“‘1 F_‘» This is to certify that the thesis entitled TELEVISION AND SOCIALIZATION ON PROSOCIAL AND ANTISOCIAL BEHAVIOR presented by M. Mark Miller has been accepted towards fulfillment of the requirements for _Eh.JL.__ degree in Wt ion MQM \ Major pro ssor DateJnmmbeM? 7 0-7639 MR 2 4 2005 ‘} ." to. . *V 1' D ' z m. ‘ ,i' $5: 9 :3 **~; a n 1'" 0 9 28‘s»- : ‘ i - u -. 033314 499.8 e [11,.» ‘3! f3. L: :x i F l TELEVISION AND SOCIALIZATION ON PROSOCIAL AND ANTISOCIAL BEHAVIOR BY M. Mark Miller A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Communication 1978 Accepted by the faculty of the Department of Communication, College of Communication Arts, Michigan State University, in partial fulfillment of the require- ee. ments for the Doctor of Ph' 0800 y do, I ‘11 j 1 2w 4 ' Directo '5'? DisserO—tion M41 / . "l , Chairman Guidance Committee: ABSTRACT TEIEVISICN AND SCEIALIZATION ON PHIJSCXZIAL AND ANTISOCIAL BEHAVIOR By M. Mark Miller This research examined effects of television exposure and iden- tification with television characters on children's performance of prosocial and antisocial behaviors. The prosocial behaviors considered were altruism, affection, and self-expression; the antisocial behaviors considered were verbal aggression and physical aggression. Past research concerning effects of television on children focused on the impact of specific televised behaviors on performance of the same behaviors. While this research considered such direct link- ages, attention was also paid to "crossed effects," i.e., effects of prosocial television on antisocial behavior, and of antisocial tele- vision on prosocial behavior. Reasoning from mediational-stiimlus contiguity theory, 14 hypotheses were derived concerning direct effects, crossed effects, and interaction effects of television exposure and of identificaticm with television characters. A questiormaire was administered to 721 fourth, sixth, and eighth graders to gather data on their exposure to 15 selected tele- vision programs, their identification with 16 selected television char- acters, and their own performance of specific social behaviors. Multiple-item indexes were constructed as indicators of the respondents' performance of the specific behaviors. Data derived from M. Nhrk Miller content analysis of the selected programs and of the programs in which the selected characters appeared mere used to weight the exposure and identification measured to form indexes. All indexes were related to sex and grade and the effects of these variables were statistically controlled in the subsequent analysis. Nbdest, but positive and significant, correlations were found between exposure to and performance of each of the specific be- haviors supporting the direct effects hypotheses. However, contrary to the crossed effects hypotheses, positive correlations were found between prosocial exposure and antisocial behavior and antisocial ex- posure and prosocial behavior. The direct effects hypotheses for identification were supported only with reference to expression and physical aggression. The crossed effects hypotheses for identification were not supported with the ap- propriate correlations being either positive or non-significant. Tests of the interaction hypotheses revealed that the anti- social exposure-behavior correlations were lowest when prosocial expo— sure was high. However, the prosocial exposure-behavior correlations were not systematically affected by levels of antisocial exposure. Prosocial and antisocial identification did not alter one another's effects. It was hypothesized that identification with characters who performed specific types of behavior would enhance the effects of ex- posure to the same behavior. 'Ihe highest prosocial exposure-behavior correlations did occur at the highest levels of prosocial identifica- tion; ravever, the relationship was markedly curvilinear. 'Ihe exposure- M. Mark Miller behavior correlations were relatively high at the lowest levels of iden- tification, and were near zero at middle levels of identification. ‘Ihe interactions of antisocial exposure and identification were less syste- matic with the highest exposure-behavior correlations occurring at moderately high levels of identification for verbal aggression, and at lowest levels of identification for physical aggression. The findings suggest multiple processes may account for the effects of both exposure and identification. Television exposure appears to lead not only to imitation, but also to heightened arousal which, in turn, increases levels of all behaviors. Identification ap— pears to operate through one process among children who want to be like television characters and through a distinctly different process among children who deny wanting to be like television characters. ACKNOWLEDGMENTS The data for this research came from the Project CASTLE research team at the Michigan State University Department of Communication. I am indebted to all members of that team and hOpe I have given back some reasonable fraction of what they contributed to me. The team, with all of its dynamism and fervor, could not have existed without financial support. It was made possible through Grant 90-C-635 from the U.S. Office of Child Development to Michigan State University which is gratefully acknowledged. I am doubly indebted to Bradley S. Greenberg and Charles K. Atkin, who, as Project CASTLE Co-directors, made the data for this dissertation available to me. Dr. Greenberg was my advisor and recognition goes first to him for his time and assistance during preparation of this dissertation and throughout my graduate career. I worked closely with Dr. Atkin in collection of this and other data sets and learned much from him. My other committee members, Thomas F. Baldwin, and F. Elaine Donelson, are to be thanked for their ii encouragement and review of this dissertation. Many others deserve thanks for their various efforts on my behalf. Those whose efforts were extraordinary con- tributions to this dissertation include Felipe Korzenny, Edward L. Fink, Dan G. Drew, and Byron Reeves. iii TABLE OF CONTENTS Chapter I INTRODUCTION . . . . . . . . Theoretic Perspective . . . . . . Attention Processes . . . . . Retention Processes . . . . . Motoric Reproduction Processes . . Reinforcement/Motivational Processes . . . . . . . Modeling Effects . . . . Inhibitory/Disinhibitory Effects . . Response Facilitation Effects . . Definition of Pro— and Antisocial Behavior. Prosocial Behavior . . . . . Altruism . . . . . . Affection . . . . Explanation of Feelings . . . Antisocial Behavior . . . . . Physical Aggression . . . . Verbal Aggression . . . . Specification of Variables and Hypotheses . Exposure to Televised Behavior . . Identification with Televised Behavior Models . . . . . Interaction Hypotheses . . Prosocial Exposure with Antisocial Exposure . . . . . . Prosocial Identification with Anti- social Identification . . . . Exposure with Identification . . . Demographic Variables . . . . Sex . . . . . . . . Age . . . . . . . . Summary . . . . . . . . . iv Page 10 11 11 12 12 12 13 13 14 14 14 15 15 15 16 17 2O 24 24 25 25 26 27 28 30 Chapter II DATA COLLECTION AND INDEX CONSTRUCTION. Questionnaire Data . . . . . Children's Social Behavior Variables Altruism Index . . . . Affection Index . . . . Self-Expression Index . . Verbal Aggression Index . . Physical Aggression Index . Television Exposure Variables . Character Identification Variables Content Analysis Data . . . . . Combined Indexes . . . . . . Exposure-Behavior Indexes. . . Identification-Behavior Indexes . Relationships Among Indexes and Control Variables . . . . . . Relationships with Sex . . . Criterion Variables by Sex Exposure Variables by Sex . Identification Variables by Sex . Relationships with Grade . . . Criterion Variables by Grade. Exposure Variables by Grade . Identification Variables with Grade . . . . . . Summary . . . . . . . . III ANALYSIS AND RESULTS . . . . . Analysis Procedures . . . . . Main Effects of Exposure . . . . Prosocial Exposure . . . . Antisocial Exposure . . . . Page 31 31 32 36 36 36 37 37 38 40 43 48 48 49 50 52 52 53 53 54 54 55 57 57 59 59 63 63 63 Chapter III (cont'd.) Crossed Effects of Exposure . . . . Prosocial Exposure . . . . . Antisocial Exposure . . . . . Main Effects for Identification . . . Prosocial Identification . . . . Antisocial Identification. . . . Crossed Effects of Identification . . . Prosocial Identification . . . . Antisocial Identification. . . . Interaction Effects . . . . . . Prosocial Exposure on Antisocial Exposure Effects . . . . . Antisocial Exposure on Prosocial Exposure Effects . . . Prosocial Identification on Antisocial Identification Effects . . . Antisocial Identification on Prosocial Identification Effects . . . . Prosocial Identification on Prosocial Exposure Effects . . . . Antisocial Identification on Anti- social Exposure Effects. . . APPENDICES O O O O O O O O O O A. Questionnaire Television Survey . . . B. Item Analysis for Criterion Measures. . REFERENCES 0 O O O O O O O O 0 vi Page 65 65 67 68 68 69 70 7O 72 73 73 81 84 84 84 86 93 125 125 136 150 Table 2.1. LIST OF TABLES Descriptive Statistics and Reliabilities for Children's Social Behavior Indexes. . Correlations Among Children's Behavior Indexes . . . . . . . . . Levels of Exposure to Selected Programs Levels of Identification with Selected Television Characters . . . . . . Frequencies and Rates of Selected Social Behaviors for a Composite Week of Television Drama . . . . . . Behavior Profiles for Selected Television Programs . . . . . . . . . Behavior Profiles for Selected Television Characters . . . . . . . . Descriptive Statistics and Correlations for Exposure-Behavior Variables . . . . Descriptive Statistics and Correlations for Identification-Behavior Variables . . . Correlations Among Exposure and Identification Variables . . . . . F Statistics for Criterion Variables by sex 0 O O O O O O O O F Statistics for Exposure Variables by Sex . . . . . . . . . F Statistics for Identification Variables by Sex . . . . . . . . . F Statistics for Criterion Variables by Grade . . . . . . . . . vii Page 37 38 39 42 45 46 47 49 50 51 52 53 54 55 Table Page 2.15. F Statistics for Exposure Variables by Grade . . . . . . . . . 56 2.16. F Statistics for Identification Variables by Grade . . . . . . . . . 58 3.1. Correlations Between Prosocial Exposure and Prosocial Behavior . . . . . 64 3.2. Correlations Between Antisocial Exposure and Antisocial Behavior . . . . . 64 3.3. Correlations Between Prosocial Exposure and Antisocial Behavior . . . . . 66 3.4. Correlations Between Antisocial Exposure and Prosocial Behavior . . . . . 67 3.5. Correlations Between Prosocial Identifi- cation and Prosocial Behavior . . . . 68 3.6. Correlations Between Antisocial Identifi- cation and Antisocial Behavior. . . . 69 3.7. Correlations Between Prosocial Identifi- cation and Antisocial Behavior. . . . 71 3.8. Correlations Between Antisocial Identifi- cation and Prosocial Behavior . . . . 72 3.9. Contingent Correlations Between Antisocial Exposure and Antisocial Behavior Within High and Low Prosocial Exposure Groups. . 74 3.9A. Contingent Correlations Between Antisocial Exposure and Antisocial Behavior Within Prosocial Exposure Quartiles . . . . 77 3.93. Contingent Correlations Between Antisocial Exposure and Antisocial Behavior Within High and Low Overall Prosocial Exposure Groups. . . . . . . . . . 80 .3.10. Contingent Correlations Between Prosocial Exposure and Prosocial Behavior Within High and Low Antisocial Exposure Groups . 82 viii Table 3.11. 3.13A. 3.13B. 3.14. 3.14A. B1. B2. BB. B4. B5. Contingent Correlations Between Antisocial Identification and Antisocial Behavior Within High and Low Prosocial Identifica- tion Groups . . . . . . . Contingent Correlations Between Prosocial Identification and Prosocial Behavior Within High and Low Antisocial Identifi- cation Groups . . . . . . . Contingent Correlations Between Prosocial Exposure and Prosocial Behavior Within High and Low Prosocial Identification Groups. . . . . . . . . Contingent Correlations Between Prosocial Exposure and Prosocial Behavior Within High and Low Prosocial Identification Median Groups . . . . . . . Contingent Correlations Between Prosocial Exposure and Prosocial Behavior Within Prosocial Identification Quartiles . Contingent Correlations Between Antisocial Exposure and Antisocial Behavior Within High and Low Antisocial Identification Groups. . . . . . . . . Contingent Correlations Between Antisocial Exposure and Antisocial Behavior Within Quartiles of Antisocial Identification. Item Analysis for Children's Altruistic BehaVior O O O O O O O 0 Item Analysis for Children's Affection Behavior . . . . . . . . Item Analysis for Children's Expressive Behavior . . . . . . . . Item Analysis for Children's Verbal Aggression. . Item Analysis for Children's Physical Aggression. . . . . . . . ix Page 85 87 89 91 92 94 95 136 139 141 144 147 CHAPTER I INTRODUCTION Much of the concern about the effects of television centers on the proposition that children will imitate tele- vised behavior. Scant attention has been paid to the possi- bility that imitation of one type of behavior would reduce enactment of other types of behavior. If television in- creases the likelihood of antisocial behavior in any specific context, it probably simultaneously decreases the likelihood of prosocial behavior in the same context. This research focuses on such "crossed effects," that is, the effects of prosocial television on antisocial behavior and of antisocial television on prosocial behavior. Both exposure to tele- vision and identification with television characters are considered as predictors of social behavior. Most of the research on children and television has examined antisocial effects. In general, this research sup- ports the conclusion that televised violence can lead to antisocial behavior in children. Comstock, gt_§l., (1975) after surveying some 30 reviews of research in the area, concluded that scholars agree that ". . . under at least some circumstances, viewing of violence increases the like- lihood of some form of subsequent aggression . . . (p. 30). 