"—3 fiv —_,______ , ONE-SIDED VS. TVVO-SSDED MESSAGES: AN EXPERIMENT IN COUNTERGONDJTIONING Thesis for fhe Degree of Ph. D. MLCWGAN STATE UNWERSH‘Y Deimer M. Hi§yard 1966 J *4 (N. fl/II/lfl/I/ll/Il/l/ /////I/// I/////I/I////I///I/I//////I///I * ' 413- ' 7- 11 State. 3 1293 10 iVnLLdbd . i THESM 364 7792 Umvermt)’ if" L— {.ti "W' " r: 43"". ‘ ‘ This is to certify that the thesis entitled One-Sided vs. Two-Sided Messages: An Experiment in Counterconditioning presented by Delmar M. Hilyard ' has been accepted towards fulfillment of the requirements for Ph.D. degree in Communication 1/ Major professor Date February 22, 1966 0-169 /)7-2-4e F19? m7 MR 0 3 ‘l 4.‘ 7 ABSTRACT ONE-SIDED VS. TWO-SIDED MESSAGES: AN EXPERIMENT IN COUNTERCONDITIONING Delmer M. Hilyard This study develOped from an assumption that the effects of one- sided and two-sided messages on receivers' attitudes are analogous to the effects of total and partial reinforcement in learning experimentation. The conditioning/counterconditioning hypothesis of Mowrer's two-factor learning theory was used to develop the rationale. A major assumption of the study was that each sentence within a persuasive message was the equivalent of a learning trial in the classical conditioning paradigm and would either condition (positively reinforce) or countercondition (negatively reinforce) a subject's pre-experimental response toward the message topic. Favoring (Pro), neutral and Opposing (Con) statements, regarding the desirability of revising existing abortion laws, were prepared and arranged into eight different combinations analogous to eight different reinforcement schedules. Neutral statements were assumed to be message equivalents of non-reinforcement conditions in learning research. The following message conditions were used to test the analogy: l. 100% Pro (34 Statements) 5. 100% Neutral (17 Statements) 2. 100% Pro (17 Statements) 6. 50% Con/50% Neutral (3n State.) 3. 50% Pro/50% Neutral (3n 7. 100% Con (17 Statements) Statements) 8. 100% Con (3n Statements) u. 50% Pro/50% Con (34 State- ments) Delmer M. Hilyard On the basis of pretest attitude scores subjects were assigned to Neutral, Con, or Pro subjects experimental treatments. 128 subjects in each experimental treatment were then randomly assigned and equally apportioned among the eight message conditions. One-sided vs. two-sided message effects were tested for both "learning," measured by the amount of pretest to immediate posttest change in attitude, and "extinction," measured by the amount of immediate posttest to delayed posttest regression. Bach attitude measurement was based on summated scores of responses to six semantic differential scales. Neutral subjects in the one-sided, 100% Pro (17 Statements) message condition showed a significantly greater amount of attitude change than did Neutral subjects in the two-sided, 50% Pro/50% Con message condition. Both experimental groups differed significantly from a Control group receiving the 100% Neutral statements message. Neutral subjects in the one-sided message condition also showed a significant amount of "regression." Neutral subjects in the two-sided message condition did not regress significantly. These findings supported the two major hypotheses that a total reinforcement message would result in a greater amount of attitude change and a greater amount of regression than a partial reinforcement message. Relevant data from Con and from Pro subjects experimental 'treatments were generally consistent with the hypotheses but failed to reach the required .05 level for statistical support. Another hypothesis, that the amount of attitude change observed increases with the amount of change advocated, was generally supported by the data for Con and for Pro subjects. There were no significant regression effects for any of the an statements message conditions.‘ 4 Delmer M. Hilyard Attitude change and regression hypotheses comparing partial reinforcement with total reinforcement, when partial reinforcement was defined by either the 50% Pro/50% Neutral or the 50% Con/50% Neutral message conditions were generally supported by the trends of the data. However, none of the comparisons reached statistically significant levels. Subjects' responses to the messages' informational content were measured by a multiple choice recall test. All message conditions for the three experimental treatments showed significant learning and ex- tinction effects. There was no correlation between attitude change scores and information recall test scores. The findings of the study, while not providing conclusive support in all instances, generally imply that the analogy between learning and attitude research is useful. Additional tests of attitude change and resistance to extinction using the adapted conditioning paradigms are suggested. ONE-SIDED VS. TWO-SIDED MESSAGES: AN EXPERIMENT IN COUNTERCONDITIONING BY Delmer M. Hilyard A THESIS Submitted to The Department of Communication Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY 1966 ACKNOWLEDGEMENTS The personal satisfaction I gain from the completion and pre- sentation of this study has been realized only because of the contributions of a great number of other people. In particular, my graduate committee chairman, Dr. Gerald R. Miller, provided encouragement, counsel, and criticism in a manner consistently equal to the best teaching traditions. Special thanks are due Dr. William T. Stellwagen, who promoted and guided my research efforts, and are also due the other members of my graduate committee -- Dr. Fred Alexander, Dr. Murray Hewgill, and Dr. Clessen Martin -- whose interest and support have been most helpful. Dr. David K. Berlo and Dr. Erwin P. Bettinghaus have frequently made crises less stressful by their considerate administrative support of my academic program. The efforts of Mrs. Shirley Sherman, who typed the manuscript as one of her many contributions, and Mr. A. Talbott, who provided essential services for data analysis, are also very much appreciated. The cheerful willingness of my special three -- Susan, Julia, and Scott -- to forego some rights and privileges of childhood in loyal support of their father's goals has been particularly encouraging. And for an even longer period of time, my wife, Shirley, has maintained the home environment that made my professional goals seem not only worthwhile but also possible. TABLE CHAPTER I RATIONALE AND HYPOTHESES Introduction . . . . . Rationale . . . . . . Hyp0theses o o o o o 0 II METHOD . . . . . . . . . SUbjeCtSooooooo Experimental Treatments Message TOpic . . . . Independent Variables Dependent Variables . Procedures . . . . . . III RESULTS . . . . . . . . Hypotheses of the Study IV CONCLUSIONS, DISCUSSION, FOR FURTHER RESEARCH . . CODClUSion o o o e o o DiscuSSion o o o o o 0 OF CONTENTS 0 O O O O O I C AND IMPLICATIONS Implications for Further Research . . . REFERENCES 0 O O O O O O O O O O APPENDICES O O O O O O O O O O O Page 20 2O 20 22 23 27 3O 35 38 67 67 71 78 80 83 TABLE 10 12 LIST OF TABLES Frequency Distributions of Responses to Pour Attitude T0pics on the (Bad-Good) Scale of the Semantic Differential Frequency Distributions of Responses to Four Attitude Topics on the (Unimportant-Important) Scale of the Semantic Differential Intercorrelation Matrix of Six Evaluative Scales and Summated Attitude Score Pretest Minus Posttestl Attitude Score Mean Differences of Neutral Subjects Experimental Treatments Relevant to Hypothesis 1. Pretest Minus Posttestl Attitude Score Mean Differences of Con Subjects Experimental Treatments Relevant to Hypothesis 1. Pretest Minus Posttestl Attitude Score Mean Differences of Pro Subjects Experimental Treatments Relevant to Hypothesis 1. Posttestl Minus Posttest2 Attitude Score Mean Differences of Neutral Subjects Experimental Treatments Relevant to Hypothesis 2. Posttestl Minus Posttest Attitude Score Mean Differences of Con Subjects Experimental Treatments Relevant to Hypothesis 2. Pretest Minus Posttest Attitude Score Mean Differences of Neutral Subjects Experimental Treatments Relevant to Hypothesis 3. Pretest Minus Posttest Attitude Score Mean Differences of Con Subjects Experimental Treatments Relevant to Hypothesis 3. Pretest Minus Posttest Attitude Score Mean Differences of Pro Subjects Experimental Treatments Relevant to Hypothesis 3. Posttestl Minus Posttest Attitude Score Mean Differences of Neutral Subjects Experimental Treatments Relevant to Hypothesis u. iv Page 22 29 38 #0 41 #2 #3 an #5 H7 48 TABLE Page 13 Posttestl Minus Posttest2 Attitude Score Mean Differences of Con Subjects Experimental Treatments Relevant to Hypothesis 4. 49 14 Pretest Minus Posttest Attitude Score Mean Differences of Neutral Subjects Experimental Treatments Relevant to Hypothesis 5. 50 15 Pretest Minus Posttestl Attitude Score Mean Differences of Con Subjects Experimental Treatments Relevant to Hypothesis 5. 51 16 Pretest Minus Posttestl Attitude Score Mean Differences of Pro Subjects Experimental Treatments Relevant to Hypothesis 5. 52 17 Posttestl Minus Posttestz Attitude Score Mean Differences of Neutral Subjects Experimental Treatments Relevant to Hypothesis 6. 