PROTOCOL: Practices and Program Components for Enhancing Prosocial Behavior in Children and Youth: A Systematic Review

Learning by doing

The role of learning by doing is underscored in many theories of prosocial development. Through experiences of providing help, care, and other prosocial actions, children are expected to learn prosocial skills, receive intrinsic rewards (e.g., feeling good about oneself) and social rewards (e.g., social approval), develop self-efficacy for prosocial conduct, and learn about others' feelings and perspectives, which, in turn, are predicted to motivate children's future prosociability (Eisenberg et al., 2006; Staub, 2003). Intervention studies that examine outcomes of prosocial behavior have examined learning by doing activities such as assigning children the responsibility for others and engaging children in cooperative learning activities (Eisenberg et al., 2006; Johnson, Johnson, Johnson, & Anderson, 1976; Staub, 1975; 1992). The influences of learning by doing activities that involve practice, rehearsal, or role-play of prosocial behavior have also been examined in the literature (Rosenhan & White, 1967; Staub, 1971a; White, 1972).

Social learning

According to social learning theory, social behavior is learned and shaped primarily through processes of observational learning (e.g., provision of positive models of behavior), positive behavioral reinforcement, behavioral expectations, and direction instruction. In these ways, socializers are theorized to provide important information regarding behavioral expectations and serve as resources and guides for learning and adopting behavior (Bandura, 1986). Studies have examined the impact on prosocial behaviors of social learning influences including provision of prosocial adult models (Staub, 1971b; Grusec & Skubiski, 1970; Rushton, 1975), praise, punishment, external rewards (Bénabou & Tirole, 2006; Bryan & Brickman, 1973; Gelfand, Hartmann, Cromer, Smith, & Page, 1975; Fabes, Fultz, Eisenberg, Plumlee, & Christopher, 1989; Ramaswamy & Bergin, 2009; Szynal-Brown & Morgan, 1983), and instructional prompts (Gelfand et al., 1975; Israel & Brown, 1979). Prosocial models in media (e.g., prosocial stories or television of programs) have also been studied (Mares & Woodard, 2005).

Emotional literacy

Emotional literacy refers to the capacity to accurately recognize and label emotions in oneself and others as well as in stories, music, etc., to understand the causes and consequences of emotions, and to express and regulate emotions in socially adaptive ways (Brackett & Rivers, 2014). According to many theoretical accounts, emotional literacy underlies prosocial behavior and other social and emotional competencies (Lemerise & Arsenio, 2000; Mayer & Salovey, 1997). With reference to prosocial behavior, if a peer is in distress, the potential enactor of help needs to be able to make sense of the peer's emotional cues, self-regulate emotional responses (e.g., distress might block helping responses), and show appropriate emotions when helping the child (e.g., expression of happiness would be inappropriate). Activities like socializers drawing attention to the feelings of a peer or storybook character, showing and labeling pictures of faces with different emotional expressions, and reading books about different feelings are examples of emotional literacy training (Joseph & Strain, 2003). Some interventions targeted at enhancement of prosocial behavior include teaching emotional literacy. For example, an evaluation by Ornaghi, Grazzani, Cherubin, Conte, and Piralli (2015) demonstrated that a conversational intervention concentrated on the nature, causes, and regulation of emotion had positive effects on the prosocial orientation of children in the treatment group versus control at post-test and the four-month follow-up.

Other-orientation

Perspective-taking, empathy, and sympathy are frequently posited sources of other-oriented motivation for prosocial behavior. These other-oriented mechanisms are theorized to enable understanding of others' thoughts, feelings, desires, motivations, and intentions, which, in turn, may lead to prosocial action if an individual is in distress or need of help (or help would result in alleviating the distress or guilt of the potential helper) (Batson, 1991; Eisenberg et al., 2006; Feshbach, 1978; Hoffman, 2000; Staub, 1979). An array of adults' techniques aimed at increasing children's consideration for others have been tested for influence upon prosocial behavior in, for example, experiments on empathic arousal (Howard & Barnett, 1981; Eisenberg-Berg & Geisheker, 1979; Ladd, Lange, & Stremmel, 1983), other-oriented inductive techniques (i.e., discipline techniques that include pointing out the consequences of children's actions on others) (Ramaswamy & Bergin, 2009; Staub, 1971b), and other-oriented preaching (Eisenberg, 1983; Eisenberg-Berg & Geisheker, 1979; Midlarsky & Bryan, 1972). It should be clarified that theoretically and empirically other-orientation and emotional literacy are related but also distinct and separable constructs. In essence, skills of emotional literacy (e.g., decoding, understanding, and regulating emotion) lay the foundation for perspective-taking, empathy, and sympathy.