1 2 On the other hand, some research supports the conclusion that television can affect such prosocial behaviors as help- ing, task persistence, delay of gratification, and cooPera- tion (e.g., Stein and Friedrich, 1973), Friedrich and Stein, 1975; Sprafkin, §t_§1,, 1975; Rubinstein, gt_§1., 1975). Under public pressure generated in light of such findings, the television industry has begun to change its programming. The changes include "santizing" the violence portrayed, introducing family hour and family-oriented pro- grams, and inserting prosocial content in the Saturday morn- ing schedule. These changes apparently have not resulted in an appreciable decline in the number of violent acts por- trayed on television (Gerbner, §£_§l., 1976); however, they may have caused a qualitative difference in the nature and intensity of such portrayals. A recent content analysis that considered both pro- and antisocial behavior reported that a substantial number of both kinds of acts are aired on television (Greenberg, gt_al., 1977). Given that television viewing can enhance the likeli- hood of both pro- and antisocial behavior in children, and that both types of content are readily available, discerning the effects of the current configuration of television is problematic. Among the possibilities that arise are (l) the medium concurrently teaches both types of behavior to all children; (2) because of differences in television use among children, the medium instills prosocial behavior in some 3 children and antisocial behavior in other children, and (3) exposure to both types of television content causes poten— tial effects to cancel out (e.g., because television shows characters resolving conflict both through violence and through reasoned discourse, the medium provides little in- formation on which is the "best" course of action, and, therefore, has little impact on children's behavior). This research examines such possibilities. Theoretic Perspective Numerous theoretic explanations have been offered for the phenomenon of learning through observation of others' behavior (e.g., Miller and Dollard, 1941; Skinner, 1953, 1957; Mowrer, 1960; Bandura, 1965, 1971). The major dif- ferences in these theories center on the number of necessary and sufficient conditions for observational learning. As an historical progression, these theories move from mechanistic explanations based on drive reduction and conditioning to more complex explanations incorporating cognitive, mediation- al processes. Miller and Dollard's theory, which is based on Hullian psychology, relies heavily on the concept of drive. Drive is defined as any strong stimulus that impels the organism to act. Drives, which are internally generated, may be pri- mary (based on biological needs for such things as food, rest, or sex) or secondary (socially modified or obscured 4 primary drives). In the general Hullian paradigm, stimulus cues stimulate internal responses or drives that impel the individual to an external response. Rewards, which are con- ceived as drive-reducing behavioral outcomes, determine whether a response will be repeated. If a response is un- rewarded, it is less likely to be repeated in the presence of the original cue. Miller and Dollard describe several processes of social influence; however, only one--matched-dependent be- havior--is particularly germaine. In matched-dependent learning, the model's response to a particular stimulus cue serves as a cue for the observer who then matches the models's behavior. The rewards accruing to the observer's matching behavior determines whether the observer will again respond by matching the model's behavior. This process, then, ac- counts more for learning to imitate than it does for learning through imitation. It is clear that Miller and Dollard's theory posits several necessary conditions for observational learning. These include a drive state, observation of modeled behavior, internal mediation processes linking the modeled cues and drive, overt behavior, and reward for behavior. Skinner's theory is very similar to Miller and Dollard's. The major difference is that Skinner drOps all reference to drive and internal mediation processes. This is accomplished through a tautological definition of reward IE (D h: In 5 as any behavioral outcome that increases the likelihood of repeated behavior in the presence of the same stimulus cues. Skinner is more explicit than Miller and Dollard con— cerning the origin of first imitative responses. Skinner holds that first imitations are chance occurrences which de- rive subsequent reinforcement. Through a series of succes- sive approximations in which increasingly rigorous criteria are established for administration of reinforcement, the observer comes to reproduce modeled behavior more accurately. Skinner also stresses the role of higher order con- ditioning in which cues take on behavior modifying properties through contiguity with previous response producing cues. After generalized imitative behavior has been develOped through consistent reward, higher order conditioning can be evoked as an explanation for increasingly complex instances of imitative behavior. Skinner's theory, like Miller and Dollard's, then is primarily a theory of learning to imitate. Skinner's theory posits as necessary conditions for learning: observation of modeled behavior, performance of behavior, and reinforcement for behavior. Cognitive or mediational processes are eliminated from consideration as explanations for behavior. Mowrer (1960),on the other hand, stresses the role of internally generated stimuli. Beginning with the assumption of a positive relationship between the model and the observer, luowrer posits that activities of the model become associated 6 with reinforcing consequences to the observer. When this re- inforcing link between the behavior and pleasurable internal states becomes strong enough, observers may generate these desired outcomes by reproducing the behavior. In short, the observer comes to imitate behavior because performance of the behavior has come to be associated with "feeling good." Mowrer distinguishes two processes of observational learning which differ basically in terms of the directness of reinforcement to the observer. The first process relies on direct reinforcement of an observer by a model and has little direct application to learning from television. How- ever, the second process, "empathetic" learning, is directly relevant. In empathetic learning, the model's behavior is overtly reinforced and the observer is assumed to be able to vicariously experience these rewards. Therefore, the likelihood of the observer reproducing the modeled behavior in the presence of similar stimulus cues is heightened. This occurs because the observer seeks the same rewards as the model received. By relying on cognitive processes, Mowrer diminishes the number of conditions assumed to be necessary for obser- vational learning. All that Mowrer's theory requires is observation of modeled behavior from which rewarding conse- quences can be derived by the observer. Since the rewarding consequences may be derived either from pleasurable internal states of the observer or from cognitive inferences by the O [‘1‘ (n ROI . u. q:& - a "Ju 7 observer, neither performance of the behavior by the observer nor externally administered reward to the observer are re- quired for observational learning. Bandura (1965 , 1971) eliminates reward, either to the model or to the observer as a necessary condition for observational learning. Bandura distinguishes acquisition of cognitive responses from performance or overt enactment of behavior. Further, he defines learning in terms of the acquisition process. In his own experimental work (e.g., Bandura, 1965), he demonstrated that external rewards to the model and/or to the observer are not essential for acquisi— tion. In this study, observers in non-reward conditions were able to reproduce modeled behavior when strong incentives were offered for them to do so. It should be noted that ex- periments like the one cited do not preclude the possibility that observers inferred reinforcing consequences for imita- tion, and thus, do not contradict Mowrer's formulation. Also, Bandura holds that continued patterns of performance are determined by reinforcement. In this regard, his formu- lation is very similar to that of Skinner who is interested only in overt enactment of behavior. Bandura's theory is clearly data based and incorpor- ates all variables and propositions for which there is strong enmdrical support. Thus, Bandura recognizes a wide range of ‘variables as facilitators of observational learning includ- iJug motivation, attention, performance ability, cognitive L“" lb any" Bio-Li V- Hng‘ “4“I L IO .1 t 9 p U :a’vvj 8 processes, and reward to both the model and the observer. Because of this empirical eclecticism, it is the most general and probably the most accepted contemporary formulation. For that reason it was chosen as the guide for this research and is outlined in more detail below. Bandura frequently writes under the label of "Social Learning Theory" by which he apparently means to encompass the entire set of perspectives discussed above. Bandura's specific formulation is usually termed "Mediational-Stimulus Contiguity Theory." The central proposition of this theory is that, ". . during the period of exposure, modeling stimuli elicit in observing subjects configurations and sequences of sensory experience which . . . become centrally integrated and struc- tured into perceptual responses" (1965, p. 10). That is, through observation individuals develop cognitive represen- tations of responses associated with specific stimuli. These stimuli can then serve as cues for the cognitive repre— sentations of the responses. Thus, when the observer is placed in a behavioral field containing the cues, they elicit the cognitive responses which may be translated into behavior. Television provides children with the Opportunity to (observe a wide range of behaviors. To the degree that these Jbehaviors are consistently performed in the contexts of (Ither stimulus cues, it is possible for observers to develop 6X. In. «V. v.- I b. ‘3‘; .‘O‘ V‘c . 'V. 9 associations between the behaviors and the cues. While the existence of behavior and contiguous stimuli are necessary for the development of cognitive associations, they are not sufficient. Bandura posits two necessary subprocesses for this development--attention and retention. Attention Processes Exposure to modeling stimuli is insufficient for de- velopment of cognitive associations unless the stimuli are attended to and registered at the sensory level. Attention is governed by several factors including incentive conditions, observer characteristics, and properties of the modeling stimuli. Television drama is designed to attract maximal audiences, and Bandura assumes that televised models are intrinsically interesting enough to attract attention. Attention may be directed by purely physical prOper- ties such as size and intensity. However, the distinctive— ness of model attributes has been shown to be more important. Among these attributes are competence, status, age sex, race, and attractiveness (summarized in Bandura, 1969, p. 138). Bandura himself eschews the term "identification" on the grounds that it generally has no meaning distinct from that encompassed in the more general term, observational learning. In this research identification has a distinctive ineaning and is defined as conscious approval of a specific .individual as an appropriate model for one's own behavior. {this general definition is meant to encompass a myriad of sc- bi rit Va 50‘ fl! \r 10 factors that have been shown to focus attention and facili- tate observational learning. These factors include model attributes indicating that the model's behavior generally de- rives reward (e.g., status, power, and prestige). It also includes relational states between model attributes and ob- server attributes indicating that the model is similar to the observer, and, therefore, similar consequences will accrue to the behavior of both (e.g., similarity in sex, age, or race). Retention Processes Retention processes are essential to contiguity theory because there may be considerable time elapsed between the observation of modeling cues and the occasions which the ob- server finds appropriate for enactment of modeled behavior. Covert rehearsal and verbal coding of behavior sequences have been found to facilitate retention. There is ample evidence linking television viewing to various attitudes and behaviors including not only various social behaviors but also such diverse phenomena as beliefs about crime (Dominick, 1974) and perceptions of sex roles (Miller and Reeves, 1975). Therefore, it is reasonable to assume that televised behavior is simple enough and repetitive enough for cognitive representations of behavior and contiguous cues to be easily retained. Bandura distinguishes the above processes which govern acquisition of cognitive responses from those which govern performance. Performance, Bandura states, is governed by now 3‘ n b4» u M’V‘fi «I A. an " 11.