53 18 Posttestl Minus Posttest Attitude Score Mean Differences of Con Subjects Experimental Treatments Relevant to Hypothesis 6. 54 19 Pretest Minus Posttestl Attitude Score Mean Differences for Five 34 Statement Message Conditions for Con Subjects Experimental Treatment Relevant to Hypothesis 7. 56 20 Critical Differences Among Posttestl Attitude Score Means for Five 34 Statement Message Conditions for Con Subjects Experimental Treatments Relevant to Hypothesis 7 56 21 Pretest Minus Posttestl Attitude Score Mean Differences for Five 34 Statement Message Conditions for Pro Subjects Experimental Treatments Relevant to Hypothesis 7 58 22 Posttestl Minus Posttestz Attitude Score Mean Differences for Five 34 Statement Message Conditions for Con Subjects Experimental Treatment Relevant to Hypothesis 8 59 23 Summary of Statistical Results for Eight Hypotheses Tested by Data for Three Experimental Treatments 60 24 Summary of Two Factor Analysis of Variance of Information Recall Test Scores for Neutral Subjects Experimental Treatment 61 TABLE , Page 25 Summary of Simple Effects Analysis of Information Recall Test Scores for Eight Message Conditions at Times of Pretest, Posttestl and Posttest2 for Neutral Subjects Experimental Treatment 62 26 Summary of Two Factor Analysis of Variance of Information Recall Test Scores for Con Subjects Experimental Treatment 63 27 Summary of Two Factor Analysis of Variance of Information Recall Test Scores for Pro Subjects Experimental Treatment 63 28 Summary of Simple Effects Analysis of Information Recall Test Scores for Eight Message Conditions at Times of Pretest, Posttest , and Posttestz for Pro SUbJeCtS Experimental Treatment 64 29 Critical Difference Test for Posttest2 Message Condition Means of Information Recall Test Scores for Pro Subjects Experimental Treatment 65 30 Pretest Minus Posttestl and Posttest1 Minus Posttest2 Difference Score Means and Standard Deviations of Attitude Test Responses of Eight Message Conditions for Neutral Subjects Experimental Treatment 114 31 Pretest Minus Posttestl and Posttestl Minus Posttestz Difference Score Means and Standard Deviation of Attitude Test Responses for Eight Message Conditions for Con Subjects Experimental Treatment 115 32 Pretest Minus Posttestl and Posttestz Minus Posttestz Difference Score Means and Standard Deviations of Attitude Test Responses for Eight Message Conditions for Pro Subjects Experimental Treatment 116 33 Pretest, Posttest , and Posttest2 Means and Standard Deviations of Attitude Test Scores for Eight Message Conditions for Neutral Subjects Experimental Treatment 117 34 Pretest, Posttest , and Posttest Means and Standard Deviations of Attitude Test Scores for Eight Message Conditions fer Con Subjects Experimental Treatment 118 TABLE 35 36 37 38 Pretest, Posttestl, and Posttest2 Means and Standard Deviations of Attitude Test Scores for Eight Message Conditions for Pro Subjects Experimental Treatment Pretest, Posttestl, and Posttestz Means and Standard Deviations of Information Recall Test Scores for Eight Message Conditions for Neutral Subjects Experimental Treatment Pretest, Posttestl, and Posttest Means and Standard Deviations of Information Recall Test Scores for Eight Message Conditions for Con Subjects Experimental Treatment Pretest, Posttest , and Posttest2 Means and Standard Deviations of In ormation Recall Test Scores for Eight Message Conditions for Pro Subjects Experimental Treatment Page 119 120 121 122 FIGURE LIST OF FIGURES Diagrams of the conditioning of word meaning Diagram of conditioning paradigm for a one-sided message Diagram of conditioning paradigm for a two-sided message Page LIST OF APPENDICES Page APPENDIX A O O O 0 O 0 O O O 0 O O O O 0 O O O 83 APPENDIX B O 0 O O O O O O O O O O O O O O O O 96 APPENDIX C O O O O O O O O O O O O O O O O O O 113 ix CHAPTER I RATIONALE AND HYPOTHESES Introduction Communication research has frequently relied on theories of learning for the development of hypotheses and possible explanations of communication events. Similarly, this study developed from an assumption that the effects of one-sided and two-sided messages on receivers' attitudes are analogous to the effects of total and partial reinforcement on the learning of overt responses. Thus, the rationale presented in this chapter equates hypotheses concerning message variables and attitude change with the general findings from studies of reinforcement effects on learning. Rationale Attitudes and Learniggf-The attitude concept has consistently served an integrative function in social science research. Hollander and Hunt note that The attractiveness of the attitude concept is readily understood. It serves as a simple, manageable representation of something quite complex: A brief summary of what has gone before in thg_individual's experience that may affect his present behavior. (Italics mine) (12) Katz has stated that modifying an old attitude or replacing it with a new one involves a process of learning. He conceptualizes attitude as the predisposition of the individual to evaluate some symbol or object or aspect of his world in a favorable or unfavorable manner. (18) Osgood, Suci. and Tannenbaum similarly state that Most authorities are_agreed that attitudes are learned and implicit. Further, they are predispositions to respond, but are distinguished from other such states of readiness in that they predispose toward an evaluative response. Thus, attitudes are referred to as 'tendencies of approach or avoidance,‘ or as 'favorable or unfavorable,‘ and so on. (30) Thus, given a general agreement that attitudes are learned and implicit, then the explicit behavior from which an attitude is inferred should presumably conform to generalizations about behavior derived from studies of learning. Learning, Reinforcement,_and the One-Sided/Two-Sided Message Issue -- Two , generalizations which have been consistently supported by animal studies (e.g., 9, l7) and additionally supported by studies using children as subjects (n) have been simply stated by Berelson and Steiner: The course and strength of conditioning is systematically related to the quantitative relationship between reinforcer and response (the 'reinforcement schedule'). (1) A 100% schedule is usually the quickest way to establish new behavior or to increase the frequency of a response; that is, learning proceeds most rapidly when every correct response is reinforced. But by the same token: (2) Responses learned on such a schedule also extinguish most rapidly. (2) Although messages apparently were not prepared according to any type of 'reinforcement schedule' criteria, previous one-sided/two- sided message research appears to be analogously related. A one-sided message might be considered a 100% reinforcement schedule; a two-sided message might be viewed as partial reinforcement. In the first study to employ the one-sided/two-sided message variable, Hovland, Lumsdaine, and Sheffield presented experimental messages to soldiers regarding the proposition that the war with Japan would be a long one. The two-sided message contained non-refuted con arguments as well as pro arguments as well as pro arguments supporting the proposition. The one-sided message contained the same pro arguments, but the non-refuted con arguments were deleted. Among men initially favorable to the proposition that it would be a long war, the one-sided message resulted in a greater percentage of individuals changing in the direction advocated. (15) Two later studies haVe also indicated that one-sided messages for initially favorable subjects result in greater change in the direction advocated when such change is based on pretest-immediate posttest differences. (23, 24) Other studies incorporating the one-sided/two- sided message variable have produced inclusive results. (22, 31, 39, #0) Although the analogy between attitude change and learning re- search is supported by the above studies, the parallelism is weakened by an apparent discrepancy between counterargument in persuasion and non-reinforcement in learning. This discrepancy is possibly most evident in extinction studies in the two research areas. In one-sided/two-sided message research, the extinguishing of an attitude or opinion has been based on a counterargument presented sometime after the message of advocation has been presented and a posttest measurement of subjects' responses to the advocacy have been made. In learning research, extinction studies have typically compared one group which first learned on a total reinforcement schedule with other groups which first learned on partial reinforcement schedules (i.e., some critérial responses were not reinforced). All groups then received no reinforcements for any response during the extinction phase of the study. The counterargument in persuasion research does not seem to be an analogical equivalent of the non-reinforcement conditions of learning research. Instead, it seems to be the substitution of an opposing rein- forcement, i.e., a punishment. However, if this disparity is disregarded, message research findings are consistent with reinforcement generalizations. Several studies have reported two-sided messages (i.e., partial reinforcement) as