Moral instruction

Although the literature on moral instruction has been confounded by definitional debate, moral reasoning is commonly referred to as evaluations adhering to principles of fairness, justice, and/or concern for the welfare of others or society (Turiel, 1983). The underlying theoretical expectation is that the development of moral reasoning contributes significantly to prosocial behavior by having an impact on social sensitivity and responsibility, internalization of moral obligations, and guilt for wrongdoing (Eisenberg, 2000; Krebs & Van Hesteren, 1994). Interventions designed to move children to higher levels of moral development predominantly center on group discussions of real-life or hypothetical moral dilemmas or on moral exhortations (e.g., socializer remarks on the merits or virtues of prosocial behavior such as “It is good to help others”). Although the potential impact of moral instruction on prosocial behavior is largely theoretical with comparatively little empirical work, a few evaluations of interventions have investigated this impact (e.g., Krivel-Zacks, 1995; Grusec, Saas-Kortsaak, & Simutis, 1978).

Mindfulness

There, as yet, is no consensus on a definition of mindfulness (Chiesa, 2013). Mindfulness is characterized herein by an ability to concentrate on and attend to sensations, thoughts, feelings, and objects; describe and label feelings and thoughts with words; bear nonreactivity to inner experiences; and approach moment-to-moment experiences with awareness, openness, acceptance, and non-judgment (Baer, 2007; Baer, Smith, Hopkins, Krietemeyer, & Toney, 2006). Processes by which mindfulness training is thought to influence prosocial behavior include self-regulation of emotions and behavior and transcendence of the self to focus on the broader context (e.g., other people) (Roeser & Pinela, 2014). Mindfulness interventions include components of meditation, relaxation, and concentration practices; acceptance-based strategies; yoga and other movement practices; sense, breath, and body awareness; optimism training; and cultivation of present awareness of feelings, thoughts and sensations. Control group evaluations that include mindfulness training are relatively recent; however, a few have investigated effects on prosocial behavior in children (e.g., Flook, Goldberg, Pinger, & Davidson, 2015). It should be noted that mindfulness interventions often embed features of emotional literacy training (e.g., emotional self-regulation), but there is a focus on self-referential emotion or the inner experience. Such intervention features will be captured under the coding category of emotional literacy.

Cognitive/behavioral self-regulation

Increasingly, cognitive and behavioral self-regulatory capacities are viewed as core processes relevant to the learning and initiation of prosocial behavior (e.g., Hay & Cook, 2010). Although definitions vary, cognitive and behavioral self-regulation broadly refer to skills to inhibit impulsivity, maintain focus and attention, retain information, delay gratification, generate adaptive cognitive appraisals, and control behavior in service of externally imposed or personal goals or social acceptance (Baumeister, Schmeichel, & Vohs, 2007; Blair, 2009). A variety of interventions will be coded under this category, including those with a focus on improving goal-directed behavior; modifying dysfunctional thinking; inhibiting inappropriate behaviors; and regulating behavior in accordance with contextual rules (for coding subcategories, see Appendix A). Interventions that incorporate features to improve self-regulatory skills are fairly common but are seldom evaluated for their impact on prosocial behavior, with some exceptions (e.g., Ooi, 2013).

Bystander intervention training

Children and youth often face situations where they witness bullying of and threats to peers. When they reach out and protect a peer from harm, they are enacting a form of prosocial behavior. Naturalistic research suggests that bystander interventions that support victims are rare (Hawkins, Pepler, & Craig, 2001). However, through increasing positive self-identity, empathy, moral evaluations of bullying, comprehension of the harm of bullying, self-efficacy for defending, and self-determination, bystander interventions are thought to increase helpful reactions in bullying situations (Thornberg et al., 2012; Tsang, Hui, & Law, 2011). Although few control group interventions have been evaluated with respect to their effects on increasing bystander help or defense of bullied victims (Polanin, Esplanage, & Pigott, 2012; Ttofi & Farrington, 2011), there have been a handful (e.g., Fonagy et al., 2009; Karna et al., 2011). Bystander interventions include components such as raising peer support for bullying intervention; increasing children's feeling and empathy for victims; teaching strategies to intervene and support the victim and to cope when victimized; raising awareness about the harmful consequences of bullying and the social responsibility to help victims; and role-playing how to effectively intervene and defend a victim.

Friendship-making skills

Approaches to strengthening peer relationship skills include teaching children how to join in and suggest activities, start and have a conversation, display manners, give and accept compliments, invite peers to play, make and sustain friendships, and engage in prosocial behavior toward peers. According to Youniss (1980), the egalitarian structure of children's peer relations offers a salient context for children to learn prosocial behaviors as compared to hierarchical social relations such as those involving adults, which invoke more compliance-oriented behavior. Interventions aimed at promoting friendship skills occur predominantly in formal social and emotional learning programs. The Early Childhood Friendship Project is an example of one such program that incorporates lessons to help children learn friendship skills and has been investigated for influence on prosocial behavior (e.g., Ostrov et al., 2008).

Caring community

Interventions that strive to create caring communities (or social/group bonding to the school, peers, and/or community) are those that work to foster caring social relations, positive shared values and norms, fairness and respect, sense of connectedness to others and of being valued and mutually supported, and commitment to relationships to others. There are various pathways through which caring communities might increase prosocial behavior. A pervasive explanation advanced by the Social Development Model is that when members of a community feel strongly committed to the community and espouse positive behaviors, norms, and values, they are likely to behave in accordance with those positive behavioral standards (Catalano & Hawkins, 1996). It has also been hypothesized that social groups that produce strong bonds of attachment meet psychological needs for belonging, safety, autonomy, and competence, which, in turn, foster caring attitudes and behavior (Deci, Vallerand, Pelletier, & Ryan 1991). The Child Development Project is an example of a program in which a major intervention component is to build caring communities in classrooms and schools. This program has been evaluated repeatedly for its causal impacts on prosocial behavior (e.g., Solomon, Battistich, Watson, Schaps, & Lewis, 2000; Solomon, Watson, Delucchi, Schaps, & Battistich, 1988).

