.

Having arrived at final attitude, the model is complete. The balance of the model is excellent, as can be seen in the joint or spurious effects of Table 9. No joint or spurious effect exceeds .02, which speaks well for the model. 'rhere are questions which remain to be answered, however, particularly as concerns effects on the individual components of the difference measure of counter and consonant argument. Tables 10 and 11 depict the multiple regression of initial attitude, reward and their interaction on both message counter and consonant argument. The results are expresses in the form of an analysis of variance table with the change variance (r^2 change) of each variable substituted for the sum of squares.

The effects, as can be seen in Tables 10 and 11, are fairly straightforward. Message consonant argument is significantly predicted by only one variable, initial attitude, while message counter-argument was significantly predicted by both reward and the interaction but not by The correlation of counter and consonant argument (r=-.51) tends to blur these results somewhat, when indirect effects are considered, but the direct effects of reward, initial attitude and their interaction on message counter and consonant argument is unmistakable. These results indicate that the level of message consonant argument is determined largely by the subjects' initial attitude. The relationship, with a correlation of .52, is positive, indicating that, as might be expected, subjects with more favorable initial attitudes used more message consonant arguments.

The influence of reward on message counter-argument was equally straightforward, with its simple association (r=,32) positive, indicating that higher rewards were accompanied by higher levels of counter-arguments. The interaction of reward and initial attitude is more difficult to interpret, however. Its negative correlation with message counter-argument (r=-.38) indicates that the highest levels of counter-argument occurred in the favorable initial attitude, low reward and the unfavorable initial attitude, high reward conditions. Together, the effects of reward and reward x initial attitude leave the unfavorable initial attitude, high reward condition with the highest level, the two favorable initial attitude conditions with smaller and approximately equal levels, and the most dissonant condition, the unfavorable initial attitude, low reward condition, with the smallest level of message counter-argument,

In chapter 5 it was hypothesized that cognitive dissonance could evidence itself in either or both of two ways. It was first hypothesized that the act of overcoming cognitive dissonance would result in increased message consonant argument. This hypothesis follows from the existing literature on cognitive dissonance. The second hypothesis, built on the theoretical construct presented in chapters 2, 3, and 4, predicted that the effort involved in overcoming the dissonance would result in decreased message counter-argument. The high positive correlation between initial attitude and message consonant argument, although a perfectly reason able result, does not support the first of these two hypotheses. This result is not nearly as important to the hypothesis as is the non-significance of reward and reward x initial attitude, at least one of which should have differentiated between levels of message consonant argument. These results fail to support the traditional dissonance viewpoint, that persuasion under dissonant conditions is the direct result of justifying actions with attitudes, of finding good reasons for one's actions. The effects of reward and reward x initial attitude on message counter-argument provide excellent support for the second hypothesis, however. The least counter-argument occurs in the $.50 dissonant condition, the most dissonant condition. Of the three variables, only initial attitude did not significantly predict counter-argument. Finally, both reward and the interaction have the correct relationship to message counter-argument.

Implicit to these results, that persuasion due to dissonance was not due to justifying actions with attitudes, but rather the result of distraction from considering existing attitudes, is an explanation for the dissuasion observed in the favorable initial attitude, low reward condition. Table 12 displays the means of the experimental groups on several of the experimental variables. Several patterns emerge from close examination of these means. The first, which can be observed in the message counter and message consonant argument columns, is the isolation of those subjects in the negative initial attitude, high reward group. All of the significant differences, as determined by t-ratios, in these columns, were with this group. The negative initial attitude encoders had the highest means for counter-argument and the lowest means for consonant argument. The second pattern, seen in the counter X consonant argument column, is the isolation of the low reward, negative initial, attitude group. The groups mean on the interaction is significantly lower than those of the other groups.

The third pattern involves the clear separation of nearly all the variables in the duration column. There is only one non-significant difference here, that between the highly persuaded low reward, negative initial attitude group and the high reward, positive initial attitude group. The high reward, negative initial attitude condition lies significantly below these groups. The low reward positive initial attitude group lies significantly above them. This pattern is important in and of itself, because it is firmly contrary to the expectations of the incentive perspective. Higher rewards, in incentive theory, yield greater effort. Here, however, the opposite holds true, with low reward yielding increased message duration in both positive and negative initial attitude groups. With respect to the dissuasion in the low reward, positive initial attitude group, this same pattern gives an important clue.

The high reward, negative initial attitude group has the shortest mean message duration, an indication, that the group didn't put as much into the message as the other groups. Incentive and dissonance theorists might argue for a lifetime over the meaningfulness of this result, but it makes sense in terms of other measures used in the experiment. Simply, the group has the highest level of counter-argument and the lowest level of con sonant argument. It might be said that the group did an excellent job of 'biased scanning,' favoring counter-arguments over consonant. Having found few consonant arguments, they recorded a short message.