having been more effective against later counterargument than were one- sided messages. (22; 23; 24) Two factors have been incorporated into the rationale in order to establish analogical consistency. First, it has been assumed that a part of a message that is neutral, i.e., that neither favors or opposes the proposition, is analogous to a non-reinforcement condition in learning ex- perimentation. Second, the mediational theory of learning, as presented in the writings of Osgood and Mowrer, has been used to account for the relationships between positive reinforcement, non-reinforcement, and negative reinforcement. Both Osgood and Mowrer offer relatively well explicated relationships between learning, reinforcement, and attitude change. As noted in the next section, some evidence of such relationships has already been es- tablished at a relatively simple level. Potential integration of attitude research and learning research requires the application and testing of mediational theory to the more complex relationships of message variables and attitude change. Attitude Development and Mediational Theory -- A central assumption of Mowrer's and Osgood's theories is that covert (i.e., mediational) responses are conditioned to stimuli, including verbal stimuli. Osgood states, A very large proportion of the verbal signs used in communication are what we have termed assigns -- their meaning is literally 'assigned' to them via association, not with the objects represented but with other signs. . . (P)reviously established signs (or assigns) elicit certain meaningful reactions, and since the new assign is temporally contiguous with these reactions, it also becomes conditioned to them. (28) Mowrer proposes that a sign or assign is placed in temporal con- tinguity with another sign or assign when the two are linked as subject and predicate; i.e., a sentence functions as a conditioning device. Thus, in his illustrative sentence, "Tom is a thief," the respondent's pre- viously established, mediational, evaluative reponse to the verbal stimulus this: becomes conditioned to the temporally contiguous verbal stimulus Isa (27). Staats, Staats, and Heard have summarized the verbal conditioning process proposed by Osgood and Mowrer through use of classical conditioning paradigms as illustrated in Figure 1. Two experiments (35; 37) have demonstrated that attitude (i.e., responses to semantic differential scales) can be acquired through classical conditioning and in accordance with predicted differences based on learning reinforcement schedules. Other studies of evaluative reactions to verbal stimuli also indirectly indicate that the conditioning paradigm is useful in accounting for verbal stimulus-attitudinal response relationships. (6; 7; 8) One implication of the verbal conditioning paradigm is that a proposition (a statement or sentence) is conditionable to another proposition, or, at least, to the 'detachable component(s)' common to a series of propositions. In other words, the likelihood of a response to (A) l. 2. (B) l. 2. Word: BAD CS UCS R BAD ................ rnv ’,,¢””””. R physical punishment BAD-—» a an Sentence: NBGROBS ARE BAD. CS UCS R NEGROES - - - - - - - - - - - - - - - - an memes; - I‘m, Fig. l -- The diagram depicts first-order conditioning of word meaning. After a number of pairings of BAD, the CS, with punishment, the UCS, BAD comes to elicit the conditionable (i,e., "detachable") components of the responses elicited by the punishment (symbolized as rnv because 0f the negative value). The components of the total response which are not stably conditioned are symbolized as R. Diagram B depicts higher-order conditioning of meaning. The negative meaning responses now elicited by BAD are conditioned to NEGROES through contiguous pairing of the two words in the sentence. (37) a proposition may be assumed to increase or decrease (to be positively or negatively reinforced) by placing that proposition in temporal contiguity with other propositions. Within the perspective of mediational theory, such conditioning arrangements appear to be central to the persuasion process. If the conditioned stimulus is considered to be a Summary Evaluative Proposition (e.g., "X is good."), which delineates an intended common component of the temporally contiguous, accompanying arguments, a persuasive message may be viewed as a particular combination of rein- forcers (unconditioned stimuli) for a response to that §;§;§, Figure 2 illustrates such a paradigm for a one-sided message. CS UCS R 10 x is 80“.. ------------- o u - - —( Trial 1. X makes us rich. (Pp1 8 (R) Trial 2. X promotes peace. (rp2 8 (R) Trial N. X saves lives. 5 8 (R) ‘n X is good. p Fig. 2 -- Conditioning paradigm for a one-sided message. The conditionable component of each total response to each learning trial is symbolized as rp to indicate a positive value relative to the CS. The components of each total response which are not stably conditioned are symbolized as R. CS UCS R l. X is gOOd-O .................... ( Trial 1. X makes us rich. (pp1 a (R) Trial 2. X requires effort. (rql a (R) Trial 3. X promotes peace. (rpz 8 (R) Trial a. X is a tedious process_.____n(r'q_2 8 (R) . (. . (. . (. Trial N. X saves lives. (rpn 8 (R) 2. X 's ood. r 1 g p-q Fig. 3 -- Conditioning paradigm for a two-sided message. The conditionable component of each total response to each learning trial is symbolized as r to indicate a positive value or r to symbolize a negative value relative to the C.S. he components of each total response which are not stably conditioned are symbolized as R. The diagram for a two-sided message differs in that the conditionable response components are indicated by another symbol (rq) with a resulting change in the conditioned response value to r (See Fig. 3) Such P'q paradigms of statements as conditioned and unconditioned stimuli are basic to the operationalization of persuasive messages as reinforcement schedules. Reinforcement and Extinction as Conditioning and Counterconditioning -- Although there is disagreement about the relationships among rewards, punishments, and non-reinforcements (ll), mediational theory assumes that evaluative responses to the three stimulus conditions are ordered along one dimension. Mowrer, for example, assumes a learning 'scale' extending from +1.0 (reward; strong hope) through 0.0 (neutrality; indifference) to -l.0 (punishment; strong fear). Non-reinforcement, within his theory, is functionally equivalent to mild punishment or mild reward; (i.e., its 'scale' value is at some point near 0.0, but does have negative or positive value). (26) Another of his major assumptions is that all learning involves conditioning principles and all nnlearning, i.e., extinction or forgetting, involves counterconditioning. Thus, reinforcers, as unconditioned stimuli, are either conditioners or counterconditioners. Whether a specific reinforcer is a conditioner or counterconditioner is, of course, relative to a respondent's prior schedule of reinforcements. Within this arrange- ment a non-reinforcement is a 'mild' counterconditioner. In terms of the conditioning/counterconditioning continuum, attitude studies may be viewed in this way: 1. Pretest: Measurement of extra-experimental pre- conditioning/ counterconditioning effects. 2. Message treatment and immediate posttest: Measurement of experimental conditioning effeCts. For example, a. 100% Conditioning Schedule. (One-sided message) b. Conditioning plus some counterconditioning. (Two-sided message) 3. Delayed posttest: Measurement of message effects confounded with non-message (typically extra-experimental) conditioning/ counterconditioning effects. a. Counterargument and immediate posttest: Measurement of experimental counterconditioning effects. The conditioning/counterconditioning continuum thus appears sufficient to render differences in resistance to negatively reinforcing counter- argument of one-sided vs. two-sided messages analogous to the extinction 10 process in learning research. Extinction, however, also refers to the weakening process generally, i.e., across time and outside the laboratory. Through its assumptions concerning extra-experimental non-reinforcement, general interference theory, which attempts to account for the generalized extinction process, provides a means for supporting the attitude/learning analogy. By accepting these assumptions, attitudinal resistance to extinction may be tested by a non-reinforcing, rather than a negatively reinforcing, condition. Extinction and General Interference Theory -- The most useful illustration of extra-experimental extinction (forgetting) is provided in a paper regarding verbal learning presented by Postman. He states, The assumed process of interference may be represented most conveniently in terms of the A-B, A-C paradigm, where A is a stimulus term in the experimental list, B is a response associated with A through linguistic usage, and C is the response to A prescribed in the experiment. Acquisition of the prescribed association requires the unlearning of extinction of the pre-experimental association, A-B, and its replacement by A-C. The evidence reviewed earlier makes it