Diversity Inclusion

Probably the most common theoretical framework that guides research investigating the effects of diversity inclusion is the Common In-Group Identity Model (Gaertner & Dovidio, 2000; Dovidio, Gaertner, & Saguy, 2009). This theory puts forth that experiences that heighten sense of belonging to more expansive groups or emphasize shared characteristics among diverse groups encourage individuals to conceive of themselves as members of a more inclusive community (e.g., school). With diminished intergroup boundaries and the subjective experience of positive group membership, interactions with members of the broader community are expected to become more positive, particularly toward members of other subgroups nested within this larger, shared identity. Enhancing the overlap between representations of the self and other is also a major mechanism thought to increase prosocial behavior (Batson, Sager, Garst, Kang, Rubchinsky, & Dawson, 1997). Interventions with a focus on diversity inclusion include activities that focus on enhancing positive intergroup attitudes and inclusiveness and raising awareness of bias and individual differences and similarities. Diversity inclusion can, at times, be similar to increasing social bonding except that it is targeted at developing sense of belonging by encouraging children to be inclusive of others of different backgrounds (e.g., ethnicity/race) or group membership (e.g., sports team). The Green Circle program is an example of a school-based intervention focused on increasing inclusiveness and respect for diversity, and it has been evaluated for influence on children's sharing with individuals of similar and different race, gender, and body type (Houlette, Gaertner, Johnson, Banker, & Riek, 2004).

Civic and political participation

From volunteering at a homeless shelter to attending a political meeting on local affairs, civic and political participation refer to the many ways in which individuals actively participate in, shape, or improve a community and/or nation (Adler & Googin, 2005; Ekman & Amná, 2009). This participation is multidimensional. For example, Ekman & Amná (2009) differentiate four forms of individual and collective civic and political participation: social involvement (e.g., attention on political or social issues), civic engagement (e.g., volunteering actions in a community), formal political participation (e.g., voting), and activism (e.g., participating in demonstrations). Developmental perspectives suggest that civic and political participation enhance social relatedness and support a shift in values toward greater social concern and prosocial norms, which subsequently foster prosocial proclivities (e.g., Yates & Youniss, 1996). Theories tend to underscore that voluntary civic and political participation increase prosocial tendencies. This presents a problem for testing causality because young people who self-select into civic and political activities may be more prosocially inclined. Therefore, it will be important to synthesize the results of studies (e.g., Caprara et al., 2014; Kahne, Chi, & Middaugh, 2006) that experimentally test whether interventions that increase such participation have an impact on prosocial behavior.

Responsiveness, emotional warmth, and nurturance

Although some theorists propose that emotionally warm and nurturant interactions with adults can increase children's prosocial behavior (e.g., Staub, 1992), there is debate about how these types of interactions influence prosociability. Different causal mechanisms have been proposed such as socializers' emotional warmth and nurturance fostering a secure attachment that provides a basis to care for others (Hastings et al., 2007), providing a secure base to confidently explore interactions that increase opportunities for children to develop social skills (White & Howe, 1998), increasing an openness to the needs of others because children feel as if their own emotional needs are being met (Hoffman, 2000), or decreasing inhibition of enacting prosocial behavior due to fear of disapproval (Staub, 1971b). It has also been argued that emotional warmth and nurturance do not directly influence prosocial behavior but instead enhance children's receptivity to influence by adults' other prosocializing practices (Hoffman, 1970). However, empirical studies provide mixed evidence toward this view, with experiments showing support for adults' nurturance and emotional warmth directly and indirectly (and sometimes not directly but only indirectly through exposure to emotionally warm adults who also use instructional practices such as modeling) increasing children's prosocial behavior (Grusec & Skubiski, 1970; Midlarsky & Bryan, 1967; Staub, 1971a; Yarrow, Scott, & Waxler, 1973).

Development of socializers' social and emotional competence and knowledge of children's social and emotional development

Jennings and Greenberg (2009) propose a model of the prosocial classroom that underscores the significance of cultivating educators' social and emotional competence and wellbeing. Specifically, the researchers outline the importance of providing educators with training in emotional awareness and emotion-related processes, mindfulness to promote reflection and reduce stress, and knowledge on how children develop socially and emotionally. This is a paradigm shift, especially in reference to teacher training, as many social and emotional learning interventions purely provide adults with the curriculum to aid children's development. Parenting interventions are more likely to focus on the social and emotional competence of socializers. The ABCD Parenting Young Adolescent Program is an intervention that is, in part, targeted to support parental wellbeing through teaching parents acceptance-based strategies. The program has been shown to have positive effects on youths' prosocial behavior (Burke, Brennan, & Cann, 2012).