-.. “N -.u , ~“F .o'ti ‘U‘ ’_ ll motoric reproduction processes and by reinforcement/motiva- tional processes. Motoric Reproduction Processes Under this rubric Bandura simply calls attention to the fact that observers must possess the requisite physical skills to reproduce modeled behavior. Reinforcement/Motivational Processes Bandura's theory emphasizes cognitive processes involv- ing inferences about the reward value of performance. No single proposition of the theory has received more support than the one stating that overt reinforcement accrued to the model increases the likelihood of imitation. In general, the theory posits that if models are reinforced for behavior, ob- servers will reason that they would be similarly reinforced for similar behavior in similar situations. In addition to obvious inferences based on overt rein- forcement, it has been found that children can infer the con- sequences of behavior from a wide variety of cues including verbal labels, emotional responses of the model, attributes of the model that indicate power and prestige, and the ob- server's own emotional and physiological states. Bandura classifies the behavioral effects of observa- tional learning into three categories depending on (1) the degree to which the outcome behaviors already exist in the observer's behavioral repertoire, and (2) the degree to which M54: oav‘v \ u. CEAC a‘ Cu 5» f». 9“ . hh‘ Al... I :u -nnu fit 5 A Q o‘.\ e .. a. :u w‘ sh. 12 the outcome behaviors are socially sanctioned. Modeling Effects Modeling effects occur when the observer acquires a new response pattern through observation of a highly novel behavior. Component parts of the novel response are assumed to already exist in the observer's repertoire, but the per- formance is in a new combination or sequence. Inhibitory/Disinhibitory Effects When observation of the consequences of a model's be- havior results in modification of an observer's performance of a negatively sanctioned behavior, inhibitory/disinhibitory effects are said to have occurred. Inhibitory effects result from punishment to the model's behavior strengthening the ob- server's cognitive association between performance and neg- ative consequences. Disinhibitory effects result from either lack of punishment or even reinforcement to the model's be- havior, weakening the observer's association between perform- ance and negative consequences. Response Facilitation Effects Response facilitation effects occur when the behavior elicited already exists in the observer's repertoire, and modeling stimuli serve as informative cues that conditions (are appropriate for performance. These effects are distin- guished from modeling effects in that novel behaviors are not involved and from inhibitory/disinhibitory effects in that 13 negative social sanctions are not involved. The focus of this research is on pro- and antisocial behaviors which for the most part probably already exist in children's behavioral repertoires. Thus, concern here is centered on inhibitory/disinhibitory effects and response facilitation effects. In general, it is argued that the vast array of models provided by television can affect chil- dren's perceptions of the appropriateness of specific class- es of behavior and thus their rates of performance of those behaviors. This section provides a theoretic framework for explanation of the effects of television on children and allows for derivation of the specific hypotheses offered be- low. Definition of Pro- and Antisocial Behavior The concepts of pro- and antisocial behavior are central to this research both with regard to the predictor variables concerning television content and with regard to the criterion variables concerning children's behavior pat- terns. This section defines these concepts in a general way that can be applied to both sets of variables. Prosocial Behavior In any particular situation involving interactions annong persons there exists a wide range of possible behaviors. Behaviors that create closeness between the person involved and thus are socially approved can be labeled prosocial. 14 To be more specific, prosocial behaviors include those acts which can be presumed to be beneficial to their recipient and which may in turn elicit reciprocal benefits to the actor. Prosocial behaviors may be divided into several dis- tinct categories, three of which are used in this research: Altruism. Altruism refers either to acts of giving physical objects to others or to acts of assistance to others (except where the other's goals are illicit). Some authors restrict the concept to situations in which the individual acts without hope of reciprocation (e.g., Bryan and London, 1970; Krebs, 1970). This restriction is not used here. Affection. Affection refers to displays of positive affect toward others. Affectionate behaviors may be either verbal (e.g., "I love you,") or physical acts (e.g., a hug, or a kiss). Explanation of Feelings. Expression of feelings con- sists of verbal statements which are made in attempts to affect positive outcomes. They include attempts to increase understanding, or to resolve strife. Several other behaviors such as cooperation, obedience to rules, delay of gratification, task persistence, and con- trol of other's antisocial behavior have been considered under the prosocial rubric. These behaviors are not con- sidered in this research because content analysis reveals that they occur with relatively low frequency in television drama (Greenberg, et al., 1977). “H o A». and . n6~ “ob. 15 Antisocial Behavior Behaviors that create or extend interpersonal distance and thus are socially disapproved can be labeled antisocial. Antisocial behaviors include those acts which can be assumed to be harmful to their recipient and which may in turn elicit harmful responses to the actor. Categories of antisocial be- havior used in this research include: Physical Aggression. Physical aggression refers to acts that result in damage or injury to other persons. Phy- sical aggression included such acts as hitting, shooting, stabbing, and throwing objects at other persons or threaten- ing such acts. Verbal Aggression. Verbal aggression refers to sym- bolic acts that result in psychological damage to other persons or hold them up to social opprobrium. These include insults, threats, acts of rejection, and general hostility. Several other behaviors such as abridgment of privacy, deceit, theft, and destruction of property are often con- sidered to be antisocial. Again, these behaviors are not con- sidered because they occur infrequently on television (Greenberg, gt_§l., 1977). The above definitions distinguish pro- and antisocial behavior on the basis of their outcomes rather than upon the motivational states of the persons performing them. Defini- tion in terms of motivation or personality properties of individuals would imply persistent pro- and/or antisocial L. 96.1: #‘h:¥ yeluh .b 3.1 l6 behavior patterns across situations. It is assumed here that such persistent patterns do not necessarily exist and that an individual may be prosocial in one situation and antisocial in another. It should be stressed that the definitions refer to performance of behavior rather than to knowledge of how to perform behavior. If television teaches children the prin- ciples of building Molotov cocktails, it is not of concern here unless they act on that knowledge and construct such devices. Specification of Variables and Hypotheses This section defines the variables of concern to this research and describes relationships among them. The vari- ables are considered in three distinct classes: (1) cri- terion variables concerning children's patterns of pro- and antisocial behavior; (2) predictor variables indicative of children's involvement with pro- and antisocial television, and (3) demographic variables--grade and sex--used as statis- tical controls. The variables and their interrelationships are de- scribed here at a general theoretical level. Methods of Operationalizing variables and subjecting hypotheses to sta- tistical tests are discussed in Chapters II and III. Hypotheses concerning main effects of predictor vari- ables are offered first followed by hypotheses concerning 17 interactions among them. Relationships among the demographic variables and the predictor and criterion variables are gen— erally well documented. These relationships necessitate use of the demographic variables as statistical controls. How- ever, no formal hypotheses are offered concerning the demo— graphic variables in order to limit the scope of this research and to focus on the substantively more interesting television variables. Exposure to Televised Behavior There is abundant evidence that exposure to a specific behavior can lead to performance of that behavior. Experi- ments have documented this relationship across an extremely broad range of behaviors including courage, aggression, al- truism, affection, and self-criticism (for a summary, see Lesser, 1975). Survey research has focused primarily on television and aggression; however, the evidence from such research is consistent with the proposition that exposure to behavior is associated with performance of that behavior (for summaries, see Baker and Ball, 1969; Chaffee, 1972; Liebert, gg_§l., 1973). Field studies of the effects of exposure to prosocial television are rare; however, Stein and Friedrich (1973) and Friedrich and Stein (1975) have found such effects. The theoretic rationale given for such findings gen- erally is that behavioral cues contained in specific situa- tions evoke cognitive responses acquired through exposure to television. If that exposure has been dominated by prosocial 18 behavior, the likelihood is increased that a prosocial re- sponse will be evoked and performed. Thus, it is straight- forward to hypothesize that: H1: Exposure to televised prosocial be- hav1or Will be p031t1vely assoc1ated with performance of prosocial behavior, and, H : Exposure to televised antisocial be- havior will be positively associated with performance of antisocial behavior. The above hypotheses essentially replicate past re- search and are not of central concern here. However, they are necessary precursors to other hypotheses offered below. Hypotheses l and 2 are entirely plausible when con- sidered separately; however, problems arise when they are considered simultaneously. Because most television programs contain both types of content, heavy exposure to television implies heavy exposure to both pro- and antisocial behavior. Thus, uncritical acceptance of Hypotheses l and 2 leads to the conclusion that heavy television viewers perform both more prosocial behavior and more antisocial behavior. This seems unlikely if for no other reason than that television viewing takes time from other activities. Thus, heavy tele- vision viewers would have less time for any social activities and could be expected to manifest less behavior of both types than would light viewers. In addition to the direct effects hypothesized above, it is likely that television has indirect effects on chil- dren's behavior which operate on their tendencies to choose 19 among behavioral alternatives (cf., Liefer and Roberts, 1972). Children exposed primarily to prosocial content may learn to associate specific cues with prosocial behavior while children exposed primarily to antisocial content may learn to associate similar cues with antisocial behavior. Thus, when faced with interpersonal conflict, for example, some children may attempt resolution through prosocial reasoned discourse while others may turn to verbal aggres- sion (cf., Roloff, 1975). If television exposure increases the likelihood of one behavior in response to specific cues, it must simultane- ously decrease the likelihood of other behaviors in response to the same cues. Therefore, increasing the likelihood of prosocial behavior must decrease the likelihood of antisocial behavior and vice versa. This reasoning leads to the hy- potheses that: H3: Exposure to televised prosocial be- hav1or w1ll be negat1vely assoc1ated with performance of antisocial be- havior, and, H4: Exposure to televised antisocial be- hav1or w1ll be negatively assoc1ated with performance of prosocial behavior. These crossed variations of Hypotheses l and 2 have rarely been considered in past research. A review of research reveals only one study that considered crossed effects, a field experiment by Stein and Friedrich (1973). These re- searchers established baseline rates of pro- and antisocial behavior through observation of pre-schoolers at play. They 20 then exposed one group to a prosocial television program and another to an antisocial program. On subsequent observation, they found decreases in task persistence and rule obedience among children exposed to antisocial television along with direct effects of both pro- and antisocial television. Pro- social effects were primarily among children from low socio- economic status families and children of high intelligence. Antisocial effects were confined to children with high base- line scores in aggression. Another possibility that comes from consideration of Hypotheses l and 2 is that mixed viewing of pro- and anti- social television causes potential effect to cancel out. Friedrich and Stein used relatively pure pro- and antisocial stimuli and did not consider this possibility. Mixed view- ing is probably the rule in natural settings so this possi- bility must be considered. It is discussed in a later section concerning interaction hypotheses. Identification with Televised Behavior Models Researchers often assume that children's affective relationships with television characters are key predictors of the medium's impact on social behavior (cf., Weiss, 1969). Several laboratory researchers have linked observers' per- ceived closeness to models with their recall or replication of observed behavior (e.g., Maccoby and Wilson, 1957; Tannen- baum and Gaer, 1965; Rosekranz, 1967). Similar findings have been reported with reference to broadcast television #5 r. QR“ c! MA ., Mi). :— L.