reasonable to assume that the extinguished habit, A-B, will gradually recover as a function of time and complete with A-C, at the time of recall. If A-B is a stable language habit, its pre-experimental strength was undoubtedly much greater than that imparted to A-C during the experiment. Thus, A-B will readily recover sufficient strength to compete effectively with A-C. Of course, if A-B is practiced after the end of the ex- periment, the process of recovery is speeded up and the probability of effective competition is increased. (33) On the basis of these statements the analOgy of attitude change with learning generalizations can be strengthened thus: A subject's pre- experimental attitude is equivalent to A—B, "stable language habit," in the interference paradigm. Given that a persuasive message is intended to produce a different association, i.e., A-C, a later counterargument would ll be analogous to A-B practice after A-C learning, and a similar return to the pro-experimental "stable language habit" (the pretest attitude) could be expected across time without counterargument, i.e., under con- ditions of non-reinforcement. The Problem of Attitude Stability -- there has been no consistency in prior research, however, to indicate whether or not attitudes do or do not regress to a pre-experimental level. A study by Hovland, Lumsdaine, and Sheffield (15) indicated that opinion change effects in some instances increased with the passage of time, the "sleeper" effect. The "sleeper" effect was accounted for by the source credibility variable. Other studies have shown maximum regression of attitudes to pre-experimental levels (5), while still other studies have found little change from posttest measure- ments to delayed posttest measurements made as long as one and one-half years later. (32) The obvious conclusion to be derived is that unknown and thus uncontrolled variables were contributing to the stability or instability of the attitudes measured. Attitude stability from pretest to measurement after presentation of a persuasive message has also been studied, principally in regard to the relationship between the amount of change advocated and the amount of change produced. Inasmuch as Mowrer prOposes that partial reinforcement does not produce as much "hope" as total reinforcement (i.e., different degrees of change are advocated), such studies are relevant to analogizing attitude research with learning theory. 12 The findings of Hovland and Pritzker (13) support the hypothesis that degree of change is directly related to the degree of difference between source and receiver positions -- under laboratory conditions. Similar findings, under conditions of high source credibility (but not with low credible sources), have also been reported. (1; 3; 10; #1) 0n the other hand, field or "naturalistic" studies of attitudes (1%; 19; 20; 34) indicate that the greater the source-receiver difference the less the degree of change. Some of the laboratory studies above also found rejection of discrepant messages under conditions of low source credibility. (l; 3) Sherif, Sherif, and Nebergall attribute the field study-laboratory differences to four variables: 1. The degree of stimulus (attitude tapic) specificity; degree of restriction on alternative modes of response or interpretation. 2. The range of stimulus values used in a study relative to the range of possible stimulus values; i.e., source-receiver discrepancies may or may not be maximal. 3. Source credibility, including reference group affiliations of source and receiver. 4. The degree of receiver (subject) familiarity and/or involvement with the issue. (an) Although all four variables are relevant to research control pro- cedures and consequent generalization of findings, the familiarity variable, in particular, indicates that field study-laboratory differences result, in part, from differences in the pre-experimental stability of attitudes. Some confusion concerning the measurement of attitudes has 13 consequently developed in attempting to relate the stability variable to attitude scales. Conditioning,_Counterconditioning, and Stability_-- In survey studies, extreme attitude scale positions denote relatively greater resistance to change (stability) than do more neutral positions. This would appear to be, on first impression, the assumption underlying the use of semantic differential scales in attitude studies, for Osgood, Suci, and Tannenbaum state that intensity (habit strength) is equated with distance from the neutral origin of a scale. (30) However, at least one study has indicated that neutral scale positions may be more resistant to change-than extreme positions. (38) Osgood and Tannenbaum consequently attempted to deal with the attitude stability dimension by incorporating an incredulity correction into the attitude congruity model. (29) In terms of Mowrer's counterconditioning hypothesis, the assumption is that extreme marking on the "hOpe-fear" scale indicates only that conditioning has been on a total or nearly total schedule. Thus, the pattern of conditioning (the ratios of positive reinforcements, non- reinforcements, and negative reinforcements to total number of trials) is indicated by scale position, but the amount of conditioning (the total number of conditioning trials) is not. The amount of conditioning (e.g., the number of statements in a message) is a relevant dimension of stability. Staats and Staats, generalizing from learning studies, have stated that resistance to ex- tinction also is affected by the extent of conditioning. A response that has been established in many respondent conditioning trials will continue 1n to be elicited for many trials when the CS is presented alone. (36) The two dimensions, amount and pattern of conditioning, are taken into account in Osgood, Suci, and Tannenbaum's theory of the congruity- learning process: Each time two signs are related in an assertion, the intensity (i.e., the pattern of prior conditioning) of the mediating reaction characteristic of each in isolation is shifted toward that characteristic of each in interaction, by a constant fraction of the difference in intensity. Since the difference in intensity decreases with each "trial," this means that the reactions characteristic of both signs must approach a point of common intensity . . . according to a negatively accelerated function. In other words, this generates a typical learning curve.1 (30) Thus, attitude change or resistance to change is assumed to be dependent upon the point at which an experiment "intersects" a subject's "learning curve," the number of experimental conditioning/counterconditioning trials, and the degree of difference between pre-experimental conditioning/ counterconditioning and the conditioning/counterconditioning schedule of the experiment. Measurement by attitude scale is assumed to summarize the pattern of conditioning/counterconditioning, but it is assumed to provide no summary of the amount of conditioning/counterconditioning. In terms of Postman's interference, A-B/A-C, paradigm, A-B specification is not achieved, although Postman considers such specification necessary to account for un- learning. (33) lCongruity-learning may be "fitted" to the CS-UCS paradigm: The point of common intensity of two interacting signs may be at any point along the continuum, or dimension. If one of the signs has been previously conditioned to a level of 'high' stability, it becomes, essentially, an unconditioned stimulus when paired with an unstable sign. Pairing a stable sign with an unstable sign will thus show maximal shift of the response to the unstable sign toward congruity with the response to the stable sign, and minimal shift (zero measurement) of the response to the stable sign toward the response to the unstable sign. 15 Prediction of Message Treatment and Extinction Effects -- As a summary of the rationale to this point, message treatment and extinction (regression) effects have been assumed to be relative to the interaction of the following five variables (among others not included within the rationale): A. Message type -- the pattern or ratio of conditioning/ counterconditioning statements. B. Message amount -- the length of the message; the number of statements or conditioning/counterconditioning trials. C. Subject's pretest attitude -- the attern of the subject's preexperimental conditioning/counterconditioning schedule as summarized by attitude scale measurement. D. Subject's pretest attitude stability -- the unspecified number of pre-experimental conditioning/counterconditioning trials. E. Time of measurement -- particularly the time lapse between posttest and delayed posttest when A-B language habits are assumed to interfere with message-induced A-C effects. Several difficulties, resulting from interaction of the five variables, restrict the ability to predict specific message and time effects. In terms of the postulated variables, relative to measurement precision, these difficulties include: 1. The more stable the pretest attitude (D), the less the effect of the message (A and B). 