Types of study designs

Studies must use an experimental or quasi-experimental design. Studies without control conditions such as one-group pre-post designs will be excluded. Eligible comparison conditions may be no treatment, treatment as usual, placebo treatment, or any other similar condition set up as a contrast to the treatment condition that should not have an impact (even a negative impact) on prosocial behavior. Studies without random assignment to treatment and control groups must include matched control groups, appropriate adjustments for pretest differences, or report of a pre-treatment variable with respect to an eligible measure of prosocial behavior or at least one social, emotional, cognitive, or academic skill variable. Social, emotional, cognitive, and academic skills are generally correlated with prosocial behavior (see Eisenberg et al., 2006).

Types of participants

The review will include studies with participants ages 3 to 18. This age range was selected to correspond to the ages of students in preschool through secondary school. One primary objective of the review is to provide relevant evidence-based indications of the instructional practices and types of programs that could be instrumental in educating students to be prosocial. The specified age range was also selected to capture the broad developmental stages of early childhood through adolescence, which are marked by high levels of brain plasticity and may provide optimal windows for intervention (Bradshaw, Goldweber, Fishbein, & Greenberg, 2012). In addition, longitudinal research indicates that prosocial acts in childhood and adolescence are predictive of prosocial acts in early adulthood (Eisenberg et al., 1999, 2002), which suggests that it may be of great consequence to focus efforts on enhancing helping behaviors in these developmental periods.

There is one exclusion criterion based on participant demographic background characteristics. Studies are ineligible if more than 15% of participants have identified learning disabilities (e.g., mental disabilities/handicaps/impairments, autism spectrum disorder, cognitive deficits, traumatic brain injury, or enrollment in special education services) unless prosocial outcomes are reported separately for participants with and without identified learning disabilities.

Types of interventions

Interventions eligible for inclusion in this review will be required to meet several eligibility criteria. Only studies of universal or targeted intervention practices or programs evaluated for influence on prosocial behavior in children and/or youth will be eligible. Interventions must include behavioral and/or psychological content. Programs or practices may occur in a single exposure or be implemented over a longer-term (e.g., school year). For eligibility, interventions must be delivered or organized by a teacher, parent, or other adult (e.g., researcher, graduate student, school psychologist). Studies that only compare general early care, school, or afterschool (e.g., Sure Start, 21st Century Community Learning Centers) attendance with non-attendance will be ineligible. Although general early care, school, and afterschool programs often have standards for social and emotional learning, benchmarks or standards are not considered adequate enough to signify that social and emotional learning was integrated into instruction. Also, interventions that include pharmaceutical treatments will be excluded.

Types of outcome measures

The outcome to be examined is prosocial behavior. Prosocial behavior represents a broad category of voluntary acts intended to help or promote the wellbeing of others (Eisenberg & Mussen, 1989). Sharing resources, providing assistance or comfort, cooperating, donating, volunteering, and community service/outreach are common examples of prosocial acts. There are no restrictions on the type of measure of prosocial behavior (e.g., observational measure, teacher report, peer report, vignette measure) or the time frame in which measures of prosocial behavior were collected following treatment.

It should be noted that prosocial behavior is sometimes referred to as altruism. However, altruism generally refers to self-sacrificing actions that benefit others and do not involve self-gain (Rushton, 1982; Staub, 1978; Wispe, 1978). While altruistic behaviors are always prosocial, acts of prosociability are not always altruistic because prosocial behaviors do not require self-sacrifice and may even involve self-gain (e.g., social approval). In this meta-analysis, altruistic behavior will be included as a type of prosocial behavior.

An eligible measure is one in which at least 75% of items reflect prosocial behavior (i.e., acts intended to help or benefit the wellbeing of others). If a measure includes items that do not reflect prosocial behavior, these items need to examine other social and emotional competencies, skills, behaviors, attitudes and/or perceptions. This means that outcome measures that include more than 25% of items on friendliness, empathy, sympathy, peer acceptance, popularity, antisocial impulse regulation, and other indices of non-prosocial social and emotional outcomes are not eligible. Item-level information of measures of social and emotional competence (e.g., moral behavior, friendship skills, compassionate behavior, ethical behavior, character development, citizenship engagement) will be reviewed to see if they are compatible with the eligibility criteria for a qualifying outcome measure.

Duration of follow-up

All follow-up measures of prosocial behavior collected on both treatment and comparison group samples will be included. It is an objective of this review to provide information on whether and which extant instructional strategies show evidence of longer-term effects.

Types of settings

Studies conducted in most natural settings are eligible (e.g., home, school, after school program, camp, neighborhood, YMCA). However, studies conducted in laboratory, incarcerated, or inpatient settings are ineligible due to external validity concerns including the potential limitation in providing reliable inferences about the effects of practices and programs on behavior outside laboratory or monitored settings.

Geographical context

Studies may have been conducted in any country. Social behaviors and relational processes may dynamically interplay with culture (LeVine et al., 1994; Weisner, Gallimore, & Jordan, 1988), so broad cross-cultural comparability may be problematic. Consequently, the country of origin in which the research was conducted will be examined as a moderator in analysis.