“ R‘J ‘I‘ .‘d .. -v “Us RUN n 4: mu. 21 characters (W. Miller, 1968, 1975; Meyer, 1973; Donohue, 1975; Greenberg, gp_§l., 1976). It should be noted that the studies dealing with broadcast television characters dealt with children's perceptions of behavioral similarity rather than with behavior considered separately from such characters. Greenberg, §£_§l,, asked children about their desires to model; the other researchers cited above compared children's responses to hypothetical situations to their per— ceptions of what television characters would do in the same situations. The affective relationships observers have with models are generally discussed under the rubric of identifi- cation. However, this term has been used in so many differ- ent contexts with different meanings that several researchers have advocated its abolition (e.g., Bandura, 1962; Sanford, 1955). While the term is too useful to abolish, it is necessary to define its use in any specific context. In this research identification is defined as a con- scious approval of a specific individual as an appropriate model for one's own behavior. It is assumed that the indi- viduals with whom the person identifies possess qualities desired by the person, and that the identifying person rea- sons that by imitating the model he/she will come to possess the same qualities. Television characters obviously possess desirable qualities in differing degrees. Thus, some characters, more 22 than others, will command attention, facilitate retention, and enhance the likelihood of inferences concerning the re- inforcement value of behaviors. Thus, identification should have direct effects on behavior: H5: Identification with televised models will be positively associated with performance of prosocial behavior to the degree that the models perform prosocial behavior, and, H : Identification with televised anti- social models will be positively as- sociated with performance of antisocial behavior to the degree that the models perform antisocial behavior. Because different characters may respond to specific situations in different ways, identification with them leads to development of differing cognitive associations between behaviors and stimulus cues. Inasmuch as these associations are systematic with reference to pro- and antisocial behavior, differential identification should lead to differential re- sponse proclivities. This reasoning, like that offered for Hypotheses 3 and 4, leads to crossed variations of Hypothes- es 5 and 6: H7: Identification with televised models will be negatively associated with performance of antisocial behavior to the degree that the models perform pro- social behavior, and, H : Identification with televised models will be negatively associated with performance of prosocial behavior to the degree that the models perform antisocial behavior. 23 While behavior type itself may be a criterion for children's choices of televised behavior models, several researchers have shown that other factors such as model's physical strength and physical attractiveness are related to such choices (Reeves and Miller, 1976; Reeves and Greenberg, 1977). Therefore, children might simultaneously identify with both pro- and antisocial models. The ramifications of this possibility are considered under discussion of inter- action hypotheses. Interaction Hypotheses An interaction occurs when the effect of one predic- tor variable on a criterion variable is different for dif- ferent values of another predictor variable (cf., Winer, 1971; Namboodiri, §E_gl,, 1975). There are numerous possi- bilities for interactions among variables. These include enhancer effects in which the impact of one predictor is in- creased at high levels of another variable, depressor effects in which the impact of one predictor is decreased at high levels of another variable, and curvilinear effects. Because of this wide range of possibilities, it is incumbent on the researcher hypothesizing interactions to specify both their nature and to offer rationales for their existence. This section specifies several interactions of the predictor variables defined above, prosocial exposure, anti- social exposure, prosocial identification, and antisocial identification. With four predictor variables and two classes 24 of criterion variables, there are 22 possible interactions. (The operational variables described in Chapter II include even more variables and, therefore, more interactions.) To limit the scope of this analysis, only two-way interactions are considered. Further, because some interactions are un- likely to exist to any appreciable degree, they are excluded. For example, it is unlikely that prosocial exposure and anti- social identification co-occur at high levels or interact in important ways. These limitations are consistent with the recommendations of methodologists who note that some inter- actions may lack substantive interest, may be difficult to interpret, and that their inclusion decreases statistical power and increases the possibility of errors of inference. Prosocial Exposure with Antisocial Exposure Low levels of exposure to pro- and antisocial tele- vision would be expected to have minimal effects on social behavior. High levels of exposure to both types of content would not be expected to allow for develOpment of consistent patterns of association between stimulus cues and social be- havior. Therefore, minimal effects would be expected at high levels of exposure to both pro- and antisocial content. However, when one type of exposure is at a high level and the other at a low level, the exposure at a high level is free to Operate and to allow development of consistent as- sociations between specific stimulus cues and behaviors. Therefore: 25 H : Exposure to prosocial television will suppress the effects of antisocial exposure on antisocial behavior, and, H ° Exposure to antisocial television will suppress the effects of prosocial television on prosocial behavior. Prosocial Identification with Antisocial Identification The rationale for the interaction of these variables is analogous to that offered for Hypotheses 9 and 10. H11: Identification with televised models who perform prosocial behavior will suppress the effects of identification with televised models on antisocial behavior attributable to the degree to which the latter perform antisocial behavior, and, H12: Identification with television models who perform antisocial behavior will suppress the effects of identification with televised models on prosocial behavior attributable to the degree to which the latter perform prosocial behavior. Exposure with Identification Since exposure to televised behavior models is neces- sary for identification with those models and identification should lead to repeated exposure, these variables should be highly related. Also, it can be argued that the main effects of identification should be minimal. It is unlikely that high identification would affect observers' behavior if the observers have not had substantial exposure to the model. Without such exposure, the observers would lack sufficient knowledge of modeled behavior to duplicate it. On the other 26 hand, heavy exposure and high identification with a specific type of behavior should allow for the development of strong, consistent associations between stimulus cues and behavior. Therefore: H13: Identification with television models will enhance the effects of prosocial exposure on prosocial behavior to the degree that the models perform pro- social behavior, and, Identification with television models will enhance the effects of antisocial exposure on antisocial behavior to the degree that the models perform anti- social behavior. 14‘ Demographic Variables Demographic variables--age and sex—-are included in this research not because they are of substantive interest, but because they are known to be related to the central vari- ables of concern. Therefore, failure to include these vari- ables could lead to obvious misinterpretations of the data. The relationships between prosocial television exposure and prosocial behavior, for example, could be overestimated if the effects of sex on both variables were not considered. The demographic variables are clearly causally prior to the other variables in the sense that they cannot be taken as the results of television use or social behavior. Rather, the causal chain must be construed to operate in the direction of age and sex affecting the other variables. The substantive and methodological ramifications of this causal priority are discussed below. 27 S35. Sex is a major predictor of social behavior. Maccoby and Jacklin (1974), after a comprehensive review of the literature on psychological sex differences, state: "The sex difference in aggression has been observed in all cultures in which the relevant behavior has been observed" (p. 352). While conventional wisdom holds that girls are more nurturant and altruistic than boys, Maccoby and Jacklin conclude that there is little evidence to support this as- sertion. The overall finding, they state, is one of sex similarity for these prosocial behaviors. Sex differences in television program preferences and exposure patterns are apparent in children as young as four years old (Lyle and Hoffman, 1972). These differences seem to continue throughout adult life (Israel and Robinson, 1972). In general, studies of sex differences in program exposure show that males view more television aggression than females. Similar differences have been found in children's choices of television characters as role models. Several researchers have reported that children have strong prefer- ences for models of their own sex. This has been reported in experiments (Maccoby and Wilson, 1957; Sprafkin, gE_§1., 1975), and in surveys (Miller and Reeves, 1976; Reeves and Miller, 1976; Greenberg, EE_21" 1976). Given that the majority of violent television char- acters are male (Gerbner, 1972), it is reasonable to assume 28 that boys will identify more frequently with violent models than will girls. Two researchers (Meyer, 1973; Donohue, 1975) explicitly report that boys are more likely to choose violent television models as "favorite characters." Conversely, content analysists have noted that female characters on television are more likely to be nurturant and affectionate than male characters (Busby, 1975; Tedesco, 1974). Thus, girls would be expected to identify with prosocial models more frequently than boys; however, there are no data available on children's choices of prosocial models. Given that sex is related to children's social be- havior and to their television use, it is obvious that the variable must be included in research attempting to link television and social behavior. Otherwise, there is danger of imputing a causal relationship between television use and social behavior when observed correlations between them could be accounted for solely by their mutual dependence on a causally prior variable--sex. 533. The relationships between age and other vari- ables in this research are far less clear than are the rela- tionships with sex. While there is substantial evidence that children's moral judgments change dramatically as they mature (Piaget, 1948; Kohlberg, 1964), it is not clear how these cognitive shifts affect behavior. Knowledge that one child's behavior is governed by external constraints and another child's, by internalized moral standards does not necessarily 29 mean that different predictions can be made about their be- havior. It is reasonable to assume that children's interpre- tations and use of televised information shifts with cogni- tive developmental stage; however, these shifts have not been well documented (cf., Roberts, 1973). Relationships between age and television exposure are well documented (Roberts, 1973), but the implications of these patterns are unclear. Young children are heavy view- ers of Saturday morning programs which now contain both high- ly prosocial and highly antisocial content. As children grow older, their Saturday morning viewing decreases and they turn more to adult programs. These programs are also quite mixed in content. Children's total television viewing increases throughout their elementary school years and begins to de- cline as they approach adolescence (Schramm, EE_El-r 1961; Roberts, 1973). The effects of age on children's patterns of identifi- cation with television characters apparently have not been investigated. Reeves (1976), in a study of children's gen- eral perceptions of television characters found that younger children tend to discriminate characters on the basis of phy- sical attributes (e.g., strength and attractiveness while older children depend more on behavioral attributes (e.g., activity).l From Reeves' study, it might be inferred that older children would base their