The low reward, positive initial attitude condition has the longest mean message duration. Here the difference does not appear to be due to doing a good job of biased scanning, but rather the opposite. Although the group has the highest mean for consonant argument, it also has the second highest mean for counter-argument. This movement is reflected in the counter X consonant argument column of Table 12. The group's score, the highest in the column, is indicative of a generally low level of bias in considering arguments.

By contrast, the low reward, negative initial attitude group did an excellent job of biased scanning. The group has the lowest level of message counter-argument and the second highest level of message consonant argument. Ultimately, however, the group's strongest evidence of good biased scanning is in the interaction of counter and consonant argument. Its score here, easily the lowest, argues that more than one subject had zero in the counter-argument column.

The explanation, as outlined above, lies in the incentive concept of biased scanning, a concept which was shown in chapter three to be strongly related to that of cognitive distraction. For the purpose of this experiment, the high reward conditions of both the positive and negative initial attitude groups serve as reference points to demonstrate how biased scanning operates. The high reward, negative initial attitude group functions pretty much the way a group might be expected to where dissonance could be reduced through channels other than attitude change. They perceive arguments supporting their position. They fail to perceive non-supportive (message consonant) arguments. In the low reward, negative attitude group, the reward left no alternative dissonance reduction channel open. They therefore channeled their available energy into finding good reasons for their actions, diverting energy from the act of counter-arguing. The imbalance created by finding the good reasons (consonant arguments) while not message counter-arguing yielded persuasion and a reduction of dissonance. This is clearly biased scanning, but it is composed of as much of a reduction in message counter-argument as it is of an increase in message consonant argument.

Using the high reward, positive initial attitude group as a base point, biased scanning in the absence of cognitive dissonance can be observed. It should be expected that subjects who initially agree with a message will be biased toward message consonant arguments. This is in fact the case in both the high and low reward conditions of the positive initial attitude group. The low reward condition, however, has higher means for both counter and consonant argument. It can be argued that the differences between the means for counter and con sonant argument are about the same for both groups and that the means do not significantly differ from each other, but the increase in counter-argument production is a signal of poor biased scanning in the low reward, positive initial attitude group.

This signal is strengthened by the interaction of counter and consonant argument. Here, the strength of the difference between the means of the low reward, positive initial attitude group and the other groups appears to vary directly with the quality of the biased scanning in the comparison group. Thus the difference between low reward, positive initial attitude and low reward, negative initial attitude, the group whose scanning bias was clearly the most effective, is significant (t=3.04, df=18, p=.007). With the next most effective group, high reward, positive initial attitude, the difference is a trend t=l. 5, df=20, p=.149). With high reward, negative initial attitude, the difference is not significant (t=l.19, df=20, p=.248).

Together these results make an interesting picture. Where there is dissonance, low reward increases biased scanning and effort. Where there is no dissonance, low reward decreases biased scanning and increases effort. The latter result may seem somewhat incongruous. How, one might ask, can effort increase as bias decreases. A look at the correlations between the various sets of means clarifies this. The interaction of counter and consonant argument, an interaction that might well go by the name biased scanning, for that is what it appears to represent, correlates -.9969 with the means for persuasion. Despite the small df, this relationship is highly significant (F=477.93, df=l/3, p=.0002), indicating that the two measures are much more highly related than was indicated in the model. Duration and consonant argument are also highly correlated (.9544). This relationship is also significant (F=30.65, df=l/3, p=.01). Finally, counter-argument is correlated -.85 with final attitude, a near significant relationship (F=7.98, df=l/3, p=.06).

Together, these correlations paint an interesting picture of a three tiered model of cause and effect in persuasion. Manipulation of reward and attitude affect counter-argument, consonant argument, and through them biased scanning. These affect, respectively, final attitude, duration, and persuasion. Although this model is simplistic, it underlines the separation between effort and biased scanning. One need not expend great effort to perceive selectively. Similarly, poor biased scanning does not preclude high degrees of effort.

The picture, however, based as it is on t-ratios and correlated means, should be viewed with great caution. The statistics with which the picture is painted are weak ones, and their use is only justified by the post hoc nature of this portion of the results. The important thing is the logic and consistency of the explanation they support. Biased scanning makes sense, not only as the mechanism which in imbalance can yield persuasion. but as a source of dissuasion when exercised poorly.

The picture is, moreover, consistent with the results surrounding the model, notably the positive relationships that initial attitude has with consonant argument and duration, and that reward x initial attitude has with counter-argument.

Foulger, Davis A. (2005). Results. 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/chapter7.htm.