Publication language

Studies must be reported in English. The choice to exclude studies from non-English language sources reflects the practical constraints in searching and translating literature not reported in English.

Date of publication

The date of publication or reporting of the study must be 1960 or later. Study of the effects on children's prosocial behavior began predominantly in the 1960's and 70's, mainly in the U.S., Canada, and Western Europe, when there was great interest in experimental work testing whether adults' actions could induce changes in children's prosocial responding. Social and psychological research on prosocial behavior rapidly accelerated in 1964 when the brutal stabbing of Katherine “Kitty” Genovese outside her apartment in Queens, New York while 38 neighbors and other witnesses did not assist or contact the police was highly publicized and ignited social scientists' exploration into reasons for why individuals act in ways that help others (Penner, Dovidio, Piliavin, & Schroeder, 2005).

General study characteristics

Study design, country of origin, date of publication or report, and study setting are study characteristics that may influence the results of a quantitative review and have important implications for external validity (Card, 2012). Particular attention will be made to study research design. Randomized experiments in which children or group of children (e.g., classrooms) were randomly assigned to treatment and control groups and data analysis was conducted at the level of random assignment will be compared to non-randomized studies (and also studies that have only a small number of groups for cluster random assignment).

Sample characteristics

One overarching question that this review seeks to address is whether most children and youth can benefit from practices and programs intended to enhance prosocial acts. Gathering a large representative body of studies will be critical in providing evidence toward this end, but it will also be essential to empirically investigate the degree to which our results can be generalized by examining whether participant characteristics change the magnitude of effect size. Sample characteristics of age/grade-level, gender, ethnicity/race, socioeconomic status, and baseline levels of prosocial behavior will be tested as moderators. These sample characteristics have been identified as factors associated with prosocial behavior (Eisenberg et al., 2006; Penner et al., 2005). For example, a meta-analysis on gender and age differences in prosocial behavior suggests general trends of females showing more prosocial behavior than males and increasing prosociability across childhood (Eisenberg & Fabes, 1998; Fabes, Carlo, Kupanoff, & Laible, 1999). However, it's unclear from prior literature whether children's background characteristics may influence treatment effects on prosocial behavior. A review of meta-analyses does provide evidence on the link between a few of children's background characteristics and efficacy of social emotional learning programs and skills for life programs. Conclusions were that children from lower economic status families benefit as much or more than their more affluent peers and that children across age groups benefit, but there may be variations in the age groups that benefit the most (Diekstra & Gravesteijn, 2008). It is also of substantive interest as to whether children with emotional or behavioral issues are receptive (or less receptive) to increases in prosocial behavior through intervention. Responsiveness to intervention may be differential for children who have emotional or behavioral issues because they may be less likely to accurately encode social information or generate relational goals (Crick & Dodge, 1996; Lemerise & Arsenio, 2000). In teasing out what works for whom, we hope to be able to formulate evidence-based recommendations, which can be further empirically evaluated, for working with populations of children with different characteristics and backgrounds.

General intervention characteristics

Instructional practices and programs vary in modality (i.e., targeted or universal intervention; single component vs. multicomponent programs; intervention in one or more contexts), format (i.e., treatment structure of individual versus group sessions), and dosage (i.e., frequency, intensity, and duration of a treatment). For practice, it is important to provide information on which service delivery variables are more conducive for increases in prosocial behavior. Meta-analyses of interventions that target teaching social and emotional skills suggest that treatments of short duration or low intensity often show smaller or insignificant effect sizes (Diekstra & Gravesteijn, 2008). Durlak et al. (2011) also found that multicomponent interventions (i.e., school programs with multiple components often supplemented with parent training and/or schoolwide initiatives) appear to be less effective than single-component programs in increasing positive social behavior of children and youth; however, these meta-analytic findings may be tempered by indications that multicomponent programs had more implementation problems.

This review will also provide evidence on the quantity and type of provider training and technical assistance associated with more effective delivery of instructional practices and programs. Durlak and DuPre's (2008) review of the research on the impact of implementation on child outcomes suggests the importance of documenting program approaches to insure provider proficiencies in the skills to conduct the intervention and to deliver technical assistance such as retraining or emotional support.

SAFE framework

Based on meta-analytic findings, interventions focused on social, emotional and/or personal skills tend to improve a range of outcomes when they employ SAFE implementation practices including sequenced activities to achieve goals, active learning techniques to help participants acquire skills, focused time on social, emotional, and/or personal development, and explicit objectives for social, emotional, and/or personal skills (Durlak et al., 2010; 2011; Durlak & Weissberg, 2012). This meta-analysis will investigate whether effects on prosocial behavior are moderated by the use of SAFE practices. However, the SAFE features “focused” and “explicit” will be adapted to reflect a focus on prosocial behavior.

Intervention content

We also plan to investigate whether practices and interventions with different intervention components appear to be more or less effective in supporting prosocial responses. This has important implications for selection and implementation of interventions that may have larger effects. These broad categories of intervention components are outlined in Appendix A and were discussed in the section on how the intervention might work.