identification patterns more 30 on the social valence of model behavior than would younger children; however, there are no data bearing directly on this point. It is clear from the above discussion that age could have an important impact on the relationship between tele- vision use and social behavior. However, its inclusion in this research should be viewed as exploratory. SummaEy The purpose of this research is to link children's use of television with their patterns of pro- and antisocial behavior. Reasoning from mediational-stimulus contiguity theory, hypotheses were offered relating four predictor variables--exposure to prosocial television, exposure to antisocial television, identification with prosocial tele- vision models, and identification with antisocial television models-~to performance of pro- and antisocial behavior. Both main effects and interactions among the predictors were considered. Reasons for inclusion of age and sex as control variables were discussed. CHAPTER II DATA COLLECTION AND INDEX CONSTRUCTION The data for this research come from the Project CASTLE Social Behaviors Questionnaire and the content analy- sis of pro- and antisocial behavior on television. The data can be categorized into five distinct sets: (1) demographic characteristics of the questionnaire respondents, (2) chil- dren's reports of their own social behavior, (3) children's reports of their exposure to selected television programs, (4) children's reports of their identification with selected television characters, and (5) descriptions of social be- haviors portrayed on television focusing on the selected programs and characters. This chapter describes procedures of data collection and index construction, and examines re- lationships among the indexes and control variables. Questionnaire Data The CASTLE Social Behaviors Questionnaire (see Appendix A) was administered to 721 fourth, sixth, and eighth graders in the Spring of 1976. The instrument was administered to all respondents in their home classrooms at school. Research assistants read the entire questionnaire to the fourth graders, and older children completed them 31 32 alone. Research assistants were available to help any chil- dren who had problems. The sample included every child in the grades sur- veyed who attended school on the days the questionnaire was administered. The school systems of Haslett, Michigan, and Verona, Wisconsin, participated in the survey. These schools offered a fair mix of rural and urban children from middle and lower socioeconomic strata. The sample included 345 boys and 376 girls. By grade, there were 227 fourth graders, 268 sixth graders, and 226 eighth graders. Children's Social Behavior Variables The children's social behavior variables were assessed with items of three distinct types: (1) hypothetical situa- tion items which asked children to imagine themselves in specific situations and to indicate their probable responses, (2) behavior report items which asked children to indicate the frequencies of specific behaviors in the past week, and (3) contingent report items which asked children to indicate their responses to real-life situations as they occur. The differing types of items were used to "surround" the constructs being measured by offering differential re- sponse constraints. The hypothetical situation items were designed to tap response proclivities independently from the frequency with which children find themselves in specific response situations; the behavior report items, to tap both 33 the frequency of specific response situations and behavior in those situations; and the contingent report items, to tap actual behavior in real-life situations regardless of the frequency of those situations. The differing types of items allowed for repeated questions on the same behavior to increase index reliability. This also increased variation in the questionnaire and served to minimize specific types of response bias (e.g., "yes" bias). All items offered closed response categories and none forced choice among behavior types. Five social behavior indexes were constructed by sum- ming all items, regardless of type, designed to tap each of the following behaviors: altruism, affection, self-expres- sion, verbal aggression, and physical aggression.* Under this procedure individual items contribute to the overall index in prOportion to their standard deviations and their average correlations with other items. Inspection of the standard deviations and correlations within each index re- veals that they are generally of the same order of magnitude (see Appendix B, Tables B1 to B5 Contingent report and be- havior report items tend to have larger variance and there- fore are weighted more heavily. Since these items are * Content analysis had revealed that these behaviors occur on television with sufficient frequency to believe that they might affect children's behavior (Greenberg, et al., 1977). 34 probably the better indicators of performance, this weight- ing is not undesirable. Descriptive statistics and reli- abilities for each of the indexes are shown in Table 2.1. The index reliabilities, which ranged from .59 to .83, were judged to be adequate for this research. An effort was made to assess the validity of the self-report items through collection of data from subsamples of Wiscon- sin repondent's classmates and mothers. Classmate data were collected on 252 respondents and mother data on 293 respon- dents. The behavior types selected for validation were altruism, verbal aggression, and physical aggression. To validate the altruism items, classmates were asked to nominate respondents who "help" others and who "share" with others. The number of nominations for each respondent were summed and these sums were correlated with the individ- ual altruism items. The average correlation between the sum and the individual items was .13. (Individual item-sum cor- relations are reported in Appendix B). The respondent's mothers were asked selected hypothetical situation items. These items were modifications of the hypothetical items asked of children asking for the mother's assessment of how their children would behave in the posed situations. The average itemeto-item correlation between the mother's and children's responses was .11. The average correlation between mother assessments and peer nomination concerning altruism was .08. 35 The validation procedure for the verbal aggression and the physical aggression items was the same as for the altruism items. For verbal aggression, children were asked to nominate respondents who "say mean things." The average correlation of the sum of these nominations was .07. The correlation between the mothers'assessments of children's responses to selected hypothetical situations and children's responses to the same situations was .05. The average cor- relation between mother assessments and peer nomination concerning verbal aggression was .06. For physical aggression, classmates were asked to nominate respondents who "hurt" others. The average correla- tion between the sums of these nominations and the items was .18. The correlation between mothers' assessments of chil- dren's responses to physical aggression situations and the children's responses to the same situations was .05. The average correlation between mother assessments and peer nomination concerning physical aggression was .16. The validity correlations are extremely low and do not allow for a high level of confidence in the self-report behavior indexes. However, the classmate and mother mea- sures used in the validation procedures are probably subject to error. The classmate nomination technique is based on a single question of unknown reliability and may be invalid because of numerous social pressures. Further, the nomina- tion procedure results in highly skewed data with the majority 36 of children receiving no nominations. This skewness prob- ably deflates the validity correlations. Mothers are cer- tainly under pressure to show that their children have socially desirable characteristics which may lead to in- validity in their assessments of respondents' behavior. Thus, the low validity correlations may result either from low validity of the self-report items or from low reliability and validity of the classmate and mother measures. The validity analysis failed to demonstrate that the self—report measures are a valid assessment of the respon- dent's social behaviors. On the other hand, because of problems with the criterion measures used in the validation procedure, it cannot be said that the procedure demonstrates invalidity. Given the face validity of the self—report mea- sures and their acceptable reliability levels, it was de- cided to use these measures in subsequent analysis. However, it should be kept in mind that the validity of the procedures has not been demonstrated. Altruism Index. This index included 10 items covering such behaviors as helping, sharing, and doing favors. Affection Index. This index included five items covering such behaviors as hugging, kissing, and verbal ex- pressions of affection. Self-Expression Index. This index included seven items covering various expressions of individuals' emotional States . 37 Table 2.1. Descriptive Statistics and Reliabilities for Children's Social Behavior Indexes Mean Standard Deviation Reliability Prosocial Indexes Altruism 27.00 4.94 .75 Affection 12.16 3.22 .59 Self-Expression 15.18 3.61 .69 Antisocial Indexes Verbal Aggression 18.29 4.09 .70 Physical Aggression 18.54 5.44 .83 Verbal Aggression Index. This index included eight items covering such behaviors as yelling or screaming at others and "saying mean things." Physical Aggression Index. This index included nine items covering such behaviors as hitting, pushing, kicking and fighting. The index reliabilities, which range from .59 to .83, were judged to be adequate for this research. No formal validity analysis was performed on these indexes; however, the pattern of correlations among them lends some credence to the procedures (see Table 2.2). The correlations are moderately high and positive within the prosocial class and quite high and positive within the antisocial class. Cor- relations between the two classes are moderate and negative. 38 Table 2.2. Correlations Among Children's Behavior Indexes l 2 3 4 5 l. Altruism ..... 2. Affection .54 _____ 3. Expressiveness .50 .50 ----- 4. Verbal Aggression -.18 -.13 -.08 ----- 5. Physical Aggression -.35 -.23 -.26 .69 ----- Television Exposure Variables The CASTLE Questionnaire contained a list of 29 tele- vision programs selected tO maximize the range of behaviors portrayed on them. The respondents were asked to indicate their frequencies of viewing each of these shows. The re- sponse categories were: every week, most weeks, some weeks, never. These categories were assigned code values of 3, 2, l, and 0, respectively. From this list, 15 shows were chosen for this analysis. Descriptive statistics for viewing of these shows are listed in Table 2.3. Use Of all the shows listed in the question- naire was precluded by the fact that several of them con- tained characters used for the identification variables. Inclusion of these shows would have confounded the analysis by having individual characters' behavior included in both the exposure index and the identification index. Thus, the shortened list was used to assure Operational independence between the two sets of indexes. 39 Tab1e2L3. Levels of Exposure to Selected Programs Percent who report watching* . . every most some Program week weeks weeks never mean sd.** Bob Newhart 20 27 28 25 1.42 1.07 Bugs Bunny 30 13 35 27 1.50 1.18 Fat Albert 23 32 22 24 1.46 1.08 Good Times 10 18 45 27 1.10 0.91 Happy Days 54 25 17 04 2.30 0.89 Hong Kong Phooey 12 11 29 47 0.87 1.03 Kojak 05 15 44 36 0.89 0.84 Little House on the Praire 14 17 38 31 1.14 1.00 Pink Panther 28 25 19 28 1.47 1.16 Rhoda 15 22 40 24 1.26 0.98 The Rockford Files 14 23 37 27 1.23 0.99 The Rookies 09 15 45 32 1.01 0.90 Sanford and Son l3 17 42 28 1.14 1.07 Shazam 18 16 30 36 1.16 1.10 Starsky and Hutch 24 12 21 40 1.16 1.22 * Percentages may not add to 100 because of rounding error. ** Means and standard deviations were calculated by scoring every week "3," most weeks "2," some weeks "1," and never "1.