2. The less stable the pretest attitude (D), the less the effect of regression (A, C and E). Therefore, as the probability of getting significant message effects increases, the probability of getting significant regression effects decreases. 3. Two subjects may differ in both the pattern (C) and stability (D) dimensions. 16 With the (D) variable unspecifiable, relative effects of a given message on each of two subjects who differ in their attitude scale responses cannot be predicted. For example, a 'neutral' subject may change more than an 'extremely opposed' subject after presentation of a one-sided message favoring a topic because the 'neutral' subject is less stable, even though he is closer to message type (A) -- less change advocated -- than is the 'extremely opposed' subject. One additional factor should be noted. In studies of reinforcement effects on learning, reinforcement trials have frequently been held constant for both a 100% reinforcement group and a 50% reinforcement/50% non- reinforcement group. As a result, the 50% reinforcement group experiences twice the number of trials of the 100% reinforcement group (e.g., Staats, Staats and Heard's study, above). Extinction or regression differences should be most apparent on this arrangement of the (A) and (B) variables. Hypotheses The major purpose of the study was to test the analogy of one-sided/ two-sided message research as learning research. Therefore, four hypotheses may be specifically and directly related to these two types of messages: 1. Total reinforcement (one-sided message) will result in a greater amount of attitude change than will partial rein- forcement (two-sided message), if (a) total reinforcement is defined by a message of either 5 number of positively reinforcing (conditioning) or x number of negatively reinforcing (counterconditioning) statements. (b) partial reinforcement is defined by a message of x_ number of positively reinforcing (conditioning) statements combined with x number of negatively reinforcing (counterconditioning) statements, and 17 (c) change is measured by pretest-immediate posttest (Posttestl) differences. 2. Total reinforcement (one-sided message) will result in a greater amount of attitude regression than will partial reinforcement (two-sided message), if (a) total reinforcement is defined as in Hypothesis 1 above. (b) partial reinforcement is defined as in Hypothesis 1 above, and (c) regression is measured by immediate posttest (Posttest1)-- delayed posttest (Posttest2) differences. 3. The greater the number of learning trials, the greater the amount of attitude change, if (a) the number of learning trials is defined by the number of statements in a total reinforcement (one-sided) message, and (b) attitude change is measured by pretest-immediate posttest (Posttestl) differences. u. The greater the number of learning trials, the less the amount of attitude regression, if (a) the number of learning trials is defined by the number of statements in a total reinforcement (one-sided) message, and (b) regression is measured by immediate posttest (Posttest1)-- delayed posttest (Posttestz) differences. Development of the analogy suggested the inclusion of messages combining non-reinforcing (neutral) statements with reinforcing ones, in addition to the one-sided and the two-sided messages. Therefore, the two following hypotheses were also tested. 5. Total reinforcement will result in a greater amount of attitude change than will partial reinforcement, if (a) total reinforcement is defined as in Hypothesis 1 above (b) partial reinforcement is defined by a message of x number of non-reinforcing (neutral) statements combined with either x number of positively reinforcing (conditioning) or x number of negatively reinforcing (countercon- ditioning) statements, and 6. 18 (c) change is measured by pretest--immediate posttest (Posttestl) differences. Total reinforcement will result in a greater amount of attitude regression than will partial reinforcement, if (a) total reinforcement is defined as in Hypothesis 1 above, (b) partial reinforcement is defined as in Hypothesis 5 above, and (c) regression is measured by immediate posttest (Posttestl)-- delayed posttest (Posttest2) differences. In accordance with conditioning/counterconditioning theory as developed in this rationale, the above six hypotheses are assumed to be subsets of the following, more general hypotheses: 7. 8. The greater the difference between a subject's pretest attitude and the amount of change advocated by a message, the greater the amount of attitude change, if (a) the amount of change advocated by a message is defined by the ratios of positively reinforcing, non-reinforcing, and negatively reinforcing statements to the total number of statements, (b) the total number of statements, is held constant for all messages, and, (c) change is measured by pretest--immediate posttest (POSttEStl) differences. The greater the difference between a subject's pretest attitude and the amount of change advocated by a message, the greater the amount of attitude regression, if (a) the amount of change advocated by a message is defined as in Hypothesis 7 above, (b) the total number of statements is held constant for all messages, and (c) regression is measured by immediate posttest (Posttestl)-- delayed posttest (Posttestz) differences. 19 9. The greater the difference between a subject's pretest attitude and the amount of change advocated by a message, the greater the amount of attitude change, if 10. (a) (b) (c) The and the (a) (b) (c) the amount of change advocated by a message is defined by both the total number of statements and the ratios of positively reinforcing, non-reinforcing, and negatively reinforcing statements to the total statements, the amount of change advocated by a message is assumed to increase with an increase in the total number of statements while the ratios are held constant, and change is measured by pretest-oimmediate posttest (Posttestl) differences. greater the difference between a subject's pretest attitude the amount of change advocated by a message, the greater amount of attitude regression, if the amount of change advocated by a message is defined as in Hypothesis 9 above, the amount of extinction is assumed to decrease with an increase in the total number of statements, and regression is measured by immediate posttest (Posttest1)-- delayed posttest (Posttest2 ) differences. CHAPTER II METHOD Subjects Subjects were undergraduate students enrolled in an introductory psychology course at Michigan State University during one of the three academic terms in which the experiment was conducted. Qualification as a subjeCt required that the student volunteer to participate and that he be in attendance during the three unannounced times that the pretest, the post- test, and the delayed posttest were administered. From 25 to #0 percent of the students who took the pretest in one of the terms were absent during one or both of the later experimental sessions; hence, they failed to qualify as subjects.2 Experimental Treatments Students participating in the experiment during the Spring and Fall terms of 196% were placed in Extreme Pro, Mild Pro, Mild Con or Extreme Con conditions on the basis of their pretest attitude scores. Students whose scores ranged from #3 to 5n on the attitude test, indicating they were the most favorably disposed toward the proposition that legalized abortion was 2The study was conducted during the Spring, 196%, Fall, 196%, and Winter, 1965, terms. The number of students pretested in each of the three terms, respectively, was 202, 169, and wen for a total of 86A. 0f the 86a pretested students, 153, 12a and 297 (total: 57k) qualified as subjects. Equalization of groups within message treatments for analytic purposes further reduced the sample size to 38” subjects. 20 21 good, were classified as Extreme Pro subjects. Students whose scores ranged from 31 to 42 were classified as Mild Pro; those whose scores ranged from 18 to 30 were classified as Mild Con; and those whose scores ranged from 6 to 17, indicating they were least favorably disposed toward the proposition, were placed in the Extreme Con condition. 0n the assumption that different pretest attitude scores indicated different pre-experimental patterns of conditioning and unknown, but possibly differing, pre-experimental amounts of conditioning, it was decided that the different experimental treatment conditions should be considered as separate and independent studies of four different subject populations. After being placed in the experimental treat- ment conditions the subjects were randomly assigned to message treatment conditions. Prior to random assignment of subjects to message treatments during the administration of the experiment in Winter term, 1965, it was decided that the small number of Mild Con subjects in the study justified re- classification of subjects. Consequently, all subjects from all three terms were reclassified as Pro, Neutral, or Con. The Pro condition included those subjects whose scores on the attitude pretest ranged from #3 to 5%; the Neutral condition included those subjects whose attitude pretest scores ranged from 18 to #23; the Con condition included those subjects whose attitude pretest scores ranged from 6 to 17. Winter term students were then randomly assigned to message treatments. 