Relationship of socializer (i.e., person providing treatment) to participant children

The relationship that adult socializers have with participants might impact treatment effects. It has been argued that parents can have strong causal effects on their children's learning and enactment of prosocial behavior (Eisenberg et al., 2006). However, there are also theoretical underpinnings that support and strengthen the case that teachers can make important contributions to facilitating children's prosocial behavior (e.g., Bandura, 1986). Because this review has an interest on whether educators can influence prosocial behavior, it will be important to provide evidence of whether the relationship of the socializer to the study participant (e.g., teacher, parent, unfamiliar adult) is a moderator of effect size. Explicitly, for generalization, future research, and practice and policy, it is key to provide indications of whether interventions used by non-familial familiar adults (e.g., teachers) and unfamiliar adults (e.g., researchers), who do not have the same kind of pre-existing affective bonds or patterns of social interchange that exist in parent-child relationships, produce positive behavioral change.

Characteristics of prosocial behavior outcome measure

For generalizing findings, it is important to examine how effect size relates to characteristics of measurement of prosocial outcomes. Toward this aim, the following moderators will be evaluated: measurement design (e.g., natural situation vs. experimentally designed or hypothetical situation; anonymous/unsupervised vs. supervised assessment of prosocial outcomes), method of report (i.e. self-report, other-report, observation), and construct measurement (i.e., 75% - 90% vs. < 90% of prosocial behavior items). Potentially, there may be limitations to external validity of prosocial outcomes assessed under the scrutiny of researchers or through the use of experimentally designed or hypothetical situations. It is also critical to examine how sustainable effects are. This has implications for gauging the amount of time required to deliver lasting interventions and for future research evaluating the importance of later booster sessions. For the studies that evaluate impacts following post-test assessment, moderator analyses will be conducted to explore the maintenance of the effects of intervention using categories of duration of follow-up (e.g., short-term, mid-term, and long-term) or timing of follow-up (e.g., months following conclusion of the intervention).

Characteristics of targets of prosocial behavior

It has been suggested that characteristics of the target of prosocial behavior may influence prosocial behavior (Padilla-Walker & Carlo, 2014). Characteristics of the relationship of the target to the enactor of prosocial behavior (e.g., familial vs. non-familial; friends, peers, teachers, and/or parents) and age, gender, ethnicity/race, and socio-economic status of the target will be examined in this review. There are different perspectives on how the identity of the target may influence prosocial behavior. Evolutionary biology perspectives would suggest that prosocial behaviors are more likely directed toward those that share kinship ties (e.g., Hastings, Zahn-Waxler, & McShane, 2005). Similar to this perspective, research has suggested that higher levels of prosocial behavior are directed toward family members and friends than strangers; however, there are also indications that children's prosocial behavior may be more frequent toward peers than family members, but this finding may vary by age and culture (de Guzman, Carlo, & Edwards, 2008; Eisenberg & Fabes, 1998; Padilla-Walker & Christensen, 2011). Other perspectives have suggested that prosocial behaviors may be influenced by whether or not a target is clearly distressed or in need of help (e.g., Hoffman, 2000). Consistent with social identity theory (Tajfel & Turner, 1986), the extent of homogeneity or heterogeneity of individuals' background characteristics may also be a contextual factor with relevance for how individuals treat one another. The interpersonal effect of perceived similarity upon helping has been widely noted in the literature on prosocial behavior (e.g., Dovidio, Gaertner, Validzic, Johnson, & Frazier, 1997). That said, the ethnic/racial similarity of participants to the targets of prosocial behavior will be explored as a moderator of effect size. Taken together, these perspectives suggest that characteristics of the target of prosocial behavior (as outlined in the outcome measure of prosocial behavior) may potentially influence the magnitude of effect size.

Methodological characteristics

Assessing the validity and reliability of the research included in a meta-analysis improves understanding of the ways in which flaws in the conduct of studies may bias the results and influence the conclusions that can be drawn (Higgins et al., 2011). An assessment of the validity of studies will be evaluated using the risk of bias procedures outlined by Higgins et al. (2011). Additional data will be collected on implementation fidelity and problems and outcome measure reliability. If sufficient data are collected on these methodological characteristics, they will be tested as moderators of effect size.

Electronic databases

British Library EThOS

Search terminology in electronic searches

The tentative search strategy that will be used to search ProQuest is outlined in Table 2. The search will be further refined by using the limiting commands to exclude studies which are retrieved by the search strategy but clearly are unrelated to the topic of the review (e.g., medical research on Hodgkin's disease). For other databases, this search strategy will be adapted using relevant Boolean logic and limiting commands. Unfortunately, it is not possible to further limit by names of instructional practices as they widely vary and comprehensive programs will also be included in the review. Many older studies are not well classified in terms of their search terminology in electronic databases. In preliminary test searches, placing more limitations on the current planned search resulted in the loss of relevant studies. For electronic searches, the database thesauri will be consulted to include any additionally relevant subject and keyword terminology in the search process. In addition, the head social sciences librarian of the Jean and Alexander Heard Libraries at Vanderbilt University will be consulted regarding the search strategy in order to identify further restrictions that could be imposed in the search process without losing potentially relevant studies.