“ 40 The final list included five Saturday morning shows (Bugs Bunny, Fat Albert, Hong Kong Phooey, Pink Panther, and Shazam), five situation comedies (Bob Newhart, Good Times, Happy Days, Rhoda, and Sanford and Son), four action- adventure shows (Kojak, The Rockford Files, The Rookies, and Starsky and Hutch), and one family drama (Little House on the Prairie). Captain Marvel from the Shazam show had been included on the list of characters for the identification index; however, this portrayal is ambiguous. This occurs because Captain Marel and his alter ego, Billy Batson, are played by different actors, so it would be unclear what identifica- tion with Captain Marvel might mean to children. Characters from Little House on the Prairie were also on the identifi- cation list; however, it was decided tO balance the program types on the program and character lists by shifting this program to the exposure indexes. Therefore, Captain Marvel from Shazam and Laura Ingalls and Charles Ingalls from Little House on the Prairie were excluded from the identifi- cation indexes and the shows were included in the exposure indexes. Character Identification Variables After the above exclusions, 16 characters remained for the identification index. These included six characters from action-adventure shows (Steve Austin Of the Six Million Dollar Man, Pepper Anderson of Police Woman, Hondo of SWAT 41 Dixie McCall of Emergency, Steve McGarrett Of Hawaii 5-0, and Jaime Sommers of the Bionic Woman), seven from situa- tion comedies (Ed Brown from Chico and the Man, LaVern DeFazio of LaVern and Shirley, Margaret Hoolahan Of MASH, George Jefferson of the Jeffersons, Gabe Kotter of Welcome Back Kotter, and Mike Stivak of All in the Family), two from family dramas (John—Boy Walton and Olivia Walton Of the Waltons) and one from Saturday morning shows (Isis of the Shazam/Isis Hour). The respondents indicated their degree of identifi- cation with those characters by circling the name Of those they "want to be like," drawing a line through the names of those they "do not want to be like," and disregarding the names of those toward whom they are neutral or "don't care." These responses were coded l, -l, and 0, respectively. Identification frequencies and descriptive statistics for these responses are shown in Table 2.4. It should be noted that the measure apparently tapped disidentification more than identification. While all the characters received at least some endorsements, only three more endorsements than rejections--Steve Austin, Gabe Kotter, and Jaime Sommers. Thus, it cannot be said that these tele- vision characters were, in general, accepted as behavior models by the respondents. 42 Table254. Levels of Identification with Selected Television Characters Percent whO* . . disiden- are Character tify neutral identify mean sd.** Steve Austin 36 13 51 0.14 0.92 Pepper Anderson 54 20 26 -0.27 0.85 Ed Brown 64 l9 16 -0.49 0.75 Lavern DeFaziO 68 16 16 —0.52 0.76 Hondo 45 23 32 -0.13 0.87 Margaret Hoolahan 70 19 11 -0.59 0.68 Isis 57 16 26 -0.31 0.86 George Jefferson 73 16 10 -0.63 0.62 Gabe Kotter 33 16 51 0.17 0.90 Dixie McCall 61 19 19 -0.42 0.79 Steve McGarrett 57 20 23 -0.34 0.83 Mary Richards 60 16 24 -0.37 0.84 Jaime Sommers 41 14 45 0.04 0.93 Mike Stivak 70 18 11 -0.60 0.68 Olivia Walton 67 18 14 -0.53 0.73 John Boy Walton 67 16 16 -0.51 0.76 * Percentages may not add to 100 because of rounding error. ** Means and standard deviations were computed by scoring identify "1," neutral "0," and disidentify "-l." 43 Content Analysis Data The television behavior profiles for this research are based on the Project CASTLE content analysis of pro- and antisocial behavior. Greenberg, §E_§l., (1977) de- scribe the procedures and results Of this analysis in detail. Therefore, the procedures are only outlined here and only results germaine to this research are reported. Specifically, interest here lies with the frequencies of social behaviors portrayed on the selected shows listed in Table 2.3 and by the selected characters listed in Table 2.4. One episode Of each Of the selected programs was video taped during the fall of 1975 for subsequent analysis by trained undergraduate coders. Essentially the analysis procedure consisted of counting instances of specific be- haviors that occurred during each episode.* The procedures required coders to categorize occurrences of social behavior and to record the identify of the character who performed them. This allowed for development of behavior profiles both for entire programs and for specific characters. Frequencies and rates of the behaviors selected for this research from the overall CASTLE content analysis are * The coders also evaluated the motives, consequences, and intensities of each act; however, these data are not used in this research. 44 reported in Table 2.5. All the selected behaviors occur frequently and, together, they comprise the majority Of these social behaviors on television. The content analysis procedures required coders to understand the conceptualizations of the behaviors and to agree on criteria for distinguishing among them. To achieve these goals, coders were provided with extensive training. Coders were asked to study training manuals which defined each of the variables for one week prior to the actual training sessions. During the training sessions, coders were shown videotaped instances of the specific behaviors which were discussed with the researchers until it was clear that coders understood the conceptualizations. Practice and discussion continued until the coders reached acceptable levels of reliability. During the actual coding, nearly 40 percent of all programs were viewed by at least two coders and agreement between them was monitored. The percentage Of agreement between coders ranged from 76 to 100 percent. Coders were able to categorize and distinguish among the behavior types with a high degree of reliability. For this analysis, the frequencies of each of the be- haviors were summed across each of the selected programs and the selected characters. This procedure provided the be- havior profiles for programs and characters shown in Tables 2.6 and 2.7. 45 Table 2.5. Frequencies and Rates of Selected Social Behaviors for a Composite Week Of Television Drama* Frequency Percent Rate per in Week of Class Hour Antisocial Behaviors Verbal Aggression 1,629 62 23.78 Physical Aggression** 828 32 12.08 Total Coded*** 2,620 94 Prosocial Behaviors Altruism 915 27 13.50 Affection 528 16 7.70 Expression 921 27 13.10 Total Coded**** 3,379 70 * Adapted from Greenberg et al., 1977. From a sample of 92 fictional television shows representing 68 1/2 hours for a composite week. ** Includes only behavior Of interest to this research, e.g., hitting, shoving, shooting, stabbing. *** Includes behavior not of interest to this research, e.g., bombing, arson, rape. **** Includes behavior not Of interest to this research, e.g., reparation, delay of gratification, control of others' bad behavior. 46 m m m o AH Seaman mH mm mm m an gonna use sxmumum m oH H m a now one euomemm mm Hm s m NH meonom was a HH s o m mmHHn euomxoom m 4H NH HH HH moose mH H o m m seaweed erm m NH H «H mH eHuHeum was so mmsom mHuuHH OH a N H a games em a s H OH second secs ago: mH mm m s o mama seams m s m mH m mmsHa eooo m mH s o m unmnHa umm mm Hm H h m hccsm mmsm H m OH mH H Humnsez hem :OHmmmHmmd conmmHmm< coflmmoumxm cofluommmm Emwsuuafl Emnmoum Hmonsem Hmnum> mpOmHmm Hoe How>mnmm mo zoomskum mEmHmonm confl>mHmB pmuowamm Mom mmemOHnH How>mnmm .m.m OHQMB 47 OOVO$OOOONNNONNV H H NVCDOWOOHHHNNNI‘HO H v m w H v N H m o o m N HH m m o H N o H H o o v o O o o H H H o H o m o o m H o N o v H o o o meEEom wEHwb OHNmmmo cum>mq Hmuuox mono :Humo< O>Oum sownmmmmo mmuoow conumpcd Hmmmmm :3onm om couHMB wom anon couHmz MH>HHO HHmuoz OHxHO mwanOHm hum: oncom mHmH pumunmwoz O>mum cmanoom uwummnmz xm>Hum mtz cOHmmmHmmfl conmmHmm< HmOHmmnm Hmnum> conmmmem coHuommmfi EmHouuH< mbOmHmm Hoe Homwmsmm mo mocmovmum Hmuomnmso mumuomumzu :onH>OHmB pmuomHmm How mmHHmoum uoH>mnmm .s.NOHan 48 Combined Indexes The predictor variables for this research are indexes constructed by combining information from the television be- havior profiles with questionnaire responses concerning ex- posure to television and identification with television characters. This section describes methods of construction Of these indexes and provides statistics concerning them and their interrelations. Exposure-Behavior Indexes The five exposure-behavior indexes were constructed by multiplying the frequency of each specific behavior for each show by each child's exposure to that show.* These products were summed across the selected programs to form the indexes. Thus, a child who watches several shows with a high frequency Of physical aggression, for example, received a high physical aggression exposure score. Another child who watched fewer shows or who watched shows containing less physical aggression would receive a lower physical aggression exposure score. Because these indexes reflect differential weightings of the same exposure behavior, they are highly intercorrelated. Ramifications of these intercorrelations are discussed in the data analysis section of the next chapter. Descriptive statistics and intercorrelations for * The CASTLE content analysis data included coding of intensity Of each act; however, these data were not used be- cause the coding scheme was not comparable across behavior types. 49 these indexes are shown in Table 2.8. Table 2.8. Descriptive Statistics and Correlations for Exposure-Behavior Variables Variable mean sd 1 2 3 4 5 l. Altruism 145.91 53.63 ----- 2. Affection 110.92 40.79 .69 ----- 3. Expression 109.71 48.27 .88 .57 ----- 4. Verbal Aggression 260.26 89.45 .86 .67 .74 ----- 5. Physical Aggression 280.99 114.69 .81 .56 .62 .91 ----- Identification-Behavior Indexes Similar procedures were used to combine information from the televised behavior profiles for characters with in- formation of the children's identification with those char- acters. The altruism identification index, for example, was formed by multiplying the frequency of altruistic behavior for each character by each child's level of identification with that character and summing across characters. It should be recalled that rejection of a character as a behavior model was coded as minus one; therefore, these indexes may take on negative values. In fact, the means of all five identification indexes are less than zero. Like the exposure indexes, the identification indexes are highly intercorrelated. Table 9 provides descriptive statistics and correlations for these indexes. 50 Table 2.9. Descriptive Statistics and Correlations for Identification-Behavior Variables Variable mean sd. 1 2 3 4 5 1. Altruism -6.26 16.05 ----- 2. Affection -9.81 10.33 .71 ----- 3. Expression -9.74 11.23 .78 .70 ----- 4. Verbal Aggression -24.31 28.84 .80 .67 .73 ----- 5. Physical Aggression -5.41 10.62 .86 .63 .71 .88 ----- The use of characters for the identification indexes appearing in the programs used for the exposure indexes apparently minimized the correlations between the two types Of variables as shown in Table 2.10. Relationships Amohg Indexes and Control Variables Although no formal hypotheses were offered, Chapter I noted that grade and sex were expected to be related tO both the predictor variables and criterion variables in this re- search. Therefore, it was necessary to test for such rela- tionships to determine if sex and grade should be used as control variables. To examine these relationships, F statistics were computed for each of the indexes by each of the control vari- ables. This section reports the results of the analysis. 51 NN. ON. mo. mo. pH. conmmHomd HMOHmmnm Hm. mm. OO. NH. OH. eonmmumma Hmnue> NH. NH. mo. 0H. NH. conmwumxm OO. HH. OH. OH. OH. eoHuommma OH. sH. OH. OH. OH. EmHsHuHa conmmHmm4 conmOHmm< conmmHOxm :oHuommm¢ EmHsHqu mmHanHm> chamomxm Hmonsnm Hmnuw> mmHQMHHm> COHHMOHMHuCObH mmHanHmS OOHOMOHMHHEOOH pom musmoaxm mucosa chHumHmHHOU .OH.Nanm.H. 52 Relationships with Sex Sex was expected to be a major determinant of social behavior patterns, exposure to television, and identifica- tion with television characters. Girls were generally ex- pected to have higher scores on the prosocial indexes, and boys, higher values on the antisocial indexes. Criterion Variables by Sex. F statistics for these variables by sex are shown in Table 2.15. All differences are significant beyond the .001 probability level and the pattern is completely consistent with expectation. Girls report more prosocial behavior, and boys, more antisocial behavior. Table 2.11. F Statistics for Criterion Variables by Sex Boys Girls Variable N=345 N=376 F(1, 719) Altruism x 25.64 28.24 53.13*** sd. 5.01 4.56 Affection X 11.37 12.88 42.09*** sd. 3.13 3.14 Expression R 14.30 15.97 40.49*** sd. 4.20 3.53 Verbal r 1906 17.60 23.66*** Aggression sd. 4.20 3.85 physical x 21.07 16.26 l74.22*** Aggression sd. 4.99 4.76 *** p < .001 Exposure Variables by Sex. 53 relationships are shown in Table 2.12. significant at the F statistics for these All differences are .05 probability level or beyond; however, in four Of the five cases, boys'mean exposure is higher than girls. The one index on which girls have a higher mean than boys--affection--is, as expected a prosocial behavior. Also as expected, boys are markedly higher in their exposure to antisocial behavior than girls are. Table 2.12. F Statistics for Exposure Variables by_Sex Boys Girls Variable N=345 N=376 F(1, 719) Altruism R 152.62 139.76 10.38** sd. 55.20 51.93 Affection R 115.20 124.25 8.95** sd. 40.74 40.40 Expression _ 113.75 106.11 4.51* sd. 48.84 47.51 Verbal — 278.75 243.30 29.38*** Aggression sd. 90.46 85.45 Physical — 310.52 253.89 46.62*** Aggression sd. 116.50 106.15 * p < .05 ** p < .01 *** p < .001 Identification Variables by Sex. these relationships are shown in Table 2.13. F statistics for It should be noted in reading the table that all means are less than zero. Therefore, higher absolute values indicate lower identifica- tion. A11 mean differences are significant beyond the .05 54 probability level, and the pattern is consistent with expec- tation. Girls identify more with prosocial characters, and boys, more with antisocial characters. Table 2.13. F Statistics for Identification Variables by Sex Boys Girls Variable N=345 N=376 F(1, 719) Altruism r -7.83 -4.82 6.40* sd. 15.84 16.13 Affection E -11.70 -8.00 22.91*** sd. 11.26 9.04 Expression E -13.24 -6.54 70.14*** sd. 10.25 11.14 Verbal E -17.71 -30.36 36.34*** Aggression sd. 29.32 27.05 Physical i —4.20 -6.52 8.66** Aggression sd. 11.50 9.60 'k p < .05 *** p < .001 Relationships with Grade While it was difficult to determine precisely what the relationships between grade and the other variables in this research would be like, there is ample