3The neutral condition included a greater number of pro subjects (87) than con subjects (41). 22 Messege Topic On the-basis of student response to four potential topics, the topic of legalized abortion (revision of existing abortion laws) was selected for use in the study. The four initial topics were rated through use of semantic differential scales by 108 students enrolled in an introductory speech course at Michigan State University during Winter term, 196%. The students were asked to judge the topics as possible issues for speech course presentations. Response distributions for the four topics on the two scales used as criteria for topic selection are found in Table l and Table 2 0 Table 1. Frequency distributions of responses to four attitude topics on the (Bad-Good) scale of the Semantic Differential Bad 1 2 3 % 5 6 7 Good Legalized Abortion ll 9 16 23 21 15 13 Trial by Jury 2 % 2 0 11 27 62 Compulsory Voting 2 13 29 19 17 19 9 Tariff on Rice % l2 7 63 5 l2 5 Table 2. Frequency distributions of responses to four attitude topics on the (Unimportant-Important) scale of the Semantic Differential Unimportant l 2 3 % 5 6 1 7 Important Legalized Abortion 2 2 l 1% 21 32 36 Trial by Jury o o o 1+ 12 2% 68 Compulsory Voting 5 7 3 19 1% 2% 36 Tariff on Rice 32 26 22 20 % 3 1 23 The Bad-Good scale was used to measure subject attitudes toward the four topics, while the Unimportant-Important scale was used to measure the level of subject interest in the tOpic. It was sssumed that a measure of importance would provide an indication of which of the topics would be most likely to involve the subject, i,e., hold his interest and promote his attention to the message. The abortion topic was selected, and the other three topics were excluded, on the basis of relatively high importance in combination with a relatively equal distribution of pro and con reSponses. Independent Variables Statements and Messages -- A major assumption underlying this study was that each statement within a message represented the equivalent of a learning trial. Basic operationalization of the message variable was therefore centered on the preparation of statements. Statements were limited to simple and complex sentences containing an explicit (i.e., non-ellipted) subject and predicate plus accompanying modifiers. Modifiers included adjectival and adverbial words, phrases, or clauses. Compound and compound-complex sentences were excluded. Each statement was first prepared and presented to four judges. Each judge categorized each complete sentence as a statement that was favorable, unfavorable, or neutral toward revision of existing abortion laws. Ambiguous sentences were reworded and all sentences were then judged by ten graduate students. No sentence was retained which was judged by fewer than eight people as being in a particular category. Following are examples of the three categories of statements: 2% Abortion laws need revision and updating. (Favorable) Whether existing abortion laws should be revised or retained without revision is an issue promoting a great deal of debate. (Neutral) Existing abortion laws are adequate and accurate expressions of majority belief and opinion. (Unfavorable) Messages were then prepared by combining sentences into texts appropriate for the different treatments. In those messages requiring more than one type of statement, e.g., pro and con statements, the order of types was randomly assigned in a manner identical to that usually employed in establishing random reinforcement schedules in learning ex- periments. Specific statements were then selected and placed according to contextual requirements, i.e., an attempt was made to make the content maximally readable within the reinforcement scheduling limits. Where it was believed necessary, transitional expressions were inserted to approxi- mate usual message patterns more closely. For the same reason, statements were arranged in paragraph form and presented to subjects as mimeographed booklets. Appendix A contains copies of the messages presented to subjects. In order to approximate reinforcement experiments for subjects with pre-experimental pro attitudes and subjects with pre-experimental con attitudes, the following messages (reinforcement schedules) were pre- pared: 25 ' Message Treatment 3% Con Statements (The one-sided message represented by the same number of statements as the two-sided message. This treatment was assumed to maintain the relationship of this study with previous one- sided/two-sided studies in which messages were kept approximately equal in length. 17 Con Statements (The one-sided message equating number of reinforcements with two-sided messages. This treatment was assumed to maintain the relationship of this study with learning research in which number of reinforcements are equalized for total and partial reinforcement con- ditions. 17 Con Statements plus 17 Neutral Statements (The one-sided message with a partial rein- forcement condition. This treatment was assumed to approximate the partial rein- forcement condition of reinforcing half the learning trials and not reinforcing the Other half.) 17 Neutral Statements (This treatment was assumed to approximate the non-rein- forcement condition usually included in learning research.) Reinforcement Schedule 100% Negative Reinforcement for Pro subjects; 100% Negative Reinforce- ment for Neutral subjects; 100% Positive Reinforcement for Con subjects. (Conditioning statements are assumed to be positive rein- forcers for Pro and for Con subjects. Counterconditioning statements are assumed to be negative reinforcers as Pro and for Con subjects. Statements favoring the topic are assumed to be positive reinforcers for neutral subjects. Statements opposing the topic are assumed to be negative reinforcers for neutral subjects.) 100% Negative Reinforcement for Pro subjects; 100% Negative Reinforce- ment for Neutral subjects; 100% Positive Reinforcement for Con 8111):) Gets 0 50% Negative Reinforcement/50% Non-Reinforcement for Pro and for Neutral subjects; 50% Positive Reinforcement/50% Non-Reinforcement for Con subjects. 0% Reinforcement for Pro, Neutral and Con subjects. (This treatment is assumed to be a mild counter- conditioner for Pro and for Con subjects; i.e., some change is advocated.) 26 17 Pro Statements plus 17 Con Statements (The two-sided message with number of reinforcing statements equated with 17 statement messages above and below.) 17 Pro Statements plus 17 Neutral Statements (The rationale for this message is the same as the rationale for the 17 Con Statements plus 17 Neutral Statements message above.) 17 Pro Statements (The one-sided message equating number of reinforcements with two-sided messages. The rationale for this treatment is the same as the rationale for the 17 Con Statement treatment above.) 3% Pro Statements (The one-sided message represented by the same number of statements as the two-sided message. The rationale for this treatment is the same as the rationale for the 3% Con Statement treatment above.) 85% Positive Reinforcement/50% Negative Reinforcement for all subjects. 50% Positive Reinforcement/50% Non- Reinforcement for Pro subjects and for Neutral subjects; 50% Negative Reinforcement/50% Non-Reinforcement for Con subjects. 100% Positive Reinforcement for Pro subjects; 100% Positive Reinforcement for Neutral subjects; 100% Negative Reinforcement far Con subjects. 100% Positive Reinforcement for Pro subjects; 100% Positive Reinforce— ment for Neutral subjects; 100% Negative Reinforcement for Con subjects. For purposes of analysis, certain message treatment conditions were designated as control conditions for the three experimental treatment groups. The 17 Neutral Statements message, as a non-reinforcing, non-conditioning treatment for Neutral subjects was designated the control or "no-change" message treatment for Neutral experimental treatments. For Con subjects, who were assumed to have experienced mostly Con learning trials prior to the experiment, the 3% Con Statements message treatment and the 17 Con Statements message treatment were combined and were designated a control or "no-change" treatment. For Pro subjects, who were 27 assumed to have experienced mostly Pro learning trials prior to the experiment, the 3% Pro Statements message treatment and the 17 Pro Statements message treatment were combined and were designated the control or "no-change" treatment. Iige_-- Operationally, this variable was defined as the time intervals between (1) pretest, (2) immediate posttest (Posttestl), and (3) delayed posttest (Posttest2). In the Spring, 196%, administration of the ex- periment, the time interval between pretest and immediate posttest was three weeks and the time interval between immediate posttest and delayed posttest was two weeks. Time intervals between pretest and immediate posttest, and between immediate posttest and delayed posttest, in Fall term, 196%, and Winter Term, 1965, were two weeks. While the variation in pretest- immediate posttest time intervals was necessary because course activities had priority in scheduling and the experiment had to be accommodated to the class schedule, it was not assumed that such variation of time prior to exposure to the message treatment would affect the experiment. Dependent Variables Attitude (Responses to Summary Evaluative Propositions) -- Six of the evaluative scales from Osgood, Suci, and Tannenbaum's factor analyses of the semantic differential were selected for the attitude measuring instrument: Negative-Positive; Bad-Good; Reject-Accept; Oppose-Favor; Dislike-Like; and Against-For. To eliminate order effects, three of the six scales were presented with the negative polar term at the left side of the scale, while three were presented with the negative polar term at the right. The concept, Legalized Abortion, was centered above the six scales. 