Bibliography search of eligible research reports

The reference lists of all eligible research reports (and near misses ineligible because of weak study designs, etc.) will be scanned and screened for research reports not already identified. In addition, forward searching will occur for all eligible research reports in order to identify additional relevant studies (i.e., via Science Citation Index).

Bibliography search of relevant meta-analyses and literature reviews

Bibliographies of relevant meta-analyses and literature reviews found via the search process will be screened for research reports to include as part of this review. Some reviews have already been located during a preliminary search (e.g., Bar-Tal, 1976; Batson, 2012; Dovidio et al., 2006; Eisenberg, 1983; Eisenberg & Fabes, 1998; Eisenberg, Fabes, & Spinrad, 2006; Eisenberg & Mussen, 1989; Hastings et al., 2007; McCullough & Tabak, 2010; Mikulincer & Shaver, 2014; Nucci, Narvaez, & Krettenauer, 2014; Padilla-Walker & Carlo, 2014; Penner et al., 2005; Staub, 2003; Stürmer, & Snyder, 2010).

Professional Organization Website Searching

The websites of the professional organizations of the International Association of Applied Psychology, International Positive Psychology Association, American Psychological Association, and Society for Research on Child Development will be searched for potentially eligible research reports. There are also planned searches of websites of identified professional organizations that develop interventions designed to promote prosocial behavior (e.g., The Developmental Studies Center website).

Hand searching

After the electronic search is complete, the journals that include many eligible studies will be manually reviewed.

Contact key authors in the field

Key investigators who are known to be active in the field and experts identified during the review process will be contacted with a request to share any published, unpublished, and ongoing research relevant to the review.

Study 1

In a quasi-experimental pretest-posttest design study, Ramaswamy and Bergin (2009) examined the effects of teacher reinforcement and induction on preschoolers' spontaneous prosocial behavior in the classroom. Eight classrooms were randomly assigned to one of three treatment groups (i.e., reinforcement only, induction only, reinforcement and induction) or control. The sample consisted of 98 boys and girls between 3 and 5 years of age. The interventions were implemented in all but one treatment classroom for 7 weeks. Observational measures of children's prosocial behavior in the classroom were taken before and after intervention, and the results were separately reported for the four conditions.

Study 2

In an experimental design study, Staub (1971b) examined the impact of role-playing and induction on children's helping and sharing behavior. Pairs of kindergarten children, randomly selected from a sample of 75 boys and girls, were assigned to one of three treatment groups (i.e., role playing, induction, role playing with induction) or a straw man comparison group. There were 2 treatment/control group sessions. A day following the 2^nd treatment/control exposure, children were either assessed on whether they helped a distressed child or whether they shared material possessions with another child and helped an adult on a task. Four to 6 days later, children were given the assessment they had not received following treatment. Results were reported separately for the four conditions.

Effect Size Metric

Standardized mean difference effect sizes (Cohen's d formula in Lipsey & Wilson, 2001) will be reported with adjustments of the small-sample correction factor to provide unbiased estimates of effect size (Hedges, 1981; Hedges & Olkin, 1985). For binary outcomes, the Cox transformation will be used to convert log odds ratios into standardized mean difference effect sizes (Sánchez-Meca, Marin-Martinez, & Chácon-Moscoso, 2003) because most outcomes are expected to be reported on a continuous scale. Effect sizes will be computed from data in primary studies in such forms as means, standard deviations, percentage, frequencies, t-tests, F-tests, p levels, and other quantitative statistics via conversion formulas provided by Lipsey and Wilson (2001). All effect sizes will be coded such that a positive effect size reflects greater prosocial behavior in the treatment group relative to the control. Effect sizes will be reported using a 95% confidence interval. Coders will document the computations used for the effect size estimates derived from each study. The distributions of effect sizes and sample sizes will be examined to identify potential outliers. The criteria for what constitutes an outlier will be calculated using the outer fence procedure outlined by Tukey (1977). If necessary, extreme values will be Winsorized, or recoded to more moderate values, to prevent distortion of the results (Lipsey & Wilson, 2001).

Clustered data analysis issues

Not accounting for clustering in studies may result in overestimating the precision of effect size (Perera & Glasziou, 2007). For clustered data (e.g., studies that use classrooms as the unit of assignment to condition), we will adhere to the procedures outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins & Green, 2011). Adjustments will be made using the intraclass correlation coefficient (ICC). Specifically, in cases of continuous outcomes, the effective sample size will be calculated by taking the study sample size and dividing it by the “design effect” (i.e., 1 + (average cluster size — 1) ∗ ICC). For dichotomous outcomes, both number of participants and number of participants experiencing the event will be divided by the design effect. Study authors will be contacted if ICC information is not available in a research report. In cases in which ICCs are unobtainable from authors, we will consider other approaches to account for the effect of clustering such as selecting a close-matching ICC from our sample of ICCs (in terms of the outcome measure and nature of clusters) or another representative ICC reported in the literature (e.g., via existing databases of ICCs).