evidence to suggest that such relationships exist. Indeed, the analysis shown below reveals several complex patterns. Criterion Variables bngrade. F statistics, including Scheffe tests for post hoc comparisons, for these relation- ships are shown in Table 2.14. In general, antisocial be- havior appears to increase with age. Fourth graders report significantly less verbal and physical aggression than sixth 55 Table 2.14. F Statistics for Criterion Variables by Grade 4th 6th 8th a Variable N=227 N=268 N=226 F(2, 718) Scheffe Altruism X 26.81 27.72 26.30 5.30** (8 4)(8 6) sd. 5.55 4.57 4.64 Affection — 13.10 12.26 11.09 23.63*** sd. 3.24 3.09 3.04 Expression X 15.27 15.22 15.02 .33 (4 6 8) sd. 3.48 3.72 3.62 Verbal E 16.88 18.60 19.38 23.81*** (6 8) Aggression sd. 3.83 4.05 3.98 Physical ’ 17.45 18.78 19.42 7.94*** (6 8) Aggression sd. 5.45 5.21 5.50 aNumbers in parentheses indicate grades not significantly different at the .05 level. * t p < .01 *** p < .001 and eighth graders. While sixth and eighth graders are not significantly different in their reports of antisocial be- havior, eighth graders' means are slightly higher for both variables. On the other hand, reports Of affection decline with age and each successive class reports significantly less of the behavior. For altruism and expression the relation- ships are curvilinear with sixth graders reporting the high- est levels. for altruism. However, the differences are significant only Exposure Variables by Grade. F statistics, including Scheffe tests for post hoc comparisons, for these relation- ships are shown in Table 2.15. Eighth graders are exposed significantly less to antisocial behavior than are fourth 56 Table 2.15. F Statistics for Exposure Variables by Grade 4th 6th 8th a Variables N=227 N=268 N=226 F(2, 718) Scheffe Altruism x 142.07 152.78 141.61 3.48* (4 6 8) sd. 52.49 58.32 48.90 Affection — 119.20 126.19 113.21 6.35** (4 8)(4 6) sd. 40.81 41.56 38.85 Expression _ 99.07 112.93 116.73 8.69** sd. 46.15 51.41 44.76 Verbal _ 270.26 272.32 235.92 12.6l*** (4 6) Aggression sd. 83.37 93.95 82.28 Physical ’ 300.39 298.75 240.43 21.78*** (4 6) Aggression sd. 106.24 121.51 104.06 aNumbers in parentheses indicate grades not significantly different at the * p < .05 t p < .01 * p < .001 * ** .05 level. and sixth graders whose means are only trivially different. Exposure to self-expression increases with age and each successive grade has a higher mean than the one preceding it. Although the overall F ratio is significant for altruism, the highly conservative Scheffe test shows no significant differences between any Of the grades. For affection, sixth graders are significantly higher than eighth graders but not significantly higher than fourth graders. Fourth and eighth graders are not significantly different for affection. Some Of these findings can be attributed to changes in program preferences with age. Younger children tend to watch relatively simple programs, particularly cartoons, 57 which stress action over verbal interchange. Therefore, they are less likely to be exposed to self-expression. Cartoons also contain the highest portion Of antisocial be— havior on television, accounting for younger children's higher exposure to physical and verbal aggression. Identification Variables with Grade. F statistics, including Scheffe tests, for these relationships are shown in Table 2.16. In general, identification declines with age for all behavior types. Fourth graders have signifi- cantly higher means for altruism identification than sixth and eighth graders whose means are not significantly dif- ferent. For affection, verbal aggression, and physical aggression identification, fourth graders have significantly higher means than the other two groups, while contiguous grades are not significantly different. NO significant dif- ferences are found for expression although the means tend to decrease while grade increases. Summary This chapter described methods Of data collection and index construction for Operationalization of the variables defined in Chapter I. Indexes Of behavior, television ex- posure, and identification with television characters were constructed for each Of the following behavior types: altruism, affection, self-expression, verbal aggression, and physical aggression. The exposure and identification 58 Table 2.16. F Statistics for Identification Variables by Grade Variable 4th 6th 8th F(2, 718) Scheffea Altruism r -2.75 -6.31 -9.73 10.98*** (6 8) sd. 15.63 16.58 15.10 Affection I -8.16 -10.16 -11.72 4.72* (4 6)(6 8) sd. 10.79 10.14 9.88 Expression r -9.11 -9.55 -10.56 0.98 (4 6 8) sd. 11.09 11.52 11.01 Verbal x -2l.08 -24.13 -27.76 3.06* (4 6)(6 8) Aggression sd. 28.66 29.81 27.55 Physical r -3.90 -5.35 -7.01 4.93** (4 6)(6 8) Aggression sd. 10.22 10.84 10.50 aNumbers in parentheses indicate grades not significantly different at the .05 level. * p < .05 * p < .01 p < .001 * *** indexes reflect both children's questionnaire responses and a content analysis of televised behavior. Relationships between the above indexes and demograph- ic variables, grade and sex, were examined. This analysis demonstrated the necessity of controlling the effects of grade and sex to isolate the effects of television on chil- dren's social behavior. CHAPTER III ANALYSIS AND RESULTS This chapter begins with a section describing analy- sis procedures. Results are then presented in five sections that conform to the order in which the hypotheses were pre- sented in Chapter I. These sections are on (1) main effects of television exposure, (2) crossed effects of television exposure, (3) main effects of identification with television characters, (4) crossed effects Of identification with tele- vision characters, and (5) interaction effects. Findings are simply presented here, and discussion is reserved for Chapter IV. Analysis Procedures Before considering specific hypotheses, it is neces- sary to recall the interrelationships within sets of predic- tor variables. Correlations within the set Of exposure indexes and within the set of identification indexes are quite high--ranging from .56 to .88. Procedures to minimize the correlations between the identification and exposure in- dexes were successful with the values falling in the teens and 203. 59 60 The high collinearity among predictors makes it like- ly that if one predictor within a set correlates with a criterion, other predictors in the same set will correlate in the same way. This statistical constraint may account for some of the anomolous findings reported below. The high collinearity precludes the use Of multiple regression statistics to test interaction hypotheses. There- fore, contingent correlation analysis was used to test these prOpositions. Under these procedures, the sample is divided into groups on the basis of respondents' scores on one predictor variable. Then correlations between another predictor vari- able and a criterion variable are computed and compared. Several problems arise with this analysis procedure the first of which is the question Of where to divide the sample. For the exposure indexes, it was decided to make an arbitrary division at the median. Where analysis of the median splits was encouraging, correlations within quartiles were computed to further examine the relationships. For the identification variables, a "natural" division point existed at zero with values above that point indicating identifica- tion and those below that point indicating disidentification. Again, where a dichotomy at zero produced encouraging results, relationships were further examined within quartiles. It should be recognized that different divisions of the sample might alter relationships somewhat and thus lead to different 61 conclusions. This in fact occurs with reference to the median and quartile splits for identification. A second problem with contingent correlation analysis is that subgroups are obviously smaller than the total sample; therefore higher correlations are needed in sub- sample to achieve statistical significance. This problem is not severe in these data because even division into quartiles results in subgroups of well over 100 cases. The most important problem in contingent correlation analysis is that when the predictor on which the divisions are made is correlated with the predictor used to calculate correlations, the variance of the correlation predictor is truncated. This suppresses the relationship between the correlation predictor and the criterion variable. Further, to the degree that the predictors are correlated the mean values of the correlation predictor will be systematically different between the subgroups. That is, the correlations in the high subgroups will tend to be based on moderate to high values of the correlation predictor, while in the low subgroup the correlations will tend to be based on low to moderate values Of the correlation predictor. Obviously, this problem becomes more severe when the sample is divided into larger numbers of smaller groups. Therefore, it af- fects the contingent analysis within quartiles more than it does median splits. 62 Given the above problems, it is clear that the tests of the interaction hypotheses must be approached with some circumspection. However, given the data available, it appears that contingent correlation analysis offers the best method of testing the interaction proposition. The last section of Chapter II demonstrates the ne- cessity of controlling the effects of grade and sex to iso- late the effect of television on social behavior. Control- ling for sex is relatively straightforward. Correlations Of interest are partialed for sex which is coded as a dummy variable. The curvilinear relationships by grade preclude par- tialing the variable as a single measure. However, this problem can be solved by treating grade as two two-level dummy variables. That is, one variable represents member- ship in the fourth grade and a second represents membership in the six grade.* A second-order partial on these two dummy variables completely controls the effects of grade on the predictor and criterion variables. Thus, the basis statistic for tests Of the hypotheses are third-order partial correlations--the relationships be- tween predictor and criterion variables controlling for sex and two dummy variables for grade. * All the information in a three-level variable is contained in two dichotomies because a third dichotomy would be completely determined by the other two (cf. Cohen and Cohen, 1975). 63 The main effects Hypotheses (l, 2, 5, and 6) and the crossed Hypotheses (3, 4, 7, and 8) are tested by examining the sign and significance of the appropriate correlations and partial correlations for the entire sample. The inter- action Hypotheses (9 through 14) are tested by comparison of the appropriate correlations and partial correlations computed for specific subgroups as described above. Main Effects Of Exposure This section deals with the proposition that exposure to a specific class of behavior will be positively associated with performance of that behavior (Hypotheses l and 2). Prosocial Exposure The correlations between prosocial exposure and pro- social behavior indexes are shown in Table 3.1. All three sets Of indexes--altruism, affection, and expression--are included. While the correlations are generally low, they are positive and, given the sample size, statistically sig- nificant with the exception of the zero-order correlation for expression. All partial correlations are significant although partialing does not dramatically affect the rela- tionships. Thus, Hypothesis 1 is supported. Antisocial Exposure The correlations for exposure and antisocial behavior are shown in Table 3.2. These include verbal aggression and 64 Table 3.1. Correlations Between Prosocial Exposure and Prosocial Behavior O-Order Partialed for . . . Behavior Correlation Sex Grade Grade and Sex Altruism .12*** .16*** .ll*** .15*** Affection .14*** .11*** .12*** .10** Expression .06 .08* .O7* .08* N=721 * p < . 05 *1: p < .01 *** p < .001 Table 3.2. Correlations Between Antisocial Exposure and Antisocial Behavior 0-Order Partialed for . . . Behavior Correlation Sex Grade Grade and Sex Verbal Aggression .10** .07* .l4*** .ll** Physical Aggression .16*** .06 .20*** .09** N=721 * p <<.05 ** p ‘<.01 *** p '<.001 physical aggression. positive and significant. Again, the correlations are generally However, partialing for sex de— creases the correlations and the level for physical aggress- ion falls below that needed for statistical significance. 65 Partialing for sex reduced the correlation because girls are lower in both aggression exposure and in aggressive behavior. Partialing for grade increases the correlation because young- er children are less aggressive but more exposed to aggres- sive television. The correlations for antisocial exposure and antisocial behavior remain positive and significant after partialing; therefore, Hypothesis 2 is supported. Crossed Effects of Exposure Hypothesis 3 states that exposure to prosocial tele- vision will be negatively associated with performance of antisocial behavior and Hypothesis 4, that exposure to anti- social television will be negatively associated with perform- ance of prosocial behavior. This section deals with tests of these propositions. Prosocial Exposure The correlations between the three prosocial exposure indexes and both of the antisocial behavior indexes are shown in Table 3.3. Contrary to the hypothesis, these cor- relations are generally positive and significant. Only the correlations between affection exposure and physical and verbal aggression are not significant, but these are far from being significantly negative. Thus, Hypothesis 3 is not supported. 