28 Because the original intention had been to categorize subjects as either Pro or Con (i.e., to have four experimental treatment groups), the semantic differential scales were modified in order to force subjects to favor or to oppose the topic. Modification consisted of presenting eight intervals between the polar terms, plus allowing for a ninth alternative of writing the word irrelevant across the scale if, for any reason, the subject did desire to avoid commitment. Each scale was then scored as a nine point scale with the written irrelevant receiving a neutral interval rating of five. Since each subject's attitude was obtained by summing his score on the six scales, the scores ranged from an extreme con score of 6, through a neutral score of 30, to an extreme pro score of 5%. A single Summary Evaluative Proposition was assumed to be the con- ditioned stimulus of the experiment. For example, it was assumed that a pro message would increase the probability of a favorable response to the §:§;§,, "Legalized abortion is good," while decreasing the probability of a favorable response to the S.E.P., "Legalized abortion is bad." A check of the response consistency of 202 subjects taking the pretest in Spring term, 196%, was conducted by computing product-moment correlations of the responses for each scale with the responses for every other scale and with the summated score of subjects' responses for the combined six scales. This intercorrelation matrix is presented in Table 3. The high correlation of each scale with the summated score (.88 or greater) indicated that each scale was a reliable predictor of the summated attitude score. At the same time, the intercorrelations among the six scales, ranging from 172 to .91, indicated that the use of the six scales did permit variability of 29 an individual subject's responses. (In accordance with the Osgood, et. al., rationale, the six evaluative scales were therefore assumed to provide finer degrees of intensity of response to the intended S.E.P. than a single scale.) Table 3. Intercorrelation matrix of six evaluative scales and summated attitude score Ne ative Against Bad Reject Dislike Qppose Summated POSitive For Good Accept ‘EIEE"" Favor Score Ne ative PoSitive --- Against For .80 --- Bad Good .78 .86 --- Reject Accept .79 .88 .88 --- Dislike Like .72 .73 .85 .80 --- se Favor .8% .91 .91 .91 .80 --- Summated . Score .88 .93 .95 .9% .88 .95 --- The measuring instrument, together with instructions for marking the scales at the time of the pretest, the time of the immediate posttest and the time of the delayed posttest, may be found in Appendix B. Information Recall Test -- A second dependent variable, an information recall test, was introduced into the study to serve as a learning criterion. Each message used in the study contained information which was assumed to be 30 equivalent to some of the information contained in every other message. Such equivalence was developed by taking twelve neutral statements and converting them to pro statements or to con statements by the addition of appropriate interpretive words, phrases or clauses. It was assumed that the value of this test as a learning criterion would be that differences in information recall between message treatments would tend to negate the assumption that statements (and, thus, "learning trials") were equal except for types of reinforcement. Twelve, five-foil multiple choice test items were prepared and presented to subjects as a part of pretest, immediate posttest, and delayed posttest measurement procedures. The twelve recall test items, plus the accompanying instructions for each of the three testing periods, are presented in Appendix B. Procedures Administration of Pretest -- During one of the first sessions at the beginning of each term that the experiment was conducted, the instructor announced that there would be a study conducted during class time and within the classroom for which students could volunteer as subjects, receiving credit for participation as a part of the instructor's evaluation of a student's course performance. No information regarding the purpose of the experiment or the dates on which it would be conducted was given. At the time of administration of the pretest, students were informed at the beginning of the class period that the study would be conducted that day. They were then told that a booklet would be distributed to each person who wished to participate and that specific instructions for 31 participation were presented on the first page of the booklet. They were asked not to talk with their classmates about the experiment and to work independently. Pretest booklets were then distributed. The pretest booklet, with complete instructions, is presented in Appendix B. In addition to instructions for marking the measuring in- struments, the instructions stated, Please print your name, student number, and the name of your instructor in the upper left hand corner of this page in the space provided. This information is absolutely necessary if you are to receive course credit for participation in this study. Please do it now before reading further instructions. The purpose of this study is to measure the way people judge topics of oral and written messages by having each person judge one topic on a series of descriptive scales. After the person has judged the topic on the descriptive scales, a second task is to answer some multiple choice information questions about the topic. Remember that your first task is to judge the topic on each of the descriptive scales in order. After all subjects had responded to all items of the two measuring instruments, the booklets were collected. After collection, the instructor began his lecture. Administration of Message Treatment Booklets -- At the time of adminis- tration of the message treatments the subjects were told that a second phase of the study of legalized abortion was to be conducted. Subjects were then told that each of them was to receive a booklet with his name on it. They were cautioned not to talk with each other about the experi- ment and to work independently. They were then asked to come to the front of the classroom to pick up their booklets and to return to their seats and proceed according to instructions contained within the book- let. 32 Prior to the beginning of the class session the booklets were arranged at the front of the classroom in an order based on an alphabetized roster of the students who had participated in the pretest. The book- lets were arranged in alphabetical groupings of no more than 15 booklets in a group. During the first two administrations of the study, only four message treatment conditions were prepared and presented to subjects. The four treatment conditions were the 3% Con Statements, the 17 Con Statements plp§.l7 Pro Statements, the 17 Pro Statements pipe 17 Neutral Statements, and the 3% Pro Statements messages. The four additional message treatment conditions were added to the study for the third administration in Winter term, 1965. The message treatment booklet included a coded cover sheet, a page of instructions regarding the subject's reading of the message, the message pages, a following page of instructions regarding the use of the attitude measuring instrument, a page containing the attitude measuring instrument, and a final page of concluding instructions. Message Treatment Control Factor: Source Credibility -- One of the major identified variables in attitude change is source credibility. In order to minimize possibilities of confounding this variable with the independent variables, no quotations or references to authoritative sources were included in the messages. In addition, the instructions to the subjects prior to their reading their messages included the following statements: This tOpic is one of continued public importance, so, quite understandably, certain arguments have been presented by various sources concerning the issue of whether existing laws relevant to the tOpic should be modified or retained. 