Handling missing or incomplete data

All reasonable attempts will be made to collect complete data on variables in the coding manual. Authors of reports will be contacted if sufficient information is not reported to calculate effect sizes or code selected moderators. If some studies continue to have missing data after contacting authors, missing moderator and effect size values will be estimated using multiple imputation methods. A series of imputed data sets will be generated using a program available for use with R statistical software (e.g., Amelia II; Honaker, King, & Blackwell, 2011). Using Rubin's rules (1987), the imputed data sets will be pooled to obtain overall estimates and standard errors. Multiple imputation has been shown to have advantages of reducing bias of estimates and standard errors and reducing loss of information and power (Schlomer, Bauman, & Card, 2010).

The data will be examined for patterns of systematic missingness to explore whether there is evidence that data may be missing not at random (MNAR). However, the application of MNAR models is of wide debate (Enders, 2010). If evidence of MNAR is found, this limitation to the findings will be presented in the discussion section but not handled in analyses.

Descriptive statistics

Descriptive statistics will be used to summarize the current state of the literature and highlight gaps in the research on the impacts of programs and practices on the prosocial behavior of children and youth. Descriptives will be synthesized across primary studies on characteristics of methodology, participants, context, instructional strategies, socializers of prosocial behavior, recipients of prosocial behavior, and outcomes.

Synthesis of Effect Sizes and Assessment of Heterogeneity of Effect Sizes

Meta-regression methods are planned to examine the overall effects and variability of effects of instructional practices and programs. If many eligible studies including several different treatment groups and one common control, multiple measures of the same underlying outcome, and multiple measures at different follow-periods, meta-regression with robust standard errors (RSE) will be used to model the dependencies in the data (see Hedges, Tipton, & Johnson 2010). This method would allow us to include multiple effect size estimates from the same study.

Random effects models are planned to account for the heterogeneity between studies given the likely variation in study characteristics and for the importance of supporting generalization of the findings beyond the included studies (Borenstein et al., 2009). Random effects weighted mean posttest effect sizes and 95% confidence intervals for each study will be calculated and graphically shown in a forest plot, which will also display the average effect size across studies. Using random effects models will allow us to account for within study sampling variance and between study variability.

Moderator Analyses

If there is significant variability in effects (i.e., if the Q statistic indicates significant heterogeneity in effect sizes), moderator analyses will explore for whom and under what conditions instructional strategies appear to enhance prosocial behavior. Estimates of the residual Cochrane's Q, I ², and τ ² will be used to assess residual variability in the effect sizes after inclusion of the potential moderators outlined in Table 1. Substantive categories of moderators (e.g., characteristics of the programs and practices and sample) will be evaluated within a regression framework to examine the association of thematically grouped moderator variables with study effect sizes (i.e., separate meta-regression models are planned for substantive groupings of moderators). However, first, relevant study characteristics (e.g., design, year of publication, country of origin) will be tested as moderators. Those methodological characteristics that are correlated with effect sizes will be included in all moderator analyses to correct for possible misidentification (e.g., overestimation or confounding with methodological variables) of the impact of substantive moderator variables (Lipsey, 2009b).

While moderator analyses can elucidate important information for informing and improving research, practice and policy, there are important considerations to address. Confounded moderators make it difficult to delineate the situations that may account for differential intervention effects (Lipsey, 2003). In this review, interrelationships of moderator variables with effect size and each other will be carefully examined to try to disentangle probable key moderators and program and practice effects from variables that simply tend to co-occur due to aspects of study methodology, etc. If there are substantial correlations among moderators, analyses will be conducted to investigate the potential interactions among the moderators. Another concern is that sufficient data may not be reported on selected moderating variables to examine their relationship with effect size. In this case, it may be necessary to exclude variables from moderator analyses and examine them descriptively.

Publication Bias

Publication bias stems from failing to detect unpublished studies. Underrepresentation of unpublished studies, which are more likely to have non-significant effects, can substantially bias effect size estimates (Borenstein et al., 2009). Although all reasonable attempts will be made to include unpublished research such as searching databases of conference proceedings and other grey literature and corresponding with experts in the field, some unpublished studies, particularly from decades past, will be unobtainable. Potential publication bias will be examined using Egger's regression-based assessment of asymmetry of funnel plots (Egger, Smith, Schneider, & Minder, 1997). If publication bias is suggested by the diagnostic test, then Duval and Tweedie's (2000) trim and fill method will be used to impute the potentially missing effect sizes and then the combined effect will be recomputed.

Sensitivity Analyses

Several types of sensitivity checks will be conducted to investigate the robustness of the results. Sensitivity analyses will be used to examine how sensitive results are to 1) imputed effect size and moderator values, 2) Winsorized outliers, 3) inclusion of non-randomized studies, and 4) duration of follow-up. The potential impact of results from the sensitivity analysis on the findings will be addressed in the final review.

Software for analyses

Analyses will be performed using meta-analysis commands that run in the R statistical environment and Stata version 12 (StataCorp, 2011). Excel spreadsheets will be used for some effect size calculations and proportion to percent calculations (and perhaps other calculations such as months to years, etc.).

Treatment of qualitative research

We do not plan to include purely qualitative research. The focus of this review is on quantifying the impact of instructional practices and programs on prosocial behavior in children and youth. We do, however, anticipate that the great majority of eligible studies will not include qualitative data.