66 Table 3.3. Correlations Between Prosocial Exposure and Antisocial Behavior O-Order Partialed for . . . Pair Correlation Sex Grade Grade and Sex Altruism with ... Verbal Aggression .ll** .09** .10** .09** Physical Aggression .15*** .ll** .15*** .ll** Affection with ... Verbal Aggression .01 .03 .02 .04 Physical Aggression -.02 .03 -.02 .03 Expression with ... Verbal Aggression .12*** .08** .09** .08* Physical Aggression .15*** .ll** .13*** .ll** N=721 .05 p < .01 p < .001 67 Antisocial EXposure The correlations between both antisocial exposure in— dexes and the three prosocial behavior indexes are shown in Table 3.4. These correlations are positive and significant with two exceptions; the correlations between physical ag- ression exposure and expression behavior are not significant for the zero—order relationship and when partialed for sex. Partialing for sex generally increases the correlations, and partialing for grade decreases them. Nearly all the corre- lations in the table are significantly positive regardless of partialing. Thus, Hypothesis 4 is not accepted. Table 3.4. Correlations Between Antisocial Exposure and Prosocial Behavior O-Order Partialed for . . . Pair Correlation Sex Grade Grade and Sex Verbal Aggression with ... Altruism .ll** .17*** .09** .16*** Affection .l4*** .19*** .10** .16*** Expression .09** .14*** .08** .l4*** Physical Aggression with ... Altruism .10** .17*** .08* .16*** Affection .13*** .21*** .09** .16*** Expression .04 .10** .04 .10** N=721 * p '<.05 ** p < .01 *** p ‘<.001 68 Main Effects for Identification This section deals with the proposition that identi- fication with characters who perform a specific class Of be- havior will be positively associated with performance of that behavior (Hypotheses 5 and 6). Prosocial Identification The correlations between prosocial identification in- dexes and prosocial behavior indexes are shown in Table 3.5. Table 3.5. Correlations Between Prosocial Identification and Prosocial Behavior O-Order Partialed for . . . Behavior Correlation Sex Grade Sex and Grade Altruism .08* .06 .07* .05 Affection .ll** .07* .09** .05 Expression .15*** .08* .15*** .08* N=721 * p ‘<.05 ** p ‘<.01 *** p ‘<.001 While the zero—order correlations are positive and signifi- cant, only the expression correlation remains significant after partialing for grade and sex. Partialing for sex alone reduces the altruism identification-behavior correla- tion to a non-significant level. While grade alone has little effect as a control variable, partialing sex in 69 combination with grade lowers the affection identification— behavior correlation to a nonsignificant level. Thus, Hypotheses 5 is supported marginally, and only with reference to expression. Antisocial Identification The correlations between the antisocial identification indexes and the antisocial behavior indexes are shown in Table 3.6. Only the correlations for physical aggression ident- ification and physical aggression behavior are consistently positive, and this relationship is not significant when partialed for sex alone. Partialing for grade increases the correlations slightly and partialing for sex decreases them. Thus, the joint partial Of sex and grade leaves the correla- tion for physical aggression at a significant level. Hypoth- esis 6 is given marginal support with reference to physical aggression only. Table 3.6. Correlations Between Antisocial Identification and Antisocial Behavior O-Order Partialed for . . . Behavior Correlation Sex Grade Grade and Sex Verbal Aggression .026 -.01 .05 .01 Physical Aggression .10** .05 .12*** .07* N=721 * p <<.05 ** p '<.01 ** * p ‘<.001 70 Crossed Effects Of Identification Hypothesis 7 states that identification with models who perform prosocial behavior will be negatively associated with performance Of antisocial behavior and Hypothesis 8, that identification with models who perform prosocial behavior will be negatively assoicated with performance of prosocial be- havior. This section deals with these hypotheses. Prosocial Identification Correlations between the three prosocial identifica- tion indexes and both antisocial behavior indexes are shown in Table 37% Partialing for grade and for sex individually makes all correlations slightly more positive. Grade and sex are only trivially correlated so their effects on the rela- tionship Operate largely independently. In fact, when par- tials are computed for both control variables, the correla- tions between altruism and affection identification with physical aggression are significant and positive, which con— tradicts the hypothesis. The zero-order correlations and partial correlations controlling for grade between expres- sion identification and physical and verbal aggression be- havior are negative and significant as predicted. However, these correlations disappear when partialing for sex or for grade and sex. Given the overall pattern, Hypothesis 7 is not accepted. 71 Table 3.7. Correlations Between Prosocial Identification and Antisocial Behavior 0-Order Partialed for . . . Pair Correlation Sex Grade Sex and Grade Altruism with ... Verbal Aggression .00 .04 .05 .06 Physical Aggression .04 .06 .06 .ll** Affection with ... Verbal Aggression -.02 .01 .01 .05 Physical Aggression .00 .02 .02 .ll** Expression with ... Verbal Aggression -.08* -.O7* -.07* -.02 Physical Aggression -.13*** .00 -.13** .00 N=721 .05 .01 .001 ** *** A A A 72 Antisocial Identification The correlations between the antisocial identification indexes and the prosocial behavior indexes are shown in Table 3.8. Contrary to the hypothesis, all these correlations are positive. The correlations Of both aggression identification indexes with altruism and expression behavior are positive and significant when partialed for sex and for sex and grade. Thus, Hypothesis 8 is not accepted. Table 3.8. Correlations Between Antisocial Identification and Prosocial Behavior 0-Order Partialed for . . . Pair Correlation Sex Grade Grade and Sex Verbal Aggression with ... Altruism .02 .08* .01 .08* Affection .02 .08 .00 .05 Expression .06 .09** .05 .09** Physical Aggression with ... Altruism .06 .09** .05 .09** Affection .03 .06 .01 .03 Expression .04 .07* .04 .07* N=721 * p ‘<.05 ** p ‘<.01 ** * p ‘<.001 73 Interaction Effects The analyses for interaction effects are discussed in the order in which they were presented as hypotheses in Chapter I. That is: (1) prosocial exposure mitigating the effects of antisocial exposure on antisocial behavior, (2) antisocial exposure mitigating the effects Of prosocial ex- posure on prosocial behavior, (3) prosocial identification mitigating the effects of antisocial identification (4) antisocial identification mitigating the effects of pro- social identification, (5) prosocial identification enhanc- ing the effects Of prosocial exposure on prosocial behavior, and (6) antisocial identification enhancing the effects of antisocial exposure on antisocial behavior. Prosocial Exposure on Antisocial Exposure Effects Hypothesis 9 states that the relationship between antisocial exposure and antisocial behavior will be suppress- ed by high levels of prosocial exposure. Contingent corre- lations for testing this hypothesis are shown in Table 3.9. The values in the table are the correlations between the antisocial exposure indexes and the antisocial behavior in- dexes within subgroups of respondents who are high or low in prosocial exposure. The high group includes those respond- ents who are above the median in a particular class of pro- social exposure, and the low group includes those below the median. Bracketed pairs of correlations are different at the .05 level Of significance. 74 Table 3.9. Contingent Correlations Between Antisocial Ex- posure and Antisocial Behavior Within High and Low Prosocial Exposure Groups O-Order Partialed for . . . Variable Correlation Sex Grade Grade and Sex Verbal Aggression High Altruism (366)@ Low Altruism (355) Physical Aggression High Altruism (366) Low Altruism (355) Verbal Aggression High Affection (361) Low Affection (360) Physical Aggression High Affection (361) Low Affection @ .00 -.02 .06 [:03 .13* .11* l..19M1': L.lG*** .09* -.03 .13* .00 .14** .06 .19*** .11* .08 .02 .12* .07 .11* .06 .l4** .10* .18** .10 .21*** .05 .18* .07 .21*** .10* Numbers in parentheses indicate N for group. Bracketed pairs of correlations are significantly different at the .05 level. * . p '<.05 * <.01 *** p < .001 75 Table 3.9 (cont'd.) O-Order Partialed for . . . Variable Correlation Sex Grade Grade and Sex Verbal Aggression High Expression -.01 -.05 .05 .01 (354) Low Expression .ll** .09* .l7** .15** (367) ' Physical Aggression High Expression .05 -.06 .ll** -.02 (354) Low Expression .15** .06 .19*** .09* (367) Bracketed pairs of correlations are significantly different at the .05 level. * p < .05 'k p < .01 * p < .001 * *‘k In general, the pattern of correlations is consistent with the hypothesis. Correlations between antisocial expo- sure and antisocial behvior are consistently higher when pro- social exposure is low. The altruism subgroup correlations between verbal aggression exposure and verbal aggression be- havior are significantly different regardless of partialing for grade and sex. The expression subgroup correlations for both physical and verbal aggression are significantly differ- ent only when partialed for both grade and sex. Given the overall pattern of correlations, and that three of six replications reveal significant differences for 76 median splits, it was decided to examine the correlations within prosocial exposure quartiles. These correlations are shown in Table 3.9A. The verbal aggression exposure-behavior correlations within altruism exposure quartiles reveal a clear pattern. Consistent with the hypothesis that prosocial exposure sup- presses the effects Of antisocial exposure, the correlations are consistently lowest in the highest altruism exposure quartile. The highest quartile correlation is significant- ly lower than the lowest quartile when partialing for sex significantly lower than the second quartile when partialing for grade, and significantly lower than the second and third quartiles when partialing for both grade and sex. The physical aggression exposure-behavior correlations within altruism quartiles are significantly different only when partialing for grade. Partialing for grade, the cor- relation in the highest quartile is significantly lower than the correlations in the first and second quartiles. While this is congruent with the hypothesis, the rest of the cor- relations do not generally conform to the expected pattern. The verbal aggression and physical aggression expo- sure-behavior correlations within affection exposure quartiles clearly do not conform to the hypothesis. The highest cor- relations in this set occur in the third quartile and there are no significant differences between quartiles. Tables 3.9 and 3.9A reveal that differing levels of exposure to each of the prosocial behaviors affects the 77 Table 3.9A. Cbntingent Correlations Between Antisocial Exposure and Antisocial Behavior‘Within Prosocial Exposure Quartiles Quartile 0-Order Partialed.fOr . . . variable variable Correlation Sex Grade Sex and Grade verbal .Altruism Aggression lowest .09 (.06) .13* .ll N:l78 lower .09 .04 (.21**) (.16*) N=l76 higher .09 .08 .18** (.l7*) N:188 highest -.09 [-.15] [-.04] [-.09] N=179 Physical .Altruism Aggression lowest .12 .04 (.15*) .08 N=178 lower .06 -.07 (.15*) .00 N=l76 higher .02 -.10 .07 -.05 N=188 highest .11 -.02 {-.15*] .01 N=179 Cbrrelations in parentheses are significantly different fromlcorre- lations in brackets in the same column» * p < .05 ** p < .01 78 Table 3.9A.(cont'd.) Quartile O-Crder Partialed for . . . variable variable Correlation Sex Grade Sex and Grade verbal .Affection Aggression lowest .l4* .09 .15* .09 N=181 lower .07 .02 .l4* .09 N=181 higher .20** .13* .24*** .18* N=l77 highest .00 -.04 .05 .01 N=182 Physical .Affection Aggression lowest .15* .05 .12* .07 N=181 lower .19** .03 .25*** .09 N=181 higher .29*** .12 .31*** .15** Ehd77 highest .13* -.04 .l7* .00 N=182 * NO significant differences appear * p < .05 * p < .01 ** p < .001 * among quartile correlations. 79 Table 3.9A (cont'd.) Quartile O-Order Partialed for . . . variable variable Correlation Sex Grade Sex and Grade verbal Expression Aggression lowest (.12) (.07) (.20**) (.16*) Ne18l lower .05 (.04) .13* (.13*) N=181 higher .00 .00 .ll .10 N=180 highest [-.11] [-.20**][-.03] [-.12] N=l77 Physical Expression Aggression lowest (.l7*) .09 .l9** (.ll) NelBl lower .09 -.02 .14* .01 EhlBl higher {-.08] -.16 .00 {-.09] NHJBO highest .08 -.06 .l4* -.03 N=l77 Correlations in parentheses are significantly different fromicorre- lations in brackets in the same column. * p < .05 ** p < .01 80 antisocial exposure-behavior correlations in parallel ways. TO summzarize this relationship, the three prosocial expo- sure indexes were summed and a median split analysis was performed for overall prosocial exposure. The results of this analysis, shown in Table 3.93 conform to those shown in Tables 3.9 and 3.9A. High levels of prosocial exposure leads to significantly lower verbal aggression exposure- behavior correlations and tend to suppress physical aggres- sion exposure—behavior correlations. Table 3.98. Contingent Correlations Between Antisocial Exposure and IvmiaxflalIkmamkanfliunIfigharfilhvammedlIkoaxfial Empmmue Gongs 0