33 On the next few pages you will find a summary of some of these arguments. Please read all of the statements since each statement is presented as a representative expression of the arguments concerning this topic. Administration of Immediate Posttest -- Within the message treatment booklet, on the page immediately following the last message statement, the instructions to the subject included: Now that you have read the representative statements your next task is to judge the topic of legalized abortion on a series of descriptive scales similar to the scales you marked a few weeks ago. Be certain that you know the correct procedure for marking, On the final page of the message treatment booklet were these in- structions: There is one more task for you to complete today. However, the next task requires your receiving another booklet. Please wait quietly until everyone is finished with the task you have just completed. It is important that you do not discuss what you have read with your neighbors. As soon as everyone is through with this booklet, it will be picked up and the next task booklet will be distributed. Thanks for your cooperation and patience. After subjects had completed their responses to the attitude measuring instrument, the message treatment booklets were collected and the booklet containing the information recall test was distributed. The information recall test was distributed as a second booklet to avoid a subject's referring to the printed message while answering the test items. When all subjects had responded to the information recall test items, the booklets were collected and the class resumed its course activity. Appendix B contains a copy of the message treatment booklet, minus the message sheets, and a copy of the information test booklet. 3% Administration of the Delayed Posttest -- At the time of administration of the delayed posttest, students were informed at the beginning of the class period that a third phase of the abortion study was to be conducted. Again they were cautioned not to discuss the experiment with their classmates and were told that a booklet containing instructions for participation would be distributed. In addition to the instructions about how to mark the attitude measuring instrument and the information recall test, the instructions stated, Once again it is necessary that you print your name, student number, and the name of your instructor in the upper left hand corner of this page in the spaces provided. Your participation has been most helpful to this point and the remaining tasks require only a few minutes of time. You undoubtedly recallthat the purpose of this study is to measure the way people judge topics by having each person judge one topic on a series of descriptive scales. After the person has judged the topic on the descriptive scales, a second task is to answer multiple choice information questions about the topic. Remember that your first task is to judge the tOpic on each of the descriptive scales in order. Be certain that ypu know the correct procedure for marking_the scales. When all subjects had responded to all items of the two measuring instruments, the delayed posttest booklets were collected and the class resumed its course activity. A copy of the delayed posttest booklet is contained in Appendix B. 35 CHAPTER III RESULTS The analyses reported below were conducted to test hypotheses concerning differences among message conditions in the acquisition and extinction of attitudes. Data from the three categories of experi- mental treatments--Neutra1 Subjects, Con Subjects and Pro Subjects ~- were analyzed independently and constitute three separate tests of each hypothesis. Since the hypotheses concerned either pretest-immediate posttest differences or immediate posttest-delayed posttest differences, E_tests for differences and for differences between mean difference scores were employed. In each £_test of the significance of a change across time (e.g., from pretest to immediate posttest) within a single message condition the error term employed was the standard error of the specific message condition's mean difference score with N-l degrees of freedom. In tests comparing mean difference scores between two message conditions error terms were based on the pooled variance of the two sets of raw difference scores under test with N1 + N2 - 2 degrees of freedom. A summary of the means and standard deviations of the difference scores for all message conditions for the Neutral Subjects experimental treatments is contained in Table 30 of Appendix C. A similar summary for the Con Subjects experimental treatments is contained in Table 31 of Appendix C, and the equivalent summary for the Pro Subjects experimental treatments is contained in Table 32 of Appendix C. 36 In addition, within cells (Time x Message) means and standard deviations for the three experimental treatments are contained in Table 33, Table 3%, and Table 35 of Appendix C for Neutral Subjects, Con Subjects and Pro Subjects respectively. In the following analyses, one-tailed tests of significance were employed in all cases for which the hypotheses predicted the direction of change. In those cases fOr which no change was predicted, two-tailed tests of significance were used. (Findings for which the probability is greater than .05 are reported as nonsignificant. The assumed relationship between attitude change and consequent regression led to establishing certain procedures for testing hypotheses and concluding whether or not the hypotheses were statistically supported. An attitude change (pretest to immediate posttest) hypothesis was assumed to be supported if one of the two following conditions prevailed: (1) If the message condition which was predicted to result in the greater amount of attitude change (e.g., a one-sided message) showed a statistically significant pretest to immediate posttest mean difference while the message condition which was predicted to show a lesser amount of attitude change (e.g., the two-sided message) failed to show a significant mean difference; or, (2), if both message conditions showed significant pretest to immediate posttest change, the mean difference of the message condition which was predicted to result in the greater amount of change was significantly greater than the mean difference of the other message condition (i.e., a significant difference between the pretest-immediate posttest mean differences.) 37 In regard to regression hypotheses, it was considered necessary that the message conditions should indicate significant pretest to immediate posttest mean differences before a measurement of immediate posttest to delayed posttest regression could be made. This was considered an analogical equivalent of the learning research use of a criterion reSponse level being attained by subjects before extinction research procedures are instituted. Therefore, if one or more of the message conditions failed to indicate significant pretest to immediate posttest change, testing the related immediate posttest to delayed posttest hypothesis was assumed to be statistically meaningless. In those instances for which a test of a regression hypothesis was made, differences in the amounts of regression between two message conditions were considered significant if the predicted greater regression of a message condition proved to be a significant immediate posttest-delayed posttest mean difference while the predicted lesser regression of the compared message condition proved to be a nonsignificant immediate posttest-delayed posttest mean difference. Data for two general hypotheses (Numbers 9 and 10), predicting increases in the amount of attitude change and in the amount of attitude regression with increases in the amount of change advocated for all message conditions, are not summarized in the study. Comparisons in- dicated that no new information was provided by such a summary, i.e., the hypotheses were accepted or rejected on the basis of the data summarized in Hypothesis 7 and Hypothesis 8. 38 Hypotheses of the'Study Hypothesis 1 Total reinforcement (one-sided message) will result in a greater amount of attitude change than will partial reinforcement (two-sided message), if: (a) total reinforcement is defined by a message of either fiknumber of positively reinforcing (conditioning) or §;number of negatively reinforcing (counterconditioning) statements. (b) partial reinforcement is defined by a message of 1‘. number of positively reinforcing (conditioning) statements combined with x number of negatively reinforcing (counterconditioning) statements, and (c) change is measured by pretest-immediate posttest (POSttGStl) differences. Neutral Subjects Experimental Treatments - Table % contains a summary of the Pretest-Posttestl mean differences for each of the five message conditions for Neutral Subjects that were relevant to testing Hypothesis 1. The 100% Neutral message condition served as a control condition for testing the 50% Pro/50% Con message condition against the 100% Pro (17 Statements) message condition and for testing the 50% Pro/50% Con message condition against the 10096 Con (17 Statements) message condition. Table %. Pretest Minus Posttestl Attitude Score Mean Differences of Neutral Subjects Experimental Treatment Relevant to Hypothesis 1. Pre- Postl- (N) ' mean mean D t p 100% Pro (17 State.) (16) 31.13 33.19 -12.06 5.9% .0005-1T (l-sided) 50% Pro/50% Con (16) 33.25 38.38 - 5.13 %.10 .0005-1T (2-sided) 100% Neutral (16) 30.56 30.69 - .13